Let Relations Speak: An End-to-End LLM-GNN Soft Prompt Framework for Fraud Detection

2026-05-27

AI Summary

Existing large language model (LLM)-based fraud detection approaches rely heavily on textual features and struggle to handle non-textual, multi-relational graph-structured data. To address this limitation, this work proposes an end-to-end LLM–GNN soft prompting framework (LGSPF) that bridges graph topology and semantic space through learnable soft prompts, avoiding the distortion-prone hard prompts. The framework incorporates a parallel GNN encoder that transforms multi-relational structures into graph tokens, enabling the LLM to capture fine-grained fraudulent patterns. LGSPF is the first method to achieve semantic alignment of graph structures without relying on textual input, effectively modeling complex relational dependencies while enhancing interpretability. Experimental results demonstrate that LGSPF achieves state-of-the-art performance across multiple fraud detection benchmarks.

Abstract

In recent years, Large Language Models (LLMs) have shown great capability in processing graph tasks such as fraud detection. However, most existing methods rely heavily on rich text attributes, which poses difficulties for this domain due to the lack of textual data. Although some pioneering methods attempt to overcome it, their textualization of graph structures via hard prompts easily leads to feature distortion. Additionally, fraud detection often exhibits multi-relational complexity, where current methods struggle to capture this deep semantic information. To address these challenges, we propose LLM-GNN Soft Prompt Framework (LGSPF). Specifically, LGSPF bridges the graph structure and semantic space using soft prompt to eliminate reliance on text. We further introduce a parallel Graph Neural Network (GNN) encoder to translate multi-relational topologies into graph tokens for fine-grained LLM fraud comprehension. Through end-to-end optimization, LGSPF enhances deep semantic alignment between LLM and GNN. Experiments across diverse fraud detection benchmarks demonstrate our method achieves state-of-the-art performance. Moreover, we further validate the contribution of LGSPF on enhancing the semantic interpretability of fraud behaviors.

Problem

• fraud detection
• Large Language Models
• graph structure
• multi-relational complexity
• textual data scarcity

Innovation

• soft prompt
• Large Language Model
• Graph Neural Network
• multi-relational graph
• fraud detection