Network flow detection of semantic relationship between flow and byte
Yiqing Luo, Mingshu He, Xiaojuan Wang, Lei Jin · 2022
Starting from the semantic relationship between network flow and bytes in the original data, this study proposes a network flow detection method based on graph convolution model and transformer structure. The network flow is encoded into a complete network flow language according to the smallest unit byte, and a topological relationship heterogeneous graph containing the relationship between flow and bytes is constructed. The graph convolution model is used to extract the global information in the topological relationship heterogeneous graph, which is named Global embedding. At the same time, A coding structure based on global embedding, learning embedding and location embedding is proposed, and the encoder in transformer structure is used to extract the characteristics of network flow. The experimental results show that this method is effective. The extracted global embedding can better characterize all kinds of network flows and better realize network flow detection and classification. While improving the detection and classification effect, text classification provides a new method for network flow detection.