Skip to main content

Advertisement

Table 1 Performance of the taggers

From: Design, implementation, and operation of a rapid, robust named entity recognition web service

# Documents Tagger: total time (s) PiTagger: total time (s)
Abstract server Patent server Abstract server Patent server
1 0.87 ± 0.32 0.84 ± 0.32 0.75 ± 0.34 0.84 ± 0.24
10 0.98 ± 0.30 0.83 ± 0.26 1.10 ± 0.28 0.87 ± 0.37
100 1.89 ± 0.29 1.48 ± 0.27 2.34 ± 0.33 1.52 ± 0.34
1000 11.31 ± 0.68 6.02 ± 0.37 15.23 ± 0.48 8.89 ± 0.48
5000 52.18 ± 2.76 26.67 ± 1.16 72.73 ± 1.81 40.83 ± 1.01
  1. For small requests the total turnaround time is ~ 1 s. Larger requests take an extra 5–10 s per 1000 abstracts to be processed on Tagger. Notably, most of this time is spent on retrieving the document texts from document identifiers, whereas the actual NER step takes only about 20% of the total time. This is reflected in the fact that the PiTagger, which runs on a Raspberry Pi 3, takes only about 50% longer to process large requests