Scholarlink

This paper investigates the application of large language models (LLMs) in email anomaly detection, proposing a federated learning framework that integrates RoBERTa’s advanced semantic capabilities with differential privacy and a dynamic reputation mechanism to address privacy and security challenges. Our approach employs adapter modules for parameter-efficient fine-tuning, enabling localized model adaptation without compromising data privacy. The framework is evaluated across four diverse datasets, demonstrating robust performance comparable to centralized models while ensuring resilience against adversarial interference. Experimental results highlight the potential of federated LLMs as a scalable and privacy-preserving solution for email anomaly detection, particularly in scenarios with limited labeled data and frequent model updates.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Spam Detection Model With Federated Learning and Large Language Model

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jiateng Zhao,
  • Bin Wen,
  • Jiashuai Yang,
  • Shang Zhou

摘要

This paper investigates the application of large language models (LLMs) in email anomaly detection, proposing a federated learning framework that integrates RoBERTa’s advanced semantic capabilities with differential privacy and a dynamic reputation mechanism to address privacy and security challenges. Our approach employs adapter modules for parameter-efficient fine-tuning, enabling localized model adaptation without compromising data privacy. The framework is evaluated across four diverse datasets, demonstrating robust performance comparable to centralized models while ensuring resilience against adversarial interference. Experimental results highlight the potential of federated LLMs as a scalable and privacy-preserving solution for email anomaly detection, particularly in scenarios with limited labeled data and frequent model updates.