florin-hf/nq_open_gold

Natural Questions Open Dataset with Gold Documents

This dataset is a curated version of the Natural Questions open dataset, with the inclusion of the gold documents from the original Natural Questions (NQ) dataset. The main difference with the NQ-open dataset is that some entries were excluded, as their respective gold documents exceeded 512 tokens in length. This is due to the pre-processing of the gold documents, as detailed in this related dataset.

The dataset is designed to facilitate research in question-answering systems, especially focusing on integrating gold documents for training and testing purposes.

Dataset Sources

The Natural Questions (NQ) dataset is a large-scale collection of real-world queries derived from Google search data. Each entry in the dataset consists of a user query and the corresponding Wikipedia page containing the answer.

The NQ-open dataset, a subset of the NQ dataset, differs by removing the restriction of linking answers to specific Wikipedia passages, thereby mimicking a more general information retrieval scenario similar to web searches.

This version of the NQ-open dataset was used in the paper The Power of Noise: Redefining Retrieval for RAG Systems.

Dataset Structure

A sample in the dataset has the following format:

{
    'example_id' (int64): an identifier for the question, consistent with the original NQ dataset,
    'question' (str): a question, that is identical to the question in the original NQ,
    'answers' (List[str]): the list of correct answers in the original NQ,
    'text' (str): gold document, associated with the question, in the original NQ,
    'idx_gold_in_corpus' (int64): index of the gold document in the full corpus.
}

Ex.
{
    'example_id': -3440030035760311385,
    'question': 'who owned the millennium falcon before han solo',
    'answers': [Lando Calrissian],
    'text': "Han Solo won the Millennium Falcon from Lando Calrissian in the card game ' sabacc ' several years before the events of the film A New Hope..."
    'idx_gold_in_corpus': 20995349
}

Dataset Splits

• Train set: 72,209 (50,2 MB)
• Validation set: 8,006 (5,57 BM)
• Test set: 2889 (1,96 MB)

Citation Information

@article{doi:10.1162/tacl\_a\_00276,
    author = {Kwiatkowski, Tom and Palomaki, Jennimaria and Redfield, Olivia and Collins, Michael and Parikh, Ankur and Alberti, Chris and Epstein, Danielle and Polosukhin, Illia and Devlin, Jacob and Lee, Kenton and Toutanova, Kristina and Jones, Llion and Kelcey, Matthew and Chang, Ming-Wei and Dai, Andrew                         M. and Uszkoreit, Jakob and Le, Quoc and Petrov, Slav},
    title = {Natural Questions: A Benchmark for Question Answering Research},
    journal = {Transactions of the Association for Computational Linguistics},
    volume = {7},
    number = {},
    pages = {453-466},
    year = {2019},
    doi = {10.1162/tacl\_a\_00276},
    URL = { 
            https://doi.org/10.1162/tacl_a_00276
        },
    eprint = { 
            https://doi.org/10.1162/tacl_a_00276
        
        },
    abstract = { We present the Natural Questions corpus, a question answering data set. Questions consist of real anonymized, aggregated queries issued to the Google search engine. An annotator is presented with a question along with a Wikipedia page from the top 5 search results, and annotates a long answer (typically a paragraph) and a short answer (one or more entities) if present on the page, or marks null if no long/short answer is present. The public release consists of 307,373 training examples with single annotations; 7,830 examples with 5-way annotations for development data; and a further 7,842 examples with 5-way annotated sequestered as test data. We present experiments validating quality of the data. We also describe analysis of 25-way annotations on 302 examples, giving insights into human variability on the annotation task. We introduce robust metrics for the purposes of evaluating question answering systems; demonstrate high human upper bounds on these metrics; and establish baseline results using competitive methods drawn from related literature. }
}

@inproceedings{lee-etal-2019-latent,
    title = "Latent Retrieval for Weakly Supervised Open Domain Question Answering",
    author = "Lee, Kenton  and
      Chang, Ming-Wei  and
      Toutanova, Kristina",
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/P19-1612",
    doi = "10.18653/v1/P19-1612",
    pages = "6086--6096",
    abstract = "Recent work on open domain question answering (QA) assumes strong supervision of the supporting evidence and/or assumes a blackbox information retrieval (IR) system to retrieve evidence candidates. We argue that both are suboptimal, since gold evidence is not always available, and QA is fundamentally different from IR. We show for the first time that it is possible to jointly learn the retriever and reader from question-answer string pairs and without any IR system. In this setting, evidence retrieval from all of Wikipedia is treated as a latent variable. Since this is impractical to learn from scratch, we pre-train the retriever with an Inverse Cloze Task. We evaluate on open versions of five QA datasets. On datasets where the questioner already knows the answer, a traditional IR system such as BM25 is sufficient. On datasets where a user is genuinely seeking an answer, we show that learned retrieval is crucial, outperforming BM25 by up to 19 points in exact match.",
}

@inproceedings{Cuconasu_2024, series={SIGIR 2024},
   title={The Power of Noise: Redefining Retrieval for RAG Systems},
   url={http://dx.doi.org/10.1145/3626772.3657834},
   DOI={10.1145/3626772.3657834},
   booktitle={Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval},
   publisher={ACM},
   author={Cuconasu, Florin and Trappolini, Giovanni and Siciliano, Federico and Filice, Simone and Campagnano, Cesare and Maarek, Yoelle and Tonellotto, Nicola and Silvestri, Fabrizio},
   year={2024},
   month=jul, collection={SIGIR 2024}
}