Social Recommendation Based On User Influence And Debiased Interaction Graph

With the development of mobile Internet technology,the data on the network has increased significantly,the recommendation system can filter out a large amount of information according to user preferences and provide decision-making help for users,however the recommendation system has always faced problems such as sparsity and cold start.The rise of social networks has generated a large number of social relationships,studies have shown that users tend to establish connections with people with similar preferences and are easy to obtain communication information with users with social relationships.Therefore,the social recommendation combined with user social relations improves the performance of the recommendation system to a certain extent.Most of the existing social recommendations are based on explicit social relations,and the singleness of explicit social relations makes the recommendation algorithm ignore the complexity of the relationship between nodes in the social network,the complexity is manifested in the social bias caused by the high-dimensional sparsity of the data and the loss of trust node information during the propagation process.Aiming at this problem this paper combines the influence of users to find the best trusted users and combines the graph neural network to alleviate the social bias between users,so as to further improve the effectiveness of recommendation.The specific research contents include :(1)Considering the difference of user preferences and the instability of trust propagation,a social recommendation algorithm based on user influence and preference consistency is proposed.The algorithm combines rating information and social information to mine the trust relationship between users from the perspective of preference consistency,and finds a path with the strongest stability of trust propagation with the help of users’ social influence.The experimental results show that this method effectively alleviates the loss of trust node information caused by trust in the propagation process and improves the prediction accuracy of recommendation.(2)Due to the high-dimensional sparsity of the data,the graph neural network has a deviation in learning the relationship between users.Therefore,a social recommendation method based on debiased interaction graph is proposed.This method learns user and item representations by vectorizing debiased rating data.At the same time,the rating threshold and item proportion are set to quantify the trust relationship between users and learn the potential factors of users’ friends in a directed way,which reduces the social bias between users.Experimental results show that this method improves the overall performance of social recommendation.(3)A personalized movie recommendation system is designed and implemented based on user influence and debiased interaction graph.The movie recommendation system analyzes the user’s preferences by mining the complex relationships between users in combination with explicit social relationships,and formulates a recommendation list for users.In addition,the system can not only constantly adjust the user’s preferences according to the number of users,but also provide users with popular movie recommendations.