CN112395515A

CN112395515A - Information recommendation method and device, computer equipment and storage medium

Info

Publication number: CN112395515A
Application number: CN202110066081.5A
Authority: CN
Inventors: 陈昊
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2021-01-19
Filing date: 2021-01-19
Publication date: 2021-02-23
Anticipated expiration: 2041-01-19
Also published as: CN112395515B

Abstract

The embodiment of the application discloses an information recommendation method, an information recommendation device, computer equipment and a storage medium, and the embodiment of the application can acquire user historical behavior data of a plurality of users; constructing a user network of interactive behavior relation among users according to the historical user behavior data, wherein the user network comprises user nodes corresponding to the users and edges used for representing interactive behaviors among different users; extracting a feature vector of each user node based on the user network; performing information propagation operation of multiple levels according to the user network and the feature vector to obtain propagation information corresponding to each propagation layer; fusing the propagation information corresponding to the adjacent propagation layers to obtain fused information; and determining the matching degree between every two user nodes based on the fusion information, and recommending information to the user according to the matching degree. The accuracy and the reliability of information recommendation are improved.

Description

Information recommendation method and device, computer equipment and storage medium

Technical Field

The application relates to the technical field of internet, in particular to an information recommendation method and device, computer equipment and a storage medium.

Background

With the development of internet technology, various types of software on the terminal are more and more abundant, for example, a user can make friends through a friend making function of friend making software, so that the friend making software becomes an immediate need for solving social problems of the user. Meanwhile, the friend making software can automatically recommend a stranger candidate list for the user so that the user can select an object to be handed in as a friend from the stranger candidate list.

At present, a friend recommendation mode of existing friend making software generally performs matching recommendation based on geographic positions: the method comprises the steps of obtaining the current geographical position of a user, obtaining strangers within a preset range from the user based on the current geographical position of the user, and recommending the strangers to the user for the user to select whether the strangers need to be added as friends or not. In the geographic position matching recommendation method, because the recommendation is performed only based on the geographic position, people who have different signs and are not suitable for the same language are difficult to have, the matching success rate is reduced, and the recommendation accuracy is reduced. Or recommending based on the user characteristic rule: matching the users through some psychological and behavioral rules by means of questionnaire interaction and other forms, and recommending strangers with high matching degree with the users to the users for the users to select whether the strangers need to be added as friends or not. In the user characteristic rule recommendation mode, the user can be matched based on certain psychological and behavioral rules, although the psychological and behavioral rules have certain effects, the rule-based method is still rigid, personalized recommendation cannot be provided for the user, the recommendation accuracy and reliability are reduced, and the satisfaction degree of the user in using friend-making software is reduced.

Disclosure of Invention

The embodiment of the application provides an information recommendation method, an information recommendation device, computer equipment and a storage medium, which can improve the accuracy and reliability of information recommendation.

In order to solve the above technical problem, an embodiment of the present application provides the following technical solutions:

the embodiment of the application provides an information recommendation method, which comprises the following steps:

acquiring user historical behavior data of a plurality of users;

constructing a user network of interactive behavior relation among users according to the historical user behavior data, wherein the user network comprises user nodes corresponding to the users and edges used for representing interactive behaviors among different users;

extracting a feature vector of each user node based on the user network;

performing information propagation operation of multiple levels according to the user network and the feature vector to obtain propagation information corresponding to each propagation layer;

fusing the propagation information corresponding to the adjacent propagation layers to obtain fused information;

and determining the matching degree between every two user nodes based on the fusion information, and recommending information to the user according to the matching degree.

According to an aspect of the present application, there is also provided an information recommendation apparatus including:

the device comprises an acquisition unit, a storage unit and a processing unit, wherein the acquisition unit is used for acquiring user historical behavior data of a plurality of users;

the building unit is used for building a user network of interactive behavior relation among users according to the historical user behavior data, wherein the user network comprises user nodes corresponding to the users and edges used for representing interactive behaviors among different users;

an extraction unit, configured to extract a feature vector of each user node based on the user network;

the propagation unit is used for carrying out information propagation operation of a plurality of levels according to the user network and the characteristic vector so as to obtain propagation information corresponding to each propagation layer;

the fusion unit is used for fusing the propagation information corresponding to the adjacent propagation layers to obtain fusion information;

the determining unit is used for determining the matching degree between every two user nodes based on the fusion information;

and the recommending unit is used for recommending information to the user according to the matching degree.

In an embodiment, the adjacent propagation layers include a current propagation layer, a first propagation layer adjacent to the current propagation layer and propagating from a first direction, and a second propagation layer adjacent to the current propagation layer and propagating from a second direction, and the merging unit is specifically configured to: determining fusion coefficients corresponding to the current propagation layer, the first propagation layer and the second propagation layer; fusing the propagation information corresponding to the current propagation layer, the first propagation layer and the second propagation layer based on the fusion coefficient to obtain initial fusion information; and performing pooling operation on the initial fusion information to obtain fusion information.

In an embodiment, the propagation unit is specifically configured to: standardizing the user network to obtain a standardized user network; and step-by-step information transmission is carried out on the standardized user network and the characteristic vector to a plurality of transmission layers through preset transmission parameters, and transmission information corresponding to each transmission layer is obtained.

In one embodiment, the extraction unit comprises:

the track extraction subunit is used for extracting at least one interaction track formed by connection between user nodes from the user network, wherein the interaction track is used for representing that interaction behaviors exist between users corresponding to the user nodes;

and the characteristic extraction subunit is used for extracting the characteristic vector of each user node based on the interaction track.

In an embodiment, the trajectory extraction subunit is specifically configured to: determining a weight value corresponding to each edge in the user network; and performing wandering clustering operation by taking each user node in the user network as a starting point according to the weight value to obtain at least one interaction track formed by the connection of the user nodes.

In an embodiment, the building unit is specifically configured to: extracting interaction times, interaction accumulated duration, interaction contents and interaction frequency among the users based on the historical behavior data of the users; and constructing a user network of the interactive behavior relation among the users according to at least one of the interactive times, the interactive accumulated time, the interactive contents and the interactive frequency.

In an embodiment, the determining unit is specifically configured to: predicting a similarity probability value between every two user nodes based on the fusion information through a multilayer perceptron; and determining the matching degree between every two user nodes according to the similarity probability value.

In an embodiment, the recommending unit is specifically configured to: screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair; and recommending the user corresponding to the user node in the target user node pair to serve as a friend to the user corresponding to the user node in the target user node pair.

In an embodiment, the recommending unit is specifically configured to: screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair; acquiring a preference item of a user corresponding to a user node in the target user node pair; recommending the preference item to a user corresponding to the other user node in the target user node pair.

According to an aspect of the present application, there is also provided a computer device, including a processor and a memory, where the memory stores a computer program, and the processor executes any one of the information recommendation methods provided by the embodiments of the present application when calling the computer program in the memory.

According to an aspect of the present application, there is also provided a storage medium for storing a computer program, which is loaded by a processor to execute any one of the information recommendation methods provided by the embodiments of the present application.

According to the method and the device, the user historical behavior data of a plurality of users can be obtained, the user network of the interactive behavior relation among the users is constructed according to the user historical behavior data, and the user network comprises user nodes corresponding to the users and edges used for representing the interactive behavior among different users; then, feature vectors of all user nodes can be extracted based on a user network, and information propagation operation of multiple levels is carried out according to the user network and the feature vectors so as to obtain propagation information corresponding to each propagation layer; at this time, the propagation information corresponding to the adjacent propagation layers can be fused to obtain fusion information, the matching degree between every two user nodes is determined based on the fusion information, and information is recommended to the user according to the matching degree. According to the scheme, the propagation information corresponding to each propagation layer is obtained based on the user network constructed by the historical behavior data of the user and the feature vectors of the user nodes obtained by extraction, the propagation information corresponding to the adjacent propagation layers is fused to obtain the fusion information, the matching degree between every two user nodes is accurately determined based on the fusion information to accurately recommend the information to the user, and the accuracy and the reliability of information recommendation are improved.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.

Fig. 1 is a scene schematic diagram of an application of an information recommendation method provided in an embodiment of the present application;

fig. 2 is a schematic flowchart of an information recommendation method provided in an embodiment of the present application;

FIG. 3 is a schematic diagram of a user network provided by an embodiment of the present application;

FIG. 4 is a schematic diagram of a user network constructed by interaction times according to an embodiment of the present application;

fig. 5 is a schematic diagram of fusion information obtained by fusing propagation information of adjacent propagation layers according to an embodiment of the present application;

fig. 6 is another schematic flowchart of an information recommendation method provided in an embodiment of the present application;

FIG. 7 is a schematic diagram of an information recommendation device provided in an embodiment of the present application;

fig. 8 is a schematic structural diagram of a server according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

The embodiment of the application provides an information recommendation method and device, computer equipment and a storage medium.

Referring to fig. 1, fig. 1 is a schematic view of a scene of an application of an information recommendation method provided in an embodiment of the present application, where the application of the information recommendation method may include an information recommendation device, the information recommendation device may be specifically integrated in a server 10, the server 10 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server that provides basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a Content Delivery Network (CDN), and a big data and artificial intelligence platform, but is not limited thereto. The server 10 and the terminal 20 may be directly or indirectly connected through wired or wireless communication, and the application is not limited herein. The terminal 20 may be a mobile phone, a tablet computer, a notebook computer, a desktop computer, or a wearable device.

The server 10 may obtain user historical behavior data of a plurality of users, and construct a user network of an interactive behavior relationship between the users according to the user historical behavior data, where the user network includes user nodes corresponding to the users and edges used for representing interactive behaviors among different users; then, feature vectors of all user nodes can be extracted based on a user network, and information propagation operation of multiple levels is carried out according to the user network and the feature vectors so as to obtain propagation information corresponding to each propagation layer; at this time, the propagation information corresponding to the adjacent propagation layers may be fused to obtain fusion information, the matching degree between every two user nodes is determined based on the fusion information, and information is recommended to the user according to the matching degree, for example, other users who may be added as friends are recommended to the terminal 20 corresponding to the user with a higher matching degree, or promotion information related to the article is recommended to the terminal 20 corresponding to the user with a higher matching degree, and the accuracy and reliability of information recommendation are improved.

It should be noted that the scenario diagram of the information recommendation method application shown in fig. 1 is only an example, and the information recommendation method application and the scenario described in the embodiment of the present application are for more clearly illustrating the technical solution of the embodiment of the present application, and do not form a limitation on the technical solution provided in the embodiment of the present application.

The following are detailed below. It should be noted that the following description of the embodiments is not intended to limit the preferred order of the embodiments.

The information recommendation method provided by the embodiment of the application can relate to technologies such as a machine learning technology in artificial intelligence, for example, the feature vectors of user nodes can be extracted through the machine learning technology in artificial intelligence, propagation information corresponding to each propagation layer is obtained based on a user network and the feature vectors, the propagation information corresponding to adjacent propagation layers is fused to obtain fusion information, the matching degree between every two user nodes is determined based on the fusion information, and the like, and the artificial intelligence technology and the machine learning technology are explained first below.

Artificial intelligence is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. Artificial intelligence infrastructures generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operating/interactive systems, and mechatronics. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

Machine Learning (ML) is a multi-domain cross discipline, involving multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal learning.

In the present embodiment, description will be made from the perspective of an information recommendation apparatus, which may be specifically integrated in a computer device such as a server.

Referring to fig. 2, fig. 2 is a flowchart illustrating an information recommendation method according to an embodiment of the present application. The information recommendation method can comprise the following steps:

s101, obtaining user historical behavior data of a plurality of users.

The user historical behavior data may be data of an interactive behavior generated between a user and other users through application software (also referred to as an application program) installed on the terminal, for example, the user historical behavior data may be data of an interactive behavior generated between a user and other users through application software such as friend-making software, instant messaging software, video software or sports software. It should be noted that the user historical behavior data may also be data of an interaction behavior generated between the user and another user through a phone call, a short message, or the like, or the user historical behavior data may also be behavior data of the user and another user connected to the same local area network (e.g., a WIFI network), or the like. The user historical behavior data can comprise the number of interactions between the user and other users, the accumulated time length of the interactions, the interaction content, the interaction frequency and the like.

After the user passes through the user historical behavior data generated between the application software installed on the terminal and other users, the user historical behavior data can be uploaded to the server, so that the server can store the user historical behavior data into a database, the database can store user historical behavior data corresponding to a plurality of different users, and for example, the user identification of the user and the user historical behavior data corresponding to the user can be stored in an associated manner. When the user historical behavior data of the multiple users needs to be obtained, the server may obtain the user historical behavior data of the multiple users from the database, for example, the user historical behavior data generated by the multiple users within a preset time interval range (for example, the last month or six months, and the like) may be obtained from the database, and for example, the user historical behavior data generated by the multiple users within a preset region range (for example, a certain province or a certain city, and the like) may be obtained from the database.

S102, constructing a user network of interactive behavior relations among users according to historical behavior data of the users, wherein the user network comprises user nodes corresponding to the users and edges used for representing interactive behaviors among different users.

In one embodiment, the constructing of the user network of the interactive behavior relationship between the users according to the historical behavior data of the users may include: extracting interaction times, interaction accumulated duration, interaction contents and interaction frequency among the users based on historical behavior data of the users; and constructing a user network of the interactive behavior relation among the users according to at least one of the interactive times, the interactive accumulated time length, the interactive contents and the interactive frequency.

In order to improve flexibility and convenience of user network construction, a server may connect a plurality of users into a user network based on user historical behavior data, for example, as shown in fig. 3, the user network may include user nodes corresponding to the users, and edges for characterizing that there is an interaction behavior between different users, each edge may be correspondingly provided with a weight (e.g., a to j), the weight of the edge may be determined by user affinity, which may be determined by any one or more of interaction times, interaction accumulated time, interaction content, interaction frequency, and the like, for example, the higher the interaction times, the higher the affinity; the longer the interactive accumulated time is, the higher the intimacy is; the higher the interaction frequency, the higher the affinity, etc. That is, if there are more interactive behaviors between two users, the weight of the connection line (i.e., edge) between the user nodes corresponding to the two users is higher, and if there are less interactive behaviors between the two users, the weight of the connection line between the user nodes corresponding to the two users is lower. And if no interactive behavior exists between the two users, no connection exists between the user nodes corresponding to the two users.

The measurement of the user interaction behavior can be jointly determined by the variables such as the interaction times, the interaction accumulated time length, the interaction content, the interaction frequency and the like among the users. For example, the number of interactions, the accumulated duration of interactions, the interaction content, and the interaction frequency between users may be extracted based on the historical behavior data of the users, the interaction frequency may include an interaction frequency ranking, the number of interaction days in a preset time period (for example, the last week), and the like, and the interaction content may include specific content of an interaction discussion and the number thereof, and the like. And then constructing a user network of the interactive behavior relationship among the users according to at least one of the interactive times, the interactive accumulated time length, the interactive contents and the interactive frequency.

For example, taking a user network for constructing an interactive behavior relationship between users based on the number of interactions as an example, taking users as user nodes, and directly taking the number of interactions between users as the weight of a connection (i.e., an edge) between the user nodes, as shown in fig. 4, the number of interactions between user u1 and user u2 is large (e.g., 1000), the weight of the connection between u1 and u2 is high, the number of interactions between user u1 and user u3 is small (e.g., 10), the weight of the connection between u1 and u3 is low, user u1 does not interact with other users, and user u1 does not have a connection with other users.

For another example, taking a user network for constructing an interaction behavior relationship among users based on the number of interactions, the cumulative duration of interactions, the interaction content, and the interaction frequency as an example, taking users as user nodes, setting corresponding weight coefficients a1, a2, a3, and a4 for the number of interactions X1, the cumulative duration of interactions X2, the interaction content X3, and the interaction frequency X4, respectively, and then calculating the weight of a connection line (i.e., an edge) among the user nodes according to the number of interactions X1, the cumulative duration of interactions X2, the interaction content X3, the interaction frequency X4, and the corresponding weight coefficients a1, a2, a3, and a 4: x1 a1+ X2 a2+ X3 a3+ X4 a 4.

It should be noted that, for convenience of subsequent calculation, the user network may be represented by a matrix, for example, taking user i and user j as an example, in order to better describe intimacy between users, the number of interactions between user i and user j is ci, j, and the interaction behavior relationship between user i and user j is described by log (1+ ci, j), where the constructed user network Ai may be represented by matrix a, and j = log (1+ ci, j).

For another example, taking user i and user j as an example, in order to better describe the intimacy between users, the number of interactions between user i and user j is c_i,jThe cumulative interactive duration of the user i and the user j is d_i,jThe interactive contents of the user i and the user j are e_i,jThe interaction frequency of the user i and the user j is f_i,jBy log (1+ c)_i,j+d_i,j+e_i,j+f_i,j) To describe the interaction behavior relationship between the user i and the user j, the matrix a can be used to represent the user network a constructed by the matrix a_i,j=log(1+c_i,j+d_i,j+e_i,j+f_i,j)。

And S103, extracting the characteristic vector of each user node based on the user network.

The feature vector may be a vector for characterizing feature information of the user, for example, the feature information of the user may include age, gender, constellation, location, hobbies, and psychological images.

In one embodiment, extracting the feature vector of each user node based on the user network may include: extracting at least one interaction track formed by connection between user nodes from a user network, wherein the interaction track is used for representing that interaction behaviors exist between users corresponding to the user nodes; and extracting the feature vector of each user node based on the interaction track.

In order to improve the accuracy of feature vector extraction, the server may extract at least one interaction track formed by connection between user nodes from the user network, where the interaction track is used to represent that there is an interaction behavior between users corresponding to the user nodes. For example, at least one interaction trajectory may be obtained by starting from each user Node in the user network in a Node embedding manner of the word vector generation model Node2Vec according to a preset policy or by random walk (e.g., clustering) based on a given user network. Wherein, for the user nodes on the same interaction track, greater similarity exists.

In one embodiment, extracting at least one interaction trajectory formed by connections between user nodes from a user network may include: determining a weight value corresponding to each edge in a user network; and performing wandering clustering operation by taking each user node in the user network as a starting point according to the weight value to obtain at least one interaction track formed by the connection of the user nodes.

For example, each edge in the user network may be correspondingly provided with a weight value, so that the weight value corresponding to each edge in the user network may be determined, and a wandering clustering operation (i.e., random wandering with a right, such as a route with a high weight value may be walked preferentially) is performed from each user node in the user network as a starting point according to the weight value, to obtain at least one interaction track formed by connection between the user nodes, so as to perform clustering based on the weight value, and may improve reliability of interaction track generation.

After the interaction trajectory is obtained, a feature vector of each user node may be extracted based on the interaction trajectory, and the feature vector may be represented by a matrix X, where the matrix X includes the feature vector of each user node. For example, the interaction trajectory may be input to a word vector embedding algorithm word2vec as a corpus, and a feature vector of each user node is extracted through the word2vec algorithm, where the feature vector is used to represent feature information of a user corresponding to the user node, and feature vectors corresponding to similar user nodes in a user network are relatively similar.

Due to the special situation of the friend-making software, the information such as the user's head portrait, speaking, and circle of friends can be converted into a feature vector describing the user's nature through image Processing technology, Natural Language Processing (NLP) technology, or user tagging technology without violating the individual privacy.

And S104, performing information propagation operation of multiple hierarchies according to the user network and the feature vector to acquire propagation information corresponding to each propagation layer.

The number of the propagation layers can be flexibly set according to actual needs. The propagation information may include information obtained by information transmission through propagation layers of multiple hierarchical layers, the propagation information merges all information such as an interaction behavior of each user node in the user network and a feature vector of each user node, and the propagation information may be referred to as graph representation information of the user.

In an embodiment, performing a plurality of levels of information propagation operations according to the user network and the feature vector to obtain propagation information corresponding to each propagation level may include: standardizing the user network to obtain a standardized user network; and step-by-step information propagation of the standardized user network and the characteristic vector to a plurality of propagation layers is carried out through preset propagation parameters, and propagation information corresponding to each propagation layer is obtained.

In order to improve the accuracy and reliability of the acquisition of the broadcast information, information can be transmitted by utilizing the characteristic vector X of the user node and the matrix A representing the user network. Firstly, the matrix a corresponding to the user network may be processed: in order to ensure that the information amount of each user can be kept at a fixed magnitude after multiple information transmissions, the matrix A corresponding to the user network can be standardized to obtain a standardized matrix corresponding to the user network

In which the standardization isThe processing mode can comprise:

normalization processing method (1):

(ii) a Normalization processing method (2):

(ii) a Normalization processing method (3):

. Where D may be a diagonal matrix,

and I may be an identity matrix.

A preset propagation parameter may then be obtained, which may be a mapping matrix M that may be used to map the user network

And mapping the feature vector X into information with higher dimensionality, and gradually transmitting the standardized user network and feature vector to a plurality of transmission layers through preset transmission parameters to obtain transmission information corresponding to each transmission layer.

For example, the information dissemination operation can be performed by the following formula:

wherein the content of the first and second substances,

is the mapping matrix (i.e. propagation parameter) of the l-th layer, the value of l may be an integer starting from 0, and if the final number of propagation layers is K, the mapping matrix may be obtained

To represent the propagation information of each layer, the value of K can beThe propagation information of the layer 0 which is not propagated can be expressed as

. X contains the characteristic information (i.e. the characteristic vector) of all users, A contains the adjacent information of all users in the user network after standardization,

the information of all users in each propagation layer is contained in the information.

And S105, fusing the propagation information corresponding to the adjacent propagation layers to obtain fused information.

The adjacent propagation layers may be flexibly set according to actual needs, for example, the adjacent propagation layers may include a current propagation layer and a previous propagation layer adjacent to the current propagation layer (e.g., a second propagation layer and a first propagation layer), for example, the adjacent propagation layers may include the current propagation layer and a next propagation layer adjacent to the current propagation layer (e.g., a second propagation layer and a third propagation layer), for example, the adjacent propagation layers may include the current propagation layer, the previous propagation layer adjacent to the current propagation layer, the next propagation layer adjacent to the current propagation layer, and the like (e.g., a second propagation layer, a first propagation layer, and a third propagation layer).

In an embodiment, the adjacent propagation layers include a current propagation layer, a first propagation layer adjacent to the current propagation layer and propagating from a first direction, and a second propagation layer adjacent to the current propagation layer and propagating from a second direction, and the fusing the propagation information corresponding to the adjacent propagation layers to obtain the fused information may include: determining fusion coefficients corresponding to a current propagation layer, a first propagation layer and a second propagation layer; fusing the propagation information corresponding to the current propagation layer, the first propagation layer and the second propagation layer based on the fusion coefficient to obtain initial fusion information; and performing pooling operation on the initial fusion information to obtain fusion information.

To avoid propagation information when the number of propagation layers K is large (3 or more)

The information which is more inclined to represent the user neighbor is not the user information, and in order to improve the reliability of the fusion, the fused information obtained by the fusion can accurately represent the context crossing information of the propagation information corresponding to each propagation layer, and the propagation information of the context crossing of the adjacent propagation layers can be fused through a Graph Neural Network (GNN), and at this time, the Graph Neural Network can be called as a context crossing Graph Neural Network. Specifically, a current propagation layer, a first propagation layer adjacent to the current propagation layer and propagating from a first direction, and a second propagation layer adjacent to the current propagation layer and propagating from a second direction may be determined, where the first direction and the second direction may be flexibly set according to actual needs, for example, as shown in fig. 5, taking 5 propagation layers as an example, the propagation layers may be taken as

As the current propagation layer, with the propagation layer

As and

a first propagation layer adjacent and propagating upward (i.e., from a first direction) to propagate the layer

As and

a second propagation layer adjacent and propagating downward (i.e., from a second direction). In fig. 5, the propagation layer is shown

Its upper layer representation may not be considered (i.e., the first propagation layer propagating upward may not be considered), and for the propagation layer

Its underlying representation may not be considered (i.e., the second propagation layer that propagates downward may not be considered).

Then, a fusion coefficient corresponding to the current propagation layer may be obtained

Fusion coefficient corresponding to the first transmission layer

And corresponding fusion coefficient of the second propagation layer

And the fusion coefficients corresponding to the current propagation layer, the first propagation layer and the second propagation layer can be flexibly set according to actual needs, and specific values are not limited here. At this time, the fusion coefficient corresponding to the current propagation layer can be obtained through the graph neural network

Fusion coefficient corresponding to the first transmission layer

And a fusion coefficient corresponding to the second propagation layer

The propagation information corresponding to the current propagation layer, the first propagation layer and the second propagation layer is respectively transmitted

Performing fusion to obtain initial fusion information

. For example, the fusion processing method may be as follows:

wherein the content of the first and second substances,

the corresponding fusion coefficients of the first propagation layer may be represented,

may represent the corresponding fusion coefficient of the current propagation layer,

the corresponding fusion coefficients of the second propagation layer may be represented,

may represent the propagation information of the first propagation layer,

it is possible to represent the propagation information of the current propagation layer,

the propagation information of the second propagation layer can be represented,

initial fusion information resulting from the current fusion may be represented.

After merging the propagation information corresponding to a plurality of pairs of adjacent propagation layers, a plurality of sets of initial merging information can be obtained, for example

. Is obtained by

The final fused information E may then be obtained by pooling the initial fused information, e.g.,

. Wherein, the poolThe mode of operation may include average pooling or maximum pooling, among others. After the fused information E is obtained, the fusion information E,

may represent the fused information representation of the ith user.

And S106, determining the matching degree between every two user nodes based on the fusion information, and recommending information to the user according to the matching degree.

The matching degree may be used to represent the similarity between users corresponding to two user nodes, and if the matching degree between two users is higher, it indicates that there is more similar information between the two users, and conversely, if the matching degree between two users is lower, it indicates that there is less similar information between the two users.

In an embodiment, determining the matching degree between each two user nodes based on the fusion information may include: predicting a similarity probability value between every two user nodes based on the fusion information through a multilayer perceptron; and determining the matching degree between every two user nodes according to the similarity probability value.

For example, the server may predict a similarity probability value between each two user nodes based on the fusion information through a Multilayer Perceptron (MLP), and determine a matching degree between each two user nodes according to the similarity probability value. For example, the similarity probability value may be subjected to a weighted operation or other operations, and the matching degree between every two user nodes is obtained through the operations, or the similarity probability value may be directly used as the matching degree between every two user nodes. For example, for user node i and user node j, fusion information based on user node i is obtained through a multi-layer perceptron

And fusion information of user node j

And calculating the matching degree between the user node i and the user node j to be

。

In one embodiment, recommending information to the user according to the matching degree may include: screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair; and recommending the user corresponding to the node of one user in the target user node pair as a friend to the user corresponding to the node of the other user in the target user node pair.

The preset threshold value can be flexibly set according to actual needs, after the matching degree between every two users is obtained, two user nodes with the matching degree larger than the preset threshold value can be screened out, a target user node pair is obtained, and the matching degree of the target user node pair is larger than the preset threshold value, which indicates that the two users of the target user node pair are relatively similar (for example, behavior habits or preferences are relatively similar). At this time, the user corresponding to the one user node in the target user node pair may be recommended as a friend. For example, for a user a and a user B corresponding to the target user node pair, the user B may be recommended to the user a for the user a to select whether to add the user B as a friend, or the user a may be recommended to the user B for the user B to select whether to add the user a as a friend, thereby realizing a personalized recommendation of a friend-making object for the user.

It should be noted that the server may receive the recommendation request sent by the terminal, calculate the matching degree between the user corresponding to the terminal and other users according to the above manner based on the recommendation request, screen out other users with higher matching degree, sort the other users with higher matching degree according to the order of matching degree from high to low, and recommend the other users with higher matching degree after sorting to the user corresponding to the terminal, for example, account information of the other users with higher matching degree may be recommended, so that the user adds other users as friends through the account information.

In one embodiment, recommending information to the user according to the matching degree may include: screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair; acquiring a preference item of a user corresponding to a user node in a target user node pair; and recommending preference items to the user corresponding to the other user node in the target user node pair.

After the matching degree between every two users is obtained, two user nodes with the matching degree larger than a preset threshold value can be screened out to obtain a target user node pair, and if the matching degree of the target user node pair is larger than the preset threshold value, it indicates that the two users of the target user node pair are relatively similar (e.g., behavior habits or preferences are relatively similar). At this time, the preference item of the user corresponding to one user node in the target user node pair can be obtained, the preference item can include various items such as video programs, musical works, entertainment venues, articles for daily use and foods, and then the preference item can be recommended to the user corresponding to the other user node in the target user node pair. For example, for a user a and a user B corresponding to the target user node pair, a movie program that the user B likes to watch may be recommended to the user a, so that the user a may select whether to watch the movie program, or a movie program that the user a likes to watch may be recommended to the user B, so that the user B may select whether to watch the movie program, thereby realizing accurate recommendation of a preference item for the user.

The method described in the above embodiments is further illustrated in detail by way of example.

In this embodiment, taking an information recommendation device integrated in a server as an example, please refer to fig. 6, and fig. 6 is a schematic flow chart of an information recommendation method provided in this embodiment of the present application. The method flow can comprise the following steps:

s201, obtaining user historical behavior data of a plurality of users.

For example, the server may obtain the historical behavior data of the users generated by a plurality of users within a preset time interval (for example, the last month or six months, etc.) from a database in which the historical behavior data of the users are stored in advance.

For another example, the server may obtain the historical behavior data of the users generated by a plurality of users within a preset region (e.g., a certain province or a certain city, etc.) from a database in which the historical behavior data of the users are stored in advance.

S202, constructing a user network of interactive behavior relation among users according to the historical behavior data of the users.

The user network comprises user nodes corresponding to the users and edges (namely connecting lines or connecting lines) used for representing the existence of interactive behaviors among different users, and each edge can determine a weight value corresponding to the edge according to at least one of interaction times, interaction accumulated time, interaction content and interaction frequency.

For example, the server may construct a user network of interaction behavior relationships between users based on the number of interactions: the user can be used as a user node, the number of times of interaction between the users is directly used as the weight of the edge connected between the user nodes, the number of times of interaction between the user i and the user j is large, the weight of the edge connected between the user i and the user j is high, the number of times of interaction between the user i and the user j is small, and the weight of the edge connected between the user i and the user j is low. If the number of the interaction times of the user i and the user j is c_i,jThen can pass through log (1 +)c_i,j) To describe the interaction behavior relationship between the user i and the user j, the matrix a can be used to represent the user network a constructed by the matrix a_i,j=log(1+c_i,j)。

And S203, standardizing the user network to obtain the standardized user network.

In order to ensure that the information amount of each user can be kept at a fixed magnitude after multiple information transmissions, the server can standardize the matrix A corresponding to the user network to obtain a standardized matrix A corresponding to the user network

Wherein, the normalization processing mode may include: (1):

；（2）：

；（3）：

. Where D may be a diagonal matrix,

and I may be an identity matrix.

S204, at least one interaction track formed by connection of the user nodes is extracted from the user network, and the feature vector of each user node is extracted based on the interaction track.

The feature vector may include a vector corresponding to feature information such as age, gender, constellation, location, and hobby of the user. The server can generate a Node embedding mode of a model Node2Vec through a word vector, starts from each user Node in the user network, and walks according to a preset strategy or randomly based on a weight value corresponding to each edge in the user network to obtain at least one interaction track. Wherein, for the user nodes on the same interaction track, greater similarity exists. It should be noted that the word vector generation model Node2Vec may be a model that is trained in advance based on training samples. After the interaction track is obtained, the interaction track is used as a corpus and input into a word vector embedding algorithm word2vec, the feature vector of each user node is extracted through the word2vec algorithm, and the feature vector can be represented by a matrix X.

And S205, performing information propagation operation of multiple hierarchies according to the normalized user network and the feature vector to acquire propagation information corresponding to each propagation layer.

For example, the server may propagate the normalized user network and feature vector to multiple propagation layers step by step through the graph neural network based on a preset propagation parameter (the propagation parameter may be the mapping matrix M), to obtain propagation information corresponding to each propagation layer, that is, the graph neural network represents the user by aggregating information of neighbors for multiple times. For example, the information dissemination operation can be performed by the following formula:

wherein the content of the first and second substances,

The propagation information of each layer is represented, the value of K can be flexibly set according to actual needs, and the propagation information of the 0 th layer which is not propagated can be represented as

S206, fusing the propagation information corresponding to the adjacent propagation layers to obtain initial fusion information.

Wherein the adjacent propagation layers may include a current propagation layer, a first propagation layer adjacent to the current propagation layer and propagating from a first direction, and a second propagation layer adjacent to the current propagation layer and propagating from a second direction.

Because the graph neural network represents the user by aggregating the neighbor information for multiple times, and continuously aggregating the neighbor information by using the graph neural network can lead the propagation information to represent more importance on the neighbor information of the user, and the propagation information of the user at the current propagation layer can be gradually forgotten, thereby bringing about a serious over-smoothing problem and further influencing the user matching prediction effect, in order to better retain the information of the user at the current propagation layer, the propagation information (such as user diagram representation information) at the context level and the current level can be comprehensively considered when describing the propagation layer representation of each propagation layer of each user based on the context cross graph neural network, the user can be represented by a context cross mode to better retain the user diagram representation information at different levels, thereby the user, friends, etc. can be more accurately retained when comparing the matching degrees of two users, And comparing the matching degree between the two users at different angles such as the neighborhood.

Specifically, the server may obtain a fusion coefficient corresponding to the current propagation layer

Fusion coefficient corresponding to the first transmission layer

And corresponding fusion coefficient of the second propagation layer

Wherein the fusion coefficient

、

And

can be flexibly set according to actual needs. At this time, the server may use a graph neural network (also referred to as a context cross-graph neural network) to perform fusion on the basis of the fusion coefficient corresponding to the current propagation layer

Fusion coefficient corresponding to the first transmission layer

And a fusion coefficient corresponding to the second propagation layer

Performing fusion to obtain initial fusion information

. For example, the fusion processing method may be as follows:

wherein the content of the first and second substances,

may represent the propagation information of the first propagation layer,

the propagation information of the second propagation layer can be represented,

The graph neural network may be a network trained in advance based on training samples, and for example, the graph neural network and the fusion coefficient W are trained by randomly screening training samples such as positive samples (pairs of users with interactive behaviors) and negative samples (pairs of users without interactive behaviors).

And S207, performing pooling operation on the initial fusion information to obtain fusion information.

. Is obtained by

The server may then pool the initial fusion information through the neural network to obtain final fusion information E, e.g.,

. The mode of the pooling operation may include an average pooling mode or a maximum pooling mode. After the fused information E is obtained, the fusion information E,

may represent the fused information representation of the ith user.

And S208, determining the matching degree between every two user nodes based on the fusion information.

The server can predict the matching degree between every two user nodes based on the fusion information through the multi-layer perceptron. For example, for user node i and user node j, fusion information based on user node i is obtained through a multi-layer perceptron

And fusion information of user node j

。

It should be noted that the multi-layer perceptron may be a model trained in advance based on training samples, for example, the multi-layer perceptron is trained by randomly screening the training samples such as positive samples (pairs of users with interactive behavior) and negative samples (pairs of users without interactive behavior).

S209, screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair.

The preset threshold value can be flexibly set according to actual needs, after the matching degree between every two users is obtained, the two user nodes with the matching degree larger than the preset threshold value can be screened out, the target user node pair is obtained, and if the matching degree of the target user node pair is larger than the preset threshold value, the two users of the target user node pair are similar.

S210, recommending the user information of the user corresponding to the user node in the target user node pair to the user corresponding to the other user node in the target user node pair.

The user information may include instant messaging account information of the user or a link address corresponding to a preferred item of the user. The server can recommend the user information of the user corresponding to the one user node in the target user node pair to the user corresponding to the other user node in the target user node pair, so that the user can select to view and receive or refuse the recommendation and the like.

In the above embodiments, the descriptions of the embodiments have respective emphasis, and parts that are not described in detail in a certain embodiment may refer to the above detailed description of the information recommendation method, and are not described herein again.

According to the method and the device, the server constructs the user network based on the historical user behavior data, extracts the obtained feature vectors of the user nodes, and conducts information propagation operation of multiple levels according to the user network and the feature vectors so as to obtain the propagation information corresponding to each propagation layer. And then, the propagation information corresponding to the adjacent propagation layers can be fused to obtain fused information, and the matching degree between every two user nodes is accurately determined based on the fused information to accurately recommend information to the user, so that intelligent and personalized recommendation is realized for the user, the accuracy and reliability of information recommendation are improved, and the matching success rate is improved.

In order to better implement the information recommendation method provided by the embodiment of the present application, the embodiment of the present application further provides a device based on the information recommendation method. The meanings of the nouns are the same as those in the information recommendation method, and specific implementation details can refer to the description in the method embodiment.

Referring to fig. 7, fig. 7 is a schematic structural diagram of an information recommendation apparatus according to an embodiment of the present application, where the information recommendation apparatus may include an obtaining unit 301, a constructing unit 302, an extracting unit 303, a propagating unit 304, a fusing unit 305, a determining unit 306, a recommending unit 307, and the like.

The acquiring unit 301 is configured to acquire user historical behavior data of a plurality of users.

The constructing unit 302 is configured to construct a user network of an interactive behavior relationship between users according to the historical behavior data of the users, where the user network includes user nodes corresponding to the users and edges used to represent that there are interactive behaviors between different users.

An extracting unit 303, configured to extract a feature vector of each user node based on the user network.

And a propagation unit 304, configured to perform information propagation operations of multiple hierarchies according to the user network and the feature vector, so as to obtain propagation information corresponding to each propagation layer.

The merging unit 305 is configured to merge propagation information corresponding to adjacent propagation layers to obtain merged information.

A determining unit 306, configured to determine a matching degree between each two user nodes based on the fusion information.

And the recommending unit 307 is configured to recommend information to the user according to the matching degree.

In an embodiment, the adjacent propagation layers include a current propagation layer, a first propagation layer adjacent to the current propagation layer and propagating from a first direction, and a second propagation layer adjacent to the current propagation layer and propagating from a second direction, and the merging unit 305 may be specifically configured to: determining fusion coefficients corresponding to a current propagation layer, a first propagation layer and a second propagation layer; fusing the propagation information corresponding to the current propagation layer, the first propagation layer and the second propagation layer based on the fusion coefficient to obtain initial fusion information; and performing pooling operation on the initial fusion information to obtain fusion information.

In an embodiment, the propagation unit 304 may be specifically configured to: standardizing the user network to obtain a standardized user network; and step-by-step information propagation of the standardized user network and the characteristic vector to a plurality of propagation layers is carried out through preset propagation parameters, and propagation information corresponding to each propagation layer is obtained.

In an embodiment, the extracting unit 303 may include:

the track extraction subunit is used for extracting at least one interactive track formed by connection between user nodes from the user network, wherein the interactive track is used for representing interactive behaviors existing between users corresponding to the user nodes;

In an embodiment, the trajectory extraction subunit may be specifically configured to: determining a weight value corresponding to each edge in a user network; and performing wandering clustering operation by taking each user node in the user network as a starting point according to the weight value to obtain at least one interaction track formed by the connection of the user nodes.

In an embodiment, the constructing unit 302 may specifically be configured to: extracting interaction times, interaction accumulated duration, interaction contents and interaction frequency among the users based on historical behavior data of the users; and constructing a user network of the interactive behavior relation among the users according to at least one of the interactive times, the interactive accumulated time length, the interactive contents and the interactive frequency.

In an embodiment, the determining unit 306 may specifically be configured to: predicting a similarity probability value between every two user nodes based on the fusion information through a multilayer perceptron; and determining the matching degree between every two user nodes according to the similarity probability value.

In an embodiment, the recommending unit 307 may specifically be configured to: screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair; and recommending the user corresponding to the node of one user in the target user node pair to serve as a friend to the user corresponding to the node of the other user in the target user node pair.

In an embodiment, the recommending unit 307 may specifically be configured to: screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair; acquiring a preference item of a user corresponding to a user node in a target user node pair; and recommending preference items to the user corresponding to the other user node in the target user node pair.

In the embodiment of the application, the obtaining unit 301 may obtain the historical user behavior data of a plurality of users, and the constructing unit 302 constructs a user network of the interactive behavior relationship between the users according to the historical user behavior data, where the user network includes user nodes corresponding to the users and edges used for representing the interactive behavior existing between different users; then, the extracting unit 303 may extract the feature vector of each user node based on the user network, and the propagation unit 304 may perform information propagation operations of multiple levels according to the user network and the feature vector to obtain propagation information corresponding to each propagation layer; at this time, the fusion unit 305 may fuse the propagation information corresponding to the adjacent propagation layers to obtain fusion information, the determining unit 306 determines the matching degree between every two user nodes based on the fusion information, and the recommending unit 307 recommends information to the user according to the matching degree. According to the scheme, the propagation information corresponding to each propagation layer is obtained based on the user network constructed by the historical behavior data of the user and the feature vectors of the user nodes obtained by extraction, the propagation information corresponding to the adjacent propagation layers is fused to obtain the fusion information, the matching degree between every two user nodes is accurately determined based on the fusion information to accurately recommend the information to the user, and the accuracy and the reliability of information recommendation are improved.

The embodiment of the present application further provides a computer device, where the computer device may be a server, as shown in fig. 8, which shows a schematic structural diagram of a server according to the embodiment of the present application, and specifically:

the server may include components such as a processor 401 of one or more processing cores, memory 402 of one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will appreciate that the server architecture shown in FIG. 8 is not meant to be limiting, and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components. Wherein:

the processor 401 is a control center of the server, connects various parts of the entire server using various interfaces and lines, and performs various functions of the server and processes data by running or executing software programs and/or modules stored in the memory 402 and calling data stored in the memory 402, thereby performing overall monitoring of the server. Optionally, processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor, which mainly handles operating systems, user interfaces, application programs, etc., and a modem processor, which mainly handles wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 401.

The memory 402 may be used to store software programs and modules, and the processor 401 executes various functional applications and data processing by operating the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data created according to the use of the server, and the like. Further, the memory 402 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 access to the memory 402.

The server further includes a power supply 403 for supplying power to each component, and preferably, the power supply 403 may be logically connected to the processor 401 through a power management system, so as to implement functions of managing charging, discharging, and power consumption through the power management system. The power supply 403 may also include any component of one or more dc or ac power sources, recharging systems, power failure detection circuitry, power converters or inverters, power status indicators, and the like.

The server may also include an input unit 404, the input unit 404 being operable to receive input numeric or character information and to generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

Although not shown, the server may further include a display unit and the like, which will not be described in detail herein. Specifically, in this embodiment, the processor 401 in the server loads the executable file corresponding to the process of one or more application programs into the memory 402 according to the following instructions, and the processor 401 runs the application program stored in the memory 402, thereby implementing various functions as follows:

acquiring user historical behavior data of a plurality of users, and constructing a user network of interactive behavior relation among the users according to the user historical behavior data, wherein the user network comprises user nodes corresponding to the users and edges used for representing interactive behaviors among different users; extracting a characteristic vector of each user node based on a user network, and performing information propagation operation of multiple levels according to the user network and the characteristic vector to obtain propagation information corresponding to each propagation layer; fusing the propagation information corresponding to the adjacent propagation layers to obtain fused information; and determining the matching degree between every two user nodes based on the fusion information, and recommending information to the user according to the matching degree.

According to an aspect of the application, a computer program product or computer program is provided, comprising computer instructions, the computer instructions being stored in a computer readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the method provided in the various alternative implementations of the above embodiments.

It will be understood by those skilled in the art that all or part of the steps of the methods of the embodiments described above may be performed by computer instructions, or by computer instructions controlling associated hardware, which may be stored in a computer-readable storage medium and loaded and executed by a processor. To this end, the present application provides a storage medium, in which a computer program is stored, where the computer program may include computer instructions, and the computer program can be loaded by a processor to execute any one of the information recommendation methods provided in the present application.

The above operations can be implemented in the foregoing embodiments, and are not described in detail herein.

Wherein the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.

Since the instructions stored in the storage medium can execute the steps in any information recommendation method provided in the embodiments of the present application, beneficial effects that can be achieved by any information recommendation method provided in the embodiments of the present application can be achieved, which are detailed in the foregoing embodiments and will not be described herein again.

The information recommendation method, apparatus, computer device and storage medium provided by the embodiments of the present application are described in detail above, and a specific example is applied in the present application to explain the principle and implementation manner of the present application, and the description of the above embodiments is only used to help understand the method and core ideas of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.

Claims

1. An information recommendation method, comprising:

acquiring user historical behavior data of a plurality of users;

extracting a feature vector of each user node based on the user network;

2. The information recommendation method according to claim 1, wherein the adjacent propagation layers include a current propagation layer, a first propagation layer adjacent to the current propagation layer and propagating from a first direction, and a second propagation layer adjacent to the current propagation layer and propagating from a second direction, and the fusing propagation information corresponding to the adjacent propagation layers to obtain fused information includes:

determining fusion coefficients corresponding to the current propagation layer, the first propagation layer and the second propagation layer;

fusing the propagation information corresponding to the current propagation layer, the first propagation layer and the second propagation layer based on the fusion coefficient to obtain initial fusion information;

and performing pooling operation on the initial fusion information to obtain fusion information.

3. The information recommendation method according to claim 1, wherein the performing a plurality of levels of information propagation operations according to the user network and the feature vector to obtain propagation information corresponding to each propagation layer comprises:

standardizing the user network to obtain a standardized user network;

and step-by-step information transmission is carried out on the standardized user network and the characteristic vector to a plurality of transmission layers through preset transmission parameters, and transmission information corresponding to each transmission layer is obtained.

4. The information recommendation method according to claim 1, wherein said extracting feature vectors of user nodes based on the user network comprises:

extracting at least one interaction track formed by connection between user nodes from the user network, wherein the interaction track is used for representing that interaction behaviors exist between users corresponding to the user nodes;

and extracting the feature vector of each user node based on the interaction track.

5. The information recommendation method according to claim 4, wherein said extracting at least one interaction track formed by connections between user nodes from the user network comprises:

determining a weight value corresponding to each edge in the user network;

and performing wandering clustering operation by taking each user node in the user network as a starting point according to the weight value to obtain at least one interaction track formed by the connection of the user nodes.

6. The information recommendation method according to claim 1, wherein the constructing a user network of interactive behavior relationships among users according to the user historical behavior data comprises:

extracting interaction times, interaction accumulated duration, interaction contents and interaction frequency among the users based on the historical behavior data of the users;

and constructing a user network of the interactive behavior relation among the users according to at least one of the interactive times, the interactive accumulated time, the interactive contents and the interactive frequency.

7. The information recommendation method according to claim 1, wherein the determining a matching degree between each two user nodes based on the fusion information comprises:

predicting a similarity probability value between every two user nodes based on the fusion information through a multilayer perceptron;

and determining the matching degree between every two user nodes according to the similarity probability value.

8. The information recommendation method according to any one of claims 1 to 7, wherein recommending information to a user according to the matching degree comprises:

screening out two user nodes with the matching degree larger than a preset threshold value to obtain a target user node pair;

and recommending the user corresponding to the user node in the target user node pair to serve as a friend to the user corresponding to the user node in the target user node pair.

9. The information recommendation method according to any one of claims 1 to 7, wherein recommending information to a user according to the matching degree comprises:

acquiring a preference item of a user corresponding to a user node in the target user node pair;

recommending the preference item to a user corresponding to the other user node in the target user node pair.

10. An information recommendation apparatus, comprising:

11. A computer device comprising a processor and a memory, the memory having a computer program stored therein, the processor executing the information recommendation method according to any one of claims 1 to 9 when calling the computer program in the memory.

12. A storage medium for storing a computer program which is loaded by a processor to execute the information recommendation method according to any one of claims 1 to 9.