<![CDATA[The implementation of QRIS Cross-border services has triggered diverse responses on YouTube; however, its analysis is hindered by imbalanced data. This study proposes a Hybrid Machine Learning model combining TF-IDF Bigram for statistical features and Bidirectional Long Short-Term Memory (Bi-LSTM) to capture deep contextual features from informal social media text. This hybrid approach is employed to overcome the limitations of single statistical features in understanding complex semantic meanings in YouTube comments. A total of 5,520 comment data points were divided into 80% training data and 20% testing data using the Stratified Split method. To address majority class bias, the SMOTE technique was applied to the training data before classification using a Random Forest algorithm optimized with 1,200 trees (n_estimators). Experimental results show that the Hybrid Bi-LSTM-Random Forest model with SMOTE achieved an accuracy of 93.48%, outperforming SVM (90.31%) and standard Random Forest (88.95%). The application of SMOTE significantly improved the minority class F1-Score from 0.73 to 0.84, with a precision of 0.96. Substantially, public complaints focused on exchange rate issues and technical glitches. The integration of contextual features and data balancing proved effective in producing an accurate and sensitive model for capturing critical public aspirations for financial regulators.]]>
Copyrights © 2026
