Information pushing method and device, computer equipment and storage medium

Abstract

The invention discloses an information pushing method and device, computer equipment and a storage medium, and the method comprises the steps: determining a service matching score of each candidate service in a plurality of candidate services based on target interaction data between a target user and the plurality of candidate services; obtaining a popularity score and an entity score of each candidate service in the plurality of candidate services, wherein the entity score is used for representing the density degree of an entity corresponding to the candidate service in the knowledge graph corresponding to the plurality of candidate services; obtaining a user score of the target user, wherein the user score is used for representing the degree of conformity of the target user; determining a recommendation score corresponding to each candidate service based on the service matching score, the popularity score, the user score and the entity score; and based on the recommendation score corresponding to each candidate service, pushing information corresponding to a target service in the plurality of candidate services to an electronic device corresponding to the target user. The method can improve the accuracy of service recommendation.

Description

Technical Field

[0001] This application relates to the field of computer technologies, and more particularly, to an information push method, apparatus, computer device, and storage medium. Background Art

[0002] With the rapid progress of the technological level and living standard, electronic devices (such as smart phones, tablet computers, etc.) have become one of the commonly used electronic products in people's lives. And currently, general electronic devices have an information push function, and through the information push function, services that users are interested in can be recommended. However, in related technologies, the accuracy of service recommendation is insufficient. Summary of the Invention

[0003] This application provides an information push method, apparatus, computer device, and storage medium, which can improve the accuracy of service recommendation.

[0004] In a first aspect, an embodiment of this application provides an information push method, where the method includes: determining a service matching score for each candidate service among the multiple candidate services based on target interaction data between a target user and the multiple candidate services, where the service matching score is used to characterize the matching degree between the candidate service and the target user; obtaining a popularity score and an entity score for each candidate service among the multiple candidate services, where the popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in a knowledge graph corresponding to the multiple candidate services; obtaining a user score of the target user, where the user score is used to characterize the conformity degree of the target user; determining a recommendation score corresponding to each candidate service based on the service matching score, popularity score, user score, and entity score; and pushing information corresponding to a target service among the multiple candidate services to an electronic device corresponding to the target user based on the recommendation score corresponding to each candidate service.

[0005] Second aspect, an embodiment of the present application provides an information push device, the device includes: a first score acquisition module, a second score acquisition module, a third score acquisition module, a fourth score acquisition module, and a service recommendation module, wherein, the first score acquisition module is configured to determine a service matching score for each candidate service among the multiple candidate services based on the target interaction data between the target user and the multiple candidate services, and the service matching score is used to characterize the matching degree between the candidate service and the target user; the second score acquisition module is configured to acquire the popularity score and the entity score for each candidate service among the multiple candidate services, the popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services; the third score acquisition module is configured to acquire the user score of the target user, and the user score is used to characterize the herd mentality of the target user; the fourth score acquisition module is configured to determine a recommendation score corresponding to each candidate service based on the service matching score, the popularity score, the user score, and the entity score; the service recommendation module is configured to push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user based on the recommendation score corresponding to each candidate service.

[0006] Third aspect, an embodiment of the present application provides a computer device, including: one or more processors; a memory; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, and the one or more application programs are configured to execute the information push method provided in the first aspect above.

[0007] Fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which program code is stored, and the program code can be called by a processor to execute the information push method provided in the first aspect above.

[0008] The solution provided by this application determines the service matching score of each candidate service among multiple candidate services based on the target interaction data between the target user and the multiple candidate services. The service matching score is used to characterize the matching degree between the candidate service and the target user; obtains the popularity score and entity score of each candidate service among the multiple candidate services. The popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services; obtains the user score of the target user, and the user score is used to characterize the conformity degree of the target user; based on the service matching score, popularity score, user score, and entity score, determines the recommendation score corresponding to each candidate service; based on the recommendation score corresponding to each candidate service, pushes the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user. Thus, when determining the recommendation score corresponding to each candidate service for the target user, on the basis of considering the matching degree between the candidate service and the target user, it also considers the popularity of the candidate service, the density of the relationship of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services, and the conformity degree of the target user, thereby improving the accuracy of the determined recommendation score, and further improving the accuracy of service recommendation when using the recommendation score corresponding to each candidate service for service recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the following briefly introduces the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of this application. For those skilled in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0010] Figure 1 FIG. shows a flowchart of an information push method according to an embodiment of this application.

[0011] Figure 2 FIG. shows a flowchart of an information push method according to another embodiment of this application.

[0012] Figure 3 FIG. shows a flowchart of an information push method according to still another embodiment of this application.

[0013] Figure 4 FIG. shows a flowchart of an information push method according to yet another embodiment of this application.

[0014] Figure 5 FIG. shows a block diagram of an information push device according to an embodiment of this application.

[0015] Figure 6It is a block diagram of a computer device for executing the information push method according to an embodiment of the present application.

[0016] Figure 7 It is a storage unit for storing or carrying program codes for implementing the information push method according to an embodiment of the present application. Detailed implementation manners

[0017] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application.

[0018] With the development of the Internet, information push has been widely used. Especially in electronic devices, the recommendation of services that users are interested in can be realized through the information push function. For example, application programs, commodities, videos, music, pictures, etc. can be recommended. When implementing service recommendation, it needs to be based on massive data mining. By analyzing the historical behaviors and user profiles of each user, the services that the user is most likely to be interested in can be recommended to the user.

[0019] In the related art, the interaction data between users and services is widely used in a large number of user data. Through the bipartite graph between users and services, the matching evaluation scores between users and services are statistically calculated, and recommendations are made according to the most matching scores. The collaborative filtering algorithm has become a classic algorithm in the recommendation technology and is still widely used today because of its complete theoretical system, strong interpretability, simple design concept, and relatively ideal effect. This algorithm can be divided into the collaborative filtering algorithm based on personalized services and the collaborative filtering algorithm based on users according to different analysis objects. The method based on personalized services generates a recommended ranking set by calculating the vector similarity between services, while the method based on users constructs a user vector through features such as the browsing records of users, calculates the users most similar to this user, and then recommends the services of the similar users to this user. With the continuous development of various deep learning methods, by jointly encoding users, services, and various auxiliary information, the implicit expressions of user and service features can be obtained, and multi-dimensional deep features can be automatically mined and characterized, so as to improve the effect of the model, and then recommendations are made through the similarity matching scores.

[0020] The interaction data between users and personalized services is widely used in a large number of user data. However, in the interaction data, the frequency distribution of services is not uniform and is affected by many factors. In most cases, the frequency distribution will show a long-tail effect, that is, most of the interaction data is occupied by a small number of popular services. When making service recommendations, the demands of users are usually inferred based on this biased interaction data. Therefore, the user interests are affected by this popularity bias to a certain extent, resulting in insufficient accuracy of service recommendations.

[0021] In view of the above problems, the inventors have proposed an information push method, apparatus, computer device, and storage medium provided in the embodiments of the present application, which can realize that when determining the recommendation score corresponding to each candidate service for a target user, on the basis of considering the matching degree between the candidate service and the target user, the popularity of the candidate service, the density of the entity corresponding to the candidate service in the knowledge graphs corresponding to multiple candidate services, and the conformity degree of the target user are also considered, so as to improve the accuracy of the determined recommendation score, and further improve the accuracy of service recommendation when using the recommendation score corresponding to each candidate service for service recommendation. Among them, the specific information push method will be described in detail in the subsequent embodiments.

[0022] Next, the information push method provided in the embodiments of the present application will be introduced in detail with reference to the accompanying drawings.

[0023] Please refer to Figure 1 , Figure 1 which shows a schematic flowchart of an information push method provided in an embodiment of the present application. In a specific embodiment, the information push method is applied to an information push device 400 as shown in Figure 5 and a computer device 100 configured with the information push device 400( Figure 6 ). Next, taking the computer device as an example, the specific process of this embodiment will be described. Of course, it can be understood that the computer device applied in this embodiment can be a physical server, a cloud server, etc., which is not limited here. Next, the process shown in Figure 1 will be elaborated in detail. The information push method may specifically include the following steps:

[0024] Step S110: Based on the target interaction data between the target user and multiple candidate services, determine the service matching score of each candidate service among the multiple candidate services, where the service matching score is used to represent the matching degree between the candidate service and the target user.

[0025] In the embodiments of the present application, when performing service recommendation for a target user, the matching degree between the candidate service and the target user can be determined to determine the service to be pushed to the target user according to the matching degrees corresponding to different candidate services. Among them, based on the target interaction data between the target user and multiple candidate services, the service matching score of each candidate service among the multiple candidate services can be determined, and the service matching score is used to represent the matching degree between the candidate service and the target user; the service matching score of the candidate service is positively correlated with the above-mentioned matching degree corresponding to it. That is to say, the higher the matching degree between the candidate service and the target user, the higher the service matching score, and the lower the matching degree between the candidate service and the target user, the lower the service matching score.

[0026] In some embodiments, the target interaction data between the target user and multiple candidate services may include the number of interactions, interaction duration, interaction frequency, etc. between the target user and each candidate service. When determining the service matching score of each candidate service, the computer device may use the above interaction data as an indicator to measure the matching degree between the user and the candidate service, and quantify the interaction data to obtain the service matching score. Among them, for the target user, the larger the interaction number, interaction duration, interaction frequency, etc. corresponding to the candidate service, the greater the matching degree between the candidate service and the target user, that is, the higher the service matching score of the candidate service.

[0027] In a possible embodiment, for each candidate service, the ratio between the parameter value in the target interaction data corresponding to the candidate service and the preset parameter value may be obtained to obtain the matching degree corresponding to the candidate service; then, based on the matching degree corresponding to the candidate service, the service matching score of the candidate service is determined. Among them, the preset parameter value may be the average value of the parameter values in the target interaction data between the target user and each candidate service; the preset parameter value may also be the average value of the parameter values in the target interaction data between multiple sample users and each candidate service.

[0028] In some embodiments, the computer device may also determine the service matching score of each candidate service through a pre-trained service matching model and based on the above target interaction data between the target user and multiple candidate services. Among them, for each candidate service, the target interaction data corresponding to each candidate service may be input into the service matching model, so that the service matching score output by the service matching model for the target interaction data corresponding to the candidate service can be obtained. The service matching model may be a tree model, such as a lightGBM model, or a neural network model, such as a deep neural network model, etc. The specific model type of the service matching model may not be limited. When the service matching model realizes outputting the service matching score for the target interaction data, it may extract data features from the target interaction data and then output the corresponding service matching score according to the data features.

[0029] Of course, in the embodiments of the present application, the specific manner in which the computer device determines the service matching score of the candidate service based on the above target interaction data may not be limited.

[0030] Step S120: Obtain the popularity score and entity score of each candidate service among the multiple candidate services, where the popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services.

[0031] In the embodiments of the present application, when a computer device makes service recommendations for a target user, in addition to obtaining the service matching scores that can characterize the matching degree between the candidate services and the target user, it can also obtain the popularity scores and entity scores of each candidate service among multiple candidate services. The popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to multiple candidate services. Among them, the popularity score can be positively correlated with the popularity of the candidate service, that is, the higher the popularity of the candidate service, the higher the popularity score, and the lower the popularity of the candidate service, the lower the popularity score; the density of any entity can characterize the number of relationships with that entity. Similarly, the entity score can also be positively correlated with the above density of the entity corresponding to the candidate service, that is, the higher the popularity corresponding to the candidate service, the higher the popularity score of the candidate service, and the lower the popularity corresponding to the candidate service, the lower the popularity score of the candidate service.

[0032] It can be understood that the popularity of a candidate service can represent the popularity of the candidate service. The higher the popularity, the more popular the candidate service is, and the higher the probability that the user is affected by its popularity; the density of the entity corresponding to the candidate service in the above knowledge graph can represent the influence of the entity in the knowledge graph on service recommendations. The more popular the service, the denser it may be in the knowledge graph. Therefore, while obtaining the above service matching scores, by obtaining the above popularity scores and entity scores, it is possible to more accurately determine the recommendation score corresponding to each candidate service subsequently.

[0033] In some embodiments, the popularity score of each candidate service can be determined based on the sample interaction data between multiple sample users and multiple candidate services. It can be understood that the sample interaction data includes the number of interactions, interaction duration, interaction frequency, etc. between the sample user and the candidate service. The higher these parameter values in the sample interaction data, to a certain extent, can indicate that the candidate service is more popular. Therefore, the popularity score of the candidate service can be determined based on the sample interaction data of multiple sample users.

[0034] In a possible embodiment, based on the sample interaction data between multiple sample users and multiple candidate services, for each candidate service, the total value of the parameter values in the above sample interaction data corresponding to each candidate service can be determined, and then this total value can be quantified as the above popularity score. Taking the number of interactions as an example, the sum of the number of interactions of multiple sample users for each candidate sample user can be obtained, and then this sum can be quantified as the popularity score. For example, the total number of interactions of multiple sample users for all candidate services can be obtained, and then the ratio of the above sum to this total number can be used as the above popularity score.

[0035] In some embodiments, the popularity score of the candidate service can also be determined by obtaining the popularity value corresponding to the candidate service from different servers where the above candidate services are located, and then determining the popularity score of the candidate service according to the obtained popularity value. For example, the obtained popularity value of the candidate service can be directly used as the above popularity score; for another example, the ratio between the obtained popularity value of the candidate service and the target value can be obtained and used as the above popularity score.

[0036] Of course, in the embodiments of the present application, the specific manner in which the computer device determines the popularity score of the candidate service may not be limited.

[0037] In some embodiments, when the computer device obtains the entity score corresponding to each candidate service, it can obtain the knowledge graph corresponding to the above multiple candidate services. The knowledge graph can be established with each candidate service as an entity and according to the relationships between the candidate services; then for each entity in the knowledge graph, determine the number of entities related to the entity to obtain the number of entities corresponding to each entity; then quantify the determined number of entities to obtain the entity score of the candidate service corresponding to the entity. For example, after determining the number of entities related to each entity, the average number of entities corresponding to all entities can be obtained, and then when quantifying the number of entities corresponding to each entity, the ratio between the number of entities corresponding to each entity and the average number can be obtained and used as the above entity score. Of course, the specific manner in which the computer device obtains the above entity score may not be limited, as long as the determined entity score can represent the above density degree of the entity corresponding to the candidate service in the above knowledge graph.

[0038] Step S130: Obtain the user score of the target user, where the user score is used to represent the degree of conformity of the target user.

[0039] In the embodiments of the present application, when the computer device makes a service recommendation for the target user, in addition to obtaining the service matching score that can represent the matching degree between the candidate service and the target user as described above, it can also obtain the user score of the target user, where the user score is used to represent the degree of conformity of the target user. Among them, the user score can be positively correlated with the degree of conformity of the user. That is to say, the higher the degree of conformity of the user, the higher the user score, and the lower the degree of conformity of the user, the lower the user score. It can be understood that the above user score can reflect the degree to which the user will interact with the service regardless of whether the user's preferences match. And considering that two users are randomly recommended the same number of services, one user may click on the service with higher exposure due to broader interests or herd mentality. Therefore, the higher the degree of conformity of the user, the higher the user score.

[0040] In some embodiments, the user score of the target user may be determined based on the user information of the target user. The user information may include the user's identity document number (Identity document, ID), name, etc. The computer device may pre-determine the user scores of different users. Based on this, the computer device may determine the user score corresponding to the user information as the user score of the target user according to the user information. Among them, the computer device may analyze the historical behavior data of the user to mine the similarities and differences between the user and the group behavior. For example, it may analyze data such as the user's search records and browsing records to obtain the degree of conformity of the user in information acquisition, and then determine the above user score according to the degree of conformity.

[0041] In some embodiments, the computer device may also determine the user score of the target user through a pre-trained user prediction model for predicting the user score of the user. The user prediction model may be trained according to the user information marked with different user scores. The user information may include the user's identity document number, name, age, birthday, occupation, educational background, purchase records, browsing records, search records, geographical location, hobbies, etc.; when determining the user score of the target user, the user information of the target user may be obtained, and the user information of the target user may be input into the above user prediction model, so as to obtain the user score of the target user output by the user prediction model.

[0042] Certainly, in the embodiments of the present application, the specific manner for the computer device to determine the user score of the target user may not be limited.

[0043] Step S140: Based on the service matching score, popularity score, user score, and entity score, determine the recommendation score corresponding to each candidate service.

[0044] In the embodiments of the present application, after determining the service matching score, popularity score, entity score of each of the above candidate services and the user score of the target user, the final recommendation score for determining the service to be recommended can be determined based on the service matching score, popularity score, entity score of each candidate service and the user score of the target user. It can be understood that in the related art, the service to be recommended is usually directly determined according to the matching score between the candidate service and the user. Since the user is easily affected by popularity, the accuracy of service recommendation is insufficient. Therefore, on the basis of the matching degree between the candidate service and the target user (i.e., the above service matching score), the popularity of the candidate service (i.e., the above popularity score), the density of the entity corresponding to the candidate service in the knowledge graph corresponding to multiple candidate services (i.e., the above entity score), and the conformity degree of the target user (the above user score) are also considered, so that the finally determined recommendation score is more accurate, and thus the service recommendation based on the recommendation score subsequently will be more accurate.

[0045] In some embodiments, when determining the recommendation score corresponding to each candidate service based on the service matching score, popularity score, entity score of each candidate service and the user score of the target user, the service matching score can be adjusted by using the popularity score, entity score and user score of the target user on the basis of the service matching score, so as to obtain the final recommendation score. Among them, the popularity score, entity score and user score can be mapped to the range of 0 to 1 by using the sigmoid function, and then the product between the service matching score and the values of the popularity score, entity score and user score mapped to the range of 0 to 1 is obtained. Thus, the influence of the popularity of the candidate service, the density of the entity corresponding to the candidate service in the above knowledge graph, and the conformity degree of the target user on the service matching score can be eliminated, so as to adjust the degree of restoring the historical interaction between the user and the candidate service depending on the service matching score. For example, in order to restore the interaction between a user with a low user score and a candidate service with a low popularity score, the above method can be used to amplify the service matching score (relative to directly comparing the service matches of different candidate services). Exemplarily, the recommendation score can be obtained through the following formula:

[0046]

[0047] Among them, represents the recommendation score corresponding to candidate service i, represents the above service matching score, represents the popularity score of candidate service i, represents the above user score, represents the entity score of candidate service i, and σ() represents the sigmoid function.

[0048] Step S150: Based on the recommendation scores corresponding to each candidate service, push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user.

[0049] In the embodiment of the present application, after obtaining the recommendation scores corresponding to each of the above candidate services, the target service to be recommended to the target user can be determined from the multiple candidate services based on the obtained recommendation scores, and then the information corresponding to the target service is pushed to the electronic device corresponding to the target user.

[0050] In some embodiments, when pushing the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user based on the recommendation scores corresponding to each candidate service, the candidate services whose recommendation scores meet the target score condition can be determined according to the recommendation scores corresponding to each candidate service, and used as the above target services to be recommended to the target user. Among them, the target score condition can be that the recommendation score is greater than the target score, or the recommendation scores are ranked among the top N in the descending order, where N is a positive integer, such as 3, 5, 10, etc.

[0051] The information push method provided by the embodiment of the present application can, when determining the recommendation scores corresponding to each candidate service for the target user, consider the popularity of the candidate service, the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services, and the conformity degree of the target user on the basis of considering the matching degree between the candidate service and the target user, so as to improve the accuracy of the determined recommendation scores, and further improve the accuracy of service recommendation when using the recommendation scores corresponding to each candidate service for service recommendation.

[0052] Please refer to Figure 2 , Figure 2 which shows a schematic flowchart of an information push method provided by another embodiment of the present application. This information push method is applied to the above computer device. The following will elaborate in detail on the Figure 2 shown process. The information push method can specifically include the following steps:

[0053] Step S210: Based on the target interaction data between the target user and multiple candidate services, the sample interaction data between multiple sample users and the multiple candidate services, the knowledge graph corresponding to the multiple candidate services, and the user information of the target user, and using a pre-trained service recommendation model, determine the service matching score, popularity score, entity score of each candidate service among the multiple candidate services, and the user score of the target user.

[0054] In the embodiments of the present application, when obtaining the service matching score, popularity score, entity score of each candidate service, and the user score of the target user, the above service matching score, popularity score, entity score, and user score can be determined simultaneously through a pre-trained service recommendation model. Among them, based on the target interaction data between the target user and multiple candidate services, the sample interaction data between multiple sample users and multiple candidate services, the knowledge graphs corresponding to multiple candidate services, and the user information of the target user, and by using the pre-trained service recommendation model, the above service matching score, popularity score, entity score, and user score can be determined.

[0055] In some embodiments, the service recommendation model may include a service matching module, a service scoring module, an entity scoring module, and a user scoring module. The service matching module is used to determine the above service matching score, the service scoring module is used to determine the above popularity score, the entity scoring module is used to determine the above entity score, and the user scoring module is used to determine the above user score. When using the service recommendation model to determine the above service matching score, popularity score, entity score, and user score, the target interaction data between the target user and multiple candidate services can be input into the service matching module to obtain the service matching score of each candidate service output by the service matching module; the sample interaction data between multiple sample users and multiple candidate services can be input into the service scoring module to obtain the popularity score of each candidate service output by the service scoring module; the knowledge graphs corresponding to multiple candidate services can be input into the entity scoring module to obtain the entity score of each candidate service output by the entity scoring module; the user information of the target user can be input into the user scoring module to obtain the user score of the target user output by the user scoring module. In the above embodiments, the service matching module, the service scoring module, the entity scoring module, and the user scoring module can be understood as sub-models in the service recommendation model, and each sub-model can be a neural network. For example, each sub-model can be implemented by a multi-layer perceptron. The training process of the above service recommendation model will be introduced in detail in subsequent embodiments.

[0056] Step S220: Determine the recommendation score corresponding to each candidate service based on the service matching score, popularity score, user score, and entity score.

[0057] In the embodiments of the present application, after obtaining the service matching score, popularity score, entity score of each candidate service, and the user score of the target user, for each candidate service, the service matching score, popularity score, user score, and entity score can be fused to obtain the recommendation score corresponding to each candidate service.

[0058] In some embodiments, the above service recommendation model further includes a score fusion module. When determining the above recommendation scores, the above service matching score, popularity score, user score, and entity score can be input into the score fusion module to obtain the recommendation scores corresponding to each candidate service output by the score fusion module. That is to say, determining the service matching score, popularity score, entity score, and user score of the target user for each candidate service is all implemented by the service recommendation model, and fusing the above obtained scores to obtain the final recommendation scores is also implemented by the service recommendation model. Thus, after only inputting the above target interaction data, sample interaction data, knowledge graph, and user information into the service recommendation model, the recommendation scores corresponding to each candidate service output by the service recommendation model can be obtained.

[0059] In a possible implementation, when determining the above recommendation scores through the score fusion module, the above popularity score, user score, and entity score can be normalized to obtain the normalized popularity score, user score, and entity score; then, the service matching score, the normalized popularity score, user score, and entity score are input into the score fusion module to obtain the recommendation scores corresponding to each candidate service obtained by the score fusion module after fusing the service matching score, the normalized popularity score, user score, and entity score. Among them, normalizing the above popularity score, user score, and entity score can be to map the popularity score, entity score, and user score to the range of 0 to 1 using the sigmoid function; the score fusion module fusing the service matching score, the normalized popularity score, user score, and entity score can be to obtain the product between the service matching score and the normalized popularity score, entity score, and user score, and use this product as the recommendation score.

[0060] Step S230: Based on the recommendation scores corresponding to each candidate service, push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user.

[0061] In the embodiments of the present application, the content of step S230 can refer to the content of other embodiments and will not be elaborated here.

[0062] The information push method provided by the embodiment of the present application can, when determining the recommendation score corresponding to each candidate service for a target user, determine a service matching score for characterizing the matching degree between the candidate service and the target user, an entity score for characterizing the density of the entity corresponding to the candidate service in the knowledge graph corresponding to multiple candidate services, a popularity score for characterizing the popularity of the candidate service, and a user score for characterizing the conformity degree of the target user through a pre-trained service recommendation model. Then, based on the service matching score, the determined popularity score, entity score, and user score can be combined to determine the final recommendation score, thereby improving the accuracy of the determined recommendation score. Furthermore, when using the recommendation score corresponding to each candidate service for service recommendation, the accuracy of service recommendation can be improved.

[0063] Please refer to Figure 3 , Figure 3 which shows a schematic flowchart of the information push method provided by another embodiment of the present application. This information push method is applied to the above computer device. The following will elaborate on the Figure 3 process shown below in detail. The information push method may specifically include the following steps:

[0064] Step S310: Input the sample interaction data between the multiple sample users and the multiple candidate services into the service matching module in the initial recommendation model, and obtain the sample matching score between each sample user and each candidate service output by the service matching module.

[0065] For the service recommendation model in the foregoing embodiment, the embodiment of the present application further includes a training method for the service recommendation model. It should be noted that the training of the service recommendation model can be pre-performed according to the obtained sample data. Subsequently, when service recommendation is required each time, the trained service recommendation model can be used, and there is no need to train the service recommendation model each time service recommendation is performed.

[0066] In the embodiment of the present application, for the training of the above service recommendation model, the initial recommendation model can be trained to obtain the service recommendation model. Similarly, the initial recommendation model can include a service matching module, a service scoring module, an entity scoring module, and a user scoring module. When training the initial recommendation model, the sample interaction data between multiple sample users and multiple candidate services can be input into the service matching module in the initial recommendation model, and the service matching score between each sample user and each candidate service output by the service matching module can be obtained as the sample matching score.

[0067] Step S320: Input the sample interaction data into the service scoring module to obtain the sample popularity scores of each candidate service output by the service scoring module.

[0068] In the embodiment of the present application, when training the initial recommendation model, the above sample interaction data can also be input into the service scoring module to obtain the popularity scores of each candidate service output by the service scoring module as the sample popularity scores.

[0069] Step S330: Input the user information of the sample user into the user scoring module to obtain the sample user score of the sample user output by the user scoring module.

[0070] In the embodiment of the present application, when training the initial recommendation model, the user information of the sample user can also be input into the user scoring module to obtain the sample user score of the sample user output by the user scoring module as the sample user score.

[0071] Step S340: Input the knowledge graph into the entity scoring module to obtain the sample entity score output by the entity scoring module.

[0072] In the embodiment of the present application, when training the initial recommendation model, the knowledge graphs corresponding to the above multiple candidate services can also be input into the entity scoring module to obtain the entity scores output by the entity scoring module as the sample entity scores.

[0073] Step S350: Determine the total loss value of the initial recommendation model based on the sample matching score, sample popularity score, sample user score, and sample entity score.

[0074] In the embodiment of the present application, after obtaining the above sample matching score, sample popularity score, sample user score, and sample entity score, the total loss value corresponding to the initial recommendation model can be determined according to the sample matching score, sample popularity score, sample user score, and sample entity score, so as to update the initial recommendation model according to the determined total loss value.

[0075] In some embodiments, when determining the total loss value corresponding to the initial recommendation model, the first loss value corresponding to the service matching module may be determined based on the sample matching score and the sample interaction data; the second loss value corresponding to the service scoring module may be determined based on the sample popularity score and the sample interaction data; the third loss value corresponding to the user scoring module may be determined based on the sample user score and the sample interaction data; the fourth loss value corresponding to the entity scoring module may be determined based on the sample entity score and the entity scoring module; and the total loss value of the initial recommendation model may be determined based on the first loss value, the second loss value, the third loss value, and the fourth loss value. It can be understood that the sample interaction data can determine whether the user actually interacts with the candidate service, so it can be used as a label to determine the above loss values.

[0076] Exemplarily, the above total loss value may be determined according to the following formula:

[0077] L = L R + α * L I + β * L U + γ * L E

[0078]

[0079]

[0080]

[0081] Wherein, L R represents the basic loss, α, β, and γ respectively represent three adjustable hyperparameters, and the basic loss L R can be calculated according to the cross-entropy loss function, represents the above service matching score, represents the popularity score of candidate service i, represents the above user score, represents the entity score of candidate service i, σ() represents the sigmoid function, and y ui refers to the true value determined based on the sample interaction data. For candidate services with interactions, the value of its y ui is 1, and for candidate services without interactions, the value of its y ui is 0. The total loss value determined by the above formula can enable the above true value to constrain the updates of the service matching module, service scoring module, entity scoring module, and user scoring module in the service recommendation model, so that the scores output by the service matching module, service scoring module, entity scoring module, and user scoring module are more and more accurate.

[0082] Step S360: Based on the total loss value, iteratively train the initial recommendation model to obtain the service recommendation model.

[0083] In the embodiment of this application, after determining the above total loss value, the initial recommendation model can be iteratively trained according to the total loss value to obtain the final service recommendation model.

[0084] In some embodiments, the model parameters of the initial recommendation model can be adjusted according to the calculated total loss value; return to step S310, and repeat steps S310 - S360 until the training end condition is satisfied to obtain the trained service recommendation model.

[0085] In a possible embodiment, the initial recommendation model can be iteratively trained using the Adam optimizer according to the total loss value until the loss value of the output result of the initial recommendation model converges, and the model at this time is saved to obtain the trained service recommendation model. Among them, the Adam optimizer combines the advantages of the AdaGra (Adaptive Gradient) and RMSProp optimization algorithms, comprehensively considers the first moment estimation of the gradient (i.e., the mean of the gradient) and the second moment estimation (i.e., the uncentered variance of the gradient), and calculates the update step size.

[0086] In some embodiments, the training end conditions for iterative training may include: the number of iterations reaches the target number; or the total loss value of the output result of the initial recommendation model meets the set conditions.

[0087] Optionally, the convergence condition is to make the total loss value as small as possible. Using the initial learning rate of 1e - 3, the learning rate decays cosine - like with the number of steps, batch_size = 8, and after training for 16 epochs, it can be considered that the convergence is completed. Among them, batch_size can be understood as the batch processing parameter, and its limit value is the total number of samples in the training set.

[0088] Optionally, the total loss value meeting the set conditions may include: the total loss value is less than the set threshold. Of course, the specific set conditions are not limited.

[0089] Step S370: Based on the target interaction data between the target user and multiple candidate services, the sample interaction data between multiple sample users and the multiple candidate services, the knowledge graphs corresponding to the multiple candidate services, and the user information of the target user, and by using a pre-trained service recommendation model, determine the service matching score of each candidate service among the multiple candidate services, the popularity score of each candidate service, the entity score of each candidate service, and the user score of the target user.

[0090] Step S380: Based on the service matching score, popularity score, user score, and entity score, determine the recommendation score corresponding to each candidate service.

[0091] Step S390: Based on the recommendation score corresponding to each candidate service, push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user.

[0092] In the embodiments of the present application, Steps S370 to S390 may refer to the content of other embodiments and will not be elaborated herein.

[0093] The information push method provided by the embodiments of the present application also provides a training method for the above service recommendation model, enabling the determination of a service matching score for characterizing the matching degree between a candidate service and a target user, an entity score for characterizing the density of the entity corresponding to the candidate service in the knowledge graphs corresponding to the multiple candidate services, a popularity score for characterizing the popularity of the candidate service, and a user score for characterizing the conformity degree of the target user. Then, based on the service matching score, the determined popularity score, entity score, and user score can be combined to determine the final recommendation score, thereby improving the accuracy of the determined recommendation score. Furthermore, when using the recommendation score corresponding to each candidate service for service recommendation, the accuracy of service recommendation can be improved.

[0094] Please refer to Figure 4 , Figure 4 which shows a schematic flowchart of an information push method provided by another embodiment of the present application. This information push method is applied to the above computer device. The following will elaborate in detail on the Figure 4 shown process. The information push method may specifically include the following steps:

[0095] Step S401: Based on the target interaction data between the target user and multiple candidate services, determine the service matching score of each candidate service among the multiple candidate services, where the service matching score is used to characterize the matching degree between the candidate service and the target user.

[0096] Step S402: Obtain the popularity score and the entity score of each candidate service among the multiple candidate services. The popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services.

[0097] Step S403: Obtain the user score of the target user. The user score is used to characterize the degree of conformity of the target user.

[0098] Step S404: Based on the service matching score, popularity score, user score, and entity score, determine the recommendation score corresponding to each candidate service.

[0099] In the embodiments of the present application, Steps S401 to S404 can refer to the content of the foregoing embodiments and will not be elaborated here.

[0100] Step S405: Obtain the average value of the service matching scores of each candidate service corresponding to multiple sample users as the average matching score.

[0101] In the embodiments of the present application, after obtaining the recommendation score corresponding to each candidate service and before performing service recommendation based on the recommendation score corresponding to each candidate service, in order to better eliminate the popularity bias, it is also possible to infer the direct influence from the popularity of the candidate service to the recommendation score corresponding to the candidate service, so as to remove this influence from the recommendation score before performing service recommendation. Among them, in the counterfactual reasoning stage, the above recommendation score can be calibrated by deducting the counterfactual prediction, so as to eliminate this bias. When calibrating the obtained recommendation score, the average value of the service matching scores of each candidate service corresponding to multiple sample users can be obtained as the average matching score.

[0102] In some embodiments, when obtaining the average value of the service matching scores of each candidate service corresponding to multiple sample users, first, for each sample user, obtain the service matching score of each candidate service by the method of obtaining the service matching score of each candidate service corresponding to the target user provided in the foregoing embodiments; then, for each candidate service, calculate the average value of the service matching scores of all sample users corresponding to the candidate service to obtain the average matching score corresponding to the candidate service.

[0103] Step S406: Obtain the average value of the popularity scores corresponding to the multiple candidate services as the average popularity score.

[0104] In the embodiments of the present application, when calibrating the obtained recommendation score, it is also possible to obtain the average value of the popularity scores corresponding to multiple candidate services as the average popularity score.

[0105] In some embodiments, when obtaining the average popularity score corresponding to multiple candidate services, first, for each sample user, obtain the popularity score of each candidate service by the method of obtaining the popularity score of each candidate service corresponding to the target user provided in the foregoing embodiments; then, for each candidate service, calculate the average of the popularity scores of all sample users corresponding to this candidate service to obtain the average popularity score corresponding to this candidate service.

[0106] Step S407: Obtain the average of the entity scores corresponding to the multiple candidate services as the average entity score.

[0107] In the embodiments of the present application, when calibrating the obtained recommendation scores, the average of the entity scores corresponding to multiple candidate services can also be obtained as the average entity score.

[0108] In some embodiments, when obtaining the average of the entity scores corresponding to multiple candidate services, first, for each sample user, obtain the entity score of each candidate service by the method of obtaining the entity score of each candidate service corresponding to the target user provided in the foregoing embodiments; then, for each candidate service, calculate the average of the entity scores of all sample users corresponding to this candidate service to obtain the average entity score corresponding to this candidate service.

[0109] Step S408: Obtain the average of the user scores corresponding to the multiple sample users as the average user score.

[0110] In the embodiments of the present application, when calibrating the obtained recommendation scores, the average of the user scores corresponding to multiple sample users can also be obtained as the average user score.

[0111] In some embodiments, when obtaining the average of the user scores corresponding to multiple sample users, first, for each sample user, obtain the user score corresponding to each sample user by the method of obtaining the user score of the target user provided in the foregoing embodiments; then calculate the average of the user scores corresponding to all sample users to obtain the average user score.

[0112] Step S409: Based on the average matching score, average popularity score, average entity score, and average user score, determine the intermediate recommendation score corresponding to each candidate service.

[0113] In an embodiment of the present application, after obtaining the average matching score corresponding to each candidate service, the average popularity score corresponding to each candidate service, the average entity score corresponding to each candidate service, and the average user score of multiple sample users, for each candidate service, based on its corresponding average matching score, average popularity score, average entity score, and average user score, an intermediate recommendation score corresponding to each candidate service can be determined. Among them, the method for determining the intermediate recommendation score corresponding to each candidate service based on the average matching score, average popularity score, average entity score, and average user score can refer to the method for determining the recommendation score corresponding to each candidate service based on the service matching score, popularity score, user score, and entity score provided in the foregoing embodiment, which will not be elaborated herein.

[0114] Step S410: Calculate the causal effect of counterfactual reasoning on the recommendation score corresponding to each candidate service based on the intermediate recommendation score corresponding to each candidate service, so as to eliminate the popularity bias of the recommendation score corresponding to each candidate service.

[0115] In an embodiment of the present application, after obtaining the intermediate recommendation score corresponding to each candidate service as above, based on the intermediate recommendation score corresponding to each candidate service, the causal effect of counterfactual reasoning can be calculated on the recommendation score corresponding to each candidate service, so as to eliminate the popularity bias of the recommendation score corresponding to each candidate service.

[0116] In some embodiments, when calculating the causal effect of counterfactual reasoning on the recommendation score corresponding to each candidate service, the product of the intermediate recommendation score corresponding to each candidate service and a preset coefficient can be obtained to obtain a deviation value representing the popularity bias corresponding to each candidate service; the difference between the recommendation score corresponding to each candidate service and the deviation value corresponding to each candidate service is obtained to obtain the recommendation score corresponding to each candidate service after the causal effect calculation of counterfactual reasoning. Among them, the preset coefficient can be less than 1. For example, the preset coefficient can be 0.1, 0.2, 0.3, etc.

[0117] Next, the principle based on which the method for calculating the causal effect of counterfactual reasoning on the recommendation score provided in the embodiment of the present application is introduced.

[0118] Among them, U represents the degree of conformity of users, E represents the density of the entity corresponding to the service in the knowledge graph, I represents the popularity of the service in the recommendation scenario, C represents the interaction between the user and the service, and Y represents the final recommendation score. u * 、i * 、e * respectively represent the reference values of U, I, and E, and are usually set to the average values of the corresponding variables.

[0119] The matching function C(U, I, E) blocks the indirect paths through which U, I, and E affect Y, and its Natural Direct Effect (NDE) can be expressed as:

[0120]

[0121] According to the formula of NDE, the Total Effect (TE) from I to Y is:

[0122]

[0123] The Total Indirect Effect (TIE) is expressed as:

[0124]

[0125] Based on the formula of the total indirect effect, it can be obtained that for the causal effect calculation of counterfactual reasoning on the recommendation scores corresponding to each candidate service, after obtaining the above average matching score, average popularity score, average entity score, and average user score, the intermediate recommendation score can be determined, and then based on the product of the intermediate recommendation score and the preset coefficient, the deviation to be eliminated can be obtained. Then, subtracting this deviation from the recommendation score can obtain the recommendation score after the causal effect calculation of counterfactual reasoning, thereby realizing the elimination of popularity bias for the recommendation score based on causal analysis. Causal analysis can distinguish the relationship between user preferences and popularity bias, thereby providing more accurate recommendations. By understanding the real motivation behind user behavior, the recommendation system can more effectively meet the personalized needs of users.

[0126] Step S411: Based on the recommendation scores corresponding to each of the candidate services, push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user.

[0127] In the embodiment of the present application, after completing the causal effect calculation of counterfactual reasoning on the obtained recommendation scores, the information corresponding to the target service among the multiple candidate services can be pushed to the electronic device corresponding to the target user based on the recommendation scores corresponding to each candidate service after the causal effect calculation of counterfactual reasoning.

[0128] The information push method provided by the embodiments of the present application can, when determining the recommendation score corresponding to each candidate service for a target user, consider not only the matching degree between the candidate service and the target user, but also the popularity of the candidate service, the density of the entity corresponding to the candidate service in the knowledge graph corresponding to multiple candidate services, and the herd mentality of the target user, thereby improving the accuracy of the determined recommendation score. Furthermore, when using the recommendation score corresponding to each candidate service for service recommendation, the accuracy of service recommendation can be improved. Additionally, before using the recommendation score corresponding to each candidate service for service recommendation, the epidemic bias of the obtained recommendation score is eliminated based on the causal relationship, thereby further improving the accuracy of the recommendation score for service recommendation, and further improving the accuracy of service recommendation.

[0129] Please refer to Figure 5 , which shows a structural block diagram of an information push device 500 provided by the embodiments of the present application. The information push device 500 applies the above computer device. The information push device 500 includes: a first score acquisition module 510, a second score acquisition module 520, a third score acquisition module 530, a fourth score acquisition module 540, and a service recommendation module 550. Among them, the first score acquisition module 510 is configured to determine the service matching score of each candidate service among the multiple candidate services based on the target interaction data between the target user and the multiple candidate services, and the service matching score is used to represent the matching degree between the candidate service and the target user; the second score acquisition module 520 is configured to acquire the popularity score and the entity score of each candidate service among the multiple candidate services, the popularity score is used to represent the popularity of the candidate service, and the entity score is used to represent the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services; the third score acquisition module 530 is configured to acquire the user score of the target user, and the user score is used to represent the herd mentality of the target user; the fourth score acquisition module 540 is configured to determine the recommendation score corresponding to each candidate service based on the service matching score, the popularity score, the user score, and the entity score; the service recommendation module 550 is configured to push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user based on the recommendation score corresponding to each candidate service.

[0130] In some embodiments, the first score acquisition module 510, the second score acquisition module 520, and the third score acquisition module 530 may be specifically configured to: based on the target interaction data between the target user and multiple candidate services, the sample interaction data between multiple sample users and the multiple candidate services, the knowledge graph corresponding to the multiple candidate services, and the user information of the target user, and by using a pre-trained service recommendation model, determine the service matching score, popularity score, entity score, and user score.

[0131] In a possible embodiment, the service recommendation model includes a service matching module, a service scoring module, an entity scoring module, and a user scoring module. The first score acquisition module 510 may be specifically configured to input the target interaction data between the target user and multiple candidate services into the service matching module, and obtain the service matching score of each candidate service output by the service matching module; the second score acquisition module 520 may be specifically configured to input the sample interaction data between the multiple sample users and the multiple candidate services into the service scoring module, and obtain the popularity score of each candidate service output by the service scoring module; the second score acquisition module 520 may also be configured to input the knowledge graph corresponding to the multiple candidate services into the entity scoring module, and obtain the entity score of each candidate service output by the entity scoring module; the third score acquisition module 530 may be specifically configured to input the user information of the target user into the user scoring module, and obtain the user score of the target user output by the user scoring module.

[0132] Optionally, the service recommendation model may further include a score fusion module. The fourth score acquisition module 540 may be specifically configured to input the service matching score, popularity score, user score, and entity score into the score fusion module, and obtain the recommendation score corresponding to each candidate service output by the score fusion module.

[0133] Optionally, the fourth score acquisition module 540 may also be configured to perform normalization processing on the popularity score, user score, and entity score to obtain the normalized popularity score, user score, and entity score; input the service matching score, the normalized popularity score, user score, and entity score into the score fusion module, and obtain the recommendation score corresponding to each candidate service obtained by the score fusion module after fusing the service matching score, the normalized popularity score, user score, and entity score.

[0134] Optionally, the information recommendation device 500 may further include a model training module. The model training module may be configured to: input the sample interaction data into the service matching module in the initial recommendation model to obtain sample matching scores between each sample user and each candidate service output by the service matching module; input the sample interaction data into the service scoring module to obtain sample popularity scores of each candidate service output by the service scoring module; input the user information of the sample user into the user scoring module to obtain sample user scores of the sample user output by the user scoring module; input the knowledge graph into the entity scoring module to obtain sample entity scores output by the entity scoring module; determine a total loss value of the initial recommendation model based on the sample matching scores, sample popularity scores, sample user scores, and sample entity scores; and perform iterative training on the initial recommendation model based on the total loss value to obtain the service recommendation model.

[0135] Optionally, the model training module may further be configured to: determine a first loss value corresponding to the service matching module based on the sample matching scores and the sample interaction data; determine a second loss value corresponding to the service scoring module based on the sample popularity scores and the sample interaction data; determine a third loss value corresponding to the user scoring module based on the sample user scores and the sample interaction data; determine a fourth loss value corresponding to the entity scoring module based on the sample entity scores and the entity scoring module; and determine the total loss value of the initial recommendation model based on the first loss value, second loss value, third loss value, and fourth loss value.

[0136] In some embodiments, the information recommendation device 500 may further include a first average value acquisition module, a second average value acquisition module, a third average value acquisition module, a fourth average value acquisition module, an intermediate score acquisition module, and a score calibration module. The first average value acquisition module may be configured to, before pushing the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user based on the recommendation scores corresponding to each candidate service, acquire the average value of the service matching scores of each candidate service corresponding to multiple sample users as the average matching score; the second average value acquisition module is configured to acquire the average value of the popularity scores corresponding to the multiple candidate services as the average popularity score; the third average value acquisition module is configured to acquire the average value of the entity scores corresponding to the multiple candidate services as the average entity score; the fourth average value acquisition module is configured to acquire the average value of the user scores corresponding to the multiple sample users as the average user score; the intermediate score acquisition module is configured to determine the intermediate recommendation score corresponding to each candidate service based on the average matching score, the average popularity score, the average entity score, and the average user score; the score calibration module is configured to perform a causal effect calculation of counterfactual inference on the recommendation score corresponding to each candidate service based on the intermediate recommendation score corresponding to each candidate service to eliminate the popularity bias of the recommendation score corresponding to each candidate service.

[0137] In a possible implementation manner, the score calibration module may specifically be configured to: acquire the product of the intermediate recommendation score corresponding to each candidate service and a preset coefficient to obtain a deviation value representing the popularity bias corresponding to each candidate service; acquire the difference between the recommendation score corresponding to each candidate service and the deviation value corresponding to each candidate service to obtain the recommendation score corresponding to each candidate service after the causal effect calculation of counterfactual inference.

[0138] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described devices and modules can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0139] In several embodiments provided in the present application, the coupling between modules may be electrical, mechanical, or other forms of coupling.

[0140] In addition, in each embodiment of the present application, each functional module may be integrated in one processing module, or each module may exist physically alone, or two or more modules may be integrated in one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0141] In summary, the solution provided by this application determines the service matching score for each candidate service among multiple candidate services based on the target interaction data between the target user and the multiple candidate services. The service matching score is used to characterize the matching degree between the candidate service and the target user. Obtain the popularity score and entity score for each candidate service among the multiple candidate services. The popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services. Obtain the user score of the target user, where the user score is used to characterize the herd mentality of the target user. Based on the service matching score, popularity score, user score, and entity score, determine the recommendation score corresponding to each candidate service. Based on the recommendation score corresponding to each candidate service, push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user. Thus, when determining the recommendation score corresponding to each candidate service for the target user, on the basis of considering the matching degree between the candidate service and the target user, the popularity of the candidate service, the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services, and the herd mentality of the target user are also considered, so that the accuracy of the determined recommendation score can be improved. Furthermore, when using the recommendation score corresponding to each candidate service for service recommendation, the accuracy of service recommendation can be improved.

[0142] Please refer to Figure 6 , which shows a structural block diagram of a computer device provided by an embodiment of the present application. The computer device 100 may be a physical server, a cloud server, etc. The computer device 100 in the present application may include one or more of the following components: a processor 110, a memory 120, and one or more application programs, where one or more application programs may be stored in the memory 120 and configured to be executed by one or more processors 110, and one or more application programs are configured to execute the method described in the foregoing method embodiment.

[0143] The processor 110 may include one or more processing cores. The processor 110 connects various parts within the entire computer device 100 using various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 120, and by invoking data stored in the memory 120, it performs various functions of the computer device 100 and processes data. Optionally, the processor 110 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 110 may integrate a combination of one or several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the displayed content; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor 110 and may be implemented separately through a communication chip.

[0144] The memory 120 may include random access memory (RAM) and may also include read-only memory. The memory 120 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 120 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for implementing at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing each of the following method embodiments, etc. The data storage area may also store data created during the use of the computer device 100 (such as phone book, audio and video data, chat record data, etc.).

[0145] Please refer to Figure 7 , which shows a structural block diagram of a computer-readable storage medium provided by an embodiment of the present application. Program code is stored in the computer-readable medium 800, and the program code can be called by the processor to execute the methods described in the above method embodiments.

[0146] The computer-readable storage medium 800 can be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. Optionally, the computer-readable storage medium 800 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 800 has a storage space for program code 810 that executes any of the method steps in the above-described methods. These program codes can be read from or written into one or more computer program products. The program code 810 can be compressed in a suitable form, for example.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not intended to limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An information push method, characterized in that, The method includes: Based on the target interaction data between the target user and multiple candidate services, determining a service matching score for each candidate service among the multiple candidate services, where the service matching score is used to characterize the matching degree between the candidate service and the target user; Obtaining a popularity score and an entity score for each candidate service among the multiple candidate services, where the popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services; Obtaining a user score of the target user, where the user score is used to characterize the degree of conformity of the target user; Based on the service matching score, popularity score, user score, and entity score, determining a recommendation score corresponding to each candidate service; Based on the recommendation score corresponding to each candidate service, pushing the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user.

2. The method according to claim 1, characterized in that The steps from determining the service matching score for each candidate service among the multiple candidate services based on the target interaction data between the target user and the multiple candidate services to obtaining the user score of the target user include: Based on the target interaction data between the target user and the multiple candidate services, the sample interaction data between multiple sample users and the multiple candidate services, the knowledge graph corresponding to the multiple candidate services, and the user information of the target user, and using a pre-trained service recommendation model, determining the service matching score, popularity score, entity score, and user score.

3. The method according to claim 2, wherein The service recommendation model includes a service matching module, a service scoring module, an entity scoring module, and a user scoring module. The process of determining the service matching score, popularity score, entity score, and user score based on the target interaction data between the target user and the multiple candidate services, the sample interaction data between multiple sample users and the multiple candidate services, the knowledge graph corresponding to the multiple candidate services, and the user information of the target user, and using a pre-trained service recommendation model includes: Inputting the target interaction data between the target user and the multiple candidate services into the service matching module to obtain the service matching score of each candidate service output by the service matching module; Inputting the sample interaction data between the multiple sample users and the multiple candidate services into the service scoring module to obtain the popularity score of each candidate service output by the service scoring module; Inputting the knowledge graph corresponding to the multiple candidate services into the entity scoring module to obtain the entity score of each candidate service output by the entity scoring module; Inputting the user information of the target user into the user scoring module to obtain the user score of the target user output by the user scoring module.

4. The method according to claim 3, characterized in that, The service recommendation model further includes a score fusion module. The process of determining the recommendation score corresponding to each candidate service based on the service matching score, popularity score, user score, and entity score includes: Input the service matching score, popularity score, user score, and entity score into the score fusion module to obtain the recommendation score corresponding to each candidate service output by the score fusion module.

5. The method according to claim 4, wherein The step of inputting the service matching score, popularity score, user score, and entity score into the score fusion module to obtain the recommendation score corresponding to each candidate service output by the score fusion module includes: Perform normalization processing on the popularity score, user score, and entity score to obtain the normalized popularity score, user score, and entity score. Input the service matching score, the normalized popularity score, user score, and entity score into the score fusion module to obtain the recommendation score corresponding to each candidate service obtained by the score fusion module after fusing the service matching score, the normalized popularity score, user score, and entity score.

6. The method according to claim 3, wherein The service recommendation model is trained in the following manner: Input the sample interaction data into the service matching module in the initial recommendation model to obtain the sample matching score between each sample user and each candidate service output by the service matching module. Input the sample interaction data into the service scoring module to obtain the sample popularity score of each candidate service output by the service scoring module. Input the user information of the sample user into the user scoring module to obtain the sample user score of the sample user output by the user scoring module. Input the knowledge graph into the entity scoring module to obtain the sample entity score output by the entity scoring module. Determine the total loss value of the initial recommendation model based on the sample matching score, sample popularity score, sample user score, and sample entity score. Based on the total loss value, perform iterative training on the initial recommendation model to obtain the service recommendation model.

7. The method according to claim 6, characterized in that, The step of determining the total loss value of the initial recommendation model based on the sample matching score, sample popularity score, sample user score, and sample entity score includes: Determine the first loss value corresponding to the service matching module based on the sample matching score and the sample interaction data. Determine the second loss value corresponding to the service scoring module based on the sample popularity score and the sample interaction data. Determine the third loss value corresponding to the user scoring module based on the sample user score and the sample interaction data. Determine the fourth loss value corresponding to the entity scoring module based on the sample entity score and the entity scoring module. Determine the total loss value of the initial recommendation model based on the first loss value, second loss value, third loss value, and fourth loss value.

8. The method according to any one of claims 1-7, characterized in that, Before pushing the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user based on the recommendation score corresponding to each candidate service, the method further includes: Obtain the average value of the service matching scores of each candidate service corresponding to multiple sample users as the average matching score. Obtain the average of the popularity scores corresponding to the multiple candidate services as the average popularity score; Obtain the average of the entity scores corresponding to the multiple candidate services as the average entity score; Obtain the average of the user scores corresponding to the multiple sample users as the average user score; Based on the average matching score, average popularity score, average entity score, and average user score, determine the intermediate recommendation score corresponding to each candidate service; Based on the intermediate recommendation score corresponding to each candidate service, perform a causal effect calculation of counterfactual reasoning on the recommendation score corresponding to each candidate service to eliminate the popularity bias of the recommendation score corresponding to each candidate service.

9. The method according to claim 8, wherein The performing a causal effect calculation of counterfactual reasoning on the recommendation score corresponding to each candidate service based on the intermediate recommendation score corresponding to each candidate service includes: Obtain the product of the intermediate recommendation score corresponding to each candidate service and a preset coefficient to obtain a deviation value for characterizing the popularity bias corresponding to each candidate service; Obtain the difference between the recommendation score corresponding to each candidate service and the deviation value corresponding to each candidate service to obtain the recommendation score corresponding to each candidate service after the causal effect calculation of counterfactual reasoning.

10. An information push device, characterized in that, The device includes: a first score acquisition module, a second score acquisition module, a third score acquisition module, a fourth score acquisition module, and a service recommendation module, where The first score acquisition module is configured to determine the service matching score of each candidate service among the multiple candidate services based on the target interaction data between the target user and the multiple candidate services, and the service matching score is used to characterize the matching degree between the candidate service and the target user; The second score acquisition module is configured to obtain the popularity score and entity score of each candidate service among the multiple candidate services, the popularity score is used to characterize the popularity of the candidate service, and the entity score is used to characterize the density of the entity corresponding to the candidate service in the knowledge graph corresponding to the multiple candidate services; The third score acquisition module is configured to obtain the user score of the target user, and the user score is used to characterize the herd mentality of the target user; The fourth score acquisition module is configured to determine the recommendation score corresponding to each candidate service based on the service matching score, popularity score, user score, and entity score; The service recommendation module is configured to push the information corresponding to the target service among the multiple candidate services to the electronic device corresponding to the target user based on the recommendation score corresponding to each candidate service.

11. A computer device, characterized in that, Comprising: One or more processors; A memory; One or more programs, where the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to execute the method according to any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, Program code is stored in the computer-readable storage medium, and the program code can be called by a processor to execute the method according to any one of claims 1-9.

Images