risposta_equivalenza

  • Descrizione :

Il set di dati sull'equivalenza delle risposte contiene valutazioni umane sulle previsioni del modello da diversi modelli nel set di dati SQuAD. Le valutazioni stabiliscono se la risposta prevista è "equivalente" alla risposta oro (tenendo conto sia della domanda che del contesto).

Più specificamente, per "equivalente" intendiamo che la risposta prevista contiene almeno le stesse informazioni della risposta oro e non aggiunge informazioni superflue. Il set di dati contiene annotazioni per: * previsioni da BiDAF su SQuAD dev * previsioni da XLNet su SQuAD dev * previsioni da Luke su SQuAD dev * previsioni da Albert su SQuAD training, dev ed esempi di test

Diviso Esempi
'ae_dev' 4.446
'ae_test' 9.724
'dev_bidaf' 7.522
'dev_luke' 4.590
'dev_xlnet' 7.932
'train' 9.090
  • Struttura delle caratteristiche :
FeaturesDict({
    'candidate': Text(shape=(), dtype=string),
    'context': Text(shape=(), dtype=string),
    'gold_index': int32,
    'qid': Text(shape=(), dtype=string),
    'question': Text(shape=(), dtype=string),
    'question_1': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'question_2': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'question_3': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'question_4': ClassLabel(shape=(), dtype=int64, num_classes=3),
    'reference': Text(shape=(), dtype=string),
    'score': float32,
})
  • Documentazione delle funzionalità :
Caratteristica Classe Forma Tipo D Descrizione
CaratteristicheDict
candidato Testo corda
contesto Testo corda
indice_oro Tensore int32
qid Testo corda
domanda Testo corda
domanda 1 ClassLabel int64
Domanda 2 ClassLabel int64
Domanda 3 ClassLabel int64
domanda_4 ClassLabel int64
riferimento Testo corda
punto Tensore galleggiante32
  • Citazione :
@article{bulian-etal-2022-tomayto,
      title={Tomayto, Tomahto. Beyond Token-level Answer Equivalence for Question Answering Evaluation},
      author={Jannis Bulian and Christian Buck and Wojciech Gajewski and Benjamin Boerschinger and Tal Schuster},
      year={2022},
      eprint={2202.07654},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}