DETEKSI RISIKO DEPRESI PADA PENGGUNA MEDIA SOSIAL INDONESIA MENGGUNAKAN INDOBERTWEET

Abstract

This study created a depression risk detection model for Indonesian-speaking social media users using IndoBERTweet, a pre-drilled transformer-based language model optimized specifically for Indonesian on Twitter. Data collection was conducted from 5,000 public Twitter posts using keywords related to negative emotional expressions and psychological terms, obtained between January and June 2024. The data was labeled using a combination of a translated PHQ-9 (Patient Health Questionnaire-9) lexicon and manual annotation by three clinical psychologists, resulting in a high level of inter-rater consistency (Cohen's Kappa of 0.82). The IndoBERTweet model was fine-tuned and compared with several baseline models, including SVM with TF-IDF, LSTM with Word2Vec, and IndoBERT-base. The results showed that IndoBERTweet achieved the highest accuracy of 91.4% and an F1 score of 0.89, significantly outperforming the baseline models (with a 7–10% difference in F1 score). These findings demonstrate that the local-language transformer model is highly effective in capturing the contextual and informal nuances of Indonesian language used to indicate depression in social media data. This research significantly contributes to the development of a Natural Language Processing (NLP)-based mental health early screening system within the Indonesian linguistic and cultural context.