Wheels Know Why You Travel: Predicting Trip Purpose via a Dual-Attention Graph Embedding Network

Trip purpose - i.e., why people travel - is an important yet challenging research topic in travel behavior analysis. Generally, the key to this problem is understanding the activity semantics from trip contexts. However, most existing methods rely on passengers' sensitive information - e.g., daily travel log or home address from surveys - to achieve accurate results, and could thus be hardly applied in real-life scenarios. In this paper, we aim to predict the passenger's trip purpose in the scenarios of door-to-door ride services (e.g., taxi trips) by only using the vehicle's GPS trajectory on roads, for which "wheels" is used as a metaphor. Specifically, we propose a novel dual-attention graph embedding model based on the vehicle's trajectory and public POI check-in data. Firstly, both data are aggregated to augment the activity semantics of trip contexts, including the spatiotemporal context and POI contexts at the origin and destination, which are important clues. Based on that, graph attention networks and soft-attention are employed to model the dependency of different contexts on the trip purpose, so as to obtain the trip's comprehensive activity semantics for the final prediction. Extensive experiments are conducted based on the large-scale labeled datasets in Beijing. The prediction results show a considerable improvement compared to state-of-the-arts. A case study demonstrates the feasibility of our study.