<![CDATA[Micro, Small, and Medium Enterprises (MSMEs) in Lhokseumawe struggle to respond to customer comments on the TikTok platform due to limited human and technological resources. This study aims to develop an automatic TikTok comment classification system using a fine-tuned IndoBERT model. The system classifies comments into two main aspects: category (product, price, location, service, etc.) and sentiment (positive, negative, neutral). The dataset used includes 329 comments for category classification and 241 for sentiment classification, collected through web scraping using Selenium and enhanced using synonym-based data augmentation. Evaluation results show that the IndoBERT model achieves an accuracy of 82.07% for category classification and 97.51% for sentiment classification. In comparison, the LSTM model only achieves 59.09% and 73.47% on the same tasks. This system significantly reduces the average response time from 12–24 hours to under 5 minutes. The research contributes to improving customer service efficiency and strengthening digital marketing strategies for local MSMEs on social media platforms.]]>
Copyrights © 2025
