Multi-hop clinical reasoning QA dataset grounded in an expanded UMLS-based biomedical knowledge graph (45,746 nodes, 702,290 directed edges). Generated via Gemini 2.5 Flash with DeepSeek-V3 validation, as described in:
One Hop at a Time: Knowledge Graph-Grounded Curriculum Learning for Multi-Hop Clinical Reasoning with Bottom-Up Superintelligence
Isha Yerramilli-Rao, Princeton University ECE, 2026. Advised by Prof. Niraj K. Jha.… See the full description on the dataset page:
https://huggingface.co/datasets/jha-lab/med-si-dataset.
