Multi-Dialogue Agents for Collaborative Learning Support Through Adaptive Opinion Integration

Abstract

This research presents a collaborative learning support system that enhances learner motivation by creating psychologically safe discussion environments using multiple LLM-powered agents. These agents are designed to validate and integrate user opinions into learning dialogues dynamically. While agent-based collaborative learning offers flexibility in time and location, achieving truly adaptive learning environments remains challenging. Our system addresses this by leveraging agents that affirm learner contributions through strategically aligned responses during discussions, fostering positive psychological effects of acceptance and belonging. Experimental evaluation with 11 participants demonstrated that 91% experienced increased learning motivation when interacting with opinion-affirming agents. This research contributes to the emerging field of LLM-based educational technology by demonstrating the effectiveness of adaptive multi-agent systems in supporting collaborative learning.