TAUR-Lab/MuSR

MuSR数据集旨在测试多步软推理的思维链极限。该数据集包含三个部分：谋杀谜题、物体放置和团队分配，分别对应不同的CSV文件。数据集的任务类别是问答，语言为英语，标签包括推理和常识。数据集的创建使用了ChatGPT，旨在创建需要多步推理和常识的谋杀谜题。数据集的大小类别为小于1K，许可证为cc-by-4.0。The MuSR dataset is designed to evaluate the chain-of-thought limits of multi-step soft reasoning. It consists of three components: murder mysteries, object placement tasks, and team assignment tasks, each corresponding to a separate CSV file. The dataset's task category is question answering, with all content in English, and its annotation labels cover two types: reasoning and commonsense. The dataset was constructed using ChatGPT, with the core goal of generating murder mysteries that demand multi-step reasoning and commonsense knowledge. The dataset is classified as having fewer than 1,000 instances, and it is released under the CC-BY-4.0 license.

原始信息汇总

数据集概述

数据集名称

MuSR

数据集描述

MuSR 数据集旨在测试多步骤软推理（Multistep Soft Reasoning）的能力，特别是在创建需要常识推理的谋杀谜题中。

数据文件

• murder_mysteries：murder_mystery.csv
• object_placements：object_placements.csv
• team_allocation：team_allocation.csv

许可证

CC-BY-4.0

任务类别

• 问答

语言

• 英语

标签

• 推理
• 常识

数据集大小

• 小于1000条记录

搜集汇总

背景与挑战

背景概述

MuSR is a dataset of 756 murder mystery scenarios that require multi-step reasoning and commonsense to solve. Each scenario includes a detailed narrative, two suspects, and clues that must be analyzed to identify the murderer. The dataset is designed to test the limits of chain-of-thought reasoning and is used for training and evaluating AI models. It features diverse settings, motives, and weapons, providing a rich and challenging set of problems for reasoning tasks.