A Cost-Effective Sequential Route Recommender System for Taxi Drivers
Information Systems
Liu, Junming; Teng, Mingfei; Chen, Weiwei; Xiong Hui
Published in INFORMS Journal on Computing, September 2023
Imagine a busy city where empty taxis easily find passengers without any hassle. This is becoming possible with a new system that helps taxis choose the best routes to pick up riders.
Professor Junming Liu from the Department of Information Systems and his team propose a system using a smart framework that helps taxis decide which road to take to earn the most money from their next passenger. Looking at how many people need rides and how many taxis are available helps drivers be in the right place at the right time.
What makes this system special? It uses technology to predict where passengers are likely to be. A combination of two models—one that focuses on where people are (called a graph convolution network) and another that looks at trends over time (called a long short-term memory model)—works together to give accurate predictions. This means taxis can know the chances of finding a passenger on different streets.
As drivers get these updates, they can make smart choices about which routes to take. The system quickly calculates the best options, allowing drivers to reach passengers faster. This approach does not only help drivers work more efficiently but also means shorter waiting times for riders.
Tests using GPS data from taxis in Beijing show that this new system works well and is efficient. With this technology, the future of taxi services looks promising, making it easier for everyone to get a ride when they need one.
Professor Liu explains, “During rush hours, taxi drivers must proactively reach the high-demand regions before a large group of taxis get crowded in a long waiting line.” This highlights the necessity of timely decision-making in urban transport. a challenge the new system effectively addresses.
He adds, “Integrating predictive analytics into the optimisation framework, the new system can achieve the balance of demand-and-supply mobility and avoid recommendation overload, a phenomenon due to the repeated recommendations of the same searching route to too many taxi drivers, causing traffic congestion and profit decline.”
This prediction-in-optimisation research method also provides an alternative data-driven approach to traditional mathematical methods for considering potential mutual interdependence between demand and optimisation outputs. It unlocks the potential of data mining techniques in solving optimisation problems, where the input parameters change dynamically with operational decisions.
