Analisis Sentimen Pengguna X dan YouTube Terhadap Carmen Hearts2Hearts Menggunakan Metode IndoBERT

Abstract

The rapid growth of social media has increased the amount of public opinion expressed online, particularly on platforms such as X and YouTube, where users actively share their views regarding public figures and entertainment topics. This study aims to analyze public sentiment toward Carmen, a member of the K-pop group Hearts2Hearts, using the IndoBERT model for sentiment classification. Data were collected from X and YouTube comments through web scraping techniques and combined into a single dataset to obtain more diverse opinions. The research process involved several stages, including text preprocessing, manual sentiment labeling, dataset splitting, model training, and evaluation. The preprocessing stage consisted of duplicate data removal, case folding, noise removal, tokenization, stopword removal, and stemming to improve data quality before classification. The dataset was categorized into three sentiment classes: positive, neutral, and negative, then divided into training and testing data using an 80:20 ratio. The IndoBERT model was trained using transformer-based deep learning to understand the context of Indonesian-language text more effectively. Evaluation results showed that the model achieved an accuracy of 72.41%, precision of 75.82%, recall of 72.41%, and F1-score of 71.15%, indicating that IndoBERT performs effectively in classifying sentiment on Indonesian social media data despite challenges such as informal language and ambiguous expressions.