Table 3 Effects of fine-tuning word embeddings in LSTM-RNN and BLSTM-RNN
From: Long short-term memory RNN for biomedical named entity recognition
Systems | Dim. | GM | JNLPBA | ||
|---|---|---|---|---|---|
LSTM-RNN | +tune | -tune | +tune | -tune | |
+SENNA | 50 | 85.69 | 82.13 | 70.56 | 69.92 |
+Biomedical | 50 | 85.33 | 84.96 | 71.78 | 72.69 |
300 | 85.65 | 83.35 | 71.13 | 70.83 | |
+Biomedical | 300 | 84.56 | 86.12 | 72.04 | 72.58 |
+Random | 300 | 84.74 | 79.94 | 71.10 | 69.13 |
BLSTM-RNN | +tune | -tune | +tune | -tune | |
+SENNA | 50 | 86.81 | 82.00 | 72.09 | 69.22 |
+Biomedical | 50 | 85.24 | 85.43 | 72.28 | 72.71 |
300 | 86.52 | 83.50 | 73.03 | 71.19 | |
+Biomedical | 300 | 84.53 | 86.55 | 73.44 | 73.79 |
+Random | 300 | 84.94 | 80.18 | 71.81 | 69.68 |