HEAMWalk: Heterogeneous Network Embedding Based on Attribute Combined Multi-view Random Walks

Heterogeneous information network (HIN) has complicated semantics information and structural information. Most current HIN embedding methods utilize meta-path to capture rich information in HIN, which still has some problems. On the one hand, these methods ignore the interaction between diverse semantic views and are unable to fully extract the difference in information contribution to different meta-paths for each node; on the other hand, they don’t fully leverage node attributes to mitigate the limitations of meta-path structural information. In this paper, a novel model named HEAMWalk (Heterogeneous network Embedding method based on Attribute combined Multi-view random Walk) is proposed. HEAMWalk employs multi-view random walk to obtain node sequences. Moreover, a selection strategy is designed to measure the contribution of different semantic views to guide the random walks. Experimental results on three HINs show the effectiveness of HEAMWalk.