Table 2.
Major parameter settings in ML-Net for the 3 tasks
| Parameter | Setting |
||
|---|---|---|---|
| Task 1 | Task 2 | Task 3 | |
| Maximum permitted labels | 5 | 8 | 70 |
| Maximum tokens in documents | All tokens | First 1500 tokens | |
| Neurons in the MLP | 128, 128, 64 | 7024, 7024, 128 | |
| Batch size | 32 | 16 | |
| Training epochs (label prediction network) | 50 | 30 | |
| RNN unit (dimension) | LSTM (50) | ||
| Attention layer dimension | 50 | ||
| Dropout rate | 0.5 | ||
| Optimizer (learning rate) | Adam (0.001) | ||
Abbreviations: LSTM, long short-term memory; MLP, multi-layer perceptron; RNN, recurrent neural networks;