Skip to main content

Table 3 Final evaluation results on the CHEMDNER test set.

From: A document processing pipeline for annotating chemical entities in scientific documents

  Entity Mention Document Indexing
System Precision Recall F-measure Precision Recall F-measure
Top scoring 89.09 85.75 87.39 87.02 89.41 88.20
Official best run 86.50 85.66 86.08 86.35 82.37 84.31
Corrected run 87.35 86.49 86.92 87.07 87.97 87.52
  +0.85 +0.83 +0.84 +0.72 +5.60 +3.21
1st order CRF 88.04 84.89 86.44 88.00 86.42 87.20
  +0.69 -1.60 -0.48 +0.93 -1.55 -0.31
2nd order CRF 88.35 83.79 86.01 88.14 86.65 87.39
  +1.01 -2.71 -0.91 +1.08 -1.32 -0.13
Post-processing 88.67 86.32 87.48 87.68 87.81 87.75
  +1.32 -0.17 +0.56 +0.61 -0.16 +0.23
  1. The 'Corrected run' line shows results obtained using the same models as in the official run, after correcting the generation of the annotation files. Results obtained with first- and second-order CRF models alone (without model combination) and after post-processing are shown with differences compared to the corrected run. Values are shown in percentage points.