超图
利用
计算机科学
成对比较
聚类分析
吸引力
同质性(统计学)
对偶(语法数字)
社会关系
关系(数据库)
理论计算机科学
社交网络(社会语言学)
数据挖掘
人工智能
机器学习
社会化媒体
数学
万维网
心理学
社会心理学
计算机安全
文学类
离散数学
精神分析
艺术
作者
Jiadi Han,Tao Qian,Youxi Tang,Yuhan Xia
标识
DOI:10.1145/3477495.3531828
摘要
Social relations are often used as auxiliary information to improve recommendations. In the real-world, social relations among users are complex and diverse. However, most existing recommendation methods assume only single social relation (i.e., exploit pairwise relations to mine user preferences), ignoring the impact of multifaceted social relations on user preferences (i.e., high order complexity of user relations). Moreover, an observing fact is that similar items always have similar attractiveness when exposed to users, indicating a potential connection among the static attributes of items. Here, we advocate modeling the dual homogeneity from social relations and item connections by hypergraph convolution networks, named DH-HGCN, to obtain high-order correlations among users and items. Specifically, we use sentiment analysis to extract comment relation and use the k-means clustering to construct item-item correlations, and we then optimize those heterogeneous graphs in a unified framework. Extensive experiments on two real-world datasets demonstrate the effectiveness of our model.
科研通智能强力驱动
Strongly Powered by AbleSci AI