Towards Robust Trajectory Representations: Isolating Environmental Confounders with Causal Learning

• DOI: 10.24963/ijcai.2024/248
• Publication Date: 2024-07-26T14:28:11Z
• Authors: Kang Luo; Yuanshao Zhu; Wei Chen; Kun Wang; Zhengyang Zhou; Sijie Ruan; Yuxuan Liang
• Keywords: Causal structure

Trajectory modeling refers to characterizing human movement behavior, serving as a pivotal step in understanding mobility patterns. Nevertheless, existing studies typically ignore the confounding effects of geospatial context, leading acquisition spurious correlations and limited generalization capabilities. To bridge this gap, we initially formulate Structural Causal Model (SCM) decipher trajectory representation learning process from causal perspective. Building upon SCM, further present framework (TrajCL) based on Learning, which leverages backdoor adjustment theory an intervention tool eliminate between context trajectories. Extensive experiments two real-world datasets verify that TrajCL markedly enhances performance classification tasks while showcasing superior interpretability.