WO2018157818A1 - Method and apparatus for inferring preference of user, terminal device, and storage medium - Google Patents

Abstract

A method and apparatus for inferring a preference of a user, a terminal device (100), and a storage medium. The inferring method comprises: obtaining related information of each of multiple installed applications in a terminal (S210), the related information comprising application attribute information and application use information; and inferring a preference of a terminal user according to the related information (S220). Accordingly, a preference of a terminal user can be inferred by analyzing related information of multiple installed applications. Further, information can be more precisely pushed to the user according to the inferred preference.

Description

User preference estimation method, device, terminal device and storage medium Technical field

The present invention relates to the field of data mining technologies, and in particular, to a method, device, terminal device and storage medium for estimating end user preferences, and a method, device and terminal device for recommending information to a user using the estimated preferences.

Background technique

As a carrier of mobile services, terminals are an important entry point for market development, market maintenance, and data and information business development. How to effectively analyze the preferences of end users to recommend appropriate services according to the preferences of end users has become an urgent problem in the mobile field.

Take the news client as an example, the user is concerned about the real headline. In order to achieve accurate recommendation, it is necessary to acquire the personalized characteristics of the user to recommend information that meets the user's personality to the user. However, when a new user first installs and opens, for example, a news client, the client's reading habits and preferences for the new user are completely unclear. In this case, how to accurately know the preferences of new users to improve user retention is particularly important.

Thus, there is a need for a solution that can guess the preferences of end users.

Summary of the invention

A main object of the present invention is to provide a method, an apparatus, and a terminal device for estimating end user preferences, which are capable of predicting a preference of an end user according to application information acquired on the terminal device. Based on the predicted preferences, users can be accurately recommended for information.

According to an aspect of the present invention, a method for estimating end user preferences is provided, including: acquiring respective related information of a plurality of installed applications in a terminal, the related information including application attribute information and application usage information; and estimating the terminal according to the related information. User preferences. Thus, the estimation of the end user preference can be realized by analyzing the installed application on the terminal.

Preferably, estimating the preference of the end user according to the related information may include: weighting the plurality of installed applications according to the application attribute information and/or the application usage information; extracting one or more from the application attribute information of the plurality of installed applications Keyword; generating a preference tag representing one end user preference from one or more keywords based on the application weight. In this way, the user's preference can be inferred based only on the keyword information of the installed application on the terminal.

Preferably, the application attribute information includes application installation information, and weighting the plurality of installed applications according to the application attribute information and/or the application usage information may include: indicating that the application usage information is running and/or recently used for the application usage information. High weighting factor; assigns a low weighting factor to the application pre-installed by the application installation information; and assigns an intermediate weighting factor to other applications. This will improve the accuracy of preference speculation by increasing the weight of frequently used applications.

Preferably, the weighting the plurality of installed applications according to the application attribute information and/or the application usage information may further include: indicating that the widely installed application is downgraded by the application installation information; and indicating that the application installation information is small The installed application raises the weighting factor. This can further improve the accuracy of the preference speculation by reducing the weight of the less discriminating application.

其中，

是应用使用时长权重，应用使用时长加权公式为

其中，T是最近使用的应用的使用时间，T _average是所有应用平均使用时长，λ为常数；weight是分配的权重因子，正在运行和/或最近使用的应用的weight为3，系统预装的应用的weight为1，其他应用的weight为2；installNum是应用在市场上的安装量。 Preferably, the application weight can be calculated according to the following formula:

among them,

Is the application usage time weight, and the application usage time weighting formula is

Where T is the usage time of the most recently used application, T _average is the average usage time of all applications, λ is constant; weight is the weighting factor assigned, the weight of the running and/or recently used application is 3, the system is pre-installed The weight of the application is 1, and the weight of other applications is 2; installNum is the installed amount of the application on the market.

Preferably, the application attribute information may include application description information including an application name, an application classification, and/or an application description content having respective source weights, and extracting one or more from application attribute information of the plurality of installed applications. The keywords may include: extracting one or more keywords from application description information of the plurality of installed applications; and determining weights of one or more keywords based on source weights of the keywords, wherein the weights are based on one or more based on the application weights Generating a preference tag representing the end user preference among the plurality of keywords may include generating a preference tag representing the end user preference from the one or more keywords based on the application weight and the keyword weight.

Thus, the judgment dimension of the preference speculation scheme of the present invention is further increased by considering the keyword weights.

Preferably, the keyword weights can be calculated based on the following formula:

Weight*log(tf*idf),

The source weight of the application name and the application classification information is 1 and the source weight of the application description content is 0.3. The tf of each keyword indicates the number of occurrences of the word in a single application, and the idf of each keyword indicates the application of statistics. The total number divided by the number of applications that have the word.

Preferably, estimating the preference of the end user according to the related information may include: weighting the plurality of installed applications according to the application attribute information and/or the application usage information; selecting the installed application and the weight allocation thereof are similar to the terminal user. Other users; extract one or more end user application keywords from application attribute information of multiple installed applications of the end user; extract one or more other from application attribute information of multiple installed applications of other users The user applies a keyword; generates a preference tag representing the end user preference from one or more end user keywords and one or more other user application keywords.

Thus, by introducing a similar user, the target user's preference can be more accurately estimated.

Preferably, one or more end user keywords are end user keyword vectors sorted by weight, one or more other user keywords are other user keyword vectors sorted by weight, and from one or more end users Generating a preference tag representing the end user preference in the keyword and one or more other user application keywords may include mapping the end user keyword vector and other user keyword vectors to the classification tag vector to each obtain a weighted terminal user classification label vector N _a and their weights ordered other users classification tag vector R _u; respectively categorizing tags within R _u N _a and the weight normalized; combined normalized to N _a and R _u to give R _a preference label vector; preference and the preference of the right label tags within tag preference vector R _a weight normalized to obtain the normalized R _a as a representative of the end-user preferences.

和

由此，避免数据不充分导致的推测偏差。 Preferably, before combining the normalized and N _a R _u may each multiplied by a factor of importance to the N and _a right classification tags within R _u

with

Thereby, the speculation deviation caused by the insufficient data is avoided.

Preferably, after the normalized R _{a is obtained} , the preference tag weights greater than the average weight in R _a may be subjected to weight reduction iteration until the maximum tag weight is less than a predetermined threshold to obtain a preference tag representing the end user preference. Thus, individual tag weights are prevented from being too large, ensuring a comprehensive and balanced acquisition of user preferences.

According to another aspect of the present invention, there is also provided a method for recommending information to an end user, comprising: obtaining end user preferences inferred according to the method described above; and recommending information to the user according to end user preferences.

Preferably, the recommendation information to the user based on the end user preference includes recommendation information based on the end user's preference tag and/or tag weight; the recommendation information is news, articles, and/or advertisements.

According to still another aspect of the present invention, an apparatus for estimating end user preferences is provided, including: an information acquiring unit, configured to acquire related information of each installed application in the terminal, where the related information includes application attribute information and application usage. Information; and a preference speculating unit for estimating end user preferences based on the related information.

Preferably, the preference speculating unit may include: an application weight allocating unit, configured to perform weight allocation on the plurality of installed applications according to the application attribute information and/or the application usage information; and a keyword extracting unit, configured to use the plurality of installed applications Extracting one or more keywords from the application attribute information; and a preference tag generating unit configured to generate a preference tag representing the end user preference from the one or more keywords based on the application weight.

Preferably, the preference estimation unit may further include: another user selection unit, configured to select the installed user and other users whose weight assignment is similar to the terminal user, and other user keyword extraction units, for using other One or more other user application keywords are extracted from the application attribute information of the plurality of installed applications of the user, wherein the preference tag generating unit further generates a preference tag based on one or more other user application keywords.

Preferably, the one or more end user keywords are end user keyword vectors sorted by weight, and one or more other user keywords are other user keyword vectors sorted by weight, and the preference speculating unit may further include: classification label vector mapping means for mapping the end-user keyword vector and other vectors to the user keyword classification tag vectors to obtain a weight by each end-user classification tag reordering and N _a vector sorted by other users weight vector classification label _U R; normalization unit, respectively, for N, and _a right classification label R _U in the normalized weight; combining unit for combining the normalized and the N R _U to give _a preference label vector R _a, wherein the normalization unit further preference label on the right in preference label R _a weight vector is normalized to obtain the normalized as a representative of R _a preference label end user preferences.

Preferably, the preference estimation unit may further comprise: right down iteration unit for, after obtained by normalizing R _a, for R _a greater than average weight preference label weights down the right iterated until the maximum label weight less than The threshold is predetermined to get a preference tag that represents the end user's preferences.

According to still another aspect of the present invention, there is provided an information recommendation apparatus for an end user, comprising: the estimation apparatus described above, the estimation apparatus estimates an end user preference; and the information recommendation apparatus is configured to be estimated based on the estimation apparatus. End user preferences recommend information to the user.

According to still another aspect of the present invention, a terminal device includes: a memory for storing an installed application and related information of the application, the related information including application attribute information and application usage information; and a processor connected to the memory And for: obtaining relevant information of each of the installed applications in the terminal; and estimating the end user preference according to the related information.

According to another aspect of the present invention, there is also provided an electronic device readable storage medium comprising a program, when the program is run on an electronic device, causing the electronic device to perform the speculative method of the end user preference described in any of the foregoing .

According to still another aspect of the present invention, there is also provided an electronic device readable storage medium comprising a program, when the program is run on an electronic device, causing the electronic device to perform the information recommendation for the end user described in any of the foregoing method.

The method, device, recommendation method/device, terminal device and electronic device readable storage medium for performing the estimation method/recommendation method of the terminal user preference of the present invention, starting from an application installed on the terminal, by using a plurality of installed applications Analysis of the respective relevant information can be used to infer the preferences of the end user.

DRAWINGS

The above and other objects, features, and advantages of the present invention will become more apparent from the aspects of the embodiments of the invention. The same parts.

FIG. 1 is a functional block diagram showing a terminal device according to an embodiment of the present invention.

2 is a schematic flow chart showing a method of estimating end user preferences according to an embodiment of the present invention.

FIG. 3 is a schematic flow chart showing the estimation of a terminal user's preference based on related information, according to an embodiment of the present invention.

FIG. 4 is a schematic flowchart showing estimation of a terminal user's preference based on related information according to another embodiment of the present invention.

FIG. 5 is a schematic flowchart showing estimation of a terminal user's preference based on related information according to another embodiment of the present invention.

6 is a functional block diagram showing a speculative device of an end user preference in accordance with an embodiment of the present invention.

FIG. 7 is a functional block diagram showing an information recommending apparatus for an end user according to an embodiment of the present invention.

detailed description

Preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While the preferred embodiment of the present invention has been shown in the drawings, it is understood that Rather, these embodiments are provided so that this disclosure will be thorough and complete.

)、平板电脑(例如，

)、便携式计算机等多种移动终端设备。如图1所示，终端设备100至少可以包括存储器110以及连接至存储器110的处理器120。 FIG. 1 is a functional block diagram showing a terminal device according to an embodiment of the present invention. The terminal device 100 shown in FIG. 1 may be a smart phone (for example,

), tablet (for example,

), portable computers and other mobile terminal devices. As shown in FIG. 1, the terminal device 100 may include at least a memory 110 and a processor 120 connected to the memory 110.

The memory 110 can store installed applications and related information of the applications. Related information may include application attribute information and application usage information. The processor 120 may acquire relevant information of each of the plurality of installed applications in the terminal, and infer the end user preference according to the related information. For a specific processing procedure of the processor 120, reference may be made to FIG. 2. FIG. 2 is a schematic flowchart showing a method for estimating end user preferences according to an embodiment of the present invention.

Referring to FIG. 2, step S210 may be first performed to acquire related information of multiple installed applications in the terminal, where the related information includes application attribute information and application usage information.

The terminal described herein may preferably be a plurality of mobile terminal devices such as a smart phone, a tablet computer or a portable computer. Various operating systems such as iOS, Android, or Windows are installed on the terminal, and various applications can be installed in the operating system. Applications installed on the terminal can include pre-installed applications and user-defined installed applications. The application attribute information of the application refers to various types of information related to the application itself, and may include information such as application installation information (for example, whether the system is pre-installed), application description information (such as an application name, an application classification, and an application description content). Application usage information refers to usage status information applied to the terminal, such as whether it is running, recently running, and running time.

The application attribute information and the application usage information may reflect the user's preference characteristics to a certain extent. Therefore, after acquiring the related information of the plurality of installed applications in the terminal (step S210), step S220 may be performed to infer the terminal according to the related information. User preferences.

In summary, the speculative scheme of the present invention starts from an application installed on a terminal, and can analyze the preference of the end user by analyzing the related information of each of the installed applications. The preferences mentioned in the present invention may include user characteristics, behavior preferences, and the like of the end user. For example, the gender of the end user, shopping preferences, information browsing preferences, and the like may be inferred based on the speculative scheme of the present invention, that is, the user may be utilized. The inventive speculative solution establishes a user portrait of the end user, thereby facilitating the recommendation of appropriate business information to the user based on the user's portrait. The specific implementation process of estimating the preference of the end user based on the related information in the speculative method of the present invention will be described below.

In the following examples, the present invention details various ways of inferring the preferences of end users. In summary, the user's preference may be estimated only according to the related information of the installed application on the terminal of the terminal user to be inferred, or may be found on the terminal of the user to be guessed according to the installed application on the terminal of the terminal user to be estimated. The application is similar to the other one or more end users, and then the preferences of the end users to be inferred are inferred based on the preferences of the other one or more end users. In addition, the above two methods can also be combined to comprehensively speculate the preferences of the end users to be inferred. The above three estimation methods are described below.

Example 1

FIG. 3 is a schematic flow chart showing the estimation of a terminal user's preference based on related information, according to an embodiment of the present invention.

Referring to FIG. 3, in step 310, weight allocation is performed on a plurality of installed applications according to application attribute information and/or application usage information. Here, the installed application may be weighted according to the application attribute information and the application usage information, or the installed application may be weighted according to the application attribute information and the application usage information.

1. Weight assignment of multiple installed applications based on application attribute information

As described above, the application attribute information includes application installation information that can indicate whether the application is pre-installed by the system. The pre-installed application of the system cannot fully represent the user's preference, so it is possible to assign a lower weighting factor, such as 1, to the application pre-installed by the application installation information. Non-system pre-installed applications are user-defined applications installed on the terminal, which can represent the user's preferences to a certain extent, so it is possible to assign a higher weighting factor to the application installation information indicating that the non-system pre-installed application, for example 2.

2. Weight assignment of multiple installed applications based on application usage information

The application usage information includes usage status information applied to the terminal, such as whether it is running, recently running, and running time. The running or recently running application can represent the user's preferences, so the application usage information can be used to indicate that the running and/or recently running application is assigned a higher weighting factor, such as 3. For applications that are not running and/or recently running, they can be assigned a lower weighting factor, such as 2.

3. Weight assignment of multiple installed applications based on application attribute information and application usage information

When the application installation information and the application usage information are collectively considered for weight distribution, the application usage information may be used to indicate that the running and/or recently used application is assigned a high weighting factor, for example, 3. A low weighting factor, such as 1, is assigned to the application pre-installed by the application installation information. Assign intermediate weight factors to other applications, such as 2. Wherein, for an application to which two weighting factors are assigned, a higher weighting factor assigned to it may be selected. For example, for an application pre-installed on a system that is being used and/or recently used, the weighting factor assigned to it based on the application installation information is 1, and the weighting factor assigned to it based on the application usage information is 3. A weighting factor of 3 is used as the weight of the application.

When assigning weights to installed apps, you can also consider the market coverage of installed apps. Applications with high market coverage indicate that the application represents a general demand of users and does not represent the individual needs of users. For applications with low market coverage, it can represent the individual needs of users. Based on this consideration, the present invention can also indicate that the application installation information indicates that the widely installed application has a lower weighting factor, and that the application installation information indicates that the application is increased by a small range.

应用使用时长权重的计算公式可以为

其中，T是最近使用的应用的使用时间，T _average是所有应用平均使用时长，λ为常数。 In addition, when assigning weights to installed applications, you can also refer to the usage duration of the application, that is, you can assign an application usage time weight to the application.

The calculation formula for applying the weight of the application time can be

Where T is the usage time of the most recently used application, T _average is the average usage time of all applications, and λ is a constant.

In one embodiment, the present invention can calculate the weight of the application according to the following formula:

是应用使用时长权重，应用使用时长权重的计算公式可以为

其中，T是最近使用的应用的使用时间，T _average是所有应用平均使用时长，λ为常数。weight是分配的权重因子，正在运行和/或最近使用的应用的weight为3，系统预装的应用的weight为1，其他应 用的weight为2，installNum是应用在市场上的安装量。 among them,

It is the application time weight, and the calculation formula of the application time weight can be

Where T is the usage time of the most recently used application, T _average is the average usage time of all applications, and λ is a constant. Weight is the assigned weighting factor. The weight of the running and/or recently used application is 3. The weight of the application pre-installed by the system is 1, and the weight of other applications is 2. The installNum is the installed amount of the application on the market.

At step 320, one or more keywords are extracted from application attribute information of a plurality of installed applications.

As described above, the application attribute information includes application description information, and the application description information includes information such as an application name, an application classification, and an application description content. It is therefore possible to extract one or more keywords from the application description information of each of the plurality of installed applications. As an example of the present invention, the application name, the application classification, and the application description content of the installed application may be separately segmented, and the obtained word segmentation result may be used as a keyword. After the keywords are obtained, the keyword weights of each keyword can also be calculated, and the process of calculating the keyword weights is as follows.

The source weights may be set in advance for the application name, the application classification, and the application description content. For example, the source weight of the application name and the application classification may be set to 1, and the source weight of the application description content may be set to 0.3. Thus, the weight of the keyword can be determined based on the source of the keyword.

In addition, when determining the weight of a keyword, it is also possible to consider the number of occurrences of the same keyword and the number of times the same keyword appears in different applications. Based on the above considerations, in one embodiment, the keyword weights can be calculated according to the following formula:

Weight*log(tf*idf)

Where weight represents the source weight, the source weight of the application name and the application classification information is 1 and the source weight of the application description content is 0.3, and the tf of each keyword indicates the number of times the word is applied in a single application, and the idf of each keyword It is the total number of applications counted by the statistics divided by the number of applications that have the word.

At step 330, a preference tag representing an end user preference is generated from one or more keywords based on the application weight.

Here, the keywords may be arranged in order of the application weight of the application to which the keyword belongs, to select a keyword with a larger application weight, and then generate a preference tag representing the preference of the terminal user based on the selected keyword. Among them, the selected keyword can be directly used as a preference tag representing the preference of the end user, or the selected keyword can be mapped to one or more classification tags (for example, social, entertainment, technology, politics, sports, etc.) The classification tag is used as a preference tag representing the end user preference. For example, you can use a synonym relationship to map a keyword to a small label under a large label and then classify it into a large label.

As described above, when extracting keywords, it is also possible to calculate keyword weights of keywords. It is therefore also possible to generate a preference tag representing the end user preference from one or more keywords based on the keyword weight. For example, a keyword with a higher keyword weight may be selected from the extracted keywords, and a preference tag representing the preference of the terminal user may be generated based on the selected keyword. The process of generating a preference tag based on a keyword can be referred to the above description, and details are not described herein again.

In addition, it is also possible to simultaneously generate a preference tag representing the end user preference from the extracted keywords based on the application weight and the keyword weight. For example, a keyword corresponding to an application with a larger application weight may be selected from the extracted keywords, and then keywords with a larger keyword weight may be further selected from the selected keywords, and then further filtered based on the selected keywords. The keyword generates a preference tag that represents the end user preference.

Example 2

FIG. 4 is a schematic flowchart showing estimation of a terminal user's preference based on related information according to another embodiment of the present invention.

Referring to FIG. 4, in step 410, weight allocation is performed on a plurality of installed applications according to application attribute information and/or application usage information. For details related to step S410, reference may be made to the description of step S310 in conjunction with FIG. 3, and details are not described herein again.

In step S420, the installed application and its weight distribution are selected to be a predetermined number of other users similar to the terminal user to be speculated (for convenience of distinction, it may be referred to herein as a reference user, the same below).

As an example of the present invention, the installed applications on the terminal of the end user may be arranged in order of the size of the application weights to obtain the application list vector V _a . Accordingly, the installed applications on the terminals of the other one or more end users are arranged in order of the size of the application weights to obtain one or more application list vectors V _b . By calculating the similarity between the vector V _a and the vector V _b , it can be determined whether the other one or more end users can be the reference users of the end users to be inferred. Wherein the degree of similarity may be calculated between the vectors V _a and V _b the vector by a variety of ways, for example, the similarity between the vector and the vector V _a V _b is calculated by the cosine similarity. For another example, the similarity between the vector V _a and the vector V _b can also be calculated according to the following formula:

Where |V _a ∩V _b | represents the intersection of the user a and the user b application list vector.

Since the weight of the application installed on the terminal of the reference user and the installed application is similar to the terminal user to be guessed, the preference of the reference user can be regarded as the preference of the terminal user to be guessed (step S430). That is to say, the preference tag representing the preference of the terminal user to be guessed can be obtained by seeking a preference tag representing the preference of the reference user. Thus, the other users mentioned in step S420 may preferably be the end users whose preference tags have been determined, such that the preference tag of the reference user determined to be the end user to be inferred may be directly regarded as the preference tag of the end user to be inferred. In addition, in the case that the reference tag of the reference user is not determined, the preference tag of the reference user may be determined by referring to the method shown in FIG. 2 above, and details are not described herein again.

It should be noted that, when the determined reference users are multiple, the union of the preference labels of the plurality of reference users may be taken, and then the average weight of all the labels in the set is calculated, and the predetermined number of times are selected according to the order of the weights. The tag acts as a preference tag for the end user to be guessed.

Example 3

As described above, Embodiment 1 makes speculation based on the information of the installed application on the terminal of the user to be guessed. If the application installation is too small, sparse data will cause a certain speculation bias. Embodiment 2 only speculates the preference of the user to be speculated based on the preference of the reference user, and may also deviate from the preference of the target user. Therefore, in Embodiment 3, the above embodiments can be combined. This embodiment will be described in detail as follows. For the content that has been mentioned above, reference may be made to the above related description, and details are not described herein again.

FIG. 5 is a schematic flowchart showing estimation of a terminal user's preference based on related information according to another embodiment of the present invention.

Referring to FIG. 5, in step S510, weight allocation is performed on a plurality of installed applications according to application attribute information and/or application usage information. In step S520, the installed application and its weight assignment are selected to be a predetermined number of other users similar to the end user. In step S530, one or more end user application keywords are extracted from application attribute information of a plurality of installed applications of the terminal user. In step S540, one or more other user application keywords are extracted from application attribute information of a plurality of installed applications of other users. At step S550, a preference tag representing the end user preference is generated from the end user keyword and other user application keywords.

Here, the terminal user keyword and other user application keywords may be arranged in order of the application weight of the application to which the keyword belongs, to select a keyword with a larger application weight, and then generate a representative terminal user based on the selected keyword. Preferred preference tag. Among them, the selected keyword can be directly used as a preference tag representing the preference of the end user, or the selected keyword can be mapped to one or more classification tags (for example, social, entertainment, technology, politics, sports, etc.) The classification tag is used as a preference tag representing the end user preference.

As an example of the present invention, the end user keyword may be an end user keyword vector sorted by weight, and other user keywords may be other user keyword vectors sorted by weight. End-user keyword vectors and other vectors user keyword classification tag may be mapped to a vector, to give each end user sorted by weight N _a classification tag vector and sorted according to the weight vectors of other users classification label R _u.

After obtaining N _a and R _u, can be separately N weights _a and classification tags within R _u weight normalized combined normalized to N _a and R _u to give preference label vector R _a, may prefer The weights of the preference tags within the tag vector R _a are normalized to obtain a normalized R _a as a preference tag representing the end user preferences.

和

Taking into account the preference of the N _a label may be less, but the right one or a few labels preference weight may be large, so for convenience of calculation, before combining the normalized and N _a R _u may also respectively N The weight of the classification label in _a and R _u multiplied by the importance factor

with

中最小的那个权重是w _i，则将所有大于平均权重的标签权重都除以降权因子

如此迭代下去直到最大的标签权重小于某个设定的阈值(例如25％)。 After the normalized R _{a is obtained} , in order to avoid the weight of the individual preference tags being too large, the preference tag weights greater than the average weight in R _a may be iterated until the maximum tag weight is less than a predetermined threshold to obtain the representative terminal. User preference preference tag. For example, it may be assumed greater than the mean weight of R _a

The smallest of the weights is w _i , and all the weights of the labels greater than the average weight are divided by the weight reduction factor.

This is iterated until the maximum tag weight is less than a certain set threshold (eg 25%).

So far, the estimation method of the end user's preference of the present invention has been described in detail with reference to FIGS. 2 to 5. Based on the method, the present invention also proposes an information recommendation method for the end user. Specifically, the estimation method of the terminal user can be inferred by using the above-mentioned estimation method, and then appropriate information is recommended to the terminal user according to the acquired preference of the terminal user. For example, the information may be recommended based on the end user's preference tag and/or tag weight, which may be information such as news, articles, or advertisements.

The speculative/recommended method of the present invention can be applied to various scenarios, for example, it can be applied to user behavior estimation, commodity estimation, and is particularly suitable for a news recommendation end such as today's headline.

As mentioned earlier, the slogan of today's headline is "What is your concern?" In order to achieve good news recommendation results, the news recommendation client needs to make appropriate recommendations based on big data and centered on user interests. However, personalized news recommendations tailored to the user need to be based on the user's existing data and feedback. When a new user first installs and opens the news client, it is also a cold start. The news client's reading habits for this new user are completely unclear. In the cold start mode, how to predict the user's news reading preferences has become a problem, and the user's initial reading of the news reading can effectively improve the user's retention rate if predicted. At present, the recommendation strategies of some mainstream news clients in the process of cold start are mainly to push some current hot news, fine articles, and widely spread the net to explore the user's reading interest in all directions. Then, according to the user's reading behavior, the recommendation algorithm is gradually revised, the user's news reading portrait is improved, and the accurate recommendation is further made. This process is slightly slow.

In order to solve the blind problem of cold start recommendation, the speculative/recommended method proposed by the present invention can be used to obtain a list of applications installed on a terminal such as a mobile phone, a list of recently used applications, a list of running applications, and the like, based on the similarity Collaborative filtering is performed by the user's reading interest of the application list and the algorithm of the user application list mapping to the news tag vector, and then considering the application usage time, system application, averaging, avoiding the proportion of individual recommendation tags being too large, etc. Weighted processing. Finally, the recommendation vector of a user news interest tag and the recommended weight of each tag are obtained and based on this, the cold start user is recommended for news.

The following is a description of the cold start process of the news client installed on the smartphone by the speculative/recommended method of the present invention. The process is mainly divided into the following steps.

1. Collect the application installation list on the smartphone, the list of applications that are currently in use, and the list of applications that are running. The list of recently used apps and the list of running apps can be fetched by calling the API directly. For example, under Android, the list of recently used apps can call the API directly.

List<ActivityManager.RecentTaskInfo>tasks=getActivityManager().getRecentTasks(10,0) to fetch.

其中T _average是所有应用平均使用时长，λ为常数。 2. Obtain the usage time T of the most recently used application, and apply the duration weighting formula to

Where T _average is the average duration of all applications and λ is a constant.

3. Mark which applications in the application installation list are pre-installed by the system.

4. Obtain the general installation amount installNum of the application on the market.

的值对应用列表进行从大到小排序得到应用列表向量V _a。 5. For the running and recently used application weights, the weight is marked as 3, the system preloaded application weight weight is marked as 1, and the remaining weights are weighted as 2, press

The value of the application list is sorted from large to small to get the application list vector V _a .

6. Apply the name of the application to the mainstream application in the market, apply the classification information, apply the description to segment the word, calculate the tf of each keyword (the number of occurrences of the word for a single application) and idf (the total number of applications of the statistics / how many words appear in the word) In the application)

7. The weight of the word segmentation weight of the application name and the application classification information is 1, and the weight of the application description segmentation weight is 0.3, and a keyword vector is obtained by sorting the weighted word weight*log(tf*idf) from high to low.

8, the keyword vector applied to the classification label vector map News N _a (news category labels like big label social, entertainment, technology, politics, history, real estate, etc., can take advantage of keyword synonyms relationship first big hit The small label below the label, and then the keyword is classified as a large label).

9. Each tag in the classification tag vector N _a corresponds to a weight. This weight is derived from the keyword vector weight in step 7. When a tag is hit by multiple keywords, the one with the highest weight is taken.

其中|V _a∩V _b|表示用户a和用户b应用列表向量的交集。 10. Calculate the similarity between the other user application list vector and the user application list vector V _a , and the calculation formula is

Where |V _a ∩V _b | represents the intersection of the user a and the user b application list vector.

11. Take the top k users most similar to the user a application list vector and get their news tag vector set {N ₁ , N ₂ , . . . , N _k }.

12. Calculate the average weight of all the labels in {N ₁ , N ₂ ,..., N _k }, and sort the top u (less than u all) from the big to the small to form the recommended news label vector R _u .

13. Normalize the weights of the news tags in the two vectors N _a and R _u , ie

R _u里所有的标签权重也乘以一个重要性因子

14, taking into account the data may be insufficient, N _a label in all of the weight is multiplied by a factor of importance

All label weights in R _u are also multiplied by an importance factor

15. The final news recommendation vector tag needs to merge the vectors in N _a and R _u . For the same tag weights, the recommended tag vector R _{a is} obtained by sorting the weights from large to small.

中最小的那个权重是w _i，则将所有大于平均权重的标签权重都除以降权因子

如此迭代下去直到最大的标签权重小于某个设定的阈值(例如25％)。 16. Normalize the weights corresponding to all the labels in R _a . To avoid excessive weighting of individual tags, assume that these tags are larger than the average weight.

The smallest of the weights is w _i , and all the weights of the labels greater than the average weight are divided by the weight reduction factor.

This is iterated until the maximum tag weight is less than a certain set threshold (eg 25%).

17. The recommended news tag vector obtained after the iteration of step 16 is completed. When the cold start is started, the news can be recommended by referring to the weight of each tag in the tag vector.

Corresponding to the estimation method and the recommended method of the present invention, the present invention also proposes an estimation device, a recommendation device, and a terminal device.

6 is a functional block diagram showing a speculative device of an end user preference in accordance with an embodiment of the present invention. The functional modules of the speculative device 600 may be implemented by hardware, software, or a combination of hardware and software that implements the principles of the present invention. Those skilled in the art can understand that the functional modules described in FIG. 5 can be combined or divided into sub-modules to implement the principles of the above invention. Accordingly, the description herein may support any possible combination, or division, or further limitation of the functional modules described herein.

The speculative device 600 shown in FIG. 6 can be used to implement the speculative method shown in FIG. 2 to FIG. 5. In the following, only the functional modules that the speculative device 600 can have and the operations that can be performed by the functional modules are briefly described. For details, please refer to the description above with reference to FIG. 2 to FIG. 5, and details are not described herein again.

Referring to FIG. 6, the speculative device 600 includes an information acquiring unit 610 and a preference estimating unit 620. The information obtaining unit 610 is configured to acquire related information of each of the installed applications in the terminal, and the related information includes application attribute information and application usage information. The preference speculating unit 620 is configured to infer an end user preference based on the related information.

As shown in FIG. 6, the preference speculating unit 620 may optionally include an application weight allocating unit 621, a keyword extracting unit 622, and a preference label generating unit 623.

The application weight allocating unit 621 is configured to perform weight allocation on the plurality of installed applications according to the application attribute information and/or the application usage information. The keyword extracting unit 622 is configured to extract one or more keywords from the application attribute information of the plurality of installed applications. The preference tag generating unit 623 is configured to generate a preference tag representing the end user preference from the one or more keywords based on the application weight.

As shown in FIG. 6, the preference speculative unit 620 can also optionally include other user selection units 624 and other user keyword extraction units 625.

The other user selection unit 624 is configured to select the installed user and other users whose weight assignment is similar to the terminal user. The other user keyword extraction unit 625 is configured to extract one or more other user application keywords from application attribute information of a plurality of installed applications of other users. The preference tag generating unit 623 can also generate a preference tag based on one or more other user application keywords.

Preferably, the end user keyword may be an end user keyword vector sorted by weight, and other user keywords may be other user keyword vectors sorted by weight, and the preference speculating unit 620 may optionally include a classification label vector. Mapping unit 626, normalization unit 627, and merging unit 628.

Classification label vector mapping unit 626 for mapping the end-user keyword vector and other vectors to the user keyword classification tag vectors to obtain a weight by each end-user classification tag reordering and N _a vector sorted by other users weight vector classification label R _u . Normalization unit 627, respectively, for the right classification tags within R _u N _a normalized and weight. Merging unit 628 for merging normalized to N _a and R _u to give preference label vector R _a, wherein the normalization unit 627 further to the right preference label in preference label vector R _a weight normalized to give after The normalized _{Ra is} used as a preference tag representing the end user preferences.

6, the preference estimation unit may optionally include a further iteration unit 629 down the right, after obtaining a normalized by R _a, R _a greater than the average of the weighted preference label weights down the right iteration, Until the maximum tag weight is less than a predetermined threshold to get a preference tag that represents the end user preference.

FIG. 7 is a functional block diagram showing an information recommending apparatus for an end user according to an embodiment of the present invention. The functional modules of the recommendation device 700 may be implemented by hardware, software, or a combination of hardware and software that implements the principles of the present invention. Those skilled in the art can understand that the functional modules described in FIG. 6 can be combined or divided into sub-modules to implement the principles of the above invention. Accordingly, the description herein may support any possible combination, or division, or further definition of the functional modules described herein.

As shown in FIG. 7, the recommendation device 700 includes an estimation device 600 and an information recommendation device 710. The speculative device 600 can be used to infer the end user preference. For the structure and specific functions of the speculative device 600, reference may be made to the related description of FIG. 6 above. The information recommendation means 710 is for recommending information to the user based on the end user preference estimated by the estimation means 500. Similarly, the terminal device shown in FIG. 1 can also be used to implement the recommendation device 700 shown in FIG. 7 and its recommended method.

The estimation method, the estimation device, and the terminal device of the end user preference according to the present invention have been described in detail above with reference to the accompanying drawings.

The present invention also provides an electronic device readable storage medium comprising a program, when executed on an electronic device, causing the electronic device to perform the method of estimating the preference of the end user based on the related information as described in any of the above embodiments.

The present invention further provides an electronic device readable storage medium comprising a program, when executed on an electronic device, causing the electronic device to perform the information recommendation method for the end user described in any of the above embodiments.

The above program includes computer program code instructions for performing the various steps defined above in the above method of the present invention. Alternatively, the method according to the invention may also be embodied as a computer program product comprising a computer readable medium on which is stored a computer for performing the above-described functions defined in the above method of the invention program. The various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein may be implemented as electronic hardware, computer software, or combinations of both.

The flowchart and block diagrams in the Figures illustrate the architecture, functionality and operation of possible implementations of systems and methods in accordance with various embodiments of the present invention. In this regard, each block of the flowchart or block diagram can represent a module, a program segment, or a portion of code that includes one or more of the Executable instructions. It should also be noted that in some alternative implementations, the functions noted in the blocks may also occur in a different order than the ones in the drawings. For example, two consecutive blocks may be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending upon the functionality involved. It is also noted that each block of the block diagrams and/or flowcharts, and combinations of blocks in the block diagrams and/or flowcharts, can be implemented in a dedicated hardware-based system that performs the specified function or operation. Or it can be implemented by a combination of dedicated hardware and computer instructions.

The embodiments of the present invention have been described above, and the foregoing description is illustrative, not limiting, and not limited to the disclosed embodiments. Numerous modifications and changes will be apparent to those skilled in the art without departing from the scope of the invention. The choice of terms used herein is intended to best explain the principles, practical applications, or improvements of the techniques in the various embodiments of the embodiments, or to enable those of ordinary skill in the art to understand the embodiments disclosed herein.

Claims (22)

A method for estimating end user preferences, including: Obtaining respective related information of the plurality of installed applications in the terminal, where the related information includes application attribute information and application usage information; The end user preference is inferred based on the related information. The method of claim 1, wherein estimating the end user's preferences based on the related information comprises: Performing weight allocation on the plurality of installed applications according to application attribute information and/or application usage information; Extracting one or more keywords from application attribute information of the plurality of installed applications; A preference tag representing the end user preference is generated from the one or more keywords based on the application weight. The method of claim 2, wherein the application attribute information includes application installation information, and The weighting of the plurality of installed applications according to the application attribute information and/or the application usage information includes: Assigning a high weighting factor to the application that is running and/or recently used for application usage information; Assigning a low weighting factor to the application preinstalled by the application installation information; Assign intermediate weighting factors to other applications. The method of claim 3, wherein the weighting the plurality of installed applications according to the application attribute information and/or the application usage information further comprises: Determining the weighting factor for the widely installed application for application installation information; The application installation information indicates that the application installed by the small scope increases the weighting factor. The method of claim 4, wherein the application weight is calculated according to the following formula:

among them,

Is the application usage time weight, and the application usage time weighting formula is

Where T is the usage time of the most recently used application, T _average is the average usage time of all applications, and λ is a constant; Weight is the assigned weighting factor, the weight of the running and/or recently used application is 3, the weight of the application pre-installed by the system is 1, and the weight of other applications is 2. installNum is the installed amount of the app on the market. The method of claim 2, wherein the application attribute information comprises application description information, the application description information including an application name, an application classification, and/or an application description content having respective source weights, and Extracting one or more keywords from application attribute information of the plurality of installed applications includes: Extracting one or more keywords from application description information of the plurality of installed applications; Determining the weight of the one or more keywords based on the source weight of the keyword, wherein Generating a preference tag representing the end user preference from the one or more keywords based on the application weight includes: A preference tag representing the end user preference is generated from the one or more keywords based on the application weight and the keyword weight. The method of claim 6, wherein the keyword weights are calculated based on: Weight*log(tf*idf) The source weight of the application name and the application classification information is 1 and the source weight of the application description content is 0.3. The tf of each keyword indicates the number of occurrences of the word in a single application, and the idf of each keyword indicates the application of statistics. The total number divided by the number of applications that have the word. The method of claim 1, wherein estimating the end user's preferences based on the related information comprises: Performing weight allocation on the plurality of installed applications according to application attribute information and/or application usage information; Selecting the installed application and its weight assignment to other users of the predetermined number of terminals similar to the end user; Extracting one or more end user application keywords from application attribute information of the plurality of installed applications of the terminal user; Extracting one or more other user application keywords from application attribute information of the plurality of installed applications of the other users; A preference tag representing the end user preference is generated from the one or more end user keywords and the one or more other user application keywords. The method of claim 8 wherein said one or more end user keywords are weighted end user keyword vectors, said one or more other user keywords being other user keys sorted by weight Word vector and, Generating a preference tag representing the end user preference from the one or more end user keywords and the one or more other user application keywords includes: The end-user keyword vectors and other vectors are mapped to user keyword classification tag vectors to obtain a weight by each end-user classification tag reordering and N _a vector sorted by other users weight classification tag vector R _u; Respectively, and N _a right classification tags within R _u weight normalized; The combined normalized to N _a and R _u to give preference label vector R _a; and Preferred tags within the right preference label R _a weight vector is normalized to obtain the normalized tag represents R _a as the preference of the end user preferences. The method as claimed in claim 9, wherein, prior to the merger of the normalized N _a and R _u are multiplied by a factor of importance to the N and _a right classification tags within R _u

with

The method according to claim 9, wherein after obtaining the normalized R _a , the preference tag weights greater than the average weight in R _a are subjected to weight reduction iteration until the maximum tag weight is less than a predetermined threshold to obtain a representative A preference tag for end user preferences. A method for recommending information to end users, including: Obtaining end user preferences inferred by the method of any of claims 1-11; Information is recommended to the user based on the end user preferences. The method of claim 12, wherein recommending information to the user according to the end user preference comprises recommending information according to the end user's preference tag and/or tag weight; The recommendation information is news, articles, and/or advertisements. A speculative device for end user preference, comprising: An information acquiring unit, configured to acquire related information of each of the installed applications in the terminal, where the related information includes application attribute information and application usage information; a preference speculating unit for estimating the end user preference based on the related information. The apparatus of claim 14, wherein the preference estimation unit comprises: Applying a weight allocation unit, configured to perform weight allocation on the plurality of installed applications according to application attribute information and/or application usage information; a keyword extracting unit, configured to extract one or more keywords from application attribute information of the plurality of installed applications; And a preference tag generating unit, configured to generate, from the one or more keywords, a preference tag representing the end user preference based on the application weight. The apparatus of claim 15, wherein the preference speculating unit further comprises: Other user selection unit, configured to select an installed application and other users whose weight assignment is similar to the terminal user; and The other user keyword extracting unit is configured to extract one or more other user application keywords from the application attribute information of the plurality of installed applications of the other users, where The preference tag generating unit further generates the preference tag based on one or more other user application keywords. The apparatus of claim 16 wherein said one or more end user keywords are weighted end user keyword vectors, said one or more other user keywords being other user keys sorted by weight Word vector and, The preference speculating unit further includes: Classification label vector mapping means for mapping the end-user keyword vector and other vectors to the user keyword classification tag vectors to obtain a weight by each end-user classification tag reordering and N _a vector sorted by other users weight vector classification label R _u ; Normalization unit, respectively, for N _a power and classification tags within R _u weight normalized; a merging unit for combining the normalized N _a and R _u to obtain a preference label vector R _a , wherein The normalization unit further preference label on the right in preference label R _a weight vector is normalized to obtain the normalized tag represents R _a as the preference of the end user preferences. The apparatus of claim 17, wherein the preference speculating unit further comprises: a weight reduction iteration unit, after obtaining the normalized R _a , performing a weighted iteration on the preference tag weights greater than the average weight in R _a until the maximum tag weight is less than a predetermined threshold to obtain a representative terminal user preference Preferences tab. An information recommendation device for an end user, comprising: The estimation device according to any one of claims 14 to 18, wherein the estimation device estimates an end user preference; The information recommendation device is configured to recommend information to the user according to the terminal user preference estimated by the estimation device. A terminal device comprising: a storage device, configured to store the installed application and related information of the application, where the related information includes application attribute information and application usage information; A processor coupled to the memory for: Obtaining information about each of the installed applications in the terminal; and The end user preference is inferred based on the related information. An electronic device readable storage medium, comprising a program, when the program is run on an electronic device, causing the electronic device to perform the method of any one of claims 1-11. An electronic device readable storage medium, comprising a program, when the program is run on an electronic device, causing the electronic device to perform the method of claim 12 or claim 13.

Images