STREET

STREET是一个多任务、多领域的自然语言推理和解释基准数据集，由AWS AI, 美国创建。该数据集包含35800个问题，涵盖数学问题、逻辑谜题问题和常识及科学问题等多个领域。每个问题都附有一个推理图，这些推理图由专家标注者或程序创建，总共包含151100个推理步骤。STREET数据集旨在通过多步骤推理和解释，提高模型的可解释性，解决自然语言处理中的复杂推理问题。STREET is a multi-task, multi-domain benchmark dataset for natural language inference and explanation, developed by AWS AI of the United States. This dataset includes 35,800 questions covering diverse domains such as mathematical problems, logical puzzles, common-sense questions and scientific questions. Each question is accompanied by a reasoning graph, and these graphs, created by expert annotators or automated programs, hold a total of 151,100 reasoning steps. The STREET dataset is designed to enhance model interpretability and tackle complex reasoning challenges in natural language processing via multi-step reasoning and explanation.

搜集汇总

背景与挑战

背景概述

STREET是一个多任务、多领域的自然语言推理和解释基准数据集，包含35800个问题，涵盖数学、逻辑谜题和常识科学等领域，每个问题都配有专家或程序创建的推理图，总计151100个推理步骤。该数据集旨在通过多步骤推理提高模型的可解释性，解决自然语言处理中的复杂推理挑战。