Research map / Learning theory and algorithms / Prediction and decision losses
A Spectral Approach to Item Response Theory
Duc Nguyen, Anderson Ye Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use CMLE instead of the spectral/MMLE/JMLE estimators on large rating data with missing entries | CMLE · BX | held-out AUC: 0.57 -> 0.503 | Hurt |
| replace MMLE with the spectral method on BX | Spectral · BX | held-out AUC: 0.577 -> 0.546 | Hurt |
| use spectral method instead of MMLE for top-K ranking on ML-1M | Spectral · ML-1M | top-10 accuracy: 0.6 -> 0.8 | Improved |
| use spectral method instead of JMLE for top-K ranking on Book-Genome | Spectral · Book-Genome | top-10 accuracy: 0.2 -> 0.6 | Improved |
| spectral method versus MMLE/JMLE on large rating data, AUC and log-likelihood | Spectral · Book-Genome | held-out AUC: 0.665 vs 0.658 | No clear effect |
| use spectral method instead of MMLE for inference time on ML-20M | Spectral · ML-20M | total inference time (s): 5.36k -> 1.58k | Improved |
| use spectral method instead of CMLE for inference time on BX | Spectral · BX | total inference time (s): 156k -> 205 | Improved |
| use spectral method on small, complete education datasets (LSAT, UCI, 3 GRADES) | Spectral · LSAT | held-out AUC: 0.707 vs 0.707 | No clear effect |
| misspecify MMLE prior (sigma=2 instead of 1) in synthetic data | MMLE · Synthetic Rasch data | l2 error of beta | Hurt |
| small number of items m relative to n | JMLE · Synthetic Rasch data | l2 error of beta | Hurt |
| moderately sized dataset with missing data | CMLE · Synthetic Rasch data | l2 error of beta | Hurt |
| use accelerated (modified Markov chain) version instead of original spectral algorithm (ablation) | Spectral · Real-life datasets | convergence speed / numerical issues | Improved |
Closest papers by meaning
- An Efficient Joint Learning Approach for Item Response Theory (ICML 2026 · similarity 0.92)
- An Estimation and Analysis Framework for the Rasch Model (ICML 2018 · similarity 0.91)
- Novel Spectral Algorithms for the Partial Credit Model (ICML 2024 · similarity 0.84)
- Item Response Scaling Laws: A Measurement Theory Approach for Efficient and Generalizable Neural Scaling Estimation (ICML 2026 · similarity 0.83)
- Ada-Retrieval: An Adaptive Multi-Round Retrieval Paradigm for Sequential Recommendations (AAAI 2024 · similarity 0.82)
- Can We Trust Item Response Theory for AI Evaluation? (NeurIPS 2026 · similarity 0.82)
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.
