CN120494942A - A method and system for intelligent product recommendation based on user behavior analysis - Google Patents
A method and system for intelligent product recommendation based on user behavior analysisInfo
- Publication number
- CN120494942A CN120494942A CN202510969220.3A CN202510969220A CN120494942A CN 120494942 A CN120494942 A CN 120494942A CN 202510969220 A CN202510969220 A CN 202510969220A CN 120494942 A CN120494942 A CN 120494942A
- Authority
- CN
- China
- Prior art keywords
- user
- data
- browsing
- purchase
- intention
- 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.)
- Granted
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明公开了一种基于用户行为分析的商品智能推荐方法及系统，涉及商品推荐技术领域。一种基于用户行为分析的商品智能推荐系统，包括有：行为分析模块和商品推荐模块。本发明通过获取用户历史的商品浏览数据并进行分类筛选，能够精确地识别用户的行为类型，从而更好地了解用户的兴趣和购买意图，这种数据的分类和分析能够有效支持个性化推荐，确保推荐内容与用户需求更加匹配；基于用户购买意图特征和浏览意图特征的对比分析，能够准确地识别出用户的潜在购买意图和浏览偏好，从而优化商品推荐内容。
The present invention discloses a method and system for intelligent product recommendation based on user behavior analysis, and relates to the technical field of product recommendation. A product intelligent recommendation system based on user behavior analysis includes: a behavior analysis module and a product recommendation module. By obtaining the user's historical product browsing data and performing classification and screening, the present invention can accurately identify the user's behavior type, thereby better understanding the user's interests and purchasing intentions. The classification and analysis of such data can effectively support personalized recommendations and ensure that the recommended content is more closely matched with user needs. Based on the comparative analysis of the user's purchase intention characteristics and browsing intention characteristics, the user's potential purchase intentions and browsing preferences can be accurately identified, thereby optimizing the product recommendation content.
Description
Technical Field
The invention relates to the technical field of commodity recommendation, in particular to an intelligent commodity recommendation method and system based on user behavior analysis.
Background
In the prior commodity recommendation technology, a collaborative filtering or content-based method is generally adopted by an e-commerce platform, interest preference is predicted and related commodities are recommended by analyzing action sequences of clicking, purchasing and the like of a user, however, the methods have obvious limitations that firstly, defects exist when distinguishing accidental actions such as deep browsing intention and invalid jump and the like of the user, so that deviation is caused to judgment of real interests of the user, secondly, commodity recommendation and preferential strategies are generally treated as two independent systems in the prior art, induced preferential strategies cannot be integrated into recommendation calculation of interests of the user, and effective means for promoting final achievement of the user are lacked in recommendation algorithms.
In order to solve the technical problems, the content interest evolution and the preferential sensitivity evolution of the user are subjected to parallel modeling and joint learning, and the optimal transformation induction scheme is actively learned and executed while the user interest is accurately predicted, so that the transformation efficiency from intention to achievement is remarkably improved.
Disclosure of Invention
The invention aims to provide an intelligent commodity recommending method and system based on user behavior analysis, which actively learn and execute an optimal transformation induction scheme while accurately predicting user interests, so that the transformation efficiency from intention to achievement is remarkably improved.
An intelligent commodity recommendation method based on user behavior analysis comprises the following steps:
Classifying and screening the commodity browsing data to obtain transaction conversion data, intention purchasing data, browsing loss data and invalid jump data;
analyzing based on the intercourse transformation data, extracting an induced preferential data set corresponding to each intercourse transformation data, analyzing based on the induced preferential data set to obtain the user purchase intention characteristic, and comparing and analyzing based on the intention purchase data and the browsing loss data to obtain the user browsing intention characteristic;
And recommending commodities to the user in a prediction period according to the user purchase interest evolution model.
As a preferred technical scheme of the invention, the method for classifying and screening commodity browsing data comprises the following specific steps:
Acquiring corresponding original behavior data and user interaction characteristics aiming at each piece of commodity browsing data, wherein the original behavior data comprises, but is not limited to, a user-commodity identification, a browsing time stamp and a follow-up behavior type;
according to the follow-up behavior type, directly dividing corresponding commodity browsing data into transaction conversion data, intention purchasing data and residual commodity browsing data;
the method comprises the steps of obtaining semantic relevance scores among recently browsed commodities of a user for residual commodity browsing data, carrying out quantitative scores based on commodity stay time and page rolling depth to obtain browsing effective scores, carrying out weighted calculation on the semantic relevance scores and the browsing effective scores to obtain interaction depth scores;
if the interaction depth score is lower than the set browsing threshold, the residual commodity browsing data are divided into invalid jump data, otherwise, the residual commodity browsing data are divided into browsing loss data.
As a preferred technical scheme of the invention, the specific steps for analyzing the user behavior based on the induced preference data set comprise the following steps:
From the conversion data of the interchange, one or more pieces of induced discount data received before the interchange are associated for each piece of the conversion data of the interchange so as to construct an induced discount data set, wherein the induced discount data set comprises at least one piece of accepted discount data and zero or more pieces of refused discount data, and the purchase inflection point feature vector in the induced discount data set is calculated;
The method comprises the steps of combining purchase inflection point feature vectors corresponding to all the transaction conversion data to obtain a purchase inflection point feature space, clustering the purchase inflection point feature space by using an unsupervised clustering algorithm to obtain a plurality of user intention type clusters, and semantically labeling each user intention type cluster to obtain user purchase intention features.
As a preferred technical scheme of the invention, the specific steps for carrying out the comparative analysis based on the intention purchasing data and the browsing loss data comprise the following steps:
The method comprises the steps of obtaining acquisition data of a commodity, analyzing intention shopping data to obtain corresponding shopping commodity characteristic data, analyzing browsing loss data to obtain corresponding loss commodity characteristic data, calculating similarity of the shopping commodity characteristic data and the loss commodity characteristic data, and combining the intention shopping data with the similarity being larger than a commodity similarity threshold value with the browsing loss data to obtain a dual behavior-commodity data set;
Extracting features of the dual behavior-commodity data set to obtain a composite interaction feature vector;
Taking the composite interaction feature vector as input, and taking the corresponding contribution data type as a target label to construct a commodity importance analysis model;
and carrying out intention rule analysis on all the composite interaction feature vectors with higher contribution degree to obtain browsing intention features of the user.
As a preferred technical scheme of the invention, the specific steps of establishing the user purchase interest evolution model based on the user purchase intention characteristics and the user browsing intention characteristics comprise the following steps:
performing differential calculation on the user browsing intention characteristic and the user purchasing intention characteristic to obtain a purchasing conversion gain sequence;
Sequencing the browsing intention features of each user and the purchasing intention features of the user according to the sequence of the browsing time stamp to obtain a purchasing interest evolution sequence of the user;
The user purchase interest evolution model comprises a content interest evolution channel and a preferential sensitivity evolution channel, wherein the content interest evolution channel is trained based on a user purchase interest evolution sequence, the output is a predicted target commodity set in the next period, the preferential sensitivity evolution channel is trained based on a purchase conversion gain sequence, and the output is a predicted preferential sensitivity state in the next period.
As a preferred technical scheme of the invention, the method comprises the specific steps of recommending commodities to a user in a prediction period according to a user purchase interest evolution sequence, and comprises the following steps:
Inputting the behavior sequence into a user purchase interest evolution model for double-channel analysis to obtain a recommended commodity list and a personalized induced preference strategy matched with the recommended commodity list;
and recommending the commodity for the user by utilizing the recommended commodity list and the personalized induced preferential strategy matched with the recommended commodity list.
An intelligent commodity recommendation system based on user behavior analysis, comprising:
The system comprises a behavior analysis module, a comparison analysis module and a comparison analysis module, wherein the behavior analysis module comprises a purchase behavior analysis unit and a browsing behavior analysis unit, the purchase behavior analysis unit is used for acquiring a plurality of commodity browsing data of a user history in the previous period, classifying and screening the commodity browsing data to obtain transaction conversion data, intention purchase data, browsing loss data and invalid jump data, the browsing behavior analysis unit is used for analyzing based on the transaction conversion data, extracting an induced preference data set corresponding to each transaction conversion data, carrying out user behavior analysis based on the induced preference data set to obtain the purchase intention characteristic of the user, and carrying out comparison analysis based on the intention purchase data and the browsing loss data to obtain the browsing intention characteristic of the user;
The commodity recommending module comprises an intelligent recommending unit and is used for establishing a user purchase interest evolution model based on user purchase intention characteristics and user browsing intention characteristics, and recommending commodities to the user in a prediction period according to the user purchase interest evolution model.
The invention has the following advantages:
1. The method and the system can accurately identify the behavior type of the user by acquiring the commodity browsing data of the user history and classifying and screening the commodity browsing data, so that the interests and the buying intention of the user can be better known, personalized recommendation can be effectively supported by classifying and analyzing the data, the recommended content is ensured to be more matched with the user demand, the potential buying intention and browsing preference of the user can be accurately identified based on the comparative analysis of the buying intention characteristic and the browsing intention characteristic of the user, so that the commodity recommended content is optimized, and commodity recommendation meeting the interests of the user can be provided for the user by establishing a commodity buying interest evolution model and predicting a commodity set output according to the model, so that the buying conversion rate of the user is improved.
2. The dynamic modeling mode enables a recommendation system to flexibly adapt to interest changes of users in different time periods, better predicts commodities possibly interested by the users in future periods, improves timeliness and accuracy of recommendation, can better identify purchasing behavior characteristics of different user groups by clustering purchasing inflection point feature vectors through an unsupervised clustering algorithm, generates personalized recommendation lists for each user group, and can effectively find potential user groups and corresponding purchasing intentions through clustering analysis, so that the recommendation system is further optimized.
Drawings
Fig. 1 is a schematic structural diagram of an intelligent commodity recommendation system based on user behavior analysis according to an embodiment of the present invention.
Detailed Description
In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.
Embodiment 1, a commodity intelligent recommendation method based on user behavior analysis, comprising the following steps:
Classifying and screening the commodity browsing data to obtain transaction conversion data, intention purchasing data, browsing loss data and invalid jump data;
The transaction conversion data refers to commodity browsing behavior data of a user finally completing purchase, wherein the data comprises records of the user finally completing the transaction through purchase or other conversion behaviors (such as bill, payment and the like) after browsing a certain commodity, the characteristics are generally directly related to the follow-up behavior types (such as purchase) and mean that the user makes a purchase decision after browsing the commodity, the intention purchasing data refers to the fact that the user generates clear purchase intention when browsing the commodity but does not complete the transaction, and the intention purchasing data generally represents the behavior of adding the commodity into a shopping cart or favorites and the like, which means that the user has a certain interest in the commodity and possibly completes the purchase at a future moment, and the intention is that the user possibly adds the commodity into the shopping cart or marks the commodity as like after browsing the commodity, but does not immediately complete the purchase.
Browsing loss data refer to data of actions of a user which do not continue other interactions after browsing a commodity, such as adding the commodity into a shopping cart, not conducting other deep browsing, and the like, and finally not completing purchasing, wherein the data show that the interest of the user in the commodity is not maintained or converted into purchasing action, the user possibly loses interest in the browsing process, loss is caused, the browsing action of the user is not further deepened (such as stay time of the commodity, low page rolling depth, and the like) and finally no conversion is caused, and invalid skip data refer to data of actions of the user which do not conduct any interaction or deep browsing when the user browses the commodity and are shown as not having any substantial interest in the commodity, and the data generally comprise extremely short stay time and extremely low page rolling depth, and the user hardly interacts with the page or stays on the commodity page and belongs to invalid commodity browsing data.
The specific step of classifying and screening the commodity browsing data comprises the following steps:
Acquiring corresponding original behavior data and user interaction characteristics aiming at each piece of commodity browsing data, wherein the original behavior data comprises, but is not limited to, a user-commodity identification, a browsing time stamp and a follow-up behavior type;
in the commodity browsing data sorting and screening process, the original behavior data and the user interaction characteristics are comprehensive records of the interaction process of the user with the commodity on the platform, and the comprehensive records help analyze the interests of the user in the commodity, the behavior patterns and whether purchasing or other conversion behaviors are generated finally.
The user-commodity identification is used for uniquely identifying the interaction relation between a user and a certain commodity, each commodity browsing data can be bound with a specific user and a specific commodity, and the combination of the user ID and the commodity ID is generally used for ensuring that the behavior of a specific user when browsing a certain commodity can be uniquely identified.
The browsing time stamp records a specific time point when a user starts browsing goods, the time stamp provides a time reference for subsequent analysis and helps to analyze whether the behavior of the user is concentrated in certain time periods, and the time stamp usually represents the number of seconds from a certain starting time in milliseconds or seconds.
The follow-up behavior types record the next behavior of the user after browsing the commodity, and the behaviors can include browsing completion (namely, the user directly leaves after browsing a certain commodity), purchasing (namely, the user adds the commodity to a shopping cart), purchasing (namely, the user completes purchasing), and the like.
The user interaction characteristics reflect the deep interaction condition of the user when browsing the commodity, help evaluate the interest intensity of the user on the commodity, are helpful for understanding the potential motivation of the user behavior, and the commodity stay time represents the actual stay time of the user when browsing a certain commodity page. This feature may reveal that a user's interest in the merchandise is a longer dwell time, which typically indicates that the user has a higher interest in the merchandise, a dwell time, which typically is recorded in seconds, e.g., 120, which indicates that the user has stayed on the merchandise page, a page scroll depth, which records the user's scroll through the merchandise page, reflecting whether the user is looking deep into other portions of the merchandise page, a greater scroll depth, which may indicate that the user has a more complete understanding of the merchandise, a page scroll depth, which typically is a percentage, which indicates the progress of the user scrolling through the page, e.g., 70% which indicates that the user has browsed through the page, which may be calculated by tracking the action of the page scroll bar, or by analyzing the user's page interaction data.
According to the follow-up behavior type, directly dividing corresponding commodity browsing data into transaction conversion data, intention purchasing data and residual commodity browsing data;
The method comprises the steps of checking follow-up behavior types of each commodity browsing data, directly dividing the data into transaction conversion data if the follow-up behavior types are purchase or similar transaction behaviors, dividing the data into intention purchasing data if the follow-up behavior types are intention behaviors such as shopping carts, collection and the like, and dividing the browsing data which do not immediately generate purchase or purchasing behaviors into residual commodity browsing data if the follow-up behavior types are browsing or jumping and no further interaction exists, wherein the residual data are further analyzed into invalid jumping data or browsing loss data according to interaction depth scores.
The method comprises the steps of obtaining semantic relevance scores among recently browsed commodities of a user for residual commodity browsing data, carrying out quantitative scores based on commodity stay time and page rolling depth to obtain browsing effective scores, carrying out weighted calculation on the semantic relevance scores and the browsing effective scores to obtain interaction depth scores;
The semantic relevance score aims at measuring similarity or relevance between commodities recently browsed by a user to reflect the interest of the user to related commodities, and specifically comprises the steps of obtaining a commodity list recently browsed by the user, wherein the commodities can be identified through commodity IDs or commodity attributes, analyzing contents such as descriptions, titles and classification labels of the commodities through Natural Language Processing (NLP) technology, calculating the semantic similarity between every two commodities, and obtaining a semantic relevance score between a group of commodities based on the similarity calculation method, wherein the higher the score value is, the stronger the semantic relevance between the commodities is indicated, and the interest of the user is concentrated on the related commodities.
The browsing effective scoring is to evaluate browsing depth and quality based on the stay time and page rolling depth of the user on the commodity page, and the specific steps are to obtain stay time and page rolling depth of each commodity browsing data, to set a reasonable threshold for stay time, to be browsing under which browsing may be considered incomplete (if browsing is considered invalid if browsing is conducted for 10 seconds), to be higher than the threshold, to indicate that the user has higher interest in the commodity page, to roll more than a certain proportion of users, to indicate that they have higher interest in the commodity content, to convert them into a unified browsing effective scoring by combining the two factors, and to adjust the weight of stay time and page rolling depth according to practical conditions.
The interaction depth score is obtained by comprehensively evaluating the interaction depth of a user by combining the semantic relevance score and the browsing effective score, and specifically comprises the steps of carrying out normalization processing on the semantic relevance score and the browsing effective score to ensure that the score ranges of the semantic relevance score and the browsing effective score are consistent, distributing weight values for the semantic relevance score and the browsing effective score, wherein the weight values can be set according to business requirements, for example, the semantic relevance score weight is 0.6, the browsing effective score weight is 0.4, and calculating the interaction depth score according to a weighted average method, namely calculating according to the formula that the interaction depth score= (semantic relevance score multiplied by 0.6) + (browsing effective score multiplied by 0.4), and specifically adjusting the weight values by people.
The method comprises the steps of dividing the browsing data of the residual commodities into invalid skip data if the interaction depth score is lower than a set browsing threshold value, dividing the browsing data of the residual commodities into browsing loss data if the interaction depth score is lower than the set browsing threshold value, otherwise, dividing the browsing data of the residual commodities into browsing loss data, setting a specific numerical value of the browsing threshold value by manpower, calculating the interaction depth score of each piece of browsing data of the residual commodities, wherein the score combines a semantic relevance score and a browsing effective score, comparing the semantic relevance score with the browsing effective score with the artificially set browsing threshold value, if the interaction depth score is lower than the set browsing threshold value, indicating that the interests and participation degree of the user on the commodities are lower, dividing the piece of data into the invalid skip data, indicating that the user does not conduct effective browsing interaction, and rapidly leaves a page, and if the interaction depth score is higher than or equal to the set browsing threshold value, indicating that the user has stronger interests on the commodities but fails to be finally converted into purchasing behavior or purchasing behavior, dividing the piece of data into browsing loss data, indicating that the user has a certain interests on the commodities, but does not continue to conduct subsequent operations, and loss.
Analyzing based on the intercourse transformation data, extracting an induced preferential data set corresponding to each intercourse transformation data, analyzing based on the induced preferential data set to obtain the user purchase intention characteristic, and comparing and analyzing based on the intention purchase data and the browsing loss data to obtain the user browsing intention characteristic;
the specific steps for user behavior analysis based on the induced preference data set comprise:
The method comprises the steps of constructing an induced discount data set by associating one or more pieces of induced discount data received before a transaction with each piece of the induced discount data from the transaction conversion data, wherein the induced discount data set comprises at least one piece of received discount data and zero or more pieces of refused discount data, calculating a purchase inflection point characteristic vector in the induced discount data set, wherein the purchase inflection point characteristic vector specifically comprises a maximum discount amount difference of the received discount data compared with all pieces of refused discount data, time from the first time of receiving the induced discount data to the final time of receiving the discount data, total times of refusing the discount data before the reception of the induced discount data, finding one or more pieces of induced discount data received before the transaction for each piece of the transaction conversion data from the transaction conversion data, possibly associating the plurality of the induced discount data with each piece of the induced discount data, and constructing an induced discount data set by the step, wherein each piece of records at least one piece of received discount data and zero or more pieces of refused discount data, the information received by a user before the transaction and the response condition of the received by the user, and records that the received discount data is refused by the final user.
The induced offer data refers to different types of offer information provided by the platform for the user before the user completes the exchange conversion and the response condition of the user to the offers, and specifically, each induced offer data records a certain offer and a selection behavior thereof received by the user before the exchange.
For example, assuming that the user receives two offers during shopping, a first 100-element minus 10-element offer and a second 200-element minus 50-element offer, the user eventually accepts the second offer, i.e., selects the 200-element minus 50-element offer, while rejecting the first offer, the induced offer data set would contain two records, one accepted offer (200-element minus 50-element) and the other rejected offer (100-element minus 10-element), which help analyze how the user makes the selection during shopping, and which offers more effectively motivate the user to complete the purchase.
The method comprises the steps of calculating a purchase inflection point feature vector in an induced offer data set, wherein the purchase inflection point feature vector is an index for measuring key behavior change of a user in the process of receiving the induced offer data, calculating the maximum offer amount difference value of the received offer data compared with all the rejected offer data, wherein the difference value can help identify which offers have larger influence on a user purchase decision, calculating the time from the first time of receiving the induced offer data to the final time of receiving the offered data, reflecting the time span of selecting among a plurality of offers of the user, calculating how many times of offered data the user refuses before receiving the induced offer data, and evaluating the selection process of the user and the decision mode of the preferential acceptance, wherein the characteristics form the purchase inflection point feature vector and are key quantization indexes for the user behavior.
For example, when the purchase inflection point feature vector is calculated in the induced coupon data set, the maximum coupon amount difference of the accepted coupon data compared with all the rejected coupon data is calculated, and it is assumed that two coupons received by the user are 100-element minus 10-element and 200-element minus 50-element respectively, and the user finally accepts the 200-element minus 50-element coupon and rejects the 100-element minus 10-element. The maximum offer amount difference is then 50 yuan (the difference between a full 200 yuan minus 50 yuan offer and a full 100 yuan minus 10 yuan offer), the time from the first time the user receives the induced offer data to the final time the user receives the offer data is calculated, the user receives the full 100 yuan minus 10 yuan offer for the first time on 1 day of xx year x month until the full 200 yuan minus 50 yuan offer is received on 5 days of xx year x month, the time span is 4 days, the number of times the user refuses the offer data before receiving the final offer is calculated, the total number of times the offer data is refused is 3 if the user refuses 3 different offer data in the decision process, and through the steps, the purchase inflection point feature vector comprises the maximum offer amount difference (50 yuan), the time span (4 days) and the number of times the refused offer (3 times), the features help evaluate the behavior mode of the user making the purchase decision in the process of receiving the offer, and further analyze which factors have significant influence on the purchase decision.
The method comprises the steps of combining purchase inflection point feature vectors corresponding to all the transaction conversion data to obtain a purchase inflection point feature space, clustering the purchase inflection point feature space by using an unsupervised clustering algorithm to obtain a plurality of user intention type clusters, and semantically labeling each user intention type cluster to obtain user purchase intention features;
The method comprises the steps of combining purchase inflection point feature vectors corresponding to all the transaction conversion data of the same user on different commodities to form a purchase inflection point feature space, reflecting behavior features of the user on different commodities in the process of receiving induced preference data, clustering the feature space by using an unsupervised clustering algorithm for better understanding of the behavior patterns and the purchase intentions of the user, wherein a common clustering algorithm such as K-means or DBSCAN is adopted, in this case, the clustering divides the behavior patterns of the same user on a plurality of commodities into a plurality of different intention clusters, each cluster represents a specific behavior feature of the user on a certain type of commodity in a purchase decision, semantically labeling the intention clusters of the user on different commodities, and giving a descriptive label to each cluster by analyzing the common behavior and the features of the user in similar commodity categories and combining the actual commodity features and the purchase intentions of the user, so as to represent the purchase intentions of the user.
For example, some clusters may represent offer-sensitive users that make purchasing decisions more easily in the face of higher discounts or large offers, especially on higher priced items, while other clusters may represent lower-priced pursuit users that prefer to accept certain offers in the face of lower priced, moderately priced items, in this way, the user's intent-to-purchase characteristics are ultimately obtained, in combination with the characteristics of the items, providing data support for subsequent accurate item recommendations, personalized recommendations, and user behavior predictions, helping the platform to understand the user's purchasing preferences and decision process.
The specific steps for performing comparative analysis based on the intention purchasing data and the browsing loss data comprise the following steps:
the method comprises the steps of obtaining intention purchasing data, obtaining corresponding purchasing commodity characteristic data, analyzing browsing loss data to obtain corresponding lost commodity characteristic data, calculating similarity of the purchasing commodity characteristic data and the lost commodity characteristic data, combining the intention purchasing data with the similarity being larger than a commodity similarity threshold value with the browsing loss data to obtain a dual behavior-commodity data set, wherein the commodity similarity threshold value is set by people;
The method comprises the steps of carrying out independent analysis on two types of data, extracting corresponding commodity characteristic data, regarding to intention purchasing data, paying attention to commodity characteristics that a user adds commodities to a shopping cart but does not purchase, for example, the types, prices, discount rates, brands, evaluations and the like of the commodities can be used as characteristic data of the purchasing commodities, meanwhile, regarding browsing loss data, analyzing commodity characteristics that the user does not continue to interact or does not purchase after browsing commodity pages, browsing the lost commodity characteristics can comprise attractive force (such as picture quality and page loading speed) of the commodities, user residence time, page rolling depth and the like, carrying out analysis on the two types of data, respectively extracting the purchasing commodity characteristic data and the lost commodity characteristic data, calculating the similarity between the purchasing commodity characteristic data and the lost commodity characteristic data, and combining the purchasing data and the lost commodity characteristic data, wherein the similarity is greater than a preset commodity similarity threshold, so as to form a dual behavior-commodity data set.
Extracting features of the dual behavior-commodity data set to obtain a composite interaction feature vector;
The composite interaction feature vector comprises three parts, namely a behavior signature vector which quantifies microscopic interaction behaviors of a user when browsing goods, wherein the microscopic interaction behaviors generally comprise information exploration depth (such as depth or stay time of browsing pages), visual confirmation intensity (such as the number of times of page scrolling or picture enlarging), social confirmation seeking degree (such as behavior of checking comments, sharing or scoring by referring to others), a goods static feature vector which directly comprises intrinsic feature data of the goods, such as information of goods category, price, brands, discounts, evaluation and the like, and finally behavior-goods interaction features which are generated by carrying out cross operation on dimensions in the behavior signature vector and dimensions in the goods static feature vector, and describe interaction features of specific behaviors occurring in specific goods background, and a group of complete composite interaction feature vector is formed through the features.
Taking the composite interaction feature vector as input, and taking the corresponding contribution data type as a target label to construct a commodity importance analysis model;
and carrying out intention rule analysis on all the composite interaction feature vectors with higher contribution degree to obtain browsing intention features of the user.
In the commodity importance analysis model, the contribution degree represents the influence degree of each composite interaction feature vector on the prediction of a target label (such as purchase, loss and the like), the composite interaction feature vector with higher contribution degree means that the relationship between the feature and the behavior (such as purchase or loss) of a user is tighter, namely the feature plays a larger role in explaining the behavior decision process of the user, the contribution degree is generally measured through weights or importance scores in the model and represents the contribution of a specific feature to a prediction result, and a feature selection method (such as decision trees, random forests, lasso regression and the like) is used for calculating the contribution of each composite interaction feature vector to the prediction of the target behavior. Feature vectors with higher contributions will have greater impact in the model, helping us identify which features or behavior patterns are most relevant to the user's final purchasing, purchasing or churn behavior.
The target tag is output data of model learning, represents a behavior result or target behavior of a user, is directly related to actual behavior (such as purchasing, purchasing and losing) of the user in commodity importance analysis, and is defined according to the analyzed behavior type, for example, the corresponding contribution data type is set as a purchasing tag, namely, whether the user adds the commodity to a shopping cart after browsing the commodity or not, and is marked with 1 to indicate that the commodity is purchased by the user and 0 to indicate that the commodity is not purchased.
Assuming that the browsing intention of a user when browsing two identical clothes is to be analyzed, the composite interaction feature vector is input into a commodity importance analysis model, and the user behavior is analyzed according to different target labels (shopping or losing). For example, there are two types of clothing, red and black, and the user's browsing behavior may be different for each type of clothing. For example, a user may show a strong purchasing intent for a red style of clothing, but may show fluid loss for a black style of clothing. The behavior of the user can be analyzed through the model to conclude that for clothes of red style, the contribution degree between the composite interaction feature vector of the user (such as longer stay time, large page scrolling depth, frequently viewing comments and the like) and the shopping tag is higher, which indicates that the user has strong interest in the red style and finally decides to add the commodity to the shopping cart, and for clothes of black style, the contribution degree of the user is higher than that of the loss tag although the behavior feature of the user is similar (such as longer stay time), and the possible reason is that although the user browses the clothes of black style for a longer time, the user does not purchase at all due to certain factors (such as color preference, lower discount strength and the like) and leaves. From these analyses, it can be inferred that the user prefers a red style of clothing because the red style has a stronger correlation with the purchasing behavior, while the black style is more likely to lead to user churn, possibly related to the user's color preferences or other shopping decision factors (e.g., discounts, recommendations, etc.), by which, in combination with the analysis of the composite interaction feature vector and the target label, the merchandise importance analysis model can help us identify the user's preferences and optimize merchandise display, personalized recommendations, and marketing strategies. For example, the platform can conduct personalized recommendation on different types of clothes according to browsing and purchasing behaviors of the user, and the conversion rate is further improved.
And recommending commodities to the user in a prediction period according to the user purchase interest evolution model.
The specific steps for establishing the user purchase interest evolution model based on the user purchase intention characteristics and the user browsing intention characteristics comprise the following steps:
performing differential calculation on the user browsing intention characteristic and the user purchasing intention characteristic to obtain a purchasing conversion gain sequence;
Sequencing the browsing intention features of each user and the purchasing intention features of the user according to the sequence of the browsing time stamp to obtain a purchasing interest evolution sequence of the user;
The user purchase interest evolution model comprises a content interest evolution channel and a preferential sensitivity evolution channel, wherein the content interest evolution channel is trained based on a user purchase interest evolution sequence, the output is a predicted target commodity set in the next period, the preferential sensitivity evolution channel is trained based on a purchase conversion gain sequence, and the output is a predicted preferential sensitivity state in the next period.
In the process of constructing the user purchase interest evolution model, firstly, differential calculation is carried out on the user browsing intention characteristic and the user purchase intention characteristic to obtain a purchase conversion gain sequence, wherein the browsing intention characteristic reflects the interest of the user when browsing commodities, the purchase intention characteristic reflects the additional purchase or the purchase intention of the user, and the purchase conversion gain of the user, namely the degree of conversion of the user into the purchase intention after browsing the commodities, can be measured through the differential calculation on the two types of characteristics. For example, if a user has a strong browsing interest in a commodity, but does not show enough buying intention in the subsequent conversion process, the buying conversion gain value is lower, otherwise, if the user shows a strong buying intention in the browsing process, the gain value is higher, the result of the differential calculation is a gain sequence, which helps us identify which commodities have a larger pushing effect on the buying conversion of the user, especially under the effect of inducing the preferential data, the buying conversion gain sequence can further reflect the influence of the preferential strategy on the conversion.
Ordering the browsing intention features and the purchasing intention features of each user according to the sequence of the browsing time stamp to form a user purchasing interest evolution sequence, wherein the step aims at simulating the interest evolution process of the user and reflecting the dynamic change of the user from preliminary browsing to final purchasing intention. By sorting the user behaviors according to the browsing time stamps, the interests and intention change tracks of the user can be better captured. For example, a user may generate a strong browsing interest for a commodity in a certain period, and over time, the browsing interest may be gradually converted into a purchasing intention due to the influence of certain sales promotion activities or preferential strategies, and the finally obtained evolution sequence of the purchasing interest of the user may be used for analyzing the variation trend of the user interest over time, so as to provide basis for subsequent personalized recommendation and marketing strategies.
The user purchase interest evolution model comprises a content interest evolution channel and a preferential sensitivity evolution channel. The content interest evolution channel is mainly trained based on the purchase interest evolution sequence of the user, and aims to predict the commodity set possibly interested by the user in the next period. By analyzing the purchase interest evolution trajectory of the user over the past period, the model is able to learn which commodity types or features are likely to attract the user again in the future. For example, a user historically shows a strong interest in sports shoe merchandise, and the model may predict that the user may continue to be interested in similar sports shoe merchandise in future periods, and thus the content interest evolution channel can help the platform make more accurate merchandise recommendations in future periods.
The preferential sensitivity evolution channel is trained based on the purchase conversion gain sequence, so that the sensitivity state of the user to the preferential is predicted in the next period, and the model can identify which preferential plays a key role in the purchase decision of the user by analyzing the purchase conversion gain sequence of the user in the past period. For example, some users may be very sensitive to a large discount, while other users may exhibit a weaker sensitivity to the discount during the promotional program. Based on the gain sequences, the preferential sensitivity evolution channel can predict the response of a user to preferential activities in a future period, thereby providing basis for formulating accurate marketing strategies, such as adjusting preferential amplitude, preferential category or pushing mode.
The constructed user purchase interest evolution model can pay attention to the dynamic change of the user on the content preference and the preferential sensitivity at the same time, and help the platform to more accurately predict commodity interests and purchase intentions of the user in a future period, so that a recommendation algorithm and a marketing strategy are optimized, and the conversion rate and loyalty of the user are improved.
The method comprises the specific steps of recommending commodities to a user in a prediction period according to a user purchase interest evolution sequence, and comprises the following steps:
Inputting the behavior sequence into a user purchase interest evolution model for double-channel analysis to obtain a recommended commodity list and a personalized induced preference strategy matched with the recommended commodity list;
and recommending the commodity for the user by utilizing the recommended commodity list and the personalized induced preferential strategy matched with the recommended commodity list.
In the process of recommending commodities to a user in a prediction period according to a user purchase interest evolution sequence, the first step is to acquire a behavior sequence of the user in the prediction period, wherein the process involves collecting and sorting all interactive behavior data of the user in the prediction period, generally comprises browsing records, purchasing behavior, interaction conditions with a sales promotion activity and the like of the user on a commodity page, and the behavior sequence can provide detailed data about user interest changes and preferences by tracking the interaction between the user and the commodities so as to help a model to know the current demands and potential purchase intentions of the user, and the behavior sequence provides key information input for subsequent recommendation. The behavior sequence of the user is input into a user purchase interest evolution model for dual-channel analysis, wherein the user purchase interest evolution model comprises two channels, namely a content interest evolution channel and a preferential sensitivity evolution channel.
In the content interest evolution channel, the model analyzes the interest evolution trend of the user, predicts the commodity type or specific commodity which is most likely to be interested in the user in the prediction period through the past behavior track, and the process is based on the historical browsing, purchasing and purchasing records of the user and the preference change of the user, so that the model is helped to judge which commodities possibly arouse the interest of the user and promote the purchase. Meanwhile, in the preferential sensitivity evolution channel, the model analyzes the response of the user to preferential activities in the past period, such as discount, full-reduction and other promotion strategies, predicts which preferential forms the user is likely to be sensitive to in the prediction period, and the joint analysis of the two channels can not only recommend commodities for the user, but also match corresponding personalized preferential strategies for each recommended commodity.
And by utilizing the recommended commodity list obtained from the double-channel analysis and the personalized induced preference strategy matched with the recommended commodity list, accurate commodity recommendation is provided for the user. When recommending commodities, the commodities conforming to the preferences of the users are pushed according to the current interests and the preferential sensitivity of the users, and matched personalized preferential strategies are provided together.
For example, a user may be recommended a brand of athletic shoes that have recently been viewed by the user, and be motivated to complete the purchase with an exclusive discount or coupon for that brand. Through the steps, highly personalized commodity recommendation and discount inducing strategies can be provided for each user in a prediction period, so that user experience is improved, and conversion and sales of a platform can be effectively promoted.
In this embodiment, it is assumed that personalized commodity recommendation and promotion of an induced preference policy are being performed for a user of a certain e-commerce platform. In the prediction period, firstly, a behavior sequence of the user is acquired, wherein the behavior sequence comprises browsing records, purchasing behavior and purchasing behavior in the past 30 days, for example, the user browses 100 commodities in the past 30 days, 20 commodities are added to a shopping cart, and finally 5 commodities are purchased, the behavior sequences are input into a user purchasing interest evolution model for dual-channel analysis, in a content interest evolution channel, the model analyzes browsing and purchasing tracks of the user, predicts that the user has higher interests on 'sports shoes' and 'fitness equipment' commodities in the future period, recommends 10 commodities meeting the interests, and in a preferential sensitivity evolution channel, the model identifies that the user is sensitive to preferential activity response of less than 50 yuan, and therefore the preferential strategy is matched with the recommended commodities. Finally, the recommendation system provides a recommendation list containing 10 commodities for the user, and a personalized preferential strategy is matched for each commodity, for example, a user can enjoy a preference of full 200 minus 50 yuan when buying sports shoes, so that the user's purchasing desire is stimulated. In the process, the composite interaction feature vector helps to analyze the interest change of the user, and the purchasing conversion gain sequence further optimizes the matching of the commodity and the preferential, so that the accuracy and the effectiveness of recommendation are ensured.
Embodiment 2, referring to fig. 1, an intelligent commodity recommendation system based on user behavior analysis includes:
The system comprises a behavior analysis module, a comparison analysis module and a comparison analysis module, wherein the behavior analysis module comprises a purchase behavior analysis unit and a browsing behavior analysis unit, the purchase behavior analysis unit is used for acquiring a plurality of commodity browsing data of a user history in the previous period, classifying and screening the commodity browsing data to obtain transaction conversion data, intention purchase data, browsing loss data and invalid jump data, the browsing behavior analysis unit is used for analyzing based on the transaction conversion data, extracting an induced preference data set corresponding to each transaction conversion data, carrying out user behavior analysis based on the induced preference data set to obtain the purchase intention characteristic of the user, and carrying out comparison analysis based on the intention purchase data and the browsing loss data to obtain the browsing intention characteristic of the user;
The commodity recommending module comprises an intelligent recommending unit and is used for establishing a user purchase interest evolution model based on user purchase intention characteristics and user browsing intention characteristics, and recommending commodities to the user in a prediction period according to the user purchase interest evolution model.
It will be understood that modifications and variations will be apparent to those skilled in the art from the foregoing description, and it is intended that all such modifications and variations be included within the scope of the following claims. Parts of the specification not described in detail belong to the prior art known to those skilled in the art.
Claims (7)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510969220.3A CN120494942B (en) | 2025-07-15 | A method and system for intelligent product recommendation based on user behavior analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202510969220.3A CN120494942B (en) | 2025-07-15 | A method and system for intelligent product recommendation based on user behavior analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN120494942A true CN120494942A (en) | 2025-08-15 |
| CN120494942B CN120494942B (en) | 2025-10-10 |
Family
ID=
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030105682A1 (en) * | 1998-09-18 | 2003-06-05 | Dicker Russell A. | User interface and methods for recommending items to users |
| CN105117418A (en) * | 2015-07-30 | 2015-12-02 | 百度在线网络技术（北京）有限公司 | Search based service information management system and method |
| US9953358B1 (en) * | 2014-12-08 | 2018-04-24 | Amazon Technologies, Inc. | Behavioral filter for personalized recommendations based on behavior at third-party content sites |
| CN112464094A (en) * | 2020-11-30 | 2021-03-09 | 泰康保险集团股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
| CN117391820A (en) * | 2023-12-01 | 2024-01-12 | 深圳市思迅网络科技有限公司 | SaaS service comprehensive management method and system |
| CN118052618A (en) * | 2024-03-15 | 2024-05-17 | 大白鲸信息科技(浙江)有限公司 | E-commerce platform snack type pushing system based on big data analysis |
| CN119273431A (en) * | 2024-12-09 | 2025-01-07 | 青岛惊喜优品科技有限公司 | Smart shopping system with recommendation ranking function |
| CN120013643A (en) * | 2025-01-22 | 2025-05-16 | 智之锚(苏州)科技有限公司 | Customer behavior prediction and automated response system based on artificial intelligence |
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030105682A1 (en) * | 1998-09-18 | 2003-06-05 | Dicker Russell A. | User interface and methods for recommending items to users |
| US9953358B1 (en) * | 2014-12-08 | 2018-04-24 | Amazon Technologies, Inc. | Behavioral filter for personalized recommendations based on behavior at third-party content sites |
| CN105117418A (en) * | 2015-07-30 | 2015-12-02 | 百度在线网络技术（北京）有限公司 | Search based service information management system and method |
| CN112464094A (en) * | 2020-11-30 | 2021-03-09 | 泰康保险集团股份有限公司 | Information recommendation method and device, electronic equipment and storage medium |
| CN117391820A (en) * | 2023-12-01 | 2024-01-12 | 深圳市思迅网络科技有限公司 | SaaS service comprehensive management method and system |
| CN118052618A (en) * | 2024-03-15 | 2024-05-17 | 大白鲸信息科技(浙江)有限公司 | E-commerce platform snack type pushing system based on big data analysis |
| CN119273431A (en) * | 2024-12-09 | 2025-01-07 | 青岛惊喜优品科技有限公司 | Smart shopping system with recommendation ranking function |
| CN120013643A (en) * | 2025-01-22 | 2025-05-16 | 智之锚(苏州)科技有限公司 | Customer behavior prediction and automated response system based on artificial intelligence |
Non-Patent Citations (2)
| Title |
|---|
| MANASA DESHPANDE ET AL.: "A Personalized Supermarket Product Recommendation System Using Augmented Reality", 《ECS TRANSACTIONS》, vol. 101, no. 1, 31 December 2022 (2022-12-31), pages 915 - 923 * |
| 毕洁昊: "基于用户感知的个性化推荐系统推荐效果评价", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》, 15 March 2023 (2023-03-15), pages 138 - 658 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102213768B1 (en) | Customer-specific product recommendation system that exposes products with high purchase conversion rate based on customer information by artificial intelligence based on big data | |
| CN113077317B (en) | Item recommendation method, device, equipment and storage medium based on user data | |
| CN112435067A (en) | Intelligent advertisement putting method and system for cross-e-commerce platform and social platform | |
| CN116894709A (en) | Advertisement commodity recommendation method and system | |
| Sharma et al. | Impacts of customer feedback for online-offline shopping using machine learning | |
| Wong et al. | Identifying valuable travelers and their next foreign destination by the application of data mining techniques | |
| CN112967116A (en) | Multi-strategy-pool-based off-line retail store commodity recommendation method and system | |
| Khodabandehlou | Designing an e-commerce recommender system based on collaborative filtering using a data mining approach | |
| Zheng et al. | A scalable purchase intention prediction system using extreme gradient boosting machines with browsing content entropy | |
| CN115809926A (en) | Innovative alliance distribution cooperation transaction platform for converging alliance resources of multiple distribution channels | |
| CN119027165A (en) | Commodity marketing method and system based on artificial intelligence | |
| CN118822700B (en) | A commodity search method based on big data and electronic shopping mall | |
| Chou et al. | The RFM Model Analysis for VIP Customer: A case study of golf clothing brand | |
| CN117035947B (en) | Agricultural product data analysis method and cloud platform based on big data processing | |
| Zuo | Prediction of consumer purchase behaviour using Bayesian network: an operational improvement and new results based on RFID data | |
| JP6910515B1 (en) | Analytical instruments, analytical methods and analytical programs | |
| KR20220014543A (en) | The measures and systems of online platform that connects both sides in a virtuous cycle through voluntary participation of users and developers. | |
| CN120494942B (en) | A method and system for intelligent product recommendation based on user behavior analysis | |
| CN120494942A (en) | A method and system for intelligent product recommendation based on user behavior analysis | |
| JP6910516B1 (en) | Analytical instruments, analytical methods and analytical programs | |
| Xiaona | Personalized recommendation model for mobile e-commerce users | |
| Gao | User Behavior Analysis and Prediction Algorithm Design of E-commerce Platform Based on Big Data Analysis | |
| KR20220120984A (en) | Goods analysis and recommendation system | |
| Chate | Behavioral modelling of customer marketing patterns and review prediction using machine learning techniques | |
| Maurya et al. | Analysis of behavioural data of customer for the E-commerce platform by using machine learning approach |
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 |