A Hybrid Approach for Sentiment Analysis of Arabic Tweets

Sentiment analysis is an emerging field that focuses on determining positive and negative sentiments, emotions, and measuring their intensity. One application of sentiment analysis is to provide a clearer picture of the general public's opinion on specific topics, such as events, campaigns, services, or specific news, which may be in the form of articles, posts, brief comments, tweets, or even images. While various sentiment analysis tools for English texts have been developed, tools for Arabic text sentiment analysis remain limited and often basic. Not only is a vast amount of content generated online in Arabic, but its strategic importance cannot be overlooked. The aim of this paper is to conduct a hybrid approach to sentiment analysis on Arabic text using the Semeval-2018 Task-1 dataset. The proposed method uses a combination of lexicon-based features, word embedding as well as character embedding as features classified by an ensemble of CNN, DNN, GBT, and bi-LSTM models. Results show a 4% improvement over the Semeval-2018 competition's winner. The knowledge gained from this method has been localized, making it implementable for future applications.