StrategyQA
收藏资源简介:
StrategyQA 是一个问答基准,专注于开放域问题,其中所需的推理步骤隐含在问题中,应该使用策略进行推断。 StrategyQA 包括 2,780 个示例,每个示例由一个策略问题、其分解和证据段落组成。 StrategyQA 数据集是通过众包管道创建的,用于引发需要隐式推理步骤的创造性和多样化的是/否问题。为了解决 StrategyQA 中的问题,应该使用策略来推断推理步骤。为了指导和评估问题回答过程,StrategyQA 中的每个示例都被注释为分解为回答它的推理步骤,以及为每个步骤的答案提供证据的 Wikipedia 段落。
StrategyQA is an open-domain question answering benchmark where the requisite reasoning steps are implicitly embedded in the question, and inference must be carried out via structured strategies. The dataset includes 2,780 examples, each consisting of a strategic question, its reasoning decomposition, and supporting evidence passages. The StrategyQA dataset was created via a crowdsourcing pipeline designed to elicit creative and diverse yes/no questions that require implicit reasoning steps. To solve the questions in StrategyQA, strategies should be leveraged to infer the reasoning steps. To guide and evaluate the question answering process, each example in StrategyQA is annotated with a decomposition of the reasoning steps required to answer the question, as well as Wikipedia paragraphs that provide evidence for the answer of each reasoning step.

- StrategyQA数据集首次发表,由Amit Malhotra等人提出,旨在评估机器在复杂推理任务中的表现。
- StrategyQA数据集首次应用于多个自然语言处理模型,包括BERT和GPT-3,以测试其在多步骤推理任务中的性能。
- StrategyQA数据集被广泛用于学术研究,特别是在推理和问答系统领域,成为评估模型推理能力的重要基准。



