A Shared Recommender System For Taxi Drivers

Nowadays,spatiotemporal trajectory data is being produced in large numbers.Many taxis with positioning devices can record their positions and report them to servers.Large amounts of data generate taxi trajectories,but how to use these trajectories on recommender systems becomes a problem.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.The main contributions of this paper are summarized as follows:· We develop a shared recommender system for taxi drivers that suggests a series of passenger seeking actions in real time with the objective of maximizing the expected profit.We propose a Q-learning approach to the problem efficiently. To the best of our knowledge,we are the first to include waiting as one kind of seeking policy.· We consider potential competitions among drivers and real-time traffic during online recommendation.We produce seeking actions based on a weighted round robin algorithm to avoid recommending routes run into traffic jams.· We conduct extensive experiments on real-world trajectories.Results show that our approach makes higher profit than the state-of-the-art solutions.