Standard and Dialectal Arabic Text Classification for Sentiment Analysis

Abstract

In social networks, the users tend to express more themselves by sharing publicly their opinions, emotions and sentiments, the benefits of analyzing such data are eminent, however the process of extracting and transforming these raw data can be a very challenging task particularly when the sentiments are expressed in Arabic language. Two main categories of Arabic are massively used in social networks, namely the modern standard Arabic, which is the official language, and the dialectal Arabic, which is itself, subdivided to several categories depending on countries and regions. In this paper, we focus on analyzing Facebook comments that are expressed in modern standard or in Moroccan dialectal Arabic; therefore we put these two language categories under the scope by testing and comparing two approaches. The first one is the classical approach that considers all Arabic text as homogeneous. The second one, that we propose, require a text classification beforehand sentiment classification, based on language categories: the standard and the dialectal Arabic. The idea behind this approach is to adapt the text preprocessing on each language category with more precision. In supervised classification, we have applied two of the most reputed classifiers in sentiment analysis applications, Naive Bayes and SVM. The results of this study are promising since good performance were obtained.