Information recommendation method and device, computer device, and storage medium

Abstract

The application relates to an information recommendation method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining an account interaction feature corresponding to a social network to which a target account belongs; the account interaction feature is obtained according to interaction behaviors between each social account in the social network; performing feature propagation mapping on an account feature corresponding to the target account through the account interaction feature, obtaining a propagation account feature corresponding to the target account according to a result of the feature propagation mapping; obtaining an information interaction feature of each to-be-recommended information; the information interaction feature is generated based on an account feature of a social account that has generated an interaction operation on the corresponding to-be-recommended information in the social network; matching the propagation account feature and each information interaction feature, and recommending a target information determined from each to-be-recommended information to a terminal corresponding to the target account according to a matching result. The method can improve the accuracy of information recommendation.

Description

TECHNICAL FIELD

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

[0002] With the rapid development of computer and Internet technology, it brings a lot of convenience to people's life, and also brings a large amount of data information, so that people are difficult to obtain the required information from a large amount of data information. For example, for a large amount of audio and video, articles, pictures, web pages and other data information in the Internet, it is difficult for people to quickly filter and obtain the required information. In order to let the user obtain the required information more accurately, the personalized information recommendation is often used to recommend the information meeting the needs of the user to the user.

[0003] At present, information recommendation mainly analyzes the historical behavior of the user, such as analyzing the video browsing behavior of the user, and recommending the corresponding information according to the user preference obtained by analysis. However, when the historical behavior data of the user is less, the analysis of the historical behavior cannot effectively obtain the user preference, resulting in limited accuracy of information recommendation for the user. SUMMARY

[0004] Therefore, it is necessary to provide an information recommendation method, device, computer equipment and storage medium capable of improving the accuracy of information recommendation to solve the above technical problems.

[0005] An information recommendation method, the method comprising:

[0006] obtaining account interaction features corresponding to a social network to which a target account belongs; the account interaction features are obtained according to the interaction behaviors between each social account in the social network;

[0007] mapping the account features corresponding to the target account through the account interaction features, and obtaining the propagation account features corresponding to the target account according to the mapping result of the feature propagation;

[0008] obtaining information interaction features of each to-be-recommended information; the information interaction features are generated based on the account features of the social accounts which have generated interaction operations on the corresponding to-be-recommended information in the social network;

[0009] matching the propagation account features and each information interaction feature, and recommending the target information determined from each to-be-recommended information to the terminal corresponding to the target account according to the matching result.

[0010] An information recommendation device, the device comprising:

[0011] The account interaction feature acquisition module is configured to acquire account interaction features corresponding to the social network to which the target account belongs, the account interaction features being obtained according to interaction behaviors between social accounts in the social network;

[0012] The feature propagation mapping module is configured to perform feature propagation mapping on the account features corresponding to the target account by using the account interaction features, and to obtain propagation account features corresponding to the target account according to a result of the feature propagation mapping;

[0013] The information interaction feature acquisition module is configured to acquire information interaction features of each to-be-recommended information, the information interaction features being generated based on account features of social accounts that have performed interaction operations on the corresponding to-be-recommended information in the social network;

[0014] The target information recommendation module is configured to match the propagation account features and the information interaction features, and to recommend a target information determined from the to-be-recommended information to a terminal corresponding to the target account according to a matching result.

[0015] In one of the embodiments, the feature propagation mapping module includes:

[0016] The target interaction feature extraction module is configured to extract target account interaction features corresponding to the target account from the account interaction features;

[0017] The interaction account feature acquisition module is configured to acquire account features of an interaction social account corresponding to the target account interaction features;

[0018] The propagation mapping iteration module is configured to perform iterative feature propagation mapping on the account features corresponding to the target account based on the target account interaction features, the account features of the interaction social account, and propagation mapping parameters, to obtain the result of the feature propagation mapping;

[0019] The propagation account feature determination module is configured to determine the propagation account features corresponding to the target account according to the result of the feature propagation mapping.

[0020] In one of the embodiments, the propagation mapping iteration module includes:

[0021] The current account feature determination module is configured to determine the account features corresponding to the target account as current account features;

[0022] The propagation mapping processing module is configured to perform feature propagation mapping on the current account feature based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping, to obtain a result of the current iteration feature propagation mapping; and return the result of the current iteration feature propagation mapping as the current account feature, and return the step of performing feature propagation mapping on the current account feature based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping, to obtain a result of the current iteration feature propagation mapping.

[0023] In one of the embodiments, the propagation mapping processing module comprises:

[0024] The feature propagation module is configured to perform feature propagation on the current account feature based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping, to obtain a feature propagation result.

[0025] The nonlinear mapping module is configured to perform nonlinear mapping on the feature propagation result, to obtain a result of the current iteration feature propagation mapping.

[0026] In one of the embodiments, the feature propagation mapping module comprises:

[0027] The network account feature determination module is configured to determine a network account feature constructed according to account features of each social account in a social network; the network account feature comprises an account feature corresponding to the target account.

[0028] The model feature propagation mapping module is configured to input the account interaction feature and the network account feature into a feature propagation mapping model to perform feature propagation mapping, to obtain a network propagation account feature output by the feature propagation mapping model.

[0029] The model output processing module is configured to extract a propagation account feature corresponding to the target account from the network propagation account feature.

[0030] In one of the embodiments, the model feature propagation mapping module comprises:

[0031] The standardization processing module is configured to perform standardization processing on the account interaction feature based on a standardization condition, to obtain a standardized account interaction feature.

[0032] The feature input module is configured to input the standardized account interaction feature and the network account feature into the feature propagation mapping model to perform feature propagation mapping.

[0033] In one of the embodiments, the target information recommendation module comprises:

[0034] The feature matching module is configured to match the propagation account feature and each information interaction feature by using a matching model, and obtain a matching result output by the matching model.

[0035] The target information determination module is configured to determine target information from each to-be-recommended information based on the matching result.

[0036] The target information processing module is configured to recommend the target information to a terminal corresponding to the target account.

[0037] In one of the embodiments, the target information determination module includes:

[0038] The recommendation condition screening module is configured to determine target matching results that meet a recommendation condition from the matching result.

[0039] The screening result processing module is configured to determine the to-be-recommended information corresponding to the target matching result as the target information.

[0040] In one of the embodiments, the apparatus further includes:

[0041] The statistical result determination module is configured to determine a statistical result of interaction behaviors between each social account in a social network to which the target account belongs.

[0042] The interaction feature obtaining module is configured to obtain interaction features between each social account in the social network based on the statistical result.

[0043] The account interaction feature generation module is configured to generate an account interaction feature corresponding to the social network according to the interaction features between each social account in the social network.

[0044] In one of the embodiments, the apparatus further includes:

[0045] The social network determination module is configured to determine a social network to which the target account belongs.

[0046] The node embedding module is configured to perform node embedding on each node in the social network to obtain a node feature corresponding to each node respectively; each node corresponds to each social account in the social network, and a node relationship between each node corresponds to an interaction behavior between each social account.

[0047] The account feature determination module is configured to determine an account feature corresponding to the target account based on the node features corresponding to each node.

[0048] In one of the embodiments, the node embedding module includes:

[0049] The weight determination module is configured to determine a walk weight between each node in the social network according to a node relationship between each node.

[0050] A walking module is configured to perform node walking in the social network based on a walking weight from each node as a starting point to form a node walking track of each node;

[0051] A walking track processing module is configured to perform feature embedding on each node walking track by using an embedding model to obtain a node feature corresponding to each node.

[0052] In one of the embodiments, the device further comprises:

[0053] An interactive social account determining module is configured to determine an interactive social account of each to-be-recommended information; the interactive social account is a social account that has generated an interaction operation on the corresponding to-be-recommended information in the social network;

[0054] An information interaction feature obtaining module is configured to perform feature aggregation on the account features corresponding to each interactive social account to obtain an information interaction feature of the corresponding to-be-recommended information.

[0055] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0056] An account interaction feature corresponding to a social network to which the target account belongs is obtained; the account interaction feature is obtained according to interaction behaviors between each social account in the social network;

[0057] The account feature corresponding to the target account is mapped by using the account interaction feature, and a propagated account feature corresponding to the target account is obtained according to a result of the feature propagation mapping;

[0058] Each information interaction feature of each to-be-recommended information is obtained; the information interaction feature is generated based on the account features of the social accounts that have generated an interaction operation on the corresponding to-be-recommended information in the social network;

[0059] The propagated account feature and each information interaction feature are matched, and a target information determined from each to-be-recommended information according to a matching result is recommended to a terminal corresponding to the target account.

[0060] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the following steps:

[0061] An account interaction feature corresponding to a social network to which the target account belongs is obtained; the account interaction feature is obtained according to interaction behaviors between each social account in the social network;

[0062] The account feature corresponding to the target account is mapped by using the account interaction feature, and a propagated account feature corresponding to the target account is obtained according to a result of the feature propagation mapping;

[0063] obtaining respective information interaction features of each to-be-recommended information; the information interaction features are generated based on account features of social accounts that have generated interaction operations on the respective to-be-recommended information in the social network;

[0064] matching the propagation account features and the information interaction features, and recommending target information determined from the to-be-recommended information to a terminal corresponding to the target account according to a matching result.

[0065] The information recommendation method, apparatus, computer device, and storage medium described above perform feature propagation mapping on account features corresponding to a target account according to account interaction features obtained based on interaction behaviors between social accounts in a social network, obtain propagation account features corresponding to the target account according to a result of the feature propagation mapping, match the propagation account features with respective information interaction features of each to-be-recommended information, the information interaction features are generated based on account features of social accounts that have generated interaction operations on the respective to-be-recommended information in the social network, and recommend target information determined from the to-be-recommended information to a terminal corresponding to the target account according to a matching result. In the information recommendation process, the feature propagation mapping is performed on the account features corresponding to the target account according to the account interaction features obtained based on the interaction behaviors between the social accounts in the social network, so that the account features of the social accounts that have the interaction behavior relationship with the target account in the social network are propagated to the target account, the propagation account features corresponding to the target account are obtained, and the propagation account features and the respective information interaction features of the to-be-recommended information are matched, the target information is determined according to a matching result, and the information is recommended. Therefore, the information is recommended by using the interaction behavior relationship between the social accounts in the social network, and the accuracy of the information recommendation is improved. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 An application environment diagram of an information recommendation method in an embodiment;

[0067] Figure 2 A flowchart of an information recommendation method in an embodiment;

[0068] Figure 3 A flowchart of feature propagation mapping in an embodiment;

[0069] Figure 4 An interaction behavior diagram between social accounts in a social network in an embodiment;

[0070] Figure 5 A flowchart of feature propagation mapping in another embodiment;

[0071] Figure 6 A flowchart of recommending target information in an embodiment;

[0072] Figure 7 This is a schematic diagram illustrating the interaction behavior between social accounts in a social network in another embodiment;

[0073] Figure 8 for Figure 7 The diagram shown is a schematic representation of the node structure of the social network in the embodiment shown.

[0074] Figure 9 This is a flowchart illustrating the information recommendation method in another embodiment;

[0075] Figure 10 This is a structural block diagram of an information recommendation device in one embodiment;

[0076] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0077] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0078] The information recommendation method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. Users A, B, C, and D have social accounts on the same social network, and user C has interacted with users A, B, and D on the social network. User C logs into the short video platform on terminal 102 using the target account on the social network. When user C clicks the short video recommendation control, triggering a short video recommendation, server 104 responds to the short video recommendation request sent by terminal 102. By using the account interaction features obtained from the interaction behavior between various social accounts on the social network, server 104 performs feature propagation mapping on the account features corresponding to the target account. Based on the result of the feature propagation mapping, server 104 obtains the propagation account features corresponding to the target account and matches these propagation account features with the information interaction features of each short video to be recommended. The information interaction features are generated based on the account features of social accounts that have interacted with the corresponding short videos to be recommended on the social network. Based on the matching results, server 104 recommends the target short videos determined from the short videos to be recommended to the terminal 102 corresponding to the target account. Terminal 102 receives the target short videos recommended by server 104 and displays them. Furthermore, the information recommended by the information recommendation system can be limited to short videos, and can also include various data information such as audio and video, images, text, web pages, and business cards.

[0079] The terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, vehicle-mounted devices, and portable wearable devices. The server 104 can be implemented by a single server or a server cluster composed of multiple servers. The server 104 can also be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, and basic cloud computing services such as big data and artificial intelligence platforms.

[0080] In a specific application, the account interaction features and information interaction features involved in the information recommendation can be chained to the blockchain for secure storage. The blockchain is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm. The blockchain is essentially a decentralized database, which is a series of data blocks associated using cryptographic methods. Each data block contains information about a batch of network transactions, which is used to verify the validity of the information (anti-fake) and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer.

[0081] The blockchain underlying platform can include user management, basic services, smart contracts, and other processing modules. The user management module is responsible for the identity information management of all blockchain participants, including maintaining public and private key generation (account management), key management, and the correspondence between user real identities and blockchain addresses (permission management). In addition, under authorization, the module supervises and audits the transaction of certain real identities, provides rules configuration for risk control (risk audit), and verifies the validity of business requests. The basic service module is deployed on all blockchain node devices to verify the validity of business requests and record them to the storage after consensus. For a new business request, the basic service module first performs interface adaptation analysis and authentication processing (interface adaptation), then encrypts the business information through a consensus algorithm (consensus management), and finally transmits and records the complete and consistent information to the shared ledger (network communication). The smart contract module is responsible for contract registration and issuance, contract triggering, and contract execution. Developers can define contract logic through a certain programming language, publish it to the blockchain (contract registration), and trigger execution according to the logic of the contract terms, complete the contract logic, and provide contract upgrade and cancellation functions.

[0082] The platform product service layer provides basic capabilities and implementation frameworks for typical applications. Developers can add business features based on these basic capabilities to complete the blockchain implementation of business logic. The application service layer provides application services based on the blockchain solution for business participants to use.

[0083] In one embodiment, asFigure 2 As shown, an information recommendation method is provided, and the method is applied to Figure 1 a server in the information recommendation method, including the following steps:

[0084] In step 202, an account interaction feature corresponding to a social network to which a target account belongs is obtained; the account interaction feature is obtained according to interaction behaviors between each social account in the social network.

[0085] The target account is a user account that needs to be recommended information, and the target account can be specifically a social account in the social network. For example, a user can log in to an information client through a social account, so that the server can accurately recommend information to the user through the social relationship of the user in the corresponding social network. The social relationship refers to the social communication between people through the social network, and is a social activity for achieving a certain purpose by using a certain way to transmit information and exchange ideas. Specifically, people can register different social accounts to interact through the social network, such as chatting, liking, commenting, and various interactive behaviors.

[0086] The account interaction feature is obtained according to the interaction behaviors between each social account in the social network, and the interaction behaviors include various interactive operations such as chatting, commenting, liking, and forwarding performed by the user through each social account. The account interaction feature can be obtained according to an interaction analysis result obtained by performing interaction analysis on the interaction behaviors between each social account in the social network. For example, the interaction behaviors between each two social accounts in the social network can be counted respectively to obtain an interaction behavior statistical result between each two social accounts, an interaction feature between each two social accounts is constructed based on the interaction behavior statistical result, and the account interaction feature corresponding to the social network is obtained according to the interaction features between each two social accounts in the social network. The account interaction feature reflects the interaction relationship between each social account in the social network.

[0087] In a specific application, for example, for a social network including social accounts A, B, C, and D, the interaction behaviors between each two social accounts can be analyzed respectively to obtain a statistical result of the interaction behaviors between each two social accounts. Based on the statistical result, feature extraction can be performed to obtain the interaction features between each two social accounts. For example, the interaction features between each two social accounts can be determined comprehensively according to the statistical results such as the number of chatting times, the frequency, and the cumulative interaction duration between each two social accounts. After obtaining the interaction features between each two social accounts in the social network, the interaction features are fused to obtain the account interaction feature corresponding to the social network. The account interaction feature can carry the interaction features between each two accounts in the social network.

[0088] Specifically, when triggering the information recommendation processing for the target account, such as when the server receives the information recommendation request uploaded by the terminal, the server obtains the account interaction features corresponding to the social network to which the target account belongs in response to the information recommendation request. The account interaction features can be determined by the server in advance according to the interaction behaviors between each social account in the social network.

[0089] In step 204, the account features corresponding to the target account are mapped by the account interaction features, and the propagation account features corresponding to the target account are obtained according to the mapping results.

[0090] The account features corresponding to the target account can be information representing the essential characteristics of the target account, which can be obtained by feature engineering processing of the target account. Feature engineering is a process of converting raw data into better features representing the essence of the problem, so that the features can be applied to the prediction model to improve the model prediction accuracy for invisible data. Feature propagation mapping is a process of describing the target account features by using the features of the social accounts having social relationships with the target account in the social network. The feature propagation mapping of the account features corresponding to the target account can transfer the features of the social accounts having social relationships with the target account to the target account, so as to further describe the account features of the target account by the features of the social accounts having social relationships with the target account, thereby obtaining the account features accurately reflecting the social relationships of the target account. The propagation account features corresponding to the target account are the features obtained after the feature propagation mapping of the account features corresponding to the target account. In addition to carrying the features of the target account itself, the propagation account features also carry the account features of the social accounts having social relationships with the target account in the social network.

[0091] Specifically, after the server obtains the account interaction features corresponding to the social network to which the target account belongs, the server further obtains the account features of the target account, and performs feature propagation mapping on the account features of the target account by using the account interaction features, to obtain the mapping results. The server obtains the propagation account features corresponding to the target account from the mapping results. The propagation account features combine the account features of the target account itself and the account features of the social accounts having social relationships with the target account in the social network.

[0092] In step 206, the information interaction features of each to-be-recommended information are obtained. The information interaction features are generated based on the account features of the social accounts having generated interaction operations on the corresponding to-be-recommended information in the social network.

[0093] The to-be-recommended information is information that can be recommended, and can be various types of information, such as audio and video, pictures, text, web pages, business cards, and various data information. The to-be-recommended information can be determined according to the scene in which the information recommendation is actually applied. For example, for a short video platform application, the to-be-recommended information can be each short video in the short video platform application, so as to accurately recommend short videos to the target account. The information interaction feature is a feature of each to-be-recommended information, and is generated based on the account features of the social accounts that have generated interaction operations on the corresponding to-be-recommended information in the social network. In specific implementation, the interaction operation can be flexibly set according to actual needs, such as browsing, liking, commenting, forwarding, collecting, and following various operations, and combinations of various operations. For example, for a short video R1, social accounts A, B, and C in the social network have all browsed or liked the short video R1, it can be considered that the social accounts A, B, and C have all generated interaction operations on the short video R1, and then the information interaction feature representing the characteristics of the short video R1 can be obtained according to the account features of the social accounts A, B, and C respectively.

[0094] Specifically, the server obtains the information interaction feature of each to-be-recommended information, which can be generated by the server in advance based on the account features of the social accounts that have generated interaction operations on the corresponding to-be-recommended information in the social network, so as to obtain the information interaction feature corresponding to each to-be-recommended information when triggering information recommendation.

[0095] In step 208, the propagation account feature and each information interaction feature are matched, and the target information determined from each to-be-recommended information according to the matching result is recommended to the terminal corresponding to the target account.

[0096] In which, the target information is the information that needs to be recommended to the target account according to the matching result from each to-be-recommended information, that is, the target information is the information selected from each to-be-recommended information for recommendation to the target account. Specifically, after obtaining the propagation account feature of the target account and the information interaction feature corresponding to each to-be-recommended information, the server matches the propagation account feature and each information interaction feature respectively, obtains the matching result of the propagation account feature and each information interaction feature, and determines the target information from each to-be-recommended information according to the matching result. The server recommends the target information to the terminal corresponding to the target account.

[0097] In specific implementation, the similarity of the propagation account feature and each information interaction feature can be determined respectively, and the to-be-recommended information corresponding to the information interaction feature with a similarity greater than a similarity threshold is determined as the target information, and the target information is recommended to the terminal corresponding to the target account.

[0098] In the information recommendation method, the account interaction features obtained according to the interaction behaviors between the social accounts in the social network are used to perform feature propagation mapping on the account features corresponding to the target account, the propagation account features corresponding to the target account are obtained according to the result of the feature propagation mapping, the propagation account features are matched with the information interaction features of each to-be-recommended information, the information interaction features are generated based on the account features of the social accounts that have generated interaction operations on the corresponding to-be-recommended information in the social network, and the target information determined from the to-be-recommended information is recommended to the terminal corresponding to the target account according to the matching result. In the information recommendation process, the account interaction features obtained according to the interaction behaviors between the social accounts in the social network are used to perform feature propagation mapping on the account features corresponding to the target account, so that the account features of the social accounts that have the interaction behavior relationship with the target account in the social network are propagated to the target account, the propagation account features corresponding to the target account are obtained, and the target information is determined for recommendation according to the matching result of the propagation account features and the information interaction features of the to-be-recommended information, so that the information recommendation is performed by using the interaction behavior relationship between the social accounts in the social network, and the accuracy of the information recommendation is improved.

[0099] In one embodiment, as shown in Figure 3 The processing of the feature propagation mapping, that is, the feature propagation mapping on the account features corresponding to the target account by using the account interaction features, and the obtaining of the propagation account features corresponding to the target account according to the result of the feature propagation mapping, include:

[0100] In step 302, the target account interaction features corresponding to the target account are extracted from the account interaction features.

[0101] The account interaction features are obtained according to the interaction behaviors between the social accounts in the social network, that is, the account interaction features include the interaction behaviors between all the social accounts in the social network. The target account interaction features are the interaction features between the target account and the social accounts that have the interaction behavior with the target account in the social network, that is, the target account interaction features are the features corresponding to the interaction behaviors that exist between the target account and the social accounts in the social network. For example, as shown in Figure 4As shown, in the social network, social accounts A, B, C, D, and E are included, each social account can be taken as a node in the social network, and the nodes are connected through edges, the edges connecting the nodes represent the interaction behaviors between the social accounts, and the width of the edge reflects the closeness of the interaction between the social accounts, the wider the edge, the more intimate the interaction between the social accounts, that is, the more or more frequent the interaction behaviors between the corresponding two social accounts. When the target account is the social account A, the target account interaction features corresponding to the social account A extracted from the account interaction features include features A_AB, A_AD, and A_AE, wherein the feature A_AB is the interaction feature corresponding to the interaction behavior between the social account A and the social account B, the feature A_AD is the interaction feature corresponding to the interaction behavior between the social account A and the social account D, and the feature A_AE is the interaction feature corresponding to the interaction behavior between the social account A and the social account E. There is no corresponding interaction feature between the social account A and the social account C, and no interaction behavior occurs.

[0102] Specifically, when information is recommended for the target account, the server extracts the target account interaction features corresponding to the target account from the account interaction features corresponding to the social network, and the target account interaction features include the interaction features obtained according to the interaction behaviors between the target account and other social accounts in the social network.

[0103] In step 304, the account features of the interaction social account corresponding to the target account interaction features are obtained.

[0104] The interaction social account is a social account in the social network that has an interaction behavior with the target account, that is, the interaction social account corresponds to the target account interaction feature, and the target account interaction feature represents the characteristics of the interaction behavior between the target account and the interaction social account. For example, as shown in the figure, when the target account is the social account A, the target account interaction features corresponding to the social account A extracted from the account interaction features include features A_AB, A_AD, and A_AE, and the interaction social account corresponding to the target account interaction features includes social accounts B, D, and E. Figure 4 As shown, when the target account is the social account A, the target account interaction features corresponding to the social account A extracted from the account interaction features include features A_AB, A_AD, and A_AE, and the interaction social account corresponding to the target account interaction features includes social accounts B, D, and E. The account features of the interaction social account can be information representing the characteristics of the interaction social account, and can be obtained by performing feature engineering processing on the interaction social account.

[0105] In a specific implementation, the server can perform feature engineering processing on each social account in the social network in advance to obtain the account features of each social account in the social network, and store them. After determining the target account, the server can directly determine the account features corresponding to the target account from the stored account features, and further, the server can filter out the account features of the interaction social account having an interaction behavior with the target account from the stored account features.

[0106] Specifically, the server determines the target account interaction feature, and acquires the account feature corresponding to the interaction social account having the interaction behavior with the target account according to the target account interaction feature.

[0107] In step 306, the account feature corresponding to the target account is iteratively feature propagation mapped based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter, to obtain a result of the feature propagation mapping.

[0108] The propagation mapping parameter is used to adjust the feature propagation mapping, and can be determined through network model training when the feature propagation mapping is implemented through the network model. Iteration is an activity of repeated feedback process, and the result obtained in each iteration is used as the initial value of the next iteration, i.e., the result of the current iteration feature propagation mapping is used as the initial value for the next feature propagation mapping after each feature propagation mapping processing, until the iteration ends, and the result of the feature propagation mapping is obtained. In a specific implementation, multiple propagation mapping parameters can be set, and the account feature corresponding to the target account is iteratively feature propagation mapped through the target account interaction feature, the account feature of the interaction social account, and the multiple propagation mapping parameters, such as iteratively feature propagation mapping through the multi-layer structure of the network model, to obtain the result of the feature propagation mapping.

[0109] Specifically, the server acquires the propagation mapping parameter when the feature propagation mapping is processed, and iteratively feature propagation maps the account feature corresponding to the target account based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter, to obtain the result of the feature propagation mapping. In one specific application, the feature propagation mapping is implemented through a pre-trained feature propagation mapping model, and the feature propagation mapping model includes at least one layer structure, each layer structure has a corresponding propagation mapping parameter, and can perform one feature propagation mapping processing. The feature propagation mapping model can iteratively feature propagation map the account feature corresponding to the target account, to obtain the result of the feature propagation mapping.

[0110] In step 308, the propagation account feature corresponding to the target account is determined according to the result of the feature propagation mapping.

[0111] After obtaining the result of the feature propagation mapping, the server can determine the propagation account feature corresponding to the target account according to the result of the feature propagation mapping. The propagation account feature is a feature obtained by processing the account feature corresponding to the target account through the feature propagation mapping. In addition to carrying the feature of the target account itself, the propagation account feature also carries the account feature of the social account having a social relationship with the target account in the social network. In a specific implementation, the server can directly extract the propagation account feature corresponding to the target account from the result of the feature propagation mapping, or can further process the propagation account feature included in the result of the feature propagation mapping, such as normalization or standardization processing, to obtain the propagation account feature corresponding to the target account.

[0112] In this embodiment, the server performs iterative feature propagation mapping on the account feature corresponding to the target account based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter, and determines the propagation account feature corresponding to the target account according to the result of the feature propagation mapping. Therefore, through the target account interaction feature corresponding to the interaction behavior and the propagation mapping parameter, the account feature of the interaction social account having the interaction behavior with the target account in the social network is transmitted to the target account, the propagation account feature capable of accurately representing the social relationship in the social network is obtained, and the information recommendation processing is performed based on the propagation account feature, thereby improving the accuracy of information recommendation.

[0113] In one embodiment, the iterative feature propagation mapping is performed on the account feature corresponding to the target account based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter to obtain the result of the feature propagation mapping, including: determining the account feature corresponding to the target account as a current account feature; performing feature propagation mapping on the current account feature based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of this iteration of feature propagation mapping to obtain the result of this iteration of feature propagation mapping; taking the result of this iteration of feature propagation mapping as the current account feature, and returning to the step of performing feature propagation mapping on the current account feature based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of this iteration of feature propagation mapping to obtain the result of this iteration of feature propagation mapping.

[0114] Wherein, in each iteration of the feature propagation mapping, the result of the last iteration of the feature propagation mapping is taken as the initial value of the current iteration of the feature propagation mapping, until the iteration is completed, and the result of the feature propagation mapping is obtained. The current account feature refers to the account feature corresponding to the target account in the current feature propagation mapping. When the feature propagation mapping is not performed, the current account feature takes the value of the account feature corresponding to the target account obtained by the server. Each iteration of the feature propagation mapping can be provided with corresponding propagation mapping parameters. The propagation mapping parameters corresponding to each iteration of the feature propagation mapping can be the same or different. The propagation mapping parameters corresponding to each iteration of the feature propagation mapping can be obtained through model training.

[0115] Specifically, in the iteration of the feature propagation mapping of the target account, the server determines the account feature corresponding to the target account as the current account feature, that is, takes the obtained account feature corresponding to the target account as the initial value of the current iteration of the feature propagation mapping. The server performs feature propagation mapping on the current account feature based on the target account interaction feature, the account feature of the interactive social account and the propagation mapping parameters of the current iteration of the feature propagation mapping. Specifically, the target account interaction feature, the account feature of the interactive social account and the current account feature can be fused and mapped through the propagation mapping parameters of the current iteration of the feature propagation mapping to obtain the result of the current iteration of the feature propagation mapping. The server takes the result of the current iteration of the feature propagation mapping as the initial value of the next iteration of the feature propagation mapping, that is, takes the result of the current iteration of the feature propagation mapping as the current account feature, and returns to the step of performing feature propagation mapping on the current account feature based on the target account interaction feature, the account feature of the interactive social account and the propagation mapping parameters of the current iteration of the feature propagation mapping to obtain the result of the current iteration of the feature propagation mapping, thereby realizing the iteration of the feature propagation mapping of the account feature of the target account.

[0116] In the embodiment, the server performs feature propagation mapping on the target account interaction feature, the account feature of the interactive social account and the current account feature through the propagation mapping parameters of each iteration of the feature propagation mapping, and takes the result of each iteration of the feature propagation mapping as the initial value of the next iteration of the feature propagation mapping. After the iteration is completed, the result of the feature propagation mapping is obtained, so that the account feature of the interactive social account having the interaction behavior with the target account in the social network is transmitted to the target account from multiple dimensions through the iteration of the feature propagation mapping, and the propagation account feature capable of accurately representing the social relationship in the social network is obtained. Based on the propagation account feature, the information recommendation processing can be performed, and the accuracy of the information recommendation can be improved.

[0117] In an embodiment, the current account feature is subjected to feature propagation mapping based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping, to obtain a result of the current iteration feature propagation mapping, including: the current account feature is subjected to feature propagation based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping, to obtain a feature propagation result; the feature propagation result is subjected to nonlinear mapping to obtain a result of the current iteration feature propagation mapping.

[0118] The feature propagation result is the result obtained by subjecting the current account feature to feature propagation based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping. The nonlinear mapping of the feature propagation result can increase the nonlinearity in the feature propagation mapping processing, thereby improving the accuracy of feature expression and obtaining a high-accuracy propagated account feature.

[0119] Specifically, after the server obtains the feature propagation result by subjecting the current account feature to feature propagation based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the current iteration feature propagation mapping, the server subjects the feature propagation result to nonlinear mapping to obtain a result of the current iteration feature propagation mapping. Specifically, the server can obtain a pre-set nonlinear function, such as a Sigmoid function, a ReLU (Rectified Linear Unit) function, etc., to perform nonlinear mapping processing on the feature propagation result, to obtain a result of the current iteration feature propagation mapping. In a specific implementation, the feature propagation mapping can be realized by a feature propagation mapping model, and in the structure of the feature propagation mapping model, each layer of the feature propagation mapping is provided with corresponding propagation mapping parameters to perform feature propagation mapping. After each feature propagation mapping, the feature propagation result of each feature propagation mapping is subjected to nonlinear mapping by a set activation layer, to obtain a result of each iteration feature propagation mapping. The activation layer can be a ReLU layer, thereby obtaining a result of the current iteration feature propagation mapping. After obtaining the result of the current iteration feature propagation mapping, the next feature propagation mapping is performed until the iteration ends, to obtain a result of the feature propagation mapping, and the propagated account feature corresponding to the target account is obtained based on the result of the feature propagation mapping.

[0120] In this embodiment, after the account characteristics corresponding to the target account are subjected to the current iteration characteristic propagation mapping to obtain the characteristic propagation result, the characteristic propagation result is further subjected to nonlinear mapping to obtain the result of the current iteration characteristic propagation mapping, thereby the result of each characteristic propagation mapping is subjected to nonlinear processing, the nonlinearity of the result of the characteristic propagation mapping is improved, the feature expression of the obtained propagated account characteristics can be enhanced, and the accuracy of information recommendation based on the propagated account characteristics can be improved.

[0121] In one embodiment, as shown in Figure 5 The processing of the characteristic propagation mapping, i.e., the characteristic propagation mapping of the account characteristics corresponding to the target account by the account interaction characteristics, obtaining the propagated account characteristics corresponding to the target account according to the result of the characteristic propagation mapping, includes:

[0122] Step 502, determining a network account characteristic constructed according to the account characteristics of each social account in the social network; the network account characteristic includes the account characteristics corresponding to the target account.

[0123] The network account characteristic is constructed according to the account characteristics of each social account in the social network, and the network account characteristic includes the account characteristics of all social accounts in the social network, specifically including the account characteristics corresponding to the target account and the account characteristics of the interactive social accounts having the interaction behavior with the target account. The account characteristics of each social account in the social network can be realized based on the feature engineering to accurately describe the characteristics of each social account in the social network.

[0124] Specifically, when the account characteristics corresponding to the target account are subjected to the characteristic propagation mapping to determine the propagated account characteristics corresponding to the target account, the server can directly perform the characteristic propagation mapping processing on each social account in the social network to which the target account belongs to obtain the network account characteristic including the corresponding account characteristics of each social account in the social network. In specific application, the server can pre-extract the account characteristics of each social account in the social network to which the target account belongs based on the feature engineering, and construct the network account characteristic corresponding to the social network. For example, each social account in the social network can be regarded as a node of the social network, and the interaction behavior between the social accounts can be regarded as an edge connecting the nodes, so as to construct a user node graph corresponding to the social network according to the interaction behavior between the social accounts, and then perform node embedding processing based on the user node graph by a node embedding algorithm such as Node2Vec algorithm to obtain the network account characteristic corresponding to the social network, the network account characteristic including the account characteristics of each social account in the social network.

[0125] Step 504, inputting the account interaction characteristics and the network account characteristics into the characteristic propagation mapping model to perform the characteristic propagation mapping, and obtaining the network propagated account characteristics output by the characteristic propagation mapping model.

[0126] The account interaction feature is obtained according to interaction behaviors between social accounts in the social network, that is, the account interaction feature includes interaction behaviors between all social accounts in the social network. The feature propagation mapping model can be a machine learning model, which can be a network model trained based on a neural network algorithm or a deep learning algorithm. The feature propagation mapping model can be pre-trained, and the feature propagation mapping model can perform feature propagation mapping according to the input account interaction feature and the network account feature, and output the network propagation account feature. The network propagation account feature includes respective propagation account features of each social account in the social network. The network propagation account feature is obtained by performing feature propagation mapping on the account interaction feature and the network account feature of the social network by the feature propagation mapping model, that is, performing feature propagation mapping on each social account in the social network to obtain the network propagation account feature including respective propagation account features of each social account.

[0127] Specifically, the server can query the feature propagation mapping model pre-trained by the neural network algorithm or the deep learning algorithm. The server inputs the account interaction feature and the network account feature into the feature propagation mapping model to perform feature propagation mapping by the feature propagation mapping model, and obtains the network propagation account feature output by the feature propagation mapping model.

[0128] Step 506, extracting the propagation account feature corresponding to the target account from the network propagation account feature.

[0129] After obtaining the network propagation account feature, the network propagation account feature includes the propagation account features of all social accounts in the social network, and then the server can extract the propagation account feature corresponding to the target account from the network propagation account feature. Specifically, the server can extract the propagation account feature corresponding to the target account from the network propagation account feature output by the feature propagation mapping model according to the mapping relationship between the target account and the network account feature.

[0130] In this embodiment, the server performs feature propagation mapping based on the account interaction feature and the network account feature of the social network by the feature propagation mapping model pre-trained, so as to perform feature propagation mapping on each social account in the social network by the feature propagation mapping model, obtain the network propagation account feature corresponding to the social network, and the network propagation account feature includes the propagation account features of each social account in the social network. The server extracts the propagation account feature corresponding to the target account from the network propagation account feature. By performing feature propagation mapping by the feature propagation mapping model, the feature propagation mapping on each social account in the social network can be accurately and quickly performed, which is beneficial to improving the processing efficiency and recommendation accuracy of information recommendation for the social accounts in the social network.

[0131] In an embodiment, the account interaction feature and the network account feature are input into a feature propagation mapping model for feature propagation mapping, including: performing standardization processing on the account interaction feature according to a standardization condition to obtain a standardized account interaction feature; and inputting the standardized account interaction feature and the network account feature into the feature propagation mapping model for feature propagation mapping.

[0132] The standardization condition can be set according to actual needs, and specifically can include a feature standardization formula. By performing standardization processing on the account interaction feature, it can be ensured that the dimension number of the output network propagation account feature can be controlled when the feature propagation mapping model is used for feature propagation mapping, the complexity of the network propagation account feature can be reduced, and the processing efficiency of information recommendation can be improved.

[0133] Specifically, after obtaining the account interaction feature and the network account feature, the server obtains a pre-set standardization condition. When the account interaction feature includes an account interaction feature matrix, the standardization processing on the account interaction feature matrix can be implemented by multiplying the diagonal matrix of the account interaction feature matrix by the account interaction feature matrix. The standardization condition can be flexibly set according to actual needs. The server performs standardization processing on the account interaction feature according to the standardization condition to obtain a standardized account interaction feature. The server inputs the standardized account interaction feature and the network account feature into the feature propagation mapping model, so that the feature propagation mapping model performs feature propagation mapping according to the standardized account interaction feature and the network account feature, and outputs a network propagation account feature.

[0134] In this embodiment, the account interaction feature is standardized by the standardization condition, and the feature propagation mapping model performs feature propagation mapping based on the standardized account interaction feature and the network account feature. When the feature propagation mapping model performs multiple feature propagation mapping processing, the dimension of the data can be effectively controlled, the complexity of the propagation account feature output by the feature propagation mapping model can be reduced, and the processing efficiency of information recommendation based on the propagation account feature can be improved.

[0135] In an embodiment, as shown in Figure 6 The processing of the recommended target information, i.e., matching the propagation account feature and each information interaction feature, and recommending the target information determined from each to-be-recommended information to the terminal corresponding to the target account according to the matching result, includes:

[0136] Step 602: matching the propagation account feature and each information interaction feature by using a matching model to obtain a matching result output by the matching model.

[0137] The matching model can be a machine learning model, such as a network model trained based on a neural network algorithm or a deep learning algorithm, such as a multi-layer perception model. The matching model can match the input propagation account features and information interaction features respectively, output the matching result of the propagation account features and the information interaction features, and the matching result can reflect the similarity between the propagation account features and the information interaction features. The higher the similarity between the propagation account features and the information interaction features, the more matched the propagation account features and the information interaction features, and the information interaction features corresponding to the to-be-recommended information can be recommended to the target account.

[0138] Specifically, when matching the propagation account features and the information interaction features, the server queries the pre-trained matching model, and the server inputs the propagation account features and the information interaction features into the matching model in sequence for matching, and obtains the matching result output by the matching model.

[0139] Step 604, determining the target information from the to-be-recommended information based on the matching result.

[0140] The target information is information determined from the to-be-recommended information based on the matching result, that is, the target information is information selected from the to-be-recommended information for recommendation to the target account. Specifically, after obtaining the matching result output by the matching model, the server determines the target information from the to-be-recommended information based on the matching result. In specific implementation, the server can compare the matching result with a preset recommendation condition, thereby determining the matching result that meets the recommendation condition, and determining the to-be-recommended information corresponding to the matching result that meets the recommendation condition as the target information. For example, when the matching result includes a matching similarity, the recommendation condition can be a similarity threshold, and the matching result with a matching similarity greater than the similarity threshold can be determined from the matching result, and the to-be-recommended information corresponding to the matching result with a matching similarity greater than the similarity threshold can be determined as the target information.

[0141] Step 606, recommending the target information to the terminal corresponding to the target account.

[0142] After determining the target information from the to-be-recommended information, the server recommends the target information to the terminal corresponding to the target account. Specifically, the server can generate an information recommendation message according to the target information, and send the information recommendation message to the terminal corresponding to the target account. In specific application, the server can recommend the target information to the terminal corresponding to the target account when receiving an information recommendation request uploaded by the terminal corresponding to the target account, or the server can actively recommend the target information to the terminal corresponding to the target account when a information recommendation trigger condition is met, such as when a information recommendation period is reached.

[0143] In this embodiment, the server matches the propagation account features and the information interaction features through the pre-trained matching model, and determines the target information from the to-be-recommended information according to the matching result, and recommends the target information to the terminal corresponding to the target account, so that the accurate matching of the propagation account features and the information interaction features can be realized, and the accuracy of information recommendation is improved.

[0144] In one embodiment, determining the target information from the to-be-recommended information based on the matching result includes: determining a target matching result that meets a recommendation condition from the matching result; and determining the to-be-recommended information corresponding to the target matching result as the target information.

[0145] The recommendation condition can be flexibly set according to actual needs, such as being set to a matching degree greater than a matching degree threshold or meeting a matching degree sorting requirement. Specifically, after the server obtains the matching result output by the matching model, the server further obtains a preset recommendation condition, filters the matching result according to the recommendation condition, determines a target matching result that meets the recommendation condition from the matching result, and determines the to-be-recommended information corresponding to the target matching result. The server determines the to-be-recommended information corresponding to the target matching result as the target information, so as to recommend the target information to the terminal corresponding to the target account.

[0146] In one specific application, the recommendation condition is that the matching degree is greater than the matching degree threshold, and then the server can compare the matching result with the preset matching degree threshold, determine the matching result with the matching degree greater than the matching degree threshold as the target matching result, and determine the to-be-recommended information corresponding to the target matching result as the target information. In another specific application, the recommendation condition is the top 20 matching results with the highest matching degree, and then the server can sort the matching results in descending order of the matching degree, take the top 20 matching results as the target matching result, and determine the to-be-recommended information corresponding to the target matching result as the target information.

[0147] In this embodiment, the matching result is filtered through the preset recommendation condition, the matching result that meets the recommendation condition is determined as the target matching result, and the target information is determined from the to-be-recommended information according to the target matching result, so that the matching result can be filtered based on the recommendation condition, the target information matched with the target account is selected for recommendation, and the accuracy of information recommendation can be improved.

[0148] In one embodiment, the information recommendation method further includes: determining a statistical result of interaction behaviors between each social account in a social network to which the target account belongs; obtaining interaction features between each social account in the social network based on the statistical result; and generating account interaction features corresponding to the social network according to the interaction features between each social account in the social network.

[0149] The target account is a user account in a social network that needs information recommendation, and the account interaction feature is obtained according to the interaction behaviors between the social accounts in the social network, and the interaction behaviors include various interaction operations such as chatting, commenting, liking, and forwarding performed by the user through the social accounts. The statistical result is statistical data obtained by counting the interaction behaviors between the social accounts, which can include the cumulative number of interaction behaviors between the social accounts, the interaction frequency, the interaction cumulative duration, etc. The statistical result can be set according to actual needs. The interaction feature is used to represent the interaction behaviors between the two social accounts in the social network, and can be generated based on the statistical result of the interaction behaviors between the two social accounts in the social network, such as feature extraction on the statistical result of the interaction behaviors between the two social accounts, to obtain the interaction feature between the two social accounts.

[0150] Specifically, the server obtains the statistical result of the interaction behaviors between the social accounts in the social network to which the target account belongs. The statistical result can be obtained by counting and processing the interaction behaviors between the social accounts, such as counting the cumulative number of interaction behaviors between the social accounts. The server determines the interaction feature between the social accounts based on the statistical result of the interaction behaviors between the social accounts, and can perform feature processing on the statistical result of the interaction behaviors between the social accounts to obtain the interaction feature between the social accounts. After obtaining the interaction feature between the social accounts in the social network, the server generates the account interaction feature corresponding to the social network according to the interaction feature between the social accounts. The account interaction feature includes the interaction feature between the social accounts. In specific implementation, the server can combine the interaction features between the social accounts to generate the account interaction feature corresponding to the social network.

[0151] In this embodiment, the interaction feature between the social accounts is determined based on the statistical result of the interaction behaviors between the social accounts in the social network, and the account interaction feature corresponding to the social network is generated based on the interaction feature between the social accounts. The account interaction feature corresponding to the social network is constructed based on the interaction behaviors between the social accounts in the social network, so that the account feature of each social account can be mapped by feature propagation through the account interaction feature. The propagation account feature obtained based on the feature propagation can realize accurate information recommendation for the target account.

[0152] In one embodiment, the information recommendation method further includes: determining a social network to which the target account belongs; performing node embedding on each node in the social network to obtain a node feature corresponding to each node respectively; each node corresponds to a social account in the social network, and the node relationship between the nodes corresponds to the interaction behaviors between the social accounts; determining the account feature corresponding to the target account based on the node features corresponding to the nodes.

[0153] The target account is a user account that needs to be recommended information, and the social network is a network corresponding to a social system to which the target account belongs. The social network is composed of social accounts corresponding to each user. The social accounts in the social network can be used as nodes of the social network, and each social account corresponds to a user. The interaction behavior between nodes can be described by connecting the edges of the nodes, so as to construct a user node graph corresponding to the social network according to the interaction behavior between the social accounts. Based on the user node graph, the node embedding algorithm such as Node2Vec algorithm can be used for node embedding processing to obtain the node features corresponding to each node in the user node graph. According to the node features corresponding to the nodes, the account features corresponding to the target account can be obtained, for example, the node features of the target account corresponding node can be used as the account features corresponding to the target account.

[0154] Specifically, the server determines the social network to which the target account belongs, maps the social accounts in the social network to corresponding nodes, and maps the interaction behavior between the social accounts to the edges between the nodes. When there is an interaction behavior between the social accounts, the nodes are connected by the edges, and the width of the edge corresponds to the statistical result of the interaction behavior. For example, as shown in FIG. 1, in the social network, there are social accounts A, B, C, D and E, Figure 7 Figure 7 The statistical data of the interaction behavior between the social accounts in a certain time period is recorded in the table, including the number of chats, the cumulative interaction time, the number of like and comment operations, etc. For example, as shown in FIG. 1, the table records the statistical data of the interaction behavior between the social accounts A, B, C, D and E in a certain time period, including the number of chats, the cumulative interaction time, the number of like and comment operations, etc. Figure 8 Figure 7 The social accounts A, B, C, D and E in the table are mapped to nodes A, B, C, D and E, and the nodes are connected by edges. The width of the edge reflects the statistical data of the interaction behavior between the social accounts. The statistical data can be obtained by weighting the weight values of various types of interaction behaviors as needed. The nodes A, B, C, D and E correspond to the social accounts A, B, C, D and E one by one. The node relationship between the nodes, such as the connection relationship of the nodes and the width of the connected edges, corresponds to the interaction behavior between the social accounts.

[0155] ​​The server performs node embedding on each node in the social network, such as performing node embedding on each node by a Node2Vec node embedding algorithm, to obtain a node feature corresponding to each node. The embedding can map an entity into a continuous vector space, so that the entity can be represented by a vector, and the node embedding can find the neighbor nodes of a certain node in the node graph through the embedding feature of the node, and can describe the corresponding node by the embedding feature. The server determines an account feature corresponding to the target account based on the node features corresponding to the nodes. In specific implementation, the server can determine a target node from the nodes in the social network according to the mapping relationship between the target account and the nodes, and obtain a target node feature corresponding to the target node according to the result of the node embedding, and the server determines the target node feature as the account feature corresponding to the target account.

[0156] In this embodiment, by performing node embedding on the nodes corresponding to the social accounts in the social network, an account feature that can accurately express the social accounts can be constructed, and information recommendation based on the account feature can improve the accuracy of information recommendation.

[0157] In one embodiment, each node in the social network is subjected to node embedding to obtain a node feature corresponding to each node, including: determining a walk weight between each node according to a node relationship between the nodes in the social network; performing node walk in the social network based on the walk weight with each node as a starting point to form a node walk track of each node; and performing feature embedding on the node walk track of each node by an embedding model to obtain a node feature corresponding to each node.

[0158] The walk weight corresponds to the node relationship between the nodes, and the node relationship reflects the interaction behavior between the social accounts. The more frequent and richer the interaction behavior between the social accounts is, the closer the node relationship is, and the greater the value of the walk weight is. The node walk track is a track obtained by performing walk according to the edges connecting the nodes with the nodes in the social network as a starting point. The embedding model can perform feature mapping on the node walk track. Specifically, the embedding model can be a network model pre-constructed based on a word2vec word vector embedding algorithm. The embedding model can perform feature embedding processing on the input node walk track to map the node walk track into a node feature of the corresponding node. The node feature reflects the interaction relationship of the corresponding node in the social network.

[0159] Specifically, the server determines node relationships between nodes in the social network, the node relationships corresponding to interaction behaviors between social accounts, and determines a walk weight between corresponding nodes according to the node relationships. For example, a mapping relationship between the node relationships and the walk weight can be established in advance, so that the walk weight corresponding to the node relationship of each node can be obtained according to the mapping relationship. The server performs node walking between nodes in the social network based on the walk weight, and specifically performs random walking between nodes according to edges connected between the nodes, to form a node walking track of each node. The server performs feature embedding processing on the obtained node walking track of each node by using a pre-trained embedding model. Specifically, the server can input the node walking track of each node into the embedding model, and the embedding model performs feature embedding processing and outputs a node feature corresponding to each node.

[0160] In this embodiment, the walk weight determined by the node relationship enables each node to perform weighted walking in the social network, and the feature embedding of the obtained node walking track by using the embedding model can obtain a node feature based on a node embedding algorithm. The node feature reflects interaction behaviors between social accounts in the social network, and can accurately express social relationships of the social accounts in the social network. Information recommendation based on the account feature of the social account can improve the accuracy of information recommendation.

[0161] In one embodiment, the information recommendation method further includes: determining interaction social accounts of each to-be-recommended information; the interaction social account is a social account in the social network that has generated an interaction operation on the corresponding to-be-recommended information; and performing feature aggregation on the account feature corresponding to each interaction social account to obtain an information interaction feature of the corresponding to-be-recommended information.

[0162] The to-be-recommended information is information that can be recommended, and can be various types of information such as audio and video, pictures, text, web pages, and business cards. The information interaction feature is a feature of each to-be-recommended information, and is generated based on the account feature of a social account in the social network that has generated an interaction operation on the corresponding to-be-recommended information.

[0163] Specifically, the server determines interaction social accounts of each to-be-recommended information, and the interaction social account is a social account in the social network that has generated an interaction operation on the corresponding to-be-recommended information. For example, in the social network, social accounts A, B, and D have all generated an interaction operation on to-be-recommended information R, such as browsing the to-be-recommended information R or performing a like operation on the to-be-recommended information R. Therefore, for the to-be-recommended information R, the interaction social accounts include the social accounts A, B, and D, and for the social account C, the to-be-recommended information R does not include the social account C because the social account C has not generated an interaction operation on the to-be-recommended information R.

[0164] The server determines the account features corresponding to each interactive social account, and aggregates the account features corresponding to each interactive social account to obtain information interaction features of the corresponding to-be-recommended information. In a specific implementation, the server can fuse the account features corresponding to each interactive social account, for example, performing average pooling processing to obtain the information interaction features of the to-be-recommended information. The information interaction features comprehensively reflect the account features of the interactive social accounts that have generated interaction operations on the to-be-recommended information, so that the features of the to-be-recommended information are described based on the account features corresponding to the interactive social accounts, to facilitate information recommendation based on the social relationships in the social network.

[0165] In this embodiment, the information interaction features of the corresponding to-be-recommended information are obtained by aggregating the account features of the interactive social accounts that have generated interaction operations on the to-be-recommended information, so that the account features corresponding to the interactive social accounts can be fused to describe the features of the to-be-recommended information based on the account features corresponding to the interactive social accounts, to facilitate information recommendation based on the social relationships in the social network, to implement information recommendation based on the social relationships in the social network, and to improve the accuracy of information recommendation.

[0166] In one embodiment, as shown in Figure 9 , an information recommendation method is provided. The method is applied to a server in Figure 1 for example, and includes the following steps.

[0167] In step 902, account interaction features corresponding to a social network to which a target account belongs are obtained. The account interaction features are obtained based on interaction behaviors between social accounts in the social network.

[0168] In this embodiment, the account interaction features are obtained based on interaction behaviors between social accounts in the social network, and specifically include: determining statistical results of interaction behaviors between social accounts in the social network to which the target account belongs; obtaining interaction features between the social accounts in the social network based on the statistical results; and generating account interaction features corresponding to the social network based on the interaction features between the social accounts in the social network. The account interaction features reflect the interaction relationships between the social accounts in the social network.

[0169] In step 904, network account features constructed based on account features of the social accounts in the social network are determined. The network account features include account features corresponding to the target account.

[0170] In step 906, the account interaction features are standardized based on a standardization condition to obtain standardized account interaction features.

[0171] Step 908, input the normalized account interaction features and the network account features into a feature propagation mapping model for feature propagation mapping, and obtain network propagation account features output by the feature propagation mapping model;

[0172] Step 910, extract the propagation account features corresponding to the target account from the network propagation account features.

[0173] In this embodiment, the network account features are constructed according to the account features of each social account in the social network, and the network account features include the account features of all social accounts in the social network, specifically including the account features of the target account and the account features of the interactive social accounts having interaction behaviors with the target account. The account interaction features include an account interaction feature matrix, and the standardization condition is a standardization processing rule through the product of the diagonal matrix of the account interaction feature matrix and the account interaction feature matrix. The feature propagation mapping model is a machine learning model pre-trained based on a neural network algorithm. The feature propagation mapping model can perform feature propagation mapping according to the input account interaction features and the input feature of the network account features, and output network propagation account features including the corresponding propagation account features of each social account. In addition to carrying the features of the target account itself, the propagation account features also carry the account features of the social accounts having social relationships with the target account in the social network.

[0174] Step 912, obtain information interaction features of each to-be-recommended information; the information interaction features are generated based on the account features of the social accounts having generated interaction operations on the corresponding to-be-recommended information in the social network.

[0175] In this embodiment, the to-be-recommended information is text, web pages, pictures, or audio and video data, and the information interaction features are generated based on the account features of the social accounts having generated interaction operations on the corresponding to-be-recommended information in the social network, and are used to describe the features of each to-be-recommended information.

[0176] Step 914, match the propagation account features and the information interaction features by using a matching model, and obtain a matching result output by the matching model;

[0177] Step 916, determine a target matching result meeting the recommendation condition from the matching result;

[0178] Step 918, determine the to-be-recommended information corresponding to the target matching result as a target information;

[0179] Step 920, recommend the target information to the terminal corresponding to the target account.

[0180] In this embodiment, the matching model is a multi-layer perceptron model pre-trained based on a neural network algorithm. The matching model can match the propagation account features and the information interaction features according to the input, and output the matching results of the propagation account features and the information interaction features. The matching results can reflect the similarity between the propagation account features and the information interaction features. The higher the similarity between the propagation account features and the information interaction features, the more matched the propagation account features and the information interaction features are. Then, the information to be recommended corresponding to the information interaction features can be recommended to the target account. The recommendation condition is that the matching degree is greater than the matching degree threshold. The target matching result is the matching result whose matching degree is greater than the matching degree threshold. The target information is the information to be recommended corresponding to the target matching result. After the target information is determined, the server recommends the target information to the terminal corresponding to the target account.

[0181] In this embodiment, the account interaction features obtained according to the interaction behaviors between the social accounts in the social network are used to perform feature propagation mapping on the account features corresponding to the target account, so that the account features of the social accounts having the interaction behavior relationship with the target account in the social network are propagated and mapped to the target account, the propagation account features corresponding to the target account are obtained, and the propagation account features and the information interaction features of the information to be recommended are matched, the target information is determined according to the matching result, and the information recommendation is performed. Therefore, the interaction behavior relationship between the social accounts in the social network is used for information recommendation, and the accuracy of information recommendation is improved.

[0182] The application further provides an application scenario applying the information recommendation method. Specifically, the information recommendation method is applied as follows in the application scenario:

[0183] The information to be recommended in the application scenario is a short video. The short video is a kind of internet content propagation mode, which is generally a video with a length of less than 30 minutes propagated on the internet new media. With the popularization of mobile terminals and the acceleration of network, short, flat and fast mass propagation content has gradually gained the favor of major platforms, fans and capital.

[0184] Due to the short time of the rise of short video entertainment, short video recommendation has become the main battlefield of major Internet companies. Although traditional processing uses collaborative filtering and other algorithms for short video recommendation, for some late entrants, due to less accumulation, although they have a large amount of user social network information, the actual browsing behavior of users on short videos is less, resulting in poor results of directly using mature algorithms for short video recommendation. For example, in traditional short video recommendation processing, one way is a user similarity-based recommendation method, for a given user, by recommending the short videos browsed by similar users to the user to give a recommendation result; another way is a short video-based recommendation method, given a user, by recommending the most similar videos to the videos browsed by the user to give a recommendation result. In addition, in traditional recommendation processing, the matching degree-based recommendation technology mainly calculates the matching degree of a given user and a candidate set of videos, then sorts the candidate set of videos according to the matching degree, and then recommends the top several videos with the highest matching degree. However, in traditional short video recommendation processing, the prerequisite for accurate recommendation is to accumulate enough user browsing and clicking data, and when the amount of user browsing data is small, accurate short video recommendation cannot be performed.

[0185] Based on this, for platforms with a large amount of social network data, user social network data can be used for short video recommendation. Compared with traditional recommendation methods, the user's social network contains a large number of friends with the same interests, in addition, users usually have similar circles and interests with friends such as: financial circle, student circle, scientific research circle, consumption level, basketball, badminton, swimming, etc. Based on the social relationship in the social network for short video recommendation, the user account can be described by aggregating the accounts with social relationship to the user account, so as to better recommend short videos through social information and improve the accuracy of short video recommendation.

[0186] Specifically, for a social network G = (A, X), A is an account interaction feature, which is obtained according to the interaction behavior between the social accounts of each user in the social network, and A can be a user network matrix in particular. If the number of all users in the social network is N, the user network matrix A includes the interaction features between the social accounts of N users. The user network matrix A is constructed according to various interaction behavior information, such as the number of user chats, the user chat frequency, the user circle interaction evaluation rate and other interaction behavior features. X is the account feature of the user's social account. For the social network G, the users in the whole network can be connected into a user network based on the user historical behavior, such as the interaction behavior between users, and the edges between users are determined by the user intimacy. The user intimacy is determined according to the interaction behavior between users. If there are more interaction behaviors between two users, the connection weight between the two users is higher. If the interaction behavior between two users is less, the connection weight between the two users is lower. If there is no interaction behavior between two users, there is no connection between the two users. The user interaction behavior can be determined by various indicators such as the number of user interaction behaviors, the cumulative duration of interaction behaviors, the number of interactions, the interaction frequency ranking, and the number of interaction days in the last week.

[0187] For example, for users u1, u2 and u3, if u1 and u2 have more interaction times, the connection weight between u1 and u2 is higher; if u1 and u3 have less interaction times, the connection weight between u1 and u3 is lower; if u1 has no interaction with other users, u1 has no connection with other users. Further, for the intimacy between users i and j, the interaction times between user i and user j can be , and the interaction feature between user i and user j can be , and the user network matrix A is obtained according to the interaction features between all users in the social network.

[0188] Further, the account feature X of the social account of the user is obtained by vector embedding processing on the social account of the user. Specifically, each user in the social network is described by a vector feature, and the physical meaning of vector embedding is to describe the social behavior of the user by a vector, so that the vector representation of users with close social relationships is relatively close, and the vector representation of users with relatively distant social relationships is relatively different. Specifically, an unsupervised user embedding method of Node2Vec node embedding is used to embed the users in the social network, for example, the social account corresponding to each user can be taken as a node, starting from each node, multiple trajectories of random walk are generated, and all the trajectories of random walk are input into a word2vec word vector embedding model as a corpus for vector embedding to obtain the account feature X of the social account of the user. The word2vec word vector embedding model can be trained based on the word2vec algorithm, and each user can be represented by a vector. In addition, similar nodes in the social network represent similar users, and since the weights of the lines between different users in the graph are different, the influence of the weights is considered when embedding, and a weighted random walk is used to obtain the account feature of the social account of the user. Specifically, wherein, is the account feature of each user's social account, X is a matrix composed of the account features of the social accounts of all users in the social network, and N is the number of users in the social network.

[0189] Further, by aggregating the features of all users accessing a short video, each short video can be represented in a vectorized manner. The physical meaning of the vector corresponding to each short video is that the vector representation of each short video is multiplied by the features of the users who like to access or browse the short video, that is, the similarity between the user and the short video is high. Specifically, for a short video, the list of users who have accessed the short video is (u1, u2, u3, u4, … uK), and K is the total number of all users who have accessed the short video. The account features of the users who have accessed the short video can be obtained by mean pooling, that is, the information interaction feature of the short video is Correspondingly, the matrix composed of the information interaction features of all short videos is n is the number of all short videos.

[0190] Further, the information of each user is better described by the way of social relationship aggregation. Based on the features aggregated by social relationship, the information of the corresponding interactive user of the user can be considered on the basis of the user's own features, and the nature and interest of the user can be better described through the description of the user's neighbors, thereby improving the accuracy of short video recommendation. Specifically, the user's social account matrix X and the user network matrix A are used for information transmission. In order to ensure that the information amount of each user after multiple feature propagation mappings is kept in a fixed order of magnitude, the user network matrix A can be standardized, and the user network matrix A can be standardized by any one of the following three formulas:

[0191]

[0192] wherein, is the result of the standardized user network matrix A, D is a diagonal matrix, and specifically I is an identity matrix.

[0193] When the matrix X composed of the account features of all users' social accounts is mapped by feature propagation, the features not mapped by feature propagation are mapped by feature propagation by the following formula,

[0194]

[0195] wherein, is the result of the first feature propagation mapping, is a nonlinear function, which can be a RELU function, is the initial value of the first feature propagation mapping, is the mapping matrix of the first feature propagation mapping.

[0196] If the propagation number of the feature propagation mapping is K, then

[0197]

[0198] wherein, is the result of K times feature propagation mapping.

[0199] After the feature propagation mapping by the social relationship in the social network, the matching degree of the user and the short video can be predicted by the multilayer perception machine, that is, the representation of the user u and the video i is input into the multilayer perception machine to obtain the prediction result. For example, for the user u and the video i, the predicted matching degree of the user-short video pair ui is as follows,

[0200]

[0201] wherein, is the matching degree between the user u and the short video i, MLP is a multi-layer perception, is the account feature of the user u, is the feature of the short video i. In training the multi-layer perception, the MLP and the mapping matrix parameter in the feature propagation mapping can be trained by randomly screening the positive samples (i.e. the user-video pairs actually occurring) and the negative samples (i.e. the user-video pairs not occurring).

[0202] Further, in recommending the short videos for the target user, the network account features U of all users in the social network can be acquired first, and then when the user u triggers the short video browsing request, the similarity between the user u and all short videos in the candidate set is compared through the above short video information recommendation method, and then a part of short videos with the highest ranking are selected for recommendation, so that the social relationship in the social network is utilized for the short video recommendation, and the accuracy of the short video recommendation is improved.

[0203] The application further provides an application scenario applying the above information recommendation method. Specifically, the information recommendation method is applied in the application scenario as follows:

[0204] The user logs in the e-book platform through the social account in the social network, and when the user triggers the e-book recommendation, the server acquires the account interaction feature corresponding to the social network to which the user account belongs, performs feature propagation mapping on the account feature corresponding to the user account according to the account interaction feature, acquires the propagation account feature corresponding to the user account according to the result of the feature propagation mapping, acquires the information interaction feature of each e-book in the e-book platform, matches the propagation account feature and each information interaction feature, and recommends the target e-book determined from each e-book to the terminal corresponding to the user account according to the matching result.

[0205] In addition, in other application scenarios, the information recommendation can also be for music, pictures, webpages and various Internet information resources.

[0206] It should be understood that, although Figure 2 , Figure 3 , Figure 5 , Figure 6 and Figure 9 the steps in the flowcharts are displayed in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, Figure 2 , Figure 3 , Figure 5 , Figure 6 and Figure 9At least one of the steps in the method can comprise a plurality of steps or stages which are not necessarily performed at the same time but can be performed at different times and which can not necessarily be performed in the order shown but can be performed alternately or in rotation with at least one of the other steps or the steps or stages of the other steps.

[0207] In one embodiment, as shown in FIG. 10, there is provided an information recommendation device 1000 which can be a software module or a hardware module or a combination of both as part of a computer device, and the device specifically comprises: an account interaction feature acquisition module 1002, a feature propagation mapping module 1004, an information interaction feature acquisition module 1006, and a target information recommendation module 1008, wherein: Figure 10 The account interaction feature acquisition module 1002 is configured to acquire account interaction features corresponding to a social network to which a target account belongs, wherein the account interaction features are obtained based on interaction behaviors between social accounts in the social network.

[0208] The feature propagation mapping module 1004 is configured to perform feature propagation mapping on account features corresponding to the target account by using the account interaction features, and to obtain propagation account features corresponding to the target account based on a result of the feature propagation mapping.

[0209] The information interaction feature acquisition module 1006 is configured to acquire information interaction features of each of the information to be recommended, wherein the information interaction features are generated based on account features of social accounts that have generated interaction operations on the corresponding information to be recommended.

[0210] The target information recommendation module 1008 is configured to match the propagation account features and the information interaction features, and to recommend target information determined from the information to be recommended to a terminal corresponding to the target account based on a result of the matching.

[0211]

[0212] ​The information recommendation device obtains the account interaction features according to the interaction behaviors between the social accounts in the social network, performs feature propagation mapping on the account features corresponding to the target account, obtains the propagation account features corresponding to the target account according to the result of the feature propagation mapping, matches the propagation account features with the information interaction features of each to-be-recommended information, and generates the information interaction features based on the account features of the social accounts that have generated interaction operations on the corresponding to-be-recommended information in the social network, and recommends the target information determined from the to-be-recommended information to the terminal corresponding to the target account according to the matching result. In the information recommendation process, the account interaction features obtained according to the interaction behaviors between the social accounts in the social network are used to perform feature propagation mapping on the account features corresponding to the target account, so that the account features of the social accounts having the interaction behavior relationship with the target account in the social network are propagated to the target account, the propagation account features corresponding to the target account are obtained, and the target information is determined according to the matching result of the propagation account features and the information interaction features of the to-be-recommended information for recommendation. Therefore, the information recommendation is performed by using the interaction behavior relationship between the social accounts in the social network, and the accuracy of the information recommendation is improved.

[0213] In an embodiment, the feature propagation mapping module 1004 includes a target interaction feature extraction module, an interaction account feature acquisition module, a propagation mapping iteration module, and a propagation account feature determination module. The target interaction feature extraction module is configured to extract target account interaction features corresponding to the target account from the account interaction features. The interaction account feature acquisition module is configured to acquire account features of an interaction social account corresponding to the target account interaction features. The propagation mapping iteration module is configured to perform iterative feature propagation mapping on the account features corresponding to the target account based on the target account interaction features, the account features of the interaction social account, and propagation mapping parameters, to obtain a result of the feature propagation mapping. The propagation account feature determination module is configured to determine the propagation account features corresponding to the target account according to the result of the feature propagation mapping.

[0214] In an embodiment, the propagation mapping iteration module includes a current account feature determination module and a propagation mapping processing module. The current account feature determination module is configured to determine the account features corresponding to the target account as current account features. The propagation mapping processing module is configured to perform feature propagation mapping on the current account features based on the target account interaction features, the account features of the interaction social account, and the propagation mapping parameters of the current iteration feature propagation mapping, to obtain a result of the current iteration feature propagation mapping. The result of the current iteration feature propagation mapping is taken as the current account features, and the step of performing feature propagation mapping on the current account features based on the target account interaction features, the account features of the interaction social account, and the propagation mapping parameters of the current iteration feature propagation mapping to obtain the result of the current iteration feature propagation mapping is returned.

[0215] In an embodiment, the propagation mapping processing module comprises a feature propagation module and a nonlinear mapping module; wherein: the feature propagation module is configured to perform feature propagation on the current account feature based on the target account interaction feature, the account feature of the interaction social account, and the propagation mapping parameter of the feature propagation mapping of the current iteration, to obtain a feature propagation result; the nonlinear mapping module is configured to perform nonlinear mapping on the feature propagation result to obtain the result of the feature propagation mapping of the current iteration.

[0216] In an embodiment, the feature propagation mapping module 1004 comprises a network account feature determination module, a model feature propagation mapping module, and a model output processing module; wherein: the network account feature determination module is configured to determine a network account feature constructed according to the account features of each social account in the social network; the network account feature comprises the account feature corresponding to the target account; the model feature propagation mapping module is configured to input the account interaction feature and the network account feature into the feature propagation mapping model for feature propagation mapping, to obtain the network propagation account feature output by the feature propagation mapping model; the model output processing module is configured to extract the propagation account feature corresponding to the target account from the network propagation account feature.

[0217] In an embodiment, the model feature propagation mapping module comprises a standardization processing module and a feature input module; wherein: the standardization processing module is configured to perform standardization processing on the account interaction feature through a standardization condition to obtain a standardized account interaction feature; the feature input module is configured to input the standardized account interaction feature and the network account feature into the feature propagation mapping model for feature propagation mapping.

[0218] In an embodiment, the target information recommendation module 1008 comprises a feature matching module, a target information determination module, and a target information processing module; wherein: the feature matching module is configured to perform matching on the propagation account feature and each information interaction feature through a matching model, to obtain a matching result output by the matching model; the target information determination module is configured to determine the target information from each to-be-recommended information based on the matching result; the target information processing module is configured to recommend the target information to the terminal corresponding to the target account.

[0219] In an embodiment, the target information determination module comprises a recommendation condition screening module and a screening result processing module; wherein: the recommendation condition screening module is configured to determine the target matching result satisfying the recommendation condition from the matching result; the screening result processing module is configured to determine the to-be-recommended information corresponding to the target matching result as the target information.

[0220] In an embodiment, the information recommendation device further comprises a statistical result determination module, an interaction feature obtaining module, and an account interaction feature generation module; wherein: the statistical result determination module is configured to determine statistical results of interaction behaviors between each social account in a social network to which the target account belongs; the interaction feature obtaining module is configured to obtain interaction features between each social account in the social network based on the statistical results; and the account interaction feature generation module is configured to generate an account interaction feature corresponding to the social network according to the interaction features between each social account in the social network.

[0221] In an embodiment, the information recommendation device further comprises a social network determination module, a node embedding module, and an account feature determination module; wherein: the social network determination module is configured to determine a social network to which the target account belongs; the node embedding module is configured to perform node embedding on each node in the social network to obtain a node feature corresponding to each node respectively; each node corresponds to each social account in the social network, and a node relationship between each node corresponds to an interaction behavior between each social account; and the account feature determination module is configured to determine an account feature corresponding to the target account based on the node features corresponding to each node.

[0222] In an embodiment, the node embedding module comprises a weight determination module, a walk module, and a walk track processing module; wherein: the weight determination module is configured to determine walk weights between each node in the social network according to a node relationship between each node in the social network; the walk module is configured to perform node walk in the social network based on the walk weights with each node as a starting point to form a walk track of each node; and the walk track processing module is configured to perform feature embedding on the walk track of each node through an embedding model to obtain the node feature corresponding to each node respectively.

[0223] In an embodiment, the information recommendation device further comprises an interaction social account determination module and an information interaction feature obtaining module; wherein: the interaction social account determination module is configured to determine an interaction social account corresponding to each to-be-recommended information; the interaction social account is a social account in the social network that has generated an interaction operation on the corresponding to-be-recommended information; and the information interaction feature obtaining module is configured to perform feature aggregation on an account feature corresponding to each interaction social account to obtain an information interaction feature of the corresponding to-be-recommended information.

[0224] Specific limitations of the information recommendation device can be referred to the limitations of the information recommendation method in the foregoing, which will not be described herein again. Each module in the information recommendation device described above can be realized by software, hardware, and combinations thereof, in whole or in part. Each module described above 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.

[0225] In an embodiment, 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. 1. Figure 11 The computer device includes a processor, a memory and a network interface connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement an information recommendation method.

[0226] Those skilled in the art can understand that Figure 11 The structure shown in FIG. 1 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.

[0227] In an embodiment, a computer device is provided, which includes a memory and a processor. The memory stores a computer program. The processor executes the computer program to implement the steps in the above method embodiments.

[0228] In an embodiment, a computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steps in the above method embodiments.

[0229] In an embodiment, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium. The processor executes the computer instructions to cause the computer device to perform the steps in the above method embodiments.

[0230] 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, storage, database or other medium used in each embodiment provided by 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 or optical memory. Volatile memory can include random access memory (RAM) or external cache memory. 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).

[0231] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of each technical feature 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.

[0232] 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. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. An information recommendation method, characterized in that, The method includes: Obtain the account interaction characteristics corresponding to the social network to which the target account belongs; the account interaction characteristics are obtained based on the interaction behavior between various social accounts in the social network; Extract target account interaction features corresponding to the target account from the account interaction features; the target account interaction features include interaction features obtained based on the interaction behavior between the target account and the interactive social account, the interactive social account being a social account in the social network that has interaction behavior with the target account; Obtain the account characteristics of the interactive social accounts corresponding to the target account's interaction characteristics; Based on the target account interaction features, the account features of the interactive social account, and the propagation mapping parameters, iterative feature propagation mapping is performed on the account features corresponding to the target account to obtain the feature propagation mapping result; wherein, in each iteration of feature propagation mapping, the result of the current iteration of feature propagation mapping is used as the initial value for the next feature propagation mapping; the propagation mapping parameters are determined through network model training; The propagation account features corresponding to the target account are determined based on the results of the feature propagation mapping. The information interaction features of each piece of information to be recommended are obtained; the information interaction features are generated based on the account features of social accounts that have interacted with the corresponding information to be recommended in the social network. The characteristics of the dissemination account and the information interaction characteristics are matched, and the target information determined from the information to be recommended based on the matching results is recommended to the terminal corresponding to the target account.

2. The method according to claim 1, characterized in that, The step of iteratively propagating and mapping the account features corresponding to the target account based on the interaction features of the target account, the account features of the interactive social account, and the propagation mapping parameters to obtain the result of the feature propagation mapping includes: The account characteristics corresponding to the target account are determined as the current account characteristics; Based on the target account interaction features, the account features of the interactive social account, and the propagation mapping parameters of the current iteration feature propagation mapping, feature propagation mapping is performed on the current account features to obtain the result of the current iteration feature propagation mapping. The steps are as follows: taking the result of the feature propagation mapping in this iteration as the current account feature, and returning the propagation mapping parameters based on the target account interaction features, the account features of the interactive social account, and the feature propagation mapping in this iteration to perform feature propagation mapping on the current account feature to obtain the result of the feature propagation mapping in this iteration.

3. The method according to claim 2, characterized in that, The step of performing feature propagation mapping on the current account features based on the target account interaction features, the account features of the interactive social account, and the propagation mapping parameters of the current iteration feature propagation mapping, to obtain the result of the current iteration feature propagation mapping, includes: Based on the target account interaction features, the account features of the interactive social account, and the propagation mapping parameters of the feature propagation mapping in this iteration, feature propagation is performed on the current account features to obtain the feature propagation result; The feature propagation results are then subjected to a nonlinear mapping to obtain the feature propagation mapping result for this iteration.

4. The method according to claim 1, characterized in that, The method further includes: Determine network account features constructed based on the account characteristics of each social account in the social network; the network account features include the account features corresponding to the target account; The account interaction features and the network account features are input into the feature propagation mapping model for feature propagation mapping to obtain the network propagation account features output by the feature propagation mapping model. Extract the propagation account features corresponding to the target account from the network propagation account features.

5. The method according to claim 4, characterized in that, The step of inputting the account interaction features and the network account features into the feature propagation mapping model for feature propagation mapping includes: The account interaction features are standardized by applying standardized conditions to obtain standardized account interaction features. The standardized account interaction features and the network account features are input into the feature propagation mapping model for feature propagation mapping.

6. The method according to claim 1, characterized in that, The step of matching the characteristics of the dissemination account with the characteristics of each of the information interaction features, and recommending the target information determined from each of the information to be recommended to the terminal corresponding to the target account based on the matching results, includes: The matching model is used to match the characteristics of the dissemination accounts and the characteristics of each information interaction to obtain the matching result output by the matching model. Target information is determined from each of the information to be recommended based on the matching results; The target information is recommended to the terminal corresponding to the target account.

7. The method according to claim 6, characterized in that, The step of determining the target information from each of the information to be recommended based on the matching results includes: From the matching results, determine the target matching result that meets the recommendation criteria; The information to be recommended corresponding to the target matching result is determined as the target information.

8. The method according to claim 1, characterized in that, The method further includes: Statistical results of interactions between various social media accounts within the social network to which the target account belongs; Based on the statistical results, the interaction characteristics between various social accounts in the social network are obtained; Based on the interaction characteristics between various social accounts in the social network, account interaction characteristics corresponding to the social network are generated.

9. The method according to any one of claims 1 to 8, characterized in that, The method further includes: Identify the social network to which the target account belongs; Node embedding is performed on each node in the social network to obtain the node features corresponding to each node; each node corresponds to each social account in the social network, and the node relationship between each node corresponds to the interaction behavior between each social account. Based on the node characteristics corresponding to each node, the account characteristics corresponding to the target account are determined.

10. The method according to claim 9, characterized in that, The step of embedding nodes in the social network to obtain the node features corresponding to each node includes: Based on the node relationships between nodes in the social network, determine the walk weights between the nodes; Starting from each node, node traversal is performed in the social network based on the traversal weight to form the traversal trajectory of each node; By embedding features into the walking trajectories of each node using an embedding model, the node features corresponding to each node are obtained.

11. The method according to claim 1, characterized in that, The method further includes: Identify the corresponding social accounts for each piece of information to be recommended; the social accounts are social accounts in the social network that have interacted with the corresponding information to be recommended. The account features corresponding to each of the aforementioned interactive social accounts are aggregated to obtain the information interaction features of the corresponding information to be recommended.

12. An information recommendation device, characterized in that, The device includes: The account interaction feature acquisition module is used to acquire the account interaction features corresponding to the social network to which the target account belongs; the account interaction features are obtained based on the interaction behavior between various social accounts in the social network; The feature propagation mapping module is used to extract target account interaction features corresponding to the target account from the account interaction features; the target account interaction features include interaction features obtained based on the interaction behavior between the target account and the interactive social account, where the interactive social account is a social account in the social network that has interaction behavior with the target account; obtain the account features of the interactive social account corresponding to the target account interaction features; based on the target account interaction features, the account features of the interactive social account, and propagation mapping parameters, perform iterative feature propagation mapping on the account features corresponding to the target account to obtain the feature propagation mapping result; wherein, in each iteration of feature propagation mapping, the result of the current iteration of feature propagation mapping is used as the initial value for the next feature propagation mapping; the propagation mapping parameters are determined through network model training; and the propagation account features corresponding to the target account are determined based on the result of the feature propagation mapping. The information interaction feature acquisition module is used to acquire the information interaction features of each piece of information to be recommended; the information interaction features are generated based on the account features of social accounts that have interacted with the corresponding information to be recommended in the social network. The target information recommendation module is used to match the characteristics of the dissemination account with the information interaction characteristics, and recommend the target information determined from the information to be recommended based on the matching results to the terminal corresponding to the target account.

13. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 11.

14. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 11.

Images