The Japanese Society for Artificial Intelligence

User Identity Linkage aims to detect the same individual or entity across different Online Social Networks, which is a crucial step for information diffusion among isolated networks. While many pair-wise user linking methods have been proposed on this important topic, the community information naturally exists in the network is often discarded. In this paper, we proposed a novel embedding-based approach that considers both individual similarity and community similarity by jointly optimize them in a single loss function. Experiments on real dataset obtained from Foursquare and Twitter illustrate that proposed method outperforms other commonly used baselines that only consider the individual similarity.