A Large-Scale Dataset for Motivational Dialogue System: An Application of Natural Language Generation to Mental Health

Tulika Saha, Saraansh Chopra, Sriparna Saha, Pushpak Bhattacharyya, Pankaj Kumar

2021 · 20 citations · 30 references

Concepts

Abstract

Major Depressive Disorder is one of the most serious mental health concerns haunting human kind. While there have been enough global efforts to reduce clinical depression, it is still not sufficient considering the grim statistics in terms of mental health. With the lopsided distribution of Mental Health Professionals (MHPs) to patients, initiatives should be taken to develop automated systems to assist MHPs in combating this grievous mental illness. In this paper, we propose a Virtual Assistant (VA) acting as a first point of contact for depressed or discouraged users. The VA should be capable of generating motivational and affirmative responses, provide users with a safe environment to share thoughts and seek help and advice anonymously. For addressing these multiple aspects, we curate a large-scale dataset, MotiVAte, consisting of dyadic conversations between the depressed user and the VA (imparting hope and motivation). The proposed dataset is evaluated on state of the art generative models. Empirical results both automatic and human evaluation based are reported for the same. To the best of our knowledge, this is the first application of Natural Language Generation to mental health.

References

30