This repository contains the implementation, evaluation scripts, and experimental data for our paper:
"BIASINSPECTOR: Detecting Bias in Structured Data through LLM Agents"

If you find this work helpful, please consider citing our paper:

BIASINSPECTOR: Detecting Bias in Structured Data through LLM Agents
Haoxuan Li, Mingyu Derek Ma, Jen-tse Huang, Zhaotian Weng, Wei Wang, Jieyu Zhao
arXiv: 2504.04855

@article{li2025biasinspector,
  title={BIASINSPECTOR: Detecting Bias in Structured Data through LLM Agents},
  author={Li, Haoxuan and Ma, Mingyu Derek and Huang, Jen-tse and Weng, Zhaotian and Wang, Wei and Zhao, Jieyu},
  journal={arXiv preprint arXiv:2504.04855},
  year={2025}
}

For questions or collaboration opportunities, please contact: Haoxuan Li – lihaoxua@usc.edu

• submission/
  • agents/
    • bias_inspector_multi/ – Multi-agent framework for bias detection
    • bias_inspector_single/ – Single-agent framework implementation
    • react_agent/ – ReAct-based agent with tool usage
    • self_reflection_agent/ – Self-reflective agent for iterative reasoning
  • evaluation/
    • result_evaluation/ – Scripts for evaluating final detection results
    • process_evaluation/ – Scripts for analyzing reasoning steps and tool usage
  • data/
    • dataset_and_taskset/ – Structured datasets and defined bias detection tasks
    • eval_results/ – Evaluation metrics in CSV format
    • logs/ – Execution logs per agent and model variant
  • README.md – Project documentation