UniKeyphrase

• code for the ACL 2021 findings paper "UniKeyphrase: A Unified Extraction and Generation Framework for Keyphrase Prediction"
• preprint paper: arxiv
• video presentation

Updates

• Update dataset preprocess and evaluate scripts. The datasets can be found in https://github.com/memray/OpenNMT-kpg-release, we also give three test sets in dataset.zip
• Update v2 version of the paper, no model code changed. See v2 for the new version, v1 for the older version
  • fix a improper tokenization on the datasets which may lead to high results on present F1@M and low results on absent F1@5 & F1@M
  • update some results of baseline(SEG-NET) from arxiv version to the newest ACL version
  • following previous work, pad the result when calculating F1@5
  • provide more detailed ablation study(layer and module)
  • update table of "average numbers of predict keyphrases". UniKeyphrase now predicts more accurately after fixing the tokenization problem
  • update case study
• Update train and test scripts, see the sciprts/ folder.

Environment:

• prepare for APEX

    . .bashrc
    apt-get update
    apt-get install -y vim wget ssh

    PWD_DIR=$(pwd)
    cd $(mktemp -d)
    git clone -q https://github.com/NVIDIA/apex.git
    cd apex
    git reset --hard 1603407bf49c7fc3da74fceb6a6c7b47fece2ef8
    python setup.py install --user --cuda_ext --cpp_ext
    cd $PWD_DIR

• other packages

    pip install --user tensorboardX six numpy tqdm path.py pandas scikit-learn lmdb pyarrow py-lz4framed methodtools py-rouge pyrouge nltk
    python -c "import nltk; nltk.download('punkt')"
    pip install -e git://github.com/Maluuba/nlg-eval.git#egg=nlg-eval

• get pretrained models from: https://unilm.blob.core.windows.net/ckpt/unilm1-base-cased.bin

Run

• see scripts