Researchers introduce DKCD, a framework for causal discovery from unstructured data in high-expertise domains like healthcare and finance. The method addresses limitations of existing approaches by incorporating specific domain knowledge to improve the identification of latent factors and the accuracy of factor annotation.
- Knowledge Mining retrieves relevant domain information based on observable factors to support reasoning.
- Knowledge-guided Causal Reasoning discovers latent causal factors and generates clues for accurate data annotation.
- Causal Structure Discovery constructs final graphs using a more complete set of factors and annotations.
Experiments on two domain-specific datasets demonstrate that DKCD significantly improves both causal factor identification and the construction of causal graphs.