Research And Implementation Of Session-Based Recommendation With Sequence Auxiliary Information

Nowadays,with the rapid advancement of information technology,the emergence of recommendation systems has solved the problem of people’s difficulty in making choices in the face of massive data.Compared with traditional recommendation systems,session-based recommendation does not require users’ individual identity information,and it satisfies people’s need for privacy protection by mining the features of short-term behavior sequences of anonymous users to make personalized recommendation services for users,thus attracting widespread attention.Although the existing session recommendation methods use techniques such as attention mechanism and graph neural network to make full use of the limited session sequence information and also obtain good recommendation results,the lack of research on the auxiliary information of the sequences limits the performance of session recommendation to a certain extent.Specifically,(1)most of the existing mainstream session-based recommendation algorithms consider only one interaction behavior of users,ignoring the auxiliary role between different interaction behaviors of users,which leads to the inability of the model to distinguish the user intention preferences corresponding to different behaviors;(2)the existing mainstream session recommendation algorithms ignore the important feature of item category,which leads to the inability of the algorithm to grasp the macroscopic interests of users as a whole;(3)Existing methods assume that all adjacent items in a session sequence have the same time interval,but this is often unrealistic,and they ignore the influence of the time interval information between items on users’ interests,leading the model to obtain suboptimal results.In order to solve the above problems,this paper conducts research on session-based recommendation based on sequence-assisted information augmentation,as follows.1.A category-enhanced dual-view contrastive learning for session recommendation(CaDVCL)model is proposed.The model explores the influence of item categories and multiple interaction behaviors on user interests.The model combines category sequence information and item sequence information to learn session representations through an attention mechanism and captures the correlation between different behaviors by maximizing the mutual information of session representations obtained from different behavioral perspectives through contrastive learning,which effectively improves the performance of the session recommendation algorithm.2.A time-aware graph neural network for session-based recommendation(T-SBR)model is proposed.The model investigates the influence of time-related information on user interest,constructs a temporal session graph based on the time interval information of items between sequences,and performs item representation learning through graph convolutional networks.The model fuses temporal information and sequence patterns through a gated attention mechanism and also integrates users’ long-term and short-term behavioral preferences to make session recommendations for users.3.Design and implement a session recommendation application system,which is based on real user-item interaction data and uses the proposed session recommendation model to provide recommendation services for users,and better validate the applicability of the proposed model in real platforms.Through the above study,the performance of the session recommendation is enhanced by making full use of serial auxiliary information such as category,multi-behavior,and time,and the proposed model has a certain degree of improvement in model performance compared to the baseline model on several publicly available datasets.