KEvent – A Semantic-Enriched Graph-Based Approach Capitalizing Bursty Keyphrases for Event Detection in OSN

Abstract

Social networks are growing quickly, and they have soon taken over as the main global source of breaking news. As a result, these platforms provide a plethora of user-generated content, which has inspired researchers to delve into and interpret data for a variety of objectives. Due to its effectiveness in locating news items hidden inside enormous amounts of voluminous data, event detection in online social network data has recently grown in prominence. In this research, we introduce KEvent, a novel graph-based technique for event detection from Twitter messages (aka tweets). The suggested method divides tweets into bins for extracting bursty keyphrases and then uses post-processing techniques to create a weighted keyphrase graph using the Word2Vec model. The keyphrase graph is then subjected to Markov clustering for the purpose of clustering and event detection. KEvent is evaluated over the Events2012 benchmark dataset, and it performs noticeably better when compared to two state-of-the-art techniques, Twevent and SEDTWik. Additionally, KEvent has the ability to find events that the aforementioned state-of-the-art techniques were unable to find.