+

CN115129979B - Data processing method and related device - Google Patents

Data processing method and related device Download PDF

Info

Publication number
CN115129979B
CN115129979B CN202210600688.1A CN202210600688A CN115129979B CN 115129979 B CN115129979 B CN 115129979B CN 202210600688 A CN202210600688 A CN 202210600688A CN 115129979 B CN115129979 B CN 115129979B
Authority
CN
China
Prior art keywords
content
historical
feature
target
recommended
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210600688.1A
Other languages
Chinese (zh)
Other versions
CN115129979A (en
Inventor
赵胜林
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN202210600688.1A priority Critical patent/CN115129979B/en
Publication of CN115129979A publication Critical patent/CN115129979A/en
Application granted granted Critical
Publication of CN115129979B publication Critical patent/CN115129979B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the application discloses a data processing method and a related device in the field of artificial intelligence, wherein the method comprises the steps of obtaining basic attribute information and historical behavior information of a target object, and target recommended content, wherein the historical behavior information comprises a plurality of historical recommended content of which the target object triggers interactive operation, performing feature coding processing on the basic attribute information, the plurality of historical recommended content and the target recommended content to obtain basic attribute characteristics, historical content characteristics of the plurality of historical recommended content and target content characteristics, performing attention weighting processing on the historical content characteristics of the plurality of historical recommended content according to the basic attribute characteristics and the target content characteristics to obtain personalized behavior characteristics, and determining target parameters for representing the interest degree of the target object in the target recommended content according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics. The method can improve the accuracy of the determined target parameters, thereby realizing accurate recommendation.

Description

Data processing method and related device
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a data processing method and a related device.
Background
For many applications, whether the recommendation system can accurately recommend the content of interest to the use object of the application is one of important factors affecting the use experience of the use object for the application.
In the related art, a recommendation system performs feature coding processing on basic attribute information and historical behavior information of a used object and content to be recommended to obtain coding features corresponding to the three information respectively, further, directly predicts the click rate or purchase rate of the used object on the content to be recommended according to the coding features corresponding to the three information respectively, and decides whether to recommend the content to be recommended to the used object according to the click rate or purchase rate of the used object on the content to be recommended.
However, the implementation effect of the above method in practical application is not ideal, and the accuracy of the click rate or purchase rate predicted by the above method is often low, which further affects the recommendation effect of the recommendation system, and it is difficult to ensure that the real interesting content is recommended to the user in preference.
Disclosure of Invention
The embodiment of the application provides a data processing method and a related device, which can improve the accuracy of a predicted parameter used for representing the interest degree of an object to the content to be recommended, thereby improving the recommendation effect of a recommendation system.
In view of this, a first aspect of the present application provides a data processing method, the method comprising:
acquiring basic attribute information and historical behavior information of a target object, and acquiring target recommended content to be recommended, wherein the historical behavior information comprises a plurality of historical recommended contents of which the target object triggers interactive operation;
performing feature coding processing on the basic attribute information, the plurality of historical recommended contents and the target recommended contents respectively to obtain basic attribute features of the basic attribute information, historical content features of the plurality of historical recommended contents and target content features of the target recommended contents;
According to the basic attribute characteristics and the target content characteristics, performing attention weighting processing on the historical content characteristics of each of the plurality of historical recommended contents to obtain personalized behavior characteristics;
And determining target parameters corresponding to the target recommended content according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics, wherein the target parameters are used for representing the interest degree of the target object in the target recommended content.
A second aspect of the present application provides a data processing apparatus, the apparatus comprising:
The information acquisition module is used for acquiring basic attribute information and historical behavior information of a target object and acquiring target recommended content to be recommended, wherein the historical behavior information comprises a plurality of historical recommended contents of which the target object triggers interactive operation;
The feature coding module is used for respectively carrying out feature coding processing on the basic attribute information, the plurality of historical recommended contents and the target recommended contents to obtain basic attribute features of the basic attribute information, historical content features of the plurality of historical recommended contents and target content features of the target recommended contents;
The attention weighting module is used for carrying out attention weighting processing on the historical content characteristics of each of the plurality of historical recommended contents according to the basic attribute characteristics and the target content characteristics to obtain personalized behavior characteristics;
and the parameter determining module is used for determining target parameters corresponding to the target recommended content according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics, wherein the target parameters are used for representing the interested degree of the target object in the target recommended content.
A third aspect of the application provides a computer apparatus comprising a processor and a memory:
The memory is used for storing a computer program;
The processor is configured to execute the steps of the data processing method according to the first aspect described above according to the computer program.
A fourth aspect of the present application provides a computer readable storage medium storing a computer program for executing the steps of the data processing method of the first aspect described above.
A fifth aspect of the application provides a computer program product or computer program comprising computer instructions stored on a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps of the data processing method according to the first aspect described above.
From the above technical solutions, the embodiment of the present application has the following advantages:
The embodiment of the application provides a data processing method, which comprises the steps of firstly obtaining basic attribute information and historical behavior information of a target object and target recommended content to be recommended, wherein the historical behavior information comprises a plurality of historical recommended content of which the target object triggers interactive operation, then respectively carrying out feature coding processing on the basic attribute information, the plurality of historical recommended content and the target recommended content which are included in the historical behavior information to obtain basic attribute characteristics of the basic attribute information, respective historical content characteristics of the plurality of historical recommended content and target content characteristics of the target recommended content, further carrying out attention weighting processing on the respective historical content characteristics of the plurality of historical recommended content according to the basic attribute characteristics and the target content characteristics to obtain personalized behavior characteristics, and finally, determining target parameters for representing the interested degree of the target object on the target recommended content according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics. When the behavior characteristics of the target object are constructed based on the historical behavior information of the target object, the basic attribute characteristics of the basic attribute information of the target object and the target content characteristics of the target recommended content are creatively utilized, the respective historical content characteristics of each historical recommended content in the historical behavior information are subjected to attention weighting processing, so that the personalized behavior characteristics of the target object are constructed, on one hand, the personalized behavior characteristics are constructed by referring to the basic attribute characteristics of the basic attribute information of the target object, so that the personalized behavior characteristics reflect the personal differences of the target object, and on the other hand, the personalized behavior characteristics are constructed by referring to the target content characteristics of the target recommended content, so that the personalized behavior characteristics reflect the preference of the target object to the target recommended content. Further, based on the personalized behavior characteristics, the target parameters used for representing the interested degree of the target object to the target recommended content are predicted, so that the predicted target parameters can be ensured to have higher accuracy, and based on the target parameters, the related recommended service is provided for the target object, and the recommended effect of the recommended service can be improved.
Drawings
Fig. 1 is a schematic diagram of an application scenario of a data processing method according to an embodiment of the present application;
FIG. 2 is a schematic flow chart of a data processing method according to an embodiment of the present application;
FIG. 3 is a schematic diagram of the working principle of a parameter prediction model according to an embodiment of the present application;
FIG. 4 is a schematic flow chart of attention weighting processing on historical content features according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a manner of determining personalized behavior features provided by an embodiment of the present application;
fig. 6 is a schematic diagram of the working principle of the PTA Unit according to the embodiment of the present application;
FIG. 7 is a software system architecture diagram of a data processing method according to an embodiment of the present application;
FIG. 8 is a schematic diagram of a data processing apparatus according to an embodiment of the present application;
Fig. 9 is a schematic structural diagram of a terminal device according to an embodiment of the present application;
fig. 10 is a schematic structural diagram of a server according to an embodiment of the present application.
Detailed Description
In order to make the present application better understood by those skilled in the art, the following description will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings, and it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments of the application described herein may be implemented in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
Artificial intelligence (ARTIFICIAL INTELLIGENCE, AI) is the theory, method, technique, and application system that simulates, extends, and extends human intelligence using a digital computer or a machine controlled by a digital computer, perceives the environment, obtains knowledge, and uses the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.
The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning, automatic driving, intelligent traffic and other directions.
The scheme provided by the embodiment of the application relates to an artificial intelligence technology, and is specifically described by the following embodiments:
In the related art, the implementation effect of methods for predicting the degree of interest of an object in contents to be recommended is generally not ideal, and the accuracy of prediction results obtained by these methods is generally low. In order to solve the above-mentioned problems, the embodiments of the present application provide a data processing method capable of effectively improving the accuracy of predicted target parameters (parameters for characterizing the interest degree of an object in content to be recommended), thereby helping to improve the recommendation effect of a recommendation system.
Specifically, in the data processing method provided by the embodiment of the application, basic attribute information and historical behavior information of a target object and target recommended content to be recommended are acquired first, wherein the historical behavior information comprises a plurality of historical recommended contents of which the target object triggers interactive operation. And then, respectively carrying out feature coding processing on the basic attribute information, the plurality of historical recommended contents and the target recommended contents included in the historical behavior information to obtain basic attribute features of the basic attribute information, historical content features of the plurality of historical recommended contents and target content features of the target recommended contents. And further, according to the basic attribute characteristics and the target content characteristics, performing attention weighting processing on the historical content characteristics of each of the plurality of historical recommended contents to obtain personalized behavior characteristics. Finally, a target parameter for characterizing the interest degree of the target object for the target recommended content can be determined according to the personalized behavior feature, the basic attribute feature and the target content feature.
According to the data processing method, the fact that the preferences of different objects for different recommended contents are different is considered, so that when the behavior characteristics of the target object are built based on the historical behavior information of the target object, the basic attribute characteristics of the basic attribute information of the target object and the target content characteristics of the target recommended contents are creatively utilized, the attention weighting processing is carried out on the historical content characteristics of each of the historical recommended contents in the historical behavior information, and personalized behavior characteristics of the target object are built. On one hand, the personalized behavior feature is constructed by referring to the basic attribute feature of the basic attribute information of the target object, so that the personalized behavior feature can reflect the personal difference of the target object, and on the other hand, the personalized behavior feature is constructed by referring to the target content feature of the target recommended content, so that the personalized behavior feature can reflect the preference of the target object to the target recommended content. Further, based on the personalized behavior characteristics obtained by construction, the target parameters used for representing the interested degree of the target object to the target recommended content are predicted, so that the predicted target parameters can be ensured to have higher accuracy, and the recommendation effect of the recommendation service can be improved by providing the related recommendation service for the target object based on the target parameters.
It should be understood that the data processing method provided by the embodiment of the present application may be performed by a computer device having data processing capability, where the computer device may be a terminal device or a server. The terminal equipment comprises, but is not limited to, a mobile phone, a computer, intelligent voice interaction equipment, intelligent household appliances, vehicle-mounted terminals, aircrafts and the like. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligent platforms. The terminal device and the server may be directly or indirectly connected through wired or wireless communication, which is not limited by the present application. In addition, the related data involved in the embodiments of the present application may be stored in a blockchain network.
In order to facilitate understanding of the data processing method provided by the embodiment of the present application, an application scenario of the data processing method is described below by taking an execution body of the data processing method as an example of a server.
Referring to fig. 1, fig. 1 is a schematic application scenario diagram of a data processing method according to an embodiment of the present application. As shown in fig. 1, the application scenario includes a server 110, a database 120 and a database 130, the server 110 may access the database 120 and the database 130 through a network, or the database 120 and the database 130 may be integrated in the server 110. The server 110 is configured to execute the data processing method provided by the embodiment of the present application to determine the interest degree of the target object for the target recommended content to be recommended, the database 120 is configured to store relevant information of the registered object on the target network platform, such as basic attribute information and historical behavior information of the registered object, and the database 130 is configured to store the content that can be recommended to the registered object on the target network platform.
In actual application, the target object may request the server 110 to provide the content recommendation service in the course of using the target application. For example, the terminal device running the target application may detect an operation triggered by the target object through the target application, and when detecting that the target object triggers to open the content recommendation interface of the target application, the target object may be considered to request the content recommendation service, and then send the content recommendation request to the server 110.
After receiving the content recommendation request sent by the terminal device, the server 110 may retrieve the basic attribute information and the historical behavior information of the target object from the database 120. The basic attribute information is information for reflecting the relevant basic attribute of the target object, and may include relevant information reserved when the target object registers the target application program, time information and frequency information of the target object registering the target application program, time and time information of the target object for using the target application program, and the like. The historical behavior information includes a plurality of historical recommended contents of the target object triggering the interactive operation, for example, the historical recommended contents of the target object which is clicked and looked up by the target application program in a specific period (such as in the last month and in the last three months), or the historical recommended contents of the target object which is triggered and consumed by the target application program in the specific period, wherein the historical recommended contents are recommended supporting contents of the target application program and can be short videos, audios, texts, commodities, game elements (such as game map, game skin, game equipment and the like), and the like.
In addition, after receiving the content recommendation request sent by the terminal device, the server 110 may also invoke the target recommended content to be recommended from the database 130. The target recommended content may be any content that the target application program can currently recommend to the use object, and similarly, the target recommended content may be content in the form of short video, audio, text, commodity, game element, and the like, and the application is not limited in this regard.
After obtaining the basic attribute information and the historical behavior information of the target object, and the target recommended content, the server 110 may perform feature encoding processing on the basic attribute information, the plurality of historical recommended contents included in the historical behavior information, and the target recommended content, to obtain respective corresponding encoding features, that is, obtain the basic attribute feature of the basic attribute information, the respective historical content features of the plurality of historical recommended contents, and the target content feature of the target recommended content.
Then, the server 110 may perform attention weighting processing on the historical content features of each of the plurality of historical recommended contents according to the basic attribute features and the historical content features obtained through the feature encoding processing, so as to obtain the personalized behavior feature of the target object. Specifically, in the process of constructing the behavior feature for representing the target object based on the respective historical content features of the plurality of historical recommended contents in the historical behavior information, the server 110 determines the respective attention weights of the respective historical recommended contents in the historical behavior information by utilizing the basic attribute features of the basic attribute information of the target object and the target content features of the target recommended contents, and further determines the personalized behavior feature of the target object based on the respective historical content features and the attention weights of the respective historical recommended contents, wherein the personalized behavior feature can reflect the personal difference of the target object on one hand and the preference condition of the target object to the target recommended contents on the other hand.
Further, the server 110 may determine a target parameter corresponding to the target recommended content according to the basic attribute feature, the target content feature, and the personalized behavior feature, where the target parameter is a parameter capable of characterizing the interest degree of the target object in the target recommended content, and the target parameter may be, for example, a predicted click rate, a purchase rate, and the like. Based on the target parameter, the server 110 may determine whether to recommend the target recommended content to the target object, or a presentation recommendation order corresponding to the target recommended content.
It should be understood that the application scenario shown in fig. 1 is only an example, and in practical application, the data processing method provided in the embodiment of the present application may also be applied to other scenarios, and the application scenario of the data processing method provided in the embodiment of the present application is not limited in any way.
The data processing method provided by the application is described in detail through the method embodiment.
Referring to fig. 2, fig. 2 is a flow chart of a data processing method according to an embodiment of the present application. For convenience of description, the following embodiments will be described by taking an execution body of the data processing method as a server as an example. As shown in fig. 2, the data processing method includes the steps of:
step 201, basic attribute information and historical behavior information of a target object are obtained, target recommended content to be recommended is obtained, and the historical behavior information comprises a plurality of historical recommended contents of which the target object triggers interactive operation.
In the embodiment of the application, when the server needs to provide the content recommendation service for the target object, the server can acquire the basic attribute information and the historical behavior information of the target object and acquire the target recommendation content to be recommended.
For example, the target object may trigger a related operation during use of the target application to request a background server of the target application to provide content recommendation services thereto. For example, the target object may trigger the target application to provide a content recommendation service for the target application by triggering and opening a content recommendation page of the target application (such as a game map download page in the game application, a character purchase interface, a skin purchase page, a prop purchase page, etc. in the game application, a commodity recommendation page in the shopping application, a video recommendation page in the video application, a music recommendation page in the music application, a news recommendation page in the news application, etc.), and after detecting that the target object triggers and opens the content recommendation page, the terminal device running the target application may generate a content recommendation request accordingly and send the content recommendation request to the server to request the server to provide the content recommendation service for the target object.
It should be appreciated that the target application may be any application that may provide content recommendation services for its use object, including but not limited to game applications, short video applications, shopping applications, news applications, video applications, audio applications, etc., and accordingly, the target object may be any object registered for use of the target application.
It should be understood that in practical application, the server may not only respond to the related operation triggered by the target object through the target application program and provide the content recommendation service for the target object, but also determine the content that can be recommended to the target object in advance for the target object, that is, the server may not be limited by whether the target object triggers the operation for requesting the content recommendation service through the target application program, and autonomously determine the content that can be recommended to the target object for the target object.
In general, for the usage object of the target application, the server uses a database to store relevant information of each usage object, such as basic attribute information and historical behavior information of the usage object, and for the recommended content delivered by the target application, the server uses a database to store each recommended content, where the database for storing relevant information of the usage object and the database for storing recommended content may be the same database or different databases. In this case, when the server needs to determine the content that can be recommended to it for the target object, the server may call the related information of the target object, that is, the basic attribute information and the history behavior information of the target object, from the database for storing the related information of the usage object, and at the same time, the server may call the target recommended content to be recommended from the database for storing the recommended content.
It should be understood that, in practical application, the server may acquire the basic attribute information and the historical behavior information of the target object in other manners, and acquire the target recommended content in other manners.
The basic attribute information of the target object is information reflecting the basic attribute related to the target object. The relevant basic attribute here may be, for example, a personal basic attribute of the target object, information for reflecting the personal basic attribute may include, but is not limited to, relevant information reserved when the target object registers the application program, interest information, and a level of the target object on the target application program, etc., and the relevant basic attribute here may be, for example, a behavior attribute generated when the target object uses the target application program, information for reflecting the behavior attribute may include, but is not limited to, time information or frequency information when the target object registers the target application program, time-of-use information of the target object for the target application program, click behavior information generated by the target object through the target application program (e.g., frequency information that the target object triggers a click view detail operation through the target application program), consumption behavior information generated by the target object through the target application program (e.g., frequency information that the target object triggers a purchase operation through the target application program), etc. The present application is not limited in any way herein to the information specifically included in the basic attribute information of the target object.
It should be noted that, the historical behavior information of the target object is information for reflecting the historical behavior characteristics of the target object, and in the embodiment of the present application, the historical behavior information includes a plurality of historical recommended contents, specifically at least two, of which the target object triggers the interactive operation. The history recommended content of the target object after triggering the interactive operation refers to the history recommended content of the target object after triggering the further interactive operation in the history recommended content which is exposed to the target object, wherein the interactive operation can be, for example, click view detail operation, purchase operation, the history recommended content is the content of the target application program which is recommended to the target object in history, and the history recommended content can be any content of the target application program which supports recommendation, and can include, for example, but not limited to game elements (such as game map, virtual role, role skin, game props and the like), video, audio, text and the like. The plurality of history recommended contents included in the history behavior information may specifically be history recommended contents in which the target object triggers the interaction operation within a preset history period, for example, history recommended contents in which the target object triggers the interaction operation within a month or three months. The present application is not limited in any way herein to the form of the history recommended content included in the history behavior information, the interactive operation corresponding to the history recommended content, and the time of the interactive operation corresponding to the history recommended content.
In one possible implementation manner, the server may distinguish different interactive operations and obtain the historical behavior sub-information corresponding to each of the multiple interactive operations, that is, the historical behavior information may include multiple historical behavior sub-information, where the multiple historical behavior sub-information corresponds to different interactive operations, and each historical behavior sub-information includes multiple historical recommended contents, where the multiple may be at least two, where the multiple historical recommended contents include the target object triggering the corresponding interactive operation.
For example, the server may distinguish the clicking operation from the purchasing operation, and obtain the historical behavior sub-information corresponding to each of the clicking operation and the purchasing operation to form the historical behavior information. For example, when the server obtains the historical behavior information for the target object, the historical recommended content of the target object triggering the clicking operation in the last month may be obtained, the historical recommended content corresponding to the clicking operation may be formed by using the historical recommended content, the historical recommended content of the target object triggering the purchasing operation in the last month may be obtained, the historical behavior sub-information corresponding to the purchasing operation may be formed by using the historical recommended content, and further, the server may form the historical behavior information of the target object by using the historical behavior sub-information corresponding to the clicking operation and the historical behavior sub-information corresponding to the purchasing operation.
It should be understood that, in practical application, the server may also acquire the corresponding historical behavior sub-information for other types of interaction operations, and the present application does not limit the type of interaction operation focused when acquiring the historical behavior sub-information.
In this way, different interactive operations are distinguished, the historical behavior information comprising the historical behavior sub-information corresponding to each of the various interactive operations is obtained, the corresponding personalized behavior characteristics are correspondingly constructed based on the historical behavior sub-information corresponding to each of the various interactive operations in a follow-up more detailed manner, the preference condition of the historical behavior of the target object on the recommended content is reflected in a finer granularity, and the accuracy of the finally predicted interested degree of the target object on the target recommended content is improved.
In the embodiment of the application, before the server acquires the basic attribute information and the historical behavior information of the target object, the target object is requested to grant the server with the authority to acquire the basic attribute information and the historical behavior information. The server may send an information acquisition authorization request to the target object before providing the content recommendation service for the target object, so as to prompt the target object to obtain the basic attribute information and the historical behavior information of the target object for providing the content recommendation service for the target object through the information acquisition authorization request, if the target object responds to the information acquisition authorization request and confirms that the server is granted permission to obtain the basic attribute information and the historical behavior information of the target object, the server may execute the method provided by the embodiment of the application, provide the content recommendation service for the target object based on the basic attribute information and the historical behavior information of the target object, and if the target object does not respond to the information acquisition authorization request or refuses to grant the server permission to obtain the basic attribute information and the historical behavior information of the target object in response to the information acquisition authorization request, the server cannot provide the content recommendation service for the target object.
It should be noted that, the target recommended content is any content that the target application program may recommend to the target object, and similar to the history recommended content, the target recommended content may include, for example, but not limited to, game elements (such as game map, virtual character, role skin, game prop, etc.), video, audio, text, etc., and the application is not limited in this regard. In addition, in the embodiment of the present application, the server may acquire one target recommended content or may acquire a plurality of target recommended contents, and the present application does not limit the number of acquired target recommended contents.
Optionally, in the embodiment of the present application, the server may further acquire target context information, so as to determine whether to recommend the target recommended content for the target object or determine a display recommendation sequence of the target recommended content for the target object by using the target context information in combination with the base attribute information and the historical behavior information of the target object and the target recommended content.
The target context information herein is at least one of information for reflecting a recent behavior state of the target object and information for reflecting a recent activity state of the target network platform (which is a platform supporting the target application program to perform network activities using the object), and specifically, may include at least one of information for reflecting an operation behavior generated recently by the target object and activity information developed recently by the target network platform. The information of the recently generated operation behavior of the target object may include at least one of time information of the target object requesting the content recommendation service, operation information of the target object generated in the first period, and activity information of the target object participating in the second period, wherein the time information of the target object requesting the content recommendation service may be, for example, time of the target object triggering opening of the content recommendation page through the target application, the operation information of the target object generated in the first period may be, for example, operation behavior information of the target object generated through the target application on the same day (such as click view details operation information of the target object generated through the target application, purchase operation information of the target object generated through the target application, etc.), the activity information of the target object participating in the second period may be, for example, game play information of the target object participating in the previous hour, and it should be understood that in practical application, the information for reflecting the recently generated operation behavior of the target object may also include other forms of information, and the application does not limit the information for reflecting the recently generated operation behavior of the target object. The activity information recently initiated by the target network platform may include activity information initiated by the target network platform in the third period, for example, commodity purchase preferential activity information of the target network platform on the same day, and the application is not limited in any way. It should be understood that the first period, the second period, and the third period may be the same period, or may be different periods, which is not limited in any way by the present application.
It should be noted that, for the information used to reflect the operation behavior generated recently by the target object in the above-mentioned target context information, the server needs to obtain the authorization of the target object before obtaining such information, that is, the server can obtain such information used to reflect the operation behavior generated recently by the target object only if obtaining the authorization granted by the target object to obtain the above-mentioned information.
Therefore, when the content recommendation service is provided for the target object, the target context information is introduced, and the accuracy of the determined interest degree of the target object for the target recommendation content can be improved by means of more information with higher reference value, so that the quality of the content recommendation service provided for the target object is improved.
And 202, respectively performing feature coding processing on the basic attribute information, the plurality of historical recommended contents and the target recommended contents to obtain basic attribute features of the basic attribute information, the historical content features of the plurality of historical recommended contents and target content features of the target recommended contents.
After the server obtains the basic attribute information and the historical behavior information of the target object and the target recommended content, the obtained basic attribute information can be subjected to feature encoding processing to obtain features for reflecting the basic attribute information, namely basic attribute features, each historical recommended content included in the obtained historical behavior information can be subjected to feature encoding processing to obtain features for reflecting each historical recommended content, namely historical content features of each historical recommended content, and the obtained target recommended content can be subjected to feature encoding processing to obtain features for reflecting the target recommended content, namely target content features.
It should be understood that the above feature encoding process is a process for converting information or content into machine-recognizable numerical information (i.e., vectors), and the converted vectors can reflect the features of their corresponding information or content to some extent. That is, the basic attribute feature obtained by performing feature encoding processing on the basic attribute information is a vector capable of reflecting the self feature of the basic attribute information, the history content feature obtained by performing feature encoding processing on the history recommended content is a vector capable of reflecting the self feature of the history recommended content, and the target content feature obtained by performing feature encoding processing on the target recommended content is a vector capable of reflecting the self feature of the target recommended content.
In the embodiment of the application, the feature coding processing can be performed on each historical recommended content and the target recommended content in the basic attribute information, the historical behavior information through the corresponding feature coding network. Referring to fig. 3, an exemplary embodiment of the present application is shown in fig. 3, where fig. 3 is a schematic diagram of a working principle of a parameter prediction model provided by the embodiment of the present application, where the parameter prediction model is a neural network model for predicting a target parameter capable of characterizing a degree of interest of a target object with respect to a target recommended content, and as shown in fig. 3, the parameter prediction model includes a basic attribute feature encoding network 301, a historical content feature encoding network 302, and a target content feature encoding network 303, where the basic attribute feature encoding network 301 is used for performing feature encoding processing on basic attribute information of the target object to obtain a corresponding basic attribute feature, the historical content feature encoding network 302 is used for performing feature encoding processing on the historical recommended content to obtain a corresponding historical content feature, and the target content feature encoding network 303 is used for performing feature encoding processing on the target recommended content to obtain a corresponding target content feature.
Optionally, if the server further obtains the target context information through step 201, the server needs to perform feature encoding processing on the target context information at this time to obtain the target context feature, where the target context feature is a vector capable of reflecting the feature of the target context information. Illustratively, as shown in fig. 3, the parameter prediction model further includes a context feature encoding network 304, and in specific operation, the feature encoding network 304 may perform feature encoding processing on the target context information to obtain a corresponding target context feature.
And 203, performing attention weighting processing on the historical content characteristics of each of the plurality of historical recommended contents according to the basic attribute characteristics and the target content characteristics to obtain personalized behavior characteristics.
The server obtains corresponding basic attribute characteristics, respective historical content characteristics of each historical recommended content and target content characteristics by respectively carrying out characteristic coding processing on each historical recommended content and target recommended content in the basic attribute information and the historical behavior information, and then the server can carry out attention weighting processing on the historical content characteristics of each historical recommended content characteristic by utilizing the basic attribute characteristics and the target content characteristics, so that personalized behavior characteristics of the target object are obtained.
It should be noted that the personalized behavior feature is a feature vector determined by comprehensively referring to the features of the three information, namely, the historical behavior information, the basic attribute information and the target recommended content, and the personalized behavior feature can embody the historical behavior feature of the target object on one hand, and the personal difference of the target object on the other hand, and can embody the preference condition of the target object on the target recommended content on the other hand, in other words, the personalized behavior feature is a vector representation of the content preference condition of the target object by the historical interaction of the target object.
As shown in fig. 3, in the parameter prediction model, the attention weighting process may be performed by the personalized target attention Unit305 (Personalized Target Attention Unit, PTA Unit), that is, the basic attribute feature, the target content feature, and the history content feature of each history recommended content may be input to the PTA Unit305, the input basic attribute feature, target content feature, and history content feature of each history recommended content may be analyzed and processed by the PTA Unit305, and the attention weighting process may be performed on the history content feature of each history recommended content by using the basic attribute feature and the target content feature, thereby outputting the personalized behavior feature of the target object.
In a possible implementation manner, if the historical behavior information acquired by the server includes the historical behavior sub-information corresponding to each of the plurality of interactions, when the server constructs the personalized behavior feature of the target object, attention weighting processing may be performed on the historical content feature of each of the plurality of historical recommended contents included in the historical behavior sub-information according to the basic attribute feature and the target content feature, so as to obtain the personalized behavior feature corresponding to the historical behavior sub-information.
Specifically, if different interactive operations are distinguished when the server obtains the historical behavior information, and the historical behavior sub-information corresponding to each of the multiple interactive operations is obtained (each of the historical behavior sub-information includes a plurality of historical recommended contents of the corresponding interactive operation triggered by the target object), the server can correspondingly distinguish the different interactive operations when generating the personalized behavior feature, and generate the personalized behavior feature corresponding to each of the different interactive operations. That is, for each piece of history behavior sub-information, the server may perform attention weighting processing on the history content features of each of the plurality of history recommended contents included in the history behavior sub-information by using the basic attribute feature and the target content feature, so as to obtain a personalized behavior feature corresponding to the history behavior sub-information, that is, a personalized behavior feature of the interaction operation corresponding to the history behavior sub-information.
Thus, different interactive operations are distinguished, personalized behavior characteristics corresponding to the different interactive operations are built, the preference condition of the historical interaction of the target object on the recommended content can be reflected in a finer granularity, the target parameter is predicted according to the personalized behavior characteristics in the fine granularity, and the accuracy of the predicted target parameter can be improved.
In one possible implementation, the server may specifically perform, through the flow shown in fig. 4, attention weighting processing on the respective historical content features of the respective historical recommended contents in the historical behavior information (or the historical behavior sub-information) according to the basic attribute features and the target content features. As shown in fig. 4, the flow of the attention weighting process includes the steps of:
step 401, determining personalized content features according to the basic attribute features and the target content features.
When the server constructs the personalized behavior feature of the target object, the personalized content feature can be firstly constructed according to the basic attribute feature and the target content feature, wherein the personalized content feature is a feature vector which is determined based on the personal attribute of the target object and can reflect the preference condition of the target object for the target recommended content.
As an example, the server may determine the personalized content feature by performing a feature fusion process on the base attribute feature and the target content feature to obtain a first fused feature, then processing the first fused feature through a first attention network to obtain a first personalized fused feature, and further determining the personalized content feature according to the target content feature and the first personalized fused feature.
Taking the attention weighting process of the PTA Unit on the respective historical content features of each historical recommended content as an example, the PTA Unit may perform feature stitching process on the basic attribute features and the target content features to obtain the first fusion feature. The PTA Unit may then input the first fused feature into a first Attention network, which processes the input first fused feature based on an Attention (Attention) mechanism accordingly, resulting in a first personalized fused feature. Furthermore, the PTA Unit may perform a summation process on the first personalized fusion feature and the target content feature, to obtain a personalized content feature.
The above manner of generating personalized content features may be exemplarily represented by the following formula (1):
t′=t+matmul(W1,[u,t]) (1)
Wherein t' is a personalized content feature, t is a target content feature, u is a basic attribute feature, W1 is a model parameter of the first attention network, [ u, t ] is a first fusion feature obtained by splicing the target content feature and the basic attribute feature, matmul () is a matrix multiplication calculation, and matmul (W1, [ u, t ]) is a first personalized fusion feature generated through the first attention network.
Accordingly, the attention weights of the historical recommended contents are further determined based on the personalized content characteristics, and the determined attention weights can be ensured to accurately reflect the importance degree of the corresponding historical content characteristics for reflecting the preference condition of the target object.
As another example, the server may determine the personalized content feature by performing feature conversion processing on the basic attribute feature to obtain the reference attribute feature, performing feature fusion processing on the basic attribute feature, the target content feature, and a product of the reference attribute feature and the target content feature to obtain a second fusion feature, further, processing the second fusion feature through a second attention network to obtain a second personalized fusion feature, and finally, determining the personalized content feature according to the target content feature and the second personalized fusion feature.
For example, taking the attention weighting process of the PTA Unit on the respective historical content features of each historical recommended content as an example, the PTA Unit may perform feature conversion processing on the basic attribute features through the feature conversion layer to obtain reference attribute features, where the reference attribute features are features with the same dimension as the target content features. Then, the PTA Unit can multiply the reference attribute feature and the target content feature to obtain the product of the reference attribute feature and the target content feature, and perform feature stitching processing on the basic attribute feature, the target content feature and the product of the target content feature to obtain a second fusion feature. Further, the PTA Unit may input the second fused feature into a second attention network, which processes the input second fused feature based on an attention mechanism accordingly, resulting in a second personalized fused feature. Finally, the PTA Unit may perform a summation process on the second personalized fusion feature and the target content feature, to obtain a personalized content feature.
The above manner of generating personalized content features may be exemplarily represented by the following formulas (2) and (3):
u′=W′*u (2)
t′=t+W2*[u,t,u′*t]) (3)
Wherein u 'is a reference attribute feature, W' is a model parameter of the feature conversion layer, and u is a basic attribute feature. t ' is a personalized content feature, t is a target content feature, W2 is a model parameter of the second attention network, u't is a product of the reference attribute feature and the target content feature, and [ u, t, u't ] is a splice base attribute feature, the target content feature, and a second fusion feature obtained by the product of the reference attribute feature and the target content feature.
In this way, the personalized content features are generated in the manner, so that the generated personalized content features can be guaranteed to better reflect the preference condition of the target object for the target recommended content, further, the attention weight determined based on the personalized content features can be guaranteed, and the importance degree of the corresponding historical content features for reflecting the preference condition of the target object can be embodied more accurately.
It should be understood that, in practical application, the server may determine the personalized content feature according to the basic attribute feature and the target content feature in other ways besides the two ways of determining the personalized content feature, and the present application is not limited in any way herein.
Step 402, determining the attention weight of each of the plurality of historical recommended contents according to the historical content characteristics and the personalized content characteristics of each of the plurality of historical recommended contents.
After the server determines the personalized content features, the server may further determine the attention weight of each historical recommended content according to the respective historical content features and the personalized content features of each historical recommended content, where the attention weight of each historical recommended content is a parameter for reflecting the importance degree of the corresponding historical content features for reflecting the preference condition of the target object for the target recommended content.
As an example, the server may determine the respective attention weights of the respective historical recommended content by, for each historical recommended content, determining a reference weight for the historical recommended content based on the personalized content features and the historical content features of the historical recommended content, then determining a total reference weight based on the respective reference weights of the respective historical recommended content, and further, for each historical recommended content, determining the attention weight of the historical recommended content based on the reference weight and the total reference weight of the historical recommended content.
Specifically, for each of the historical recommended contents, the server may first determine, according to the personalized content features determined in step 401 and the historical content features of the historical recommended content itself, a parameter for reflecting the importance degree of the historical content features of the historical recommended content to reflect the preference condition of the target object for the target recommended content, that is, determine a reference weight of the historical recommended content, where the reference weight is an intermediate parameter generated in the process of calculating the attention weight.
In one possible implementation, the server may determine its reference weight for each historical recommendation content by performing a feature fusion process on the personalized content feature, the historical content feature of the historical recommendation content, the difference between the personalized content feature and the historical content feature, and the product of the personalized content feature and the historical content feature to obtain a third fused feature, and further, processing the third fused feature through a third attention network to obtain the reference weight of the historical recommendation content.
Taking the PTA Unit as an example to determine the reference weight of each history recommended content, the PTA Unit may calculate the difference between the personalized content feature and the history content feature of the history recommended content, calculate the product of the personalized content feature and the history content feature of the history recommended content, then, the PTA Unit may perform a stitching process on the personalized content feature, the history content feature of the history recommended content, and the difference and the product to obtain a third fused feature, and further, the PTA Unit may input the third fused feature into a third attention network, and the third attention network may output the reference weight of the history recommended content correspondingly by performing an analysis process on the third fused feature.
The above-described manner of generating the reference weight of the history recommended content can be exemplarily represented by the following formula (4):
Att_j=matmul(W3,[t′,i_j,t′-i_j,t′*i_j]) (4)
Wherein att_j is a reference weight of the jth historical recommended content, W3 is a model parameter of the third attention network, t 'is a personalized content feature generated in step 401, i_j is a historical content feature of the jth historical recommended content, t' -i_j is a difference between the personalized content feature and the historical content feature of the jth historical recommended content, t '. Times.i_j is a product of the personalized content feature and the historical content feature of the jth historical recommended content, [ t', i_j, t '-i_j, t'. Times.i_j ] is a third fusion feature obtained by splicing the personalized content feature, the historical content feature of the jth historical recommended content, and the difference and the product of the personalized content feature and the historical content feature of the jth historical recommended content, and matmul () is a matrix multiplication calculation.
After the reference weights of the historical recommended contents are obtained by calculation, the server can further calculate the total reference weight according to the reference weights of the historical recommended contents, for example, the reference weights can be processed through an exp function (i.e. an exponential function based on e), and then the sum of the parameters obtained by processing is calculated as the total reference weight. Further, for each history recommended content, the attention weight of the history recommended content is calculated based on the reference weight and the total reference weight of the history recommended content, and for example, the attention weight of the history recommended content may be obtained by dividing a parameter obtained by processing the reference weight of the history recommended content by an exp function by the total reference weight.
The above manner of calculating the attention weight of the history recommended content can be exemplarily represented by the following formula (5):
a_j=exp(Att_j)/sum(exp(Att_1)+exp(Att_2)+...+exp(Att_n)) (5)
Wherein a_j is the attention weight of the jth history recommended content, att_j is the reference weight of the jth history recommended content, att_1, att_2, att_n are the reference weight of the first history recommended content, the reference weight of the second history recommended content, att_n are the reference weight of the nth history recommended content, n is the total number of history recommended contents included in the history behavior information, or the total number of history recommended contents included in the history behavior sub information, respectively, exp () represents an exponential function based on e.
It should be understood that, in the case of distinguishing the historical behavior sub-information corresponding to different interactions in the historical behavior information, when the server executes step 402, the attention weights of the historical recommended content in the different historical behavior sub-information may be determined separately. The server can determine the total reference weight corresponding to the historical behavior sub-information according to the reference weight of each historical recommended content in the historical behavior sub-information, and further determine the attention weight of the historical recommended content according to the reference weight of the historical recommended content and the total reference weight corresponding to the historical behavior sub-information according to each historical recommended content in the historical behavior sub-information.
In this way, by determining the respective attention weights of the historical recommended contents in the historical behavior information or the historical behavior sub-information in the manner, the determined attention weights can be ensured to more accurately reflect the importance degree of the corresponding historical content characteristics for reflecting the preference condition of the target object for the target recommended contents, and accordingly, attention weighting processing is performed on each historical content characteristic based on the attention weights determined in this way, so that the obtained personalized behavior characteristics can be ensured to more accurately reflect the preference condition of the target object for the target recommended contents.
Step 403, determining personalized behavior characteristics according to the historical content characteristics and the attention weights of the historical recommended contents.
After determining the attention weights of the historical recommended contents through step 402, the server may perform attention weighting processing on the historical content features according to the historical content features and the attention weights of the historical recommended contents, so as to obtain personalized behavior features of the target object. It should be understood that if the historical behavior sub-information corresponding to different interactions is distinguished in the historical behavior information, when the server executes step 403, the corresponding personalized behavior feature of each historical behavior sub-information may be determined, that is, the server may determine, for each historical behavior sub-information, the personalized behavior feature corresponding to the historical behavior sub-information according to the historical content feature and the attention weight of each historical recommended content included in the historical behavior sub-information.
As an example, the server may determine the personalized behavior feature by weighting and summing the historical content features of each of the plurality of historical recommended content based on the attention weights of each of the plurality of historical recommended content to obtain the personalized behavior feature.
Namely PTAUnit can directly utilize the respective attention weights of the historical recommended contents to carry out weighted summation processing on the respective historical content characteristics of the historical recommended contents so as to obtain personalized behavior characteristics. This way of generating personalized behavioral characteristics can be represented by the following formula (6):
h=a_1*i_1+a_2*i_2+...+a_n*i_n (6)
Wherein h is a personalized behavior feature, a_1, a_2, & gt, a_n are the attention weights of the first to nth historical recommended contents in the historical behavior information or the historical behavior sub-information respectively, i_1, i_2, & gt, i_n are the historical content features of the first to nth historical recommended contents in the historical behavior information or the historical behavior sub-information respectively, and n is the total number of the historical recommended contents included in the historical behavior information or the historical behavior sub-information.
In this way, the personalized behavior characteristics determined in the above manner can better characterize the preference condition of the target object for the target recommended content, and the personalized behavior characteristics are particularly suitable for scenes which do not need to be particularly concerned with the historical behavior time sequence, wherein the scenes which do not need to be particularly concerned with the historical behavior time sequence refer to scenes in which the interest preference of the object does not change obviously with time.
As another example, the server may determine the personalized behavior feature by, where possible, a time arrangement of a plurality of history recommended contents in the history behavior information (or history behavior sub-information) on which the target object triggers the interactive operation, and a history content feature of a first history recommended content in the history behavior information (or history behavior sub-information) is determined based on the first history recommended content itself, and a history content feature of each history recommended content in the history behavior information (or history behavior sub-information) other than the first history recommended content is determined based on the history recommended content itself and a previous history recommended content adjacent to the history recommended content in the history behavior information. At this time, for a first one of the history behavior information (or history behavior sub-information), the server may determine a reference personalized feature of the first one of the history recommended content according to the history content feature and the attention weight of the first one of the history recommended content, and for each one of the history recommended content except the first one of the history behavior information (or history behavior sub-information), determine a reference personalized feature of the history recommended content according to the history content feature and the attention weight of the history recommended content and a reference personalized feature of a previous one of the history behavior information (or history behavior sub-information) adjacent to the history recommended content, and finally, may determine a reference personalized feature of a last one of the history behavior information (or history behavior sub-information) as a personalized behavior feature corresponding to the history behavior sub-information.
Fig. 5 is an implementation schematic diagram of the manner of determining the personalized behavior feature. As shown in fig. 5, the historical behavior information acquired by the server includes a plurality of historical recommended contents, the historical recommended contents are arranged in order from front to back according to the time of triggering the interactive operation by the target object, for example, for the historical recommended content 1, the historical recommended content 2, the historical recommended content n in the historical behavior information, the target object triggers the interactive operation for the historical recommended content 1 first, then triggers the interactive operation for the historical recommended content 2, and so on, and triggers the interactive operation for the historical recommended content n at the latest.
When the server executes the feature encoding processing of each history recommended content in the history behavior information in step 202, the server may perform feature encoding processing of the history recommended content through the gating circulation unit (Gate Recurrent Unit, GRU) for the first history recommended content in the history behavior information, to obtain the history content feature of the first history recommended content. For each history recommended content in the history behavior information except for the first history recommended content, the server can determine the history content characteristics of the history recommended content according to the history recommended content itself and the history content characteristics of the previous history recommended content adjacent to the history recommended content in the history behavior information through the GRU, for example, when determining the history content characteristics of the history recommended content for the second history recommended content, the server can determine the content characteristics of the second history recommended content according to the history content characteristics of the second history recommended content itself and the first history recommended content through the GRU, and so on.
When the server determines the personalized behavior feature, for the first historical recommended content, a reference personalized feature of the first historical recommended content may be generated by a gating loop unit (Attention Gate Recurrent Unit, AGRU) based on the attention mechanism, according to the attention weight of the first historical recommended content determined by step 402, and the historical content feature of the first historical recommended content. For example, the server may first perform weighting processing by using the historical content feature and the attention weight of the first historical recommended content to obtain the corresponding attention feature, and then generate, by the AGRU, the reference personalized feature of the first historical recommended content according to the attention feature and the historical content feature of the first historical recommended content.
For each history recommended content except the first history recommended content in the history behavior information, the reference personalized feature of the history recommended content can be determined through the AGRU according to the attention weight and the history content feature of the history recommended content and the reference personalized feature of the previous history recommended content adjacent to the history recommended content in the history recommended information, for example, for the second history recommended content in the history behavior information, the weighting processing can be performed by using the history content feature and the attention weight of the second history recommended content to obtain the corresponding attention feature, and further, the reference personalized feature of the second history recommended content can be determined through the AGRU according to the attention feature, the history content feature of the second history recommended content and the reference personalized feature of the first history recommended content.
Finally, the server can use the reference personalized feature of the last historical recommended content in the historical behavior information as the personalized behavior feature of the target object.
In this way, the personalized behavior characteristics determined by the method can better represent the preference condition of the target object for the target recommended content, and the personalized behavior characteristics are particularly suitable for scenes needing special attention for the historical behavior time sequence, wherein the scenes needing special attention for the historical behavior time sequence refer to scenes in which the interest preference change of the object can change obviously with time.
Fig. 6 is a schematic diagram of an operating principle of a PTA Unit according to an embodiment of the present application, where the operating principle shown in fig. 6 corresponds to the manner of performing attention weighting processing on the historical content characteristics of each of the historical recommended content in the historical behavior information (or the historical behavior sub-information) shown in fig. 4. As shown in fig. 6, the PTA Unit determines the attention weights of the respective historic recommended contents according to the basic attribute features, the target content features and the historic content features of the respective historic recommended contents, and further, the PTA Unit processes the respective historic content features of the respective historic recommended contents by using the attention weights of the respective historic recommended contents, thereby obtaining the personalized behavior features.
By generating the personalized behavior feature in the above manner, the generated personalized behavior feature can be well fused with the basic attribute feature and the historical behavior feature of the target object and the target content feature of the target recommended content, so that the personalized behavior feature can accurately reflect the historical behavior feature of the target object, the personal difference of the target object and the preference condition of the target object to the target recommended content.
And 204, determining target parameters corresponding to the target recommended content according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics, wherein the target parameters are used for representing the interest degree of the target object in the target recommended content.
After determining the personalized behavior feature of the target object through step 203, the server may further determine the target parameter corresponding to the target recommended content according to the personalized behavior feature, the basic attribute feature and the target content feature. The target parameter is a parameter for characterizing the interest degree of the target object in the target recommended content, the target parameter may be, for example, a determined click rate (i.e., a probability that the target object triggers a click view detail operation on the target recommended content), the target parameter may be, for example, a determined purchase rate (i.e., a probability that the target object triggers a purchase operation on the target recommended content), and of course, the target parameter may be other parameters, which is not specifically limited in the present application herein.
The server may perform feature fusion processing on the personalized behavior feature, the basic attribute feature, and the target content feature, for example, performing processing such as splicing (concat) and leveling (flat) to obtain a fusion feature, and further, performing analysis processing on the fusion feature by using a parameter prediction model (i.e., a neural network model for predicting a target parameter), and outputting the target parameter.
In one possible implementation manner, if the historical behavior information includes historical behavior sub-information corresponding to each of a plurality of different interactions, and the personalized behavior characteristics corresponding to each of the historical behavior sub-information are determined in step 203, when the server predicts the target parameter, the server may determine the target parameter according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics corresponding to each of the historical behavior sub-information.
The server can perform feature fusion processing, such as splicing, leveling and the like, on the personalized behavior features, the basic attribute features and the target content features corresponding to each historical behavior sub-information to obtain fusion features, and further analyze and process the fusion features through a parameter prediction model to output target parameters.
In this way, in the process of determining the target parameters, different interactive operations are distinguished, and the target parameters are determined by utilizing personalized behavior characteristics corresponding to the historical behavior sub-information with finer granularity, so that the accuracy of the determined target parameters is improved.
In another possible implementation manner, if the server obtains the target context information through step 201 and performs feature encoding processing on the target context information to obtain the target context feature through step 202, when the server predicts the target parameter, the server may determine the target parameter according to the personalized behavior feature, the basic attribute feature, the target content feature and the target context feature.
For example, as shown in fig. 3, the server may perform feature fusion processing on the personalized behavior feature, the basic attribute feature, the target content feature, and the personalized behavior feature, such as performing stitching and leveling processing, to obtain a fusion feature, and further, analyze and process the fusion feature by using a parameter prediction model (i.e., a neural network model for predicting the target parameter), and output the target parameter.
When the fusion feature is analyzed and processed by utilizing the parameter prediction model, the fusion feature can be input into a full connection layer and an activation layer in the reference prediction model, and a sigmoid function is accessed after stacking, so that the predicted target parameter is obtained. Assuming that the full-connection layer is a full-connection layer based on Prelu activation functions, the full-connection layer is denoted as fc (u, i) = Prelu (w×u, i, c, h ] +b), wherein W and b are model parameters of the full-connection layer, u is a basic attribute feature, i is a target content feature, c is a target context feature, h is a personalized behavior feature, and [ u, i, c, h ] is a fusion feature obtained by performing feature fusion processing on the personalized behavior feature, the basic attribute feature, the target content feature and the personalized behavior feature, and the fusion feature can be processed through stacked multi-layer full-connection layers in the parameter prediction model. After the fusion feature is processed by the multi-layer stacked full-connection layer, the fusion feature can enter a sigmoid layer, the sigmoid layer is used for performing linear conversion processing on an input feature vector, and a target parameter to be predicted is obtained through a sigmoid function, wherein the target parameter is represented by p=sigmoid (W '. Times. vfc +b'), p is the target parameter, vfc is the feature vector obtained through the multi-layer full-connection layer processing, and W 'and b' are model parameters of the sigmoid layer.
In this way, in the process of determining the target parameter, the target context characteristics of the target context information are introduced, so that more characteristics with higher reference value can be introduced, thereby being beneficial to improving the accuracy of the determined target parameter.
It should be understood that, in practical applications, when the above-mentioned parameter prediction model is trained, the parameter prediction model may be trained based on a cross entropy (Cross Entropy Loss, CELoss) loss function, specifically, cross entropy loss values of all training samples may be averaged, and the parameter prediction model is trained based on an average value of the cross entropy loss values, which is denoted as loss=mean (y ' log (1-p) + (1-y ') logp), where loss is a loss value utilized when the parameter prediction model is trained, mean () is a function of averaging, y ' is a label in the training samples, and p is a target parameter predicted by the trained parameter prediction model. Of course, in practical applications, the parameter prediction model may be trained based on other types of loss values, which is not limited in this way.
In one possible application scenario, if the target recommended content acquired by the server through step 201 includes a plurality of target parameters, and the respective target parameters are determined for each target recommended content through steps 202 to 204, the server may rank each target recommended content according to the respective target parameters of each target recommended content, and determine a recommendation display order of each target recommended content, and further, the server may recommend each target recommended content to the target object based on the recommendation display order.
For example, assuming that each target recommended content obtained by the server through step 201 is a content that needs to be recommended to the target object, the server may determine, through steps 202 to 204, a target parameter corresponding to each obtained target recommended content, and further order each target recommended content according to the order of the target parameters from large to small, to obtain a recommended display order of each target recommended content. Accordingly, the server can recommend respective target recommended contents to the target recommendation based on the recommendation presentation order.
Therefore, based on the target parameters determined by the method provided by the embodiment of the application, the target recommended contents to the target object are ordered, the recommendation display sequence of the obtained target recommended contents can be ensured to be matched with the preference condition of the target object, namely the target recommended contents which are interested in the target object are recommended preferentially, and the target recommended contents which are interested in the target object to a lower degree are recommended later, so that the use experience of the target object for the target application program is improved.
It should be understood that, in the embodiment of the present application, if the target recommended content acquired by the server through step 201 is not all the content to be recommended to the target object, the server may first screen the acquired target recommended content according to the target parameters corresponding to each target recommended content, and filter the target recommended content with no interest or low interest degree of the target object, and further, for each retained target recommended content, determine the corresponding recommended display sequence by combining the target parameters corresponding to each target recommended content.
Optionally, for each target recommended content recommended to the target object, the server may acquire a behavior of the target object on the target recommended content as a training annotation behavior, construct an optimization training sample according to the basic attribute information and the historical behavior information of the target object, the target recommended content and the training annotation behavior, and then perform optimization training on the parameter prediction model for predicting the target parameter based on the thus constructed optimization training sample.
For example, for each target recommended content exposed to the target object, the server may collect behavior of the target object triggered for the target recommended content as training annotation behavior, where the collected behavior may include, for example, no click, click to view details, purchase, and so forth. Correspondingly, the server can construct an optimized training sample according to the basic attribute information and the historical behavior information of the target object, the target recommended content and the behavior (namely training labeling behavior) of the target object triggered by the target recommended content aiming at each target recommended content, and store the optimized training sample in a related file system. Furthermore, the server may retrieve the optimized training samples generated in the corresponding model training period from the file system according to the predicted model training period (such as one day, one week, one month, etc.), and perform optimized training on the parameter prediction model for predicting the target parameter by using the optimized training samples, so as to continuously improve the model performance of the parameter prediction model, and ensure that the related content recommendation service has a good recommendation effect.
According to the data processing method, the fact that the preferences of different objects for different recommended contents are different is considered, so that when the behavior characteristics of the target object are built based on the historical behavior information of the target object, the basic attribute characteristics of the basic attribute information of the target object and the target content characteristics of the target recommended contents are creatively utilized, the attention weighting processing is carried out on the historical content characteristics of each of the historical recommended contents in the historical behavior information, and personalized behavior characteristics of the target object are built. On one hand, the personalized behavior feature is constructed by referring to the basic attribute feature of the basic attribute information of the target object, so that the personalized behavior feature can reflect the personal difference of the target object, and on the other hand, the personalized behavior feature is constructed by referring to the target content feature of the target recommended content, so that the personalized behavior feature can reflect the preference of the target object to the target recommended content. Further, based on the personalized behavior characteristics obtained by construction, the target parameters used for representing the interested degree of the target object to the target recommended content are predicted, so that the predicted target parameters can be ensured to have higher accuracy, and the recommendation effect of the recommendation service can be improved by providing the related recommendation service for the target object based on the target parameters.
In order to further understand the data processing method provided in the embodiment of the present application, the data processing method is taken as an example for supporting the content recommendation service provided by the game application, and the overall exemplary description of the data processing method is described below with reference to the software system architecture diagram shown in fig. 7.
As shown in fig. 7, in practical application, the target object may trigger a content recommendation request through a front-end presentation page in the game application, for example, if the terminal device detects that the target object opens a game map download page in the game application, a content recommendation request for requesting the server to recommend a game map may be generated accordingly, and for example, if the terminal device detects that the target object opens a skin purchase page in the game application, a content recommendation request for requesting the server to recommend virtual character skin may be generated accordingly. Specifically, the terminal device may send a content recommendation request to the server via a hypertext transfer protocol (Hyper Text Transfer Protocol, http) to request access to the ordering service deployed in a k8s (Kubernetes) container, where http is a communication protocol supporting communication between the terminal device and the server, where k8s is an open source platform capable of automatically implementing Linux container operations, the container is a virtualized form of an operating system, and typically, a container may be used to run all content from a small micro-service or software process to a large application, where the container may include all necessary executable files, binary codes, libraries, and configuration files.
After receiving the content recommendation request, the ranking service deployed in the k8s (Kubernetes) container may obtain, in response to the content recommendation request, the basic attribute information and the historical behavior information of the target object, the target context information, and the target recommended content to be recommended. For example, assuming that the received content recommendation request is a content recommendation request for requesting the server to recommend a game map, the obtained basic attribute information of the target object may include basic information reserved when the target object registers the game application, login time information of the target object for the game application, a use duration, etc., the obtained historical behavior information may include respective related information (such as names, categories, authors, etc. of the game map) of a plurality of game maps used by the target object, the obtained target context information may include a time when the target object this time triggers the content recommendation request, and the obtained target recommendation content may include related information (such as names, categories, authors, etc. of the target game map) of the target game map to be recommended. For another example, assuming that the received content recommendation request is a content recommendation request for requesting the server to recommend virtual character skin, the obtained basic attribute information of the target object may include basic information reserved when the target object registers the game application, login time information of the target object for the game application, a use duration, a level of the target object, etc., the obtained historical behavior information may include related information of multiple types of skin used by the target object (such as a category of virtual character to which the target object belongs, a skin level, a skin price, etc.), the obtained target context information may include a time when the target object this time triggers the content recommendation request, and the obtained target recommendation content may include related information of target skin to be recommended (such as a category of virtual character to which the target object belongs, a skin level, a skin price, etc.).
Further, the ranking service may request a model service deployed in another k8s (Kubernetes) container, scoring each target recommendation content based on its acquired base attribute information and historical behavior information of the target object, target context information, and each target recommendation content to be recommended. The model service scores each target recommended content, namely, the model service essentially predicts target parameters (such as click rate or purchase rate) for representing the interest degree of the target object in each target recommended content; when the model service specifically works, feature coding structures in a pre-trained parameter prediction model can be utilized to respectively perform feature coding processing on basic attribute information, each historical recommended content, target context information and target recommended content of a target object to obtain corresponding basic attribute features, each historical recommended content feature, each historical content feature, each target context feature and target content feature, then, attention weighting processing can be performed on each historical content feature of each historical recommended content by utilizing PTA Unit in the parameter prediction model according to the basic attribute features and the target content features to obtain personalized behavior features for representing the preference condition of the historical interaction of the target object on the target recommended content, further, the basic attribute features, the target content features, the target context features and the personalized behavior features can be subjected to carrying out and leveling operation to obtain fusion features, the fusion features are further input into a full-connection layer of a multi-layer stacked band prelu activation function, and feature id functions obtained after the processing of the multi-layer full-connection layer are further, and then, the personalized behavior features for representing the preference condition of the target recommended content of the target object are obtained through the moid function prediction.
Furthermore, the model service may transmit the target parameters corresponding to the predicted target recommended contents to the ranking service, and the ranking service may rank the target recommended contents according to the target parameters corresponding to the target recommended contents, so as to determine a recommendation display sequence of the target recommended contents, and control to recommend the target recommended contents to the target object based on the recommendation display sequence. For example, assuming that the content recommendation request is a content recommendation request for requesting the server to recommend the game map, the ranking service may determine a recommendation presentation order corresponding to each target game map according to a target parameter corresponding to each target game map to be recommended, and recommend each target game map to the target object accordingly. For another example, assuming that the content recommendation request is a content recommendation request for requesting the server to recommend virtual character skins, the ranking service may determine a recommendation presentation order corresponding to each target skin according to a target parameter corresponding to each target skin to be recommended, and recommend each target skin to the target object according to the recommendation presentation order.
In the embodiment of the application, the sequencing service can report the acquired basic attribute information and historical behavior information of the target object, the target context information and the target recommended content of the target object to the sample spelling service deployed in another k8s container, the front-end display page can report the behavior triggered by the target object aiming at each target recommended content recommended for the target object to the sample spelling service, and correspondingly, the sample spelling service can construct an optimized training sample according to the sequencing service and the information reported by the front-end display page and store the constructed optimized training sample in a related file system. Furthermore, according to a preset model training period, the parameter prediction model used by the model service can be trained offline by using the optimized training samples stored in the file system, and the parameter prediction model used by the model service is updated by using the trained parameter prediction model, so that the model service can continuously provide high-quality parameter prediction service.
In order to verify the effectiveness and reliability of the method provided by the embodiment of the application, the inventor performs an offline experiment based on a prop recommendation service of a certain game application program, and the offline experiment uses prop exposure data and prop purchase data of relevant players of the game application program in a certain period of time to train a parameter prediction model in the embodiment of the application. And comparing the result of AUC (Area Under Curve) of the parameter prediction model provided by the embodiment of the application with the result of several models commonly seen at present, the AUC result is the area under the working characteristic curve (receiver operating characteristic curve, ROC) of the subject, the ROC is a curve drawn according to a series of different classification modes (demarcation values or decision thresholds) by taking the true positive rate (sensitivity) as the ordinate and the false positive rate (specificity) as the abscissa, and the AUC value can reflect the effect of the classifier, and in general, the larger the AUC value is, the higher the effect of the classifier is. The specific experimental results are shown in table 1 below:
TABLE 1
DNN DeepFM Wide&Deep DIN PDIN
AUC 0.928 0.93 0.92 0.93 0.93
Loss 0.19 0.19 0.19 0.18 0.16
Wherein DNN, deepFM, wide & Deep and DIN are models of how interesting several predictive objects are for specific content, and PDIN is a parametric predictive model in the embodiments of the application. From the experimental results shown in table 1, it can be found that the parametric prediction model in the embodiments of the present application is substantially equal in AUC but improved in Loss with respect to several existing models.
The application also provides a corresponding data processing device for the data processing method, so that the data processing method can be practically applied and realized.
Referring to fig. 8, fig. 8 is a schematic diagram of a data processing apparatus 800 corresponding to the data processing method shown in fig. 2 above. As shown in fig. 8, the data processing apparatus 800 includes:
The information acquisition module 801 is configured to acquire basic attribute information and historical behavior information of a target object, and acquire target recommended content to be recommended;
the feature encoding module 802 is configured to perform feature encoding processing on the basic attribute information, the plurality of historical recommended contents, and the target recommended content, respectively, to obtain basic attribute features of the basic attribute information, historical content features of the plurality of historical recommended contents, and target content features of the target recommended content;
The attention weighting module 803 is configured to perform attention weighting processing on the historical content features of each of the plurality of historical recommended content according to the basic attribute features and the target content features, so as to obtain personalized behavior features;
The parameter determining module 804 is configured to determine a target parameter corresponding to the target recommended content according to the personalized behavior feature, the basic attribute feature, and the target content feature, where the target parameter is used to characterize the interest degree of the target object in the target recommended content.
Optionally, the attention weighting module 803 includes:
A content feature determination submodule for determining personalized content features according to the basic attribute features and the target content features;
The weight determining sub-module is used for determining the attention weight of each of the plurality of historical recommended contents according to the historical content characteristics and the personalized content characteristics of each of the plurality of historical recommended contents;
And the behavior characteristic determining submodule is used for determining the personalized behavior characteristic according to the historical content characteristics and the attention weight of each of the plurality of historical recommended contents.
Optionally, the content feature determining submodule is specifically configured to:
performing feature fusion processing on the basic attribute features and the target content features to obtain first fusion features;
processing the first fusion feature through a first attention network to obtain a first personalized fusion feature;
And determining the personalized content characteristics according to the target content characteristics and the first personalized fusion characteristics.
Optionally, the content feature determining submodule is specifically configured to:
performing feature conversion processing on the basic attribute features to obtain reference attribute features;
Performing feature fusion processing on the basic attribute feature, the target content feature and the product of the reference attribute feature and the target content feature to obtain a second fusion feature;
processing the second fusion feature through a second attention network to obtain a second personalized fusion feature;
And determining the personalized content characteristics according to the target content characteristics and the second personalized fusion characteristics.
Optionally, the weight determining submodule is specifically configured to:
For each history recommended content, determining a reference weight of the history recommended content according to the personalized content characteristics and the history content characteristics of the history recommended content;
Determining total reference weights according to the reference weights of the historical recommended contents;
For each of the historical recommended content, determining an attention weight of the historical recommended content according to the reference weight of the historical recommended content and the total reference weight.
Optionally, the weight determining submodule is specifically configured to:
Performing feature fusion processing on the personalized content features, the historical content features, differences between the personalized content features and the historical content features and products of the personalized content features and the historical content features to obtain third fusion features;
and processing the third fusion characteristic through a third attention network to obtain the reference weight of the history recommended content.
Optionally, the behavior feature determining submodule is specifically configured to:
And carrying out weighted summation processing on the historical content characteristics of each of the plurality of historical recommended contents based on the attention weights of each of the plurality of historical recommended contents to obtain the personalized behavior characteristics.
Optionally, the plurality of historical recommended contents in the historical behavior information are based on a time arrangement of the target object for triggering the interactive operation, a historical content characteristic of a first historical recommended content in the historical behavior information is determined according to the first historical recommended content, a historical content characteristic of each historical recommended content except the first historical recommended content in the historical behavior information is determined according to the historical recommended content and a previous historical recommended content adjacent to the historical recommended content in the historical behavior information, and the behavior characteristic determining submodule is specifically configured to:
determining a reference personalized feature of the first historical recommended content according to the historical content feature and the attention weight of the first historical recommended content aiming at the first historical recommended content;
For each history recommended content except the first history recommended content, determining a reference personalized feature of the history recommended content according to the history content feature and the attention weight of the history recommended content and the reference personalized feature of the previous history recommended content adjacent to the history recommended content in the history behavior information;
and determining a reference personalized feature of the last historical recommended content in the historical behavior information as the personalized behavior feature.
Optionally, the historical behavior information includes a plurality of historical behavior sub-information, the plurality of historical behavior sub-information respectively corresponds to different interaction operations, each of the historical behavior sub-information includes a plurality of historical recommendation contents that the target object triggered the corresponding interaction operation, and the attention weighting module 803 is specifically configured to:
For each piece of history behavior sub-information, performing attention weighting processing on each history content characteristic of a plurality of history recommended contents included in the history behavior sub-information according to the basic attribute characteristic and the target content characteristic to obtain a personalized behavior characteristic corresponding to the history behavior sub-information;
the parameter determining module 804 is specifically configured to:
and determining the target parameters according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics which are respectively corresponding to the historical behavior sub-information.
Optionally, the information obtaining module 801 is further configured to:
the method comprises the steps of acquiring target context information, wherein the target context information comprises at least one of time information of a target object requesting a content recommendation service, operation information generated by the target object in a first period, activity information participated by the target object in a second period and activity information initiated by a network platform for providing the content recommendation service in a third period;
the feature encoding module 802 is further configured to:
performing feature coding processing on the target context information to obtain target context features;
the parameter determining module 804 is specifically configured to:
And determining the target parameters according to the personalized behavior characteristics, the basic attribute characteristics, the target content characteristics and the target context characteristics.
Optionally, the apparatus further includes:
And the content recommendation module is used for sequencing the plurality of target recommended contents according to the target parameters corresponding to the target recommended contents when the target recommended contents comprise a plurality of target recommended contents, determining the recommendation display sequence of the plurality of target recommended contents, and recommending the plurality of target recommended contents to the target object based on the recommendation display sequence.
Optionally, the target parameter is determined by a parameter prediction model, and the device further comprises:
the sample acquisition module is used for acquiring the behavior of the target object on the target recommended content aiming at each target recommended content as a training labeling behavior; constructing an optimized training sample according to the basic attribute information, the historical behavior information, the target recommended content and the training annotation behavior;
and the model training module is used for carrying out optimization training on the parameter prediction model based on the optimization training sample.
The data processing device considers that different objects have different preferences for different recommended contents, so that when the behavior characteristics of the target object are constructed based on the historical behavior information of the target object, the basic attribute characteristics of the basic attribute information of the target object and the target content characteristics of the target recommended contents are innovatively utilized, and the respective historical content characteristics of each historical recommended content in the historical behavior information are subjected to attention weighting processing, so that the personalized behavior characteristics of the target object are constructed. On one hand, the personalized behavior feature is constructed by referring to the basic attribute feature of the basic attribute information of the target object, so that the personalized behavior feature can reflect the personal difference of the target object, and on the other hand, the personalized behavior feature is constructed by referring to the target content feature of the target recommended content, so that the personalized behavior feature can reflect the preference of the target object to the target recommended content. Further, based on the personalized behavior characteristics obtained by construction, the target parameters used for representing the interested degree of the target object to the target recommended content are predicted, so that the predicted target parameters can be ensured to have higher accuracy, and the recommendation effect of the recommendation service can be improved by providing the related recommendation service for the target object based on the target parameters.
The embodiment of the application also provides a computer device, which can be a terminal device or a server, and the terminal device and the server provided by the embodiment of the application are introduced from the aspect of hardware materialization.
Referring to fig. 9, fig. 9 is a schematic structural diagram of a terminal device according to an embodiment of the present application. As shown in fig. 9, for convenience of explanation, only the portions related to the embodiments of the present application are shown, and specific technical details are not disclosed, please refer to the method portions of the embodiments of the present application. The terminal may be any terminal device including a mobile phone, a tablet computer, a Personal digital assistant (Personal DIGITAL ASSISTANT, PDA), a Point of Sales (POS), a vehicle-mounted computer, and the like, and the terminal is taken as a computer as an example:
Fig. 9 is a block diagram showing a part of the structure of a computer related to a terminal provided by an embodiment of the present application. Referring to fig. 9, the computer includes Radio Frequency (RF) circuitry 910, memory 920, input unit 930 (including touch panel 931 and other input devices 932), display unit 940 (including display panel 941), sensor 950, audio circuitry 960 (which may connect speaker 961 and microphone 962), wireless fidelity (WIRELESS FIDELITY, wiFi) module 970, processor 980, and power source 990. Those skilled in the art will appreciate that the computer architecture shown in fig. 9 is not limiting and that more or fewer components than shown may be included, or that certain components may be combined, or that different arrangements of components may be provided.
The memory 920 may be used to store software programs and modules that the processor 980 performs various functional applications and data processing by operating on the software programs and modules stored in the memory 920. The memory 920 may mainly include a storage program area which may store an operating system, an application program required for at least one function (such as a sound playing function, an image playing function, etc.), etc., and a storage data area which may store data created according to the use of a computer (such as audio data, a phonebook, etc.), etc. In addition, memory 920 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device.
Processor 980 is a control center of the computer, connecting various portions of the overall computer using various interfaces and lines, performing various functions of the computer and processing data by running or executing software programs and/or modules stored in memory 920, and invoking data stored in memory 920. Optionally, the processor 980 may include one or more processing elements, and preferably the processor 980 may integrate an application processor with a modem processor, wherein the application processor primarily handles operating systems, user interfaces, application programs, and the like, and the modem processor primarily handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 980.
In the embodiment of the present application, the processor 980 included in the terminal is further configured to perform the steps of any implementation manner of the data processing method provided in the embodiment of the present application.
Referring to fig. 10, fig. 10 is a schematic structural diagram of a server 1000 according to an embodiment of the present application. The server 1000 may vary considerably in configuration or performance and may include one or more central processing units (central processing units, CPUs) 1022 (e.g., one or more processors) and memory 1032, one or more storage mediums 1030 (e.g., one or more mass storage devices) storing applications 1042 or data 1044. Wherein memory 1032 and storage medium 1030 may be transitory or persistent. The program stored on the storage medium 1030 may include one or more modules (not shown), each of which may include a series of instruction operations on a server. Further, central processor 1022 may be configured to communicate with storage medium 1030 to perform a series of instruction operations in storage medium 1030 on server 1000.
The Server 1000 may also include one or more power supplies 1026, one or more wired or wireless network interfaces 1050, one or more input/output interfaces 1058, and/or one or more operating systems, such as Windows Server TM,Mac OS XTM,UnixTM,LinuxTM,FreeBSDTM, or the like.
The steps performed by the server in the above embodiments may be based on the server structure shown in fig. 10.
The CPU 1022 may also be used to perform steps of any implementation of the data processing method provided in an embodiment of the present application.
The embodiments of the present application also provide a computer readable storage medium storing a computer program for executing any one of the data processing methods described in the foregoing embodiments.
Embodiments of the present application also provide a computer program product or computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform any one of the data processing methods described in the foregoing respective embodiments.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application. The storage medium includes a U disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, an optical disk, or other various media capable of storing a computer program.
It should be understood that in the present application, "at least one (item)" means one or more, and "a plurality" means two or more. "and/or" is used to describe an association relationship of an associated object, and indicates that three relationships may exist, for example, "a and/or B" may indicate that only a exists, only B exists, and three cases of a and B exist simultaneously, where a and B may be singular or plural. The character "/" generally indicates that the context-dependent object is an "or" relationship. "at least one of" or the like means any combination of these items, including any combination of single item(s) or plural items(s). For example, at least one of a, b or c may represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.
While the application has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that the foregoing embodiments may be modified or equivalents may be substituted for some of the features thereof, and that the modifications or substitutions do not depart from the spirit and scope of the embodiments of the application.

Claims (25)

1.一种数据处理方法，其特征在于，所述方法包括：1. A data processing method, characterized in that the method comprises: 获取目标对象的基础属性信息和历史行为信息，获取待推荐的目标推荐内容；所述历史行为信息包括所述目标对象触发过交互操作的多个历史推荐内容；Obtaining basic attribute information and historical behavior information of the target object, and obtaining target recommended content to be recommended; the historical behavior information includes a plurality of historical recommended content on which the target object has triggered interactive operations; 对所述基础属性信息、所述多个历史推荐内容和所述目标推荐内容分别进行特征编码处理，得到所述基础属性信息的基础属性特征、所述多个历史推荐内容各自的历史内容特征、以及所述目标推荐内容的目标内容特征；Performing feature coding processing on the basic attribute information, the multiple historical recommended contents, and the target recommended content respectively, to obtain basic attribute features of the basic attribute information, historical content features of each of the multiple historical recommended contents, and target content features of the target recommended content; 根据所述基础属性特征和所述目标内容特征，对所述多个历史推荐内容各自的历史内容特征进行注意力加权处理，得到个性化行为特征，具体包括：根据所述基础属性特征和所述目标内容特征，确定个性化内容特征；根据所述多个历史推荐内容各自的历史内容特征和所述个性化内容特征，确定所述多个历史推荐内容各自的注意力权重，所述注意力权重用于体现对应的历史内容特征对于反映目标对象偏好情况的重要程度；根据所述多个历史推荐内容各自的历史内容特征和注意力权重，确定所述个性化行为特征；According to the basic attribute features and the target content features, attention weighting processing is performed on the historical content features of each of the multiple historical recommended contents to obtain personalized behavior features, specifically including: determining personalized content features according to the basic attribute features and the target content features; determining attention weights of each of the multiple historical recommended contents according to the historical content features of each of the multiple historical recommended contents and the personalized content features, the attention weights being used to reflect the importance of the corresponding historical content features for reflecting the preferences of the target object; determining the personalized behavior features according to the historical content features and the attention weights of each of the multiple historical recommended contents; 根据所述个性化行为特征、所述基础属性特征和所述目标内容特征，确定所述目标推荐内容对应的目标参数；所述目标参数用于表征所述目标对象对于所述目标推荐内容的感兴趣程度。According to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics, a target parameter corresponding to the target recommended content is determined; the target parameter is used to characterize the degree of interest of the target object in the target recommended content. 2.根据权利要求1所述的方法，其特征在于，所述根据所述基础属性特征和所述目标内容特征，确定个性化内容特征，包括：2. The method according to claim 1, characterized in that the step of determining the personalized content feature according to the basic attribute feature and the target content feature comprises: 对所述基础属性特征和所述目标内容特征进行特征融合处理，得到第一融合特征；Performing feature fusion processing on the basic attribute feature and the target content feature to obtain a first fusion feature; 通过第一注意力网络处理所述第一融合特征，得到第一个性化融合特征；Processing the first fused feature through a first attention network to obtain a first personalized fused feature; 根据所述目标内容特征和所述第一个性化融合特征，确定所述个性化内容特征。The personalized content feature is determined according to the target content feature and the first personalized fusion feature. 3.根据权利要求1所述的方法，其特征在于，所述根据所述基础属性特征和所述目标内容特征，确定个性化内容特征，包括：3. The method according to claim 1, characterized in that the step of determining the personalized content feature according to the basic attribute feature and the target content feature comprises: 对所述基础属性特征进行特征转换处理，得到参考属性特征；Performing feature conversion processing on the basic attribute features to obtain reference attribute features; 对所述基础属性特征、所述目标内容特征、以及所述参考属性特征与所述目标内容特征的乘积进行特征融合处理，得到第二融合特征；Performing feature fusion processing on the basic attribute feature, the target content feature, and the product of the reference attribute feature and the target content feature to obtain a second fused feature; 通过第二注意力网络处理所述第二融合特征，得到第二个性化融合特征；Processing the second fused feature through a second attention network to obtain a second personalized fused feature; 根据所述目标内容特征和所述第二个性化融合特征，确定所述个性化内容特征。The personalized content feature is determined according to the target content feature and the second personalized fusion feature. 4.根据权利要求1所述的方法，其特征在于，所述根据所述多个历史推荐内容各自的历史内容特征和所述个性化内容特征，确定所述多个历史推荐内容各自的注意力权重，包括：4. The method according to claim 1, characterized in that the step of determining the attention weight of each of the plurality of historical recommended contents according to the respective historical content features of the plurality of historical recommended contents and the personalized content features comprises: 针对每个所述历史推荐内容，根据所述个性化内容特征和所述历史推荐内容的历史内容特征，确定所述历史推荐内容的参考权重；For each of the historical recommended contents, determining a reference weight of the historical recommended content according to the personalized content feature and the historical content feature of the historical recommended content; 根据所述多个历史推荐内容各自的参考权重，确定总参考权重；Determining a total reference weight according to the respective reference weights of the plurality of historical recommended contents; 针对每个所述历史推荐内容，根据所述历史推荐内容的参考权重和所述总参考权重，确定所述历史推荐内容的注意力权重。For each of the historical recommended contents, the attention weight of the historical recommended content is determined according to the reference weight of the historical recommended content and the total reference weight. 5.根据权利要求4所述的方法，其特征在于，所述根据所述个性化内容特征和所述历史推荐内容的历史内容特征，确定所述历史推荐内容的参考权重，包括：5. The method according to claim 4, characterized in that the step of determining the reference weight of the historically recommended content according to the personalized content feature and the historical content feature of the historically recommended content comprises: 对所述个性化内容特征、所述历史内容特征、所述个性化内容特征与所述历史内容特征的差值、以及所述个性化内容特征与所述历史内容特征的乘积进行特征融合处理，得到第三融合特征；Performing feature fusion processing on the personalized content feature, the historical content feature, the difference between the personalized content feature and the historical content feature, and the product of the personalized content feature and the historical content feature to obtain a third fusion feature; 通过第三注意力网络处理所述第三融合特征，得到所述历史推荐内容的参考权重。The third fusion feature is processed by a third attention network to obtain a reference weight of the historical recommendation content. 6.根据权利要求1所述的方法，其特征在于，所述根据所述多个历史推荐内容各自的历史内容特征和注意力权重，确定所述个性化行为特征，包括：6. The method according to claim 1, characterized in that the step of determining the personalized behavior feature according to the historical content features and attention weights of the plurality of historical recommended contents comprises: 基于所述多个历史推荐内容各自的注意力权重，对所述多个历史推荐内容各自的历史内容特征进行加权求和处理，得到所述个性化行为特征。Based on the attention weights of the plurality of historical recommended contents, weighted sum processing is performed on the historical content features of the plurality of historical recommended contents to obtain the personalized behavior features. 7.根据权利要求1所述的方法，其特征在于，所述历史行为信息中所述多个历史推荐内容基于所述目标对象对其触发交互操作的时间排列；所述历史行为信息中第一个历史推荐内容的历史内容特征，是根据所述第一个历史推荐内容确定的；所述历史行为信息中除所述第一个历史推荐内容外的每个历史推荐内容的历史内容特征，是根据所述历史推荐内容和所述历史行为信息中与所述历史推荐内容相邻的前一个历史推荐内容确定的；7. The method according to claim 1 is characterized in that the multiple historical recommended contents in the historical behavior information are arranged based on the time when the target object triggers an interactive operation thereon; the historical content feature of the first historical recommended content in the historical behavior information is determined based on the first historical recommended content; the historical content feature of each historical recommended content in the historical behavior information except the first historical recommended content is determined based on the historical recommended content and the previous historical recommended content adjacent to the historical recommended content in the historical behavior information; 所述根据所述多个历史推荐内容各自的历史内容特征和注意力权重，确定所述个性化行为特征，包括：The determining the personalized behavior feature according to the historical content features and attention weights of the plurality of historical recommended contents includes: 针对所述第一个历史推荐内容，根据所述第一个历史推荐内容的历史内容特征和注意力权重，确定所述第一个历史推荐内容的参考个性化特征；For the first historical recommended content, determining a reference personalized feature of the first historical recommended content according to the historical content feature and the attention weight of the first historical recommended content; 针对除所述第一个历史推荐内容外的每个历史推荐内容，根据所述历史推荐内容的历史内容特征和注意力权重、以及所述历史行为信息中与所述历史推荐内容相邻的前一个历史推荐内容的参考个性化特征，确定所述历史推荐内容的参考个性化特征；For each historical recommended content except the first historical recommended content, determining a reference personalized feature of the historical recommended content according to the historical content feature and attention weight of the historical recommended content and the reference personalized feature of a previous historical recommended content adjacent to the historical recommended content in the historical behavior information; 确定所述历史行为信息中最后一个历史推荐内容的参考个性化特征，作为所述个性化行为特征。A reference personalized feature of the last historical recommended content in the historical behavior information is determined as the personalized behavior feature. 8.根据权利要求1所述的方法，其特征在于，所述历史行为信息中包括多个历史行为子信息，所述多个历史行为子信息分别对应不同的交互操作，每个所述历史行为子信息包括所述目标对象触发过其对应的交互操作的多个历史推荐内容；8. The method according to claim 1, characterized in that the historical behavior information includes a plurality of historical behavior sub-information, the plurality of historical behavior sub-information respectively corresponding to different interactive operations, and each of the historical behavior sub-information includes a plurality of historical recommended contents of the corresponding interactive operation triggered by the target object; 所述根据所述基础属性特征和所述目标内容特征，对所述多个历史推荐内容各自的历史内容特征进行注意力加权处理，得到个性化行为特征，包括：The step of performing attention weighting processing on the historical content features of each of the plurality of historical recommended contents according to the basic attribute features and the target content features to obtain personalized behavior features includes: 针对每个所述历史行为子信息，根据所述基础属性特征和所述目标内容特征，对所述历史行为子信息包括的多个历史推荐内容各自的历史内容特征进行注意力加权处理，得到所述历史行为子信息对应的个性化行为特征；For each of the historical behavior sub-information, according to the basic attribute feature and the target content feature, attention weighted processing is performed on the historical content features of the multiple historical recommended contents included in the historical behavior sub-information to obtain the personalized behavior feature corresponding to the historical behavior sub-information; 所述根据所述个性化行为特征、所述基础属性特征和所述目标内容特征，确定所述目标推荐内容对应的目标参数，包括：The determining, according to the personalized behavior feature, the basic attribute feature and the target content feature, a target parameter corresponding to the target recommended content includes: 根据所述多个历史行为子信息各自对应的个性化行为特征、所述基础属性特征和所述目标内容特征，确定所述目标参数。The target parameter is determined according to the personalized behavior characteristics corresponding to each of the plurality of historical behavior sub-information, the basic attribute characteristics and the target content characteristics. 9.根据权利要求1所述的方法，其特征在于，所述方法还包括：9. The method according to claim 1, characterized in that the method further comprises: 获取目标上下文信息；所述目标上下文信息包括以下至少一种信息：所述目标对象请求本次内容推荐服务的时间信息、所述目标对象在第一时段内产生的操作信息、所述目标对象在第二时段内参与的活动信息、用于提供所述内容推荐服务的网络平台在第三时段内发起的活动信息；Acquire target context information; the target context information includes at least one of the following information: time information when the target object requests the current content recommendation service, operation information generated by the target object in a first time period, activity information in which the target object participates in a second time period, and activity information initiated by a network platform for providing the content recommendation service in a third time period; 对所述目标上下文信息进行特征编码处理，得到目标上下文特征；Performing feature encoding processing on the target context information to obtain target context features; 所述根据所述个性化行为特征、所述基础属性特征和所述目标内容特征，确定所述目标推荐内容对应的目标参数，包括：The determining, according to the personalized behavior feature, the basic attribute feature and the target content feature, a target parameter corresponding to the target recommended content includes: 根据所述个性化行为特征、所述基础属性特征、所述目标内容特征以及所述目标上下文特征，确定所述目标参数。The target parameter is determined according to the personalized behavior feature, the basic attribute feature, the target content feature and the target context feature. 10.根据权利要求1所述的方法，其特征在于，所述方法还包括：10. The method according to claim 1, characterized in that the method further comprises: 当所述目标推荐内容包括多个时，根据多个所述目标推荐内容各自对应的目标参数，对多个所述目标推荐内容排序，确定多个所述目标推荐内容的推荐展示顺序；When the target recommended content includes multiple ones, sorting the multiple target recommended contents according to the target parameters corresponding to the multiple target recommended contents respectively, and determining the recommendation display order of the multiple target recommended contents; 基于所述推荐展示顺序，向所述目标对象推荐多个所述目标推荐内容。Based on the recommendation display order, a plurality of the target recommended contents are recommended to the target object. 11.根据权利要求10所述的方法，其特征在于，所述目标参数是通过参数预测模型确定的；所述方法还包括：11. The method according to claim 10, characterized in that the target parameter is determined by a parameter prediction model; the method further comprises: 针对每个所述目标推荐内容，获取所述目标对象对所述目标推荐内容的行为，作为训练标注行为；根据所述基础属性信息、所述历史行为信息、所述目标推荐内容和所述训练标注行为，构建优化训练样本；For each of the target recommended contents, obtaining the behavior of the target object on the target recommended content as a training annotated behavior; constructing an optimized training sample according to the basic attribute information, the historical behavior information, the target recommended content and the training annotated behavior; 基于所述优化训练样本，对所述参数预测模型进行优化训练。Based on the optimized training samples, the parameter prediction model is optimized and trained. 12.一种数据处理装置，其特征在于，所述装置包括：12. A data processing device, characterized in that the device comprises: 信息获取模块，用于获取目标对象的基础属性信息和历史行为信息，获取待推荐的目标推荐内容；所述历史行为信息包括所述目标对象触发过交互操作的多个历史推荐内容；An information acquisition module is used to acquire basic attribute information and historical behavior information of a target object, and acquire target recommended content to be recommended; the historical behavior information includes a plurality of historical recommended content on which the target object has triggered interactive operations; 特征编码模块，用于对所述基础属性信息、所述多个历史推荐内容和所述目标推荐内容分别进行特征编码处理，得到所述基础属性信息的基础属性特征、所述多个历史推荐内容各自的历史内容特征、以及所述目标推荐内容的目标内容特征；a feature encoding module, configured to perform feature encoding processing on the basic attribute information, the plurality of historical recommended contents, and the target recommended content, respectively, to obtain basic attribute features of the basic attribute information, historical content features of each of the plurality of historical recommended contents, and target content features of the target recommended content; 注意力加权模块，用于根据所述基础属性特征和所述目标内容特征，对所述多个历史推荐内容各自的历史内容特征进行注意力加权处理，得到个性化行为特征；An attention weighting module, configured to perform attention weighting processing on the historical content features of each of the plurality of historical recommended contents according to the basic attribute features and the target content features, so as to obtain personalized behavior features; 参数确定模块，用于根据所述个性化行为特征、所述基础属性特征和所述目标内容特征，确定所述目标推荐内容对应的目标参数；所述目标参数用于表征所述目标对象对于所述目标推荐内容的感兴趣程度；A parameter determination module, used to determine a target parameter corresponding to the target recommended content according to the personalized behavior characteristics, the basic attribute characteristics and the target content characteristics; the target parameter is used to characterize the degree of interest of the target object in the target recommended content; 所述注意力加权模块包括：The attention weighting module comprises: 内容特征确定子模块，用于根据所述基础属性特征和所述目标内容特征，确定个性化内容特征；A content feature determination submodule, configured to determine personalized content features according to the basic attribute features and the target content features; 权重确定子模块，用于根据所述多个历史推荐内容各自的历史内容特征和所述个性化内容特征，确定所述多个历史推荐内容各自的注意力权重，所述注意力权重用于体现对应的历史内容特征对于反映目标对象偏好情况的重要程度；A weight determination submodule, used to determine the attention weight of each of the plurality of historical recommended contents according to the respective historical content features of the plurality of historical recommended contents and the personalized content features, wherein the attention weight is used to reflect the importance of the corresponding historical content features in reflecting the preference of the target object; 行为特征确定子模块，用于根据所述多个历史推荐内容各自的历史内容特征和注意力权重，确定所述个性化行为特征。The behavior feature determination submodule is used to determine the personalized behavior feature according to the historical content features and attention weights of each of the multiple historical recommended contents. 13.根据权利要求12所述的装置，其特征在于，所述内容特征确定子模块具体用于：13. The device according to claim 12, wherein the content feature determination submodule is specifically used for: 对所述基础属性特征和所述目标内容特征进行特征融合处理，得到第一融合特征；Performing feature fusion processing on the basic attribute feature and the target content feature to obtain a first fusion feature; 通过第一注意力网络处理所述第一融合特征，得到第一个性化融合特征；Processing the first fused feature through a first attention network to obtain a first personalized fused feature; 根据所述目标内容特征和所述第一个性化融合特征，确定所述个性化内容特征。The personalized content feature is determined according to the target content feature and the first personalized fusion feature. 14.根据权利要求12所述的装置，其特征在于，所述内容特征确定子模块具体用于：14. The device according to claim 12, wherein the content feature determination submodule is specifically used for: 对所述基础属性特征进行特征转换处理，得到参考属性特征；Performing feature conversion processing on the basic attribute features to obtain reference attribute features; 对所述基础属性特征、所述目标内容特征、以及所述参考属性特征与所述目标内容特征的乘积进行特征融合处理，得到第二融合特征；Performing feature fusion processing on the basic attribute feature, the target content feature, and the product of the reference attribute feature and the target content feature to obtain a second fused feature; 通过第二注意力网络处理所述第二融合特征，得到第二个性化融合特征；Processing the second fused feature through a second attention network to obtain a second personalized fused feature; 根据所述目标内容特征和所述第二个性化融合特征，确定所述个性化内容特征。The personalized content feature is determined according to the target content feature and the second personalized fusion feature. 15.根据权利要求12所述的装置，其特征在于，所述权重确定子模块具体用于：15. The device according to claim 12, characterized in that the weight determination submodule is specifically used for: 针对每个所述历史推荐内容，根据所述个性化内容特征和所述历史推荐内容的历史内容特征，确定所述历史推荐内容的参考权重；For each of the historical recommended contents, determining a reference weight of the historical recommended content according to the personalized content feature and the historical content feature of the historical recommended content; 根据所述多个历史推荐内容各自的参考权重，确定总参考权重；Determining a total reference weight according to the respective reference weights of the plurality of historical recommended contents; 针对每个所述历史推荐内容，根据所述历史推荐内容的参考权重和所述总参考权重，确定所述历史推荐内容的注意力权重。For each of the historical recommended contents, the attention weight of the historical recommended content is determined according to the reference weight of the historical recommended content and the total reference weight. 16.根据权利要求15所述的装置，其特征在于，所述权重确定子模块具体用于：16. The device according to claim 15, characterized in that the weight determination submodule is specifically used for: 对所述个性化内容特征、所述历史内容特征、所述个性化内容特征与所述历史内容特征的差值、以及所述个性化内容特征与所述历史内容特征的乘积进行特征融合处理，得到第三融合特征；Performing feature fusion processing on the personalized content feature, the historical content feature, the difference between the personalized content feature and the historical content feature, and the product of the personalized content feature and the historical content feature to obtain a third fusion feature; 通过第三注意力网络处理所述第三融合特征，得到所述历史推荐内容的参考权重。The third fusion feature is processed by a third attention network to obtain a reference weight of the historical recommendation content. 17.根据权利要求12所述的装置，其特征在于，所述行为特征确定子模块具体用于：17. The device according to claim 12, characterized in that the behavior feature determination submodule is specifically used for: 基于所述多个历史推荐内容各自的注意力权重，对所述多个历史推荐内容各自的历史内容特征进行加权求和处理，得到所述个性化行为特征。Based on the attention weights of the plurality of historical recommended contents, weighted sum processing is performed on the historical content features of the plurality of historical recommended contents to obtain the personalized behavior features. 18.根据权利要求12所述的装置，其特征在于，所述历史行为信息中所述多个历史推荐内容基于所述目标对象对其触发交互操作的时间排列；所述历史行为信息中第一个历史推荐内容的历史内容特征，是根据所述第一个历史推荐内容确定的；所述历史行为信息中除所述第一个历史推荐内容外的每个历史推荐内容的历史内容特征，是根据所述历史推荐内容和所述历史行为信息中与所述历史推荐内容相邻的前一个历史推荐内容确定的；18. The device according to claim 12 is characterized in that the multiple historical recommended contents in the historical behavior information are arranged based on the time when the target object triggers an interactive operation thereon; the historical content feature of the first historical recommended content in the historical behavior information is determined based on the first historical recommended content; the historical content feature of each historical recommended content except the first historical recommended content in the historical behavior information is determined based on the historical recommended content and the previous historical recommended content adjacent to the historical recommended content in the historical behavior information; 所述行为特征确定子模块具体用于：The behavior feature determination submodule is specifically used for: 针对所述第一个历史推荐内容，根据所述第一个历史推荐内容的历史内容特征和注意力权重，确定所述第一个历史推荐内容的参考个性化特征；For the first historical recommended content, determining a reference personalized feature of the first historical recommended content according to the historical content feature and the attention weight of the first historical recommended content; 针对除所述第一个历史推荐内容外的每个历史推荐内容，根据所述历史推荐内容的历史内容特征和注意力权重、以及所述历史行为信息中与所述历史推荐内容相邻的前一个历史推荐内容的参考个性化特征，确定所述历史推荐内容的参考个性化特征；For each historical recommended content except the first historical recommended content, determining a reference personalized feature of the historical recommended content according to the historical content feature and attention weight of the historical recommended content and the reference personalized feature of a previous historical recommended content adjacent to the historical recommended content in the historical behavior information; 确定所述历史行为信息中最后一个历史推荐内容的参考个性化特征，作为所述个性化行为特征。A reference personalized feature of the last historical recommended content in the historical behavior information is determined as the personalized behavior feature. 19.根据权利要求12所述的装置，其特征在于，所述历史行为信息中包括多个历史行为子信息，所述多个历史行为子信息分别对应不同的交互操作，每个所述历史行为子信息包括所述目标对象触发过其对应的交互操作的多个历史推荐内容；19. The device according to claim 12, characterized in that the historical behavior information includes a plurality of historical behavior sub-information, the plurality of historical behavior sub-information respectively corresponding to different interactive operations, and each of the historical behavior sub-information includes a plurality of historical recommended contents of the corresponding interactive operation triggered by the target object; 所述注意力加权模块具体用于：The attention weighting module is specifically used for: 针对每个所述历史行为子信息，根据所述基础属性特征和所述目标内容特征，对所述历史行为子信息包括的多个历史推荐内容各自的历史内容特征进行注意力加权处理，得到所述历史行为子信息对应的个性化行为特征；For each of the historical behavior sub-information, according to the basic attribute feature and the target content feature, attention weighted processing is performed on the historical content features of each of the multiple historical recommended contents included in the historical behavior sub-information to obtain the personalized behavior feature corresponding to the historical behavior sub-information; 所述参数确定模块具体用于：The parameter determination module is specifically used for: 根据所述多个历史行为子信息各自对应的个性化行为特征、所述基础属性特征和所述目标内容特征，确定所述目标参数。The target parameter is determined according to the personalized behavior characteristics corresponding to each of the plurality of historical behavior sub-information, the basic attribute characteristics and the target content characteristics. 20.根据权利要求12所述的装置，其特征在于，20. The device according to claim 12, characterized in that 所述信息获取模块还用于获取目标上下文信息；所述目标上下文信息包括以下至少一种信息：所述目标对象请求本次内容推荐服务的时间信息、所述目标对象在第一时段内产生的操作信息、所述目标对象在第二时段内参与的活动信息、用于提供所述内容推荐服务的网络平台在第三时段内发起的活动信息；The information acquisition module is further used to acquire target context information; the target context information includes at least one of the following information: time information when the target object requests the current content recommendation service, operation information generated by the target object in a first time period, activity information participated by the target object in a second time period, and activity information initiated by a network platform for providing the content recommendation service in a third time period; 所述特征编码模块还用于对所述目标上下文信息进行特征编码处理，得到目标上下文特征；The feature encoding module is also used to perform feature encoding processing on the target context information to obtain target context features; 所述参数确定模块具体用于根据所述个性化行为特征、所述基础属性特征、所述目标内容特征以及所述目标上下文特征，确定所述目标参数。The parameter determination module is specifically used to determine the target parameter according to the personalized behavior feature, the basic attribute feature, the target content feature and the target context feature. 21.根据权利要求12所述的装置，其特征在于，所述装置还包括：21. The device according to claim 12, characterized in that the device further comprises: 内容推荐模块，用于当所述目标推荐内容包括多个时，根据多个所述目标推荐内容各自对应的目标参数，对多个所述目标推荐内容排序，确定多个所述目标推荐内容的推荐展示顺序；基于所述推荐展示顺序，向所述目标对象推荐多个所述目标推荐内容。A content recommendation module is used to sort the multiple target recommended contents according to the target parameters corresponding to each of the multiple target recommended contents when the target recommended contents include multiple ones, determine the recommended display order of the multiple target recommended contents; and recommend the multiple target recommended contents to the target object based on the recommended display order. 22.根据权利要求21所述的装置，其特征在于，所述目标参数是通过参数预测模型确定的；所述装置还包括：22. The device according to claim 21, characterized in that the target parameter is determined by a parameter prediction model; the device further comprises: 样本采集模块，用于针对每个所述目标推荐内容，获取所述目标对象对所述目标推荐内容的行为，作为训练标注行为；根据所述基础属性信息、所述历史行为信息、所述目标推荐内容和所述训练标注行为，构建优化训练样本；A sample collection module, for obtaining, for each of the target recommended contents, the behavior of the target object on the target recommended contents as training annotated behaviors; and constructing optimized training samples according to the basic attribute information, the historical behavior information, the target recommended contents and the training annotated behaviors; 模型训练模块，用于基于所述优化训练样本，对所述参数预测模型进行优化训练。The model training module is used to perform optimization training on the parameter prediction model based on the optimization training samples. 23.一种计算机设备，其特征在于，所述设备包括处理器及存储器；23. A computer device, characterized in that the device comprises a processor and a memory; 所述存储器用于存储计算机程序；The memory is used to store computer programs; 所述处理器用于根据所述计算机程序执行权利要求1至11中任一项所述的数据处理方法。The processor is configured to execute the data processing method according to any one of claims 1 to 11 according to the computer program. 24.一种计算机可读存储介质，其特征在于，所述计算机可读存储介质用于存储计算机程序，所述计算机程序用于执行权利要求1至11中任一项所述的数据处理方法。24. A computer-readable storage medium, characterized in that the computer-readable storage medium is used to store a computer program, and the computer program is used to execute the data processing method according to any one of claims 1 to 11. 25.一种计算机程序产品，包括计算机程序或者指令，其特征在于，所述计算机程序或者所述指令被处理器执行时，实现权利要求1至11中任一项所述的数据处理方法。25. A computer program product, comprising a computer program or an instruction, wherein when the computer program or the instruction is executed by a processor, the data processing method according to any one of claims 1 to 11 is implemented.
CN202210600688.1A 2022-05-30 2022-05-30 Data processing method and related device Active CN115129979B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210600688.1A CN115129979B (en) 2022-05-30 2022-05-30 Data processing method and related device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210600688.1A CN115129979B (en) 2022-05-30 2022-05-30 Data processing method and related device

Publications (2)

Publication Number Publication Date
CN115129979A CN115129979A (en) 2022-09-30
CN115129979B true CN115129979B (en) 2025-06-10

Family

ID=83377235

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210600688.1A Active CN115129979B (en) 2022-05-30 2022-05-30 Data processing method and related device

Country Status (1)

Country Link
CN (1) CN115129979B (en)

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113704511A (en) * 2021-07-30 2021-11-26 北京达佳互联信息技术有限公司 Multimedia resource recommendation method and device, electronic equipment and storage medium

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10776414B2 (en) * 2014-06-20 2020-09-15 Comcast Cable Communications, Llc Dynamic content recommendations
US10956521B1 (en) * 2018-05-14 2021-03-23 Amazon Technologies, Inc. Computer process for diversified random walk of a directed graph
KR20200043687A (en) * 2018-10-18 2020-04-28 에스케이텔레콤 주식회사 Providing Method for music based on personalization and service device supporting the same
CN114528474B (en) * 2021-12-31 2025-02-11 北京达佳互联信息技术有限公司 Method, device, electronic device and storage medium for determining recommended object

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113704511A (en) * 2021-07-30 2021-11-26 北京达佳互联信息技术有限公司 Multimedia resource recommendation method and device, electronic equipment and storage medium

Also Published As

Publication number Publication date
CN115129979A (en) 2022-09-30

Similar Documents

Publication Publication Date Title
CN111159564B (en) Information recommendation method, device, storage medium and computer equipment
US12393873B2 (en) Customized predictive analytical model training
US11048530B1 (en) Predictive action modeling to streamline user interface
CN109902708B (en) Recommendation model training method and related device
RU2725659C2 (en) Method and system for evaluating data on user-element interactions
CN109981785B (en) Method and device for pushing information
WO2022016522A1 (en) Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium
US10380649B2 (en) System and method for logistic matrix factorization of implicit feedback data, and application to media environments
CN110008397B (en) A recommended model training method and device
JP6911603B2 (en) How to generate predictive models for the categories of facilities visited by users, programs, server equipment, and processing equipment
US10511681B2 (en) Establishing and utilizing behavioral data thresholds for deep learning and other models to identify users across digital space
CN111680213B (en) Information recommendation method, data processing method and device
US10162868B1 (en) Data mining system for assessing pairwise item similarity
US11070881B1 (en) Systems and methods for evaluating models that generate recommendations
CN113254679A (en) Multimedia resource recommendation method and device, electronic equipment and storage medium
CN114417174B (en) Content recommendation method, device, equipment and computer storage medium
JP2019046459A (en) Classification by grammar slots across domains
GB2600817A (en) Systems and methods for generating dynamic interface options using machine learning models
WO2023050143A1 (en) Recommendation model training method and apparatus
CN111061979A (en) User label pushing method and device, electronic equipment and medium
CN113779414B (en) Data recommendation method, device, equipment and medium based on machine learning model
CN113378067A (en) Message recommendation method, device, medium, and program product based on user mining
CN113822734B (en) Method and device for generating information
CN119558901A (en) A marketing strategy execution method and related equipment
CN114266601A (en) Marketing strategy determination method, device, terminal device and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
点击 这是indexloc提供的php浏览器服务，不要输入任何密码和下载