Sentiment Analysis of Tweets to predict Sri Lankan Election Results using Supervised Learning Techniques

Abstract

Due to the current pandemic situation in the world, people are spending their leisure time mostly on social media platforms. They more tend to express their selves openly when they are behind the keyboard. This user behavior has created a huge advantage for researchers and analyzers to analyze people’s opinions, behaviors and predict certain outcomes. This research study is used to get the best out of aforesaid user behavior and conduct the prediction-based analysis using the Twitter — social media platform. When we consider the election prediction using sentiment analysis, there were many researches done based on the languages English, Chinese, Arabic, Hindi etc. But this is a novel application area for Sinhala language(de Silva, 2020). Even though there are several studies available for Sinhala language, they cannot be directly used for Election prediction since sentiment analysis is highly application dependent. Applications which develop for one domain cannot be used for another domain in sentiment analysis. And another issue is, same methodologies and technologies which are used for other languages, cannot be directly used for Sinhala language due to language differences. So, the focus here is to create a domain specific research for the area of Election prediction and to introduce new resources to the text analysis community which will be helpful for their further studies. In this research, prediction — based system was developed using Sinhala tweets. Automatic labelling was used to predict the election results for each candidate. These predicted results were compared with the actual presidential election results in Sri Lanka – 2019. Suitable model was developed by applying text preprocessing and feature extraction techniques. Supervised learning classifiers were trained against the developed model to find the best classifiers for predictive sentiment analysis in Sinhala language.