LLM-Enhanced Graph Learning Improves Financial Fraud Detection

Researchers have developed SemHGT, a novel framework that enhances financial fraud detection by integrating Large Language Models (LLMs) with heterogeneous transaction graphs. This approach explicitly models the business semantics of structured attributes based on entity roles, a feature often overlooked by existing methods. SemHGT uses role-aware prompts to generate business descriptions from attributes, enriches node representations with semantic vectors, and employs a unified transformer to process path evidence and neighbor summaries for improved fraud detection accuracy across multiple benchmark datasets. AI

IMPACT Enhances financial fraud detection by explicitly modeling business semantics in transaction graphs using LLMs.

RANK_REASON The item is a research paper detailing a new framework for financial fraud detection using LLMs and graph learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Heterogeneous Graph Neural Networks
• Hugging Face
• Large Language Models
• ScienceCast
• SemHGT

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →