
DMCD: LLM-Powered Causal Discovery
DMCD is a two-phase causal discovery framework that combines LLM-based semantic drafting from variable metadata with statistical validation on observational data. In Phase I, an LLM generates a sparse draft DAG as a semantic prior; Phase II refines it using conditional independence tests. It excels on real-world benchmarks in engineering, environment, and IT, with strong gains in recall and F1.





