Datasets:

SallyTan
/

MathToF

license: cc-by-4.0
task_categories:
  - text-generation
language:
  - zh
size_categories:
  - n<1K
tags:
  - chinese
  - math-word-problems
  - true-false
  - mathematical-reasoning
modalities:
  - text
libraries:
  - Datasets

Introduction

MathToF is a Chinese mathematical reasoning dataset introduced in the paper Teaching-Inspired Integrated Prompting Framework: A Novel Approach for Enhancing Reasoning in Large Language Models.

It contains 1,000 Chinese true-or-false math problems, each annotated with a binary label (True/False) and a detailed rationale.

Dataset Structure

Data Fields

qtype: question type. In MathToF, this is typically "JUDGE".
quest_stem: the main question content.
- quest_stem.text: the problem statement in Chinese.
quest_ref: reference answers and explanations.
- quest_ref.texts: the ground-truth label(s), typically "True" or "False".
- quest_ref.analyses: the explanation(s), stored as a list.

Example

{
    "qtype": "JUDGE",
    "quest_stem": {
        "text": "三位数减三位数差一定是三位数。"
    },
    "quest_ref": {
        "texts": [
            "False"
        ],
        "analyses": [
            "如两个三位数分别为:150，100。则150-100=50，其差50是两位数，故题干描述错误。"
        ]
    }
}

Dataset Statistics

According to the paper, the question-type distribution of MathToF is:

Arithmetic: 675
Algebra: 61
Geometry: 197
Statistics: 37
Reasoning: 13
Others: 17

Total: 1,000 questions.

Citation

If you use this dataset, please cite the following paper:

@inproceedings{tan-etal-2025-teaching,
  title = {Teaching-Inspired Integrated Prompting Framework: A Novel Approach for Enhancing Reasoning in Large Language Models},
  author = {Tan, Wenting and Chen, Dongxiao and Xue, Jieting and Wang, Zihao and Chen, Taijie},
  booktitle = {Proceedings of the 31st International Conference on Computational Linguistics: Industry Track},
  pages = {827--839},
  year = {2025},
  publisher = {Association for Computational Linguistics}
}