OpenPI: Open-Domain Procedural Inference

OpenPI dataset for tracking entities in open domain procedural text (EMNLP 2020).

For OpenPI2.0, go to v2.0.

Paper: https://www.aclweb.org/anthology/2020.emnlp-main.520.pdf

Project page: https://allenai.org/data/openpi

Dataset

OpenPI Dataset files are available in JSON format under openpi-dataset/data/gold/ . There are four files:

• id_question.jsonl: each line is a json with an id, and the input sentence and its past sentences i.e., "x"
• id_question_metadata.jsonl: the metadata corresponding to the question such as topic. Each line is a json with an id, and the metadata
• id_answers_metadata.jsonl: each line is a json with an id, and the a list of answers i.e., "y"
• id_answers.jsonl: the metadata corresponding to the answer. Each line is a json with an id, and the metadata (such as entity, attribute, before value, after value).

Training

You can modify the hyperparameters in this script to train the model.

sh scripts/training_bash.sh

Run Prediction

To run predictions on a single file:

python src/model/generation.py \
      --model_path /path/to/trained_model \
      --test_input_file /path/to/input_file \
      --unformatted_outpath /path/to/store/unformatted_predictions \
      --formatted_outpath /path/to/store/formatted/predictions \
      --max_len max_len_say_200

To run predictions on multiple files, you can use this bash script:

sh scripts/predictions_bash.sh

Run Evaluation

python eval/simple_eval.py 
    -g data/gold/test/id_answers.jsonl
    -p /path/to/formatted/predictions 
    --quiet

(no diagnostics file generated when using --quiet)

To run evaluation on multiple files, you can use this bash script:

sh scripts/evaluations_bash.sh

Hyperparameters

To match the results published in the paper, please use the following hyperparameters. https://github.com/allenai/openpi-dataset/blob/main/hyperparams.md

Leaderboard

coming soon... (we are now working on openpi v2 which clusters

Citation

If you use this dataset in your work, please cite:

@inproceedings{tandon-etal-2020-dataset,
    title = "A Dataset for Tracking Entities in Open Domain Procedural Text",
    author = "Tandon, Niket  and
      Sakaguchi, Keisuke  and
      Dalvi, Bhavana  and
      Rajagopal, Dheeraj  and
      Clark, Peter  and
      Guerquin, Michal  and
      Richardson, Kyle  and
      Hovy, Eduard",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.emnlp-main.520",
    doi = "10.18653/v1/2020.emnlp-main.520",
    pages = "6408--6417"
}