Score Prediction Of High-order Social Influence And Interest Dissemination In Social Commerce

The Internet is developing rapidly and penetrating into people’s daily lives,users can obtain information very conveniently with the help of smart devices such as mobile phones and computers,the cost of information acquisition is getting lower and lower and data is increasing exponentially.In the face of the vast amount of information,it is becoming increasingly important for users to be able to find the information they want in the shortest possible time.Information filtering has become an indispensable part of people’s daily lives,and recommendation system is one of the most effective methods of information filtering.The core idea of the recommender system is to learn accurate user and item embedding vector representations for user historical behavior data to better reflect user preferences.Although,traditional collaborative filtering-based recommendation models propose a method for learning user and item embeddings from user-item history interaction data.However,the model performance is limited by the lack of sufficient data support due to the sparse historical user behavior data compared to the number of users and items.With the development of online social networks,the model considers user proximity information as auxiliary data,incorporating historical user rating data and social information to model user characteristics.In real-world scenarios,users are influenced in their decision making by their friends in addition to their own interest preferences,which in turn are influenced by the preferences of their trusted friends.Social influence propagates and spreads recursively through multiple layers of social relationships,and users’ interests change during this diffusion process,and the items that users interact with are not independent of each other and are somewhat related.However,current social recommendation model considers the user’s social network as static and only considers the user’s local social network,using the information propagation between the user and its first-order immediate neighbors to model the information of the user’s first-order social immediate neighbors,often ignoring the deep recursive propagation of information in the social network,and the user’s firstorder social relationships are also sparse,resulting in poor recommendation performance of the model.As users are at the core of multiple relationship networks,they are critical in both social networks and interest interaction networks,and each user’s potential item interest is not only reflected in the items that the user has interactively rated,but also influenced by similar users’ interest in the items.Therefore,this paper proposes a propagation model(GraphSI)that integrates higher-order social influence and potential synergistic interest information to model users’ higher-order social relationships and potential synergistic interests under a designed unified framework for better recommendation.Specifically,1)The multiple relationship networks between users and items are constructed into a heterogeneous graph containing user social networks,interest networks and item association networks,capturing the nearest neighbor influence in the user-user social network graph and mining potential synergistic interests in the user-item interaction network for input into the user embedding learning process.2)Users are better represented by iteratively aggregating the embeddedness of each user from three aspects:initial user embeddedness,high-level social relations,and user interest propagation and aggregation.3)According to the user-item interaction information,the item-item association network is built to capture the association information between items.The item embedding is updated iteratively from three aspects:the initial embedding of the item,the dissemination of users’ deep interests,and the aggregation of item association information to better represent the item.A simplified graph convolutional neural network is used to perform node feature propagation in the heterogeneous graph,where node features are recursively propagated through iterative convolutional aggregation from neighboring nodes to update the node embedding representation.Three sub-graph networks are involved in the constructed heterogeneous graphs,where different users have different biases towards social and interests,by setting up a multi-layer attention network to learn the influence of neighboring nodes in the graph on the weight of the target node as well as to distinguish the importance of different networks on the user’s decision.Under the same dataset,comparison experiments are conducted with models based on collaborative filtering ideas and recommendations based on deep learning techniques respectively to analyze the influence of higher-order social influence and interest propagation as well as item association information on recommendations,and the effectiveness of the proposed model is demonstrated through experiments.