Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MetaCD: A Meta Learning Framework for Cognitive Diagnosis based on Continual Learning

Jin Wu, Chanjin Zheng

TL;DR

MetaCD tackles cognitive diagnosis under long-tailed and non-stationary data by combining meta-learning to learn favorable initializations with continual learning to preserve knowledge across evolving tasks. It introduces a knowledge-base (KB) module that is trained and queried through meta-learning stages (KB Train Support, KB Train Query, KB Test Support) and protected by a Parameter Protection Mechanism (PPM) to balance stability and plasticity, aided by a per-class head to sharpen decisions. A KL-divergence-based boundary sharpening step reduces fuzzy boundaries between diagnoses, further improving accuracy. Experiments on five real-world datasets show MetaCD achieving superior ACC and AUC and strong generalization to sparse data, with reduced forgetting in task-incremental settings; ablations confirm the centrality of the KB component and the benefit of PPM. The framework offers a practical approach for robust cognitive diagnostics in online education where data distributions shift over time.

Abstract

Cognitive diagnosis is an essential research topic in intelligent education, aimed at assessing the level of mastery of different skills by students. So far, many research works have used deep learning models to explore the complex interactions between students, questions, and skills. However, the performance of existing method is frequently limited by the long-tailed distribution and dynamic changes in the data. To address these challenges, we propose a meta-learning framework for cognitive diagnosis based on continual learning (MetaCD). This framework can alleviate the long-tailed problem by utilizing meta-learning to learn the optimal initialization state, enabling the model to achieve good accuracy on new tasks with only a small amount of data. In addition, we utilize a continual learning method named parameter protection mechanism to give MetaCD the ability to adapt to new skills or new tasks, in order to adapt to dynamic changes in data. MetaCD can not only improve the plasticity of our model on a single task, but also ensure the stability and generalization of the model on sequential tasks. Comprehensive experiments on five real-world datasets show that MetaCD outperforms other baselines in both accuracy and generalization.

MetaCD: A Meta Learning Framework for Cognitive Diagnosis based on Continual Learning

TL;DR

MetaCD tackles cognitive diagnosis under long-tailed and non-stationary data by combining meta-learning to learn favorable initializations with continual learning to preserve knowledge across evolving tasks. It introduces a knowledge-base (KB) module that is trained and queried through meta-learning stages (KB Train Support, KB Train Query, KB Test Support) and protected by a Parameter Protection Mechanism (PPM) to balance stability and plasticity, aided by a per-class head to sharpen decisions. A KL-divergence-based boundary sharpening step reduces fuzzy boundaries between diagnoses, further improving accuracy. Experiments on five real-world datasets show MetaCD achieving superior ACC and AUC and strong generalization to sparse data, with reduced forgetting in task-incremental settings; ablations confirm the centrality of the KB component and the benefit of PPM. The framework offers a practical approach for robust cognitive diagnostics in online education where data distributions shift over time.

Abstract

Cognitive diagnosis is an essential research topic in intelligent education, aimed at assessing the level of mastery of different skills by students. So far, many research works have used deep learning models to explore the complex interactions between students, questions, and skills. However, the performance of existing method is frequently limited by the long-tailed distribution and dynamic changes in the data. To address these challenges, we propose a meta-learning framework for cognitive diagnosis based on continual learning (MetaCD). This framework can alleviate the long-tailed problem by utilizing meta-learning to learn the optimal initialization state, enabling the model to achieve good accuracy on new tasks with only a small amount of data. In addition, we utilize a continual learning method named parameter protection mechanism to give MetaCD the ability to adapt to new skills or new tasks, in order to adapt to dynamic changes in data. MetaCD can not only improve the plasticity of our model on a single task, but also ensure the stability and generalization of the model on sequential tasks. Comprehensive experiments on five real-world datasets show that MetaCD outperforms other baselines in both accuracy and generalization.
Paper Structure (11 sections, 8 equations, 2 figures, 5 tables)

This paper contains 11 sections, 8 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Our Proposed MetaCD Method
  3. Task Overview
  4. The Structure of MetaCD
  5. Experimental analysis and results
  6. Conclusion and Limitations
  7. Technical Appendices
  8. Dataset Description
  9. Experimental Setup
  10. Evaluation metrics
  11. Detailed Results of Experiments

Figures (2)

  • Figure 1: The overall structure of MetaCD.
  • Figure 2: Performance comparison on long-tailed data. The horizontal coordinate represents the number of times that a question has been answered by different students. For example, 5-10 represents a question that has been answered by different students between 5 and 10 times.