Hugging Face's logo

Datasets:
tasksource
/
abstrct-spans

Modalities:
Text
Formats:
parquet
Languages:
English
Size:
1K - 10K
Tags:
tasksource
extraction
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
abstrct-spans / README.md
sileod's picture
sileod
Publish AbstRCT argument-component span mirror
c4b6010 verified
preview code
|
Raw History Blame Contribute Delete
3.37 kB
metadata 
license: cc-by-nc-sa-4.0
language:
  - en
tags:
  - tasksource
  - extraction
configs:
  - config_name: neoplasm
    data_files:
      - split: train
        path: train.parquet
      - split: validation
        path: validation.parquet
      - split: test
        path: test.parquet
  - config_name: glaucoma
    data_files:
      - split: test
        path: glaucoma/test.parquet
  - config_name: mixed
    data_files:
      - split: test
        path: mixed/test.parquet

AbstRCT argument-component spans

Major claim, claim and premise argument-component spans in MEDLINE randomized-controlled-trial (RCT) abstracts (Mayer, Cabrio and Villata, ECAI 2020), expert-annotated with the brat tool (two annotators, adjudicated). Every context is the full abstract, not an isolated sentence: whether a sentence is a claim or a premise is judged against the rest of the abstract, which a per-sentence view would strip away.

Data-only mirror of the original brat distribution (revision f856f1ca7514caa4094194b5623e84c159d9bf1d, archive SHA256 32150598d2e9). Reproduce with scripts/repackage_dataset/abstrct_spans.py; provenance.json records the archive hash, revision, native abstract counts and settings. License: CC BY-NC-SA 4.0 (evidence).

Schema: context (the abstract), question, answers (text, answer_start: every span, not alternatives) and JSON-string metadata (field is one of major_claim, claim, premise; group is {config}:{PMID}). Every field is always asked; an abstract with none of a kind answers with null. Used by the extraction config of tasksource-jev-typed-decisions (neoplasm config, as abstrct/neoplasm).

neoplasm has the native train/dev/test split used for training. glaucoma and mixed are test-only cross-domain transfer sets (no training abstracts) and are evaluation-only.

Citation

Please cite the original dataset:

@inproceedings{mayer2020ecai,
  author    = {Tobias Mayer and Elena Cabrio and Serena Villata},
  title     = {Transformer-Based Argument Mining for Healthcare Applications},
  booktitle = {{ECAI} 2020 - 24th European Conference on Artificial Intelligence},
  series    = {Frontiers in Artificial Intelligence and Applications},
  volume    = {325},
  pages     = {2108--2115},
  publisher = {{IOS} Press},
  year      = {2020},
}

Citing tasksource

If you use tasksource, its standardized task annotations, or datasets derived from them in published research, please cite the tasksource paper (Sileo, LREC-COLING 2024).

@inproceedings{sileo-2024-tasksource,
    title = "tasksource: A Large Collection of {NLP} tasks with a Structured Dataset Preprocessing Framework",
    author = "Sileo, Damien",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.1361",
    doi = "10.63317/2h4pkr685mkq",
    pages = "15655--15684",
}