Representation Learning Based on Influence of Node for Multiplex Network

摘要

The research of social network and mining of multi-source/ multi-view network has gradually been a focus in the field of social network recently. Existing studies on building social network are mainly based on the single source data, instead of the multi-source data. In this paper, we use multiplex network (e.g. multi-relation network) to model multi-source data, then propose a node learning representation of multiplex network. First, we propose a method of extracting node influence of multiplex network. Next, taking account into the influence of node and the random walk in multiplex network, we propose a biased random walk method to learn the embedding of node in multiplex network. Finally, we compare existing state-of-the-art techniques on edge reconstruction accuracy and link prediction in several real-world networks from diverse domains. Experiments on real datasets validate the effectiveness of our network representation method, enrich the quantity of conserving multilayer network structure information, and make the description of the node embedding in multiplex network more accurate.