Use of Sentiment Analysis for Predicting Public Opinion on Referendum: A Feasibility Study

ABSTRACT This study explored the potential of using sentiment analysis of tweets to predict referendum choices (Brexit). The feasibility of using StreamKM++ in the massive online analysis framework was examined over five categories, ranging from strongly agree to strongly disagree (to exit). A Naïve Bayes classifier was used to classify people’s opinions according to these categories. The prediction model resulted in high accuracy (97.98%), making it possible to use it in predicting opinions about public events and issues. The findings from this study may help practitioners, and policymakers understand the importance of sentiment analysis of social media in assessing public opinion and, accordingly, making certain voting predictions.