<![CDATA[In Indonesia's dynamic digital economy, customer reviews on marketplace platforms like TikTok Shop, Shopee, and Tokopedia are strategic assets for understanding consumer loyalty and online shopping interest. However, extracting information from thousands of informal reviews presents a significant challenge for rapid business decision-making. This study aims to implement an automated sentiment analysis system by comparing three major machine learning algorithms: Logistic Regression (LR), Naive Bayes (NB), and K-Nearest Neighbors (KNN), utilizing the sentiment strength feature of the Indonesian SentiStrength method. The research dataset consists of 881 reviews collected through crawling techniques and subjected to text preprocessing stages including case folding, cleaning, tokenization, stemming, and stop word removal. Automatic labeling using SentiStrength resulted in a sentiment distribution consisting of Neutral (41.9%), Positive (40.2%), and Negative (17.9%). The data was then divided into training and test data to evaluate the performance of the three algorithms. Experimental results show that all three models performed very reliably in classifying customer opinions. Based on an evaluation using the Classification Report, K-Nearest Neighbors (KNN) provided the most optimal results with an accuracy rate of 99%, followed by Naive Bayes with 96% accuracy, and Logistic Regression with 94%. The high performance of these three models demonstrates that using SentiStrength sentiment scores as input features is highly effective in minimizing language ambiguity. Managerially, this research contributes to digital business practitioners' ability to monitor public perception in real-time to formulate more responsive marketing strategies and maintain customer retention in the marketplace ecosystem]]>
Copyrights © 2026
