A Sentiment Analysis Approach based on Arabic Social Media Platforms

Abstract

Apart from the major outstanding research issues facing Arabic social media sentiment analysis which includes handling of vernacular Arabic, slang vocabulary and shorthand writings. There is also a lack of comprehensive framework for Arabic social media sentiment analysis as existing works often focus on particular platforms (like twitter and Facebook). As such, models developed on one platform often perform poorly on other platforms due to lack of a representative feature space. To this regard we adopted a comprehensive approach utilizing a broad array of Arabic social media platforms to establish more generalized sentiment models using random subspace ensembles of MLP base learners. More importantly, we introduced a new sentiment classification scale and we classified sentiments as Highly Positive (HP), Fairly Positive (FP), No Sentiment (NS), Fairly Negative (FN) and Highly Negative (HN). The approach has been tested in a series of experiments and the results demonstrate significant improvements in terms of both classification accuracy and generalizing ability.