Enhancing text summarization with AI: a multi-agent system and human comparison in educational contexts

This paper introduces the Mixture-of-Agents (MoA) framework, a novel system designed to enhance text summarization by leveraging the complementary strengths of multiple large language models (LLMs). The framework dynamically integrates specialized agents, enabling the generation of summaries that excel in coherence, factual accuracy, and brevity. Evaluated on a dataset of 10 educational scenarios encompassing human-written narratives and AI-generated summaries, the MoA framework demonstrated a 15 % improvement in narrative coherence and a 12 % gain in factual accuracy compared to 20 state-of-the-art summarization models.

The framework’s educational application was tested through comparative analysis of human- and AI-generated summaries in both 50-word and 15-word formats. Results highlight that while AI-generated summaries excel in factual consistency, human summaries retain greater creativity and narrative depth. By iteratively refining outputs, the MoA framework approaches human-level performance in long-form summarization tasks, bridging the gap between human and AI capabilities.

This study contributes to text summarization research by introducing an adaptive multi-agent framework, conducting an in-depth analysis of humanAI differences in summarization, and demonstrating the potential of AIdriven tools to enhance creative writing and learning in educational settings.