Research On Graph Neural Network Based Recommendation Model In Social Network Environment

With the rapid development of the Internet,the recommendation system has gradually become an indispensable part of all kinds of online platforms,affecting every aspect of People’s Daily life in a variety of ways.The movie recommendation engine provides the users with the movie recommendation which may be of interest to them in the movie online website.The music recommendation module provides the most likely music recommendation for users.Friend recommendations help users find potential friends in social networks and so on.With the explosive growth of the total amount of network information,efficient and accurate personalized recommendation system has become a research hotspot.How to build a satisfactory recommendation system on the online platform still has some difficulties.Firstly,there are two kinds of entities on the online platform:users and goods.From the perspective of network,the network composed by users and goods is a heterogeneous network.Secondly,the preferences of users change dynamically.Users’ interest is influenced by social networks,and different friends have different influence.Finally,items on an online platform are related to each other,and the feature of of the items may change over time.In response to the three problems above,we propose a graph neural network based recommendation model for online communities.Specifically,the framework of our model is divided into three parts.The first part is user modeling which obtains the user’s dynamic interest from the user’s historical behavior by using LSTM.Then,GAT is used to learn user’s social network representation,and user node embedding is obtained.The second part is item modeling which obtains the feature representation of items from the historical interaction between items and users by using LSTM.Then the GAT is used to learn the representation of the user co-occurrence network and obtain the embedding of items.The third part is rating prediction.The user’s embedding is integrated with the item’s embedding to predict the user’s rating of the item.Finally,we use Douban movie data to test the model,and select different categories of recommendation models to evaluate on the same data set using consistent evaluation criteria.It shows that our model achieves the best performance among all the algorithms.Our proposed model starts from the user social network,item co-occurrence network and user item heterogeneous network,and at the same time takes into account the dynamic change of user preferences.On the theoretical level,it enriches and improves the existing social recommendation system model.On the practical level,our proposed model has been tested by real datasets and the results show that the model can effectively obtain users’dynamic interest,and provides users with relatively satisfactory recommendation content.Therefore,it is highly feasible in actual recommendation scenarios.It can not only provide users with more appropriate recommendations,but also meet the needs of online platform management under the current big data environment.Not only that,but it can also provide support for enterprises to understand user preferences.Generally,it provides a new way of thinking for social recommendation.