LAUSR

Route planning and recommendation have drawn increasing attention recently as traveling is now an essential part of our daily life. However, existing route recommendation researches focus on single transportation mode (i.e., taking taxi or riding bicycle only), which leave recommending routes with multiple transportation modes largely untouched. Inspired by the observation of the first/last-mile congestion problem when traveling with vehicles, in this paper we study a novel route recommendation problem, namely Route Recommendation with Hybrid Transportation Modes (RRHTM), which aims to minimize the travel time for a trip by taking taxis and/or shared bicycles. To address this problem, we propose a two-phase data-driven framework called Predictive Hybrid Route Recommendation (PHRR), which integrates a prediction and recommendation phase. The first phase proposes different learning models (i.e., Deep Neural Network model, and Transfer and Multi-Task learning model based on Deep Neural Network) to predict the travel time by taxi/bicycle for a given trip and the station status in Bicycle Sharing System. In the route recommendation phase, we develop a network-based travel time comparison algorithm to efficiently find the optimal route. Moreover, an evaluation criterion called Empirical Risk of Travel Time (ERTT) is designed to evaluate the reliability of our recommendation results. Finally, we verify the effectiveness of proposed approaches through extensive experiments based on real datasets.