---
dataset_info:
- config_name: papyrus-a
  features:
  - name: doc_id
    dtype: int64
  - name: title
    dtype: string
  - name: input_text
    dtype: string
  - name: keyphrases
    sequence: string
  - name: lang
    dtype: string
  splits:
  - name: train
    num_bytes: 48856197
    num_examples: 11290
  - name: test
    num_bytes: 14237516
    num_examples: 3261
  - name: validation
    num_bytes: 7101302
    num_examples: 1638
  download_size: 39852407
  dataset_size: 70195015
- config_name: papyrus-e
  features:
  - name: doc_id
    dtype: int64
  - name: title
    dtype: string
  - name: input_text
    dtype: string
  - name: keyphrases
    sequence: string
  - name: lang
    dtype: string
  splits:
  - name: train
    num_bytes: 23220234
    num_examples: 10508
  - name: test
    num_bytes: 6777041
    num_examples: 3046
  - name: validation
    num_bytes: 3394239
    num_examples: 1539
  download_size: 19090105
  dataset_size: 33391514
- config_name: papyrus-f
  features:
  - name: doc_id
    dtype: int64
  - name: title
    dtype: string
  - name: input_text
    dtype: string
  - name: keyphrases
    sequence: string
  - name: lang
    dtype: string
  splits:
  - name: train
    num_bytes: 26332755
    num_examples: 10299
  - name: test
    num_bytes: 7691101
    num_examples: 2981
  - name: validation
    num_bytes: 3820763
    num_examples: 1488
  download_size: 20986924
  dataset_size: 37844619
- config_name: papyrus-m
  features:
  - name: doc_id
    dtype: int64
  - name: title
    dtype: string
  - name: input_text
    dtype: string
  - name: keyphrases
    sequence: string
  - name: lang
    dtype: string
  splits:
  - name: train
    num_bytes: 49906922
    num_examples: 20963
  - name: test
    num_bytes: 14543415
    num_examples: 6061
  - name: validation
    num_bytes: 7242231
    num_examples: 3040
  download_size: 40019743
  dataset_size: 71692568
configs:
- config_name: papyrus-a
  data_files:
  - split: train
    path: papyrus-a/train-*
  - split: test
    path: papyrus-a/test-*
  - split: validation
    path: papyrus-a/validation-*
- config_name: papyrus-e
  data_files:
  - split: train
    path: papyrus-e/train-*
  - split: test
    path: papyrus-e/test-*
  - split: validation
    path: papyrus-e/validation-*
- config_name: papyrus-f
  data_files:
  - split: train
    path: papyrus-f/train-*
  - split: test
    path: papyrus-f/test-*
  - split: validation
    path: papyrus-f/validation-*
- config_name: papyrus-m
  data_files:
  - split: train
    path: papyrus-m/train-*
  - split: test
    path: papyrus-m/test-*
  - split: validation
    path: papyrus-m/validation-*
license: apache-2.0
language:
- en
- fr
tags:
- text-to-text
- keyphrase-generation
pretty_name: Papyrus
size_categories:
- 10K<n<100K
---


# Dataset Card for Papyrus

- **Paper:** [A new dataset for multilingual keyphrase generation](https://proceedings.neurips.cc/paper_files/paper/2022/hash/f88709551258331f9ab31b33c71021a4-Abstract-Datasets_and_Benchmarks.html)
- **Github:** <https://github.com/smolPixel/French-keyphrase-generation>


## Dataset Description

### Dataset Summary

The datasets are derived from Papyrus, a repository at Université de Montréal containing various types of documents, mainly theses with abstracts in multiple languages, primarily French and English. The entries are provided in four different configurations based on the languages of abstracts, allowing for generating keyphrases in French, English, or multiple languages.
- **Papyrus-f:** From the French abstracts, generate French keyphrases.
- **Papyrus-e:** From the English abstracts, generate English keyphrases.
- **Papyrus-m:** From one abstract in any language, generate keyphrases in that same
language (one language to one language).
- **Papyrus-a:** From the multiple abstracts of a document, generate keyphrases in the
same languages as the abstracts (many to many languages).

### Languages

- **Main languages:** English, French
- **Others:** Spanish, German, Italian, Portuguese, Arabic, Tagalog, Catalan, Greek, Turkish, Russian, Polish, Farsi, Indonesian, Lingala, Swedish, Finnish, Romanian, Korean


## Dataset Structure

### Dataset content

| Config    | Train set size | Valid. set size | Test set size |
| --------- | -------------- | --------------- | ------------- |
| papyrus-m | 20963          | 3040            | 6061          |
| papyrus-e | 10508          | 1539            | 3046          |
| papyrus-f | 10299          | 1488            | 2981          |
| papyrus-a | 11290          | 1638            | 3261          |

### Data fields

- **doc_id:** a unique id for the original document.
- **title:** title of the thesis or article (the language of the title does not always match the language of the abstract/keyphrases).
- **input_text:** abstract of the document.
- **keyphrases:** associated keyphrases.
- **lang:** language of the abstract/keyphrases.


## Citation

    @inproceedings{NEURIPS2022_f8870955,
        author = {Piedboeuf, Fr\'{e}d\'{e}ric and Langlais, Philippe},
        booktitle = {Advances in Neural Information Processing Systems},
        editor = {S. Koyejo and S. Mohamed and A. Agarwal and D. Belgrave and K. Cho and A. Oh},
        pages = {38046--38059},
        publisher = {Curran Associates, Inc.},
        title = {A new dataset for multilingual keyphrase generation},
        url = {https://proceedings.neurips.cc/paper_files/paper/2022/file/f88709551258331f9ab31b33c71021a4-Paper-Datasets_and_Benchmarks.pdf},
        volume = {35},
        year = {2022}
    }