南京邮电大学学报（自然科学版）2026，Vol.46Issue(3)：1-13,13.DOI:10.14132/j.cnki.1673-5439.2026.03.001
PGCTC:基于图对比学习的PCDN流量识别
PGCTC:graph contrastive learning for PCDN traffic classification
摘要
Abstract
With the increasing demand for large-scale content services,the PCDN architecture that inte-grates P2P mechanisms has significantly improved network resource utilization and reduced operational costs for content providers,thanks to its distributed content transmission mode and lower service costs.But at the cost of sacrificing the upstream bandwidth of home broadband users,PCDN seriously affects the Internet experience of broadband users across the entire network.Therefore,precise identification and effective control of PCDN traffic have become key issues that communication operators urgently need to address.However,PCDN traffic recognition faces three major challenges:hybridity,encryption,and homogenization,which make it difficult for traditional recognition methods to distinguish them finely.To address these challenges,this paper proposes a graph comparison learning method named PGCTC for PCDN traffic.This method mines inter stream structural dependencies to enhance the ability to express implicit semantics in encrypted communication.It models indirect collaborative relationships between neighboring nodes to effectively alleviate recognition interference caused by mixed flow of PCDN.Fur-ther,a graph comparison mechanism is introduced to enhance the fine-grained discrimination ability for homogeneous traffic,achieving unified modeling and accurate classification of encrypted flows,collab-orative flows,and homogeneous flows.Experimental results on the CICIDS2017 dataset and a self built real PCDN dataset show that our method outperforms existing comparison methods in all indicators,dem-onstrating good effectiveness and cross scenario generalization performance.关键词
流量识别/P2P-CDN/图神经网络/图对比学习Key words
traffic classification/P2P-CDN/graph neural network/graph contrastive learning分类
信息技术与安全科学引用本文复制引用
王攀,付虹蕾,李泽一,马媛媛,张桂玉..PGCTC:基于图对比学习的PCDN流量识别[J].南京邮电大学学报（自然科学版）,2026,46(3):1-13,13.基金项目
国家自然科学基金(61972211)资助项目 （61972211）
