Research map / Label-efficient and robust learning / Knowledge distillation
Boosting Neural Cognitive Diagnosis with Student’s Affective State Modeling
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 Affect-aware Cognitive Diagnosis (ACD), which predicts students' affective states from student affect traits and exercise difficulty, uses them to estimate personalized guessing and slipping in a DINA-style formula on top of neural cognitive diagnosis, and extends to an unsupervised contrastive variant (CACD).
Problem
Cognitive diagnosis modeling infers student proficiency from response logs, but existing interaction functions ignore students' affective states (e.g., boredom, confusion, concentration) that correlate with performance.
Approach
An affect perception module predicts student affect from a latent affect trait and exercise difficulty; the predicted affect drives personalized guessing and slipping parameters inserted into the DINA formula over a neural CDM backbone, with a contrastive variant when affect labels are missing.
Key results
- On ASSIST17, NCD with ACD improves ACC from 69.21 to 72.42 and AUC from 75.34 to 79.49 over the NCD baseline.
- On ASSIST12, NCD-based ACD reaches AUC 76.02 vs 74.78 for NCD; RCD-based ACD reaches AUC 76.19 vs 75.22.
- On Junyi (no affect labels), CACD with NCD improves AUC from 79.09 to 82.73 and ACC from 74.43 to 77.35.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add affect perception to NCD | NCD · ASSIST17 | AUC: 75.34±1.01 -> 79.49±0.10 | Helped |
| add affect perception to NCD | NCD · ASSIST12 | AUC: 74.78±0.04 -> 76.02±0.16 | Helped |
| add contrastive affect perception to NCD | NCD · Junyi | AUC: 79.09 -> 82.73 | Helped |
| use only student affect traits for affect perception, rather than both student traits and exercise difficulty (ablation) | ACD on ASSIST17 · ASSIST17 | ACC: 72.42 -> 63.54 | Hurt |
| use only exercise difficulty for affect perception, rather than both student traits and exercise difficulty (ablation) | ACD on ASSIST17 · ASSIST17 | ACC: 72.42 -> 65.51 | Hurt |
| remove the affect prediction loss (ablation) | ACD on ASSIST17 · ASSIST17 | ACC: 72.42 -> 71.54 | Hurt |
| replace supervised affect perception with contrastive affect perception (ablation) | ACD on ASSIST17 · ASSIST17 | ACC: 72.42 -> 71.76 | Hurt |
| increase the affect-loss trade-off parameter λ (ablation) | ACD · ASSIST17 | ACC | Mixed |
Surprising findings
- Modeling only exercise difficulty for affect (DINA-like) performed better than using only student affect traits, suggesting different students may show similar affect on exercises of the same difficulty.
- Graph-based RCD and SCD gain smaller improvements from ACD, which the authors attribute to GNN smoothing weakening personalized feature discrimination.
Limitations
- Affect labels are required for the supervised variant; unsupervised CACD relies on the assumption that positive response-correlated pairs share affective state.
- Predicted affect deviates from ground truth, and responses are influenced by combinations of affects and ability rather than a single affect.
- Evaluated only on three datasets with NCD/RCD/SCD-style backbones.
Experimental setup
Datasets: ASSIST17 ASSIST12 Junyi
Benchmarks: ASSIST17 ASSIST12 Junyi
Metrics: ACC RMSE AUC MAE
Baselines: DINA IRT MIRT NCD RCD SCD
Builds on: NCD DINA RCD FuzzyCDM IRT
Closest papers by meaning
- Knowledge Is Power: Harnessing Large Language Models for Enhanced Cognitive Diagnosis (AAAI 2025 · similarity 0.87)
- Enhancing Cognitive Diagnosis Using Un-interacted Exercises: A Collaboration-Aware Mixed Sampling Approach (AAAI 2024 · similarity 0.87)
- Multi-Perspective Consolidation Enhanced Cognitive Diagnosis via Conditional Diffusion Model (AAAI 2025 · similarity 0.87)
- AD4CD: Causal-Guided Anomaly Detection for Enhancing Cognitive Diagnosis (AAAI 2025 · 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)
- Symbolic Cognitive Diagnosis via Hybrid Optimization for Intelligent Education Systems (AAAI 2024 · similarity 0.85)
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.