Enhanced Chatbot for Suicide Risk Detection Using NLP with Explainable AI

Abstract

This chatbot is powered with advanced NLP models, combined with XAI, to allow the detection not only of direct expressions of suicidal thoughts but also the indirect, subtle cues from a person that might indicate poor mental health. Inclusion of XAI provides for the transparency of decision-making so that the process becomes understood by the end user or a professional in the field. The present chatbot, by embedding an empathetic responding mechanism into the conversation, creates an enabling environment where one would feel free to share openly. this could even become a tool for early intervention. It acquires data from the open mental health dataset, pre-processes the information in proper structure so that the quality of the information can be assured. In this paper, the performance of the chatbot can be measured with the help of different metrics-accuracy, precision, recall, and benchmark performance from already proposed solutions for the identification of suicidal ideation. In addition, feedback is solicited from the users to evaluate the usability of the system, its transparency, and perceived empathetic capabilities.