Analisis Sentimen Sikap Publik terhadap Konten Deepfake di Tiktok melalui Model Deep Learning Indobert

Abstract

The rapid advancement of artificial intelligence has contributed to the increasing spread of deepfake content on social media, raising concerns regarding misinformation, opinion manipulation, and digital ethics. This study aims to analyze public sentiment toward deepfake content on TikTok using the IndoBERT model. This research employs a quantitative approach with deep learning-based sentiment analysis. The dataset consists of 4,000 Indonesian-language comments collected from TikTok through a web scraping technique, which involves the automated extraction of public comments using the Apify TikTok Scraper API based on deepfake-related keywords. Data analysis was conducted through several stages, including text preprocessing, sentiment labeling using a lexicon-based approach, dataset splitting into training and testing sets, fine-tuning the IndoBERT model, and performance evaluation using accuracy, precision, recall, and F1-score metrics. The results indicate that negative sentiment dominates with 48.8% of the total comments, followed by neutral sentiment (34.2%) and positive sentiment (17.0%). The IndoBERT model achieved an accuracy score of 78.16% in classifying sentiment. These findings suggest that most TikTok users are concerned about the potential misuse of deepfake technology and demonstrate the effectiveness of IndoBERT for sentiment analysis of Indonesian social media data