Content pushing method and device, computer device and storage medium

Abstract

The application relates to a content pushing method and device, computer equipment and a storage medium. It relates to the technical field of artificial intelligence and comprises the following steps: determining candidate pushing content of a target object and a content associated object; the content associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object; performing feature fusion on object features of the content associated object and content features of the candidate pushing content to obtain an associated social feature; determining an associated object of the target object; the associated object has an association relationship with the target object; performing feature fusion on object features of the associated object and object features of the target object to obtain a target fusion feature; and obtaining a matching degree between the candidate pushing content and the target object based on the associated social feature, the target fusion feature and content features of the candidate pushing content. The method can improve the accuracy of content pushing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to a content pushing method and device, computer equipment and storage medium. BACKGROUND

[0002] With the development of computer technology and Internet technology, more and more content is pushed through the network, such as pushing articles, videos or pictures through the network.

[0003] In the traditional content pushing method, when pushing content, the content pushed to the user is determined from each content to be pushed according to the content the user paid attention to at the historical time.

[0004] However, the content pushed to the user according to the content the user paid attention to at the historical time has great limitations, resulting in low accuracy of content pushing. SUMMARY

[0005] Therefore, it is necessary to provide a content pushing method and device, computer equipment, computer readable storage medium and computer program product capable of improving the accuracy of content pushing.

[0006] In one aspect, the present application provides a content pushing method. The method comprises: determining candidate pushing content of a target object and a content associated object; the content associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object; performing feature fusion on object features of the content associated object and content features of the candidate pushing content to obtain an associated social feature; determining an associated object of the target object; the associated object has an association relationship with the target object; performing feature fusion on object features of the associated object and object features of the target object to obtain a target fusion feature; predicting a matching degree between the candidate pushing content and the target object based on the associated social feature, the target fusion feature and the content features of the candidate pushing content; and the matching degree is used for content pushing processing of the target object.

[0007] In another aspect, the present application also provides a content pushing device. The device comprises: an object determining module, configured to determine a candidate pushing content of a target object and a content associated object; the content associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object; a feature determining module, configured to perform feature fusion on an object feature of the content associated object and a content feature of the candidate pushing content to obtain an associated social feature; an associated object determining module, configured to determine an associated object of the target object; the associated object has an association relationship with the target object; a feature fusion module, configured to perform feature fusion on an object feature of the associated object and an object feature of the target object to obtain a target fusion feature; and a matching degree determining module, configured to predict a matching degree between the candidate pushing content and the target object based on the associated social feature, the target fusion feature, and a content feature of the candidate pushing content; and the matching degree is used for content pushing processing of the target object.

[0008] In some embodiments, the content associated object is a plurality of, and the matching degree determining module is further configured to perform feature fusion on the associated social feature of each of the content associated objects for the target object to obtain a comprehensive association feature; and predict the matching degree between the candidate pushing content and the target object based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate pushing content.

[0009] In some embodiments, the matching degree determining module is further configured to, for each of the content objects, perform feature fusion on the associated social feature of the content object for the target object and an object feature of the target object to obtain an attention weight corresponding to the associated social feature; and perform feature fusion on each of the associated social features based on the attention weight corresponding to each of the associated social features to obtain a comprehensive association feature.

[0010] In some embodiments, the matching degree determining module is further configured to perform feature splicing based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate pushing content to obtain a matching degree prediction feature; and predict the matching degree between the candidate pushing content and the target object based on the matching degree prediction feature.

[0011] In some embodiments, the matching degree determining module is further configured to perform feature splicing on the comprehensive association feature, the target fusion feature, the content feature of the candidate pushing content, and the object feature of the target object to obtain a matching degree prediction feature.

[0012] In some embodiments, the object feature of the target object is a target object feature, and the object feature of the associated object is an associated object feature; the feature fusion module is further configured to splice the associated object feature and the target object feature to obtain a spliced feature; determine an attention coefficient of the target object feature to the associated object feature based on the spliced feature; and perform feature fusion on the associated object feature and the target object feature by using the attention coefficient to obtain a target fusion feature.

[0013] In some embodiments, the feature fusion module is further configured to perform dimension transformation on the associated object feature to obtain a first transformed feature, and perform dimension transformation on the target object feature to obtain a second transformed feature; splice the first transformed feature after the second transformed feature to obtain a spliced feature.

[0014] In some embodiments, the feature fusion module is further configured to perform weighted calculation on each feature value in the spliced feature to obtain an attention coefficient of the target object feature to the associated object feature.

[0015] In some embodiments, the associated social feature is extracted by a trained associated social feature extraction network, and the apparatus is further configured to determine a sample content associated object corresponding to a sample object; the sample content associated object has an interaction relationship with a sample push content and an associated relationship with the sample object; perform feature splicing on an object feature of the sample content associated object and a content feature of the sample push content to obtain a sample spliced feature; and train the to-be-trained associated social feature extraction network based on the sample spliced feature to obtain the trained associated social feature extraction network.

[0016] In some embodiments, the apparatus is further configured to determine an object attribute feature of the target object based on attribute information of the target object; determine an object state feature of the target object based on device state information corresponding to the target object; and determine the object feature of the target object based on the object attribute feature and the object state feature.

[0017] In some embodiments, the candidate push content belongs to a content social platform, and the candidate push content is multiple; the apparatus is further configured to obtain a matching degree between each candidate push content and the target object; select a target push content from each candidate push content based on the matching degree between each candidate push content and the target object; and push the target push content to a terminal of the target object after the terminal of the target object receives a triggering operation on a platform entry of the content social platform, so that the terminal of the target object displays the target push content on a page of the content social platform.

[0018] In another aspect, the present application provides a computer device. The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the content pushing method when executing the computer program.

[0019] In another aspect, the present application provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the content pushing method.

[0020] In another aspect, the present application provides a computer program product. The computer program product comprises a computer program, and the computer program is executed by a processor to implement the steps of the content pushing method.

[0021] The content pushing method, device, computer device, storage medium and computer program product determine the candidate pushing content of the target object and the content associated object, the content associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object, the object features of the content associated object and the content features of the candidate pushing content are fused to obtain an associated social feature, the associated object of the target object is determined, the associated object has an association relationship with the target object, the object features of the associated object and the object features of the target object are fused to obtain a target fusion feature, and the matching degree between the candidate pushing content and the target object is predicted based on the associated social feature, the target fusion feature and the content features of the candidate pushing content. The matching degree is used for content pushing processing of the target object. Since the content associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object, and the associated object has an association relationship with the target object, the associated social feature obtained by fusing the object features of the content associated object and the content features of the candidate pushing content, and the target fusion feature obtained by fusing the object features of the associated object and the object features of the target object, are used to predict the matching degree between the candidate pushing content and the target object, thereby improving the accuracy of the matching degree. Since the matching degree is used for content pushing processing of the target object, the accuracy of the content pushing is improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 An application environment diagram of the content pushing method in some embodiments;

[0023] Figure 2 A flowchart of the content pushing method in some embodiments;

[0024] Figure 3 An interface diagram of a platform entrance of a content social platform in some embodiments;

[0025] Figure 4 Interface diagram showing a platform entry of a content social platform in some embodiments;

[0026] Figure 5 Interface diagram showing target push content in some embodiments;

[0027] Figure 6 Structure diagram of a model for predicting matching degree in some embodiments;

[0028] Figure 7 Principle diagram of a graph attention mechanism in some embodiments;

[0029] Figure 8 Flowchart of training a model in some embodiments;

[0030] Figure 9 Flowchart of a content push method in some embodiments;

[0031] Figure 10 Structure block diagram of a content push device in some embodiments;

[0032] Figure 11 Internal structure diagram of a computer device in some embodiments;

[0033] Figure 12 Internal structure diagram of a computer device in some embodiments. DETAILED DESCRIPTION

[0034] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0035] The content push method provided by the embodiments of the present application can be applied in an application environment as shown in Figure 1 The terminal 102 communicates with the server 104 through a network. The data storage system can store data required to be processed by the server 104. The data storage system can be integrated on the server 104, or placed on a cloud or other server.

[0036] Specifically, the server 104 can determine candidate push content of a target object and a content-associated object, the content-associated object having an interaction relationship with the candidate push content and an association relationship with the target object, perform feature fusion on object features of the content-associated object and content features of the candidate push content to obtain an association social feature, determine an associated object of the target object, the associated object having an association relationship with the target object, perform feature fusion on object features of the associated object and object features of the target object to obtain a target fusion feature, and predict a matching degree between the candidate push content and the target object based on the association social feature, the target fusion feature, and the content features of the candidate push content, the matching degree being used for content push processing on the target object. The candidate push content can be at least one, and the target push content is selected from the candidate push contents based on the matching degree between each candidate push content and the target object, and the target push content is pushed to the terminal 102 of the target object.

[0037] The terminal 102 can be, but is not limited to, various desktop computers, notebook computers, smart phones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle-mounted device, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, an aircraft, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0038] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use, and processing of the relevant data need to comply with relevant national and regional laws, regulations, and standards. For example, the associated object of the target object and the content-associated object in the present application are obtained with the authorization of the target object and the associated object.

[0039] In some embodiments, the associated social feature is extracted by a trained associated social feature extraction network. The associated social feature extraction network can be based on artificial intelligence and machine learning, for example, can be a neural network model. Among them, artificial intelligence (AI) is the theory, method, technology and application system of using digital computer or digital computer controlled machine to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have perception, reasoning and decision-making functions.

[0040] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc. several major directions.

[0041] Machine learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a subject that studies how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and example-based learning.

[0042] With the research and progress of artificial intelligence technology, artificial intelligence technology has been researched and applied in many fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned vehicles, autonomous vehicles, drones, robots, smart medical care, intelligent customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0043] The scheme provided by the embodiments of the present application relates to neural network and other technologies of artificial intelligence, which is specifically explained by the following embodiments:

[0044] In some embodiments, as Figure 2As shown, a content pushing method is provided, which can be executed by a terminal or a server, and can also be executed by the terminal and the server together, and the method is applied to Figure 1 The server 104 in the server 104 is taken as an example for illustration, and the method comprises the following steps:

[0045] In step 202, the candidate pushing content of the target object and the content associated object are determined; the content associated object has an interactive relationship with the candidate pushing content and has an associated relationship with the target object.

[0046] The content can be any information that can be pushed, and can include at least one of video, voice, picture, text or multimedia information, for example, can be an article, a song, a movie, a short video, etc. The candidate pushing content is the candidate content for pushing, and the candidate pushing content can be, for example, the content for pushing in the content social platform. The content social platform can correspond to an independent application program, and can also correspond to a sub-application program. The parent application program corresponding to the sub-application program can be, for example, a social application program, and the social application program can be, for example, an instant messaging application program. The sub-application program can be, for example, a small program. The sub-application program is an application program that can be implemented in the environment provided by the parent application program. The parent application program is an application program that carries the sub-application program and provides an environment for the implementation of the sub-application program. The parent application program is a native application program. The native application program is an application program that can run directly on the operating system. The content social platform can be a platform that provides a Feeds content service.

[0047] The content social platform is a platform for socializing through content. Users interact with other users on the content social platform by writing articles, posting, or publishing videos, etc. Users can also interact with other users by commenting, following, reading, or replying to the content shared by other users on the content social platform. For example, the content social platform can be a platform for asking and answering questions through content. The candidate push content can be a question published in the question and answer platform. The content social platform can also be a platform for implementing instant sharing or dissemination interaction of content, such as a social media platform for implementing instant sharing or dissemination interaction of text, pictures, or videos. The content social platform can also be a video platform for creating and sharing videos, including but not limited to at least one of short videos, medium videos, or long videos. Short videos refer to short video clips, and the video duration of short videos is less than that of medium videos, and the video duration of medium videos is less than that of long videos. The type of video includes but is not limited to at least one of animation, game, or life record, etc. For example, the content social platform can be a short video social platform, and the candidate push content can be a short video created in the short video social platform. Various content can be presented on the content social platform. Users can log in to the content social platform through an account and share content in the content social platform. The content shared in the content social platform can be pushed to the terminal of other users, so that the other users can present the pushed content in the content social platform through the terminal, and can comment, like, or forward at least one of the presented content.

[0048] The platform entry of the content social platform can be an independent entry or located in other application programs, for example, the platform entry of the content social platform can be located in the instant messaging application program. The platform entry of the content social platform can be, for example, Figure 3 “abc points” in 302 of Figure 3 The interface displayed in the instant messaging application program is logged in by the account of the user “Zhang San”, and “abc points” is the name of the content social platform. When the content social platform corresponds to a sub-application belonging to an independent application program, the platform entry of the content social platform can also include the entry of the sub-application. The content social platform can also have a public account, and the platform entry of the content social platform can also include the entry of the public account. The entry of the public account can be located in the instant messaging application, and the entry of the public account can be used to enter the content social platform. The entry of the public account corresponding to the content social platform can be, for example, Figure 4 “Enter public account” displayed in the instant messaging application in 402.

[0049] The object refers to a user, the target object refers to a user who needs to determine the pushed content, the content associated object corresponding to the target object refers to an associated object of the target object that has an interaction relationship with the candidate push content, the association relationship refers to a social relationship, the associated object of the target object refers to a user who has a social relationship with the target object, including but not limited to at least one of the following: a friend of the target object in an instant messaging application, a contact in a mobile phone address book, and the like. The interaction relationship includes but is not limited to at least one of the following: clicking, forwarding, commenting, and the like.

[0050] The candidate push content can be at least one, and the candidate push content can be content published to the content social platform, can be content pre-stored in the server, and can also be content obtained by the server from other devices. The server determines the content to be pushed to the target object from the candidate push content. For each candidate push content, the content associated object corresponding to the target object can be at least one.

[0051] Specifically, the terminal of the target object can display a platform entry of the content social platform, and in response to a triggering operation on the platform entry of the content social platform, a content push request is sent to the server, and the content push request can carry an object identifier of the target object. After the server receives the content push request, the object identifier is extracted from the content push request, and the associated object of the target object is determined based on the object identifier of the target object. The server can store the social network information of the target object, and the social network information of the target object can be uniquely identified by the object identifier of the target object. The social network information of the target object can include the object identifiers of the respective associated objects corresponding to the target object, and the server can determine the social network information of the target object based on the object identifier of the target object, so as to obtain the object identifiers of the respective associated objects from the social network information of the target object.

[0052] In some embodiments, the server can determine the candidate push content corresponding to the target object by the object identifier of the obtained associated object. Specifically, the server can store content interaction information of each object, the content interaction information can be uniquely identified by an object identifier, and the content interaction information can store identifiers of contents having an interaction relationship with the object, including but not limited to identifiers of contents that have been clicked, forwarded or commented by the object. The server can obtain the content interaction information of the associated object based on the object identifier of the associated object, obtain the content identifier from the content interaction information of the associated object, determine the content represented by the obtained content identifier as the candidate push content of the target object, and determine the associated object having an interaction relationship with the candidate push content as the content associated object. That is, the content associated object can be a subset of all associated objects of the target object. For example, for user A, his friends user B and user C like article Z, then user B and user C are the content associated objects of user A on article Z, and user B and user C are not all associated objects of user A, but only two of them.

[0053] In some embodiments, the server can determine the content having an interaction relationship with the associated object of the target object from the content library, determine the content having an interaction relationship with the associated object of the target object in the content library as the candidate push content of the target object, and determine the associated object having an interaction relationship with the candidate push content as the content associated object. The content library includes a plurality of contents. It should be noted that the content associated object can be the same or different for different candidate push contents, and for each candidate push content, there is at least one content associated object.

[0054] Step 204, the object features of the content associated object and the content features of the candidate push content are fused to obtain the associated social feature.

[0055] Each content associated object can correspond to an associated social feature for the target object, the associated social feature of the content associated object for the target object, the associated social feature of the content associated object for the target object, and the feature reflecting the social relationship between the content associated object and the target object.

[0056] The features can include three types of features, namely, variable-length sparse features, fixed-length sparse features, and dense features. The variable-length sparse features generally refer to sequence features composed of identifiers, including but not limited to at least one of image-text click sequences (sequences composed of identifiers of clicked images or articles), video play sequences (sequences composed of identifiers of played videos), author like sequences (sequences composed of identifiers of liked authors), author follow sequences (sequences composed of identifiers of followed authors), and the like. The fixed-length sparse features generally refer to identifier features, including but not limited to at least one of user identifiers, user gender identifiers, user province identifiers, user city identifiers, article identifiers, article author identifiers, or article classification identifiers, and the like. The dense features generally refer to continuous value features, including but not limited to at least one of user age, user historical image-text click times, user historical video play times, user historical like author numbers, user historical follow author numbers, article total browse times, article total like times, or article total follow times, and the like. The object features can include at least one of the variable-length sparse features, the fixed-length sparse features, or the dense features. The content features can include at least one of the variable-length sparse features, the fixed-length sparse features, or the dense features. The variable-length sparse features, the fixed-length sparse features, and the dense features can be in the form of Embedding (embedded features). Among them, the dense features are embedded by weighting the Embedding with continuous values. Taking the dense feature of “age” as an example, the “age” corresponds to a preset Embedding, and if the user’s age is 59 years old, the result of 59*the preset Embedding is determined as the feature corresponding to the age. Embedding the dense features in the form of Embedding can enhance the expression ability of the dense features and improve the memory of the model.

[0057] Specifically, for each content-associated object, the server can perform feature fusion on the object features of the content-associated object and the content features of the candidate push content to obtain the associated social features of the content-associated object for the target object, and the feature fusion includes but is not limited to at least one of feature splicing or feature multiplication. For example, the server can multiply the object features of the content-associated object and the content features of the candidate push content to obtain the associated social features, wherein the multiplication refers to multiplying the feature values at the same positions, and the result of the multiplication is arranged in order according to the positions of the feature values to obtain the associated social features. Alternatively, the server can splice the object features of the content-associated object and the content features of the candidate push content to obtain the associated social features.

[0058] In some embodiments, the server can obtain a trained associated social feature extraction network, the associated social feature extraction network being configured to extract an associated social feature. The server can concatenate the object feature of the content associated object and the content feature of the candidate push content to obtain a first concatenated feature, and input the first concatenated feature into the trained associated social feature extraction network to perform feature extraction, and obtain the associated social feature of the content associated object with respect to the target object. The associated social feature extraction network can be a neural network, for example, a DNN (Deep Neural Networks).

[0059] In step 206, the associated object of the target object is determined; the associated object and the target object have an association relationship.

[0060] The association relationship can be a social relationship, and the associated object of the target object can be a user having a social relationship with the target object, including but not limited to at least one of a friend of the target object in an instant messaging application, a contact in a mobile phone address book, etc. The associated object of the target object can be at least one.

[0061] Specifically, the server can store social network information of the target object, the social network information of the target object can be uniquely identified by an object identifier of the target object, and the social network information of the target object can include object identifiers of each associated object corresponding to the target object. The server can determine the social network information of the target object based on the object identifier of the target object, and obtain the object identifiers of each associated object from the social network information of the target object, so as to determine each associated object of the target object.

[0062] In step 208, the object feature of the associated object and the object feature of the target object are fused to obtain a target fusion feature.

[0063] Specifically, the feature fusion includes but is not limited to at least one of concatenation or weighted calculation, etc. The server can concatenate the object feature of the associated object and the object feature of the target object, and determine the concatenation result as the target fusion feature. For example, the server can concatenate the object feature of the associated object after the object feature of the target object, and determine the concatenation result as the target fusion feature.

[0064] In some embodiments, the associated object is multiple, and the server can perform weighted calculation on the object feature of each associated object and the object feature of the target object, obtain the target fusion feature based on the result of the weighted calculation, for example, can determine the result of the weighted calculation as the target fusion feature, or can further process the result of the weighted calculation to obtain the target fusion feature.

[0065] At step 210, a matching degree between the candidate push content and the target object is predicted based on the associated social feature, the target fusion feature, and the content feature of the candidate push content; the matching degree is used for content push processing of the target object.

[0066] The matching degree is used to represent the acceptance degree of the target object to the candidate push content, i.e., the interest degree; the greater the matching degree, the greater the acceptance degree, i.e., the greater the interest degree.

[0067] Specifically, the server can splice the associated social feature, the target fusion feature, and the content feature of the candidate push content to obtain a matching degree prediction feature, and predict the matching degree between the candidate push content and the target object based on the matching degree prediction feature. The matching degree prediction feature uses a feature for predicting the matching degree.

[0068] In some embodiments, the content-associated object is multiple, i.e., the object that has an interaction relationship with the candidate push content and has a social relationship with the target object is multiple. Multiple means at least two. The server can obtain the associated social feature of each content-associated object for the target object, splice each associated social feature, the target fusion feature, and the content feature of the candidate push content to obtain a matching degree prediction feature, and predict the matching degree between the candidate push content and the target object based on the matching degree prediction feature.

[0069] In some embodiments, the candidate push content is multiple, and the server can obtain the matching degree between each candidate push content and the target object, select a target push content from each candidate push content based on the matching degree between each candidate push content and the target object, and push the target push content to the terminal of the target object. The terminal of the target object can display the target push content. Taking the content in the content social platform as an example, the terminal sends a content push request to the server in response to a click operation on 302 in Figure 3 or a click operation on 402 in Figure 4 The server returns the target push content to the terminal of the target object in response to the content push request, and the terminal of the target object displays the target push content. The target push content is, for example, 502 in Figure 5

[0070] ​In the content pushing method, the candidate pushing content of the target object and the content-associated object are determined, the content-associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object, the object feature of the content-associated object and the content feature of the candidate pushing content are fused to obtain an association social feature, the association object of the target object is determined, the association object has an association relationship with the target object, the object feature of the association object and the object feature of the target object are fused to obtain a target fusion feature, and the matching degree between the candidate pushing content and the target object is predicted based on the association social feature, the target fusion feature, and the content feature of the candidate pushing content. The matching degree is used for content pushing processing of the target object. Since the content-associated object has an interaction relationship with the candidate pushing content and an association relationship with the target object, and the association object has an association relationship with the target object, the association social feature obtained by fusing the object feature of the content-associated object and the content feature of the candidate pushing content, and the target fusion feature obtained by fusing the object feature of the association object and the object feature of the target object, are used to predict the matching degree between the candidate pushing content and the target object, thereby improving the accuracy of the matching degree. Since the matching degree is used for content pushing processing of the target object, the accuracy of the content pushing is improved.

[0071] In some embodiments, the content-associated object is multiple, and predicting the matching degree between the candidate pushing content and the target object based on the association social feature, the target fusion feature, and the content feature of the candidate pushing content includes: fusing the association social feature of each content-associated object for the target object to obtain a comprehensive association feature; and predicting the matching degree between the candidate pushing content and the target object based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate pushing content.

[0072] Specifically, the feature fusion can be a weighted calculation of the features. The server can determine a weight of each associated social feature, and perform a weighted calculation on each associated social feature based on the weight of each associated social feature to obtain a comprehensive association feature. The weight of the associated social feature can be determined based on the object feature of the target object, for example, the weight of the associated social feature can be an attention weight of the target object feature to the associated social feature, where the target object feature refers to the object feature of the target object. The attention weight of the target object feature to the associated social feature can be calculated by an attention calculation network, which can be at least one of a Feed-Forward Attention network or a Self-Attention network, and the like. For example, the object feature of the associated object is the associated object feature, the content associated object is 3, which are content associated object 1, content associated object 2 and content associated object 3, the object feature of the content associated object 1 is the associated object feature 1, the object feature of the content associated object 2 is the associated object feature 2, and the object feature of the content associated object 3 is the associated object feature 3, as shown in FIG. 17, the social association module is a module for generating a comprehensive association feature, the social association module includes an attention calculation network and an associated social coupling network, the associated social coupling network is the associated social feature extraction network described above, the associated social coupling network obtains the associated social feature 1 based on the associated object feature 1 and the content feature, the associated social feature 1 is the associated social feature of the content associated object 1 for the target object, the associated social coupling network obtains the associated social feature 2 based on the associated object feature 2 and the content feature, the associated social feature 2 is the associated social feature of the content associated object 2 for the target object, the associated social coupling network obtains the associated social feature 3 based on the associated object feature 3 and the content feature, the associated social feature 3 is the associated social feature of the content associated object 3 for the target object, the associated social feature 1, the associated social feature 2, the associated social feature 3 and the target object feature are input into the attention calculation network, the attention calculation network calculates the attention weight of the associated social feature 1, the attention weight of the associated social feature 2 and the attention weight of the associated social feature 3, and then performs a weighted calculation on the three associated social features based on the three attention weights to obtain a comprehensive association feature. Figure 6

[0073] ​In some embodiments, the candidate push content corresponds to associated social information, and the associated social information includes an identification of an object having an interaction relationship with the candidate push content. The content association object of the target object can be determined through the associated social information. The associated object can also be referred to as a social friend, and the content association object can also be referred to as an associated social friend. Taking an article as an example, the social association module is responsible for modeling the associated social information of the current article. Through the associated social coupling network, the social association module can extract the associated social embedding (i.e., the associated social feature) of each associated social friend based on the article information and the associated social information. Then, through the attention mechanism, the associated social embedding of each associated social friend is weighted by assigning a certain attention. The social friend having a greater association degree with the current user can often obtain more attention, because the social friend having a greater association degree usually occupies a more important proportion in the conversion of the current user. Through the associated social coupler and the attention mechanism, the final overall associated social embedding (i.e., the comprehensive association feature) can be obtained, which will be input to the subsequent output layer to improve the modeling efficiency.

[0074] In some embodiments, the server can splice the comprehensive association feature, the target fusion feature, and the content feature of the candidate push content to obtain a matching degree prediction feature, and predict the matching degree between the candidate push content and the target object based on the matching degree prediction feature. For example, the server can obtain a trained matching degree generation network, the matching degree generation network is used to generate a corresponding matching degree based on the matching degree prediction feature, input the matching degree prediction feature into the trained matching degree generation network, and generate the matching degree between the candidate push content and the target object.

[0075] In this embodiment, the associated social features of each content association object for the target object are fused to obtain a comprehensive association feature, and the matching degree between the candidate push content and the target object is predicted based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate push content, so that the matching degree is predicted based on the comprehensive association feature obtained by fusing a plurality of associated social features, and the accuracy of predicting the matching degree is improved.

[0076] In some embodiments, the feature fusion of the associated social features of each content association object for the target object to obtain a comprehensive association feature includes: for each content object, the associated social features of the target object are fused to obtain the attention weight corresponding to the associated social features; and based on the attention weight corresponding to each associated social feature, the feature fusion of each associated social feature is performed to obtain the comprehensive association feature.

[0077] Specifically, for each associated social feature, the server can perform feature fusion on the associated social feature and the object feature of the target object to obtain an attention weight corresponding to the associated social feature. For example, the associated social feature is a vector, and the object feature is also a vector. In the case where the dimensions of the associated social feature and the object feature are consistent, the server can perform dot product operation on the vectors of the associated social feature and the object feature of the target object, and determine the result of the dot product operation as the attention weight corresponding to the associated social feature. In the case where the dimensions of the associated social feature and the object feature are inconsistent, the server can transform the dimension of the associated social feature or the dimension of the object feature so that the dimensions of the associated social feature and the object feature are consistent, and then perform dot product operation to obtain the attention weight.

[0078] In this embodiment, for each associated social feature of the target object for each content object, the server performs feature fusion on the associated social feature and the object feature of the target object to obtain an attention weight corresponding to the associated social feature. Based on the attention weight corresponding to each associated social feature, the server performs feature fusion on each associated social feature to obtain a comprehensive association feature. Since a larger attention weight indicates a greater degree of association between the associated social feature and the object feature of the target object, performing feature fusion on each associated social feature based on the attention weight can improve the accuracy of the comprehensive association feature.

[0079] In some embodiments, predicting the matching degree between the candidate push content and the target object based on the comprehensive association feature, the target fusion feature, and the content feature of the push content comprises: performing feature splicing based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate push content to obtain a matching degree prediction feature; and predicting the matching degree between the candidate push content and the target object based on the matching degree prediction feature.

[0080] Specifically, the server can perform feature splicing on the comprehensive association feature, the target fusion feature, and the content feature of the candidate push content to obtain a matching degree prediction feature, and input the matching degree prediction feature into a matching degree generation network to predict the matching degree between the candidate push content and the target object. The content feature refers to the feature of the candidate push content, and the target object feature refers to the object feature of the target object.

[0081] In this embodiment, feature splicing is performed based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate push content to obtain a matching degree prediction feature, and the matching degree between the candidate push content and the target object is predicted based on the matching degree prediction feature, thereby improving the accuracy of the matching degree prediction feature and thus improving the accuracy of the predicted matching degree.

[0082] In some embodiments, the feature splicing is performed based on the comprehensive association feature, the target fusion feature, and the content feature of the candidate push content, to obtain the matching degree prediction feature.

[0083] Specifically, the server can perform feature splicing on the comprehensive association feature, the target fusion feature, the content feature of the candidate push content, and the target object feature to obtain the matching degree prediction feature, and input the matching degree prediction feature into the matching degree generation network to predict the matching degree between the candidate push content and the target object.

[0084] As shown in Figure 6 , the target fusion feature, the target object feature, the content feature of the candidate push content, and the comprehensive association feature are spliced and input into the matching degree generation network to predict the matching degree between the candidate push content and the target object. The matching degree generation network can also be referred to as an output layer (Output Layer). The input of the output layer is composed of Embeddings modeled by the shared embedding layer, the graph attention module, and the social association module of the lower layer, respectively. The multiple Embeddings are spliced, nonlinearly transformed by the multi-layer fully connected layer of the output layer, and finally the probability is calculated using the Sigmoid function to obtain the matching degree. The matching degree output by the output layer is a probability, for example, the matching degree can be the reading target probability of the user for the article, and the reading target probability includes but is not limited to at least one of the click rate or conversion rate of the article, etc.

[0085] Figure 6 Each entire network in the matching degree prediction model can be referred to as a matching degree prediction model, and each network in the matching degree prediction model can be independently trained or jointly trained. When training the matching degree prediction model, the sample object and the associated object of the sample object can be obtained, the associated object having an interaction relationship with the sample content is determined from the associated object of the sample object for the sample content, to obtain the sample content associated object of the sample object, and the real matching degree between the sample object and the sample content is obtained. The object feature of the sample object, the object feature of the associated object, and the object feature of the sample content associated object are input into the matching degree prediction model to obtain the predicted matching degree. The model loss value is determined based on the real matching degree and the predicted matching degree, the parameters of the network in the matching degree prediction model are adjusted in the direction of reducing the model loss value, until the model converges, to obtain the trained matching degree prediction model, and the matching degree is predicted using the trained matching degree prediction model. Taking the candidate push content as a social article as an example, the flowchart of training the matching degree prediction model is as shown in Figure 8As shown, including feature and label reporting, sample set construction, training set and validation set division, model training and model export stages. Among them, in the feature and label reporting stage, based on the client's point reporting, the user's behavior log in the recommended product is reported in real time, the background service is called based on the request, and all feature data of the user at the request time is reported, including variable length sparse features, fixed length sparse features and dense features. The label extraction of the user's behavior log is used. The label refers to the long click behavior of the user when seeing social articles, such as the user's click behavior when entering the social article detail page for more than 3 seconds, the feature data is preprocessed, and the feature data will have a certain missing and abnormal, which can be processed by mean filling missing, rejecting abnormal samples and other operations. In the sample set construction stage, based on the extracted label and processed feature, it is spliced and constructed into a usable sample set. In the training set and validation set division stage, based on a certain proportion, the sample set is divided into a training set and a validation set, the training set is used to train the model, and the validation set is used to verify the model effect. Based on the divided training set and validation set, the model is trained and the model effect is evaluated. In the model training stage, based on the early stop (Early Stop) strategy, the model is trained until the effect converges, that is, the available model can be exported in the model export stage, which is used for online service. The loss function used in model training can be any loss function, for example, it can be a cross entropy loss function (CrossEntropy Loss Function). The formula of cross entropy loss function is:

[0086]

[0087] Where L is the total loss function; N is the sample number; yi is the true value of the i-th sample; pi is the predicted value of the i-th sample. The optimization algorithm can be Adam optimization algorithm, which performs first-order gradient optimization on the objective function through Adam optimization algorithm, estimates the first and second moments, calculates the adaptive learning rate of different parameters, so that the model can converge faster, find the local optimal solution of the optimization problem, and achieve good optimization performance.

[0088] The trained matching degree prediction model can be used for online prediction. For example, when the server determines that pushing is needed, the server can calculate all features required by the model based on user basic information, user social information, article information, and context state information, construct multiple to-be-predicted samples, input the multiple to-be-predicted samples into the model concurrently, enable the model to quickly perform prediction, obtain a prediction score of the to-be-predicted sample, and perform article recommendation according to the prediction score of the to-be-predicted sample in descending order, and preferentially recommend articles with higher scores (i.e., matching degrees). The context state information refers to device state information. The device state information includes, but is not limited to, at least one of the following: a network state (2G / 3G / 4G / 5G / WIFI) currently used by the user, a current mobile phone operating system (Android / iOS) of the user, and a current mobile phone brightness. The user basic information includes, but is not limited to, at least one of the following: age, gender, or a region to which the user belongs. The user social information includes social friends (i.e., associated objects of the target object) of the user. The article information includes, but is not limited to, at least one of the following: an identifier, a theme, a character, a plot, or key information of the article.

[0089] In this embodiment, the matching degree prediction features are obtained by splicing the associated features, the target fusion features, content features of the candidate push content, and the target object features, so that the matching degree prediction features further include the features of the target object, further improving the accuracy of the matching degree prediction features and the accuracy of the predicted matching degrees.

[0090] In some embodiments, the object features of the target object are target object features, and the object features of the associated object are associated object features. The target fusion features are obtained by fusing the object features of the associated object and the object features of the target object, including: splicing the associated object features and the target object features to obtain spliced features; determining an attention coefficient of the target object features to the associated object features based on the spliced features; and fusing the associated object features and the target object features by using the attention coefficient to obtain the target fusion features.

[0091] The attention coefficient of the target object features to the associated object features is used to reflect the degree of association of the associated object features to the target object features. The greater the attention coefficient, the greater the degree of association of the associated object features to the target object features.

[0092] Specifically, the server can concatenate the associated object feature and the target object feature to obtain a second concatenated feature, for example, the associated object feature can be concatenated after the target object feature to obtain the second concatenated feature, the target object feature is concatenated with the target object feature to obtain a third concatenated feature, the attention coefficient of the target object feature to the associated object feature is determined based on the second concatenated feature to obtain a first attention coefficient, the attention coefficient of the target object feature to the target object feature is determined based on the third concatenated feature to obtain a second attention coefficient, the first attention coefficient is the attention coefficient of the target object feature to the associated object feature, and the second attention coefficient is the attention coefficient of the target object feature to itself. The server can take the first attention coefficient as the weight of the associated object feature, and take the second attention coefficient as the weight of the target object feature, and perform weighted calculation on the associated object feature and the target object feature to obtain a target fusion feature.

[0093] In the embodiment, the associated object feature and the target object feature are concatenated to obtain a concatenated feature, the attention coefficient of the target object feature to the associated object feature is determined based on the concatenated feature, and the associated object feature and the target object feature are fused by using the attention coefficient to obtain a target fusion feature, thereby improving the accuracy of the target fusion feature.

[0094] In some embodiments, concatenating the associated object feature and the target object feature to obtain a concatenated feature comprises: performing dimension transformation on the associated object feature to obtain a first transformed feature, and performing dimension transformation on the target object feature to obtain a second transformed feature; and concatenating the first transformed feature after the second transformed feature to obtain the concatenated feature.

[0095] The dimensions of the first transformed feature and the second transformed feature can be the same. The concatenated feature is a second concatenated feature.

[0096] Specifically, the server can obtain a dimension transformation matrix, perform dimension transformation on the associated object feature by using the dimension transformation matrix to obtain a first transformed feature, and perform dimension transformation on the target object feature by using the dimension transformation matrix to obtain a second transformed feature. For example, the dimensions of the associated object feature h1 and the target object feature h0 are consistent, both are vectors with a length of M, the dimension of the dimension transformation matrix W is N*M, then the first transformed feature h1_1=W h1, and the second transformed feature h0_1=W h0, so that the first transformed feature h1_1 and the second transformed feature h0_1 are both vectors with a length of N. The dimension transformation matrix can be obtained by training, for example, the dimension transformation matrix can be a parameter matrix in a trained fusion feature generation network, and the fusion feature generation network is used to generate a target fusion feature.

[0097] In some embodiments, after obtaining the first transformed feature and the second transformed feature, the server can concatenate the first transformed feature after the second transformed feature to obtain a second concatenated feature.

[0098] In this embodiment, the first transformed feature is obtained by performing dimension transformation on the associated object feature, the second transformed feature is obtained by performing dimension transformation on the target object feature, and the concatenated feature is obtained by concatenating the first transformed feature after the second transformed feature. Therefore, the dimension of the concatenated feature can be flexibly set, and the flexibility of feature processing is improved.

[0099] In some embodiments, determining the attention coefficient of the target object feature to the associated object feature based on the concatenated feature includes performing weighted calculation on each feature value in the concatenated feature to obtain the attention coefficient of the target object feature to the associated object feature.

[0100] Specifically, the server can perform weighted calculation on each feature value in the concatenated feature (i.e., the second concatenated feature), and determine the result of the weighted calculation as the attention coefficient of the target object feature to the associated object feature.

[0101] In some embodiments, the server can obtain a coefficient generation vector, the dimension of the coefficient generation vector is consistent with the dimension of the concatenated feature (i.e., the second concatenated feature), perform vector dot product operation on the coefficient generation vector and the second concatenated feature, and determine the result of the dot product operation as the attention coefficient of the target object feature to the associated object feature. The coefficient generation vector can be obtained by training, for example, the coefficient generation vector is a parameter vector in the trained fusion feature generation network, for example, it can be a vector composed of parameters of a single-layer feedforward neural network in the fusion feature generation network.

[0102] In some embodiments, the fusion feature generation network includes at least one network, each network includes a plurality of feature fusion networks, and each network includes a feature fusion network for fusing features. The feature fusion network can include a dimension transformation network and a coefficient generation network, the parameters of the dimension transformation network form a dimension transformation matrix, the parameters of the coefficient generation network form a coefficient generation vector, and the feature fusion network is further used for performing weighted calculation on the associated object feature and the target object feature by using the attention coefficient.

[0103] In some embodiments, the fusion feature generation network includes a feature fusion network. The server can input the target object feature into a dimension transformation network of the feature fusion network for dimension transformation to obtain a second transformed feature, input the associated object feature into the dimension transformation network of the feature fusion network for dimension transformation to obtain a first transformed feature, concatenate the first transformed feature after the second transformed feature to obtain a second concatenated feature, input the second concatenated feature into a coefficient generation network of the feature fusion network to generate a first attention coefficient of the target object feature to the associated object feature, concatenate the second transformed feature with the second transformed feature to obtain a third concatenated feature, input the third concatenated feature into the coefficient generation network of the feature fusion network to generate a second attention coefficient of the target object feature to the target object feature. The server can take the first attention coefficient as the weight of the associated object feature, and take the second attention coefficient as the weight of the target object feature, and perform weighted calculation on the associated object feature and the target object feature to obtain a target fusion feature. When the associated object feature is multiple, the second attention coefficient of the target object feature to each associated object feature is generated respectively.

[0104] In some embodiments, the fusion feature generation network includes a multi-layer feature fusion network. Multi-layer refers to at least two layers, each layer including a feature fusion network corresponding to the target object feature and a feature fusion network corresponding to each associated object feature. The feature fusion network corresponding to the target object feature is used to update the target object feature to obtain an updated target object feature, and the feature fusion network corresponding to the associated object feature is used to update the associated object feature to obtain an updated associated object feature.

[0105] Taking the k-layer feature fusion network included in the fusion feature generation network as an example, in the feature fusion network corresponding to the target object feature, the target object feature or the updated associated object feature is taken as the main feature, and the associated object feature is taken as the auxiliary feature. For the feature fusion network A1 corresponding to the target object feature in the first layer, the server can input the target object feature (i.e., the main feature) into the dimension transformation network of the feature fusion network A1 to perform dimension transformation to obtain a second transformed feature, input the associated object feature (i.e., the auxiliary feature) into the dimension transformation network of the feature fusion network to perform dimension transformation to obtain a first transformed feature, splice the first transformed feature after the second transformed feature to obtain a second spliced feature, input the second spliced feature into the coefficient generation network of the feature fusion network to generate a first attention coefficient of the target object feature to the associated object feature, splice the second transformed feature with the second transformed feature to obtain a third spliced feature, and input the third spliced feature into the coefficient generation network of the feature fusion network to generate a second attention coefficient of the target object feature to the target object feature. The server can take the first attention coefficient as the weight of the associated object feature, take the second attention coefficient as the weight of the target object feature, and perform weighted calculation on the associated object feature and the target object feature to obtain the updated target object feature in the first layer.

[0106] In a feature fusion network corresponding to a certain associated object feature, the associated object feature or the updated associated object feature is used as the main feature, and the target object feature and other associated object features are used as auxiliary features. For the feature fusion network B1 corresponding to the associated object feature h1 in the first layer, the server can input the associated object feature h1 (i.e., the main feature) into the dimension transformation network of the feature fusion network B1 for dimension transformation to obtain the third transformed feature. Each auxiliary feature is input into the dimension transformation network of the feature fusion network B1 for dimension transformation to obtain the fourth transformed feature corresponding to each auxiliary feature. The fourth transformed feature is concatenated after the third transformed feature to obtain the fourth concatenated feature. The fourth concatenated feature is input into the coefficient generation network of the feature fusion network to generate the third attention coefficient of the main feature to the auxiliary feature. The third transformed feature is concatenated with the third transformed feature to obtain the fifth concatenated feature. The fifth concatenated feature is input into the coefficient generation network of the feature fusion network to generate the fourth attention coefficient of the main feature to the main feature. The third attention coefficient is used as the weight of the auxiliary feature, and the fourth attention coefficient is used as the weight of the main feature. The main feature and the auxiliary feature are weighted and calculated to obtain the updated associated object feature h1 in the first layer. Similarly, for the feature fusion network B2 in the second layer corresponding to the associated object feature h1, the associated object feature h1 updated in the first layer is used as the main feature, and the target object feature updated in the first layer and other associated object features updated in the first layer are used as auxiliary features to obtain the associated object feature h2 updated in the first layer.

[0107] Similarly, for the feature fusion network A2 corresponding to the target object features in the second layer, the server can use the target object features updated in the first layer as the primary features and the associated object features updated in the first layer as auxiliary features to obtain the target object features updated in the second layer. For the feature fusion network Ak corresponding to the target object features in the k-th layer, the server can use the target object features updated in the (k-1)-th layer as the primary features and the associated object features updated in the (k-1)-th layer as auxiliary features to obtain the target object features updated in the k-th layer, and determine the target fusion features updated in the k-th layer as the target fusion features.

[0108] In some embodiments, the feature fusion network can be a graph attention network, which utilizes a graph attention mechanism to update features. For example... Figure 7 The diagram shown illustrates the principle of the attention mechanism. Figure 7 Used for Updating is understandable. As the main feature, To The updated result for Its own weight, 、 、 and is an auxiliary feature, is the weight of , is the weight of , is the weight of , is the weight of , , , , and , .

[0109] For example, as shown in Figure 6 , the graph attention module (GATModule) includes a multi-layer graph attention network, refers to the object feature of the target object, refers to the object feature of the associated object, |N(u)| refers to the number of associated objects, u is the abbreviation of user, and N is the abbreviation of Neighbor, refers to the updated target object feature output by the feature fusion network corresponding to the target object feature in the last layer, i.e., the target fusion feature. The graph attention module can extract friend information in the user's social network. The feature fusion network A1 corresponding to the target object feature in the first layer is, for example, the graph attention network 1_u in Figure 6 , and the feature fusion network A1 corresponding to the target object feature in the second layer is, for example, the graph attention network 2_u in Figure 6 . Among them, in the propagation process of the network, the attention mechanism can calculate the hidden state of each node, and by assigning different attention on different neighbor nodes, it can distinguish the sharing degree of different friend information to the self representation, and then update the hidden state of the current node. The input of the graph attention module is the shared hidden layer user Embedding, which can serve as the initial hidden state of each node in the graph attention module. By stacking multiple layers of graph attention networks, the friend information in the user's social network is extracted and aggregated layer by layer, and then the overall social Embedding of the user in the entire social network is modeled, which will be input to the subsequent output layer to improve the final modeling target.

[0110] In this embodiment, the attention coefficient of the target object feature to the associated object feature is obtained by weighting calculation on each feature value in the spliced feature, so that the attention coefficient is quickly determined, and the efficiency of calculating the attention coefficient is improved.

[0111] In some embodiments, the association social feature is extracted by the trained association social feature extraction network, and the method further comprises: determining a sample content association object corresponding to the sample object; the sample content association object has an interaction relationship with the sample push content and has an association relationship with the sample object; performing feature splicing on the object feature of the sample content association object and the content feature of the sample push content to obtain sample spliced features; training the to-be-trained association social feature extraction network based on the sample spliced features to obtain the trained association social feature extraction network.

[0112] In some embodiments, the association social feature is extracted by the trained association social feature extraction network, and the method further comprises: determining a sample content association object corresponding to the sample object; the sample content association object has an interaction relationship with the sample push content and has an association relationship with the sample object; performing feature splicing on the object feature of the sample content association object and the content feature of the sample push content to obtain sample spliced features; training the to-be-trained association social feature extraction network based on the sample spliced features to obtain the trained association social feature extraction network.

[0113] Specifically, the server can perform feature splicing on the object feature of the sample content association object and the content feature of the sample push content to obtain sample spliced features, input the sample spliced features into the to-be-trained association social feature extraction network to predict the association social feature, obtain the predicted association social feature, input the predicted association social feature into the association degree generation network to predict the association degree, obtain the predicted association degree, adjust the network parameters of the association social feature extraction network based on the difference between the predicted association degree and the real association degree, and determine that the training is completed when the association social feature extraction network converges, to obtain the trained association social feature extraction network. If the association degree generation network is an untrained network, the network parameters of the association degree generation network can also be adjusted based on the difference between the predicted association degree and the real association degree, and the training is determined to be completed when both the association degree generation network and the association social feature extraction network converge, to obtain the trained association social feature extraction network.

[0114] In this embodiment, the sample content association object corresponding to the sample object is determined, the sample content association object has an interaction relationship with the sample push content and has an association relationship with the sample object, feature splicing is performed on the object feature of the sample content association object and the content feature of the sample push content to obtain sample spliced features, and the to-be-trained association social feature extraction network is trained based on the sample spliced features to obtain the trained association social feature extraction network. Thus, a network capable of predicting the association social feature is trained, and the efficiency and accuracy rate of predicting the association social feature are improved.

[0115] In some embodiments, the method further comprises: determining, based on the attribute information of the target object, an object attribute feature corresponding to the target object; determining, according to the device state information corresponding to the target object, an object state feature corresponding to the target object; and determining, based on the object attribute feature and the object state feature, an object feature of the target object.

[0116] The attribute information includes, but is not limited to, at least one of age, gender, province, city, and the like. The object attribute feature refers to a feature representing the attribute information. For example, the object attribute feature can include a feature representing the age. The device state information includes, but is not limited to, at least one of a network state (2G / 3G / 4G / 5G / WIFI) currently used by the user, a mobile phone operating system (Android / iOS) currently used by the user, and a mobile phone brightness currently used by the user. The object state feature refers to a feature representing the device state information of the object. For example, the object state feature can include a feature representing the network state.

[0117] Specifically, the server can concatenate the object attribute feature and the object state feature, and determine the concatenation result as the object feature of the target object.

[0118] In this embodiment, the object state feature corresponding to the target object is determined according to the device state information corresponding to the target object, and the object feature of the target object is determined based on the object attribute feature and the object state feature, thereby improving the richness of information contained in the object feature and further improving the accuracy of the prediction matching degree.

[0119] In some embodiments, the candidate push content belongs to a content social platform, and the candidate push content is multiple. The method further comprises: obtaining a matching degree between each candidate push content and the target object; selecting a target push content from the candidate push contents based on the matching degree between each candidate push content and the target object; and pushing the target push content to the terminal of the target object after the terminal of the target object receives a triggering operation on a platform entry of the content social platform, so that the terminal of the target object displays the target push content on a page of the content social platform.

[0120] Specifically, the target push content can be at least one. The server can determine the candidate push content with the largest matching degree as the target push content. Alternatively, the server can determine the candidate push content with a matching degree greater than a matching degree threshold as the target push content. The matching degree threshold can be preset or set as needed.

[0121] In some embodiments, the server can arrange the candidate push contents in descending order of the matching degrees to obtain a candidate push content sequence, and the greater the matching degree, the higher the ranking of the candidate push content in the candidate push content sequence. The server can determine the candidate push content before the content ranking threshold as the target push content. The content ranking refers to the ranking of the candidate push content in the candidate push content sequence, and the ranking threshold can be preset or set as needed, for example, can be any one of 2 or 4.

[0122] In some embodiments, the terminal of the target object displays a platform entry of the content social platform, and when the terminal detects a triggering operation on the platform entry of the content social platform, sends a content push request to the server. The server determines the target push content in response to the content push request, and returns the target push content to the terminal. The terminal displays the target push content on a page of the content social platform.

[0123] In this embodiment, the candidate push content belongs to the content social platform, the matching degree between each candidate push content and the target object is obtained, the target push content is selected from each candidate push content based on the matching degree between each candidate push content and the target object, and after the terminal of the target object receives a triggering operation on the platform entry of the content social platform, the target push content is pushed to the terminal of the target object, so that the terminal of the target object displays the target push content on a page of the content social platform, and the accuracy of content push on the content social platform is improved.

[0124] In some embodiments, as shown in Figure 9 A content push method is provided, in this embodiment, the content is a social article, the method can be executed by a terminal or a server, and can also be executed by a terminal and a server together. Taking the case that the method is applied to the server as an example, the method includes the following steps:

[0125] Step 902, determining a candidate push article of a target object and an article associated object, the article associated object having an interaction relationship with the candidate push article and an association relationship with the target object.

[0126] Step 904, inputting the object features of the article associated object and the article features of the candidate push article into an associated social feature extraction network in a social association module to extract the associated social features of the article associated object for the target object.

[0127] Step 906, for the associated social features of each article object for the target object, performing feature fusion on the associated social features and the object features of the target object to obtain the attention weight corresponding to the associated social features.

[0128] At step 908, the feature fusion is performed on the respective associated social features based on the attention weights of the respective associated social features, to obtain a comprehensive associated feature.

[0129] At step 910, an associated object of the target object is obtained, and the associated object has an association relationship with the target object.

[0130] At step 912, a graph attention network in a graph attention module is used to perform dimension transformation on the associated object feature to obtain a first transformed feature, and perform dimension transformation on the target object feature to obtain a second transformed feature, and then the first transformed feature is spliced after the second transformed feature to obtain a spliced feature.

[0131] At step 914, the attention coefficient of the target object feature to the associated object feature is obtained by performing weighted calculation on each feature value in the spliced feature, and the feature fusion is performed on the associated object feature and the target object feature by using the attention coefficient, to obtain a target fusion feature.

[0132] At step 916, the comprehensive associated feature, the target fusion feature, the article feature of the candidate push article, and the target object feature are spliced to obtain a matching degree prediction feature.

[0133] At step 918, the matching degree prediction feature is input into a matching degree generation network, and the matching degree between the candidate push article and the target object is predicted.

[0134] At step 920, the target push article is selected from the respective candidate push articles based on the matching degree between each candidate push article and the target object, and the target push article is pushed to the terminal of the target object.

[0135] The embodiment realizes a reading target improvement method based on social network information, solves the problem of inefficient use of social network information in the recommendation system sorting scene, and further improves the reading target conversion efficiency of the recommendation system with social network information. The application mines social network information, deeply extracts the implicit semantics in the social network information through the graph attention module and the social association module, fully models the conversion probability of users with social relationships, improves the prediction accuracy of the recommendation model, and further improves the key indicators such as DAU (Daily Active User, daily active user number), consumption time, user retention, etc. in the recommendation system. Compared with the traditional scheme, the application not only utilizes user social information, but also deeply mines social network information by using special modeling means. The application introduces a novel network module design based on a multi-layer graph attention network, so that the graph attention module can extract friend information in the user's social network, and the associated social information of the current article is modeled through the social association module, the user's social network information is fully explored, and the accuracy and efficiency of the recommendation system are improved. In the application, the matching degree generation network (output layer) can also be designed as a network for implementing multi-target tasks, for example, a multi-target task based on Share-Bottom (shared bottom layer feature), MMoE (Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts, modeling task relationships in multi-task learning with multi-gate mixture-of-experts), PLE (Progressive Layered Extraction, shared structure design of progressive layered extraction model).

[0136] The content pushing method provided by the application can be applied to a scene of pushing content in the form of text, and can also be applied to a scene of pushing videos or short videos. Taking the application in the short video scene as an example, the application of the content pushing method in this application scene is as follows: the candidate pushing content can be a short video of a short video social platform, when the server determines that a short video needs to be pushed to a target user, the candidate pushing short video of the target user and the short video associated user can be determined, the short video associated user has an interaction relationship with the candidate pushing short video and has a social relationship with the target user. The server can determine the associated social features of the short video associated user for the target user based on the features of the short video associated user and the short video features of the candidate pushing short video, determine the associated user of the target user, fuse the features of the associated user and the features of the target object to obtain target fusion features, and predict the matching degree between the candidate pushing short video and the target user based on the associated social features, the target fusion features and the video features of the candidate pushing short video. The candidate pushing short video with high matching degree is pushed to the terminal of the target user.

[0137] It should be understood that although the steps in the flowcharts involved in the embodiments described above are shown in sequence according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of the steps is not strictly limited in sequence, and the steps can be executed in other orders. Moreover, at least some of the steps in the flowcharts involved in the embodiments described above can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of the steps or stages is not necessarily sequential, but can be alternately or alternately executed with at least part of other steps or steps or stages in other steps.

[0138] Based on the same inventive concept, the embodiments of the present application also provide a content pushing device for implementing the content pushing method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more content pushing device embodiments provided below can refer to the limitations of the content pushing method described above, which will not be repeated here.

[0139] In some embodiments, as shown in FIG. 10, a content pushing device is provided, including an object determining module 1002, a feature determining module 1004, an associated object determining module 1006, a feature fusion module 1008, and a matching degree determining module 1010, wherein: Figure 10 The object determining module 1002 is configured to determine candidate pushing content of a target object and a content associated object. The content associated object has an interaction relationship with the candidate pushing content and has an association relationship with the target object.

[0140] The feature determining module 1004 is configured to perform feature fusion on object features of the content associated object and content features of the candidate pushing content to obtain associated social features.

[0141] The associated object determining module 1006 is configured to determine an associated object of the target object. The associated object has an association relationship with the target object.

[0142] The feature fusion module 1008 is configured to perform feature fusion on object features of the associated object and object features of the target object to obtain target fusion features.

[0143]

[0144] ​The matching degree determination module 1010 is configured to predict a matching degree between the candidate push content and the target object based on the association social feature, the target fusion feature, and a content feature of the candidate push content; and the matching degree is used for content push processing on the target object.

[0145] In some embodiments, the content association object is a plurality of objects, and the matching degree determination module is further configured to perform feature fusion on the association social feature of each content association object for the target object to obtain a comprehensive association feature; and predict the matching degree between the candidate push content and the target object based on the comprehensive association feature, the target fusion feature, and a content feature of the candidate push content.

[0146] In some embodiments, the matching degree determination module is further configured to, for each content object, perform feature fusion on the association social feature of the target object and an object feature of the target object to obtain an attention weight corresponding to the association social feature; and perform feature fusion on each association social feature based on the attention weight corresponding to each association social feature to obtain a comprehensive association feature.

[0147] In some embodiments, the matching degree determination module is further configured to perform feature splicing based on the comprehensive association feature, the target fusion feature, and a content feature of the candidate push content to obtain a matching degree prediction feature; and predict the matching degree between the candidate push content and the target object based on the matching degree prediction feature.

[0148] In some embodiments, the matching degree determination module is further configured to perform feature splicing on the comprehensive association feature, the target fusion feature, a content feature of the candidate push content, and a target object feature to obtain a matching degree prediction feature.

[0149] In some embodiments, the object feature of the target object is a target object feature, and the object feature of the association object is an association object feature; the feature fusion module is further configured to splice the association object feature and the target object feature to obtain a spliced feature; determine an attention coefficient of the target object feature to the association object feature based on the spliced feature; and perform feature fusion on the association object feature and the target object feature by using the attention coefficient to obtain a target fusion feature.

[0150] In some embodiments, the feature fusion module is further configured to perform dimension transformation on the association object feature to obtain a first transformed feature, and perform dimension transformation on the target object feature to obtain a second transformed feature; and splice the first transformed feature after the second transformed feature to obtain a spliced feature.

[0151] In some embodiments, the feature fusion module is further configured to perform weighted calculation on each feature value in the spliced feature to obtain an attention coefficient of the target object feature to the association object feature.

[0152] In some embodiments, the association social feature is extracted by the trained association social feature extraction network, and the apparatus is further configured to determine a sample content association object corresponding to the sample object; the sample content association object has an interaction relationship with the sample push content and has an association relationship with the sample object; perform feature splicing on object features of the sample content association object and content features of the sample push content to obtain sample spliced features; train the association social feature extraction network to be trained based on the sample spliced features to obtain the trained association social feature extraction network.

[0153] In some embodiments, the apparatus is further configured to determine an object attribute feature corresponding to the target object based on attribute information of the target object; determine an object state feature corresponding to the target object according to device state information of the target object; and determine the object feature of the target object based on the object attribute feature and the object state feature.

[0154] In some embodiments, the candidate push content belongs to a content social platform, and the candidate push content is multiple; the apparatus is further configured to obtain a matching degree between each candidate push content and the target object; select a target push content from each candidate push content based on the matching degree between each candidate push content and the target object; and push the target push content to the terminal of the target object after the terminal of the target object receives a triggering operation on a platform entry of the content social platform, so that the terminal of the target object displays the target push content on a page of the content social platform.

[0155] Each module in the content push apparatus can be realized by software, hardware, and a combination thereof in whole or in part. Each module can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform operations corresponding to each module.

[0156] In some embodiments, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in FIG. 8. Figure 11As shown in the figure. 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 the 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 ability. 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 the data involved in the content pushing method. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with the external terminal through network connection. The computer program is executed by the processor to realize a content pushing method.

[0157] In some embodiments, a computer device which can be a terminal is provided, and its internal structure diagram can be as shown in the figure. Figure 12 As shown in the figure. 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 the 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 ability. 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 the data involved in the content pushing method. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with the external terminal through network connection. The computer program is executed by the processor to realize a content pushing method.

[0158] Those skilled in the art can understand that, Figure 11 and Figure 12The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0159] In some embodiments, a computer device is provided, including a memory and a processor, the memory has stored therein a computer program, and the processor implements the steps in the content pushing method described above when executing the computer program.

[0160] In some embodiments, a computer readable storage medium is provided, having stored thereon a computer program, and the computer program implements the steps in the content pushing method described above when executed by a processor.

[0161] In some embodiments, a computer program product is provided, including a computer program, and the computer program implements the steps in the content pushing method described above when executed by a processor.

[0162] 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 for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant national and regional laws, regulations and standards.

[0163] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric 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 but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device, etc., without being limited thereto.

[0164] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0165] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A content push method characterized by, The method comprises: determining candidate push content of a target object and a content-associated object; the content-associated object has an interaction relationship with the candidate push content and an association relationship with the target object; performing feature fusion on object features of the content-associated object and content features of the candidate push content to obtain an associated social feature; determining an associated object of the target object; the associated object has an association relationship with the target object; performing feature fusion on object features of the associated object and object features of the target object to obtain target fusion features; based on the associated social feature, the target fusion feature and the content features of the candidate push content, determining a matching degree prediction feature, and based on the matching degree prediction feature, predicting a matching degree between the candidate push content and the target object; the matching degree is used for content push processing of the target object.

2. The method of claim 1, wherein, The content-associated object is multiple, and the matching degree prediction feature is determined based on the associated social feature, the target fusion feature and the content features of the candidate push content, comprising: performing feature fusion on the associated social feature of each content-associated object for the target object to obtain comprehensive association features; determining a matching degree prediction feature based on the comprehensive association features, the target fusion features and the content features of the candidate push content.

3. The method of claim 2, wherein, The feature fusion on the associated social feature of each content-associated object for the target object to obtain comprehensive association features comprises: for each content-associated object, performing feature fusion on the associated social feature of the target object to obtain an attention weight corresponding to the associated social feature; based on the attention weight of each associated social feature, performing feature fusion on each associated social feature to obtain comprehensive association features.

4. The method of claim 2, wherein, The determination of the matching degree prediction feature based on the comprehensive association features, the target fusion features and the content features of the candidate push content comprises: performing feature splicing based on the comprehensive association features, the target fusion features and the content features of the candidate push content to obtain a matching degree prediction feature.

5. The method of claim 4, wherein, The feature splicing based on the comprehensive association features, the target fusion features and the content features of the candidate push content to obtain a matching degree prediction feature comprises: performing feature splicing on the comprehensive association features, the target fusion features, the content features of the candidate push content and the object features of the target object to obtain a matching degree prediction feature.

6. The method of claim 1, wherein, The object features of the target object are target object features, and the object features of the associated object are associated object features; the feature fusion on the object features of the associated object and the object features of the target object to obtain target fusion features comprises: splicing the associated object features and the target object features to obtain spliced features; determining an attention coefficient of the target object features to the associated object features based on the spliced features; Fuse the associated object feature and the target object feature based on the attention coefficient to obtain a target fusion feature.

7. The method of claim 6, wherein, The splicing of the associated object feature and the target object feature to obtain a spliced feature comprises: dimensional transformation of the associated object feature to obtain a first transformed feature, and dimensional transformation of the target object feature to obtain a second transformed feature; splicing the first transformed feature after the second transformed feature to obtain a spliced feature.

8. The method of claim 6, wherein, The determination of the attention coefficient of the target object feature to the associated object feature based on the spliced feature comprises: weighting calculation of each feature value in the spliced feature to obtain the attention coefficient of the target object feature to the associated object feature.

9. The method of claim 1, wherein, The associated social feature is extracted by a trained associated social feature extraction network, and the method further comprises: determining a sample object corresponding sample content associated object; the sample content associated object has an interactive relationship with a sample push content and an associated relationship with the sample object; feature splicing of the object feature of the sample content associated object and the content feature of the sample push content to obtain a sample spliced feature; training the to-be-trained associated social feature extraction network based on the sample spliced feature to obtain the trained associated social feature extraction network.

10. The method of claim 1, wherein, The method further comprises: determining an object attribute feature corresponding to the target object based on attribute information of the target object; determining an object state feature corresponding to the target object according to device state information of the target object; determining an object feature of the target object based on the object attribute feature and the object state feature.

11. The method of claim 1, wherein, The candidate push content belongs to a content social platform, and the candidate push content is multiple; the method further comprises: obtaining a matching degree between each candidate push content and the target object; selecting a target push content from each candidate push content based on the matching degree between each candidate push content and the target object; after receiving a trigger operation for a platform entrance of the content social platform on the terminal of the target object, pushing the target push content to the terminal of the target object, so that the terminal of the target object displays the target push content on the page of the content social platform.

12. A content pusher device, characterized by The device comprises: an object determination module for determining a candidate push content of a target object and a content associated object; the content associated object has an interactive relationship with the candidate push content and an associated relationship with the target object; a feature determination module for feature fusion of an object feature of the content associated object and a content feature of the candidate push content to obtain an associated social feature; an associated object determination module for determining an associated object of the target object; the associated object has an associated relationship with the target object; a feature fusion module for feature fusion of an object feature of the associated object and an object feature of the target object to obtain a target fusion feature; The matching degree determination module is configured to determine a matching degree prediction feature based on the association social feature, the target fusion feature, and a content feature of the candidate push content, and predict a matching degree between the candidate push content and the target object based on the matching degree prediction feature. The matching degree is used for content push processing of the target object.

13. The apparatus of claim 12, wherein, The content association objects are multiple, and the matching degree determination module is further configured to: perform feature fusion on the association social feature of each content association object for the target object to obtain a comprehensive association feature; obtain a matching degree prediction feature based on the comprehensive association feature, the target fusion feature, and a content feature of the candidate push content.

14. The apparatus of claim 13, wherein, The matching degree determination module is further configured to: perform feature fusion on the association social feature of each content association object for the target object to obtain a comprehensive association feature; obtain a matching degree prediction feature based on the comprehensive association feature, the target fusion feature, and a content feature of the candidate push content.

15. The apparatus of claim 13, wherein, The matching degree determination module is further configured to: perform feature fusion on the association social feature of each content association object for the target object to obtain a comprehensive association feature; 16. The apparatus of claim 15, wherein, obtain a matching degree prediction feature based on the comprehensive association feature, the target fusion feature, and a content feature of the candidate push content. The matching degree determination module is further configured to:

17. The apparatus of claim 12, wherein, perform feature fusion on the association social feature of each content association object for the target object to obtain a comprehensive association feature; obtain a matching degree prediction feature based on the comprehensive association feature, the target fusion feature, and a content feature of the candidate push content. The object feature of the target object is a target object feature, and the object feature of the association object is an association object feature. The feature fusion module is further configured to: splice the association object feature and the target object feature to obtain a spliced feature; 18. The apparatus of claim 17, wherein, determine an attention coefficient of the target object feature to the association object feature based on the spliced feature; perform feature fusion on the association object feature and the target object feature by using the attention coefficient to obtain a target fusion feature. The feature fusion module is further configured to:

19. The apparatus of claim 17, wherein, perform dimension transformation on the association object feature to obtain a first transformed feature, and perform dimension transformation on the target object feature to obtain a second transformed feature; splice the first transformed feature after the second transformed feature to obtain a spliced feature.

20. The apparatus of claim 12, wherein, The feature fusion module is further configured to: perform weighted calculation on each feature value in the spliced feature to obtain an attention coefficient of the target object feature to the association object feature. The association social feature is extracted by a trained association social feature extraction network. The apparatus is further configured to: determine a sample content association object corresponding to a sample object; the sample content association object has an interaction relationship with a sample push content and has an association relationship with the sample object; perform feature splicing on an object feature of the sample content association object and a content feature of the sample push content to obtain a sample spliced feature; and perform feature fusion on the sample spliced feature and the target fusion feature to obtain the matching degree prediction feature. Train the associated social feature extraction network to be trained based on the sample splicing features, to obtain a trained associated social feature extraction network.

21. The apparatus of claim 12, wherein, The device is also used for: Determining an object attribute feature corresponding to the target object based on attribute information of the target object; Determining an object state feature corresponding to the target object according to device state information of the target object; Determining an object feature of the target object based on the object attribute feature and the object state feature.

22. The apparatus of claim 12, wherein, The candidate push contents belong to a content social platform, and the candidate push contents are multiple; the device is also used for: Obtaining a matching degree between each candidate push content and the target object; Selecting a target push content from each candidate push content based on the matching degree between each candidate push content and the target object; After a terminal of the target object receives a trigger operation for a platform entrance of the content social platform, pushing the target push content to the terminal of the target object, so that the terminal of the target object displays the target push content on a page of the content social platform. 23.A computer device, comprising a memory and a processor, wherein the memory stores a computer program. The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 11.

24. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 11.

25. A computer program product comprising a computer program, characterised in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 11.

Images