OntologyAligner uses a three-stage framework of aligned retrieval, LLM reranking, and hierarchy-guided refinement to map free-text to biomedical concepts. It achieved 88.78% Macro Top-1 Accuracy on the new PhenoNormBench, outperforming existing baselines by nearly 5%.
HOW THIS AFFECTS YOU
●
builderYou can implement this framework to improve the accuracy of biomedical entity extraction in your pipelines.
●
healthThis improves the reliability of integrating diverse clinical datasets through standardized terminology.