Even though using taxi booking services is one of the major method for users to take a taxi, we argue that the customers may not satisfy with these services especially during high demand periods. Current taxi dispatching systems employed by these booking services mainly perform user-taxi matching based on real-time demand and available taxis. Hence, there are high chances that users fail to book for a taxi during periods of high taxi demand and have to repeat the requests multiple times without any guarantee of waiting time. In this paper, we propose data-driven techniques to effectively identify the taxi that the system should dispatch to a pick up location as requested by a user. Specifically, we build and leverage predictive models of taxi travel time and incorporate them into the dispatching system for realizing a much better user-taxi matching strategy which matches users with taxis that will be available in the near future near the users' locations. The benefits of such a predictive taxi dispatching system includes (1) less irritation for users due to uncertain waiting time and having to make repeated requests, (2) higher taxi utilization and revenue for taxi companies, and (3) last but not least better income for taxi drivers. We evaluate our proposed predictive user-taxi matching algorithm with real taxi data. Experimental results confirm that the predictive electronic dispatching system can efficiently handle highdemand periods while reducing repeated requests from customers.
展开▼
