trec

  • תיאור :

מערך השאלות של ועידת אחזור הטקסט (TREC) מכיל 5500 שאלות עם תווית בערכת הדרכה ועוד 500 עבור ערכת מבחנים. למערך הנתונים יש 6 תוויות, 47 תוויות ברמה 2. אורך ממוצע של כל משפט הוא 10, גודל אוצר מילים של 8700. הנתונים נאספים מארבעה מקורות: 4,500 שאלות באנגלית שפורסמו על ידי USC (Hovy et al., 2001), כ-500 שאלות שנבנו באופן ידני לכמה שיעורים נדירים, 894 TREC 8 ו TREC 9 שאלות, וגם 500 שאלות מ-TREC 10 המשמש כמערך המבחן.

לְפַצֵל דוגמאות
'test' 500
'train' 5,452
  • מבנה תכונה :
FeaturesDict({
    'label-coarse': ClassLabel(shape=(), dtype=int64, num_classes=6),
    'label-fine': ClassLabel(shape=(), dtype=int64, num_classes=47),
    'text': Text(shape=(), dtype=string),
})
  • תיעוד תכונה :
תכונה מעמד צוּרָה Dtype תיאור
FeaturesDict
תווית-גסה ClassLabel int64
תווית בסדר ClassLabel int64
טֶקסט טֶקסט חוּט
  • ציטוט :
@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",
}