Recommendation model training method, information recommendation method and related products

Abstract

The invention relates to a recommendation model training method, an information recommendation method and device, a readable storage medium and a program product, relates to the technical field of artificial intelligence, and can effectively improve information recommendation efficiency. The recommendation model training method comprises the following steps: inputting account attributes of a plurality of resource acquisition users at different clients with different application types and account behaviors triggered by the plurality of resource acquisition users for resource release accounts of the different clients into a feature extraction module; based on the output of the feature extraction module, acquiring a first similarity of account attribute features of the same resource acquisition user at different clients, a second similarity of account behavior features of the same resource acquisition user at different clients, and a third similarity of account attribute features of different resource acquisition users at different clients; and determining expression enhancement loss according to the first similarity, the second similarity and the third similarity, and adjusting model parameters of the recommendation model according to the expression enhancement loss to obtain a trained recommendation model.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a recommendation model training method, an information recommendation method, an apparatus, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the development of computer technology and artificial intelligence technology, recommendation systems are increasingly valued in daily information recommendation. Recommendation systems provide users with accurate and personalized content push by performing relevant analysis. In related technologies, recommendation systems are mainly built around information provided by a single client. By utilizing the information corresponding to different business scenarios provided by a single client, relevant content in different scenarios can be accurately pushed to users.

[0003] However, the recommendation system constructed in the above manner can only recommend information to users of a single client. When a user uses a new client, the recommendation system needs to be rebuilt, and efficient information recommendation cannot be achieved. Summary of the invention

[0004] Based on this, it is necessary to provide a recommendation model training method, information recommendation method, computer device, computer readable storage medium and computer program product to address the above technical problems.

[0005] In a first aspect, the present application provides a recommendation model training method, comprising:

[0006] Input the account attributes of multiple resource acquisition users in different clients of different application types, and the account behaviors triggered by the resource publishing accounts of the multiple resource acquisition users for the different clients, into the feature extraction module of the recommendation model to obtain the account attribute features and account behavior features corresponding to each of the resource acquisition users and each of the clients;

[0007] Based on the account attribute characteristics and account behavior characteristics corresponding to each of the resource acquisition users and each of the clients, a first similarity of the account attribute characteristics of the same resource acquisition user on different clients, a second similarity of the account behavior characteristics of the same resource acquisition user on different clients, and a third similarity of the account attribute characteristics of different resource acquisition users on different clients are obtained; wherein the different resource acquisition users are resource acquisition users who trigger the same account behavior on associated resource publishing accounts on different clients;

[0008] A representation enhancement loss is determined according to the first similarity, the second similarity, and the third similarity, and a model parameter of the recommendation model is adjusted according to the representation enhancement loss to obtain a trained recommendation model.

[0009] In one embodiment, determining the representation enhancement loss according to the first similarity, the second similarity and the third similarity comprises:

[0010] Determining a weight of the third similarity according to account behaviors triggered by different resource acquisition users on different clients for associated resource publishing accounts, and adjusting the third similarity according to the weight to obtain an adjusted third similarity;

[0011] A representation enhancement loss is determined according to the first similarity, the second similarity and the adjusted third similarity; the representation enhancement loss is positively correlated with the first similarity and negatively correlated with the second similarity and the adjusted third similarity.

[0012] In one embodiment, the weight of the third similarity is determined according to the account behaviors triggered by the different resource acquisition users on different clients for publishing accounts of associated resources, including:

[0013] If the account behaviors triggered by the different resource acquisition users on the associated resource publishing accounts in different clients include multiple types, determining the behavior priority of each of the account behaviors;

[0014] Determine the target account behavior whose behavior priority satisfies the priority condition, and obtain the weight of the third similarity according to the weight corresponding to the target account behavior.

[0015] In one embodiment, the account attributes of the multiple resource acquisition users on different clients include account attribute information of each of the multiple resource acquisition users on a first client, and account attribute information of each of the resource acquisition users on a second client;

[0016] The account behaviors triggered by the multiple resource acquisition users for the resource publishing accounts of the different clients include: the account behaviors triggered by each of the multiple resource acquisition users for the resource publishing account of the first client, and the account behaviors triggered by each of the resource acquisition users for the resource publishing account of the second client.

[0017] In one embodiment, the same resource obtaining the account attribute characteristics of the user in different clients includes: the account attribute characteristics of the first client account and the account attribute characteristics of the second client account corresponding to the same resource obtaining the user;

[0018] The account behavior characteristics of the same resource acquisition user on different clients include: account behavior characteristics of a first client account and account behavior characteristics of a second client account corresponding to the same resource acquisition user;

[0019] The account attribute characteristics of different clients of different resource acquisition users include: account attribute characteristics of a first client account and account attribute characteristics of a second client account corresponding to different resource acquisition users.

[0020] In one embodiment, the method of acquiring the account attribute characteristics and account behavior characteristics corresponding to each of the resources and each of the clients, obtaining a first similarity of the account attribute characteristics of the user acquired from the same resource on different clients, a second similarity of the account behavior characteristics of the user acquired from the same resource on different clients, and a third similarity of the account attribute characteristics of the user acquired from different resources on different clients, comprises:

[0021] Determine a first similarity between the account attribute characteristics of the first client account and the account attribute characteristics of the second client account corresponding to the same resource acquisition user;

[0022] Determine a second similarity between the account behavior characteristics of the first client account and the account behavior characteristics of the second client account corresponding to the same resource acquisition user;

[0023] A third similarity between the account attribute characteristics of the first client account corresponding to different resource acquisition users and the account attribute characteristics of the second client account is determined.

[0024] In a second aspect, the present application also provides an information recommendation method, the method comprising:

[0025] In response to an account recommendation request for a first client target account of a resource acquisition user, determining a plurality of candidate resource publishing accounts of the first client; the account recommendation request is generated after a second client target account of the same resource acquisition user on the second client triggers a preset event;

[0026] For each of the candidate resource publishing accounts, according to the resource publishing account attribute information of the candidate resource publishing account, and the account attribute information and account behavior information of the first client target account, obtain an account information combination corresponding to the candidate resource publishing account;

[0027] The account information combination is input into a trained recommendation model, and a feature extraction module of the recommendation model obtains a feature combination corresponding to the account information combination, and determines a recommendation index of the candidate resource publishing account according to the feature combination; wherein the recommendation model is trained by the recommendation model training method according to any one of claims 1 to 6;

[0028] According to the recommendation index of each of the candidate resource publishing accounts, resource publishing account recommendation information provided to the first client target account is determined.

[0029] In one embodiment, the account recommendation request is generated and sent by the first client after the second client target account of the same resource acquisition user triggers a preset event in the second client;

[0030] and / or,

[0031] The account recommendation request is generated and sent by the second client target account after the same resource acquisition user triggers a preset event in the second client target account of the second client.

[0032] In a third aspect, the present application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the training steps of the recommendation model training method as described in any one of the above items or the steps of the information recommendation method.

[0033] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the training steps of the recommendation model training method as described in any one of the above items or the steps of the information recommendation method.

[0034] In a fifth aspect, the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the training steps of the recommendation model training method as described in any of the above items or the steps of the information recommendation method.

[0035] The above-mentioned recommendation model training method, information recommendation method, computer device, computer-readable storage medium and computer program product, by obtaining the first similarity of the account attribute characteristics of the user on different clients obtained by the same resource, the second similarity of the account behavior characteristics of the user on different clients obtained by the same resource, and the third similarity of the account attribute characteristics of the user on different clients obtained by different resources, adjust the recommendation model parameters according to the first similarity, the second similarity and the third similarity, and can characterize the similarity of user behavior on multiple terminals, the difference of user attributes and the interoperability of user attributes through comparative learning, which helps the model to supplement and strengthen the characteristics of the current client user with the information of similar users learned from other clients, and accurately and flexibly push information on multiple clients based on the strengthened characteristics, effectively improving the efficiency of information recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 A flowchart of a recommendation model training method in an embodiment;

[0038] Figure 2 A schematic diagram of the architecture of a recommendation model in an embodiment;

[0039] Figure 3 A schematic diagram of a flow chart of an information recommendation method in an embodiment;

[0040] Figure 4a is a schematic diagram of a music application interface in one embodiment;

[0041] Figure 4b A schematic diagram of a karaoke application interface in one embodiment;

[0042] Figure 4c A schematic diagram of a live broadcast interface of a music application in one embodiment;

[0043] Figure 5 is a flow chart of another information recommendation method in one embodiment;

[0044] Figure 6 is an internal structure diagram of a computer device in one embodiment;

[0045] Figure 7 The figure is a diagram of the internal structure of another computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, played data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0048] In one embodiment, Figure 1 As shown, a recommendation model training method is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0049] S101, input the account attributes of multiple resource acquisition users in different clients with different application types, as well as the account behaviors triggered by the resource publishing accounts of multiple resource acquisition users for different clients, into the feature extraction module of the recommendation model to obtain the account attribute features and account behavior features corresponding to each resource acquisition user and each client.

[0050] Mobile applications (Application, app) can provide users with diversified services and resources through their clients, and improve the deficiencies and personalization of the original system. In related technologies, since the data in the business scenario provided by the client can accurately describe the style of the single client, and the data of other clients are often difficult to align and use, there will be less overlap of users of different clients and less overlap of services / items and other resources provided by different clients. Therefore, the traditional recommendation system is mainly built around the information provided by a single client. By using the information corresponding to different business scenarios provided by a single client, relevant content in different scenarios can be accurately pushed to users. For example, multi-scenario joint modeling of the same client is based on the joint modeling of the same content data in different scenarios of the same client, and the commonality and difference between scenarios are mined; this recommendation system construction method is in line with the style and tonality of most clients. Of course, in some related technologies, it also involves the use of joint modeling of multi-terminal apps, such as recommendation systems for marketing information or promotion information. However, since the promotion platform will strictly limit the information format related to marketing promotion information to ensure the consistency of the data format of promotion information in different clients, its recommendation system construction method is more similar to multi-scenario modeling of a single client. The feature system and user behavior in different business scenarios are relatively large, and the use of scenario information is focused on.

[0051] However, the recommendation system constructed in the above manner can only recommend information to users of a single client, or recommend information to multiple clients with basically the same feature systems or user behaviors. When users use new clients with different application types, if the user behaviors or client feature systems differ greatly, the recommendation system often needs to be rebuilt due to the small number of overlapping users.

[0052] In this regard, in an embodiment of the present application, multiple different clients for performing multi-terminal joint modeling may be determined first.

[0053] Among them, different clients can be application clients of different application types. In some examples, different clients can include clients developed by different developers, or can include clients developed by the same developer for different purposes. Different clients can be used in the same or different scenarios; multiple different clients are two or more, and the multiple different clients can trust each other.

[0054] Each client can provide an account for the user. For the convenience of distinction in the following text, in this embodiment, the account for obtaining resources is called a resource acquisition account, the user who uses the resource acquisition account is called a resource acquisition user, the account for publishing or providing resources is called a resource publishing account, and the user who uses the resource publishing account is called a resource acquisition user; accordingly, the resource acquisition user can obtain the required resources from the resources published by the resource publishing account through the resource acquisition account it has on the client. The resources can be real-time or pre-generated resources, such as audio live broadcast, video live broadcast, recorded audio, recorded video, image, text, etc. One or more. It can be understood that the resource acquisition user, resource publishing user, resource acquisition account, and resource publishing account in this embodiment are used to distinguish between users and accounts. In some examples, the types of users and accounts can change. For example, when user A uses his account A to publish resources, he can be classified as a resource publishing user and a resource publishing account. When user A stops publishing resources and uses account A to browse the media resources pushed by the platform, he can be classified as a resource acquisition user and a resource acquisition account.

[0055] In addition, multiple different clients can provide different services or items, and the behavior types of account behaviors triggered by resource acquisition users on multiple different clients for resource publishing accounts can be the same. For example, for client A and client B, client A is a singing app that can provide a variety of services for real users to interact with each other, such as users can enter the video virtual space corresponding to the show video anchor to watch live broadcasts; client B is a song listening app that can provide recorded songs published by the resource publishing account, and can also provide an audio virtual space for resource acquisition users to enter the audio virtual space through their accounts to listen to live songs. In this scenario, there are differences in the resources obtained by users from client A and client B: client A provides video live broadcast and user interaction services, while client B provides audio live broadcast and song listening services. However, resource acquisition users can use their accounts on both client A and client B to trigger corresponding account behaviors for the resource publishing account in the client, such as one or more account behaviors such as following, browsing, and evaluating.

[0056] After determining multiple different clients for multi-terminal joint modeling, for each client, multiple account attributes of resource acquisition users on the client and account behaviors triggered by resource publishing accounts of multiple resource acquisition users on the client can be obtained.

[0057] Among them, account attributes are the attribute information corresponding to the account of the resource acquisition user on the corresponding client. Account attributes can represent the characteristics of the user on the corresponding client, such as the basic portrait information of the account, including one or more of the account ID, age, gender and education level. Account attributes can also include statistical features for the account, such as the viewing time of the account, the number of virtual resources (such as virtual props) sent to the resource publishing account, etc. Account behavior can include one or more of the following: click behavior, viewing behavior, interactive behavior and payment behavior.

[0058] Furthermore, the account attributes of multiple resource acquisition users on different clients and the account behaviors triggered by the resource publishing accounts of multiple resource acquisition users for different clients can be input into the feature extraction module of the recommendation model for feature extraction to obtain the account attribute features and account behavior features corresponding to each resource acquisition user and each client.

[0059] S102, based on the account attribute characteristics and account behavior characteristics corresponding to each resource acquisition user and each client, obtain a first similarity of the account attribute characteristics of the same resource acquisition user in different clients, a second similarity of the account behavior characteristics of the same resource acquisition user in different clients, and a third similarity of the account attribute characteristics of different resource acquisition users in different clients; wherein different resource acquisition users are resource acquisition users who trigger the same account behavior for associated resource publishing accounts on different clients.

[0060] In some examples, the associated resource publishing accounts may be resource publishing accounts owned by the same resource publishing user on different clients.

[0061] After obtaining the account attribute characteristics and account behavior characteristics corresponding to each resource acquisition user and each client, multi-terminal app joint modeling can be performed based on contrastive learning. Contrastive learning uses similarities and dissimilarities to enable the model to map similar instances together closely in the latent feature space, while distinguishing different or unrelated instances. The embodiment of the present application uses contrastive learning to enhance the impact of overlapping user behaviors, including the similarity, difference, and interoperability of user behaviors.

[0062] Specifically, although multiple different apps have inconsistent feature systems, large differences in user behaviors, and few overlapping users, the inventors have found in practice that the same user has similar behaviors on multiple terminals, and the user attributes of different clients are different, while the user attributes are interoperable. Among them, the similarity of the same user means that the same user will have similar behaviors on different clients, which can be characterized by the similarity of the account behaviors of the same user on different clients; the differences in user attributes are introduced by users on different clients, which is caused by the differences in the essential attributes of different clients, and accordingly, the information carried by user attributes is naturally different; the interoperability of user attributes can be understood as when different users trigger the same behavior on the associated resource publisher, it can be considered that the user attributes of these different users are similar, that is, because the user attributes of different users are similar, they will perform the same account behavior on the account of the same resource publisher in different clients. In some embodiments, the degree of similarity of users can be characterized by the behavior type of the same account behavior.

[0063] Based on the above concept, on the one hand, the embodiment of the present application can obtain the similarity of the account attribute characteristics of the same resource acquisition user on different clients. For the convenience of distinction, this similarity is called the first similarity. On the other hand, the similarity of the account behavior characteristics of the same resource acquisition user on different clients can be obtained. For the convenience of distinction, this similarity is also called the second similarity. On the other hand, it is also possible to determine specific different resource acquisition users, and then obtain the similarity of the account attribute characteristics of different resource acquisition users on different clients. For the convenience of distinction, this similarity is called the third similarity, wherein the different resource acquisition users in this embodiment are resource acquisition users that trigger the same account behavior for associated resource publishing accounts on different clients.

[0064] S103, determining a representation enhancement loss according to the first similarity, the second similarity, and the third similarity, and adjusting a model parameter of the recommendation model according to the representation enhancement loss to obtain a trained recommendation model.

[0065] Specifically, the representation enhancement loss can be determined according to the first similarity, the second similarity and the third similarity. , and adjust the model parameters of the recommendation model according to the representation enhancement loss and the back propagation algorithm to obtain the trained recommendation model.

[0066] In practical applications, the recommendation model can be a multi-objective model, which is used to extract features from input information and determine the recommendation index of the resource publishing account to recommend the resource acquisition user when the specified account behavior is expected to be triggered. For example Figure 2The recommendation model shown includes a first recommendation module set for the first client and a second recommendation module set for the second client. The first recommendation module and the second recommendation module can both output the recommendation index for the resource publishing account when the resource acquisition user triggers click behavior, viewing behavior, interactive behavior and payment behavior on the resource publishing account. Among them, the structure of the recommendation module (the first recommendation module and / or the second recommendation module) can be flexibly set according to actual conditions. For example, the recommendation module may include a scene customized gate control network layer (Customized Gate Control, CGC) and a crosslayer. Among them, the scene CGC is a multi-objective processing method based on the end scene. The scene expert module and the public expert module are allocated according to different scenes. The crosslayer is used to process the input features based on the feature cross method, such as the cross method combined with the factorized machine (FM). In some examples, a representation layer (Representation Layer) can be set in the recommendation model, and the first similarity, the second similarity and the third similarity are determined by the representation layer.

[0067] In some exemplary embodiments, when training a recommendation model, a recommendation index label can also be introduced. The recommendation module loss is determined based on the difference between the recommendation index output by the first recommendation module and the second recommendation module and the recommendation index label. The model loss value is determined by combining the recommendation module loss and the representation enhancement loss to adjust the model parameters of the recommendation model.

[0068] In the above-mentioned recommendation model training method, by obtaining the first similarity of the account attribute characteristics of users with the same resources in different clients, the second similarity of the account behavior characteristics of users with the same resources in different clients, and the third similarity of the account attribute characteristics of users with different resources in different clients, the recommendation model parameters are adjusted according to the first similarity, the second similarity and the third similarity. The similarity of user behaviors on multiple terminals, the difference of user attributes and the interoperability of user attributes can be characterized through comparative learning, which helps the model to supplement and strengthen the characteristics of the current client user with information of similar users learned from other clients, and to push information accurately and flexibly on multiple clients based on the strengthened characteristics, effectively improving the efficiency of information recommendation.

[0069] In an exemplary embodiment, in step S103, determining the representation enhancement loss according to the first similarity, the second similarity and the third similarity may include the following steps:

[0070] S201, determining a weight of a third similarity according to account behaviors triggered by different resource acquisition users on different clients for publishing accounts of associated resources, and adjusting the third similarity according to the weight to obtain an adjusted third similarity.

[0071] In this step, different resource acquisition users can be determined, that is, resource acquisition users who have triggered account behaviors for associated resource publishing accounts in different clients. For example, for resource publishing user U, anchor accounts are opened in clients A, B, and C, namely anchor account 1, anchor account 2, and anchor account 3. Different resource acquisition users have triggered account behaviors for anchor account 1, anchor account 2, and anchor account 3 through their client accounts on the corresponding clients. Resource acquisition user 1 has triggered a follow behavior on anchor account 1, resource acquisition user 2 has triggered a payment behavior on anchor account 2, and resource acquisition user 3 has also triggered a follow behavior on anchor account 3. Therefore, resource acquisition user 1 and resource acquisition user 3 can be determined as different resource acquisition users, and the triggered same account behavior can be determined as a follow behavior.

[0072] In actual applications, there are differences in the degree of characterization of similar behaviors for different account behaviors. Specifically, for simple behaviors or behaviors that are of concern to business scenarios, such as interactive behaviors or payment behaviors, the behavioral similarities of different users can be constructed more accurately. For complex behaviors or behaviors that are not of concern to business scenarios, the accuracy in constructing behavioral similarities can be relatively reduced. For example, for specific browsing behaviors or click behaviors, since the browsing time and number of clicks of different users are difficult to accurately approximate, their weights can be adjusted down accordingly.

[0073] Based on this, after determining the account behaviors triggered by different resource acquisition users on different clients for associated resource publishing accounts, the weight of the third similarity can be determined based on the account behavior. In some examples, the weight of the third similarity can be positively correlated with the level of accuracy of the description of the account behavior, that is, the higher the level of accuracy of the description of a certain account behavior, the greater its weight. Then adjust the third similarity based on the weight to obtain the adjusted third similarity. For example, the product of the third similarity and the weight of the third similarity can be obtained, and the product is used as the adjusted third similarity.

[0074] S202, determining a representation enhancement loss according to the first similarity, the second similarity and the adjusted third similarity; the representation enhancement loss is positively correlated with the first similarity, and negatively correlated with the second similarity and the adjusted third similarity.

[0075] In this step, after obtaining the first similarity, the second similarity and the adjusted third similarity, the representation enhancement loss can be determined. In some exemplary embodiments, taking the joint modeling of two clients as an example, the representation enhancement loss It can be as follows:

[0076]

[0077] in, is the first similarity, is the second similarity, is the third similarity, is the weight of the third similarity, Get user U for resources in the first client The account attribute characteristics in Get user U for resources in the second client The account attribute characteristics in Get user U for resources in the first client Account behavior characteristics in Get user U for resources in the second client Account behavior characteristics in Get user U' for resource in first client The account attribute characteristics in Get user U'' for resources in the second client In the account attribute characteristics, the resource acquisition user U'' and the resource acquisition user U' are resource acquisition users who have triggered the same account behavior on the associated resource publishing account in the second client and the first client respectively. The resource acquisition user U' and the resource acquisition user U can be the same user or different users. The resource acquisition user U'' and the resource acquisition user U are different users.

[0078] In this embodiment, on the one hand, by making the representation enhancement loss positively correlated with the first similarity and negatively correlated with the second similarity and the third similarity, the account behavior characteristics output by the recommendation model for the same user on different clients can be made closer, and the account attribute characteristics of different users with the same attributes on different clients can be made closer, so as to facilitate closely mapping the behavior characteristics of the same user or the attribute characteristics of similar users together, and to make a more accurate customer of user characteristics and user behavior; on the other hand, it can also make the account attribute characteristics output by the recommendation model for the same user on different clients move away from each other, increase the differentiation of the account attribute characteristics in different clients, and facilitate the recommendation model to recommend information matching the client characteristics in combination with the attributes of a specific client; on the other hand, the influence of different account behaviors on the loss can be adjusted differentially by weighting, which is helpful to characterize the similarity of multiple user behaviors based on different accuracies.

[0079] In an exemplary embodiment, in step S201, determining the weight of the third similarity according to the behavior type of the same type of account behavior may include the following steps:

[0080] If there are multiple account behaviors triggered by different resource acquisition users on associated resource publishing accounts in different clients, determine the behavior priority of each account behavior; determine the target account behavior whose behavior priority meets the priority condition, and obtain the weight of the third similarity based on the weight corresponding to the target account behavior.

[0081] In actual applications, different resource acquisition users may have triggered multiple identical account behaviors on related resource publishing accounts. For example, in the above example, resource acquisition user 1 has triggered follow and watch behaviors on anchor account 1, and resource acquisition user 3 has also triggered follow and watch behaviors on anchor account 3. In the case where there are multiple triggered account behaviors, the behavior priority of each account behavior can be determined, and then the target account behavior that meets the priority conditions can be determined, and then the weight of the third similarity can be determined based on the weight of the target account behavior. For example, the behavior priorities of the four known account behaviors of payment (pay), interaction (interaction), watch (time) and click (click) are as follows:

[0082]

[0083] When both interactive behaviors and click behaviors exist, the interactive behavior with the highest priority can be determined as the target account behavior, and then the corresponding weight can be obtained.

[0084] In this embodiment, the weight of the third similarity is obtained by obtaining the weight corresponding to the target account behavior that meets the priority condition according to the behavior priority. When there are multiple account behaviors of the same type, the third similarity can be adjusted differentially according to the priority, thereby improving the matching of the third similarity with the actual scenario.

[0085] In an exemplary embodiment, the account attributes of multiple resource acquisition users on different clients include: the account attribute information of each resource acquisition user among the multiple resource acquisition users on the first client, and the account attribute information of each resource acquisition user on the second client; the account behavior triggered by the resource publishing accounts of the multiple resource acquisition users on different clients includes: the account behavior triggered by each resource acquisition user among the multiple resource acquisition users on the resource publishing account of the first client, and the account behavior triggered by each resource acquisition user on the resource publishing account of the second client.

[0086] In a specific implementation, a resource acquisition user may be determined, and the resource acquisition user holds a corresponding resource acquisition account on the first client and the second client at the same time. To facilitate the distinction between resource acquisition accounts on different clients, the resource acquisition account on the first client is referred to as the first client account, and the resource acquisition account on the second client is referred to as the second client account. It can be understood that since the first client and the second client are different clients, although the first client account and the second client account correspond to the same resource acquisition user, the account attributes and account behaviors corresponding to each account may be different.

[0087] Then, for the resource acquisition user, the relevant information on different clients can be obtained respectively. Specifically, the account attribute information of the resource acquisition user in the first client account and the account behavior triggered by the first client account for the resource publishing account of the first client can be obtained, and the account attribute information of the resource acquisition user in the second client account and the account behavior triggered by the second client account for the resource publishing account of the second client can be obtained at the same time. Then, the above information corresponding to each of the multiple resource acquisition users can be input into the feature extraction module of the recommendation model for feature extraction.

[0088] Take the resource publishing account as an example. Figure 2 For each resource acquisition user, the account ID of the resource acquisition user in client A and client B, the account ID of the anchor account that triggered the account behavior, and the account behavior in the live broadcast room in different clients can be spliced ​​together as the sample data of the resource acquisition user and input into the embedding vector layer (i.e., feature extraction module) of the resource acquisition model for feature extraction.

[0089] In this embodiment, by obtaining the account attribute information of the user in the first client and the second client and the account behavior triggered by the resource publishing account, the related data of the same user in the two clients can be obtained, providing a basis for cross-application multi-terminal modeling of the first client and the second client.

[0090] Accordingly, in some embodiments, the account attribute characteristics of the user acquired from the same resource on different clients include: the account attribute characteristics of the first client account and the account attribute characteristics of the second client account corresponding to the user acquired from the same resource; the account behavior characteristics of the user acquired from the same resource on different clients include: the account behavior characteristics of the first client account and the account behavior characteristics of the second client account corresponding to the user acquired from the same resource; the account attribute characteristics of the user acquired from different resources on different clients include: the account attribute characteristics of the first client account and the account attribute characteristics of the second client account corresponding to the user acquired from different resources.

[0091] In a specific implementation, after obtaining the account attribute characteristics of each first client account and the account attribute characteristics of each second client account output by the feature extraction module, the account attribute characteristics of the first client account and the account attribute characteristics of the second client account corresponding to the same resource acquisition user can be determined through device information or other identification information as the account attribute characteristics of the same resource acquisition user in different clients.

[0092] On the other hand, the account behavior characteristics of the first client account and the second client account corresponding to the same resource acquisition user can also be determined through device information or other identification information as the account behavior characteristics of the same resource acquisition user on different clients.

[0093] On the other hand, the resource publishing accounts corresponding to the same resource publishing user in different clients can be determined through device information or other identification information. For example, the resource publishing user has opened anchor accounts on both client A and client B, namely anchor account 1 and anchor account 2, and different resource acquisition users who have triggered the same type of account behavior on the above-mentioned anchor accounts can be determined, and the account attribute characteristics of different resource acquisition users on different clients can be obtained. For example, different resource acquisition users 1 have triggered the follow behavior on anchor account 1, and resource acquisition user 2 has triggered the follow behavior on anchor account 2. In this way, the account attribute characteristics of the first client account of resource acquisition user 1 on client A can be obtained, and the account attribute characteristics of the second client account of resource acquisition user 2 on client B can be obtained as the account attribute characteristics of different resource acquisition users on different clients.

[0094] Therefore, the account attribute characteristics and account behavior characteristics obtained above can provide a basis for building user behavior similarities, user attribute differences and user attribute interoperability between two clients, and the implementation method is simple, which helps to improve the efficiency of building the recommendation model.

[0095] Accordingly, step S102 may include the following steps:

[0096] Determine a first similarity between account attribute characteristics of a first client account and account attribute characteristics of a second client account corresponding to the same resource acquisition user; determine a second similarity between account behavior characteristics of a first client account and account behavior characteristics of a second client account corresponding to the same resource acquisition user; determine a third similarity between account attribute characteristics of a first client account and account attribute characteristics of a second client account corresponding to different resource acquisition users.

[0097] Specifically, for the difference of user attributes of the same user on different clients, the feature similarity between the account attribute characteristics of the first client account and the account attribute characteristics of the second client account can be determined, and the similarity can be used as the first similarity. For the similarity of user behaviors on multiple terminals, the feature similarity between the account behavior characteristics of the first client account corresponding to the same resource acquisition user and the account behavior characteristics of the second client account can be determined, and the feature similarity can be used as the second similarity. For the interoperability of different user attributes, the feature similarity between the account attribute characteristics of the first client account corresponding to different resource acquisition users and the account attribute characteristics of the second client account can be determined, and the feature similarity can be used as the third similarity. Thus, the account behavior characteristics output for the same user in the first client and the second client and the account attribute characteristics of similar users on the first client and the second client can be brought as close as possible, so that the user information of other clients can be used to supplement and strengthen the user information of the current client, while keeping the account attribute characteristics of the same user on different clients away, increasing the distinguishability of the account attribute characteristics under a specific client, and improving the accuracy of subsequent information recommendation.

[0098] In one embodiment, Figure 3 As shown, an information recommendation method is also provided. This embodiment uses the method applied to a server as an example for illustration. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0099] S301, in response to an account recommendation request for a first client target account of a resource acquisition user, determine multiple candidate resource publishing accounts of the first client; the account recommendation request is generated after a second client target account of the same resource acquisition user triggers a preset event in the second client.

[0100] Among them, the first client target account is the account of the first client, the second client target account is the account of the second client, and the first client target account and the second client target account correspond to the same resource acquisition user; the first client and the second client can be clients that provide training account data during recommendation model training.

[0101] In a specific implementation, a first client and a second client are deployed on the terminal. The resource acquisition user can perform related account behaviors on the first client through the first client target account he owns, and can also perform related account behaviors on the second client through the second client target account.

[0102] When the second client detects that the second client target account triggers a preset event, a corresponding account recommendation request may be obtained, and the account recommendation request may request to recommend a related resource publishing account for the first client target account owned by the resource acquisition user.

[0103] In some embodiments, the account recommendation request can be generated and sent by the first client after the second client target account of the same resource acquisition user triggers a preset event.

[0104] In a specific implementation, the first client and the second client communicate with the corresponding servers respectively, and the communications are independent of each other. For example, the first client communicates with the server corresponding to the first client, and the second client does not communicate with the server corresponding to the first client. At this time, the second client can avoid other clients from communicating with non-corresponding servers by triggering the first client to generate an account recommendation request, thereby achieving cross-client information push while ensuring the communication security and data security between the client and the server. For example, when "Music Application 1" detects that a user clicks Figure 4a When the button "Open" is clicked in the second client, it can be determined that the second client target account triggers a preset event, and triggers "K song application 2" to send an account recommendation request. For example, when "K song application 2" detects that the user clicks Figure 4b When the button 401 "Receive" is clicked, it can be determined that the second client target account triggers a preset event, and triggers "Music Application 1" to send an account recommendation request. According to the information recommendation method provided in the embodiment of the present application, the resource publishing account of "Music Application 1" is recommended and the corresponding live broadcast room is jumped (such as Figure 4c as shown).

[0105] Of course, the account recommendation request can also be directly generated and sent by the second client target account after the same resource acquisition user triggers a preset event in the second client target account of the second client. For example, when the server and the second client trust each other, the account recommendation request can be directly generated and sent to the server by the second client target account. In this way, the flexibility of the second client recommending the first client resource publishing account can be improved.

[0106] S302, for each candidate resource publishing account, obtain an account information combination corresponding to the candidate resource publishing account according to the resource publishing account attribute information of the candidate resource publishing account and the account attribute information and account behavior information of the first client target account.

[0107] After obtaining multiple candidate resource publishing accounts, for each of the resource publishing accounts, the resource publishing account attribute information of the candidate resource publishing account, as well as the account attribute information and account behavior information of the target account can be obtained to obtain the account information combination corresponding to the candidate resource publishing account. Figure 5 As shown, the first client requests the corresponding online anchor list for the first client target account, combines the first client target account, the client (i.e., the first client) and each anchor into a request pair, and then queries the feature platform of the corresponding client to obtain the account attribute information, account behavior information (such as follow sequence, comment sequence and payment sequence), client information and anchor account attribute information of the first client target account, and generates an account information combination (i.e., Figure 5 ).

[0108] S303, input the account information combination into the trained recommendation model, and the feature extraction module of the recommendation model obtains the feature combination corresponding to the account information combination, and determines the recommendation index of the candidate resource publishing account according to the feature combination.

[0109] The recommendation model may be trained by the recommendation model training method in one or more of the above embodiments.

[0110] After obtaining the account information combinations corresponding to multiple candidate resource publishing accounts, the account information combinations can be input into the trained recommendation model, and the feature extraction module of the recommendation model performs feature extraction on the account information combination to obtain the feature combination corresponding to the account information combination, and determines the recommendation index of the candidate resource publishing account based on the feature combination.

[0111] S304: Determine resource publishing account recommendation information to be provided to the first client target account according to the recommendation index of each candidate resource publishing account.

[0112] After obtaining the recommendation indexes of multiple candidate resource publishing accounts, the resource publishing account recommendation information provided to the first client target account can be obtained based on the multiple recommendation indexes. For example, in some embodiments, the recommendation model can score multiple targets in different scenarios at the same time. In actual applications, the scores of multiple targets can be automatically obtained based on the scenario, and then the scores of the multiple targets are weighted and fused to obtain the recommendation index, and then the fused scores are sorted from high to low, and then pushed to the user in this order for display.

[0113] The above-mentioned information recommendation method determines multiple candidate resource publishing accounts of the first client in response to an account recommendation request for a first client target account of a resource acquisition user, wherein the account recommendation request is generated after a preset event is triggered by a second client target account of the same resource acquisition user in a second client; and then for each candidate resource publishing account, the account information combination corresponding to the candidate resource publishing account can be obtained based on the resource publishing account attribute information of the candidate resource publishing account, as well as the account attribute information and account behavior information of the first client target account; and then the account information combination is input into a trained recommendation model, and a feature extraction module of the recommendation model obtains a feature combination corresponding to the account information combination, and determines the recommendation index of the candidate resource publishing account based on the feature combination; and then the resource publishing account recommendation information provided to the first client target account can be determined based on the recommendation index of each candidate resource publishing account. In this embodiment, since the recommendation model can be used to construct the similarity of user behaviors on multiple terminals, the differences in user attributes, and the interoperability of user attributes, after obtaining the user's account information combination on the first client, the recommendation model can use the information of similar or identical users learned from other terminals to supplement and strengthen the feature information of the current account, so that information recommendations can be made more accurately based on the strengthened features, thereby improving the accuracy and efficiency of information recommendations.

[0114] The inventors have found in practice that by making recommendations through the recommendation model provided in the embodiments of the present application, the information recommendation effect for different users can be effectively optimized, and users can also obtain resource information that matches their expected preferences and habits more quickly.

[0115] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0116] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 6As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store account data of multiple clients. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a recommendation model training method or an information recommendation method is implemented.

[0117] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a recommendation model training method or an information recommendation method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0118] Those skilled in the art will understand that Figure 6 and Figure 7The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0119] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0120] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0121] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0122] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0123] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0124] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A recommendation model training method, characterized in that: The method comprises: Input the account attributes of multiple resource acquisition users in different clients of different application types, and the account behaviors triggered by the resource publishing accounts of the multiple resource acquisition users for the different clients, into the feature extraction module of the recommendation model to obtain the account attribute features and account behavior features corresponding to each of the resource acquisition users and each of the clients; Based on the account attribute characteristics and account behavior characteristics corresponding to each of the resource acquisition users and each of the clients, a first similarity of the account attribute characteristics of the same resource acquisition user on different clients, a second similarity of the account behavior characteristics of the same resource acquisition user on different clients, and a third similarity of the account attribute characteristics of different resource acquisition users on different clients are obtained; wherein the different resource acquisition users are resource acquisition users who trigger the same account behavior on associated resource publishing accounts on different clients; A representation enhancement loss is determined according to the first similarity, the second similarity, and the third similarity, and a model parameter of the recommendation model is adjusted according to the representation enhancement loss to obtain a trained recommendation model.

2. The method according to claim 1, characterized in that The determining of the representation enhancement loss according to the first similarity, the second similarity and the third similarity includes: Determining a weight of the third similarity according to account behaviors triggered by different resource acquisition users on different clients for associated resource publishing accounts, and adjusting the third similarity according to the weight to obtain an adjusted third similarity; A representation enhancement loss is determined according to the first similarity, the second similarity and the adjusted third similarity; the representation enhancement loss is positively correlated with the first similarity and negatively correlated with the second similarity and the adjusted third similarity.

3. The method according to claim 2, characterized in that Determining the weight of the third similarity according to the account behaviors triggered by the different resource acquisition users on different clients for the associated resource publishing accounts includes: If the account behaviors triggered by the different resource acquisition users on the associated resource publishing accounts in different clients include multiple types, determining the behavior priority of each of the account behaviors; Determine the target account behavior whose behavior priority satisfies the priority condition, and obtain the weight of the third similarity according to the weight corresponding to the target account behavior.

4. The method according to claim 1, characterized in that: The account attributes of the multiple resource acquisition users on different clients include account attribute information of each of the multiple resource acquisition users on a first client, and account attribute information of each of the resource acquisition users on a second client; The account behaviors triggered by the multiple resource acquisition users for the resource publishing accounts of the different clients include: the account behaviors triggered by each of the multiple resource acquisition users for the resource publishing account of the first client, and the account behaviors triggered by each of the resource acquisition users for the resource publishing account of the second client.

5. The method according to claim 4, characterized in that The same resource acquisition user's account attribute characteristics on different clients include: account attribute characteristics of a first client account and account attribute characteristics of a second client account corresponding to the same resource acquisition user; The account behavior characteristics of the same resource acquisition user on different clients include: account behavior characteristics of a first client account and account behavior characteristics of a second client account corresponding to the same resource acquisition user; The account attribute characteristics of different clients of different resource acquisition users include: account attribute characteristics of a first client account and account attribute characteristics of a second client account corresponding to different resource acquisition users.

6. The method according to claim 5, characterized in that The method of acquiring the account attribute characteristics and account behavior characteristics corresponding to each of the resources and each of the clients, obtaining a first similarity of the account attribute characteristics of the user on different clients acquired by the same resource, a second similarity of the account behavior characteristics of the user on different clients acquired by the same resource, and a third similarity of the account attribute characteristics of the user on different clients acquired by different resources, includes: Determine a first similarity between the account attribute characteristics of the first client account and the account attribute characteristics of the second client account corresponding to the same resource acquisition user; Determine a second similarity between the account behavior characteristics of the first client account and the account behavior characteristics of the second client account corresponding to the same resource acquisition user; A third similarity between the account attribute characteristics of the first client account corresponding to different resource acquisition users and the account attribute characteristics of the second client account is determined.

7. An information recommendation method, characterized in that: The method comprises: In response to an account recommendation request for a first client target account of a resource acquisition user, determining a plurality of candidate resource publishing accounts of the first client; the account recommendation request is generated after a second client target account of the same resource acquisition user on the second client triggers a preset event; For each of the candidate resource publishing accounts, according to the resource publishing account attribute information of the candidate resource publishing account, and the account attribute information and account behavior information of the first client target account, obtain an account information combination corresponding to the candidate resource publishing account; The account information combination is input into a trained recommendation model, and a feature extraction module of the recommendation model obtains a feature combination corresponding to the account information combination, and determines a recommendation index of the candidate resource publishing account according to the feature combination; wherein the recommendation model is trained by the recommendation model training method according to any one of claims 1 to 6; According to the recommendation index of each of the candidate resource publishing accounts, resource publishing account recommendation information provided to the first client target account is determined.

8. The method according to claim 7, characterized in that The account recommendation request is generated and sent by the first client after the second client target account of the same resource acquisition user triggers a preset event, and the second client target account triggers the first client to generate and send; and / or, The account recommendation request is generated and sent by the second client target account after the same resource acquisition user triggers a preset event in the second client target account of the second client.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the training steps of the recommendation model training method described in any one of claims 1 to 6 or the steps of the information recommendation method described in any one of claims 7 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the training steps of the recommendation model training method according to any one of claims 1 to 6 or the steps of the information recommendation method according to any one of claims 7 to 8 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the training steps of the recommendation model training method according to any one of claims 1 to 6 or the steps of the information recommendation method according to any one of claims 7 to 8 are implemented.

Images