摘要
针对序列推荐模型对用户长期兴趣建模过程中,并未考虑与侧边信息的深度联系以及常常忽略用户近期内的多次交互行为的问题,提出一种基于时序感知和长短期兴趣融合的序列推荐方法。结合项目的侧边信息,设计全新的虚拟类目的自由路由机制对用户的长期兴趣进行建模,增强模型对用户长期行为的建模能力。考虑用户近期内的多次交互并结合属性预测,提升模型对用户短期行为的建模效果。在3个公开数据集上的实验结果表明,各项评估性能均优于其它序列推荐模型。
In the process of modeling user long-term interests in sequence recommendation models,the deep connection with side information is not considered,and the problem of frequently ignoring multiple interaction behaviors of users in the near future is often addressed.A sequence recommendation method based on temporal perception and fusion of long term and short-term inte-rests was proposed.The side information of the project was combined and a new free routing mechanism for virtual categories was designed to model the long-term interests of users,enhancing the model's ability to model long-term user behavior.Consi-dering multiple recent user interactions and combining attribute prediction,the modeling effects on user short-term behavior were improved.Results of experiments on three public datasets show that all evaluation performances are superior to that of other sequence recommendation models.
作者
侯亚飞
荀亚玲
杨海峰
李砚峰
HOU Ya-fei;XUN Ya-ling;YANG Hai-feng;LI Yan-feng(College of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,China)
出处
《计算机工程与设计》
北大核心
2025年第3期804-811,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(62272336)
山西省自然科学基金项目(202103021224286)。
关键词
推荐系统
序列推荐
自注意力机制
长短期兴趣编码
自由路由
时序感知
项目属性
recommendation system
sequence recommendation
self-attention mechanism
short-term and long-term interests coding
feature routing
temporal perception
project properties
