SENTIMENT ANALYSIS OF LIBYAN TWEETS USING MACHINE LEARNING ALGORITHMS

Abstract

Sentiment analysis is a highly active field of study in natural language processing, also known as opinion mining. Social media is a communication tool between internet users. Thus, these platforms become valuable data resources that can be exploited and used efficiently to support decision-making. Many researchers are still working on improving the processing of sentiment analysis in textual data (positive or negative comments). Although there are several studies in Arabic dialects using the machine learning approach, no prior work has been conducted on the Libyan dialect. In this paper, a training dataset of Libyan tweets for sentiment analysis is used to train three machine-learning algorithms. The aim is to determine which algorithm has the best accuracy for our dataset. We use Cohen’s Kappa measure to evaluate the quality and measure the reliability of the sentiment annotations which observed agreement is 89.1%. The experiments of the three algorithms showed that the decision tree algorithm achieved better results compared to the other algorithms in terms of accuracy. The results are (72%, 69% and 65%) in (Decision Tree, Support Vector Machine and Naive Bayes) respectively.