Lexicon Based Extraction of Twitter Sentiments and Comparison of Machine Learning Algorithms for Improved Classification of Polarities

Abstract

Twitter is amongst globally used micro-blogging applications by millions of users on daily basis for sharing their thoughts regarding diverse topics of different occasions as well as their opinions on the hottest trends. Which makes it a rich source for decision making and sentiment analysis. Over the recent years, multiple frameworks and models have been proposed to extract people’s sentiments against a topic, an individual or an organization with to help decision making process and still a lot of work is on the way to get accurate models. Sentiment analysis focuses on the polarity calculation from a tweet/text and classifying them as positive, neutral and negative. The primary focus of this methodology is to address the neutral tweets along with positive and negative tweets and their automated polarity generation. Here we proposed a system which extracts the tweets from Twitter against a profile and after applying preprocessing techniques, calculating sentiments and applying Random Forest, Naïve Bayes, Support Vector Machine, Multinomial Logistic Regression Classifier and XGBoost for prediction of sentiments from the tweets. The highest accuracy achieved with this methodology is 81% accuracy.