An Improved Approach for Sentiment Analysis of Arabic Tweets in Twitter Social Media

Abstract

Recently, sentiment analysis of social media contents is very important for opinion mining in several applications and different fields. Arabic sentiment analysis is one of the more complicated sentiment analysis tools of social media due to the informal noisy contents and the rich morphology of Arabic language. There is a number of works has been proposed for Arabic sentiment analysis. However, these works need an improvement in terms of effectiveness and accuracy. Consequently, in this paper, a corpus-based approach is proposed for Arabic sentiment analysis of tweets annotated as either negative or positive in twitter social media. The approach is based on a Discriminative multinomial naive Bayes (DMNB) method with N-grams tokenizer, stemming, and term frequency-inverse document frequency (TF-IDF) techniques. The experiments are conducted using a set of performance evaluation metrics on a public twitter dataset to test the proposed sentiment analysis approach. Experimental results demonstrated the usefulness of the proposed approach. Furthermore, the comparison results showed that the approach outperformed the related work and improved the accuracy with 0.3%.