From: Readers’ affect: predicting and understanding readers’ emotions with deep learning
Parameters | Kim’s CNN | GRU | LSTM | Bi-LSTM |
---|---|---|---|---|
Filter size | 3, 4 and 5 | – | – | – |
Number of filters | 100 | – | – | – |
Number of RNN Stack | – | 1 | 1 | 1 |
Neurons in Stack | – | 100 | 100 | 100 |
Embedding | Pre-trained GloVe | Pre-trained GloVe | Pre-trained GloVe | Pre-trained GloVe |
Embedding dimension | 100 | 100 | 100 | 100 |
Regularizer | l2(0.01) | l2(0.01) | l2(0.001) | l2(0.001) |
Dropout | 0.5 | 0.25 | 0.5 | 0.5 |
Loss | MSE | MSE | MSE | MSE |
Optimiser | Adam | Adam | Adam | Adam |
Learning rate | 0.0005 | 0.005 | 0.0005 | 0.0005 |
Dense layer activation | softmax | softmax | softmax | softmax |
Batch size | 64 | 64 | 128 | 128 |
Epoch | 100 | 100 | 100 | 100 |