Twitter Sentiment Analysis: Lexicon Method, Machine Learning Method and Their Combination

This paper presents a step-by-step methodology for Twitter sentiment analysis. Two approaches are tested to measure variations in the public opinion about retail brands. The first, a lexicon-based method, uses a dictionary of words with assigned to them semantic scores to calculate a final polarity of a tweet, and incorporates part of speech tagging. The second, machine learning approach, tackles the problem as a text classification task employing two supervised classifiers Naive Bayes and Support Vector Machines. We show that combining the lexicon and machine learning approaches by using a lexicon score as a one of the features in Naive Bayes and SVM classifications improves the accuracy of classification by 5%.