<![CDATA[This study examines sentiment classification for Indonesian ride-hailing user reviews, which often contain informal expressions, ambiguity, and strong contextual dependency. Existing studies commonly rely on either traditional machine learning or transformer-based models, while limited attention has been given to integrating heterogeneous feature representations. To address this gap, this study proposes a feature-level hybrid integration strategy combining TF-IDF and IndoBERT embeddings. This approach enables the model to capture statistical term importance and contextual semantic meaning within a unified representation. A quantitative experimental design was applied to approximately 20,000 reviews collected from Gojek, Grab, and Maxim. Sentiment labels were generated through rating-based mapping and manually validated for consistency. The dataset, which was relatively balanced across positive, neutral, and negative classes, was divided into training and testing sets using an 80:20 split. Model performance was evaluated on the test set using accuracy, precision, recall, and F1-score. The proposed hybrid model achieved the highest accuracy of 93.5%, outperforming IndoBERT (91.8%) and traditional machine learning models (78.4%–87.6%). The results show that feature-level integration improves sentiment classification performance, although neutral sentiment remains challenging due to contextual ambiguity.]]>
Copyrights © 2026
