Keywords: Forecasting, Natural Language Processing, Sentiment Analysis, Social Media, Elections, Opinion Polls.
Abstract: In Zambia, like many other countries, opinion polls have been used to predict the outcome of elections since 1999. During the run up to the 2021 general elections, two opinion polls were conducted. One poll suggested that HH would emerge victorious whilst the other predicted that ECL would emerge victorious. The variance in the two opinion polls leaves room for alternative approaches to predicting election results. This study proposes sentiment analysis as part of the initial stage to building an alternative solution to predicting the outcome of an election. The study analysed sentiments shared on social media during the build up to the August 2021 general elections. The findings of the study reveal that as the election day drew closer, there was an exponential increase in the number of tweets that were posted on a daily basis. Further, our analysis of the tweets revealed that the majority of the tweets were neither positive nor negative (they were neutral) in line with the Afrobarometer opinion poll. Topic modelling was subsequently also performed on the tweets using BERTopic. Some of the topics learnt include voter engagement, the shutdown of the internet and the election day. Initial findings are promising to drive towards election forecasting using sentiment analysis.
Submission Category: Machine learning algorithms
Submission Number: 43
Loading