Skip to main content

Advertisement

Table 1 Results for the disease-treatment corpus.

From: Extraction of semantic biomedical relations from text using conditional random fields

  NER SRE
  Recall Precision F-score Accuracy (Entities given) Accuracy (Entities hidden)
Best GM - - 71.0 91.6 74.9
Multilayer NN - - - 96.6 79.6
cascaded CRF 69.0 75.3 72.0 96.9 79.5
  1. NER and SRE performance based on evaluation scores proposed by [6]. Relation classification accuracy for seven types of relations is shown for two settings: (1) when the entities are given as gold standard and (2) when the entities have to be extracted. The cascaded CRF outperforms the best GM approach and it shows similar performance to the multilayer NN, where the latter approach can not be applied to the NER task, due to the large feature vectors.