BERT4Traj: Transformer Based Trajectory Reconstruction for Sparse Mobility Data
This addresses the challenge of sparse mobility data for applications in public health, transportation, and urban planning, though it appears incremental as it adapts existing transformer techniques to a specific domain.
The paper tackles the problem of reconstructing complete mobility trajectories from sparse data like GPS or call detail records, proposing BERT4Traj, a transformer-based model that significantly outperforms traditional methods such as Markov Chains and RNNs on real-world datasets from Kampala, Uganda.
Understanding human mobility is essential for applications in public health, transportation, and urban planning. However, mobility data often suffers from sparsity due to limitations in data collection methods, such as infrequent GPS sampling or call detail record (CDR) data that only capture locations during communication events. To address this challenge, we propose BERT4Traj, a transformer based model that reconstructs complete mobility trajectories by predicting hidden visits in sparse movement sequences. Inspired by BERT's masked language modeling objective and self_attention mechanisms, BERT4Traj leverages spatial embeddings, temporal embeddings, and contextual background features such as demographics and anchor points. We evaluate BERT4Traj on real world CDR and GPS datasets collected in Kampala, Uganda, demonstrating that our approach significantly outperforms traditional models such as Markov Chains, KNN, RNNs, and LSTMs. Our results show that BERT4Traj effectively reconstructs detailed and continuous mobility trajectories, enhancing insights into human movement patterns.