Research Of Recommendation Method Based On Graph Neural Network

With the rapid increase of the amount of information on the Internet,the problem of information overload has become an important factor restricting the development of the network.As an effective means to solve the problem of information overload,personalized recommendation system has been paid more and more attention and research by industry and academia.As an important data structure,graph can represent a group of objects and their complex relationships.With the development of artificial intelligence,graph neural network which can effectively extract the feature representation from graph has been proposed,and remarkable research results have been achieved in the fields of biochemistry,economy and finance.Most of the data in the recommendation system has graph structure in nature.Applying the graph neural network to the recommendation system can more effectively understand the preferences and needs of users from various data.Therefore,the recommendation system based on graph neural network has become an important research direction in the field of recommendation system.Another impact of the sharp increase of the amount of information is that user-item interaction data becomes more and more sparse.Compared with the huge number of items,the number of items interacted by each user is almost negligible,which greatly affects the performance of the recommendation system.In this paper,based on graph neural network technology and combined with multimodal data,the recommendation algorithm is studied to alleviate the problem of data sparsity and enhance the accuracy of personalized recommendation.The main research contents of this paper are as follows:(1)Current recommendation methods based on graph neural network mainly deal with structured data such as user-item interaction graph,but cannot deal with unstructured review text well.Item reviews are a unique way for users to choose to purchase the item.It contains users' evaluations of the rich features of items.Therefore,combining user-item interaction graph with related review text will obtain better recommendation performance.At the same time,most of the recommendation methods that have been proposed simply concatenate the representations from different modalities to make predictions,which cannot take advantage of the information from different modalities.To solve these problems,a Mutual Attention graph neural Network(MAN)is proposed for personalized recommendation.MAN first extracts user/item node representation on user-item interaction graph through node feature extraction module,and extracts user/item review text representation through review feature extraction module.Then a mutual attention module is used to correlate node representation and review text representation,so as to capture the correlation between the node representation and the review text representation during the training process.(2)Current recommendation methods based on graph neural network usually construct user-item interaction graph using all items that users interacted in the past.Then it generates long-term preference representations for users by performing neighbor aggregation on the graph.However,user preferences are dynamic in nature.With the passage of time and some trends become popular,in addition to relatively stable long-term preferences,users may also have some short-term preferences.Short-term preferences may also cause interaction between users and items.To solve the above problems,a graph neural network based on long-term and short-term user preferences(LSGNN)is proposed for personalized recommendation.At the same time,item title information and item description information are taken into consideration to alleviate the problem of data sparsity.Experimental results on three real-world datasets show that the proposed model is better than the state-of-the-art personalized recommendation method.To sum up,this paper focuses on the problems and challenges existing in the current recommendation methods based on graph neural network,puts forward the solutions and ideas,and carries out the corresponding scientific experiments.Some possible directions for the research in the field of recommendation system are put forward.