CQNet: A Clustering-Based Quadruplet Network for Decentralized Application Classification via Encrypted Traffic

Abstract

Decentralized applications (DApps), along with the development of blockchain technology, are increasingly developed and deployed on blockchain platforms. DApps based on the same platform usually adopt similar traffic encryption settings and the same communication interface, leading to traffic less distinguishable. However, existing classification methods either require manual-design features or need lots of data to train the classifier, otherwise suffering from low accuracy. In this paper, we apply metric learning to DApps encrypted traffic classification problem and propose the clustering-based quadruplet network (CQNet). The CQNet can filter out useless samples to reduce the training dataset’s redundancy data by utilizing the proposed algorithm, thereby improving the classifier’s efficiency. Moreover, we adopt a quadruplet structure that can mine more restrictive relationships among quadruplets and provide rich information to the classifier. Our comprehensive experiments on the real-world dataset covering 60 DApps indicate that CQNet can achieve excellent performance with high efficiency and is superior to the state-of-the-art methods in terms of accuracy and efficiency.