<![CDATA[Amid the growing demand for digital identity solutions, applications like Privy, VIDA, and Xignature offer integrated digital signature and e-stamp services, generating extensive user feedback on platforms like Google Play Store and App Store. Extracting meaningful insights from thousands of reviews is challenging, necessitating effective sentiment analysis. Aspect-Based Sentiment Analysis (ABSA) enables detailed sentiment evaluation by linking user feedback to specific aspects and sentiments. However, ABSA faces challenges with imbalanced datasets where label distributions are uneven. This study explores the application of three resampling techniques, including MLROS, MLSMOTE, and REMEDIAL, to address this issue in multilabel classification. Using multilabel classifiers, including Binary Relevance, Label Powerset, and Classifier Chains, the study systematically evaluates their performance. Results reveal that resampling significantly enhances outcomes, with MLROS and Classifier Chains under a 70:30 split achieving the best performance, reducing Hamming Loss to 0.0401 or 95% accuracy. This marks a 34.2% improvement over baseline models without resampling or classifiers. The model generalizes well to unseen data with minimal overfitting, as indicated by validation results. These results underscore the importance of imbalanced data resampling and multilabel classification techniques in advancing ABSA, offering valuable insights for improving sentiment analysis in real-world applications.]]>
