<![CDATA[The Female Daily platform predominantly features consumer reviews characterized by informal Indonesian and English code-mixing. This specific linguistic complexity presents a significant impediment to traditional classification methods, such as classical machine learning algorithms (e.g., Naive Bayes and SVM) and lexicon-based approaches, which often fail to accurately capture semantic nuances and contextual dependencies in unstructured text. To fill this research gap, this study uses a deep learning method with fine-tuned IndoBERT for multi-brand sentiment analysis. Using a dataset of 12,418 reviews across five popular skincare brands, the model achieved an accuracy of 85%, an F1-score of 0.84, a precision of 0.85, and a recall of 0.84. A key contribution of this research is the multi-brand analysis, which reveals distinct consumer perception patterns: Wardah and Emina achieved the highest proportions of positive sentiment, while Skintific and Garnier demonstrated a more balanced distribution between positive and negative reviews. In contrast, MS Glow exhibited more varied and diverse consumer opinions. These findings confirm that IndoBERT’s self-attention mechanism is highly effective and adaptive in processing the informal, code-mixed vocabulary of the beauty community, outperforming traditional methodologies in both robustness and contextual understanding.]]>
Copyrights © 2026
