norakassner / bert-knn

This repository contains code for "BERT-kNN: Adding a kNN Search Component to Pretrained Language Models for Better QA". The repository is forked from https://github.com/facebookresearch/LAMA and adapted accordingly.

Setup

1. Create conda environment and install requirements

(optional) It might be a good idea to use a separate conda environment. It can be created by running:

conda create -n bert_knn -y python=3.7 && conda activate bert_knn
pip install -r requirements.txt

export PYTHONPATH=${PYTHONPATH}:/path-to-project

2. Download the data

Download the LAMA data from:

wget https://dl.fbaipublicfiles.com/LAMA/data.zip
unzip data.zip
rm data.zip

Pre-processing

1. get Wikipedia corpus by DrQA

git clone https://github.com/facebookresearch/DrQA.git

2. preprocess wikipedia

python batch_wikipdia.py --path_db_wirkipedia_drqa "DrAQ/data/wikipedia/docs.db"

3. context embeddings

python encode_context.py --dump 0

(write a shell script in order to run multiple dumps in parallel)

4. database target words

python db_target_creator.py

Run BERT-kNN

python scripts/main.py

Eval BERT-kNN

python scripts/eval.py

Citation

If you use this code, please cite:

@inproceedings{kassner-schutze-2020-bert,
    title = "{BERT}-k{NN}: Adding a k{NN} Search Component to Pretrained Language Models for Better {QA}",
    author = {Kassner, Nora  and
      Sch{\"u}tze, Hinrich},
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2020",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.findings-emnlp.307",
    doi = "10.18653/v1/2020.findings-emnlp.307",
    pages = "3424--3430",
    abstract = "Khandelwal et al. (2020) use a k-nearest-neighbor (kNN) component to improve language model performance. We show that this idea is beneficial for open-domain question answering (QA). To improve the recall of facts encount>
}

@inproceedings{petroni2019language,
  title={Language Models as Knowledge Bases?},
  author={F. Petroni, T. Rockt{\"{a}}schel, A. H. Miller, P. Lewis, A. Bakhtin, Y. Wu and S. Riedel},
  booktitle={In: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2019},
  year={2019}
}

@inproceedings{chen2017reading,
  title={Reading {Wikipedia} to Answer Open-Domain Questions},
  author={Chen, Danqi and Fisch, Adam and Weston, Jason and Bordes, Antoine},
  booktitle={Association for Computational Linguistics (ACL)},
  year={2017}
}

bert-knn's Issues

No module named 'bert_knn.options'

hello, there maybe something wrong with your code

Traceback (most recent call last):
  File "main.py", line 1, in <module>
    from interpolated import main as run_evaluation
  File "/home/event-extraction/BERT-kNN/scripts/interpolated.py", line 13, in <module>
    import bert_knn.options as options
ModuleNotFoundError: No module named 'bert_knn.options'

can you provide this file, thanks

bert_knn.options

where is bert_knn.options?