Supervised Machine Learning Entity Sentiment Analysis: Prediction of Support for 2024 Indonesian Presidential Candidates

2024 is a political year in Indonesia as it marks the presidential general election. The proliferation of survey institutions attempting to capture the electability levels of each candidate may not invariably yield accurate results, as evidenced by the events of the 2016 United States Presidential election. The loyal support creates tight competition and a narrow margin in electability levels among the three contending candidates. Opinion mining on social media offers an alternative that addresses the challenges often encountered when measuring electability using traditional survey methods. This study aims to build entity-level sentiment classifiers as a new approach for predicting electability of presidential candidates based on citizen support on social media Twitter within the framework of the CRISP-DM model. The study compares 9 different algorithms with 3 vectorization techniques. Evaluation measurement with 4 metrics: accuracy, precision, recall and f1-score is performed. As a result, TF-IDF 3-gram Random Forest achieves the highest fi-score 0.84486. The selected model is then employed to measure the presidential candidates' electability levels over time. Besides streamlining the process, social media’s opinion mining enables the candidates and their constituents to monitor electability levels affordably in real-time and on-demand manner, which is advantageous compared to traditional surveys.