User Identity Linkage via Graph Convolutional Network Across Location-Based Social Networks

In the past few decades, we have witnessed the flourishing of location-based social networks (LBSNs), where many users tend to create different accounts on multiple platforms to enjoy various services. Benefiting from the large-scale check-in data generated on LBSNs, the task of location-based user identity linkage (UIL) has attracted increasing attention recently. Despite the great contributions made by existing work on location-based UIL, they usually investigate the task with data mining methods, which are hard to extract and utilize the latent features contained by check-in records for more precise user identity linkage. In view of the deficiencies of existing studies, we propose a graph convolutional network (GCN) based model namely GCNUL that consists of a GCN-based encoder, an interaction layer, and a classifier, to fully exploit the spatial features hidden in check-in records. Specifically, the GCN-based encoder aims to exploit the spatial proximity of check-in records and mine user mobility patterns. The interaction layer is developed to capture deep correlations between users' behaviors explicitly. The extensive experiments conducted on two real-world datasets demonstrate that our proposed model GCNUL outperforms the state-of-the-art methods.