Detecting spatiotemporal traffic events using geosocial media data

Abstract

Social media platforms enable efficient traffic event detection by allowing users to produce geo-tagged content (e.g., tweets) known as geosocial media data. Geosocial media data improve road safety by providing timely updates for traffic flow and traffic control. Recent studies on traffic event detection with geosocial media data have been focused around keyword-based query approaches, where the event content was inferred by predetermined categories, to retrieve relevant traffic events. Spatiotemporal features associated with traffic-related posts have not been fully investigated. In this study, we filtered irrelevant posts with association rules. A spatiotemporal clustering-based method was then used to retrieve traffic events from these filtered posts, where the content of detected events was automatically inferred with a set of representative terms. For comparison, a typical text classification-based method was also used by classifying the posts filtered from association rules into different categories. By validating the detection results with vehicle travel speed data, we demonstrate that the former outperforms the latter in terms of the number of correctly detected traffic events from one-year of Twitter data in Toronto, Canada. Our proposed approach helps organizations and governments to be aware of when and where traffic events occur by identifying event hotspots and peak periods, which improves both traffic management and urban planning.