Semi-supervised Domain Specified Event Extraction from Social Media

Abstract

Social media has quickly become popular as an important means that people, organizations use to spread information of divert events for various purposes, ranging from business intelligence to nation security. However, the language used in Twitter is heavily informal, ungrammatical, short and dynamic. Automatically detecting and categorizing events using streamed data is a difficult task, due to the presence of noise and irrelevant information. Therefore, as an emerging research area, event analysis from social media, Twitter has attracted much attention since 2010 and there are many attempts to detect and categorize events from social media. This paper proposes a framework to identify the events from twitter in a semi-supervised manner for targeted domain in specific location with SVM in combination with the corpus. The experimental results show that the semi-supervised SVM model outperforms a strong state-of-the-art semi-supervised classification model of Logic Regression, Navebays and Decision Tree.