Research And Implementation Of Time Interval-aware Session-based Recommendation System Based On GNN

As a common information retrieval tool,recommendation system can help people obtain useful information from large-scale information.In today’s big data era,for the sake of user privacy security,users’ personal information and other data cannot be accessed in many real environments,and only the user behavior data of the current session is available.Therefore,as a new paradigm of recommendation system,conversational recommendation has important research value.At present,session-based recommendation is mainly based on deep learning technology.How to fully explore the implicit representation of items and how to more accurately learn the transfer information between items are the main difficulties and research points in the task of conversational recommendation.At present,methods based on recurrent neural network and attention mechanism can not effectively use item transfer information,and the recommendation effect is also poor.Therefore,this research studies the session recommendation algorithm based on graph neural network and the session recommendation algorithm combined with self-attention mechanism,and explores the above problems.The main research contents and achievements of this paper are as follows:(1)A session-based recommendation algorithm Ti-GNN based on graph neural network with time interval awareness is proposed,which utilizes a novel time interval-aware embedding to improve the recommendation performance.Ti-GNN also incorporates sequential and transition information to enrich session representation.Different from previous session graph-based methods,we introduce a master node for each graph to explicitly model the long-range dependencies among behaviors.Finally,some classic session-based recommendation algorithms are used as baselines to verify the effectiveness of our model on three benchmark datasets.(2)Since the same item may appear in different sessions,these sessions will have some relevance.For the information between sessions,self-supervised learning is introduced as an auxiliary task,and a new session recommendation algorithm S2-TiGNN is proposed.The algorithm constructs local graph and global graph,creates self-supervised signals and conducts comparative learning.In order to verify the effectiveness of the improvement,experiments were conducted on a variety of indicators,and the experimental results proved the superiority of S2-TiGNN.(3)A visualization platform of recommendation system based on graph neural network is implemented.Based on the above research,this paper uses B/S network architecture to develop front-end pages and backend systems for the recommendation algorithm.Users can enter interaction sequences in the platform,or directly click to generate session sequences,and finally get the recommendation results of the recommendation algorithm.