Cruising or waiting: A shared recommender system for taxi drivers

Abstract

Recent efforts have been made on mining mobility of taxi trajectories and developing recommender systems for taxi drivers. Existing systems focused on recommending seeking routes to the place with the highest passenger pick-up possibility. They mostly ignore that waiting at nearby taxi stands may also help increase the profit. Furthermore, the recommended results seldom consider potential competitions among drivers and real-time traffic. In this paper, we propose a shared recommender system for taxi drivers by including waiting as one kind of seeking policy. We model a seeking process as a Markov Decision Process, and propose a novel Q-learning algorithm to train the model based on massive trajectory data efficiently. During online recommendation, we update the model using feedbacks from drivers and recommend the optimal seeking policy by taking competitions among drivers and real-time traffic into account. Experimental results show that our system achieves better performance than the state-of-the-art approaches.