Abstract:
"Sri Lanka faces a growing concern regarding mental health, with rising rates of depression and suicidal ideation. Traditional methods of detecting such conditions often rely on clinical settings or self-reported data, which may not capture those at risk due to social stigma or lack of access to mental health services. With the increasing use of social media platforms, user comments provide a valuable source of real-time emotional expression. This research aims to address the critical problem of identifying early signs of suicidal ideation and depression through sentiment analysis of user-generated comments in the Sinhala language. By focusing on a non-clinical and scalable method, this study hopes to contribute to improving mental health detection systems in Sri Lanka.
To predict suicidal ideation and depression, we employed natural language processing (NLP) techniques, focusing on sentiment analysis tailored to the Sinhala language. We collected and pre-processed a dataset of user comments from various Sinhala language forums and social media platforms. Utilizing machine learning models, specifically Support Vector Machine (SVM) and Logistic Regression, we trained the sentiment classifier to detect emotional states related to depression and suicidal thoughts. Our methodology emphasizes the use of annotated data, capturing both suicidal ideation and depressive sentiment, to train the model effectively within the linguistic and cultural context of Sinhala speakers.
Initial findings provide key insights into suicidal ideation and depressive sentiments in Sinhala and Singlish social media comments. The sentiment analysis model achieved 85% accuracy for depressive sentiment and 82% for suicidal ideation. Additional metrics, like precision, recall, and F1 score, further assess the model's performance. These results support the development of early mental health detection strategies for timely interventions in Sri Lanka’s online communities."
