Skip to main content

Table 2 Details of the 3-layer network architecture for the PICO recognition model

From: Improving reference prioritisation with PICO recognition

1aWord embedding200[21], not updated
1bCharacter embedding28trained from random initialisation
1cCharacter-based word representation2 ×28biLSTM applied to 1b
1dCombined embedding256concatenation of 1a and 1c
2Recurrent layer2 ×128biLSTM over 1d
3Linear layer41affine projection of 2
 CRF output1most likely sequence of tags