Research Summary: EduAgentQG: Multi-Agent Personalized Mathematics Question Generation with Explicit Diversity and Objective-Aware Evaluation

Research introduces EduAgentQG, a multi-agent framework designed to enhance personalized mathematics question generation in intelligent education. This system addresses limitations of current LLM-based methods by integrating explicit diversity and objective-aware evaluation within a collaborative, closed-loop process of planning, writing, and evaluation. It aims to improve both the educational relevance and adaptive assessment capabilities of generated questions.

Why it matters

This development is crucial for advancing personalized learning technologies and intelligent education systems. It offers a structured approach to generate highly relevant and diverse educational content, which can significantly enhance assessment efficacy and student learning outcomes across various educational contexts.

Key insights

• Existing LLM-based single-agent and multi-agent systems for personalized mathematics question generation often lack joint objective alignment and controllable diversity.
• EduAgentQG is a multi-agent collaborative framework that aims to produce personalized mathematics questions.
• The framework incorporates explicit diversity and objective-aware evaluation to overcome current limitations.
• Question generation within EduAgentQG is structured as a closed-loop process involving planning, writing, and evaluation.
• The primary goal is to satisfy educational requirements and support adaptive assessment and learning through generated questions.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2511.11635