Research On Anonymous Session Recommendation Algorithm Based On Graph Neural Networks

Session recommendation is the prediction of the user’s next click action based only on the user’s current session interaction information when the user logs in anonymously in the website.To address the problems that current session recommendation methods rely on conversion relationships between items and insufficient information acquisition on the sequence,this paper proposes an enhanced session recommendation method based on the interaction time interval and a session recommendation method based on the sequence enhancement of Bi LSTM.Firstly,to address the problem that the user’s intention in the session is not obvious and the information of inter-item dependence transition is not sufficiently obtained,this paper proposes an interaction time interval enhanced graph neural networks for anonymous session-based recommendation.The information about the user’s browsing time on the items is introduced into the weights of the edges,and graph neural networks and attention mechanisms are used to generate more accurate and effective session-level potential feature vectors that better represent the global overall interest of the user’s session.In addition,using dropout for all node hidden vectors in a session effectively improves the accuracy of recommendations and prevents the generation of overfitting.Secondly,graph neural network ignores the order information between nodes and longdistance dependencies when aggregating information about nodes,and items can have different meanings for users in different scenarios.To address the above problems,this paper proposes a Bi LSTM sequence enhanced graph neural network for anonymous session-based recommendation.Information about the relevant attributes of the items is introduced into the session graph,and all nodes are trained bi-directionally using Bi LSTM.For the case of inconsistent number of nodes in the sequence,the processing of data lengthening is used to obtain more accurate item feature vectors containing contextual scene information.We use attention mechanism to aggregate the sessions so that the model gets more useful information,ignore the effect of irrelevant information on the model effect,and improve the user experience.Finally,the above two models were subjected to multiple sets of experiments on two publicly available datasets using different evaluation metrics,and compared with other baseline models for comparative analysis,ablation experimental analysis,etc.,to verify the effectiveness and rationality of the algorithms in the field of session-based recommendations.