HealthCareAI

HealthCareAI for Web Semantics

The steps to reproducing results in our paper

1. (Do not need to run) the dataset is quried from the original knowledge graph (KG) using the jupyter notebook data/query_data.ipynb. We remove the orignial dataset: data/1808_original.csv since we do not have the right to make it publicly avaiable.

2. (Run the code) The jupyter notebook casual/casual_discovery.ipynb is used to simulate KGs (with different settings of patient number N, which is the sample_size in the code) and counterfactuals from dataset queried from the the original KG, and learn causal graph from the simulated KGs. The Horn rules are mined using rule_mining/rule_mining.ipynb. Rules with PCA confident = 1 are used as a part of the domain knowledge. The LLM prompts are presented in casual/casual_discovery.ipynb. The final learned causal graphs for each N (sample_size) are stored in causal/structures_N.pkl.

3. (Run the code) The python file causal_reasoning.py is used to esitmate counterfactuals from the simulated KGs. Specify the parameter sample_size for each patient number N.