Simulating Classroom Education with LLM-Empowered Agents

TL;DR

This paper addresses the challenge of simulating authentic classroom dynamics using LLM-powered multi-agent systems with real user participation. It introduces SimClass, a framework that defines Teaching and Classmate Agent roles and a Session Controller to manage classroom flow, enabling teacher-student, student-student, and peer interactions. The authors validate SimClass in two university courses with 400+ participants and perform ablation studies; they analyze interactions via Flanders Analysis and learning experiences via Community of Inquiry, showing enhanced engagement and learning outcomes. The work demonstrates the potential of LLM-empowered multi-agent classrooms and provides datasets and methodological insights for education and AI researchers.

Abstract

Large language models (LLMs) have been applied across various intelligent educational tasks to assist teaching. While preliminary studies have focused on task-specific, independent LLM-empowered agents, the potential of LLMs within a multi-agent collaborative framework for classroom simulation with real user participation remains unexplored. In this work, we propose SimClass, a multi-agent classroom simulation teaching framework. We recognize representative class roles and introduce a novel class control mechanism for automatic classroom teaching, and conduct user experiments in two real-world courses. Using the Flanders Interactive Analysis System and Community of Inquiry theoretical frameworks from educational analysis, we demonstrate that LLMs can simulate a dynamic learning environment for users with active teacher-student and student-student interactions. We also observe group behaviors among agents in SimClass, where agents collaborate to create enlivening interactions in classrooms to improve user learning process. We hope this work pioneers the application of LLM-empowered multi-agent systems in virtual classroom teaching.

Figures (5)

• Figure 1: An overview of the SimClass framework. Note that the upper portion of the framework is visible to student users, while the lower portion is hidden from them. In the classroom, users can view the current slide, configure class roles, and engage in real-time conversations with the agents.
• Figure 2: The FIAS matrix sum of users in TAGI (left) and HSU (right). Numbers $1\text{--}9$ represent the corresponding categories. $N$ in location $( x, y )$ means that there are $N$ transitions from $x$ to $y$ in the classroom. The matrix is divided into four parts based on the type of interaction between actors.
• Figure 3: The joint plot of students' normalized average quiz scores, against the logarithm of message lengths per message (left) and against the logarithm of average number of messages per chapter (right).
• Figure 4: User Results based on the CoI framework. The black lines represent the standard error of the data statistics.
• Figure 5: The FIAS matrix sum of users in TAGI (left) and HSU (right) without interactions.