关键词: Hadoop, 轨迹数据, 推荐算法, 乘客推荐服务

Abstract: In order to improve the efficiency of passenger recommendation algorithm for??wisdom city, this paper uses classical probability to count the percentage of the empty taxi passing days in the total days to recommend the probability of passengers waiting for empty taxi, and uses least squares to fit arrival rate curve, namely, the relation curve between time and the number of empty taxis to predict the time passengers have to wait to take an empty taxi from the time that they reach the right road. To improve the efficiency of recommendation, the paper has done three more work：Select Hadoop as data storage and computing platform to improve the ability of data processing; put forward a new road network storage structure based on rasterizing maps to improve the search speed of the map; reform a map matching algorithm which based on computational geometry to improve the matching accuracy. Field testing experiments show that, the correct probability of empty taxi recommendation algorithm accuracy can reach 87% while the accuracy of waiting time recommendation algorithm is about  88.4 %, which shows the feasibility of mining trajectory data to provide recommendation service for passengers.

Key words: Hadoop, trajectory data, recommendation algorithm, recommendation service for passengers