Skip to main content

Table 2 Results of our TL strategy in terms of precision (P), recall (R) and f-measure (F)

From: Syntax-based transfer learning for the task of biomedical relation extraction

Test Corpus Model Train corpus P R F σF
SNPPhenA TreeLSTM SNPPhenA alone 58.9 73.8 65.5 4.1
   + SemEval 2013 DDI 65.2 71.1 68.0 4.7
   + ADE-EXT 62.8 72.1 67.2 3.4
   + reACE 61.8 74.3 67.1 3.6
  MCCNN SNPPhenA alone 55.1 75.0 63.3 4.8
   + SemEval 2013 DDI 55.3 74.4 63.3 4.9
   + ADE-EXT 56.1 73.2 63.2 4.8
   + reACE 53.2 70.9 60.6 4.1
EU-ADR drug-disease TreeLSTM EU-ADR drug-disease alone 74.8 84.1 79.1 12.3
   + SemEval 2013 DDI 74.8 90.6 82.0 13.1
   + ADE-EXT 73.9 88.2 80.4 13.7
   + reACE 74.3 91.1 79.3 14.3
  MCCNN EU-ADR drug-disease alone 73.3 94.7 80.2 14.2
   + SemEval 2013 DDI 72.6 87.9 76.6 14.3
   + ADE-EXT 73.0 85.5 76.0 14.5
   + reACE 74.1 91.5 79.2 13.8
EU-ADR drug-target TreeLSTM EU-ADR drug-target alone 72.4 90.6 80.2 10.9
   + SemEval 2013 DDI 71.9 95.5 82.5 8.5
   + ADE-EXT 70.2 96.7 80.9 9.2
   + reACE 70.4 96.5 80.8 9.3
  MCCNN EU-ADR drug-target alone 74.5 92.3 81.0 9.3
   + SemEval 2013 DDI 74.9 88.8 80.0 10.6
   + ADE-EXT 76.3 87.4 80.3 10.1
   + reACE 73.4 92.1 80.5 7.8
EU-ADR target-disease TreeLSTM EU-ADR target-disease alone 77.0 89.7 82.7 6.4
   + SemEval 2013 DDI 77.4 91.6 83.9 8.2
   + ADE-EXT 77.7 89.5 83.3 6.9
   + reACE 75.9 91.7 83.0 7.7
  MCCNN EU-ADR target-disease alone 76.9 91.8 82.6 7.7
   + SemEval 2013 DDI 77.6 90.6 82.5 7.1
   + ADE-EXT 75.5 87.4 81.8 10.1
   + reACE 77.1 91.2 82.0 6.8
  1. σF is the standard deviation of the f-measure. The + in the column Train corpus indicates that we trained our model using the target corpus plus one additional source corpus. Bold numbers correspond to the best performing models