Characterization of Disaster Related Tweets According to Its Urgency

Abstract

With the pervasive utilization of social media platforms such as microblogging site eg. Twitter to express and share information, it has also become a very useful and helpful tool in times of evolving crisis situations. Extracting interesting and meaningful patterns in the context of disaster is very helpful to determine relationships among them. Association rule mining aims to discover frequent patterns, relationships among set of items in the database. This paper describes text mining technique for extracting association rules from disaster related tweets. It sought to characterize disaster related tweets in terms of urgency, whether they are carrying information that requires immediate attention or not. They were harvested from Twitter using Rapid Miner and from existing collection of tweets. We employed association rules using the FP-Growth algorithm for discovering significant and interesting patterns on disaster related tweets for the emergency responders to determine behavior of users in times of mass emergencies. We used support and confidence as statistical measure to observe the usefulness of the association rules. Based on the result, we discover meaningful patterns of urgent tweets, however, for the not urgent Twitter posts, we only discovered 4 interesting patterns.