luffycodes/DUPEd_StrategyQA

该数据集是为论文《Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models》创建的，用于研究大型语言模型在扰动证据下的推理能力。数据集中的QID以math_dupe或nlp_dupe结尾的条目分别对应数学和自然语言处理的扰动。QID可以映射到原始的StrategyQA数据集，并且答案需要反转。

This dataset is created for the paper Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models, which investigates the reasoning capabilities of large language models under the influence of perturbed evidence in mathematics and natural language processing. The question-answer pairs in this dataset are mapped from the original strategyQA dataset, and the answers need to be reversed for factual questions.

原始信息汇总

数据集概述

数据集名称

Dataset for the paper: Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models

数据集说明

• 该数据集用于论文《Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models》。
• QID以math_dupe/nlp_dupe结尾的分别指代数学/自然语言处理的扰动。
• QID可以映射到原始的strategyQA数据集，该数据集可从这里下载。
• 对于事实类问题，答案需要反转，这指的是strategyQA数据集中的原始答案字段。

引用信息

如果使用该数据集，请引用以下信息：

@article{sonkar2023deduction, title={Deduction under Perturbed Evidence: Probing Student Simulation Capabilities of Large Language Models}, author={Sonkar, Shashank and Baraniuk, Richard G}, journal={arXiv preprint arXiv:2305.14507}, year={2023} }