HAMAgent: human assisted multiagent system for emotion recognition and digital health-a survey and preliminary study

This study surveys existing multi-large language model (LLM) agent applications, comparing systems that incorporate active human participation with those that operate fully autonomously within digital health contexts. Building on this analysis, we conduct an early exploration of human-in-the-loop feedback in multi-LLM interactions for emotion and human behaviour understanding using a subset of the FairytaleQA dataset. We further propose the HAMAgent framework to investigate how human feedback influences multi-agent reasoning and performance. Our preliminary experiments demonstrate the potential benefits of integrating human guidance into multi-LLM workflows and provide insights for designing effective human-in-the-loop systems in digital health. We conclude by discussing the key implications of our findings and outlining future directions for multi-agent LLM systems enhanced with human input.