UMNSRS, developed by Pakhomov, et al., consists of 725 clinical term pairs whose semantic similarity and relatedness.
The similarity and relatedness of each term pair was annotated based on a continuous scale by having the resident touch
a bar on a touch sensitive computer screen to indicate the degree of similarity or relatedness.
The following subsets are available:
- similarity: A set of 566 UMLS concept pairs manually rated for semantic similarity (e.g. whale-dolphin) using a
continuous response scale.
- relatedness: A set of 588 UMLS concept pairs manually rated for semantic relatedness (e.g. needle-thread) using a
continuous response scale.
- similarity_mod: Modification of the UMNSRS-Similarity dataset to exclude control samples and those pairs that did not
match text in clinical, biomedical and general English corpora. Exact modifications are detailed in the paper (Corpus
Domain Effects on Distributional Semantic Modeling of Medical Terms. Serguei V.S. Pakhomov, Greg Finley, Reed McEwan,
Yan Wang, and Genevieve B. Melton. Bioinformatics. 2016; 32(23):3635-3644). The resulting dataset contains 449 pairs.
- relatedness_mod: Modification of the UMNSRS-Relatedness dataset to exclude control samples and those pairs that did
not match text in clinical, biomedical and general English corpora. Exact modifications are detailed in the paper
(Corpus Domain Effects on Distributional Semantic Modeling of Medical Terms. Serguei V.S. Pakhomov, Greg Finley,
Reed McEwan, Yan Wang, and Genevieve B. Melton. Bioinformatics. 2016; 32(23):3635-3644).
The resulting dataset contains 458 pairs.
