A Multi-modal Approach for Context-aware Network Traffic Classification

Network traffic classification is important for network security and management. State-of-the-art classifiers use deep learning techniques to automatically extract feature vectors from the traffic, which however lose important context of the communication sessions and encapsulated text semantics. In this paper, we present a Multi-Modal Classification method named \textit{MTCM} to systematically exploit the context for the classification task. We build an adaptive context-aware feature extraction framework over varying-length and dynamic packet sequences, based on the attention-aware graph neural networks and BERT. We next automatically fusion multi-modal features with the Multi-Layer Perception (MLP) that unifies the graph and semantic features for the packet stream. Extensive evaluation with real-world application and abnormal network datasets show that MTCM outperforms state-of-the-art deep learning methods, and is robust for different classes of traffic data sets.