Skip to main content

Table 3 Average estimated cosine similarities for sentence pairs included in the negation and antonym subset and a reference set of highly similar sentences per model. Lower values indicate lower estimated semantic similarity; higher values indicate higher estimated semantic similarities

From: Neural sentence embedding models for semantic similarity estimation in the biomedical domain

  Sent2vec Skip-thoughts PV-DM PV-DBOW fastText CBOW fastText skip-gram
Subset of highly similar sentences (n = 11) 0.706 0.899 0.652 0.568 0.938 0.971
Negation subset (n = 13) 0.967 0.999 0.930 0.936 0.945 0.979
Antonym subset (n = 7) 0.983 0.999 0.968 0.960 0.976 0.989
  1. PV-DM Paragraph Vector Distributed Memory, PV-DBOW Paragraph Vector Distributed Bag of Words, CBOW Continuous Bag of Words