Social Event Extraction: Task, Challenges and Techniques

Hao Li, Heng Ji, Lin Zhao

Abstract: Social media (e.g., Facebook and Twitter) serves as a popular platform for online communication and information dissemination, where users can effectively share information such as their recent activities and plans. This kind of information is extremely valuable for building recommendation systems. For example, a user might wish to receive alerts whenever a concert takes place near his current location or when a party will be held in his neighborhood. However, such events may not be widespread across social networks, they have not received sufficient attention. Moreover, traditional event discovery and event extraction techniques trained from formal genres cannot be effectively adapted to this domain. In this paper, we present the first formal definition of social events, discuss the annotation challenges and release a benchmark for the research community. Further more, we propose two novel solutions for extracting elements from social events: (1) an unsupervised content segmentation framework to extract event phrases (2) utilize external knowledge bases to detect fine-grained event locations and unveil their background information. Experimental results convincingly demonstrate that our approach can accurately extract social events from social media.