README.md

metadata

dataset_info:
  - config_name: default
    features:
      - name: id
        dtype: int64
      - name: query
        dtype: string
      - name: document_text
        dtype: string
      - name: acid
        dtype: string
    splits:
      - name: train
        num_bytes: 3902798162
        num_examples: 100000
      - name: validation
        num_bytes: 77013209
        num_examples: 1968
      - name: test
        num_bytes: 288050960
        num_examples: 7830
    download_size: 2465211077
    dataset_size: 4267862331
  - config_name: nq-100-tokenized
    features:
      - name: input_ids
        sequence: int32
      - name: attention_mask
        sequence: int8
      - name: labels
        sequence: int64
    splits:
      - name: train
        num_bytes: 1315687778
        num_examples: 1556605
      - name: validation
        num_bytes: 1191406
        num_examples: 1968
      - name: test
        num_bytes: 4743398
        num_examples: 7830
    download_size: 121309967
    dataset_size: 1321622582
  - config_name: nq-100k-raw
    features:
      - name: id
        dtype: int64
      - name: BM25_keywords
        sequence: string
      - name: query
        dtype: string
      - name: document_text
        dtype: string
    splits:
      - name: train
        num_bytes: 3918423056
        num_examples: 100000
      - name: validation
        num_bytes: 77322775
        num_examples: 1968
      - name: test
        num_bytes: 289276447
        num_examples: 7830
    download_size: 2277058620
    dataset_size: 4285022278
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
  - config_name: nq-100-tokenized
    data_files:
      - split: train
        path: nq-100-tokenized/train-*
      - split: validation
        path: nq-100-tokenized/validation-*
      - split: test
        path: nq-100-tokenized/test-*
  - config_name: nq-100k-raw
    data_files:
      - split: train
        path: nq-100k-raw/train-*
      - split: validation
        path: nq-100k-raw/validation-*
      - split: test
        path: nq-100k-raw/test-*

Abstractive Content-Based Document IDs for Generative Retrieval

Dataset for Summarization-Based Document IDs for Generative Retrieval with Language Models.

Update

[03/04/2025] Upload validation and test set of ACID. Add tokenized subset.

@misc{li2024summarizationbaseddocumentidsgenerative,
      title={Summarization-Based Document IDs for Generative Retrieval with Language Models}, 
      author={Haoxin Li and Daniel Cheng and Phillip Keung and Jungo Kasai and Noah A. Smith},
      year={2024},
      eprint={2311.08593},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2311.08593}, 
}