Comparative Analysis of Feature Extraction Techniques for Event Detection from News Channels' Facebook Page

Abstract

Event detection from the Social Network sites (SNs) has attracted significant attention of many researchers to understand user perceptions and opinions on certain incidents that have occurred. Facebook is the most famous SNs among internet users to express their opinions, emotions and thoughts. Due to its popularity, many news channels such as BBC have created a Facebook page to allow reader to comment on news reported, which has led to an explosion of user-generated data posted on the Internet. Monitoring and analyzing this rich and continuous user-generated content can yield unprecedentedly valuable information, enabling users and organizations to acquire actionable knowledge. Previously, in the context of text mining research, various feature extraction techniques have been proposed to extract relevant key features that could be used to detect the news posts into corresponding event. However, these techniques are separately tested on different data. Moreover, analyzing large number of news posts over a period of time is a challenging task due to its complex properties and unstructured data. Thus, this paper has proposed a comparative analysis on various types of feature extraction techniques on three different classifiers, namely Support Vector Machine (SVM), Naïve Bayes (NB) and KNearest Neighbor (kNN). The aim of this research is to discover the appropriate feature extraction technique and classifier that could correct detect event and offer optimal accuracy result. This analysis has been tested on three news channels datasets, namely, BBC, Aljazeera, and Al-Arabiya news channels. The experimental results have shown that Chisquare and SVM has proven to be a better extraction and classifier technique compared to other techniques with optimal accuracy of 92.29%, 87.12%, 87.00% have been observed in BBC, Aljazeera, and Al-Arabiya news channels respectively