xdwang0726 / KenMeSH

KenMeSH: Knowledge-enhanced End-to-end Biomedical Text Labelling

ACL 2022

Required Packages

• Python 3.7
• numpy==1.11.1
• dgl-gpu==0.6.1
• nltk==3.5
• scikit-learn==0.23.0
• scipy==1.4.1
• sklearn==0.0
• spacy==2.2.2
• tokenizers==0.9.3
• torch==1.6.0
• torchtext==0.6.0
• tqdm==4.60.0
• transformers==3.5.1
• faiss

Usage

Build a graph

python -u build_graph.py --meSH_pair_path MeSH_name_id_mapping_2019.txt --mesh_parent_children_path MeSH_parent_child_mapping_2019 --word2vec_path BioWord2Vec_standard.w2v --graph_type 'GCN' --output output_dir

Get journal name and MeSH term corelations

python -u journal_info.py --data train.json --save journal_info.pkl

Get MeSH mask using the training data

Steps (see details in get_mesh_mask.py):

• get idf vector for each document

python -u get_mesh_mask.py --allMeSH train.json --meSH_pair_path MeSH_name_id_mapping_2019.txt --  save_path_idf idf.json

• get masks using KNN

python -u get_mesh_mask.py --allMeSH train.json --meSH_pair_path MeSH_name_id_mapping_2019.txt --save_path_idf idf.json

• get masks from journal and merge the masks generated from neighbours

python -u get_mesh_mask.py --allMeSH train.json --meSH_pair_path MeSH_name_id_mapping_2019.txt --neigh_path neigh.json --journal_info journal_info.pkl --threshold 0.5 --save_path dataset.json

Training

python -u run_classifier_multigcn.py --title_path pmc_title.pkl --abstract_path pmc_abstract.pkl --label_path pmc_meshLabel.pkl --mask_path mesh_mask.pkl --meSH_pair_path MeSH_name_id_mapping_pmc_2020.txt --word2vec_path BioWord2Vec_standard.w2v --graph gcn_pmc.bin --save-model-path model.pt --batch_sz 32 --model_name 'Full'

Evaluation

python -u run_classifier_multigcn.py --title_path pmc_title.pkl --abstract_path pmc_abstract.pkl --label_path pmc_meshLabel.pkl --mask_path mesh_mask.pkl --meSH_pair_path MeSH_name_id_mapping_pmc_2020.txt --word2vec_path BioWord2Vec_standard.w2v --graph gcn_pmc.bin model model.pt --batch_sz 32 --model_name 'Full'

Citing

If you use KenMeSH in your work, please consider citing our paper：

@inproceedings{wang-etal-2022-kenmesh,
    title = "{K}en{M}e{SH}: Knowledge-enhanced End-to-end Biomedical Text Labelling",
    author = "Wang, Xindi  and
      Mercer, Robert  and
      Rudzicz, Frank",
    booktitle = "Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = may,
    year = "2022",
    address = "Dublin, Ireland",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.acl-long.210",
    pages = "2941--2951",
    abstract = "Currently, Medical Subject Headings (MeSH) are manually assigned to every biomedical article published and subsequently recorded in the PubMed database to facilitate retrieving relevant information. With the rapid growth of the PubMed database, large-scale biomedical document indexing becomes increasingly important. MeSH indexing is a challenging task for machine learning, as it needs to assign multiple labels to each article from an extremely large hierachically organized collection. To address this challenge, we propose KenMeSH, an end-to-end model that combines new text features and a dynamic knowledge-enhanced mask attention that integrates document features with MeSH label hierarchy and journal correlation features to index MeSH terms. Experimental results show the proposed method achieves state-of-the-art performance on a number of measures.",
}