Enhanced Classification of Sentiment Analysis of Arabic Reviews

Abstract

Sentiment analysis is the process of mining textual data in order to extract the author’s opinion, typically expressed as a positive, neutral, or negative attitude towards the written text. It is of great interest and has been extensively studied in the English language. However, sentiment analysis in the Arabic language has not received wide attention and most of the research done on Arabic either focuses on introducing new datasets or new sentiment lexicons. In this paper, we introduce a preprocessing suite that includes morphological processing, emoticon extraction, and negation processing to improve the sentiment analysis. Furthermore, we conduct experiments on sentiment analysis of hotel reviews that target two classification tasks: positive/negative and positive/negative/neutral. Our experimental results using various supervised learning algorithms, including deep learning algorithm, demonstrate the effectiveness of the proposed techniques.