Analisis Sentimen Review Brand Kecantikan Lokal pada Platform X Menggunakan Algoritma Support Vector Machine dan IndoBERTweet

The beauty industry in Indonesia has experienced rapid growth along with the increasing public interest in skincare and makeup products. Social media, particularly platform X (Twitter), has become a primary medium for consumers to share their opinions about local beauty brands. These consumer reviews can serve as valuable sources of information, yet they are often subjective and unstructured, making sentiment analysis necessary. This study aims to compare the performance of the Support Vector Machine (SVM) algorithm with the transformer-based IndoBERTweet model in classifying consumer sentiment toward five local beauty brands (Wardah, Scarlett Whitening, Avoskin, Skin Game, and Somethinc).
Data were collected through crawling on platform X, resulting in 13,934 tweets, from which 4,000 samples were randomly selected, cleaned, and manually labeled into four sentiment classes: positive, negative, neutral, and others. The analysis was conducted under three classification scenarios: two-class (others vs. no_others), three-class (positive, negative, neutral), and four-class (positive, negative, neutral, others). To address class imbalance, Random Oversampling was applied, while SVM was optimized using TF-IDF and Grid Search, and IndoBERTweet was fine-tuned for improved performance.
The experimental results demonstrate that IndoBERTweet consistently outperforms SVM. In the two-class scenario, IndoBERTweet achieved an accuracy of 95% compared to SVM’s 93%. In the three-class scenario, IndoBERTweet obtained 92% accuracy while SVM reached 89%, and in the four-class scenario IndoBERTweet achieved 88% while SVM achieved 86%. These findings highlight IndoBERTweet’s superior ability to understand Indonesian-language context on social media, making it a more suitable approach for multi-class sentiment analysis of local beauty brand reviews.