计算机应用研究2025，Vol.42Issue(9)：2653-2659,7.DOI:10.19734/j.issn.1001-3695.2025.02.0025
基于多序列交互与对比学习的侧信息集成序列推荐模型
Side-information integrated sequential recommendation model based on multi-sequence interaction and contrastive learning
摘要
Abstract
Existing side-information integrated sequence recommendation models suffer from insufficient user representation learning and optimization.To solve this problem,this paper proposed an MICL.Firstly,it introduced a multi-sequence interac-tion attention mechanism to construct deep intra-sequence and inter-sequence associations for item sequences and side-informa-tion sequences.This mechanism captured user preferences from both item and side-information perspectives and generated user representations from two viewpoints.Secondly,this method used a user representation optimization module and a dynamic hard negative sampling strategy to construct positive and negative sample pairs.It employed self-supervised signals to optimize user representations.Finally,it adopted a multi-task dynamic weight adjustment strategy to achieve a dynamic balance between re-commendation and attribute prediction tasks,thus enhancing the model's robustness and generalization ability.The model was tested on four public datasets,such as Beauty,Sports,Toys,and Yelp.Compared to well-performing baseline models,the pro-posed model improves the recall rate of MICL and the normalized discount rate(NDCG)of MICL by 1.63％and 2.35％on average.Experimental results verify the effectiveness of MICL in learning and optimizing user representations.关键词
序列推荐/侧信息/多序列交互注意力/用户表示优化/动态难负采样策略/对比学习/多任务动态权重调整策略Key words
sequence recommendation/side information/multi-sequence interaction attention/user representation optimi-zation/dynamic hard negative sampling strategy/contrastive learning/multi-task dynamic weight adjustment strategy分类
信息技术与安全科学引用本文复制引用
赵伟,孙福振,张文轩,王澳飞,王绍卿..基于多序列交互与对比学习的侧信息集成序列推荐模型[J].计算机应用研究,2025,42(9):2653-2659,7.基金项目
国家自然科学基金资助项目(61841602) （61841602）
山东省自然科学基金资助项目(ZR2020MF147) （ZR2020MF147）
