Aladağ, Ahmet Emre, et al. "Detecting suicidal ideation on forums: proof-of-concept study." Journal of medical Internet research 20.6 (2018): e9840.

This study aims to develop a classifier to differentiate between suicidal and nonsuicidal forum posts using text mining methods. The research analyzed 508,398 Reddit posts longer than 100 characters from the SuicideWatch, Depression, Anxiety, and ShowerThoughts subreddits, selecting 10,785 posts for annotation, with 785 manually labeled as suicidal or nonsuicidal. Features were extracted using term frequency-inverse document frequency, linguistic inquiry and word count, and sentiment analysis. Logistic regression, random forest, and support vector machine (SVM) classifiers were applied. Results showed that logistic regression and SVM classifiers achieved 80% to 92% accuracy and F1 scores, significantly outperforming the baseline ZeroR algorithm. The study suggests that these classifiers can be used to detect suicidal ideation in real-time on forums, potentially offering immediate online counseling.