核心概念
統一された情報抽出と大規模言語モデルは、定義バイアスを解決できない。
統計資料
Definition bias is a negative phenomenon that can mislead models.
Bias among IE datasets and Bias between IE and instruction tuning datasets are identified.
Fleiss’ Kappa is used to measure dataset definition bias and type definition bias.
引述
"Definition bias in IE refers to the tendency of an information extraction system to favor certain interpretations of data over others."
"Performance remains subpar in handling information extraction tasks, which we believe is due to definition bias."