<![CDATA[Digital transformation in the public sector encourages local governments to enhance service quality through online complaint management systems. However, the high volume of incoming complaints and significant data imbalance across 31 Organisasi Perangkat Daerah (OPD) pose challenges for efficient manual classification, often resulting in delays and misclassification. This study proposes an automated text classification model that integrates Long Short-Term Memory (LSTM), Word2Vec, and Random Oversampling (ROS), optimized using the Adam algorithm. The novelty of this research lies in the integration of sequential modeling and imbalance handling to address an extreme multi-class classification problem involving 31 OPD categories within a highly imbalanced dataset. The research stages include text preprocessing, word embedding construction using Word2Vec, data balancing through ROS, and model training using LSTM. Experimental results show that the proposed model achieves an accuracy of 0.72, with macro-average precision, recall, and F1-score of 0.67, 0.67, and 0.66, respectively. Considering the complexity of classifying 31 classes and the presence of severe data imbalance, the macro F1-score of 0.66 indicates that the model is reasonably effective in capturing classification patterns, although performance is not yet evenly distributed across all classes. Overall, the combination of LSTM, Word2Vec, and ROS demonstrates potential as a baseline approach for automating public complaint classification in complex multi-class scenarios. The proposed model can improve the accuracy and speed of complaint distribution to the appropriate OPD, thereby enhancing the efficiency and responsiveness of public service delivery compared to conventional manual methods.]]>
Copyrights © 2026
