摘要
网络黑灰色产业正面临技术对抗升级与隐蔽传播的双重挑战,传统检测方法面临语义解析维度缺失、特征更新滞后等系统性瓶颈。提出了基于大语言模型(LLM)的智能分类框架,通过角色认知强化与上下文情感增强的双轮驱动机制,突破传统方法的性能天花板,实现F1分数0.913的分类效能。该研究将LLM技术与动态对抗场景深度融合,不仅构建了“语义消解—知识进化—多模态推理”三位一体的技术防御体系,更开发出涵盖“数据采集—模型优化—场景模拟”全流程的AI+安全教育实践平台,为网络空间治理提供了技术创新与人才培养的协同解决方案。
The black and grey industries on the Internet are facing the dual challenges of technological confrontation escalation and covert dissemination.Traditional detection methods are confronted with systemic bottlenecks such as the lack of semantic analysis dimensions and the lag in feature updates.This research innovatively proposes an intelligent classification framework based on a Large Language Model(LLM).Through the dual-driven mechanism of role cognition enhancement and context emotional enhancement,it breaks through the performance ceiling of traditional methods and achieves a classification effectiveness with an F1 score of 0.913.This study deeply integrates LLM technology with dynamic confrontation scenarios.It not only constructs a trinity technical defense system of“semantic resolution-knowledge evolution-multimodal reasoning”,but also develops an AI+safety education practice platform covering the whole process of“data collection-model optimization-scenario simulation”,providing a collaborative solution for both technological innovation and talent cultivation in cyberspace governance.
作者
谢园园
郭燕慧
彭成智
陈浩然
李蔚然
毛文莉
Xie Yuanyuan;Guo Yanhui;Peng Chengzhi;Chen Haoran;Li Weiran;Mao Wenli(China Information Technology Designing&Consulting Institute Co.,Ltd.,Beijing 100048,China;Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《邮电设计技术》
2025年第9期1-8,共8页
Designing Techniques of Posts and Telecommunications
关键词
黑灰产
网页分类
大语言模型
角色设定
情感引导
Black and grey industries
Webpage classification
Large language model
Role setting
Emotional guidance
