COM2

该数据集名为COM2，是通过从现有的常识知识图谱（CSKG）中抽取多跳逻辑查询，并将它们转化为多选题和文本生成问题而创建的。该数据集旨在解决常识推理任务中的数据稀缺挑战，允许在不依赖昂贵的人工标注的情况下训练语言模型。该数据集支持的任务包括多选题回答（Mcqa）和生成性常识推理。The dataset is named COM2. It is created by extracting multi-hop logical queries from existing Commonsense Knowledge Graphs (CSKG) and converting them into multiple-choice questions and text generation problems. This dataset aims to address the challenge of data scarcity in commonsense reasoning tasks, enabling language models to be trained without relying on costly manual annotations. The tasks supported by this dataset include multiple-choice question answering (Mcqa) and generative commonsense reasoning.

搜集汇总

背景与挑战

背景概述

COM2数据集通过从常识知识图谱中抽取多跳逻辑查询并转化为多选题和文本生成问题而构建，旨在解决常识推理任务中的数据稀缺问题，无需人工标注即可训练语言模型，支持多选题回答和生成性常识推理任务。