Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EduDial: Constructing a Large-scale Multi-turn Teacher-Student Dialogue Corpus

Shouang Wei, Min Zhang, Xin Lin, Bo Jiang, Zhongxiang Dai, Kun Kuang

TL;DR

EduDial advances education-focused large language model evaluation by introducing a large-scale, multi-turn teacher-student dialogue dataset aligned with Bloom's taxonomy and ten questioning strategies. It couples MTI and PDTS data with a two-stage training pipeline (SFT then DPO) to train EduDial-LLM 32B, evaluated via an 11-dimensional framework capturing both teaching quality and content coverage. Across 18 baselines, EduDial-LLM achieves superior teaching capabilities, especially in student-centered interactions and adaptability, highlighting the value of pedagogy-guided data and optimization in LLM-based tutoring. The work provides a reproducible benchmark, a specialized model, and practical insights for deploying LLMs in intelligent education settings.

Abstract

Recently, several multi-turn dialogue benchmarks have been proposed to evaluate the conversational abilities of large language models (LLMs). As LLMs are increasingly recognized as a key technology for advancing intelligent education, owing to their ability to deeply understand instructional contexts and provide personalized guidance, the construction of dedicated teacher-student dialogue benchmarks has become particularly important. To this end, we present EduDial, a comprehensive multi-turn teacher-student dialogue dataset. EduDial covers 345 core knowledge points and consists of 34,250 dialogue sessions generated through interactions between teacher and student agents. Its design is guided by Bloom's taxonomy of educational objectives and incorporates ten questioning strategies, including situational questioning, zone of proximal development (ZPD) questioning, and metacognitive questioning-thus better capturing authentic classroom interactions. Furthermore, we design differentiated teaching strategies for students at different cognitive levels, thereby providing more targeted teaching guidance. Building on EduDial, we further develop EduDial-LLM 32B via training and propose an 11-dimensional evaluation framework that systematically measures the teaching abilities of LLMs, encompassing both overall teaching quality and content quality. Experiments on 17 mainstream LLMs reveal that most models struggle in student-centered teaching scenarios, whereas our EduDial-LLM achieves significant gains, consistently outperforming all baselines across all metrics. The code is available at https://github.com/Mind-Lab-ECNU/EduDial/tree/main.

EduDial: Constructing a Large-scale Multi-turn Teacher-Student Dialogue Corpus

TL;DR

EduDial advances education-focused large language model evaluation by introducing a large-scale, multi-turn teacher-student dialogue dataset aligned with Bloom's taxonomy and ten questioning strategies. It couples MTI and PDTS data with a two-stage training pipeline (SFT then DPO) to train EduDial-LLM 32B, evaluated via an 11-dimensional framework capturing both teaching quality and content coverage. Across 18 baselines, EduDial-LLM achieves superior teaching capabilities, especially in student-centered interactions and adaptability, highlighting the value of pedagogy-guided data and optimization in LLM-based tutoring. The work provides a reproducible benchmark, a specialized model, and practical insights for deploying LLMs in intelligent education settings.

Abstract

Recently, several multi-turn dialogue benchmarks have been proposed to evaluate the conversational abilities of large language models (LLMs). As LLMs are increasingly recognized as a key technology for advancing intelligent education, owing to their ability to deeply understand instructional contexts and provide personalized guidance, the construction of dedicated teacher-student dialogue benchmarks has become particularly important. To this end, we present EduDial, a comprehensive multi-turn teacher-student dialogue dataset. EduDial covers 345 core knowledge points and consists of 34,250 dialogue sessions generated through interactions between teacher and student agents. Its design is guided by Bloom's taxonomy of educational objectives and incorporates ten questioning strategies, including situational questioning, zone of proximal development (ZPD) questioning, and metacognitive questioning-thus better capturing authentic classroom interactions. Furthermore, we design differentiated teaching strategies for students at different cognitive levels, thereby providing more targeted teaching guidance. Building on EduDial, we further develop EduDial-LLM 32B via training and propose an 11-dimensional evaluation framework that systematically measures the teaching abilities of LLMs, encompassing both overall teaching quality and content quality. Experiments on 17 mainstream LLMs reveal that most models struggle in student-centered teaching scenarios, whereas our EduDial-LLM achieves significant gains, consistently outperforming all baselines across all metrics. The code is available at https://github.com/Mind-Lab-ECNU/EduDial/tree/main.
Paper Structure (17 sections, 6 equations, 4 figures, 8 tables)

This paper contains 17 sections, 6 equations, 4 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. The EduDial Dataset
  4. Task Definition
  5. Preparation
  6. Dataset Construction
  7. Two-stage Training Strategy
  8. Evaluation Metrics.
  9. Statistics
  10. Experiments
  11. Experimental Setup
  12. Performance Comparison
  13. Ablation Study
  14. Conclusion
  15. Appendix
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Comparison of "(a) single-turn ” and "(b) mutil-turn teacher-student” dialogue.
  • Figure 2: Based on the teaching syllabus and role profiles, generate high-quality multi-turn teaching instruction data through a five-stage teaching process to train one teacher and three student models.
  • Figure 3: Develop corresponding teaching strategies based on the three student response types, construct preference datasets by generating chosen and rejected responses, and further optimize the teacher model using direct preference optimization.
  • Figure 4: Machine-Human Rating Consistency: Pearson and Spearman Correlation Coefficients Between Machine and Human Evaluations for 18 Models Across 11 Dimensions.