Research map / Label-efficient and robust learning / Recognition and long-tail
Towards Accurate and Fair Cognitive Diagnosis via Monotonic Data Augmentation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes CMCD, a monotonic data augmentation framework that flips student responses to create fake students constrained by the monotonicity assumption, improving accuracy and fairness of cognitive diagnosis under data sparsity.
Problem
Data sparsity in cognitive diagnosis (students interacting with few exercises) leads to inaccurate and unfair proficiency diagnoses.
Approach
Generate fake students by flipping one response per student and enforce monotonicity-based ordering constraints (regularization terms) between original and augmented students; model-agnostic and applicable to IRT, MIRT, and NCDM backbones, with theoretical consistency and convergence-speed analysis.
Key results
- On Math, CMCD improves IRT RMSE from 0.409 to 0.394 and ACC from 0.751 to 0.772 versus Origin, with the best fairness gaps across IRT and MIRT backbones.
- CMCD gains are larger on IRT than on NCDM, which the authors attribute to NCDM's stronger data-fitting capacity.
- Ablation on Math shows removing either hypothesis degrades both fairness and utility.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add monotonic data augmentation | IRT · Math | ACC: 0.751 -> 0.772 | Helped |
| Add monotonic data augmentation | MIRT · Math | ACC: 0.757 -> 0.767 | Helped |
| Add monotonic data augmentation | NCDM · Math | ∆ACC: 0.054 -> 0.050 | Helped |
| Add monotonic data augmentation | IRT · ASSIST | ACC: 0.693 -> 0.715 | Helped |
| Add monotonic data augmentation | MIRT · ASSIST | ACC: 0.707 -> 0.724 | Helped |
| Add monotonic data augmentation | NCDM · ASSIST | RMSE: 0.433 -> 0.439 | Hurt |
| Remove Hypothesis 1 constraint (ablation) | CMCD · Math | ACC | Hurt |
| Remove Hypothesis 2 constraint (ablation) | CMCD · Math | ACC | Hurt |
| Increase constraint strength c (ablation) | CMCD · Math | ACC | Mixed |
| Increase number of generated students |Ci| (ablation) | CMCD with IRT · Math | ACC | Helped |
| Increase number of generated students |Ci| (ablation) | CMCD with NCDM · Math | ACC | Hurt |
Surprising findings
- Fairness-aware baselines (CD+EO, CD+DP) often reduce utility, while the augmentation method improves both utility and fairness in many cases.
- Larger numbers of augmented students |Ci| hurt NCDM utility, unlike IRT where more augmentation consistently helps.
Limitations
- Relies on student response records that may be inaccessible due to privacy concerns.
- Gains are less pronounced for neural backbones such as NCDM and can be hurt by excessive augmentation or constraint strength.
- Fairness results against fairness-aware baselines are mixed on some datasets; ASSIST uses a non-sensitive proxy group (response count > 50).
Experimental setup
Datasets: Math (online education system, junior high students) ASSIST (ASSISTments 2009-2010 skill builder)
Benchmarks: Math ASSIST
Metrics: RMSE MAE AUC ACC fairness gap (delta RMSE, MAE, AUC, ACC)
Baselines: CF-IRT CF-MIRT CF-NCDM CD+GF CD+EO CD+DP
Builds on: Item Response Theory NCDM MIRT counterfactual data augmentation for recommendation
Closest papers by meaning
- AD4CD: Causal-Guided Anomaly Detection for Enhancing Cognitive Diagnosis (AAAI 2025 · similarity 0.88)
- Enhancing Cognitive Diagnosis Using Un-interacted Exercises: A Collaboration-Aware Mixed Sampling Approach (AAAI 2024 · similarity 0.87)
- Zero-1-to-3: Domain-Level Zero-Shot Cognitive Diagnosis via One Batch of Early-Bird Students towards Three Diagnostic Objectives (AAAI 2024 · similarity 0.87)
- Knowledge Is Power: Harnessing Large Language Models for Enhanced Cognitive Diagnosis (AAAI 2025 · similarity 0.87)
- Multi-Perspective Consolidation Enhanced Cognitive Diagnosis via Conditional Diffusion Model (AAAI 2025 · similarity 0.86)
- Symbolic Cognitive Diagnosis via Hybrid Optimization for Intelligent Education Systems (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.