EventStory: Event Detection Using Twitter Stream Based on Locality

Abstract

Increased popularity of social media sites such as Twitter, Facebook, Flickr, etc. have produced an enormous amount of spatio-temporal data. One of the application of this type of data is event detection. Most of event detection techniques have focused on temporal feature of data for detecting an event. However, location associated with data has to be taken into consideration to detect locality based event (local event) such as local festival, sporting event or emergency situations. Users in proximity of the location of an event are more likely to post messages about an event compared to users distant from the location of that event. In this paper, we are proposing a framework, called EventStory. Our framework first identifies locally significant key-words (LSK) by monitoring changes in the bursty nature of keywords in both local and global regions. Candidate event clusters are created based on co-occurrence of locally significant keywords (LSK) in the each keyword cluster. A cluster scoring scheme is used which uses the features of cluster to filter irrelevant clusters. A case study is presented to show effectiveness of our approach.