Session-based Recommendation Incorporated With Micro-behavior Feature Analysis

In recent years,the Internet has developed rapidly,more and more people have begun to use the Internet to obtain information.However,due to the huge amount of information,it is sometimes difficult for users to effectively filter among the massive information.At this time,a recommendation system is required to provide users with choices.As one of the subtasks of sequential recommendation,session-based recommendation has been extensively studied and discussed.The existing session-based recommendation methods mainly have the following problems:First,only the embedding of the item is used to construct the representation of the session,so as to predict the next item that the user will interact with,without incorporating other additional information.Second,there is a lack of embedding of user behavior information,and different operations of users on commodities will also indicate different preferences of users for commodities.In order to solve the above-mentioned problems,this paper makes the following research:(1)In order to construct the item embedding more reasonably,a deep matrix factorization method is adopted for modeling,and a item-item interaction matrix is constructed.The row vector of the matrix is used as the item embedding,and the column vector of the matrix is used as the prediction embedding of the item.(2)A session-based recommendation model with graph neural network and user’s microbehavior(GNNMB)is proposed.At the same time,item embedding and user’s operation embedding are used to construct session representation,self-attention mechanism is used to construct operation embedding,and graph neural network with gate control is used to construct item embedding.The network constructs item embeddings and captures the transition pattern between successive items at a fine-grained level,thereby incorporating more information.(3)Aiming at the question of how to construct item embeddings,the method proposed in(1)is adopted,and a session-based recommendation model with deep matrix factorization and user’s micro-behavior(DMFMB)is proposed.The deep matrix factorization method is used to construct the item embedding,and the operation embedding obtained by using the gated recurrent unit is combined to form a session representation,and the item is scored through MLP to make recommendations.(4)In response to the above problems,this paper has conducted a lot of comparative experiments on real datasets.By comparing with multiple baseline models,we verify the performance effect of the session-based recommendation model fused with micro-behavior feature analysis,and find the best performance effect that the model can obtain by adjusting different parameters.After extensive evaluation and verification of real datasets,it is proved that the session-based recommendation method incorporated with micro-behavior feature analysis proposed in the article has better performances.