Heterogeneous Network Representation Learning Guided by Community Information

Network representation learning usually aims to learn low-dimensional vector representations for nodes in a network. However, most existing methods often ignore community information of networks. Community structure is an important topology feature in complex networks. Nodes belonging to a community are more densely connected and tend to share more common attributes. Preserving community structure of network during network representation learning has positive effects on learning results. This paper proposes a community-enhanced heterogeneous network representation learning algorithm. It introduces the community information of a heterogeneous network into its node representation learning, so that the learned results can maintain both the properties of the micro-structure and the community structure. The experiment results show that our algorithm can greatly improve the quality of heterogeneous network representation learning.