Sentiment Analysis of Twitter Feeds

Abstract

This paper focuses on classifying tweets based on the sentiments expressed in them, with the aim to classify them into three categories: positive, negative and neutral. In particular, we investigate the relevance of using a two-step classifier and negation detection in the space of Twitter Sentiment analysis. An efficient sentiment analyzer is deemed to be a must in the era of big data where preponderance of electronic communication is a major bottleneck. Major difficulties in handling of tweets are, their limited size, and the cryptic style of writing that makes them difficult to comprehend at times. We have used different datasets publicly available online and designed a comprehensive set of pre-processing steps that make the tweets more amenable to Natural Language Processing techniques. Two classifiers are designed based on Naive-Bayes and Maximum Entropy classifiers, and their accuracies are compared on different feature sets. We feel that such classifiers will help business or corporate houses, political parties or analysts etc. to evaluate public sentiments about them and design appropriate policies to address their concerns.