Ensemble Learning Sentiment Classification for Un-labeled Arabic Text

Abstract

Sentiment classification has become one of the most trending research topics, due to the rapid growth of social media platforms and applications. It is the process of determining the opinion or the feeling of a piece of text and assigning a label to it (positive, negative or neutral). One of the issues in sentiment classification is the need for labeled data – that is often carried out manually - in order to train the classifiers which is a time consuming task. In this paper we consider the lexicon-based classification as labeling technique instead of the manual labeling. In addition, for an effective sentiment classification we investigate the using of multiple ensemble learning methods - where multiple classifiers are combined - in order to improve the performance of the classification. Experiments have been run on datasets of reviews written in Modern Standard Arabic. Results show that the labeling technique is effective and promising and the use of ensemble learning has clearly improved the accuracy for the sentiment classification compared to the traditional methods.