trivia_qa

  • Description:

TriviaqQA is a reading comprehension dataset containing over 650K question-answer-evidence triples. TriviaqQA includes 95K question-answer pairs authored by trivia enthusiasts and independently gathered evidence documents, six per question on average, that provide high quality distant supervision for answering the questions.

FeaturesDict({
    'answer': FeaturesDict({
        'aliases': Sequence(Text(shape=(), dtype=string)),
        'matched_wiki_entity_name': Text(shape=(), dtype=string),
        'normalized_aliases': Sequence(Text(shape=(), dtype=string)),
        'normalized_matched_wiki_entity_name': Text(shape=(), dtype=string),
        'normalized_value': Text(shape=(), dtype=string),
        'type': Text(shape=(), dtype=string),
        'value': Text(shape=(), dtype=string),
    }),
    'entity_pages': Sequence({
        'doc_source': Text(shape=(), dtype=string),
        'filename': Text(shape=(), dtype=string),
        'title': Text(shape=(), dtype=string),
        'wiki_context': Text(shape=(), dtype=string),
    }),
    'question': Text(shape=(), dtype=string),
    'question_id': Text(shape=(), dtype=string),
    'question_source': Text(shape=(), dtype=string),
    'search_results': Sequence({
        'description': Text(shape=(), dtype=string),
        'filename': Text(shape=(), dtype=string),
        'rank': int32,
        'search_context': Text(shape=(), dtype=string),
        'title': Text(shape=(), dtype=string),
        'url': Text(shape=(), dtype=string),
    }),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
answer FeaturesDict
answer/aliases Sequence(Text) (None,) string
answer/matched_wiki_entity_name Text string
answer/normalized_aliases Sequence(Text) (None,) string
answer/normalized_matched_wiki_entity_name Text string
answer/normalized_value Text string
answer/type Text string
answer/value Text string
entity_pages Sequence
entity_pages/doc_source Text string
entity_pages/filename Text string
entity_pages/title Text string
entity_pages/wiki_context Text string
question Text string
question_id Text string
question_source Text string
search_results Sequence
search_results/description Text string
search_results/filename Text string
search_results/rank Tensor int32
search_results/search_context Text string
search_results/title Text string
search_results/url Text string
@article{2017arXivtriviaqa,
       author = { {Joshi}, Mandar and {Choi}, Eunsol and {Weld},
                 Daniel and {Zettlemoyer}, Luke},
        title = "{triviaqa: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension}",
      journal = {arXiv e-prints},
         year = 2017,
          eid = {arXiv:1705.03551},
        pages = {arXiv:1705.03551},
archivePrefix = {arXiv},
       eprint = {1705.03551},
}

trivia_qa/rc (default config)

  • Config description: Question-answer pairs where all documents for a given question contain the answer string(s). Includes context from Wikipedia and search results.

  • Download size: 2.48 GiB

  • Dataset size: 14.99 GiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'test' 17,210
'train' 138,384
'validation' 18,669

trivia_qa/rc.nocontext

  • Config description: Question-answer pairs where all documents for a given question contain the answer string(s).

  • Download size: 2.48 GiB

  • Dataset size: 196.84 MiB

  • Auto-cached (documentation): Yes (test, validation), Only when shuffle_files=False (train)

  • Splits:

Split Examples
'test' 17,210
'train' 138,384
'validation' 18,669

trivia_qa/unfiltered

  • Config description: 110k question-answer pairs for open domain QA where not all documents for a given question contain the answer string(s). This makes the unfiltered dataset more appropriate for IR-style QA. Includes context from Wikipedia and search results.

  • Download size: 3.07 GiB

  • Dataset size: 27.27 GiB

  • Auto-cached (documentation): No

  • Splits:

Split Examples
'test' 10,832
'train' 87,622
'validation' 11,313

trivia_qa/unfiltered.nocontext

  • Config description: 110k question-answer pairs for open domain QA where not all documents for a given question contain the answer string(s). This makes the unfiltered dataset more appropriate for IR-style QA.

  • Download size: 603.25 MiB

  • Dataset size: 119.78 MiB

  • Auto-cached (documentation): Yes

  • Splits:

Split Examples
'test' 10,832
'train' 87,622
'validation' 11,313