FedLSG: LLM-Enhanced Semantic Calibration for Federated Graph Backdoor Defense

2026-07-21

AI Summary

This work addresses the vulnerability of federated graph neural networks to stealthy backdoor attacks, noting that existing rule-based defenses struggle to identify triggers due to limited semantic understanding and often inadvertently corrupt benign graph structures. To overcome these limitations, the paper proposes FedLSG, a novel framework that integrates large language models (LLMs) into federated graph backdoor defense for the first time. FedLSG translates graph topology and client behavior into natural language to establish a semantic alignment mechanism and employs a lightweight teacher–student architecture: the server leverages an LLM to assess the semantic plausibility of model updates, while clients deploy LoRA-tuned student models to suppress the influence of malicious edges. This approach enables semantics-driven adaptive defense, significantly enhancing backdoor robustness while preserving the integrity of legitimate graph structures.

Abstract

Federated Graph Neural Networks (FedGNNs) are highly vulnerable to backdoor poisoning, yet existing defenses typically rely on rule-based approaches that lack semantic understanding, making them vulnerable to stealthy triggers and harmful to benign structures. To solve this, we present FedLSG, the first framework that integrates large language models (LLMs) into federated graph backdoor defense. FedLSG introduces a graph and behavior to text grounding scheme that transforms local graph structures and client update behaviors into semantically rich natural language representations. The framework further adopts a lightweight student-teacher architecture. On the server side, a full scale LLM serves as a teacher, providing global contextual guidance and evaluating client updates during aggregation to identify potentially malicious participants. On the client side, a LoRA-based student is maintained to perform semantic reasoning, to suppress the influence of edges associated with backdoor triggers. By enabling semantic interpretation of both graph patterns and client behaviors, the framework adaptively incorporates rule-based signals into message passing and client aggregation for defense. Experiments demonstrate that FedLSG significantly improves resistance to backdoor attacks without compromising graph integrity.

Problem

• Federated Graph Neural Networks
• Backdoor Poisoning
• Semantic Understanding
• Graph Integrity
• Stealthy Triggers

Innovation

• LLM-enhanced semantic calibration
• federated graph backdoor defense
• graph-to-text grounding
• student-teacher architecture
• LoRA-based reasoning