L2IR: Revealing Latent Intent in Graph Fraud Detection

2026-05-25

AI Summary

This work addresses the significant performance degradation of existing graph neural network (GNN)-based fraud detection methods in highly adversarial scenarios where fraudsters dilute genuine signals through massive fabricated connections and labeled data is scarce. To tackle this challenge, the paper proposes L2IR, a novel framework that introduces latent intent reasoning into graph-based fraud detection for the first time. L2IR leverages large language models (LLMs) to interpret user behaviors and suspicious links, extracting intent-aware representations to differentiate supportive from misleading edges. It further incorporates a plug-and-play adaptive self-training mechanism to enhance generalization under weak supervision. Experiments on two real-world datasets with high camouflage density demonstrate that L2IR substantially outperforms strong baselines and, as a universal plugin, improves the AUPRC of various GNN detectors by up to 8.27%.

Abstract

Graph fraud detection has long depended on Graph Neural Networks (GNNs) to propagate and aggregate information across relational data. A critical obstacle in practice, however, is that fraudsters frequently disguise themselves by forging numerous connections with benign users, causing fraud signals to be progressively diluted during neighborhood aggregation and undermining detection reliability. While recent efforts have used Large Language Models (LLMs) to provide rich semantic cues for fraud detection, the underlying intent behind suspicious connections remains insufficiently explored. Compounding this issue, the scarcity of annotated fraud samples makes it difficult to train detectors that remain robust under heavy camouflage. To address these gaps, we propose L2IR, an LLM-driven Latent Intent Revealing framework for graph fraud detection. By uncovering latent intent from both user behaviors and suspicious connections, L2IR extracts intent-aware representations from raw behavioral traces and reasons about the true purpose behind individual connections, effectively distinguishing supportive links from misleading ones. It further incorporates adaptive self-training to enhance robustness under limited supervision. Evaluations on two real-world datasets characterized by pervasive camouflage demonstrate that L2IR surpasses strong baselines and can function as a plug-in enhancement for a range of GNN-based detectors, improving AUPRC by up to 8.27%.

Problem

• graph fraud detection
• latent intent
• fraud camouflage
• limited supervision
• suspicious connections

Innovation

• Latent Intent
• Graph Fraud Detection
• Large Language Models
• Adaptive Self-Training
• Intent-Aware Representation