Station-to-User Transfer Learning: Towards Explainable User Clustering Through Latent Trip Signatures Using Tidal-Regularized Non-Negative Matrix Factorization.

Urban areas provide us with a treasure trove of available data capturing almost every aspect of a population's life. This work focuses on mobility data and how it will help improve our understanding of urban mobility patterns. Readily available and sizable farecard data captures trips in a public transportation network. However, such data typically lacks temporal signatures and as such the task of inferring trip semantics, station function, and user clustering is quite challenging. While existing approaches either focus on station-level or user-level signals only, we propose a Station-to-User (S2U) transfer learning framework, which augments user-level learning with shared temporal patterns learned from station-level signals. Our framework is based on a novel, so-called "Tidal-Regularized Non-negative Matrix Factorization" method, which incorporates a-priori tidal traffic patterns in generic Non-negative Matrix Factorization. To evaluate our model performance, a user clustering stability test based on the classical Rand Index is introduced as a metric to benchmark different unsupervised learning models. Using this metric, quantitative evaluations on three real-world datasets show that S2U outperforms two baselines methods by 7-21%. We also provide a qualitative analysis of the user clustering and station functions for the Washington D.C. metro and show how S2U can support spatiotemporal urban analytics.