TikTok store affiliate performance sentiment analysis using support vector machine and gradient boosting machine methods

Abstract

The development of social media-based e-commerce, particularly, opens new opportunities for digital affiliate systems. This study examines public perception of affiliate performance through comment sentiment analysis (positive, negative, neutral) using support vector machine (SVM) and gradient boosting machine (GBM). Data was collected from TikTok Shop comments, processed through text preprocessing, manual labeling, and then analyzed using Python. Evaluation using accuracy, precision, recall, and F1-score metrics showed that the combination of the synthetic minority oversampling technique (SMOTE) with SVM and GBM improved classification performance, although negative sentiment remained challenging. SVM achieved the highest accuracy (84%) with a ratio of 90:10, while GBM excelled in detecting neutral sentiment (F1 0.91). These findings are useful for sentiment-based marketing strategies and natural language processing (NLP) development for Indonesian-language texts on TikTok Shop.