<![CDATA[Depression is a severe issue in Indonesia, where sufferers often do not seek professional help and prefer to express themselves through social media such as the X. This study uses a text mining approach to classify potential depression using a dataset of 5,000 Indonesian-language tweets from October 2024 to January 2025. The preprocessing steps involves case folding, cleaning, normalization, and stopword removal. The dataset was labeled into two classes: potentially depressive and normal, then divided into 80% training data and 20% test data. A pre-trained IndoBERT model was adjusted with a learning rate of 2e-05, batch size of 8, and epoch of 2 for this depression potential classification task. The evaluation results showed that the IndoBERT model performed well with an accuracy of 87%, precision of 87%, recall of 87%, and f1 score of 87%. However, the model’s performance affected by class imbalance, so it tended to be better at predicting the majority label (normal) than the minority label (depression). Therefore, rebalancing is recommended to prevent similar occurrences. The IndoBERT model used in this study was initialized from an emotion classification model, manual labelling was conducted by researchers in collaboratoin with psychiatrists to ensure clinical relevance. Finally, the trained model was deployed into a web-based application using Streamlit. This application was created as a preliminary screening tool to assist psychiatrists, not as a diagnostic system.]]>
Copyrights © 2025
