A social recommendation method and system for enhancing neighbor relationships

Abstract

The present invention relates to the technical field of social recommendation, and provides a social recommendation method and system for enhancing neighbor relationships, comprising: obtaining a user set, an item set, and user ratings of items in a social network; and performing social recommendations for each user using a social recommendation model that enhances neighbor relationships based on the obtained user set, item set, and user ratings of items. The social recommendation model that enhances neighbor relationships establishes a mapping between users and items, calculates the intimacy between the neighbors of the user and the item, samples the neighbors of the user and the item based on the intimacy, and calculates the weights of the sampled neighbors based on the intimacy using a self-attention mechanism to obtain a project social space representation and a user social space representation. The system then combines the project's own feature representation and the user's own feature representation to obtain a project feature representation and a user feature representation. This reduces the influence of invalid neighbors and improves the accuracy of social recommendations.

Description

Technical Field

[0001] The present invention belongs to the technical field of social recommendation, and in particular relates to a social recommendation method and system for enhancing neighbor relationships. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of social networks, people are increasingly influenced by their social connections when choosing social interactions. Therefore, social connections, as auxiliary information, play an important role in improving the accuracy of social recommendations, such as Taobao product recommendations and Weibo information recommendations. However, due to the complexity of social connections, existing models face many problems:

[0004] 1. Interference caused by invalid neighbor relationships. Due to the low cost of establishing online social relationships, people are becoming more casual in choosing social connections. For example, although Taobao friends and Weibo friends have established neighbor relationships on the Internet, most of them are different from the user's preferences. The social recommendation model based on the Graph Neural Networks for Social Recommendation (GraphRec) does not consider the interference of invalid neighbors, which has a significant impact on the accuracy of social recommendations. Therefore, how to identify and enhance the weight of valid neighbors and reduce the weight of invalid neighbors in the model is a problem that needs to be solved.

[0005] 2. Lack of effective means to select valid neighbors. The original data contains a large number of trust relationships, and the reliability of trust relationships varies from user to user; therefore, how to judge the validity of trust relationships is a problem that needs to be solved.

[0006] 3. Lack of completeness of the project's social network. For example, in the Deep Social Collaborative Filtering (DSCF) model, although the user-project interaction method is proposed, the project's social network is not modeled. The lack of this auxiliary information has a significant impact on the accuracy of recommendations. Therefore, how to complete the project's social information is also a problem that needs to be solved. Summary of the Invention

[0007] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a social recommendation method and system for enhancing neighbor relationships, which weakens the influence of invalid neighbors through intimacy calculation and self-attention mechanism, thereby improving the accuracy of social recommendations.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] A first aspect of the present invention provides a social recommendation method for enhancing neighbor relationships, comprising:

[0010] Obtain the user set, item set, and user ratings of items in the social network;

[0011] Based on the obtained user set, item set, and user ratings of items, social recommendations are made for each user through a social recommendation model that enhances neighbor relationships.

[0012] Among them, the social recommendation model that enhances neighbor relationships establishes a mapping between users and items, calculates the intimacy of the neighbors of users and items, samples the neighbors of users and items based on the intimacy, and calculates the weights of the sampled neighbors according to the intimacy through the self-attention mechanism. Then, the project social space representation and the user social space representation are calculated, and the project feature representation and the user feature representation are obtained by combining the project's own feature representation and the user's own feature representation.

[0013] Furthermore, for each item, the social recommendation model that enhances neighbor relationships concatenates the embeddings and rating embeddings of the interacting users, obtains the interactive opinions between the item and the user through a multi-layer perceptron, aggregates the interactive opinions between the item and different users, and obtains the feature representation of the item itself.

[0014] Furthermore, for each user, the social recommendation model that enhances neighbor relationships concatenates the embeddings and score embeddings of the interacted items, obtains the user's perceived interaction with the items through a multi-layer perceptron, aggregates the user's perceived interaction with different items, and obtains the user's own feature representation.

[0015] Furthermore, the social recommendation model for enhancing neighbor relationships aggregates the user's sampled neighbors by weights of the user's sampled neighbors, and obtains the user's social space representation based on the aggregated information through a single-layer perceptron.

[0016] Furthermore, the social recommendation model for enhancing neighbor relationships aggregates the sampled neighbors of an item through the weights of the sampled neighbors of the item, and obtains the social space representation of the item based on the aggregated information through a single-layer perceptron.

[0017] Furthermore, the social recommendation model for enhancing neighbor relationships integrates the project social space representation and the project's own feature representation through a multi-layer perceptron to obtain the project feature representation;

[0018] or,

[0019] The social recommendation model for enhancing neighbor relationships integrates the user social space representation and the user's own feature representation through a multi-layer perceptron to obtain the user feature representation.

[0020] Furthermore, the social recommendation model for enhancing neighbor relationships concatenates the item feature representation and the user feature representation, and predicts a score through a multi-layer perceptron.

[0021] A second aspect of the present invention provides a social recommendation system for enhancing neighbor relationships, comprising:

[0022] A data acquisition module is configured to: acquire a user set, an item set, and user ratings of items in a social network;

[0023] A social recommendation module is configured to: make social recommendations for each user based on the obtained user set, item set, and user ratings of the items by using a social recommendation model that enhances neighbor relationships;

[0024] Among them, the social recommendation model that enhances neighbor relationships establishes a mapping between users and items, calculates the intimacy of the neighbors of users and items, samples the neighbors of users and items based on the intimacy, and calculates the weights of the sampled neighbors according to the intimacy through the self-attention mechanism. Then, the project social space representation and the user social space representation are calculated, and the project feature representation and the user feature representation are obtained by combining the project's own feature representation and the user's own feature representation.

[0025] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the social recommendation method for enhancing neighbor relationships as described above.

[0026] A fourth aspect of the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the social recommendation method for enhancing neighbor relationships as described above are implemented.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention provides a social recommendation method for enhancing neighbor relationships. It reduces the proportion of invalid neighbors in social recommendations through similar neighbors and self-attention mechanisms, solves the problem of cheap trust relationships; selects neighbors with the same preferences as valid neighbors, thereby solving the problem of selecting valid neighbors; and improves the accuracy of social recommendations.

[0029] The present invention provides a social recommendation method for enhancing neighbor relationships, which establishes a project social network through duality, completes the project social information, and solves the problem of incomplete auxiliary information. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0031] Figure 1 4 is a structural diagram of a social recommendation model for enhancing neighbor relationships according to a first embodiment of the present invention;

[0032] Figure 2 Schematic diagram of a social network according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0034] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0035] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0036] Explanation of terms:

[0037] GraphRec, a social recommendation model based on graph neural network model: adopts user-item embedding expression as the underlying embedding, which is more detailed; in addition, GraphRec provides a modeling idea, which is to construct its own characteristics and social characteristics from two modules, namely user and item.

[0038] Example 1

[0039] This embodiment provides a social recommendation method for enhancing neighbor relationships, which specifically includes the following steps:

[0040] Step 1: Obtain a data set, that is, obtain the user set, item set, trust relationship between users, and user ratings of items in a social network (graph network or relationship graph network).

[0041] Specifically, a social network can be a social relationship on Weibo. Each user in the user set C can be a registered user on Weibo. Each item in the item set B can be a Weibo post (information) published by a user on Weibo. Different users have different opinions (comments) on a given item. Preprocessing (emotion recognition) converts user comments into numerical ratings, which are the user's ratings of the item.

[0042] like Figure 2 As shown, it is a relationship graph network with user u1 as the central node, which includes: explicit user-user relationship, with two opposite relationships; explicit user-item relationship, with 5 levels of ratings; implicit user-user relationship; implicit item-item relationship.

[0043] Step 2: Based on the obtained user set, item set, and user ratings of items, a social recommendation model that enhances neighbor relationships is used to perform social recommendations for each user.

[0044] Since social models require auxiliary information such as trust relationships, but the trust relationships in the dataset are unreliable, this paper builds a social recommendation model Sim-GraphRec based on the GraphRec model to enhance neighbor relationships. The main contribution is to build more reliable neighbor relationships based on the GraphRec model and refer to the consistency method in the ConsisRec model. The mapping is innovative, and the mapping is to construct nonlinear relationships through a multi-layer perceptron.

[0045] like Figure 1 As shown in the figure, Sim-GraphRec mainly consists of three parts: embedding layer (using multi-layer perceptron (a small multi-layer neural network model)), social layer (divided into attention network, graph neural network, etc., each network can be divided into L layers) and scoring layer (using multi-layer perceptron).

[0046] The embedding layer is divided into two parts: project space and user space.

[0047] Item Space: For each item in item set B, we aggregate item information using the users who interacted with the item and their ratings. Because different users may have different opinions about the same item, collecting these opinions helps model item features. It also helps filter neighboring user relationships and enrich hidden information.

[0048] Interaction opinion f between item j and interacting user t jt is represented as follows:

[0049]

[0050] in, Represents the connection operation between two vectors, p t is the embedding of user t, r e is the score embedding, g u For multi-layer perceptron, better fusion of initial embedding is achieved.

[0051] Since the same project may have different user interactions, the different interaction opinions of the project are aggregated and expressed as That is, the feature representation of project j is obtained:

[0052]

[0053] Among them, I represents the characteristic representation of itself, and the corresponding S below represents the characteristic representation of society, ∑ t∈B(j) is the aggregation function, μ jt and b are weight and bias parameters respectively; W is the weight matrix, and σ is the activation function.

[0054] User space: Similarly, an approximate method is used to construct different interactive opinion representations of users Users can rate different items. Rating different items helps capture user preferences and is more conducive to item modeling.

[0055] For item a that user i interacts with, we introduce the perceived interaction x ia 、Project embeddingq a and score embedding r e , the combination of item embedding and rating embedding is passed through a multi-layer perceptron g v (), output perception interaction:

[0056]

[0057] in, Represents the connection operation between two vectors; q a is the embedding of item a, r e To score embedding, better fusion of initial embedding is achieved.

[0058] In order to reduce the impact of special values, Each pair (q a ,r e ) assign different weights α ia , then aggregate the different interactive opinions of users and express them as

[0059]

[0060] Among them, ∑ a∈C(i) is the aggregation function, α iaand b are weight and bias parameters; W is the weight matrix, and σ is the activation function.

[0061] In the social layer, user-item interactions are used as the baseline and ratings as auxiliary information to filter out low-intimacy neighbor relationships. The self-attention mechanism weakens the influence of these relationships to prevent the model from overfitting due to invalid data interference. Then, the social influence of the project and the social influence of the user are obtained. Finally, the results of the embedding layer and the social layer are linearly combined to obtain user features and item features, such as Figure 1 shown.

[0062] (1) The social impact of the project.

[0063] The similarity between two items is usually calculated by users clicking or commenting on the items. Therefore, the intimacy between items (microblogs) is calculated by using the ratings between users and items, and the intimacy relationship is used to construct the social space of the model. Specifically, the neighboring items of the items in the graph network are used to construct the social influence. First, the social space item representation of node (item) j is defined as (superscript s denotes social influence), which is represented by the different social neighbors of item j in the lth hidden layer (information representation embedding) (i.e., item j’s social neighbor item j in layer l * The self-characteristic representation ) aggregation, M(j) is the set of items similar or related to item j (i.e., the set of neighbors of item j), and the item social space is represented by as follows:

[0064]

[0065] Among them, k() is a single-layer perceptron, AGG (l) represents the aggregation function of the lth hidden layer, and σ represents the activation function.

[0066] Among them, there is no accurate standard for the neighbor relationship in the project, and the existing models rarely involve the neighbor selection of the project, so a mapping module is added to establish a direct connection between users and projects, which makes it easier to find the neighbors of the project. Define a mapping function q ij , used to dynamically select the sampled neighbors, the established mapping formula is as follows:

[0067]

[0068] Among them, q ij is the mapping embedding between item j and user i, are the embeddings of user node i and item node j respectively. g() is a multi-layer perceptron, which is more conducive to discovering the nonlinear relationship between items and user nodes.

[0069] The mapping between project j and all interacting users is embedded into the query q that constitutes project j. Afterwards, the strong and weak neighboring projects should be distinguished. The goal is to select neighbors with strong intimacy, so the consistency formula is used to judge the strength of similarity. The neighbor node j of project j * The sampling at level l is defined by the strong and weak closeness of all neighbors of the mapping q of item j as p (l) (j * ;q):

[0070]

[0071]

[0072] Among them, s (l) (k * ; q) represents the neighbor k of item j in the lth layer network * and the closeness score of query q, Represents neighbor item node k * The information at layer l is represented as embedded.

[0073] Get item j and all neighbor items j * The strong and weak intimacy (i.e., the probability representation of neighbors), set the ratio of sampled neighbors, and calculate the number of sampled neighbors M for the project * , select M with the greatest closeness to project j * The number of sampled neighbors is proportional to the total number of neighbors, where the ratio is 0 ≤ γ ≤ 1. Therefore, if a node is connected to more nodes, more neighbors will be sampled.

[0074] In order to integrate the neighbor item information with strong intimacy, the aggregation function AGG (l) Further analysis is carried out, specifically: the self-attention mechanism is used for each sampled neighbor item node j of item j * Assigned attention weight β j :

[0075]

[0076] in, is the jth * The importance of sampled neighbors, M * To sample the total number of neighbor items, a self-attention mechanism is used to collect all neighbor node information of the previous layer. Calculate and find the attention weight

[0077]

[0078] in, is the importance factor, l is the index of the hidden layer, l = 1, 2, ..., L. The rating of the project is embedded in r e express. is a trainable parameter of the self-attention layer, and σ is the activation function. It should be noted that the self-attention layer here refers to the attention mechanism in the hidden layer of the graph network structure, and the self-attention layer can also contain hidden layers.

[0079] The aggregated information can be used as user social embedding Then, the project social embedding and project relationship embedding are integrated through a multi-layer perceptron to obtain the project feature representation z j , as follows:

[0080]

[0081] Where L is the total number of hidden layers, W l and b l are weight and bias parameters.

[0082] (2) User’s social influence.

[0083] Similarly, there are connections between users, and users are more likely to be influenced by their friends. This paper uses the self-attention mechanism to aggregate the differences in social influence. Assume that the user influence in the social space is It aggregates user u i Different social neighbor representations at level l (i.e. user i’s social neighbor user i in layer l * The self-characteristic representation ), the user social space representation formula is as follows:

[0084]

[0085] Among them, N(i) is the number of users u i A collection of similar / related items. In real life, it's relatively easy to establish connections between users, but trust between neighbors isn't always reliable. This is aimed at trust relationships between users in the original data.

[0086] The established mapping formula is as follows:

[0087]

[0088] Among them, q ij is the mapping embedding between item j and user i, are the embeddings of user node i and item node j respectively. g() is a multi-layer perceptron, which is more conducive to discovering the nonlinear relationship between items and user nodes.

[0089] The mapping embedding between user i and all interaction items constitutes the query q' of user i.

[0090] Use the same method to use user i's neighbor node i * The sampling at layer l is done by mapping q' and all the neighbor users i * Intimacy score (l) (i * ; q'), the reference consistency formula is as follows:

[0091]

[0092]

[0093] Among them, s (l) (o * ; q') represents the user neighbor o of user i in the layer l network * and the closeness score of query q', represents user i's neighbor o * The information at layer l is represented as embedded.

[0094] Get user i and all neighbor users i * The strong and weak intimacy (i.e., the probability representation of neighbors), set the ratio of sampled neighbors, and calculate the number of sampled neighbors N of the user * , select N with the largest intimacy with user i * neighbor users as the sampled neighbors of user i. The number of sampled neighbors is proportional to the total number of neighbors, where the ratio is 0 ≤ γ ≤ 1. Therefore, if a node is connected to more nodes, more neighbors will be sampled.

[0095] The self-attention mechanism is used to calculate and different neighbor users i * Embedded relevance and calculating impact factor Then, we aggregate the impact factors. The aggregated information is as follows:

[0096]

[0097] Among them, l is the index of the hidden layer, and the attention weight is calculated

[0098]

[0099] Where l is the index of the hidden layer, r e For scoring embedding, is a trainable parameter of the attention layer. The aggregated information can be used as user social embedding

[0100] Then, the user social influence embedding and user relationship embedding are integrated through a multi-layer perceptron to obtain the user feature representation h i , which is expressed as follows:

[0101]

[0102] Where L is the total number of hidden layers, W′ l and b′ l are weight and bias parameters.

[0103] After propagation through the L layer calculation, the user feature h is obtained. i and item characteristics z j The scoring layer concatenates the embedded representations and then calculates them through a multi-layer perceptron to obtain the predicted score r′ ij , as follows:

[0104]

[0105] Among them, W l and b″ l are weight and bias parameters, The meaning of is the weight matrix.

[0106] After predicting each user's ratings for all items, we recommend items (information) to users based on the predicted ratings.

[0107] The loss function of the social recommendation model with enhanced neighbor relations is defined as ij and the true score r ij The difference between , the loss function is:

[0108]

[0109] where |O| is the number of ratings in the dataset.

[0110] In addition, to address the long-standing overfitting problem when optimizing deep neural network models, a dropout strategy was adopted and an early stopping method was used in the experiment.

[0111] Comparative experiments on two public datasets, Epinions and Ciao, show that compared to the latest GraphRec+, the mean absolute error (MAE) and root mean squared error (RMSE) values have increased by 0.81%-1.09% and 1.15%-1.41%, respectively, demonstrating that the social recommendation model with enhanced neighbor relationships outperforms other models. Because the social recommendation model with enhanced neighbor relationships incorporates a neighbor sampling method and selects a 60% neighbor sampling ratio, the runtime of the social recommendation model with enhanced neighbor relationships is significantly reduced. Experimental results demonstrate that not all trust relationships in the dataset are reliable, and that the social recommendation model with enhanced neighbor relationships addresses the issue of unreliable trust relationships among friends to a certain extent.

[0112] The present invention uses similarity neighbors and self-attention mechanisms to reduce the proportion of invalid neighbors in social recommendations, solving the problem of cheap trust relationships; by constructing user-item pairs to select neighbors with the same preferences as valid neighbors, thereby solving the problem of selecting valid neighbors; through duality, a project social network is established to complete the project social information and solve the problem of incomplete auxiliary information.

[0113] Example 2

[0114] This embodiment provides a social recommendation system for enhancing neighbor relationships, which specifically includes the following modules:

[0115] A data acquisition module is configured to: acquire a user set, an item set, and user ratings of items in a social network;

[0116] A social recommendation module is configured to: make social recommendations for each user based on the obtained user set, item set, and user ratings of the items by using a social recommendation model that enhances neighbor relationships;

[0117] Among them, the social recommendation model that enhances neighbor relationships establishes a mapping between users and items, calculates the intimacy of the neighbors of users and items, samples the neighbors of users and items based on the intimacy, and calculates the weights of the sampled neighbors according to the intimacy through the self-attention mechanism. Then, the project social space representation and the user social space representation are calculated, and the project feature representation and the user feature representation are obtained by combining the project's own feature representation and the user's own feature representation.

[0118] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation processes are the same, which will not be repeated here.

[0119] Example 3

[0120] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the social recommendation method for enhancing neighbor relationships as described in the first embodiment above are implemented.

[0121] Example 4

[0122] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the social recommendation method for enhancing neighbor relationships as described in the first embodiment above are implemented.

[0123] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer-usable program code.

[0124] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0125] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0127] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0128] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A social recommendation method for enhancing neighbor relationships, characterized in that: include: Obtain the user set, item set, and user ratings of items in the social network; Based on the obtained user set, item set, and user ratings of items, social recommendations are made for each user through a social recommendation model that enhances neighbor relationships. Among them, the social recommendation model that enhances neighbor relationships establishes a mapping between users and items, calculates the intimacy between the neighbors of users and items, samples the neighbors of users and items based on the intimacy, and calculates the weights of the sampled neighbors according to the intimacy through the self-attention mechanism. It then calculates the project social space representation and the user social space representation, and combines the project's own feature representation and the user's own feature representation to obtain the project feature representation and the user feature representation; The social recommendation model with enhanced neighbor relationships, for each item, concatenates the embeddings of interacting users and the embeddings of ratings, obtains the interaction opinions between the item and the user through a multi-layer perceptron, and aggregates the interaction opinions between the item and different users to obtain the feature representation of the item itself; The social recommendation model for enhancing neighbor relationships, for each user, concatenates the embeddings and ratings of the interacted items, obtains the user's perceived interactions with the items through a multi-layer perceptron, and aggregates the user's perceived interactions with different items to obtain the user's own feature representation; The social recommendation model for enhancing neighbor relationships aggregates the user's sampled neighbors by weights of the user's sampled neighbors, and obtains the user's social space representation based on the aggregated information through a single-layer perceptron; The social recommendation model for enhancing neighbor relationships aggregates the sampled neighbors of an item through the weights of the sampled neighbors of the item, and obtains the social space representation of the item based on the aggregated information through a single-layer perceptron.

2. A social recommendation method for enhancing neighbor relationships according to claim 1, characterized in that: The social recommendation model for enhancing neighbor relationships integrates the project social space representation and the project's own feature representation through a multi-layer perceptron to obtain the project feature representation; or, The social recommendation model for enhancing neighbor relationships integrates the user social space representation and the user's own feature representation through a multi-layer perceptron to obtain the user feature representation.

3. The social recommendation method for enhancing neighbor relationships according to claim 1, wherein: The social recommendation model for enhancing neighbor relationships concatenates the item feature representation and the user feature representation, and predicts a score through a multi-layer perceptron.

4. A social recommendation system for enhancing neighbor relationships, characterized in that: include: A data acquisition module is configured to: acquire a user set, an item set, and user ratings of items in a social network; A social recommendation module is configured to: make social recommendations for each user based on the obtained user set, item set, and user ratings of the items by using a social recommendation model that enhances neighbor relationships; Among them, the social recommendation model that enhances neighbor relationships establishes a mapping between users and items, calculates the intimacy between the neighbors of users and items, samples the neighbors of users and items based on the intimacy, and calculates the weights of the sampled neighbors according to the intimacy through the self-attention mechanism. It then calculates the project social space representation and the user social space representation, and combines the project's own feature representation and the user's own feature representation to obtain the project feature representation and the user feature representation; The social recommendation model with enhanced neighbor relationships, for each item, concatenates the embeddings of interacting users and the embeddings of ratings, obtains the interaction opinions between the item and the user through a multi-layer perceptron, and aggregates the interaction opinions between the item and different users to obtain the feature representation of the item itself; The social recommendation model for enhancing neighbor relationships, for each user, concatenates the embeddings and ratings of the interacted items, obtains the user's perceived interactions with the items through a multi-layer perceptron, and aggregates the user's perceived interactions with different items to obtain the user's own feature representation; The social recommendation model for enhancing neighbor relationships aggregates the user's sampled neighbors by weights of the user's sampled neighbors, and obtains the user's social space representation based on the aggregated information through a single-layer perceptron; The social recommendation model for enhancing neighbor relationships aggregates the sampled neighbors of an item through the weights of the sampled neighbors of the item, and obtains the social space representation of the item based on the aggregated information through a single-layer perceptron.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the social recommendation method for enhancing neighbor relationships as described in any one of claims 1 to 3 are implemented.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the social recommendation method for enhancing neighbor relationships according to any one of claims 1 to 3 are implemented.

Images