Let Relations Speak: An End-to-End LLM-GNN Soft Prompt Framework for Fraud Detection
This work addresses the challenge of applying LLMs to fraud detection where text attributes are scarce, by enabling LLMs to process graph structures without textualization, which is a key bottleneck for this domain.
LGSPF proposes an end-to-end LLM-GNN soft prompt framework for fraud detection that eliminates reliance on text attributes by using soft prompts to bridge graph structure and semantic space, and introduces a parallel GNN encoder to translate multi-relational topologies into graph tokens. It achieves state-of-the-art performance on diverse fraud detection benchmarks.
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.