ANALISIS SENTIMEN PUBLIK TERHADAP REGULASI TIKTOK SHOP DI INDONESIA PADA PLATFORM YOUTUBE MENGGUNAKAN ALGORITMA LSTM DAN INDOBERT

The rise of TikTok Shop as an e-commerce feature within the TikTok social media platform has sparked public debate, especially following the issuance of Ministry of Trade Regulation No. 31 of 2023, which prohibits direct commercial transactions on social media. This study aims to analyze public sentiment toward the regulation based on user comments on the YouTube platform. A total of 4,563 comments were collected using the YouTube API and manually labeled into four sentiment categories: pro, contra, neutral, and others. Two modeling approaches were implemented: Long Short-Term Memory (LSTM) with FastText text representation, and IndoBERT, a Transformer-based pre-trained model for the Indonesian language. Experimental results show that IndoBERT consistently outperformed LSTM in both binary and multiclass classification tasks, achieving peak accuracy of 95% in the binary scenario and 85% in the multiclass setting. Error analysis revealed key challenges such as difficulty in understanding sarcasm, interpreting implicit meanings, and handling seemingly neutral comments that contain emotionally polarized words. These findings highlight the effectiveness of pre-trained language models like IndoBERT in more accurately capturing public opinion on digital policy issues in the Indonesian context.