融合协同知识图谱的群组推荐方法
Group recommendation method based on collaborative knowledge graph
张宇星 1刘学军 1王步美2
作者信息
- 1. 南京工业大学计算机科学与技术学院,江苏南京 211816
- 2. 江苏省特种设备安全监督检验研究院直属分院安全评价室,江苏南京 210002
- 折叠
摘要
群组推荐是当前推荐系统研究领域的热点问题之一,针对现有方法忽略了对群组本身的偏好特征的学习以及辅助信息利用不充分的问题,提出一个融合协同知识图谱的群组推荐方法.构建协同知识图谱,使用TransH对图谱上的节点进行预训练,使用关系感知的图注意力网络迭代聚合领域信息,获得增强的用户和项 目表示;使用注意力模块和自注意力模块分别获取群组成员融合偏好和群组基础偏好,经过门控网络获得最终的群组表示;基于群组表示和项 目表示得到预测评分进行Top-K推荐.通过在多个数据集上进行实验分析,验证了该方法的有效性.
Abstract
The group recommendation is one of the focused areas in recommendation system research.To address the problem that existing methods ignore the learning of basic preference features of the group itself and insufficient use of auxiliary informa-tion,a group recommendation method based on collaborative knowledge graph was proposed.The collaborative knowledge graph was constructed by combining user interaction records and knowledge graph,the nodes'representation on the knowledge graph was pretrained,and relation-aware graph attention network on the knowledge graph was used to iteratively aggregate domain information to obtain enhanced user and item representations.The group member fusion preferences and group base preferences were obtained using the attention module and the self-attention module,respectively.The final group representation was obtained through the gating network.The prediction scores were obtained based on the group representation and item represen-tation for Top-K recommendation.The effectiveness of the proposed method was demonstrated through experimental analysis on several datasets.
关键词
推荐系统/知识图谱/注意力机制/图注意力网络/门控机制/自注意力机制/深度学习Key words
recommendation system/knowledge graph/attention mechanism/graph attention network/gating mechanism/self attention mechanism/deep learning引用本文复制引用
基金项目
国家重点研发计划(2018YFC0808500)
出版年
2024
NETL
NSTL
维普
万方数据