/ACL2019-ODEE

This is the code for our ACL 2019 paper "Open Domain Event Extraction Using Neural Latent Variable Models"

Primary LanguagePython

Open Domain Event Extraction Using Neural Latent Variable Models (ODEE)

This is the python3 code for the paper "Open Domain Event Extraction Using Neural Latent Variable Models" in ACL 2019.

Prepare ELMo model

Modify the Line 24 and 25 in cache_features.py.

The fine-tune process need 2 * GTX 1080Ti, if the fine-tune process is costly or somehow failed to complete, please use the initial parameters in allennlp.

Please note that it is optional to finetune the ELMo model if you just want to complete the whole procedure or use the model in somewhere else.

Prepare Data and Train Model

The data is HERE.

  1. run the dataprocessor
  2. run sudo chown [YOUR_UERS] [PROCESSED_DIR] and specify the directories in setting.yaml manually
  3. run pip install -r requirements.txt to install required packages
  4. run python cache_features.py
  5. run python train_avitm.py
  6. run python generate_slot_topN.py
  7. run python decode.py
  8. run cd slotcoherence && ./run-oc.sh
  9. run visualize_test.ipynb

Produced Data

  1. *.json.pt: cached features of ODEE input
  2. *.json.answer: decoded full results of a news group
  3. *.json.template: decoded template of a news group
  4. *.json.events.topN: decoded top-N events of a news group
  5. *.json.labeled: labeled events of test split
  6. slotcoherence/slot_head_words.txt: generated topN head words for each slot

Cite

Please cite our ACL 2019 paper:

@proceedings{liu2019open,
    title={Open Domain Event Extraction Using Neural Latent Variable Models},
    author={Liu, Xiao and Huang, Heyan and Zhang, Yue},
    booktitle={Proceedings of Annual Meeting of the Association for Computational Linguistics},
    year={2019},
    publisher={Association for Computational Linguistics}
}