Research On Session-Based Recommendation Algorithm With Collaborative Enhancement Of Local And Global Features

Session-based recommendation aims to predict the potential items the user will interact with next time based on the given anonymous sessions.Previous works that apply graph neural networks to integrate a variety of assistance recommendation information have achieved favorable performance,but there are still many challenges.First,existing methods(e.g.,Collaborative Session-based Recommendation Machine(CSRM),Global Context Enhanced Graph Neural Networks for session-based recommendation(GCE-GNN),etc.)fusing global collaborative information are susceptible to noise and their performance is not efficient and stable to some extent.Second,many advanced approaches(e.g.,Star Graph Neural Networks with Highway Networks for session-based recommendation(SGNN-HN),GCE-GNN,etc.)focus only on using long-range dependencies or global context information to enhance recommendation,without considering how they can implement collaborative enhancement.Moreover,it is difficult to achieve precise recommendations based on the user’s interest or purpose,since potential information from recent interactions and historical sessions is ignored when representing the user’s current interest.Third,the usability and security of the recommended method in real-world applications are still important issues.To address the above issues,the main research efforts in this paper are as follows:1.To address the over-smoothing problem caused by noise and repetitive propagation,this research proposes to incorporate global context into multi-task learning for sessionbased recommendation,which is called GCM-SR.GCM-SR takes the global context as side information to assist recommendation and integrates it into the model by a global learning task.This method effectively decouples the global context information and the dependencies in the session,and achieves efficient and stable recommendation performance improvement.2.To realize long-range dependencies and global context information to collaboratively enhance recommendation,this research proposes a method called LDGC-SR,which integrates long-range dependencies and global context information for session-based recommendation.This method employs a normalization and adaptive weight fusion mechanism to address the problems of numerical bias and weight skewing in the fusion process to obtain a more accurate user’s long-term preference.In addition,LDGC-SR employs a global context enhanced short-term memory module to capture user’s current interest more accurately.3.To verify the real-world practicability of the proposed recommendation method,this research first compares LDGC-SR with the existing state-of-the-art recommendation methods in terms of time and space complexity and extends it to streaming recommendation scenarios.Secondly,this research analyzes the application of the recommendation method in online learning course content recommendation,and discusses its extension to education data privacy and security.For the above-mentioned recommendation methods,relevant experiments are conducted on four public datasets,and the experimental results reveal the superiority and effectiveness of the proposed recommendation methods.The research results can provide a theoretical basis and practical application support for the realization of more accurate personalized recommendations.