Classifying Sentiment of Dialectal Arabic Reviews: A Semi-Supervised Approach

Arab Internet users tend to use dialectical words to express how they feel about products, services, and places. Although, dialects in Arabic derived from the formal Arabic language, it differs in several aspects. In general, Arabic sentiment analysis recently attracted lots of researchers' attention. A considerable amount of research has been conducted in Modern Standard Arabic (MSA), but little work has focused on dialectal Arabic. The presence of the dialect in the Arabic texts made Arabic sentiment analysis is a challenging issue, due to it usually does not follow specific rules in writing or speaking system. In this paper, we implement a semi-supervised approach for sentiment polarity classification of dialectal reviews with the presence of Modern Standard Arabic (MSA). We combined dialectal sentiment lexicon with four classifying learning algorithm to perform the polarity classification, namely Support Vector Machines (SVM), Naive Bayes (NB), Random Forest, and K-Nearest Neighbor (K-NN). To select the features with which the classifiers can perform the best, we used three feature evaluation methods, namely, Correlation-based Feature Selection, Principal Components Analysis, and SVM Feature Evaluation. In the experiment, we applied the approach to a data set which was manually collected. The experimental results show that the approach yielded the highest classification accuracy using SVM algorithm with 92.3 %.