Characterization of traffic events using social media

Abstract

Accidents, malfunctioning matrix signs, oil on the road and a bridge that is not able to close are examples of traffic events that happen every day on the Dutch roads. In the Netherlands we are able to measure traffic speeds and calculate travel times on highways and most important secondary roads using a network of connected traffic measuring points. The data that is collected using these points is available for anyone interested and distributed every minute. This distribution frequency makes it possible to detect near real time traffic disturbances. However, the traffic data does not provide information about what is actually happening on the road. During rush-hour or particular city events traffic disturbances are expected, but there also exist many disturbances of the unexpected kind: accidents and car or truck breakdowns for example. In this study we focus on the characterization of traffic events by using Twitter as an information source. Using open traffic data as a traffic event provider we link tweets to traffic events using different linkage strategies in order to extract traffic information. Related traffic tweets can than be used to extract cause types and enrich traffic events. We developed a demonstrating system for the Netherlands that is able to extract traffic cause types using two different Dutch Twitter datasets. The system uses a set of detected traffic events as its input.