Manipulating sentiment analysis challenges in morphological rich languages

Abstract

Sentiment analysis is the extraction of sentiments and emotions expressed in text to adjust the polarity (positive or negative opinions) of a specific statement. This can help in many applications such as to collect feedback about products. There are many methods to perform sentiment analysis for English language, but it’s difficult to apply it for morphologically rich languages, such as Arabic in which information is expressed at the word-level. Some methods translate from Arabic to English in order to manipulate the challenges of Arabic sentiment analysis, which leads to lose the language originality and beauty. In this paper, we developed a complete lexicon of standard Arabic words roots and its classification (positive or negative), and then we applied different classifiers models for sentiment analysis on Arabic language directly to compare between supervised and unsupervised learning. Finally, we introduce a new hybrid sentiment analysis algorithm enhanced to handle neutral sentences. The experiments show that preprocessing and analysis of original Arabic sentences greatly reduces the noise of the text and increases the efficiency. In addition, adapting supervised learning classifiers gives more accurate results which directly proportional to the size of the training corpus.