trec

  • Descrição :

O conjunto de dados Text REtrieval Conference (TREC) Question Classification contém 5.500 questões rotuladas no conjunto de treinamento e outras 500 para o conjunto de teste. O conjunto de dados tem 6 rótulos, 47 rótulos de nível 2. O comprimento médio de cada frase é 10, o tamanho do vocabulário é de 8.700. Os dados são coletados de quatro fontes: 4.500 questões de inglês publicadas pela USC (Hovy et al., 2001), cerca de 500 questões construídas manualmente para algumas classes raras, 894 TREC 8 e TREC 9 questões e também 500 questões do TREC 10 que serve como conjunto de teste.

Dividir Exemplos
'test' 500
'train' 5.452
  • Estrutura de recursos :
FeaturesDict({
    'label-coarse': ClassLabel(shape=(), dtype=int64, num_classes=6),
    'label-fine': ClassLabel(shape=(), dtype=int64, num_classes=47),
    'text': Text(shape=(), dtype=string),
})
  • Documentação do recurso :
Recurso Aula Forma Tipo D Descrição
RecursosDict
rótulo grosseiro ClassLabel int64
rótulo fino ClassLabel int64
texto Texto corda
  • Citação :
@inproceedings{li-roth-2002-learning,
    title = "Learning Question Classifiers",
    author = "Li, Xin  and
      Roth, Dan",
    booktitle = "{COLING} 2002: The 19th International Conference on Computational Linguistics",
    year = "2002",
    url = "https://www.aclweb.org/anthology/C02-1150",
}
@inproceedings{hovy-etal-2001-toward,
    title = "Toward Semantics-Based Answer Pinpointing",
    author = "Hovy, Eduard  and
      Gerber, Laurie  and
      Hermjakob, Ulf  and
      Lin, Chin-Yew  and
      Ravichandran, Deepak",
    booktitle = "Proceedings of the First International Conference on Human Language Technology Research",
    year = "2001",
    url = "https://www.aclweb.org/anthology/H01-1069",
}