UnbiasedMWP

UnbiasedMWP是由中山大学深圳校区创建的一个大型数学应用题数据集，旨在减少解决偏差。该数据集包含10264个多样化的数学问题，通过改变基础表达式并手动注释相应的新问题来构建，以覆盖所有MWP的每个问题叙述的多样化问题。数据集的创建过程涉及从在线教育网站收集原始数据，并进行预处理和问题重写，以确保数据的公正性。UnbiasedMWP的应用领域在于公平评估解题器的推理技能，解决现有数据集中的数据偏差和学习偏差问题。

UnbiasedMWP is a large-scale mathematical word problem (MWP) dataset developed by the Shenzhen Campus of Sun Yat-sen University, with the goal of mitigating solving biases. It consists of 10,264 diverse mathematical problems, which are constructed by modifying underlying mathematical expressions and manually annotating corresponding new questions to cover diversified problem narratives across all MWP types. The dataset creation workflow includes collecting raw data from online education platforms, followed by preprocessing and question rewriting to ensure data fairness. The primary application scenario of UnbiasedMWP is to fairly evaluate the reasoning capabilities of mathematical problem solvers, addressing the data bias and learning bias issues present in existing datasets.