DeepRead/kleister-charity-dev0

Kleister-Charity, dev-0 split

The 440 documents and labels of the dev-0 split of Kleister-Charity (Stanisławek et al., 2021): annual reports filed with the UK Charity Commission, with eight fields each — charity name and number, the street, town and postcode of the charity's own address, the report date, annual income and spending. This copy exists so the benchmark can be downloaded in one step; nothing in the original data was changed.

File	What	Origin
pdfs/<name>.pdf	the 440 documents	git-annex objects of the Kleister repo at commit 2309f486 (Applica's public S3 bucket)
documents.txt	the 440 file names in the official order	first column of dev-0/in.tsv.xz
expected.tsv	the official labels, one line per document, field=value tokens	dev-0/expected.tsv at the same commit, unchanged
expected.corrected.tsv	the same labels with DeepRead's errata applied	this dataset's maintainers
label_fixes.csv	every label changed: old, new (empty = removed), confidence, reason	this dataset's maintainers
schema.json	the JSON Schema sent to every system in DeepRead's benchmark	this dataset's maintainers

test-A labels were never released and the official scoring server (gonito.net) is offline, so dev-0 is the largest split anyone can score.

Benchmark

How the documents are scored (the benchmark's own F1 over upper-cased field=value labels), how several document AI APIs compare on them, and the code to run it with your own API keys: https://github.com/deepread-tech/deepread-public-benchmarks

pip install git+https://github.com/deepread-tech/deepread-public-benchmarks
drbench fetch charity          # downloads this dataset

Licence and provenance

The Kleister-Charity repository has no licence file; its README states the data is not treated as sensitive, and the documents are public filings. They are redistributed here as published, for benchmarking. DeepRead's additions (expected.corrected.tsv, label_fixes.csv, schema.json) are CC-BY-4.0.

@inproceedings{stanislawek2021kleister,
  title={Kleister: Key Information Extraction Datasets Involving Long Documents with Complex Layouts},
  author={Stanis{\l}awek, Tomasz and Grali{\'n}ski, Filip and Wr{\'o}blewska, Anna and Lipi{\'n}ski, Dawid and Kaliska, Agnieszka and Rosalska, Paulina and Topolski, Bartosz and Biecek, Przemys{\l}aw},
  booktitle={International Conference on Document Analysis and Recognition},
  year={2021}
}