Comparative Evaluation of Lexicons in Performing Sentiment Analysis

Abstract

Twitter is one of the fastest growing social media platforms which allows users to express themselves in short text messages on a wide range of topics. The amount of text produced allows for the understanding of human behaviour. One of the analysis that can be performed is sentiment analysis. Even though sentiment analysis has been researched for many years, there are still several difficulties in performing it such as in handling internet slangs, abbreviations, and emoticons which is common in social media. This paper investigates the performance of two lexicons which are VADER and TextBlob in performing sentiment analysis on 7,997 tweets. Out of the 7,997 tweets, 300 tweets were then randomly selected and three experts in psychology and human development were asked to classify the tweets manually based on three polarities. From the study, it is found that both lexicons have an acceptable accuracy rate of 79% for VADER and 73% for TextBlob. Considering all of the performance score, VADER emerged as a better lexicon as compared to TextBlob. The result of this study serves to help researches in deciding which lexicon to use in performing sentiment analysis for social media texts including microblogs.