Analisis Sentimen Pada Komentar YouTube Terhadap Video Clash of Champions Season 2 Menggunakan Metode Xgboost

Abstract

This study aims to evaluate the performance of the Extreme Gradient Boosting (XGBoost) algorithm in sentiment classification of YouTube comments on the educational video series “Clash of Champions Season 2.” User comments on social media platforms produce large volumes of unstructured textual data with imbalanced sentiment distributions, creating challenges for automated sentiment analysis. The dataset was collected using the YouTube Data Application Programming Interface (API) through Google Colaboratory. Text preprocessing was conducted to improve data quality, including cleaning, case normalization, tokenization, stopword removal, stemming, and text normalization. Sentiment labeling was performed automatically using a polarity score approach to categorize comments into positive, neutral, and negative sentiments. Textual data were transformed into numerical representations using the Term Frequency–Inverse Document Frequency (TF-IDF) method. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to the training dataset. The XGBoost model was trained under two schemes, namely without data balancing and with SMOTE, to compare classification performance. Model evaluation was conducted using accuracy, precision, recall, and F1-score metrics. The results indicate that XGBoost achieved high and stable classification performance on testing data. The implementation of SMOTE improved the model’s ability to recognize minority sentiment classes without significantly reducing overall performance. Furthermore, wordcloud visualization demonstrated distinct linguistic patterns across sentiment categories. Overall, the combination of XGBoost, TF-IDF, and SMOTE provides an effective and reliable approach for sentiment analysis of Indonesian YouTube comments.