StepMathAgent

StepMathAgent is a novel mathematical process evaluation agent based on Tree-of-Error, which incorporates four internal core operations: logical step segmentation, step scoring, score aggregation and error tree generation, along with four external extension modules: difficulty calibration, simplicity evaluation, completeness validation and format assessment.

We also introduce StepMathBench, a benchmark comprising 1,000 step-divided process evaluation instances, derived from 200 high-quality math problems grouped by problem type, subject category and difficulty level.

StepMathBench

Mathematical Problem: data/question.jsonl

StepMath: data/stepmath.jsonl

Quick Start

Use StepMathAgent

python agent.py

Use Four External Extension Modules in StepMathAgent

# Modify model_name in agent.py
python agent.py

Run Baseline

# Modify api_name and model_name in agent.py
python agent.py

Evaluation

python evaluator.py

Case Study

Citation

If you find this work useful in your research, please consider citing:

@misc{yang2025stepmathagentstepwiseagentevaluating,
      title={StepMathAgent: A Step-Wise Agent for Evaluating Mathematical Processes through Tree-of-Error}, 
      author={Shu-Xun Yang and Cunxiang Wang and Yidong Wang and Xiaotao Gu and Minlie Huang and Jie Tang},
      year={2025},
      eprint={2503.10105},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2503.10105}, 
}