JMIR Publications

ABSTRACT

Background:

The importance of mental health has been increasingly highlighted, yet many individuals still face barriers to accessing suitable interventions. Although AI-based dialogue systems for mental health enhancement have advanced notably to address this issue, comprehensive surveys in this area, particularly those considering studies that adopt large language models (LLMs), remain scarce.

Objective:

This study aims to conduct a quantitative and qualitative review of current research trends in AI-driven dialogue systems for enhancing mental health.

Methods:

This study performed a bibliometric analysis and a trend review analysis of AI-driven dialogue systems for mental health, covering literature from 2020 to May 2024 across three citation databases—WoS, Scopus, and ACM Digital Library. The bibliometric analysis statistically assessed the distribution of publications, while the qualitative trend review focused on three key areas: (i) highly cited publications, (ii) those using the ESConv dataset, and (iii) those employing LLMs.

Results:

We reviewed 146 papers published between 2020 and 2024, observing a steady increase in publications over the last five years. Our bibliometric analysis examined publication distribution across sources, countries, institutions, and authors, while keyword network analysis highlighted major themes. Most of the top 10 highly cited papers focused on empathetic response generation, incorporating psychological approaches within deep learning models. For the ESConv dataset’s application in counseling, prominent techniques included multi-task learning and the integration of external knowledge. Lastly, we identified notable advantages of LLMs over traditional deep learning models and explored strategies to overcome their limitations as counseling tools.

Conclusions:

Our study identifies key areas for developing counseling dialogue systems, such as incorporating psychological knowledge, improving data access, applying LLMs, and refining evaluation methods. By examining current research trends and establishing a foundational framework, this work offers future directions to enhance the effectiveness of AI counseling systems, contributing to both the machine learning and psychology fields.