A Comparative Study of Feature Selection Methods for Informal Arabic

Abstract

The advent of web 2.0 and new Big Data technologies has created a diversity of data and information that can be used in many fields of application. The case of opinion mining is of increasing interest to researchers because of its impact on policy, marketing, etc. Through this document, we are interested in the study of sentiments more specifically in informal Arabic. We present a new approach of processing and analysis that is improved through feature selection methods. The experiments we have carried out are based on the comparison of 3 feature selection methods combined with several machine learning algorithms applied on a twitter dataset. Our paper reports the enhanced results (Accuracy of 98%) and shows the importance of feature selection for Arabic Sentiment Analysis.