Research On Session-Based Recommendation Model Based On Long-Short Term Interests

With the continuous development of big data technology,the total amount of information is experiencing explosive growth.In this context,recommendation systems play an important role.They can analyze and mine users’ historical behaviors,interests,and preferences to help people find the most suitable content from a vast amount of information.However,in many practical application scenarios,users typically click on multiple consecutive items within a period of time,making it difficult for traditional recommendation models to capture the contextual information contained in the session.To solve this problem,session-based recommendation systems have emerged.These systems model users’ limited behavioral sequences,identify the current context,and predict the user’s next click based on that context.Currently,the methods for obtaining session representations in session-based recommendation systems are relatively simple.They construct the item as a project graph and use graph neural networks to obtain the representation of the project,which is then summed or averaged to obtain the session representation.This approach loses structural information between sessions and projects,resulting in inaccurate session representations and inaccurate recommendation results.To address these issues,this article proposes a session-based recommendation model called Session-Item Graph based Graph Neural Network(SIG-GNN)that expresses user’s short and long-term interests.This model constructs a Session-Item Graph,and then uses self-attention mechanisms and graph neural networks to obtain high-dimensional vector representations of both projects and sessions,effectively extracting structural information between sessions and projects.Finally,the encoding layer generates users’ short and long-term interests,providing personalized recommendations.The accuracy of recommendation system is closely related to the user’s deep interest,which is composed of the user’s long-term and short-term interests.In order to obtain the user’s deep interest,based on the session-based recommendation system,a contrastive learning task is designed on the basis of SIG-GNN to further separate the user’s long-term and short-term interests,so as to obtain the user’s deep interest.Specifically,the long-term and short-term interest representations constructed by the original interaction sequence are used as pseudo labels,and then pairwise contrastive tasks are designed to supervise the similarity between the interest representation and its corresponding interest proxy.Finally,the long-term and short-term interests are fused by attention mechanism to generate the user’s deep interest,and then recommend to the user.To verify the effectiveness of the proposed model in this paper,we conducted comparative analysis on real datasets and compared it with classical recommendation models through cross-comparison experiments,ablation experiments,and parameter visualization analysis,which all demonstrated the effectiveness of the proposed model.