Can LLMs Find Fraudsters? Multi-level LLM Enhanced Graph Fraud Detection

2025-07-16

AI Summary

Existing graph-based fraud detection methods rely on preprocessed node embeddings and fixed graph structures, neglecting rich semantic cues inherent in raw textual attributes. While large language models (LLMs) excel at textual representation learning, their effective multimodal integration with graph structural information remains challenging. To address this, we propose a novel multi-level LLM-enhanced framework that introduces, for the first time, dedicated text-graph co-enhancers at both the node-type level and relation level. Specifically, LLMs extract fine-grained semantic representations and external knowledge from node texts, which are then dynamically fused into graph neural networks via learnable gating mechanisms—enabling semantic-guided, structure-aware node representations. Extensive experiments on four real-world datasets demonstrate that our method consistently outperforms state-of-the-art approaches, achieving average F1-score improvements of 3.2–7.8 percentage points. These results validate the effectiveness and generalizability of multi-level semantic-structural joint modeling for graph fraud detection.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Graph fraud detection has garnered significant attention as Graph Neural Networks (GNNs) have proven effective in modeling complex relationships within multimodal data. However, existing graph fraud detection methods typically use preprocessed node embeddings and predefined graph structures to reveal fraudsters, which ignore the rich semantic cues contained in raw textual information. Although Large Language Models (LLMs) exhibit powerful capabilities in processing textual information, it remains a significant challenge to perform multimodal fusion of processed textual embeddings with graph structures. In this paper, we propose a extbf{M}ulti-level extbf{L}LM extbf{E}nhanced Graph Fraud extbf{D}etection framework called MLED. In MLED, we utilize LLMs to extract external knowledge from textual information to enhance graph fraud detection methods. To integrate LLMs with graph structure information and enhance the ability to distinguish fraudsters, we design a multi-level LLM enhanced framework including type-level enhancer and relation-level enhancer. One is to enhance the difference between the fraudsters and the benign entities, the other is to enhance the importance of the fraudsters in different relations. The experiments on four real-world datasets show that MLED achieves state-of-the-art performance in graph fraud detection as a generalized framework that can be applied to existing methods.

Problem

• Enhancing fraud detection using LLMs and graph structures
• Integrating textual embeddings with multimodal graph data
• Improving fraudster differentiation via multi-level LLM enhancement

Innovation

• LLMs extract external knowledge from text
• Multi-level enhancer integrates LLMs with graphs
• Type and relation enhancers distinguish fraudsters