Applications of Natural Language Processing to Predict Suicidal Tendencies from Social Media Textual Input

Many teenagers have increasingly utilized social media to convey their suicidal tendencies, or to seek information on how to commit suicide to the extent of participating in suicide pacts. The objective of the study is to apply natural language processing in a social media messaging platform to recognize suicidal tendencies in user messages. The preprocessed textual message data was then trained to detect suicide using three commonly used machine learning models namely Naïve Bayes, logistic regression, and extreme gradient boosting. Logistic regression obtained the highest diagnostic capability and hence was integrated in a Discord bot that detects suicidal intent on the posted message. The Discord bot is a server-owner-only accessible channel that can automatically receive notifications of probable suicidal intentions and thus was designed to serve as the medium for possible intervention. The bot scans the messages sent to a specific server-owner-only channel, determines whether the message exhibits suicidal disposition or objective, and then outputs the username of the message sender, the actual message, and the likelihood of suicidal tendency. Natural language processing offered new prospects to significantly improve suicide risk assessment contributing heavily to prevention measures and highlighting its clinical utility in the practice of mental health professionals.