Skip to main content

Table 9 Average speed in sentence pairs per second (sent/s) and CUI pairs per second (CUIs/s) for the evaluation of the UBSM [39] sentence similarity measure combined with three representative ontology-based similarity measures based on MeSH (Nov, 2019) in 30 sentence pairs extracted from the MedSTS [135] sentence similarity dataset, and 1 million sentence pairs extracted from BioC corpus [136]

From: HESML: a real-time semantic measures library for the biomedical domain with a reproducible survey

Pairwise sentence comparison based on MeSH UMLS::Sim (30 pairs) SML (30 pairs) HESML (30 pairs) \({{HESML\, (10^6\, pairs)}}\)
Similarity measure Avg. speed
(sent/s)
Avg. speed
(CUIs/s)
Avg. speed
(sent/s)
Avg. speed
(CUIs/s)
Avg. speed
(sent/s)
Avg. speed
(CUIs/s)
Avg. speed
(sent/s)
Avg. speed
(CUIs/s)
Rada et al. [71] 0.441 36.63 0.126 10.478 2830.189 235000 7982.222 337843.826
AncSPL-Rada
(this work)
2542.373 211101.695 7958.742 336850.041
Lin-Seco [87, 110] 0.782 64.956 2586.207 214741.379 3125 259479.167 8166.185 345629.98
Wu-Palmer\(_{fast}\) [72] 0.181 15.067 3125 259479.167 7892.959 334065.805
  1. We provide the average evaluation in normalized CUI pairs per second to allow a fair and unbiased comparison of the results reported for 30 and 1 million sentence pairs. The dataset with 30 sentence pairs requires 2491 pairwise CUI comparisons, whilst the 1 million sentence pairs dataset requires 42324534 pairwise CUI comparisons. Best performing values are shown in bold. Non-implemented methods (–)