We envision a world in which talk therapy can be given by a computer, exactly where and when it is needed. Recent advances in Natural Language Processing (NLP) and Deep Learning, have provided new ways to bring this vision to a reality.
It is a collaboration with the Nicotine Dependence Clinic at the Centre for Addiction and Mental Health and with the iSchool at the University of Toronto.
The first prototype chatbot was developed by Fahad Almusharraf in his M.A.Sc. thesis the first part of which (describing the design of the bot) was published in this paper. This bot makes use of an NLP classifier to provide specific relevant reflections and responses to subjects who describe the reasons for and against smoking. The second paper (which describes the effect of the intervention) is in preparation.
That work provided a rich dataset of conversations that is being used as the basis for a more advanced bot, in which we would like to generate motivational-interviewing style reflections that are not pre-written, using few shot generation techniques with transformer architectures. It it is related to the work on reflection generation from the University of Michigan.
The Edward S. Rogers Sr. Department of Electrical and Computer Engineering,
Faculty of Applied Science and Engineering, University of Toronto