摘要
现有的基于异构信息网络的推荐方法主要通过节点相似性挖掘推荐辅助信息,受元路径线性结构及对可见路径依赖的影响,用户和项目特征并不能被充分捕获。提出基于异构信息网络表征学习的推荐方法,通过在给定元结构上进行截断随机游走学习用户和项目节点的低维向量表示,并将其直接融入推荐样本,结合FFM模型进行评分预测。实验表明,该方法有较高性能。
The existing recommendation methods based on heterogeneous information network mainly mine the recommendation auxiliary information through the similarity of nodes.However,due to the influence of the linear structure of the meta-path and the dependence on the visible path,the characteristics of users and items cannot be fully captured.A recommendation method based on heterogeneous information net⁃work representation learning is proposed.After learning the low-dimensional vector representation of user and item nodes through truncat⁃ed random walk on the given multiple meta-structures,it directly integrates them into the recommendation samples,and combines them with FFM model for scoring prediction.Experiments show that this method has higher performance.
作者
李亚莹
LI Ya-ying(College of Computer Science,Sichuan University,Chengdu 610065)
出处
《现代计算机》
2020年第4期7-10,共4页
Modern Computer
关键词
异构信息网络
网络表征学习
推荐
Heterogeneous Information Network
Network Representation Learning
Recommendation