Standardized examinations are typically treated as uniform syllabus coverage problems, but a new research paper argues they are better understood as adversarial systems with stable latent cognitive structures that diverge systematically from official syllabi. The paper, titled "LearnOpt: Recovering the Latent Cognitive Structure of Standardized Examinations via Knowledge Graphs and Constrained Optimization" by Bose, Joy, Thomas, and Om, presents LearnOpt—a system that recovers this hidden structure from historical question papers and generates personalized, time-bounded study plans.
How LearnOpt Works
LearnOpt builds an exam knowledge graph from LLM-tagged questions. It then extracts a five-category latent skill distribution and formulates study planning as a knapsack-variant optimization over prerequisite-aware subgraphs, incorporating Bayesian Knowledge Tracing. The system was applied to nine years of NEET questions (2016–2024, n=1,496).
Key Findings from NEET Analysis
The central finding is that NEET's latent skill distribution is stable within a syllabus regime. The consecutive-year KL divergence ranges from 0.004 to 0.032 for 2016–2021, and is non-significant under permutation testing. However, a significant shift occurs with NCERT's 2023 syllabus rationalization: pooling 2016–2021 (n=1,072) vs 2023–2024 (n=392) gives KL=0.040 (p=0.0005), with Elimination/Negation questions rising from ~20–29% to ~31–35%. Latent structure, while not permanently stationary, is piecewise stable, with shifts detectable and attributable to curricular events. Within either regime, subject predicts skill profile more strongly than year.
Comparison with JEE Advanced
Applying the same pipeline to JEE Advanced reveals a different profile. A comparison table highlights the divergence:
| Metric | NEET | JEE Advanced |
|---|---|---|
| Multi-concept Integration | 33.3% | 80.9% |
| KL divergence vs NEET | — | 0.505 |
This KL divergence (0.505) exceeds NEET's largest cross-subject divergence. The researchers conclude that exam tier shapes latent cognitive structure more than subject, which shapes it more than time within a regime.
"Latent structure, while not permanently stationary, is piecewise stable, with shifts detectable and attributable to curricular events."
Optimization Evaluation and Public Release
An optimization evaluation using one real and two synthetic mastery profiles shows that the skill-weighted objective produces a modest but real reordering of recommended topics over a mastery-conditioned frequency baseline. The code, knowledge graph, and annotated dataset are released publicly.
For enterprise technology leaders, LearnOpt demonstrates a powerful combination of knowledge graphs, constrained optimization, and Bayesian Knowledge Tracing that could be adapted to other domains where latent structures underlie observable performance data. The methodology's ability to detect regime shifts and attribute them to specific events (like syllabus changes) offers a template for monitoring complex systems.