为解决知识图谱推荐方法中存在的忽略用户个人信息,或将用户和项目采用相同注意力机制,致使用户和项目的潜在语义表达不充分的问题,提出了一种知识增强的双注意力机制推荐方法.采用交叉压缩融合单元获取用户个人信息和交互历史的潜在特征,以增强用户特征表示;使用不同注意力机制关注用户和项目的重要邻居,以增强知识图谱中的结构信息和语义信息表示.为了验证方法的有效性,在MovieLens-1M、MovieLens-20M、Book-Crossing和Last.FM这4个数据集上进行实验,并与RippletNet、KGAT、CKAN等6种方法进行对比分析.结果表明,本文方法与RippletNet、KGCN、LKGR等方法相比,受试者工作特征曲线下面积(area under the receiver operator characteristic curve,AUC)性能平均提升了5.34%.
Recommendation method for knowledge-enhanced dual-attention mechanism
The ignorance of users'personal information and the usage of the same attention mechanism for users and items lead to insufficient potential semantic expression of users and items in knowledge graph recommendation methods.To address this problem,a knowledge-enhanced double-attention mechanism recommendation method was proposed.The latent features of users'personal information and interaction history was obtained by a cross-compression fusion unit to enhance user feature representation.Important neighbors of users and items were focused by different attention mechanisms to enhance the structural and semantic information repre-sentation in the knowledge graph.To verify the effectiveness of this method,experiments were conducted on four datasets,includ-ing MovieLens-1M,MovieLens-20M,Book-Crossing,and Last.FM.The results were compared with six other methods,includ-ing RippletNet,KGAT,CKAN,et al.The experimental results show that,compared with RippletNet,KGCN,LKGR,and other methods,the area under the receiver operator characteristic curve(AUC)performance of the method proposed herein is improved averagely of 5.34% .
NETL
NSTL
万方数据
