SSRLBot: Designing and Developing an LLM-based Agent using Socially Shared Regulated Learning

2025-05-02

AI Summary

Large language model (LLM) agents for multi-specialty medical team decision-making lack theoretical grounding and real-time regulatory capability. Method: We propose SSRLBot, the first LLM agent grounded in Social Shared Regulation of Learning (SSRL) theory. It integrates an SSRL system into agent architecture to enable dialogue-level SSRL marker identification, cross-dimensional (metacognitive/motivational/affective) behavioral impact assessment, and dynamic interpersonal regulation modeling. The agent comprises an SSRL rule engine, multidimensional diagnostic performance analyzer, and reflective suggestion generator. Results: Evaluated on GPT-3.5, Gemini-1.5, and DeepSeek-R1, SSRLBot significantly outperforms baselines, achieving an F1 score of 0.89 for SSRL marker identification. It establishes an interpretable causal chain from observed behavior → regulatory dimension → diagnostic outcome and generates actionable, theory-informed recommendations for team collaboration improvement—thereby addressing a critical gap in theory-driven, collaborative AI agent research.

Technology Category

• Multiagent Systems: Teamwork
• Cognitive Modeling & Cognitive Systems: Social Cognition And Interaction
• Humans and AI: Planning and Decision Support for Human-Machine Teams

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Agentic search
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Large language model (LLM)-based agents are increasingly used to support human experts by streamlining complex tasks and offering actionable insights. However, their application in multi-professional decision-making, particularly in teamwork contexts, remains underexplored. This design-based study addresses that gap by developing LLM functions to enhance collaboration, grounded in the Socially Shared Regulation of Learning (SSRL) framework and applied to medical diagnostic teamwork. SSRL emphasizes metacognitive, cognitive, motivational, and emotional processes in shared learning, focusing on how teams manage these processes to improve decision-making. This paper introduces SSRLBot, a prototype chatbot designed to help team members reflect on both their diagnostic performance and key SSRL skills. Its core functions include summarizing dialogues, analyzing SSRL behaviors, evaluating diagnostic outcomes, annotating SSRL markers in conversation, assessing their impact on performance, and identifying interpersonal regulatory dynamics. We compare SSRLBot's capabilities with those of Gemini-1.5, GPT-3.5, and Deepseek-R1 in a case study. SSRLBot demonstrates stronger alignment with SSRL theory, offering detailed evaluations that link behaviors to regulatory dimensions and suggesting improvements for collaboration. By integrating SSRL theory with LLM capabilities, SSRLBot contributes a novel tool for enhancing team-based decision-making and collaborative learning in high-stakes environments, such as medical education.

Problem

• Developing LLM-based agent for multi-professional teamwork decision-making
• Enhancing collaboration using Socially Shared Regulation of Learning framework
• Improving medical diagnostic teamwork through reflective chatbot functions

Innovation

• LLM-based agent for teamwork using SSRL framework
• Chatbot evaluates diagnostic performance and SSRL skills
• Integrates SSRL theory with LLM for medical collaboration