Predicting Voter Behaviour Through AI Techniques

This research is a sentiment analysis of Twitter towards two of the most popular political figures, Donald Trump and Joe Biden. Aspects of a narrative where sentiment evolves overtime, are considered and the role of those changes for a political narrative are evaluated. On social networks, the research similarly measures the influence of political figures by determining sentiment ratios. It seeks to better understand the complexity, and dynamics of political opinion as it is found on twitter in the modern age. The data contained in tweets was subjected to Natural Language Processing (NLP) techniques such as tokenization and VADER-model for sentiment analysis. For example, the VADER tool processes every tweet and provides a compound score which could be negative, positive, or neutral depending on the overall score of the tweet. After selecting the variables, the datasets are prepared by cleaning and filling in missing values as well as changing the tweet column to text. Each politician has their own training and testing dataset obtained from their respective politicians. The dataset has 1000 entries with a dataset that has 40 descriptive attributes per entry.The dataset attributes (detailed in Table 1) included numerical and categorical features, providing a comprehensive set for analysis. Multinomial Naive Bayes model is used to predict the tweet sentiment and the same is used for the training and evaluation of both politicians. Joe Biden's sentiment analysis resulted in a 72% accuracy rate. Donald Trump’s accuracy level is 73%. On the basis of the sentiment analysis conducted, the ratio of positive to negative tweets for Joe Biden and Donald Trump is 1.27 and 1.16, respectively. These numbers correspond to the election outcome where Joe Biden won. Our study effectively analyzed sentiment in a politically sensitive environment, highlighting the reliability of the methods employed. The models used in our work accurately distinguished between positive, negative, and neutral sentiments, indicating their potential for broader use in sentiment analysis. By leveraging machine learning and natural language processing techniques, our models can identify subtle nuances in sentiment with impressive accuracy.