Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agent4Edu: Generating Learner Response Data by Generative Agents for Intelligent Education Systems

Weibo Gao, Qi Liu, Linan Yue, Fangzhou Yao, Rui Lv, Zheng Zhang, Hao Wang, Zhenya Huang

TL;DR

Agent4Edu tackles the offline-online gap in personalized education by introducing LLM-powered generative agents that simulate individual learners through a structured profile-memory-action design. Each agent inherits a learner profile (explicit styles and implicit ability), memory (factual, short-term, long-term with reflection and forgetting), and an action module that reads, analyzes, and solves exercises while aligning with CAT-based learning environments. The framework supports interaction with personalized algorithms, enabling comprehensive agent-based evaluation and data augmentation (EduData+), which in turn improves cognitive diagnosis models used in CAT. Across extensive experiments, agents demonstrate competitive fidelity to real learners, useful zero-shot capabilities, and the potential to enhance personalized learning services, with the code and data publicly available for replication. This work provides a versatile platform for generating high-fidelity learner data and benchmarking intelligent tutoring systems in real-world educational settings.

Abstract

Personalized learning represents a promising educational strategy within intelligent educational systems, aiming to enhance learners' practice efficiency. However, the discrepancy between offline metrics and online performance significantly impedes their progress. To address this challenge, we introduce Agent4Edu, a novel personalized learning simulator leveraging recent advancements in human intelligence through large language models (LLMs). Agent4Edu features LLM-powered generative agents equipped with learner profile, memory, and action modules tailored to personalized learning algorithms. The learner profiles are initialized using real-world response data, capturing practice styles and cognitive factors. Inspired by human psychology theory, the memory module records practice facts and high-level summaries, integrating reflection mechanisms. The action module supports various behaviors, including exercise understanding, analysis, and response generation. Each agent can interact with personalized learning algorithms, such as computerized adaptive testing, enabling a multifaceted evaluation and enhancement of customized services. Through a comprehensive assessment, we explore the strengths and weaknesses of Agent4Edu, emphasizing the consistency and discrepancies in responses between agents and human learners. The code, data, and appendix are publicly available at https://github.com/bigdata-ustc/Agent4Edu.

Agent4Edu: Generating Learner Response Data by Generative Agents for Intelligent Education Systems

TL;DR

Agent4Edu tackles the offline-online gap in personalized education by introducing LLM-powered generative agents that simulate individual learners through a structured profile-memory-action design. Each agent inherits a learner profile (explicit styles and implicit ability), memory (factual, short-term, long-term with reflection and forgetting), and an action module that reads, analyzes, and solves exercises while aligning with CAT-based learning environments. The framework supports interaction with personalized algorithms, enabling comprehensive agent-based evaluation and data augmentation (EduData+), which in turn improves cognitive diagnosis models used in CAT. Across extensive experiments, agents demonstrate competitive fidelity to real learners, useful zero-shot capabilities, and the potential to enhance personalized learning services, with the code and data publicly available for replication. This work provides a versatile platform for generating high-fidelity learner data and benchmarking intelligent tutoring systems in real-world educational settings.

Abstract

Personalized learning represents a promising educational strategy within intelligent educational systems, aiming to enhance learners' practice efficiency. However, the discrepancy between offline metrics and online performance significantly impedes their progress. To address this challenge, we introduce Agent4Edu, a novel personalized learning simulator leveraging recent advancements in human intelligence through large language models (LLMs). Agent4Edu features LLM-powered generative agents equipped with learner profile, memory, and action modules tailored to personalized learning algorithms. The learner profiles are initialized using real-world response data, capturing practice styles and cognitive factors. Inspired by human psychology theory, the memory module records practice facts and high-level summaries, integrating reflection mechanisms. The action module supports various behaviors, including exercise understanding, analysis, and response generation. Each agent can interact with personalized learning algorithms, such as computerized adaptive testing, enabling a multifaceted evaluation and enhancement of customized services. Through a comprehensive assessment, we explore the strengths and weaknesses of Agent4Edu, emphasizing the consistency and discrepancies in responses between agents and human learners. The code, data, and appendix are publicly available at https://github.com/bigdata-ustc/Agent4Edu.
Paper Structure (23 sections, 2 figures, 4 tables)

This paper contains 23 sections, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Agent4Edu
  4. Task Formulation
  5. LLM-powered Agent
  6. Learner Profile Module
  7. Memory Module
  8. Action Module
  9. Personalized Learning Scenarios
  10. Experiment
  11. Dataset
  12. Experimental Setup
  13. LLM-based Agent Simulation Evaluation
  14. Motivation:
  15. Learner Simulation Evaluation
  16. ...and 8 more sections

Figures (2)

  • Figure 1: The overall framework of Agent4Edu.
  • Figure 2: (a) Comparison between the success rate distributions of ground-truth and agent-simulated response data. (b) Using LLMs as judges to identify whether the records of the agent’s simulations originate from real humans. (c) The ablation studies.