A Two-Step Method for Sentiment Analysis of Tweets

Abstract

Sentiment Mining refers to extraction of public sentiment, typically from a set of texts or documents. The present work aims at gaining insights into sentiment mining on tweets using Machine Learning and Natural Language Processing techniques. We propose a lexicon-based sentiment scoring method, and use it in combination with Support Vector Machine (SVM) classifier as a two-step method for sentiment classification. The proposed scheme creates a sentiment lexicon from Sentiment140 corpus using Point-wise Mutual Information (PMI) measure. The lexicon is used to score tweets with positive or negative polarity. Tweets with low polarity strength are passed to a Support Vector Machine (SVM) classifier. This system is compared with several supervised statistical learning algorithms. The classification performance of the proposed scheme has been found to be better than commonly used one-step methods. We also discuss efficacy of several linguistic features, such as Part-of-Speech (POS) tags and higher order n-grams in sentiment mining.