TWCM: Twitter Word Co-occurrence Model for Event Detection

Abstract

Increasing popularity of Twitter data indicates that social media is up-to-date source of information in terms of recent topics, trends or events. Although many existing techniques provide useful insights from social media data, but word co-occurrence structure based analysis is still unexplored. A word co-occurrence network based event detection approach has been proposed in this research work. The proposed technique is independent of static parameters and is an unsupervised technique. The proposed approach use directed and weighted graph for analysis and topological sorting for identifying event phrases. The proposed approach is validated on the First Story Detection standard dataset. The results are compared with the existing techniques namely Frequent Pattern Mining using Dynamic Support Values (FPM-DSV), Soft Frequent Pattern Mining (SFPM), High utility pattern clustering (HUPC) and Transaction-based Rule Change Mining (TRCM). The results show that the proposed technique outperforms FPM-DSV, SFPM, HUPC and TRCM in terms of recall rate and redundancy rate.