Research On Session-Based Recommendation Method Based On Graph Convolutional Network

With the consideration of user experience on various internet platforms,various platforms have improved the experience of anonymous access to services.For example,in Amazon?Taobao?Zhihu platforms,users can obtain information in the platform without logging in.However,when the user completes the current visit and closes the session,the platform will lose the identity of the user.Therefore,every time a new session is added,user preference need to be re-established.In order to quickly establish the user's preference corresponding to the session and provide high-quality recommendation services for them,many works have studied session-based recommendation task,aiming to quickly establish user preference based on the user's short-term interaction records and predict the user's subsequent behavior.Therefore,how to model the collaborative relationship between items in the session and how to generate a more effective session representation has become the focus of solving such problems.This paper will study the session-based recommendation method for these two key issues.The specific research work is as follows:(1)proposes a new session-based recommendation model named as IHGCN to jointly explore the adjacent interactive information and the historical interactive information by analyzing the interaction sequence of items in the session.To learn the item feature from these interactive information,the proposed method utilizes the graph structure to model the association between items,and then leverages the graph convolutional networks to the item features from the item correlation graph.The proposed IHGCN method introduces a fine-grained attention mechanism in the feature dimension level to discover the global preference of users,and then integrates the local preference of users.The final recommendation results are obtained based on the learned item features and user preference features.(2)proposes a session-based recommendation model based on overall sequence modeling named as TemSRL.By exploring the importance of each item to the session representation as a whole,a more effective session representation is learned.To well uncover the importance of each item for the session-based recommendation,the paper jointly explores the position of the item in the history session as well as the relationship between the item and the latest item in the session.The desired session representation is obtained by fusing the information of the items in the history session.(3)designs a music session-based recommendation system based on graph convolutional network.The system provides users with personalized music recommendation services based on the user's interactive music sequence.The background algorithm used by the system combines the two models proposed in this paper,which can not only effectively model the collaborative relationship between items,but also generate more effective session representation,thereby clearly deconstructing user preference in music session sequences.