Research map / Label-efficient and robust learning / Recognition and long-tail
Enhancing Cognitive Diagnosis Using Un-interacted Exercises: A Collaboration-Aware Mixed Sampling Approach
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 CMES, a model-agnostic sampling framework that mines un-interacted exercises for cognitive diagnosis via collaboration-aware clustering, attention-based mixing, and ranking-based pseudo feedback labels.
Problem
Cognitive diagnosis models use only students' answered exercises and ignore informative un-interacted exercises tied to broader knowledge concepts, and lack true response labels for such exercises.
Approach
Cluster students by response ability and collaboration, sample un-interacted exercises from other clusters, mix them with interacted exercises via scaled dot-product attention with concept-weighted embeddings, and assign pseudo labels via a BPR ranking loss on the same CD model.
Key results
- On ASSISTments, CMES-IRT raises AUC from 70.45% to 74.40% and lowers RMSE from 0.4684 to 0.4547 versus original IRT.
- On MATH, CMES-MIRT improves ACC from 72.99% to 74.60% and AUC from 75.31% to 78.04%.
- CMES-NCD trained on 50% of ASSISTments data surpasses original NCD trained on 70% of the data (per the case study claim).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply CMES to RCD | RCD · ASSISTments | ACC: 72.99% -> 73.06% | Helped |
| Apply CMES to RCD | RCD · ASSISTments | RMSE: 0.4243 -> 0.4237 | Helped |
| Apply CMES to RCD | RCD · ASSISTments | AUC: 76.40% -> 76.51% | Helped |
| Apply CMES to RCD | RCD · MATH | ACC: 74.86% -> 75.16% | Helped |
| Apply CMES to RCD | RCD · MATH | RMSE: 0.4063 -> 0.4055 | Helped |
| Apply CMES to RCD | RCD · MATH | AUC: 78.34% -> 78.64% | Helped |
| Replace collaboration-aware mixed sampling with random sampling (RSS) (ablation) | NCD · ASSISTments and MATH | predictive performance | Hurt |
| Vary the sampling number n (ablation) | CMES-MIRT and CMES-NCD · ASSISTments and MATH | predictive performance | Mixed |
| Vary the student cluster number W (ablation) | CMES-NCD · ASSISTments and MATH | predictive performance | Mixed |
Surprising findings
- Random sampling of un-interacted exercises degrades NCD performance, indicating that uninformed extra samples can harm diagnosis.
Limitations
- Sampling strategy is tailored to general CD models and not to CD-specific characteristics, as the authors acknowledge.
- Performance is sensitive to the sampling number n and student cluster number W, which need dataset-specific tuning.
- MATH dataset is proprietary, limiting reproducibility.
Experimental setup
Datasets: ASSISTments MATH
Benchmarks: ASSISTments MATH
Metrics: ACC RMSE AUC
Baselines: IRT MIRT NCD CDGK ECD RCD RSS
Builds on: NCD EIRS BPR
Closest papers by meaning
- Knowledge Is Power: Harnessing Large Language Models for Enhanced Cognitive Diagnosis (AAAI 2025 · similarity 0.90)
- Multi-Perspective Consolidation Enhanced Cognitive Diagnosis via Conditional Diffusion Model (AAAI 2025 · similarity 0.89)
- Towards Accurate and Fair Cognitive Diagnosis via Monotonic Data Augmentation (NeurIPS 2024 · similarity 0.87)
- Collaborative Cognitive Diagnosis with Disentangled Representation Learning for Learner Modeling (NeurIPS 2024 · similarity 0.87)
- Boosting Neural Cognitive Diagnosis with Student’s Affective State Modeling (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 0.86)
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.