Spatiotemporal Representation Learning for Taxi Pick-up Point Recommendation

Abstract

Taxi services play an important role in the public transportation system of large cities. In this work, we study the problem of pick-up point recommendation to idle taxi drivers. To this end, we propose a novel spatiotemporal representation learning based on graph convolutional networks (GCNs) on taxi trips and POI data. Specifically, we construct a POI interaction graph in each time slice by creating directed edges from end POI of the first trip to start POI of the second trip for each pair of consecutive trips, and model the relationship strength on edges by incorporating various factors related to drivers’ revenue. The representation vectors of POI nodes are then learned via GCN in an unsupervised manner. Next we use cosine similarly of POIs’ representation embeddings to recommend the potential pick-up points for taxi drives. Experiments on the real-world dataset in New York city demonstrate the effectiveness of the proposed recommendation model.