Skip to main content

Table 4 The effect of the different ratios of positive and negative documents

From: A neural network approach to chemical and gene/protein entity recognition in patents

Ratio (positive:negative) CEMP Dev GPRO Dev
Precision Recall F-score Precision Recall F-score
1:0 87.58 92.20 89.83 60.90 88.27 72.07
1:0.5 66.06 85.76 74.63
1:1 67.97 86.06 75.95
1:2 70.03 77.79 73.71
All training set 87.58 92.50 89.97 68.32 82.44 74.72
  1. On the CEMP corpus, only the ratio (1:0) and all training set were tested since the number of positive documents is more than the number of negative documents
  2. Italic values denote the highest values