融合用户兴趣表征与注意力机制的推荐算法  

Recommendation algorithm based on user interest representation and attention mechanism

作　　者：孙静 孙静宇[1] 李璨 魏东 SUN Jing;SUN Jing-yu;LI Can;WEI Dong(College of Software,Taiyuan University of Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学软件学院,山西太原030024

出　　处：《计算机工程与设计》2021年第3期814-821,共8页Computer Engineering and Design

基　　金：山西省科技厅重点研发计划基金项目(201803D31226);山西省研究生教育创新基金项目(2019SY117)。

摘　　要：为利用用户行为挖掘用户的兴趣,提出一种融合用户兴趣表征与注意力机制的推荐算法。利用CVR算法将传统的用户-项目表征转换为用户-兴趣表征;构建一种应用于用户兴趣预测的深度森林模型,引入兴趣簇重要性作为特征选择权重,融合时间注意力机制进行兴趣预测,将用户-兴趣模型结合基于用户的协同过滤算法预测推荐结果。两个数据集上的实验结果表明,该算法能够提高用户兴趣预测准确率,提升推荐效果。To mine user interest tags by user behavior,a recommendation algorithm was proposed,in which user interest representation and attention mechanism were integrated.The concept vector representation algorithm was used to transform user-item representation into user-interest representation.A deep forest model for user interest prediction was constructed,in which the importance of interest cluster was taken as feature selection weight and the time attention mechanism was integrated for interest prediction.The user-interest model was combined with the user-based collaborative filtering recommendation algorithm.Experimental results show that the algorithm can improve the accuracy of user interest prediction and the recommendation effect.

关 键 词：深度森林 向量表征 停留时间 兴趣预测 推荐算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]