南京邮电大学学报（自然科学版）2025，Vol.45Issue(2)：77-85,9.DOI:10.14132/j.cnki.1673-5439.2025.02.009
基于图神经网络和用户长短期偏好的会话推荐
Session-based recommendation using graph neural networks and user long short-term preferences
摘要
Abstract
Since existing session-based recommendation methods usually ignore user long-term prefer-ences and correlations between different items,this paper proposes a session-based recommendation model based on graph neural networks and user long short-term preferences(GNN-LSTUP).First,a global session graph is constructed based on all sessions,and the user long-term preferences are mined by incorporating a graph neural network with correlation encoding and attention mechanisms.Second,the user short-term preferences are captured by constructing a local session graph and utilizing graph neural networks and attention mechanisms.Finally,the user long-term and short-term preferences are fused through sum-pooling operations to accurately predict the user's next interaction behavior.Experiments are conducted on the Diginetica,Tmall and Nowplaying datasets,and the results show that the proposed GNN-LSTUP achieved P@20 of 54.19％,34.68％,23.32％with MRR@20 of 18.94％,16.96％,8.62％on the Diginetica,Tmall,and Nowpaying datasets,respectively,which are superior to those of the exist-ing session-based recommendation models by other researchers.关键词
会话推荐/图神经网络/用户偏好/相关性Key words
session-based recommendation/graph neural network(GNN)/user preferences/relevance分类
计算机与自动化引用本文复制引用
卢官明,柯润宇,卢峻禾,丁佳伟,魏金生..基于图神经网络和用户长短期偏好的会话推荐[J].南京邮电大学学报（自然科学版）,2025,45(2):77-85,9.基金项目
国家自然科学基金(72074038)和南京邮电大学引进人才自然科学研究启动基金(NY223030)资助项目 （72074038）
