A data-driven approach to automatic tweet generation about traffic incidents

Traffic congestion continues to be a major problem in large cities around the world and a source of frustration for drivers. Previous studies show that providing drivers with real-time traffic information will help them make better route planning and avoid congestion. In this research, we examine the use of data-driven natural language generation (NLG) techniques to automatically generate tweets from traffic incident data. From the task of automatic tweet generation, we discuss and propose a design of a traffic notification system that can deliver personalized and location-relevant real-time traffic information to drivers. The domain of our NLG work is novel with respect to the previous work in different domains including weather forecasts, educational reports and clinical reports. We evaluate the automatic generated tweets using BLEU-4. Our experimental results show that a well-prepared training corpus is important for better quality output, however, it is currently limited in traffic-related domains.