Twitter Sentiments: Pattern Recognition and Poll Prediction

Abstract

Purpose

Do public opinion and political sentiments expressed on Twitter during election campaign have a meaning and message? Are they inferential, that is, can they be used to estimate the political mood prevailing among the masses? Can they also be used to reliably predict the election outcome? To answer these in the Indian context, the 2014 general election was chosen.

Methodology/approach

Tweets posted on the leading parties during the voting and crucial campaign periods were mined and manual sentiment analysis was performed on them.

Findings

A strong and positive correlation was observed between the political sentiments expressed on Twitter and election results. Further, the Time Periods during which the tweets were mined were found to have a moderating effect on this relationship.

Practical implications

This study showed that the month preceding the voting period was the best to predict the vote share with Twitter data – with 83.9% accuracy.

Social implications

Twitter has become an important public communication tool in India, and as the study results reinstate, it is an ideal research tool to gauge public opinion.

Citation

Barclay, F.P., Pichandy, C., Venkat, A. and Sudhakaran, S. (2016), "Twitter Sentiments: Pattern Recognition and Poll Prediction", Communication and Information Technologies Annual (Studies in Media and Communications, Vol. 11), Emerald Group Publishing Limited, Leeds, pp. 141-167. https://doi.org/10.1108/S2050-206020160000011017