CogEvolution performs comparably to the KT-based PEERS model on mastery prediction (AUC 0.80 vs 0.82) while achieving higher mistake precision (76.8%) on CogMath-948¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 2 of 2 entries) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? CogEvolution performs comparably to the knowledge tracing-based PEERS model in mastery perception precision (AUC 0.80 vs 0.82), and both significantly outperform static persona models. → Wei Zhang 2026 q2 i? CogEvolution achieves higher Mistake Precision (76.8%) than PEERS (64.5%) and static agents, reproducing typical misconceptions consistent with real student cognitive gaps. → Wei Zhang 2026 (2)

Evidence¶

Wei Zhang 2026¶

Wei Zhang, Yihang Cheng, Zhirong Ye & Kezhen Huang. (2026). CogEvolution: A Human-like Generative Educational Agent to Simulate Student's Cognitive Evolution. https://arxiv.org/abs/2604.14786 doi:10.48550/arxiv.2604.14786

q2 · i? · design · r2

Comparative evaluation on the CogMath-948 dataset (Table 2) reports AUC and RMSE for CogEvolution versus static LLM agents and the PEERS (BKT+LLM) baseline; the article states "performs comparably to the knowledge tracing-based PEERS model (AUC: 0.80 vs 0.82)". No effect size is printed.

"In terms of mastery perception precision (AUC/RMSE), Co-gEvolution performs comparably to the knowledge tracing-based PEERS model (AUC: 0.80 vs 0.82), and both signifi-cantly outperform static persona models."

Wei Zhang 2026 (2)¶

Wei Zhang, Yihang Cheng, Zhirong Ye & Kezhen Huang. (2026). CogEvolution: A Human-like Generative Educational Agent to Simulate Student's Cognitive Evolution. https://arxiv.org/abs/2604.14786

q2 · i? · design · r2

Task-performance analysis for RQ1 on CogMath-948 reports Mistake Precision of 76.8% for CogEvolution, versus 64.5% for PEERS and 55.4% for GPT-4o in Table 2; the article attributes this to ICAP perception and memory retrieval reproducing "typical Misconceptions consistent with real student cognitive gaps". No effect size is printed.

"More importantly, CogEvolution achieves a sig-nificantadvantageinMistakePrecision(76.8%)."

Discussion¶

Related Claims¶

• Ablations show each CogEvolution module contributes: removing ICAP perception, structured retrieval, or evolutionary update degrades mistake precision, learning-curve fit, and alignment — related
• CogEvolution's simulated learning trajectory fits the human power-law trajectory far better than baselines (R²LC 0.92 vs 0.78 PEERS, 0.45 static agents) — related
• Mined misconception labels match personas' assigned misconceptions at F1 ≈ 0.56, a score that does not test whether generated questions elicit the named misconception — related
• The simulated learner tracks a real-student KT model with calibration error 0.049, and the helpfulness rubric transfers to real classroom transcripts — related
• Knowledge-graph prototype mapping outperforms existing behavior-prediction baselines for student simulation — related