Recognizing Suicidal Intent in Depressed Population using NLP: A Pilot Study

Depression is a prevalent form of mental disorder that can affect productivity in daily activities and might lead to suicidal thoughts or attempts. Conventional diagnostic techniques performed by mental health professionals can help identify the level of depression present in a person. To facilitate such a diagnostic approach, in this paper, we present an automated conversational platform that was used as a preliminary method of identifying depression associated risks. The platform was developed to understand conversations using Natural Language Processing (NLP) via machine learning technique. In the proposed two-phased platform, the initial intent recognition phase would analyze conversation and identify associated sentiments into four categories of `happy', `neutral', `depressive' and `suicidal' states. In the final emotion nurturing phase, the platform continued with supportive conversations for the first three states while triggering a local call to a suicide prevention helpline for `suicidal' state as a preventive measure. This multi-layer platform integrated Google Home mini, Google Dialogflow Machine Learning (ML) algorithm and Twilio API. Dialogflow ML obtained classification accuracy of 76% in recognizing user's mental state via NLP and was found efficient over the classic SVM classifier. As a pilot study, current focus of this paper was solely based on the usage of words and intent of the user and was found effective.