《计算机应用研究》|Application Research of Computers

一种基于自注意力机制的组推荐方法

Group recommendation method based on self-attention mechanism

免费全文下载 (已被下载 次)  
获取PDF全文
作者 刘浩翰,任洪润,贺怀清
机构 中国民航大学 计算机科学与技术学院,天津 300300
统计 摘要被查看 次,已被下载
文章编号 1001-3695(2020)12-010-3572-06
DOI 10.19734/j.issn.1001-3695.2019.08.0519
摘要 基于自注意力网络和神经协同过滤模型(neural collaborative filtering,NCF)提出一种基于自注意力机制的组推荐系统模型SAGR(self-attention group recommendation),用于建模用户交互数据以及学习群组潜在偏好的表示。通过在用户级和项目级分别使用自注意力机制,动态调整组中每个用户的权重,解决偏好融合问题从而得到组表示。再通过多层神经网络框架NCF从数据中挖掘组和项目之间的交互,最终完成群组推荐。在CAMRa2011和MovieLens数据集上与同类方法进行对比,实验结果表明SAGR方法能够取得更好的组推荐结果。
关键词 群组推荐; 自注意力机制; 协同过滤; 深度学习; 融合策略
基金项目
本文URL http://www.arocmag.com/article/01-2020-12-010.html
英文标题 Group recommendation method based on self-attention mechanism
作者英文名 Liu Haohan, Ren Hongrun, He Huaiqing
机构英文名 College of Computer Science & Technology,Civil Aviation University of China,Tianjin 300300,China
英文摘要 Based on self-attention network and NCF, this paper proposed a SAGR model based on self-attention mechanism to model user interaction data and learn group potential preference representation. By using the self-attention mechanism at the user level and the item level respectively, SAGR could dynamically adjust the weight of each user in a group and solve the preference fusion problem to obtain the group representation. Then it used the multi-layer neural network framework NCF to mine the interaction between the group and the item from the data, and finally completed the group recommendation. Experimental results on the CAMRa2011 and MovieLens datasets show that the SAGR method performs better than other similar group recommendation methods.
英文关键词 group recommendation; self-attention mechanism; collaborative filtering; deep learning; fusion strategy
参考文献 查看稿件参考文献
 
收稿日期 2019/8/20
修回日期 2019/10/12
页码 3572-3577
中图分类号 TP301.6
文献标志码 A