clustering-hal-s2s / README.md
mciancone's picture
Update README.md
3192f1c verified
metadata
license: apache-2.0
task_categories:
  - text-classification
language:
  - fr
size_categories:
  - 10K<n<100K
configs:
  - config_name: default
    data_files:
      - split: test
        path: test.jsonl
  - config_name: raw
    data_files:
      - split: test
        path: test.jsonl
  - config_name: mteb_eval
    data_files:
      - split: test
        path: mteb_eval.jsonl

Clustering HAL

This dataset was created by scrapping data from the HAL platform. Over 80,000 articles have been scrapped to keep their id, title and category.

It was originally used for the French version of MTEB, but it can also be used for various clustering or classification tasks, or even evaluate the general knowledge of a model.

⚠️ This dataset contains 2 subsets. IT IS STRONGLY ADVISED TO USE THE CLEANED UP mteb_eval SUBSET:

  • "raw" subset : contains the data originally scrapped, without any cleaning. The data contains mostly titles in French, but also titles in other languages (english, italian, ...)
  • "mteb_eval" subset : is the subset used for the MTEB evaluation. It is a cleaned up version of the raw dataset. Notably, samples have been removed if :
    • their "domain" were in a minor class (less than 500 samples were available)
    • their "title" were less than or equal 2 words
    • the language was not French

Usage

To use this dataset, you can run the following code :

from datasets import load_dataset
dataset = load_dataset("lyon-nlp/clustering-hal-s2s", name="mteb_eval", split="test") # for MTEB eval subset

Citation

If you use this dataset in your work, please consider citing:

@misc{ciancone2024extending,
      title={Extending the Massive Text Embedding Benchmark to French}, 
      author={Mathieu Ciancone and Imene Kerboua and Marion Schaeffer and Wissam Siblini},
      year={2024},
      eprint={2405.20468},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}