Machine Learning and Semantic Orientation Ensemble Methods for Egyptian Telecom Tweets Sentiment Analysis

Abstract

The vast amount of data currently available online attracted many parties to analyze sentiments expressed in these data extracting valuable knowledge. Many approaches have been proposed to classify the posted content utilizing a single classifier. However, it has been proven that ensemble learning and combining multiple classifiers may enhance classification performance. The aim of this study is to improve the Egyptian sentiment classification by combining different classification algorithms. First, we investigated the benefit of combining multiple SO classifiers using different subsets from SATALex Egyptian lexicon. Second, we investigated the benefit of combining three classification algorithms; Naïve Bayes, Maximum Entropy and Support Vector Machines, adopted as base-classifiers. The experimental results show that combining classifiers can effectively improve the accuracy of Egyptian dataset sentiment classification. However, building these ensembles require more time for processing than the individual classifiers. The time needed depends on the number of classifiers used and the combination method used to combine these classifiers. Thus, the more classifiers used, the more time needed.