PRISM-BN Dataset for Text-to-Bayesian Network Extraction
September 21, 2026
PRISM-BN provides a controlled corpus of 5,054 text descriptions paired with discrete Bayesian Networks. The dataset includes variables, directed edges, and full multi-parent CPDs across five domains to enable training neurosymbolic systems for probabilistic graphical model extraction.
HOW THIS AFFECTS YOU
●
researcherYou can use this benchmark to evaluate models on symbolic reasoning and probabilistic parameter extraction.