会议专题

Representation Learning for Heterogeneous network with Multiple Link Attributes

  The strategies of learning node representation in network,which try to reduce the dimension of adjacency matrix in network,have been drawn much attention recently.Existing approaches focus on preserving structure property and heterogeneity in a low dimen-sional space with simplex link.However,in the real world,link between each object containing rich attributes information will lead to treat links as multi-variables.In this paper,we split net-work into multiple aspects according to multiple link attributes and preserve structural and semantic properties for each aspect simulta-neously.Then,we introduce an attention strategy,which is capable of combining each aspect with relevant weight,to extract more informative aspect for nodes.Results of experiment verified with link prediction task on a real world dataset show that our approach is able to capture the most contributory attribute of link,which outperforms the state of the art network representation learning techniques regarding each attribute equally.

Heterogeneous Network Representation Learning Multi-Aspect Link Attributes

Kaiwen Song Xinao Wang Yidan Zhang Jie Zuo

Sichuan University Chengdu,China

国际会议

2019国图灵大会(ACM Turing Celebration conference-China 2019 )

成都

英文

41-45

2019-05-17(万方平台首次上网日期,不代表论文的发表时间)