Reproduce “AuditAgent: Expert-Guided Multi-Agent Reasoning for Cross-Document Fraudulent Evidence Discovery” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Songran Bai, Bingzhe Wu, Yiwei Zhang, Chengke Wu, Xiaolong Zheng, Yaze Yuan, Ke Wu, Jianqiang Li
AuditAgent, a multi-agent LLM framework, and FinFraud-Real, a new real-world dataset, were developed for localizing explicit fraudulent evidence across multiple financial documents. The AuditAgent framework significantly improves issue-level recall by up to 100.9% and evidence-level recall by up to 107.6% compared to general-purpose agent models by integrating auditing domain expertise and prior risk modeling.
Key insights
- Integrating deep auditing domain expertise and structured risk priors into AI models is crucial for effective and interpretable fine-grained financial fraud evidence localization.
- General-purpose LLM agent frameworks, without explicit domain-specific guidance, struggle to achieve precise risk recall in complex financial fraud scenarios due to generating irrelevant search paths.
- Reasoning-augmented LLMs, such as DeepSeek-R1, and financial domain-specific LLMs demonstrate superior capabilities in integrating and associating subtle, contextually relevant clues for fraud detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.