Table 2 Hyperparameters used in our experiments
From: Family history information extraction via deep joint learning
Hyperparameters | Value |
|---|---|
Dimension of word embeddings | 50 |
Dimension of POS embeddings | 20 |
Dimension of label embeddings | 10 |
Number of LSTM hidden states | 100 |
Optimizer | SGD |
Learning rate | 0.005 |
Dropout rate in entity recognition | 0.5 |
Dropout rate in relation extraction | 0.3 |
Epoch number | 20/25 |
Combination coefficient (α) | 0.4/0.5/0.6 |