Intelligent Subevent Detection Based on Social Network Data

Event Detection and Tracking has been a big challenge with the growth and diversification of data sources, especially in large urban centers. As components of a complex event, subevents have not received many attempts of detection by applications and studies. This work uses a human as a sensor approach via social network data to automatically detect subevents within events reported by media. These kind of data usually consists of short texts with colloquial language which provides an additional challenge for automated applications. To overcome these challenges, a scalable topic modeling based algorithm is proposed to identify and label the subevents to ease the analysis by humans. Evaluation is made in Brazil's 2013 political protests scenario, comparing subevents detected with news published by mainstream media regarding the same subjects. Results show that even with the particular data extracted from social media, the majority of subevents detected corresponded to real subevents published in official and traditional media. These results can open new applications in event tracking and emergence scenarios as social media usually has a minimum delay between an event and a report.