XGBOOST-BASED SENTIMENT ANALYSIS OF MACROECONOMIC POLICY YOUTUBE COMMENTS

Abstract

Public reaction to macroeconomic policy is increasingly expressed through YouTube comment sections, yet this unstructured text is rarely used to gauge sentiment toward government decisions such as exchange-rate management. This study builds and evaluates an XGBoost-based classifier to automatically label the sentiment of YouTube comments discussing Indonesia's macroeconomic policy, particularly the depreciation of the Rupiah. A dataset of 17,838 manually labeled Indonesian-language comments (positive, negative, neutral) was collected and preprocessed through case folding, text cleaning, Indonesian stopword removal, and stemming using the Sastrawi library, then represented as 5,000-dimensional TF-IDF feature vectors. An XGBoost multi-class classifier was trained on an 80:20 stratified train-test split. The model achieved an overall accuracy of 85.68%, with the highest F1-score obtained for the neutral class (0.90) and a comparatively lower recall for the positive class (0.70) due to class imbalance. Word cloud visualization revealed that terms such as “pemerintah,” “rupiah,” and “ekonomi” dominated across all sentiment classes, while “melemah” and “krisis” distinctly marked negative sentiment. These findings demonstrate that XGBoost combined with TF-IDF is an effective and computationally efficient approach for monitoring public sentiment toward macroeconomic policy on social media.