| 摘 要: 随着加密流量成为网络传输的主流,传统的单流分类方法面临着特征稀疏与上下文缺失的挑战。现有的数据增强技术大多局限于样本层面,难以恢复流量复杂的时空结构特征。针对这一问题,本文提出了一种基于语义感知聚类的加密流量结构化数据增强方法。通过融合语义特征与时间特征的聚类算子,将离散的原始流量重构为具有完整行为语义的流簇。该方法利用深度学习模型提取的高维语义特征,有效解决了传统时序聚类在多应用并发场景下的混淆问题,实现了高纯度的上下文聚合。实验结果表明,经本方法增强后的结构化数据,能够使 Random Forest、LSTM、ResNet 等多种基准模型在准确率上有较大提升。 |
| 关键词: 加密流量分类 结构化数据增强 流聚类 并发流量分析 |
|
中图分类号:
文献标识码:
|
|
| Structured Data Augmentation Method for Encrypted Traffic Based on Semantic-Aware Clustering |
|
zhouli, chenwei, zhangyiting
|
Nanjing University of Posts and Telecommunications
|
| Abstract: With encrypted traffic becoming the mainstream of network transmission, traditional single-flow classification methods face challenges such as sparse features and missing contextual information. Most existing data augmentation techniques are limited to the sample level and struggle to restore the complex spatiotemporal structural characteristics of traffic. To address this issue, this paper proposes a semantic-aware clustering-based structured data augmentation method for encrypted traffic. By integrating clustering operators that combine semantic and temporal features, discrete raw traffic is reconstructed into flow clusters with complete behavioral semantics. This method leverages high-dimensional semantic features extracted by deep learning models, effectively resolving the confusion issues of traditional temporal clustering in multi-application concurrent scenarios and achieving high-purity context aggregation. Experimental results demonstrate that the structured data augmented by this method significantly improves the accuracy of various benchmark models, including Random Forest, LSTM, and ResNet. |
| Keywords: encrypted traffic classification structured data augmentation flow clustering concurrent traffic analysis |
