Linking Check-in Data to Users on Location-aware Social Networks

Linking check-in data to their owners can benefit many downstream tasks, such as POI (Point of Interest) recommendation, destination prediction, and route planning, since we can obtain redundant information for each user after linking. Consequently, we formulate and investigate the novel problem CUL (Check-in-User Linking) in this work. Notably, the main difference between CUL and the existing problem TUL (Trajectory-User Linking) is that the trajectories used in TUL are continuous, while the check-in records in CUL are discrete. To tackle the problem CUL effectively, we develop a model entitled CULVAE (Check-in-User Linking via Variational Autoencoder). Firstly, a well-designed grid index is applied to organize the input check-in records. Then, an encoding module is developed to embed a user with corresponding grids. Next, a decoding module is proposed to generate a low-dimensional representation of each user. Finally, a multi-class classifier is proposed to link check-in records to users based on the output of the decoding module. We conduct extensive experiments on four real-world datasets, and the results demonstrate that our proposed model CULVAE performs better than all state-of-art approaches.