End-to-end dialogue with sentiment analysis features

Abstract

Psychiatric assistance for suicide prevention does not have a wide enough reach to help the number of victims who commit suicide every year. To help people cope with suicidal thoughts when formal care is unavailable, we propose an artificial intelligence, text-based conversational agent that generates responses similar to those of a counselor. The application will offer a temporary channel for expression that serves as a transition to speaking with a professional psychiatrist. We expand upon existing approaches by utilizing sentiment analysis data, or scores that rank the emotional content of users’ text input, when generating responses. We also train a response generation system based on a dataset of counseling and therapy transcripts. We posit that inclusion of sentiment analysis data provides marginally better responses based on quantitative metrics of quality. We hope our results will advance realistic conversation modeling and promote further research into its humanitarian applications.