KG-SoftMAP: Soft Knowledge-Graph Priors for Bayesian Network Structure Learning from Sparse Discrete Data
Not on the map. Why?
Abstract (arXiv)
Learning Bayesian network (BN) structure from sparse discrete data is hard: when each instance records only a few variables, most variable pairs lack the joint observations needed for reliable scoring, and data-only methods recover little structure. Imperfect domain knowledge, expressible as a weighted directed knowledge graph (KG), is often available. KG-SoftMAP encodes such a KG as a finite-strength, confidence-weighted edge prior and maximizes a MAP objective that adds this logit-form prior to the BDeu score. With an informative but imperfect KG, KG-SoftMAP recovers partial directed structure even at observation rate rho=0.05, with directed F1 (DF1) of 0.19-0.32 across benchmarks. At higher observation rates within this sparse grid, DF1 reaches 0.44-0.66 at rho=0.20 and 0.46-0.64 at rho=0.40. Across the same three rates, KG-SoftMAP without the KG prior averages DF1 0.00, 0.19, and 0.21. Stress tests that corrupt, remove, or blur the KG signal, together with checks on LLM-extracted graphs beyond canonical benchmarks, show that recovery rises and falls with KG quality. On three real sparse educational datasets without ground-truth DAGs, we evaluate prediction, calibration, and KG-consistency. On Short Answer Feedback (SAF), KG-SoftMAP+VE reaches Fail-class F1 0.75 versus 0.78 for logistic regression while also providing an inspectable concept graph, calibrated Fail probabilities, and posterior queries from partially observed concept evidence. The remaining datasets sharpen the operating picture: weak heuristic KG signal leaves prediction unchanged, while an independent expert ontology moves the learned graph toward expert relatedness.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 31 of its 61 references that are in the map. First posted on arXiv 9 Jun 2026.
Builds on
31 of 61 references.
References · 31
- 2018DAGs with NO TEARS: Continuous Optimization for Structure Learningcited by 752
- 2015Deep Knowledge Tracingcited by 332
- 2020On the Role of Sparsity and DAG Constraints for Learning Linear DAGscited by 191
- 2013Learning Bayesian Networks: The Combination of Knowledge and Statistical Datacited by 187
- 2020Context-Aware Attentive Knowledge Tracingcited by 174
- 2019A Self-Attentive model for Knowledge Tracingcited by 172
- 2013Causal Inference and Causal Explanation with Background Knowledgecited by 170
- 2022DAGMA: Learning DAGs via M-matrices and a Log-Determinant Acyclicity Characterizationcited by 140
- 2021DiBS: Differentiable Bayesian Structure Learningcited by 106
- 2020Towards an Appropriate Query, Key, and Value Computation for Knowledge Tracingcited by 81
- 2020Instructions and Guide for Diagnostic Questions: The NeurIPS 2020 Education Challengecited by 64
- 2022Greedy Relaxations of the Sparsest Permutation Algorithmcited by 64
19 more
- 2021DAGs with No Curl: An Efficient DAG Structure Learning Approachcited by 53
- 2022pyKT: A Python Library to Benchmark Deep Learning based Knowledge Tracing Modelscited by 52
- 2023simpleKT: A Simple But Tough-to-Beat Baseline for Knowledge Tracingcited by 47
- 2023Fast Scalable and Accurate Discovery of DAGs Using the Best Order Score Search and Grow-Shrink Treescited by 38
- 2024Causal Inference with Large Language Model: A Surveycited by 32
- 2023Stable Differentiable Causal Discoverycited by 31
- 2013The Bayesian Structural EM Algorithmcited by 26
- 2024Large Language Models for Causal Discovery: Current Landscape and Future Directionscited by 25
- 2024Large Language Models are Effective Priors for Causal Graph Discoverycited by 24
- 2013Learning Bayesian Network Structure from Massive Datasets: The "Sparse Candidate" Algorithmcited by 22
- 2023Self-Compatibility: Evaluating Causal Discovery without Ground Truthcited by 22
- 2023Causal Structure Learning Supervised by Large Language Modelcited by 21
- 2014Bayesian Network Constraint-Based Structure Learning Algorithms: Parallel and Optimised Implementations in the bnlearn R Packagecited by 19
- 2021Greedy structure learning from data that contain systematic missing valuescited by 13
- 2013A Construction of Bayesian Networks from Databases Based on an MDL Principlecited by 5
- 2016Learning Bayesian Networks with Incomplete Data by Augmentationcited by 5
- 2025Bayesian Network Structure Discovery Using Large Language Modelscited by 5
- 2025Realizing LLMs' Causal Potential Requires Science-Grounded, Novel Benchmarkscited by 4
- 2018Efficient Learning of Bounded-Treewidth Bayesian Networks from Complete and Incomplete Data Setscited by 2
Built on by
Citing papers per year: 2026: 1.
1 paper.