Real-Time Incident Detection Using Social Media Data

Traditional incident-detection methods face constraints due to limited sensor networks and labor-intensive emergency reporting processes. This research proposes leveraging tweet analysis to identify traffic incidents on highways and arterial roads as a practical, economical option.

The methodology involves collecting tweets via Twitter's REST API in real time, establishing a keyword dictionary indicating traffic incidents, and converting tweets into binary vectors for classification. Geocoding then determines incident locations and categories.

Testing occurred in Pittsburgh and Philadelphia. Results showed that roughly 5% of acquired tweets contained geocodable incident information, with 60–70% contributed by institutional users and the remainder by individuals. Weekend reporting exceeded weekday reporting, with peaks during daytime and rush hours. Individual users concentrated incident reports near city centers, with information density declining outward.