A social recommendation method and system based on social group influence

Abstract

The present invention discloses a social recommendation method and system based on the influence of social groups. In view of the influence of social groups on users, a random walk algorithm is used to generate a peripheral subgraph of a user node as a candidate social group, and a memory network is used to calculate the candidate social groups, thereby capturing the main potential social group information in the social network and calculating the influence of these groups on the user; then, a gating mechanism is used to calculate the representation vector of the user under social influence, and a multi-layer perceptron is used to fuse the representation of the user under social influence and the user's project preference representation to obtain the user's final representation vector; the user's representation vector and the project's representation vector are input into a prediction module to obtain a user behavior prediction result. The present invention obtains a more accurate user representation vector by mining the social group information in the user's social network and the influence of the social group on the user, thereby achieving the purpose of improving the performance of the recommendation system.

Description

Technical Field

[0001] The present application relates to the technical field of information recommendation, and in particular to a social recommendation method and system based on the influence of social groups. Background Art

[0002] In recent years, with the development and rapid iteration of Internet technology and terminal devices, people can carry out rich and colorful activities on the Internet. At the same time, the amount of information on the Internet has also exploded, which has brought a variety of content to users. However, in the face of such complex information, it has become more difficult for users to obtain interesting and useful information. In the era of big data, although people can obtain a lot of information, most of the information is inefficient or even useless, which is called "information overload". For information service providers and merchant platforms, they need to consider how to provide users with more effective and valuable information, while generating user traffic and creating profits for the company. Therefore, in an era of vast information, everyone's motivation and demand to find more useful information drives the emergence and development of recommendation systems.

[0003] Today, recommendation systems are widely used in online information services. They solve the problem of information overload by filtering inefficient information and recommending information that users may be interested in. When users are shopping online, e-commerce companies can use recommendation systems to recommend products that users may buy and display corresponding advertisements, thereby facilitating a transaction more quickly. When users are listening to music, service providers need to find music that users are interested in from a massive music library to provide users with an excellent entertainment experience. In these usage scenarios, recommendation algorithms are an important part of recommendation systems. The problem faced by researchers has always been how to make full use of existing data to make recommendations for users.

[0004] The research on recommendation algorithms has been continuously improving with the development of Internet applications and data collection methods. In the early recommendation algorithms for recommending products to users, the system could only recommend the next possible product to the user based on the user's product browsing history data as modeling information. In this process, the system can continuously collect the interaction data between users and projects. Therefore, the collaborative filtering technology using user and project interaction data has developed rapidly at this stage. However, there is a problem that new users cannot get a good recommendation experience due to the lack of project interaction history, which is the cold start problem. At the same time, as the number of projects becomes larger and larger, the sparsity problem of user and project interaction data becomes more and more serious. In the face of these problems, the cold start problem of the system and the sparsity problem of data can be solved by introducing the user's social attribute information, thereby improving the performance of the recommendation system. This type of method is called social recommendation algorithm (SocialRecommendation).

[0005] With the emergence and rise of social media, the structure of user social networks has become more and more complex and diverse, and their social information has become more and more abundant. Therefore, social recommendation algorithms that use the information brought by user social networks to improve the performance of recommendation systems have received more and more research. In social recommendation systems, the reason why users' social networks can provide useful information to the recommendation system in user representation learning is that users' behaviors, interests, and emotions are affected by users with whom they have social relationships. At the same time, in their daily lives, users may also establish social relationships with users who have similar ideas and interests. The above behavior is called the "homogeneity" phenomenon in sociology, and this has been explained by relevant sociological theories. Based on this fact, introducing users' social networks into recommendation algorithms can obtain more accurate user representations, thereby improving the performance of recommendation systems. Therefore, the research focus of social recommendation systems is how to make full use of users' social network information for recommendation.

[0006] In social networks, there is a distinction between social individuals and social groups. A social individual refers to each individual user in a social network; a social group refers to a collection of users who are connected through social relationships because of some common attributes or purposes. According to the theory of social influence, the influence under social networks can be divided into four types: the influence between social individuals, the influence of social groups on social individuals, the influence of social individuals on social groups, and the influence between social groups. Since in the social recommendation algorithm, the main research purpose is to use the user's social network information to better represent the user's portrait. It is worth noting that the influence of social groups on users is holistic and multifaceted, and will affect the user's behavior, interests and emotions, etc., but this influence cannot be explained by the influence between social individuals. With the development of social media and the emergence of social e-commerce, the present invention has important practical significance and huge application prospects. Summary of the invention

[0007] In view of the above problems, the present invention provides a social recommendation method and system based on social group influence, which obtains a more accurate user representation vector by mining the social group information in the user's social network and the influence of the group on the user, thereby achieving the purpose of improving the performance of the recommendation system.

[0008] A first aspect of the present invention is a social recommendation method based on social group influence, comprising the following steps:

[0009] The user-item interaction graph is modeled by considering different ratings of users on items, and the user preference representation vector and item representation vector in the user-item interaction graph are obtained using a graph attention network with edges.

[0010] Model the influence of social individuals on users in the user-user social graph, and use the graph attention network to obtain the user representation vector under the influence of social individuals;

[0011] Modeling the impact of social groups on users in the user-user social graph mainly includes three parts:

[0012] Random walk layer: Use the random walk algorithm to generate multiple user sequences for each user as candidate social groups, and use the average pooling operator to calculate the representation vector of the candidate social group;

[0013] Memory network layer: Use memory components to store the main social group information in the candidate social groups;

[0014] Attention layer: Use the attention mechanism to calculate the representation vector of the user under the influence of the main social groups;

[0015] The gating mechanism is used to select the influence of social individuals and social groups on users in the social network, and the multi-layer perceptron is used to fuse the representation vector of the user's main social group and the preference representation vector to obtain the user's final representation vector;

[0016] The user's final representation vector and the item's representation vector are input into the prediction module to obtain the user behavior prediction result.

[0017] Furthermore, the method also includes using the objective function to train the model established by the entire method.

[0018] Furthermore, the user-item interaction graph is modeled by considering different ratings of users on items, and the edge-based graph attention network is used to obtain the user u in the user-item interaction graph. i The preference representation vector and project v j The representation vector The specific expression is:

[0019]

[0020]

[0021]

[0022]

[0023]

[0024] Among them, N_v(i) represents user u i The set of items that have interacted, Nu(j) is the number of items that have interacted with item v j The set of users who have interacted with each other, W (l)is the linear transformation weight parameter of the node information in the graph neural network layer, is the weighted linear transformation matrix for calculating the attention coefficient of the l-th layer graph, softmax i and softmax j Yes According to user u i and project v j Perform normalization operation of softmax function, σ is the activation function, represents the user's preference representation vector in the l-th layer of the graph neural network, is the representation vector of the item in the l-th layer of the graph neural network, r ij Ratings given to items by users.

[0025] Furthermore, the influence of social groups on users is modeled in the user-user social graph. The specific steps include:

[0026] In the random walk layer, multiple user sequences are generated for each user As a candidate social group, the average pooling operator is used to calculate the representation vector of the candidate social group:

[0027]

[0028] in Represents user u i Item preference representation vector;

[0029] In the memory network layer, the memory component is used to store the main social group information in the candidate social groups. The specific expression is:

[0030]

[0031]

[0032] Among them, k j is the main social group in the memory network layer, v j is the main social group representation vector, softmax ik For social groups The score is operated by softmax function;

[0033] At the attention layer, the attention mechanism is used to calculate the representation vector of the user under the influence of the main social groups. The specific expression is:

[0034]

[0035]

[0036]

[0037] where v T and W are learnable parameters in the attention mechanism calculation.

[0038] Furthermore, the gating mechanism is used to select the influence of social individuals on users in the social network. and the impact of social groups on users And use multi-layer perceptron to fuse the representation vectors of the user's main social groups and preference representation vector Get the user's final representation vector h i , the specific expression is:

[0039]

[0040]

[0041]

[0042] Among them, W h and W g They are and The weight parameter for linear transformation, σ is the activation function, W P，S is the weight parameter and b is the bias.

[0043] Furthermore, the user's final representation vector and the item's representation vector are input into the prediction module to obtain the user behavior prediction result. The specific expression is:

[0044]

[0045] Among them, r′ ij is the predicted user's rating of the item, h i is the user’s final representation vector, z j is the representation vector of the item.

[0046] A second aspect of the present invention provides a social recommendation system based on social group influence, comprising:

[0047] The user-item interaction graph modeling module is used to model the user-item interaction graph by considering different ratings of users on items, and use the edge-based graph attention network to obtain the user's preference representation vector and the item representation vector in the user-item interaction graph;

[0048] The module for modeling the influence of social individuals on users is used to model the influence of social individuals on users in the user-user social graph, and use the graph attention network to obtain the user representation vector under the influence of social individuals;

[0049] The module for modeling the impact of social groups on users is used to model the impact of social groups on users in the user-user social graph. It mainly includes three parts:

[0050] Random walk layer: Use the random walk algorithm to generate multiple user sequences for each user as candidate social groups, and use the average pooling operator to calculate the representation vector of the candidate social group;

[0051] Memory network layer: Use memory components to store the main social group information in the candidate social groups;

[0052] Attention layer: Use the attention mechanism to calculate the representation vector of the user under the influence of the main social groups;

[0053] The user information fusion module is used to select the influence of social individuals and social groups on users in the social network by using the gating mechanism, and use the multi-layer perceptron to fuse the representation vector of the user's main social group and the preference representation vector to obtain the user's final representation vector;

[0054] The prediction module is used to input the user's final representation vector and the item's representation vector into the prediction module to obtain the user behavior prediction result.

[0055] Furthermore, the system also includes: a training module, which is used to train the model established by the entire method using the objective function.

[0056] According to a third aspect of the present invention, a social recommendation system based on the influence of social groups is provided, comprising: a processor; and a memory, wherein a computer executable program is stored in the memory, and when the computer executable program is executed by the processor, the above-mentioned social recommendation method based on the influence of social groups is executed.

[0057] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which instructions are stored. When the instructions are executed by a processor, the processor executes the above-mentioned social recommendation method based on social group influence.

[0058] The present invention provides a social recommendation method and system based on social group influence, and proposes a social recommendation algorithm (SGIMN) based on social group influence of memory network, which can make full use of social network information for user modeling. For the user's social group, a restarted random walk is used to obtain the user's surrounding social subgraph as a candidate social group. At the same time, a memory network is introduced to further process the social subgraph obtained in the previous step to obtain the main potential social groups in the social network. For the influence of social groups on users, the present invention uses an attention mechanism to calculate its influence on user modeling. Finally, a gating mechanism is used to select user representations under the influence of social individuals and user representations under the influence of social groups to obtain the final user representation vector under the social network. The present invention obtains a more accurate user representation vector by mining the social group information in the user's social network and the influence of the social group on the user, thereby achieving the purpose of improving the performance of the recommendation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flow chart of a social recommendation method based on social group influence in an embodiment of the present invention;

[0060] Figure 2 It is a framework diagram of extracting information of main potential social groups in a social network by a memory network in an embodiment of the present invention;

[0061] Figure 3 is a schematic diagram of the structure of a social recommendation system based on social group influence in an embodiment of the present invention;

[0062] Figure 4 It is an architecture diagram of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to further explain the technical solution of the present invention in detail, this embodiment is implemented on the premise of the technical solution of the present invention, and provides a detailed implementation method and specific steps.

[0064] The present invention provides a social recommendation method and system based on social group influence, and provides the following embodiments:

[0065] Embodiment 1 based on the present invention

[0066] like Figure 1 As shown, it is a flow chart of a social recommendation method based on social group influence according to Embodiment 1 of the present invention, and the specific steps are as follows:

[0067] S1. Model the user-item interaction graph by considering different ratings of users on items, and use the edge-based graph attention network to obtain the user's preference representation vector and the item representation vector in the user-item interaction graph;

[0068] In the specific implementation process, the different ratings of users on items are taken into consideration when modeling the user-item interaction graph, and a graph attention network with edges is used for training to obtain the representation vector of each node in the interaction graph.

[0069] In the step of calculating the attention score between nodes, the edge features, i.e. the score values, are concatenated and introduced into the calculation process, and then input into the information aggregation process of the graph attention network. The specific expression is:

[0070]

[0071]

[0072]

[0073]

[0074]

[0075] Among them, W (l) is the linear transformation weight parameter of the node information in the graph neural network layer, N_v ( i) indicates user u i The set of items that have interacted, Nu(j) is the number of items that have interacted with item v j The set of users who have interacted with each other. and They are user u i and project v j The normalized attention score, and softmax i and softmax j Yes According to user u i and project v j Perform normalization operation of the softmax function. is the weighted linear transformation matrix for calculating the attention coefficient of the lth layer graph, σ is the activation function, and the LeakyBeLU function is used in the algorithm. The final result is the user u of the (l+1)th layer. i Preference expression and project v j Representation

[0076] S2. Model the influence of social individuals on users in the user-user social graph, and use the graph attention network to obtain the user representation vector under the influence of social individuals;

[0077] In the specific implementation process, in order to fully capture the social influence of social networks on users and thus improve the accuracy of the recommendation system, we first model the influence of social individuals on users.

[0078] The graph attention network is used to select those social friends with important influence, and then their information is aggregated to obtain the user representation under the influence of social individuals. The specific expression is:

[0079]

[0080] in, That is, user u i After the item preference representation vector is calculated by the user-item interaction network module, is the linear transformation matrix of the neighbor features of the lth layer, N(i) represents the linear transformation matrix of the neighbor features of user u i A collection of friends with direct social connections, and α i，j Is user u i With user u j The attention score between them measures the influence contribution between social individuals. The specific expression is:

[0081]

[0082]

[0083] in, is the weighted linear transformation matrix in the calculation of the attention coefficient of the l-th layer graph, is a learnable vector parameter, σ is an activation function, and the LeakyReLU function is used in the embodiment.

[0084] S3. Modeling the influence of social groups on users in the user-user social graph mainly includes three parts:

[0085] Random walk layer: Use the random walk algorithm to generate multiple user sequences for each user as candidate social groups, and use the average pooling operator to calculate the representation vector of the candidate social group;

[0086] Memory network layer: Use memory components to store the main social group information in the candidate social groups;

[0087] Attention layer: Use the attention mechanism to calculate the representation vector of the user under the influence of the main social groups;

[0088] In the specific implementation process, in order to fully capture the social impact of social networks on users, it is necessary to model the impact of social groups on users. Since there are no explicit social groups in social networks, this step is divided into a random walk layer, a memory network layer, and an attention layer. The specific introduction is as follows:

[0089] (1) In the random walk layer, multiple user sequences are generated for each user. To obtain the surrounding social subgraph, use the average pooling operator in graph representation learning to calculate user u i The kth social group The representation vector of is:

[0090]

[0091] (2) In the memory network layer, use the memory component to store the main social group information in the social network and recalculate the user u i The representation vector of the kth social group The specific expression is:

[0092]

[0093]

[0094] in, Is user u i The kth social group, k j is the main social group in the memory network module, and v j is its representation vector, softmax ik About social groups The score is calculated by softmax function. Figure 2 It is a framework diagram of the algorithm process.

[0095] (3) At the attention layer, the attention mechanism is used to calculate the influence of different social groups on users. Here, the attention score used by the algorithm is calculated using spliced ​​attention, and finally the representation vector of the user under the influence of the social group is obtained based on the attention score. The specific expression is:

[0096]

[0097]

[0098]

[0099] Among them, v T and W are learnable parameters in the concatenated attention calculation, softmax i For user u i The attention scores of the social groups are subjected to softmax function operation to obtain the normalized β ik .

[0100] S4. Use the gating mechanism to select the influence of social individuals and social groups on the user in the social network, and use the multi-layer perceptron to fuse the representation vector of the user's main social group and the preference representation vector to obtain the user's final representation vector;

[0101] In the specific implementation process, in the user information fusion stage, the user's social information and project preference information are fused to obtain the user's final representation vector. Use the gating mechanism to select the influence of social individuals on users in social networks and the impact of social groups on users And use multi-layer perceptron to fuse user's social information and item preference information. The specific expression is:

[0102]

[0103]

[0104]

[0105] Among them, gate I，G is the coefficient calculated in the gating mechanism, W h and W g For the characteristics and The weight parameter for linear transformation, σ is the activation function; W P，S is the weight parameter, b is the bias, is the user's preference representation vector, is the user’s expression under the influence of social network, h i Represents the vector for the end user.

[0106] S5. Input the final representation vector of the user and the representation vector of the item into the prediction module to obtain the user behavior prediction result.

[0107] In the specific implementation process, the user's representation vector and the item's representation vector are input into the prediction module to obtain the user behavior prediction result. The final user representation vector and the item's representation vector are concatenated and input into the prediction module, which is implemented by a multi-layer perceptron MLP. The calculation process is as follows:

[0108]

[0109] Among them, h i Represents the vector and z for the user j is the representation vector of the item.

[0110] Embodiment 2 of the present invention

[0111] This embodiment is used to execute S6 on the basis of Embodiment 1, and use the objective function to train the model established by the entire method.

[0112] In the specific implementation process, the mean square error of the user's rating of the project in the real scenario and the calculation results is used as the loss function:

[0113]

[0114]

[0115] The overall objective function is the mean square error loss, |R| is the number of ratings, r′ ij is the predicted rating, r ij is the true score, λ is the proportion of the regularization term in the algorithm objective function, and Φ is the trainable parameter in the entire algorithm model.

[0116] Embodiment 3 based on the present invention

[0117] Below, refer to Figure 3 To describe the embodiment 1 and the embodiment 2 of the present disclosure Figure 1 , Figure 2The system corresponding to the method shown is a social recommendation system for social group influence, wherein the system 100 includes a user-item interaction modeling module 101, a social individual influence modeling module 102, a social group influence modeling module 103, a user information fusion module 104, and a prediction module 105, wherein the user-item interaction graph modeling module 101 is used to model the user-item interaction graph by considering different scores of users on the projects, and use the edge-attached graph attention network to obtain the user's preference representation vector and the project representation vector in the user-item interaction graph; the social individual influence modeling module 102 is used to model the influence of social individuals on users in the user-user social graph, and use the graph attention network to obtain the user representation vector under the influence of social individuals; the social group influence modeling module 103 is used to Modeling the influence of social groups on users in the intersection graph mainly includes three parts: random walk layer: using the random walk algorithm to generate multiple user sequences for each user as candidate social groups, and using the average pooling operator to calculate the representation vector of the candidate social group; memory network layer: using the memory component to store the main social group information in the candidate social group; attention layer: using the attention mechanism to calculate the representation vector of the user under the influence of the main social group; user information fusion module 104, used to use the gating mechanism to select the influence of social individuals on the user and the influence of social groups on the user in the social network, and use the multi-layer perceptron to fuse the representation vector of the user's main social group and the preference representation vector to obtain the user's final representation vector; prediction module 105, used to input the user's final representation vector and the representation vector of the project into the prediction module to obtain the user behavior prediction result. In addition to the above five modules, the system 100 also includes a training module 106, which is used to train the model established by the entire method using the objective function.

[0118] The specific working process of the social recommendation system 100 based on social group influence is described in the above-mentioned social recommendation method embodiment 1 and embodiment 2 based on social group influence, and will not be repeated here.

[0119] Embodiment 4 based on the present invention

[0120] The device according to the embodiment of the present invention can also be used by Figure 4 The architecture of the computing device shown is implemented. Figure 4 The architecture of the computing device is shown. Figure 4 As shown, computer system 201, system bus 203, one or more CPUs 204, input / output components 202, memory 205, etc. Memory 205 can store various data or files used for computer processing and / or communication and program instructions including the methods of Embodiments 1-2 executed by the CPU. Figure 4The architecture shown is only exemplary and can be adjusted according to actual needs when implementing different devices. Figure 4 One or more components in .

[0121] Embodiment 5 based on the present invention

[0122] The embodiment of the present invention can also be implemented as a computer-readable storage medium. According to the computer-readable storage medium of embodiment 5, computer-readable instructions are stored. When the computer-readable instructions are executed by the processor, the social recommendation method based on social group influence according to embodiments 1-2 of the present invention described with reference to the above figures can be executed.

[0123] Based on the above embodiments, the present invention provides a social recommendation method and system based on social group influence, and proposes a social recommendation algorithm (SGIMN) based on social group influence of memory network, which can make full use of the information of social network for user modeling. For the user's social group, the restarted random walk is used to obtain the user's surrounding social subgraph as a candidate social group. At the same time, the memory network is introduced to further process the social subgraph obtained in the previous step to obtain the main potential social groups in the social network. For the influence of social groups on users, the present invention uses an attention mechanism to calculate its influence on user modeling. Finally, a gating mechanism is used to select the user representation under the influence of social individuals and the user representation under the influence of social groups to obtain the final user representation vector under the social network. The present invention obtains a more accurate user representation vector by mining the social group information in the user's social network and the influence of the social group on the user, thereby achieving the purpose of improving the performance of the recommendation system.

[0124] In this document, the terms "comprises," "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, such that a step or method that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such step or method.

[0125] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A social recommendation method based on social group influence, characterized in that: The following steps are involved: The user-item interaction graph is modeled by considering different ratings of users on items, and the user preference representation vector and item representation vector in the user-item interaction graph are obtained using a graph attention network with edges. Model the influence of social individuals on users in the user-user social graph, and use the graph attention network to obtain the user representation vector under the influence of social individuals; Modeling the influence of social groups on users in the user-user social graph includes three parts: Random walk layer: Use the random walk algorithm to generate multiple user sequences for each user as candidate social groups, and use the average pooling operator to calculate the representation vector of the candidate social group; Memory network layer: Use memory components to store the main social group information in the candidate social groups; Attention layer: Use the attention mechanism to calculate the representation vector of the user under the influence of the main social groups; The gating mechanism is used to select the influence of social individuals and social groups on users in the social network, and the multi-layer perceptron is used to fuse the representation vector of the user's main social group and the preference representation vector to obtain the user's final representation vector; The user's final representation vector and the item's representation vector are input into the prediction module to obtain the user behavior prediction result.

2. The social recommendation method based on social group influence according to claim 1, characterized in that: The method also includes using the objective function to train the model established by the entire method.

3. The social recommendation method based on social group influence according to claim 1, characterized in that: The user-item interaction graph is modeled by considering different ratings of users on items, and the edge-based graph attention network is used to obtain the user u in the user-item interaction graph. i The preference representation vector and project v j The representation vector The specific expression is: Among them, N_v(i) represents user u i The set of items that have interacted, Nu(j) is the number of items that have interacted with item v j The set of users who have interacted with each other, W (l) is the linear transformation weight parameter of the node information in the graph neural network layer, is the weighted linear transformation matrix for calculating the attention coefficient of the l-th layer graph, softmax i and softmax j Yes According to user u i and project v j Perform normalization operation of softmax function, σ is the activation function, represents the user's preference representation vector in the l-th layer of the graph neural network, is the representation vector of the item in the l-th layer of the graph neural network, r ij Ratings given to items by users.

4. The social recommendation method based on social group influence according to claim 1, characterized in that: Modeling the influence of social groups on users in a user-user social graph involves the following steps: In the random walk layer, multiple user sequences are generated for each user As a candidate social group, the average pooling operator is used to calculate the representation vector of the candidate social group: in Represents user u i Item preference representation vector; In the memory network layer, the memory component is used to store the main social group information in the candidate social groups. The specific expression is: Among them, k j is the main social group in the memory network layer, v j is the main social group representation vector, softmax ik For social groups The score is operated by softmax function; At the attention layer, the attention mechanism is used to calculate the representation vector of the user under the influence of the main social groups. The specific expression is: where v T and W are learnable parameters in the attention mechanism calculation.

5. The social recommendation method based on social group influence according to claim 1, characterized in that: Use the gating mechanism to select the influence of social individuals on users in social networks and the impact of social groups on users And use multi-layer perceptron to fuse the representation vectors of the user's main social groups and preference representation vector Get the user's final representation vector h i , the specific expression is: Among them, W h and W g They are and The weight parameter for linear transformation, σ is the activation function, W P，S is the weight parameter and b is the bias.

6. The social recommendation method based on social group influence according to claim 1, characterized in that: The user's final representation vector and the item's representation vector are input into the prediction module to obtain the user behavior prediction result. The specific expression is: Among them, r′ ij is the predicted user's rating of the item, h i is the user’s final representation vector, z j is the representation vector of the item.

7. A social recommendation system based on social group influence, characterized in that: The system comprises: The user-item interaction graph modeling module is used to model the user-item interaction graph by considering different ratings of users on items, and use the edge-based graph attention network to obtain the user's preference representation vector and the item representation vector in the user-item interaction graph; The module for modeling the influence of social individuals on users is used to model the influence of social individuals on users in the user-user social graph, and use the graph attention network to obtain the user representation vector under the influence of social individuals; The module for modeling the impact of social groups on users is used to model the impact of social groups on users in the user-user social graph. It includes three parts: Random walk layer: Use the random walk algorithm to generate multiple user sequences for each user as candidate social groups, and use the average pooling operator to calculate the representation vector of the candidate social group; Memory network layer: Use memory components to store the main social group information in the candidate social groups; Attention layer: Use the attention mechanism to calculate the representation vector of the user under the influence of the main social groups; The user information fusion module is used to select the influence of social individuals and social groups on users in the social network by using the gating mechanism, and fuse the representation vector of the user's main social group and the preference representation vector by using the multi-layer perceptron to obtain the user's final representation vector; The prediction module is used to input the user's final representation vector and the item's representation vector into the prediction module to obtain the user behavior prediction result.

8. The social recommendation system based on social group influence according to claim 7, characterized in that: The system further comprises: The training module is used to train the model established by the entire method using the objective function.

9. A social recommendation system based on social group influence, characterized in that: include: Processor; and A memory, wherein a computer executable program is stored in the memory, and when the computer executable program is executed by the processor, the social recommendation method based on social group influence according to any one of claims 1 to 6 is executed.

10. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the social recommendation method based on social group influence as described in any one of claims 1 to 6 is implemented.

Images