ESBLA: An Edge-Significance Based Modified Louvain Algorithm for Event Detection from Tweets

Abstract

Background: Social media is a platform where people shares various information. Twitter is the most popular platform to share information about real-world events. The relevant information for people based on their interest in an event, attracts more and more public attention. The shared information could pertain to either local events (such as natural disasters, protests, or accidents) or global events (such as political events, wars, or international news with high impact). Objective: Event detection via tweets saves a lot of time to read thousands of tweets and informs users about the major events. Methods: This article proposes a graph modeling approach, which automatically extracts important information from tweets concerning local or global events. The proposed approach consists of three major components. The first component selects the candidate terms for an event using a modified Term Frequency-Inverse Document Frequency (TF-IDF) weighting scheme, which is suitable for Twitter dynamics. The second component proposes an event detection algorithm - Edge-Significance Based modified Louvain Algorithm (ESBLA), which clusters the event terms using modified Louvain algorithm. ESBLA improvise Louvain 1 through the utilization of significant edges. The last component detects the location of an event using the content-based technique. Results: The proposed approach is compared with existing approaches for the detection of events from tweets. The experimental results on labeled and real-time data show that the proposed approach is better than the compared approaches. Conclusion: The proposed event detection approach extracts events with a high degree of quality and run-time performance over the existing methods used for the event detection.