Social recommendation method and system based on lightweight graph convolutional network

Abstract

The application discloses a social recommendation method and system based on a lightweight graph convolution network, and belongs to the technical field of social recommendation. The method comprises the following steps: obtaining user historical interaction data, obtaining user initial embedding representation and item initial embedding representation according to the user historical interaction data; and processing the user initial embedding representation and the item initial embedding representation through a trained social recommendation model to generate and output an item recommendation list. The method can jointly capture high-order collaborative semantic signal reflected by a user social domain and a user interest domain, improve the accuracy of social recommendation, introduce a self-attention mechanism to realize focusing on important nodes, and solve the problem that, in the prior art, information in a social domain is not fully learned, node features are ignored, and the final recommendation result is affected.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of social recommendation, in particular to a social recommendation method and system based on a lightweight graph convolution network. BACKGROUND

[0002] The statements in this section merely refer to the background of the present application and do not necessarily constitute the prior art. The development of science and information technology has led to an "explosive" development of data storage, and the proportion of online life in people's daily life is becoming larger and larger, leading to information overload. This makes it difficult for users to effectively select information of interest in such complex data, and also brings great pressure to the processing capacity of computers. In order to alleviate this predicament, the recommendation system aims to help users obtain the information they want from the vast amount of information, and as an auxiliary means to help users achieve information filtering, it has been widely studied. It analyzes the user's historical interaction data, mines the user's potential consumption intention and recommends it to the target user, providing accurate and personalized recommendation services for users.

[0003] Traditional recommendation system algorithms are mainly based on collaborative filtering, which has achieved great success, but still faces the problems of data sparsity and cold start. In order to improve the accuracy of the recommendation algorithm, researchers have mined other effective information to add to the recommendation system, such as knowledge graph, social relationship, context information, etc.

[0004] Among them, due to the increasing number of online social platforms, the social relationship between users has gradually attracted widespread attention. On the other hand, as the homogeneity of social theory assumes, user preferences are easily influenced by their social friends. Users are always happy to accept the recommended items of their friends, and users with similar social relationships often have similar preferences. Therefore, social recommendation attracts a lot of research, which tries to use the properties and characteristic information hidden in the user's social network to better model the user's preference for items and improve the recommendation quality of the recommendation system.

[0005] In fact, there have been many research works dedicated to social recommendation, the core of which is how to effectively integrate user social information to more accurately mine user interest preferences. Most of the early research on social recommendation focuses on using matrix factorization techniques to add the influence of user social neighbors to user embedding representation. However, they all only consider the influence of the first-order local neighbors of each user, ignoring the useful information of high-order neighbors. Therefore, the learning process is difficult to extract enough collaborative signals for user representation.

[0006] Subsequently, graph neural networks are widely used in social recommendation due to their strong ability to capture complex relationships between nodes. Although they are effective, the existing social recommendation methods in the prior art still have two problems that cannot be ignored:

[0007] (1) Most social recommendation methods usually learn high-order collaborative semantic signals in user interest domain and user social domain in a complex way, and fail to well coordinate the fusion between the two domains;

[0008] (2) The information in the social domain (user side node features) is not fully learned, for example, the EGLN method uses an enhanced graph structure to strengthen the node embedding representation, but this learning process is carried out in units of graph structures, although the expression ability of the graph is enhanced, the quality of the node features is ignored, which may directly affect the final recommendation result. SUMMARY

[0009] To solve the problems in the prior art, the present application provides a social recommendation method based on a lightweight graph convolution network, a system, an electronic device and a computer storage medium, a lightweight graph convolution network is trained to optimize node representation and fuse high-order collaborative semantic signals reflected in the two domains; at the same time, a propagation fusion strategy is designed to fully learn user embedding, and a self-attention mechanism is introduced to distinguish the contribution of neighbor nodes. In addition, a simple contrast learning task is designed to promote the perception representation of the intrinsic relationship between user nodes by enhancing the uniformity of the user social domain.

[0010] In a first aspect, the present application provides a social recommendation method based on a lightweight graph convolution network;

[0011] The social recommendation method based on a lightweight graph convolution network comprises:

[0012] Obtaining user historical interaction data, and obtaining user initial embedding representation and item initial embedding representation according to the user historical interaction data;

[0013] Processing the user initial embedding representation and the item initial embedding representation through the trained social recommendation model to generate and output an item recommendation list;

[0014] The processing of the user initial embedding representation and the item initial embedding representation through the trained social recommendation model comprises:

[0015] Performing an embedding propagation operation on the user initial embedding representation and the item initial embedding representation through a lightweight graph convolution network;

[0016] Performing a focusing operation on the neighborhood information of the user initial embedding representation and the item initial embedding representation through a self-attention mechanism to obtain user embedding representation and item embedding representation;

[0017] Obtaining an item recommendation list according to the user embedding representation and the item embedding representation.

[0018] Further, the obtaining the user initial embedding representation and the item initial embedding representation according to the user historical interaction data comprises:

[0019] constructing a user-item interaction graph and a user-user social graph according to the user historical interaction data;

[0020] encoding the user nodes and the item nodes in the user historical interaction data based on a user interest domain and a user social domain, to obtain the user initial embedding representation and the item initial embedding representation.

[0021] Further, the performing embedding propagation operations on the user initial embedding representation and the item initial embedding representation by the lightweight graph convolutional network comprises:

[0022] performing embedding propagation operations on the user initial embedding representation and the item initial embedding representation in the user interest domain according to a propagation rule of the user initial embedding representation and the item initial embedding representation in the user interest domain by the lightweight graph convolutional network;

[0023] performing embedding propagation operations on the user initial embedding representation in the user social domain according to a propagation rule of the user initial embedding representation in the user social domain by the lightweight graph convolutional network.

[0024] Further, the propagation rule of the user initial embedding representation in the user interest domain is

[0025]

[0026] wherein, is the user embedding representation of the Lth layer in the user interest domain obtained through l times of graph convolutional operations, Nu is a set of user initial embedding representations that have interacted with the item node i in the user-item interaction graph, Ni is a set of item initial embedding representations that have interacted with the user node u, is a decay factor;

[0027] the propagation rule of the item initial embedding representation in the user interest domain is

[0028]

[0029] wherein, is the item embedding representation of the Lth layer in the user interest domain obtained through l times of graph convolutional operations, Nu is a set of user initial embedding representations that have interacted with the item node i in the user-item interaction graph, Ni is a set of item initial embedding representations that have interacted with the user node u, is a decay factor;

[0030] the propagation rule of the user initial embedding representation in the user social domain is

[0031]

[0032] wherein, is the user embedding representation of the l-th layer in the user social domain, N u is the neighbor set of the user node u in the user social domain.

[0033] Further, the focusing operation on the neighborhood information of the user initial embedding representation and the item initial embedding representation through the self-attention mechanism comprises:

[0034] The first user embedding representation is obtained by calculating the attention weight between the neighbor nodes of the user initial embedding representation in the user interest domain through the self-attention mechanism.

[0035] The item embedding representation is obtained by calculating the attention weight between the neighbor nodes of the item initial embedding representation in the user interest domain through the self-attention mechanism.

[0036] The second user embedding representation is obtained by calculating the attention weight between the neighbor nodes of the user initial embedding representation in the user social domain through the self-attention mechanism.

[0037] The user embedding representation is obtained by performing an integration operation on the first user embedding representation and the second user embedding representation.

[0038] Further, the obtaining of the item recommendation list according to the user embedding representation and the item embedding representation comprises:

[0039] The user final embedding and the item final embedding are obtained by performing an average operation on the user embedding representation and the item embedding representation.

[0040] The score matrix of the user to the item is generated by performing an inner product operation on the user final embedding and the item final embedding.

[0041] The score matrix of the user to the item is used to obtain the item recommendation list.

[0042] Further, when training the social recommendation model, the BPR loss function and the contrastive learning task loss function are used to jointly optimize the social recommendation model.

[0043] In a second aspect, the present application provides a social recommendation system based on a lightweight graph convolutional network;

[0044] The social recommendation system based on the lightweight graph convolutional network comprises:

[0045] The obtaining module is configured to obtain user historical interaction data, and obtain a user initial embedding representation and an item initial embedding representation according to the user historical interaction data.

[0046] The item recommendation list generation module is configured to: process the user initial embedding representation and the item initial embedding representation through the trained social recommendation model to generate and output an item recommendation list.

[0047] The processing of the user initial embedding representation and the item initial embedding representation through the trained social recommendation model comprises:

[0048] performing an embedding propagation operation on the user initial embedding representation and the item initial embedding representation through a light-weight graph convolution network;

[0049] performing a focusing operation on neighborhood information of the user initial embedding representation and the item initial embedding representation through a self-attention mechanism to obtain a user embedding representation and an item embedding representation;

[0050] obtaining an item recommendation list according to the user embedding representation and the item embedding representation.

[0051] In a third aspect, the present application provides an electronic device;

[0052] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and running on the processor, when the computer instructions are run by the processor, the steps of the above-mentioned social recommendation method based on a light-weight graph convolution network are completed.

[0053] In a fourth aspect, the present application provides a computer readable storage medium;

[0054] A computer readable storage medium is used to store computer instructions, when the computer instructions are executed by a processor, the steps of the above-mentioned social recommendation method based on a light-weight graph convolution network are completed.

[0055] Compared with the prior art, the present application has the following beneficial effects:

[0056] 1. The technical scheme provided by the present application is based on a light-weight graph convolution network for social recommendation, which can jointly capture high-order collaborative semantic signals reflected by user social domains and user interest domains, and improve the accuracy of social recommendation.

[0057] 2. The technical scheme provided by the present application improves the fusion method of user interest and user social information, and introduces a self-attention mechanism in the process of network propagation to realize focusing on important nodes.

[0058] 3. The technical scheme provided by the present application integrates a simple contrastive learning into social recommendation, and uses enhanced self-supervised signals to strengthen user social dependency modeling. BRIEF DESCRIPTION OF DRAWINGS

[0059] The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the application, and are incorporated in and constitute a part of this specification, the embodiments of the application, and their description, are to explain the application without imposing any undue limitations.

[0060] Figure 1 A flowchart of a social recommendation method based on a lightweight graph convolution network provided for an embodiment of the application;

[0061] Figure 2 A network structure diagram of the social recommendation method based on the lightweight graph convolution network provided for an embodiment of the application. DETAILED DESCRIPTION

[0062] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0063] It should be noted that the terms used herein are only intended to describe specific embodiments and are not intended to limit the exemplary embodiments according to the application. As used herein, the singular form is intended to include the plural form unless the context clearly indicates otherwise, and it should also be understood that the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device comprising a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to the process, method, product or device.

[0064] The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0065] Embodiment one

[0066] The social recommendation method in the prior art fails to well fuse the high-order collaborative semantic signals reflected in the user interest domain and the user social domain, ignores the unique self-supervision signals existing in the user social domain, and is difficult to learn high-quality node representations; therefore, the present application provides a social recommendation method based on a lightweight graph convolution network, trains a lightweight graph convolution network to optimize node representations, and fuses the high-order collaborative semantic signals reflected in the two domains; at the same time, a propagation fusion strategy is designed to fully learn user embeddings, and a self-attention mechanism is introduced to distinguish the contributions of neighbor nodes.

[0067] Next, the social recommendation method based on the lightweight graph convolution network disclosed in the present embodiment will be described in detail. Figures 1-2 The social recommendation method based on the lightweight graph convolution network disclosed in the present embodiment will be described in detail. The social recommendation method based on the lightweight graph convolution network includes the following steps:

[0068] S1, acquire user historical interaction data, and acquire user initial embedding representation and item initial embedding representation according to the user historical interaction data.

[0069] Specifically, first, a user-item interaction graph (user interest domain) and a user-user social graph (user social domain) are constructed according to the user historical interaction data; wherein the user historical interaction data includes a user node set and a corresponding item node set.

[0070] Then, the user nodes and the item nodes in the user historical interaction data are encoded based on the user-item interaction graph and the user-user social graph, to acquire the user initial embedding representation and the item initial embedding representation.

[0071] For example, let U = {u1, u2,...u m} and I = {i1, i2,...i n} represent the user node set and the item node set respectively, wherein m = |U|, n = |I|; let G ui = {U, I, E r} represent the user-item interaction graph, wherein E r represents the set of observed user-item interaction edges, and let G uu = {U, E u} represent the user-user social graph, wherein E u represents the edge set of social relationships between users.

[0072] Similar to existing graph models, in this embodiment, one-hot encoding is used as input to convert the item nodes and the user nodes into low-dimensional embedding vectors , which describe the initial representation of a user and an item, wherein d is the size of the embedding dimension; that is, the user initial embedding representation and the item initial embedding representation can be represented by embedding matrices and .

[0073] Each one-hot encoding is projected into a two-dimensional vector through a fully connected layer, as follows:

[0074]

[0075] wherein Dum is a hot feature vector representation of the user embedding, and Din is a hot feature vector representation of the item embedding.

[0076] S2, processing the user initial embedding representation and the item initial embedding representation through the trained social recommendation model to generate and output an item recommendation list; wherein the social recommendation model comprises a lightweight graph convolution network self-attention mechanism and contrast learning connected in turn. Specifically, the following steps are included:

[0077] S201, performing embedding propagation operations on the user initial embedding representation and the item initial embedding representation through a lightweight graph convolution network. Including:

[0078] (1) performing embedding propagation operations on the user initial embedding representation and the item initial embedding representation in the user interest domain according to the propagation rules of the user initial embedding representation and the item initial embedding representation in the user interest domain through the lightweight graph convolution network;

[0079] (2) performing embedding propagation operations on the user initial embedding representation in the user social domain according to the propagation rules of the user initial embedding representation in the user social domain through the lightweight graph convolution network.

[0080] Specifically, in the embodiment, further representation learning is performed based on the lightweight graph convolution network (i.e. removing the nonlinear activation function and feature transformation in the graph convolution operation), and embedding propagation operations are performed on the obtained user initial embedding representation and item initial embedding representation.

[0081] For user u and item i, the propagation rules in the user interest domain are as follows:

[0082]

[0083]

[0084] wherein, is the user embedding representation of the Lth layer obtained through l times of graph convolution operations, is the item embedding representation, Nu represents the set of users who have interacted with the item i in the user-item interaction graph, and Ni represents the set of items that have interacted with the user u, is a decay factor used to normalize the influence of propagation with node degree, wherein the superscript I represents the user interest domain, and the following also represents the user interest domain. The superscript S represents the user social domain. The superscript (l) in the above represents the number of layers of graph convolution, such as l = 1, 2, 3, 4, etc.

[0085] For the user representation aggregation of the user social domain G uu , a method based on a lightweight graph convolution network is also used. For user u, the embedding propagation process in the user social domain is as follows:

[0086]

[0087] wherein, denotes G uu the user embedding representation of the l-th layer in G u is the neighbor set of user u in the social graph G uu .

[0088] S202, performing focusing operation on the neighborhood information of the user initial embedding representation and the item initial embedding representation by using the self-attention mechanism, to obtain the user embedding representation and the item embedding representation;

[0089] Specifically, on the basis of S201, the self-attention mechanism is used to perform effective focusing operation on the field information of the user initial embedding representation and the item embedding representation in the user interest domain, and the user node in the user social domain, to enrich the semantics of the node itself. It includes the following steps:

[0090] (1) calculating the attention weight between the neighbor nodes of the user initial embedding representation in the user interest domain by using the self-attention mechanism, to obtain the first user embedding representation;

[0091] (2) calculating the attention weight between the neighbor nodes of the item initial embedding representation in the user interest domain by using the self-attention mechanism, to obtain the item embedding representation;

[0092] (3) calculating the attention weight between the neighbor nodes of the user initial embedding representation in the user social domain by using the self-attention mechanism, to obtain the second user embedding representation;

[0093] (4) performing integration operation on the first user embedding representation and the second user embedding representation, to obtain the user embedding representation.

[0094] In this embodiment, considering that the interests of the user for different items interacted are not the same, the influence of this intensity cannot be ignored, therefore, the self-attention mechanism is introduced to perform focusing learning, the representative item is selected by calculating the attention weight between the neighbor nodes, to obtain the updated first user embedding representation, which is as follows:

[0095]

[0096]

[0097]

[0098] wherein, K, V are learnable parameters, is the attention weight, φ(·) is the tanh activation function, σ(·) is the softmax activation function, and || represents the splicing operation.

[0099] Similarly, an updated item embedding representation is obtained by using similar operations As shown below:

[0100]

[0101]

[0102]

[0103] where K, V are learnable parameters, is the attention weight, φ(·) is the tanh activation function, σ(·) is the softmax activation function, and || represents the concatenation operation.

[0104] In addition, in the embodiment, the heterogeneity of the strength of the user social relationship is considered, that is, the contributions of different friends should be different when the user interacts with social friends. Similar to the operation of the user-item interaction graph, a self-attention mechanism is also introduced to depict this influence.

[0105]

[0106]

[0107] where, represents the attention weight of the target user u and the source user u'.

[0108] The first user embedding representation and the second user embedding representation contain different aspects of information of the user u, and then an integration operation is performed to generate a fused user embedding representation

[0109]

[0110]

[0111] where,

[0112] represents a learnable transformation matrix, γ(·) is an integration function, φ(·) is a tanh activation function, and P·P2 represents an L2 norm, which plays a role of inhibition to make the value of the user u not too large.

[0113] The integrated user embedding representation is propagated at each layer, which makes the fusion between the user social information and the interest information become more and more close in the convolution process, and is more helpful to capture the high-order personalized preferences of the user.

[0114] S203, obtaining a project recommendation list according to the user embedding representation and the item embedding representation. Including the following steps:

[0115] (1) performing an average operation on the user embedding representation and the item embedding representation to obtain a user final embedding and an item final embedding;

[0116] (2) performing an inner product operation on the user final embedding and the item final embedding to generate a user-item score matrix;

[0117] (3) the user-item score matrix, obtaining a project recommendation list.

[0118] Specifically, after the L-layer graph convolution embedding propagation operation, the embedding representation of the user u and the item i is obtained, that is, The representation obtained at different layers emphasizes the information received from different degrees of neighbors, therefore, in order to avoid over-smoothing phenomenon, here, the most effective average operation is used to combine all embedding propagation layers to generate the final embedding of the user and the item:

[0119]

[0120] Given the representation p u, q i of the user u and the item i, the most widely used inner product operation is used to predict the possibility score of the user u interacting with the item i:

[0121]

[0122] According to the user-item score matrix generated after the inner product, a recommendation list is formulated.

[0123] As an implementation mode, the social recommendation model further includes a contrastive learning framework, the social recommendation model is trained by contrastive learning, so as to further enhance the perception representation of the inherent relationship between the user social domain nodes, and make the explicit semantic representation between the users with social connection more similar. In training the social recommendation model, a simple contrastive learning method is used, random noise is directly added in the user social embedding space to achieve efficient enhancement effect, and the similarity between users is enhanced.

[0124] Formally, given the representation of the user u in the d-dimensional social embedding space Data enhancement is realized by the following operation:

[0125]

[0126] Here, is the added noise value, we control the size of by three constraint conditions, as shown in Figure 2 , wherein is a small constant.

[0127] In the experiment, it is set to Good performance can be achieved. In The enhanced embedding representation is obtained by adding a scalable and random noise value to the original representation and rotating it by two small angles β1, β2 It is worth noting that the noise added to each node in the embedding space is randomly different, and the enhanced representation not only retains most of the information of the original representation, but also has some differences. In each layer, different noise values are added to the current learned nodes to obtain the final perturbed node representation, and then the objective function is expressed as:

[0128]

[0129] Here, τ is the temperature parameter.

[0130] L c is used to update the parameters of the user social domain in reverse.

[0131] In this embodiment, in order to better learn the model parameters, the most widely used BPR loss is used to optimize the model. This loss considers the relative order between observed and unobserved user-item interactions, and it assumes that observed interactions can better reflect user preferences and should be given higher prediction value than those unobserved.

[0132]

[0133] Here, O + ,O - represents the training data set, wherein O + represents the observed interaction, O - represents the unobserved interaction. σ(·) is the sigmoid activation function, Θ is the trainable parameter, and β is the L2 normalization coefficient used to prevent overfitting.

[0134] Finally, in order to improve the recommendation performance of the model, a joint optimization strategy is used to optimize the main loss L rec and the contrastive learning task loss L c , which is used to update the parameters of the user interest domain and the user social domain in reverse, and is expressed as:

[0135] L=L rec +λL c

[0136] Here, λ is a hyperparameter used to control the training size of the contrastive learning task.

[0137] Embodiment two

[0138] The embodiment discloses a social recommendation system based on a lightweight graph convolution network, comprising:

[0139] An acquisition module is configured to acquire user historical interaction data, and acquire a user initial embedding representation and an item initial embedding representation according to the user historical interaction data;

[0140] An item recommendation list generation module is configured to process the user initial embedding representation and the item initial embedding representation through the trained social recommendation model to generate and output an item recommendation list;

[0141] The processing of the user initial embedding representation and the item initial embedding representation through the trained social recommendation model comprises:

[0142] Performing an embedding propagation operation on the user initial embedding representation and the item initial embedding representation through a lightweight graph convolution network;

[0143] Performing a focusing operation on neighborhood information of the user initial embedding representation and the item initial embedding representation through a self-attention mechanism to acquire a user embedding representation and an item embedding representation;

[0144] Acquiring an item recommendation list according to the user embedding representation and the item embedding representation.

[0145] It should be noted that the above acquisition module and item recommendation list generation module correspond to the steps in Embodiment One, and the above modules have the same examples and application scenarios as the corresponding steps, but are not limited to the content disclosed in Embodiment One. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.

[0146] Embodiment Three

[0147] Embodiment Three of the present application provides an electronic device comprising a memory and a processor, and computer instructions stored on the memory and running on the processor. When the computer instructions are executed by the processor, the steps of the above-mentioned social recommendation method based on a lightweight graph convolution network are completed.

[0148] Embodiment Four

[0149] Embodiment Four of the present application provides a computer readable storage medium for storing computer instructions. When the computer instructions are executed by the processor, the steps of the above-mentioned social recommendation method based on a lightweight graph convolution network are completed.

[0150] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0151] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0152] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0153] The above description of the various embodiments can have emphasized certain aspects of the various embodiments, which description is not to be taken to mean that other aspects of the various embodiments are not equally valid and important. Those skilled in the art will readily observe that numerous modifications and alterations of the device and method can be made without departing from the scope of the application.

[0154] The specific embodiments described herein are illustrative and not restrictive. Many variations of the described embodiments will become apparent to those of skill in the art upon reviewing the above description. The scope of the application should, therefore, be determined not with reference to the above description, but instead with reference to the appended claims, along with their full scope of equivalents.

Claims

1. A social recommendation method based on a lightweight graph convolutional network, characterized in that, The method comprises the following steps: obtaining user historical interaction data, and obtaining user initial embedding representation and item initial embedding representation according to the user historical interaction data; processing the user initial embedding representation and the item initial embedding representation through the trained social recommendation model to generate and output an item recommendation list; wherein the processing of the user initial embedding representation and the item initial embedding representation through the trained social recommendation model comprises: performing embedding propagation operation on the user initial embedding representation and the item initial embedding representation through a lightweight graph convolution network, including: performing embedding propagation operation on the user initial embedding representation and the item initial embedding representation in the user interest domain according to the propagation rule of the user initial embedding representation and the item initial embedding representation in the user interest domain through the lightweight graph convolution network; performing embedding propagation operation on the user initial embedding representation in the user social domain according to the propagation rule of the user initial embedding representation in the user social domain through the lightweight graph convolution network; focusing operation is performed on the neighborhood information of the user initial embedding representation and the item initial embedding representation through a self-attention mechanism to obtain user embedding representation and item embedding representation; and an item recommendation list is obtained according to the user embedding representation and the item embedding representation; The user initial embedding represents a propagation rule in a user interest domain as ; wherein, is the user embedding representation of the L-th layer of the user interest domain obtained by l times of graph convolution operation, Nu is the initial embedding representation set of the user who has interacted with the item node i in the user-item interaction graph, Ni is the initial embedding representation set of the item that has interacted with the user node u, is the decay factor; The project initial embedding represents a propagation rule in a user interest domain as ; wherein, is the item embedding representation of the L-th layer in the user interest domain obtained by l times of graph convolution operation, Nu is the initial embedding representation set of the users who interacted with item node i in the user-item interaction graph, Ni is the initial embedding representation set of the items that are interacted by user node u, is the decay factor; The user initial embedding represents propagation rules in a user social domain as ; wherein, is a user embedding representation of a user node u in a user social domain, is a user embedding representation of a user node u in a user social domain, is a set of neighbors of the user node u in the user social domain; denotes a target user and a source user strength of attention weight. 2.The social recommendation method based on lightweight graph convolutional network according to claim 1, characterized in that, the obtaining of the user initial embedding representation and the item initial embedding representation according to the user historical interaction data comprises: constructing a user interest domain and a user social domain according to the user historical interaction data; encoding user nodes and item nodes in the user historical interaction data based on the user interest domain and the user social domain to obtain the user initial embedding representation and the item initial embedding representation. 3.The lightweight graph convolution network-based social recommendation method of claim 1, wherein, the focusing operation performed on the neighborhood information of the user initial embedding representation and the item initial embedding representation through the self-attention mechanism to obtain the user embedding representation and the item embedding representation comprises: calculating the attention weight between neighbor nodes of the user initial embedding representation in the user interest domain through the self-attention mechanism to obtain a first user embedding representation; calculating the attention weight between neighbor nodes of the item initial embedding representation in the user interest domain through the self-attention mechanism to obtain an item embedding representation; calculating the attention weight between neighbor nodes of the user initial embedding representation in the user social domain through the self-attention mechanism to obtain a second user embedding representation; performing integration operation on the first user embedding representation and the second user embedding representation to obtain the user embedding representation. 4.The social recommendation method based on lightweight graph convolutional network according to claim 1, characterized in that, the obtaining of the item recommendation list according to the user embedding representation and the item embedding representation comprises: performing average operation on the user embedding representation and the item embedding representation to obtain user final embedding and item final embedding; performing inner product operation on the user final embedding and the item final embedding to generate a user score matrix of an item; the user score matrix of the item is used to obtain the item recommendation list. 5.The social recommendation method based on lightweight graph convolutional network according to claim 1, characterized in that, When training the social recommendation model, the BPR loss function and the contrastive learning task loss function are used to jointly optimize the social recommendation model.

6. A social recommendation system based on a lightweight graph convolutional network, characterized in that, The method comprises the following steps: an obtaining module configured to obtain user historical interaction data, and obtain user initial embedding representation and item initial embedding representation according to the user historical interaction data; The item recommendation list generation module is configured to: process the user initial embedding representation and the item initial embedding representation through the trained social recommendation model to generate and output an item recommendation list; The processing of the user initial embedding representation and the item initial embedding representation through the trained social recommendation model comprises: performing embedding propagation operations on the user initial embedding representation and the item initial embedding representation through a lightweight graph convolution network, comprising: performing embedding propagation operations on the user initial embedding representation and the item initial embedding representation in the user interest domain according to the propagation rules of the user initial embedding representation and the item initial embedding representation in the user interest domain through the lightweight graph convolution network; performing embedding propagation operations on the user initial embedding representation in the user social domain according to the propagation rules of the user initial embedding representation in the user social domain through the lightweight graph convolution network; performing focusing operations on the neighborhood information of the user initial embedding representation and the item initial embedding representation through a self-attention mechanism to obtain a user embedding representation and an item embedding representation; obtaining an item recommendation list according to the user embedding representation and the item embedding representation; The user initial embedding represents a propagation rule in a user interest domain as ; wherein, is the user embedding representation of the L-th layer of the user interest domain obtained by l times of graph convolution operation, Nu is the initial embedding representation set of the users who have interacted with the item node i in the user-item interaction graph, Ni is the initial embedding representation set of the items that have interacted with the user node u, is the decay factor; The project initial embedding represents a propagation rule in the user interest domain as ; wherein, is the item embedding representation of the L-th layer in the user interest domain obtained by l times of graph convolution operation, Nu is the initial embedding representation set of users who have interacted with item node i in the user-item interaction graph, Ni is the initial embedding representation set of items that have interacted with user node u, is the decay factor; The user initial embedding represents propagation rules in the user social domain as ; wherein, is a user embedding representation of a user in a user social domain, is a set of neighbors of the user node u in the user social domain; represents a target user and a source user strength of attention weight.

7. An electronic device, comprising: The computer program product comprises a memory and a processor, and computer instructions stored in the memory and run on the processor, and when the computer instructions are run by the processor, the steps of the method in any one of claims 1-5 are completed.

8. A computer-readable storage medium, characterized in that, The computer program product is used for storing computer instructions, and when the computer instructions are executed by the processor, the steps of the method in any one of claims 1-5 are completed.

Images