Research On Fresh Product Recommendation Based On Graph Neural Network

At present,online shopping of goods has gradually become a mainstream,and fresh goods,as essential materials in life,are undergoing such changes.There are many kinds of commodities in various fresh food e-commerce platforms,and the information is complex,and it is often difficult for users to make effective choices.The introduction of recommendation systems can effectively alleviate such problems.The recommendation system mainly analyzes the user’s historical behavior and recommends the products that may be of interest to the user,thereby improving the user experience.Graph neural networks have become a hotspot in recommender system research because they can learn node embedding representations through the propagation of node information on the graph.However,the existing recommendation methods based on graph neural network mainly consider the dissemination of information between users and item nodes,less consideration is given to the dissemination of information between items and item nodes.Considering the information dissemination between similar fresh commodities,this dissertation proposes a graph neural network recommendation model incorporating fresh commodity association graphs.The model extracts the labels that can represent the characteristics of fresh commodities from the review text to construct the fresh commodity association graph,and extracts the node features on the user-fresh commodity interaction graph and fresh commodity association graph to predict the user’s response to the fresh commodity.The preference of fresh products to improve the recommendation effect.The specific work of this dissertation is as follows:(1)In view of the fact that the traditional recommendation model seldom considers the information dissemination between similar commodities,a commodity association graph reflecting the similarity of fresh commodities is constructed.The dissertation first uses Text Rank to extract keywords from users’ comments on fresh products,then vectorizes the keywords through the Word2 vec algorithm,and finally calculates the similarity between keywords to generate labels for fresh products.On this basis,a fresh commodity-fresh commodity association graph is constructed according to the co-occurrence of labels.(2)Aiming at the problem that the traditional recommendation model cannot handle the graph structure data in the non-European space well,a graph neural network recommendation model integrated with the fresh commodity association graph is proposed.The model uses the graph attention network to aggregate and update the feature information of nodes in the user-fresh product interaction graph and the fresh product association graph.And use the random walk method to sample high-order neighbors on the fresh commodity association graph,and then use the GRU unit to capture the information of the high-order neighbors to provide a more comprehensive feature representation for the product.Finally,experiments are carried out on the real data set crawled by the JD e-commerce platform,which verifies the advantages of a graph neural network recommendation model integrated with fresh commodity association graph proposed in this dissertation in fresh commodity recommendation.