SAMLink: a Mobility Signature Augmentation Model for Trajectory-User Linking
Chao-Xiong Chen,Wanyi Zhang,Bo Yu,Chao Chen
DOI: https://doi.org/10.1007/s00521-023-09049-z
2023-01-01
Neural Computing and Applications
Abstract:Trajectory-user linking (TUL) aims to link trajectories to users who generate them, based on the historical trajectories of a set of users (e.g., social media users or drivers). It is a fundamental task for many downstream applications of trajectory data. However, there is a lack of a general method to solve the TUL problem for different types of trajectory data. Existing studies usually focus on specific trajectory data, such as check-in data. Unlike subjectively self-selected check-in data, objectively recorded trajectory data is more prevalent and voluminous in real life. To address the issue, we propose a mobility signature augmentation model named SAMLink to solve the TUL problem for different types of trajectory data. It contains three components: feature embedding, mobility signature learning, and linking. Specifically, there are three modules in the key component of mobility signature learning, i.e., local mobility signature learning, attention-based signature augmentation, and adaptive fusion. The first module encodes the input trajectory to learn the local mobility signature, while the second module generates a global mobility signature from the historical trajectories using an attention mechanism. Furthermore, the last module adaptively fuses local and global mobility signatures to yield better identification results. We evaluate our proposed model on three real-world trajectory datasets, i.e., the check-in dataset, the TCD dataset, and the ERI taxi dataset. Our results demonstrate the effectiveness and generalization of our proposed model for solving the TUL problem across different types of trajectory data.