Skip to main content

Table 4 Experimental results.

From: Automatic recognition of topic-classified relations between prostate cancer and genes using MEDLINE abstracts

    Baseline with NE filter   RR with NER (feature) RR with NER (filter)
Topic-classified Relations (%) Baseline w/o NER Automatic Manual RR w/o NER Automatic Manual Automatic Manual
Any relation P 81.1 91.8 96.7 90.9 91.5 97.0 92.1 97.1
(3196) R 100.0 97.0 100.0 96.1 96.1 99.6 96.5 99.6
Study description P 26.7 30.2 31.8 66.9 67.5 70.8 67.6 70.6
(1050) R 100.0 97.2 100.0 56.7 57.6 63.0 62.9 62.9
Genetic variation P 7.1 8.1 8.4 79.3 78.6 81.9 79.4 83.1
(278) R 100.0 98.9 100.0 69.1 67.3 70.1 73.6 73.6
Gene expression P 27.1 30.8 32.3 73.4 73.0 76.2 73.5 76.8
(1067) R 100.0 97.4 100.0 61.4 61.4 64.5 63.5 64.9
Epigenetics P 1.3 1.6 1.6 85.7 86.0 85.7 88.1 88.1
(53) R 100.0 100.0 100.0 67.9 69.8 67.9 69.8 69.8
Pharmacology P 9.1 10.3 10.9 65.7 66.1 66.7 63.7 67.2
(360) R 100.0 96.1 100.0 43.1 44.7 45.0 44.4 45.3
Clinical marker P 31.5 35.9 37.5 77.4 77.9 78.2 76.6 78.3
(1240) R 100.0 97.8 100.0 72.3 73.2 74.0 73.6 75.4
  1. Notes) Numbers in the first column: frequency of correct predictions, NER: ML-based NER, RR: ML-based topic-classified relation recognition, Automatic: experiments using ML-based NER results, Manual: experiments using human-generated NER annotation results, P: precision, R: relative recall.