Skip to main content Skip to navigation

Deep Learning Mental Health Dialogue System

Project Overview

The document explores the innovative use of generative AI in education, particularly through the development of a mental health dialogue system named Serena, which aims to enhance access to mental health counseling for students. Leveraging a Seq2Seq Transformer model fine-tuned on therapy session transcripts, Serena engages users in therapeutic conversations, addressing significant barriers such as cost, stigma, and time constraints often associated with traditional therapy. By providing limited free online services, Serena seeks to democratize mental health support in educational settings. However, the implementation of this AI system is not without challenges, including issues of response accuracy and maintaining long-term coherence in dialogue. Ultimately, the findings underscore the potential of generative AI to transform educational support services by making mental health resources more accessible while highlighting the need for ongoing improvement in AI reliability and effectiveness.

Key Applications

Serena - a deep learning dialogue system for mental health counseling

Context: Mental health counseling for individuals experiencing psychological distress

Implementation: Deployed as a web application using Google Kubernetes Engine, utilizing a Seq2Seq Transformer model trained on therapy transcripts.

Outcomes: Increased accessibility to mental health resources, low-cost alternative to human counselors, and a tool for assessing therapy needs.

Challenges: Occasional hallucinations in responses, long-term incoherence, and user feedback indicating annoyance with response style.

Implementation Barriers

Accessibility Barrier

High cost of traditional mental health counseling and time constraints for users.

Proposed Solutions: Serena provides a low-cost alternative that can be accessed anytime and anywhere without the need for scheduling.

Stigma Barrier

Fear and stigma associated with seeking mental health counseling from human therapists.

Proposed Solutions: Using a virtual counselor may reduce feelings of shyness and stigma, allowing users to engage more comfortably.

Technical Challenge

Hallucination of knowledge and long-term coherence issues in AI-generated responses.

Proposed Solutions: Implementing post-processing algorithms to detect contradictions and inappropriate responses, and refining training data.

Project Team

Lennart Brocki

Researcher

George C. Dyer

Researcher

Anna Gładka

Researcher

Neo Christopher Chung

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Lennart Brocki, George C. Dyer, Anna Gładka, Neo Christopher Chung

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies