Graph-aware Chained Trip Purpose Inference

Trip purpose is one of the significant attributes for understanding the cause of traveller's activity and travel demand. Conducting better travel demand analysis for planning and management of mobility systems also requires sufficiently understanding trip purpose. However, the explicitly described trip purpose is difficult to observe from recent automated travel data collections such as records of GPS and mobile phones. This study proposes a generalized expression of trip chain by considering the relationship between geographic and travel contexts, named as graph-Aware Chained TrIp purpOse inference Network (ACTION), and employed it to estimate trip purpose. The ACTION consists of two components, graph aggregation and tree aggregation. The ACTION aggregates the geographic contextual information by constructing a geographic adjacency graph, and leverages the interactive information between trip and travelled region by tree construction. We implemented the proposed model on the data obtained by the Person Trip (PT) survey in 2018, Japan, and conducted extensive experiments to uncover the effects of hyper-parameters. Furthermore, we compared the model performance with the most applied baseline. The results reveal that the proposed model has acceptable overall accuracy.