IfQA

IfQA数据集由圣母大学和艾伦人工智能研究所共同创建，包含超过3,800个基于反事实前提的开放领域问答问题。每个问题都通过一个“如果”子句来定义反事实前提，要求模型不仅从网络检索直接事实知识，还需识别正确的信息检索并推理一个想象的情况，这可能与模型参数中构建的事实相悖。该数据集通过众包工作者在相关维基百科文章上进行标注，旨在推动开放领域问答研究在检索和反事实推理方面的发展。The IfQA dataset was co-created by the University of Notre Dame and the Allen Institute for AI, containing over 3,800 open-domain question-answering pairs based on counterfactual premises. Each question defines its counterfactual premise via an "if" clause, requiring models to not only retrieve direct factual knowledge from the web, but also identify valid information and perform reasoning over an imaginary scenario that may conflict with the factual knowledge encoded in the model's parameters. Annotated by crowdworkers on relevant Wikipedia articles, this dataset aims to advance open-domain question-answering research in the domains of information retrieval and counterfactual reasoning.

搜集汇总

背景与挑战

背景概述

IfQA是一个由圣母大学和艾伦人工智能研究所创建的开放领域问答数据集，包含超过3,800个基于反事实前提的问题，每个问题通过'如果'子句定义反事实前提，要求模型进行检索和推理以处理想象情况。该数据集旨在推动开放领域问答在检索和反事实推理方面的研究，通过众包工作者在维基百科文章上标注构建。