Research On Social Recommendation With Graph Attention Mechanism

With the rapid development of the Internet and information technology,a large amount of information quickly floods into the Internet.While abundant information brings convenience to users,it also leads to trouble for people.In the face of mass information,it is difficult for people to find the information they really need,which is the problem of information overload in the information age.The key to prevent information overload is information filtering.Compared with the traditional information filtering technology,the recommendation system is the most effective technical to deal with mass information.It connects users and items,and uses user-item interaction information to discover user's preferences,then provides personalized services for users.The recommendation system can not only help users find the items that they are interested in,but also help merchants to carry out accurate promotion and improve revenue.Therefore,the research of recommendation system is very meaningful.In reality,the number of users and items is very large,but the interaction information between them is extremely limited.Therefore,the traditional recommendation system always faces the problems of data sparsity and cold start.In order to alleviate these problems,the social recommendation system makes use of the theory of social homogeneity,which holds that adjacent users in social network usually have similar preferences.However,the existing social recommendation still has some problems.From the viewpoint of model design,only the first-order social relations of users in social network are considered,or the influence among all users is simply considered to be equivalent.From the perspective of model optimization,the quality of negative sampling will directly affect the convergence speed and accuracy of the model,and the algorithm complexity will be increased if the data set size is too large or additional parameters are introduced.In the view of the problems existing in the social recommendation,the main contributions of this paper are as follows:(1)For the design of the model,we propose a fusion graph attention mechanism of social recommendation model(Graph Attentive Matrix Factorization,GAMF),which considers the user of social networks as node and structure a social graph,and using adaptive learning graph attention mechanism to learn the influence between different users.We seamlessly integrate social network into the matrix factorization approach and effectively alleviate the problem of data sparsity and cold start.(2)From the perspective of model learning,we devise an efficient optimization algorithm based on the all-data.In the training process,our optimization method does not need negative sampling,and optimizes the matrix factorization and the graph network respectively,which not only improves convergence speed and accuracy of the model,but also reduces the complexity of the algorithm.(3)We conduct extensive experiments on four real-world datasets,and the results show the performance of our model is substantially better than state-of-the-art recommendation algorithm in HR and NDCG.This demonstrates the effectiveness of incorporating the graph attention mechanism into social networks.