README.md

metadata

dataset_info:
  features:
    - name: model_name
      dtype: string
    - name: dataset_path
      dtype: string
    - name: num_docs_pr
      dtype: int64
    - name: random_seed
      dtype: int64
    - name: queries
      sequence: string
    - name: positives
      sequence:
        sequence: string
    - name: negatives
      sequence:
        sequence: string
    - name: scores
      sequence:
        sequence: float64
    - name: targets
      sequence:
        sequence: float64
  splits:
    - name: test
      num_bytes: 2218701030
      num_examples: 132
  download_size: 1121022682
  dataset_size: 2218701030
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*

Towards Trustworthy Reranking: A Simple yet Effective Abstention Mechanism

Dataset

This dataset provides the evaluation benchmark used in the paper "Towards Trustworthy Reranking: A Simple yet Effective Abstention Mechanism", accepted at TMLR (September 2024).

Paper

For more details on this work, you can read the full paper here.

Citation:

@article{gisserot2024towards,
  title={Towards Trustworthy Reranking: A Simple yet Effective Abstention Mechanism},
  author={Gisserot-Boukhlef, Hippolyte and Faysse, Manuel and Malherbe, Emmanuel and Hudelot, C{\'e}line and Colombo, Pierre},
  journal={arXiv preprint arXiv:2402.12997},
  year={2024}
}