Treasures Outside Contexts: Improving Event Detection via Global Statistics

Preprocessing (for ACE data)

Generate train.json, dev.json and test.json according to data_list.csv. Read ace2005-preprocessing for details.

Preprocessing (for all datasets)

Move train.json, dev.json and test.json to ACE / KBP14 / KBP15 dir.

Note: word_class_label.txt is deprecated, although runner.py reads it.

Requirements

See requirements.txt.

SSJDN Code

Train the model:

bash scripts/train.sh

The result can be seen in models/*/log.

For more details, see enet.