Social Media Sentiment Classification for Tunisian Dialect: A Deep Learning Approach.

Social media becomes nowadays a valuable resource for posting and expressing opinions about services or products. Sentiments about a product or a service offered to companies are too valuable in this era than ever before. Indeed, knowing whether their customers are expressing positive, neutral or negative sentiments toward their products can be of a high importance for these companies. Hence, extracting the sentiments from comments (or reviews) in Tunisian dialect and written with Tunisian Arabizi script is a challenge compared to other languages. Although, we are getting better with each passing year, but it is still a long way to go to handle Tunisian dialect. In this paper, we focus on sentiment analysis of Tunisian dialect comments posted on social media. More precisely, we propose an interesting deep learning approach for sentiment classification. Thus, we used three corpora, experimented and evaluated four deep learning models: CNN, LSTM, Bi-LSTM and GRU. Our deep neural networks have achieved good results for Tunisian dialect sentiment classification. Best results are obtained based on GRU model.