Text

theses

Recommender systems aim to provide personalized suggestions to users based on their backgrounds and interests. The suggestions can be made in a variety of application areas, such as movies, music, news, books, and products. Recommender systems are primarily developed for individuals who lack of the sufficient personal experiences or competence to evaluate an overwhelming number of alternatives. Therefore, recommender systems are usually personalized, and face substantial challenges in coping with information overloaded environments. This dissertation focuses on building mobile recommender systems with business effective strategies. Due to the explosive growth of GPS trajectory and urban geographical data, mobile recommender systems have been extensively utilized to offer various types of recommendation services. Indeed, recent efforts have been made to develop mobile recommender systems for taxi drivers based on the analysis of taxi GPS traces. In general, there are three ways to provide such recommendation services. The first is to focus on choosing the fastest driving route from the current location to the destination. The second is to provide a sequence of pick-up points for taxi drivers. The goal of this approach is to allow the taxi driver to find a customer within the shortest driving distance. Finally, the third method attempts to strike a balance between the needs of taxi drivers and passengers. However, in the real world, the income of a taxi driver is strongly related to effective driving hours than to the actual driving distance. To this end, in this dissertation, we aim to address the challenges involved in providing business effective recommendations in mobile environments from both theoretical and practical perspectives. Specifically, we first develop a cost-effective mobile recommender system that is capable of recommending an entire driving route for taxi drivers and helping them to find a passenger with the highest possible net profit. Experiments based on real-world data demonstrate the efficiency and effectiveness of our systems. Moreover, we develop a virtual station waiting strategy which suggests the right waiting time and locations for taxi drivers in a business effective way. Then, we design an enhanced recommender system by combining the virtual waiting and driving route search strategies. In this enhanced system, we provide a joint learning framework to evaluate the potential profits derived from different strategies and find the optimal solution. Also, we exploit a recursive algorithm to efficiently generate optimal driving route recommendations. Meanwhile, we introduce Top-K route recommendations and a dynamic maximum Net Profit strategy to provide better load balance for recommendations happened at the same location. Finally, the experimental results clearly validate the effectiveness of the enhanced recommender system for taxi drivers, and show that our recommender system can help to substantially increase the income of inexperienced taxi drivers.

ETD_8234

Ph.D.

Includes bibliographical references

by Meng Qu

doi:10.7282/T31839M3

ETD doctoral

The author owns the copyright to this work.