Arabic Text Classification Using Modified Artificial Bee Colony Algorithm for Sentiment Analysis: The Case of Jordanian Dialect

Abstract

AbstractArab customers give their comments and opinions daily, and it increases dramatically through online reviews of products or services from companies, in both Arabic, and its dialects. This text describes the user’s condition or needs for satisfaction or dissatisfaction, and this evaluation is either negative or positive polarity. Based on the need to work on Arabic text sentiment analysis problem, the case of the Jordanian dialect. The main purpose of this paper is to classify text into two classes: negative or positive which may help the business to maintain a report about service or product. The first phase has tools used in natural language processing; the stemming, stop word removal, and tokenization to filtering the text. The second phase, modified the Artificial Bee Colony (ABC) Algorithm, with Upper Confidence Bound (UCB) Algorithm, to promote the exploitation ability for the minimum dimension, to get the minimum number of the optimal feature, then using forward feature selection strategy by four classifiers of machine learning algorithms: (K-Nearest Neighbors (KNN), Support vector machines (SVM), Naïve-Bayes (NB), and Polynomial Neural Networks (PNN). This proposed model has been applied to the Jordanian dialect database, which contains comments from Jordanian telecom company’s customers. Based on the results of sentiment analysis few suggestions can be provided to the products or services to discontinue or drop, or upgrades it. Moreover, the proposed model is applied to the database of the Algerian dialect, which contains long Arabic texts, in order to see the efficiency of the proposed model for short and long texts. Four performance evaluation criteria were used: precision, recall, f1-score, and accuracy. For a future step, in order to build on or use for the classification of Arabic dialects, the experimental results show that the proposed model gives height accuracy up to 99% by applying to the Jordanian dialect, and a 82% by applying to the Algerian dialect.KeywordsNatural language processingText classificationSentiment analysisFeature selectionInspired algorithmsABCUBCKNNSVMPNNNaïve Bayes