Skip to main content

Table 1 Main characteristics of our target and source corpora

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

  Corpus name Subcorpus Train Size Test Size #Entity Types #Relation Types
    sent. rel. sent. rel.   
Target SNPPhenA 362 935 121 365 2 3
  EU-ADR drug-disease 244 176    4 3
   drug-target 247 310 4 3
   target-disease 355 262    4 3
Source SemEval DrugBank 5,675 3,805 973 889 4 4
  2013 DDI MEDLINE 1,301 232 326 95 4 4
  ADE-EXT 5,939 6,701 2 1
  reACE 5,984 2,486 4 5
  1. Two corpora are divided into subcorpora. The sizes of the training and test corpora are reported in term of number of sentences (sent.) and annotated relationships (rel.). EU-ADR, ADR-EXT and reACE have no proper test corpus