A Method for User Identification Across Multiple Transportation Systems Using Trajectories

This paper investigates a method for identifying users based on trajectories collected by different public transportation systems, where a user has different IDs in different systems. Identifying users is crucial for completing their trajectories and optimizing or recommending travel routes. The problem is similar to calculating the similarity between two trajectories from two data sources. Existing trajectory similarity methods based on trajectory point alignment or travel pattern similarity primarily focus on scenarios where trajectories are spatiotemporally synchronous. These methods are not applicable to here, as the trajectories collected by different public transport systems for same user are asynchronous. To address the issue, this paper proposes a user identification method (GLDT) for the scenario based on global semantic rationality and local contributions of co-occurrence points (e.g transfer point). Specifically, the method first filters out impossible pairs of trajectories by considering the uniqueness of a user in certain spatiotemporal contexts from a global perspective. Then it calculates the similarity of two trajectories by aggregating the novel defined local similarity of each co-occurrence point, which considers multi-view attributes of a co-occurrence point helpful to uniquely identify a user (e.g., individual and group visit frequency at a location, rationality of transfer routes, and so on). We validated our method using real-world data from Shenzhen, China, the results show that the matching accuracy improves by at least 10% compared to existing methods.