Table 2.

The pretrained word embeddings used in the study

Name	Corpus	Dimension	Vocab. size
GloVea	Wikipedia and English Gigaword	200	400 000
fastTextb	Wikipedia	300	2 519  370
nlplabc	PubMed and PMC	200	2 231  684
word2vecGN	Google News	300	3 000  000
Numberbatchd	Hybrid of ConceptNet, word2vecGN and GloVe	300	417 194
BioWordVece	PubMed and MIMIC-III	200	16 545  451
word2vecMIMIC	MIMIC-III	300	320 313
ConcatenatedVec	Hybrid of GloVe, fastText, and word2vecMIMIC	700	228 763
AddedVec	Hybrid of fastText and word2vecMIMIC	300	46 404
PurifiedVec	Postprocessed GloVe vectors	200	400 000

MIMIC-III: Medical Information Mart for Intensive Care III; PMC: PubMed Central.

^a https://nlp.stanford.edu/projects/glove/.

^b https://github.com/facebookresearch/fastText/blob/master/pretrained-vectors.md.

^cThis is the embedding we used during the n2c2 ADME track, which is available at http://evexdb.org/pmresources/vec-space-models/.

^d https://github.com/commonsense/conceptnet-numberbatch.

^e https://github.com/ncbi-nlp/BioSentVec#biowordvec-biomedical-word-embeddings-with-fasttext.