Exploring the Credibility of Large Language Models for Mental Health Support: Protocol for a Scoping Review

Area of Science:

• Artificial Intelligence
• Mental Health Technology
• Natural Language Processing

Background:

• Large language models (LLMs), including Bidirectional Encoder Representations from Transformers (BERT) and Generative Pre-trained Transformers (GPT), represent significant advancements in natural language processing.
• These AI technologies are increasingly integrated into mental health support applications.
• The credibility, reliability, and explainability of LLMs in mental health contexts require further investigation.

Purpose of the Study:

• To systematically map factors influencing the credibility of LLMs in mental health support.
• To analyze reliability, explainability, and ethical considerations of LLMs for mental health applications.
• To provide insights for practitioners, researchers, and policymakers on responsible LLM integration.

Main Methods:

• Scoping review adhering to PRISMA-ScR and Joanna Briggs Institute (JBI) methodology.
• Eligibility criteria include studies using transformer-based generative language models (e.g., BERT, GPT) in mental health support.
• Systematic search of major databases (PsycINFO, MEDLINE, Web of Science, IEEE Xplore, ACM Digital Library) from 2019 to October 2024, with qualitative data synthesis.

Main Results:

• The systematic search has been completed, and the screening phase is ongoing as of September 2024.
• Data extraction is anticipated by early November 2024, with synthesis expected by late November 2024.
• Current progress indicates an ongoing review process.

Conclusions:

• This review will map existing evidence on LLM credibility in mental health.
• It will identify key factors affecting LLM reliability, explainability, and ethical implications.
• Findings will inform future research, policy, and practice for safe LLM deployment in mental health services.