Subsumption Matching Results for SNOMED-NCIT (Pharm)
⇕System ⇕Use Training Maps ⇕MRR ⇕Hits@1 ⇕Hits@5 ⇕Hits@10
BERTSubs (IC) 0.4360.2350.7120.908
BERTSubs (IC) 0.3780.1710.6550.877
OWL2Vec* + RF 0.448 0.2550.6990.886
SORBETMtch 0.760 0.659 0.880 0.912
Word2Vec + RF 0.355 0.1790.5510.793
Word2Vec + RF (del) 0.276 0.1440.3690.563

Note: (del) is a trial setting that uses the pruned target ontology similar to OAEI 2022.

Note: OWL2Vec* uses structure, literal and mixture documents, without using pre-trained Word2Vec. RF indicates Random Forest Classifier with n_estimators=50.