README.md

metadata

dataset_info:
  features:
    - name: image_file_name
      dtype: string
    - name: image_path
      dtype: string
    - name: object_id
      sequence: int64
    - name: headline
      dtype: string
    - name: article
      dtype: string
    - name: byline
      dtype: string
    - name: bbox_list
      sequence:
        sequence: float64
    - name: bbox
      sequence: float64
    - name: full_article_id
      dtype: int64
    - name: id
      dtype: string
    - name: imageid
      dtype: int64
    - name: query
      dtype: string
    - name: idx
      dtype: int64
    - name: cluster
      dtype: int64
    - name: duplicates
      sequence: int64
  splits:
    - name: test
      num_bytes: 23946859
      num_examples: 14211
    - name: val
      num_bytes: 8647243
      num_examples: 4988
  download_size: 19407100
  dataset_size: 32594102
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
      - split: val
        path: data/val-*
license: unknown

NEWS COPY

This dataset contains the evaluation and test sets for the NEWS COPY dataset. Original source can be found at Github. The license is unclear.

It contains the following data:

• Historical Newspapers

Training datasets can be found at chenghao/NEWS-COPY-train.

Citation

@inproceedings{silcock-etal-2020-noise,
  title = "Noise-Robust De-Duplication at Scale",
  author = "Silcock, Emily and D'Amico-Wong, Luca and Yang, Jinglin and Dell, Melissa",
  booktitle = "International Conference on Learning Representations (ICLR)",
  year = "2023",
}