CRS:A Website Fingerprinting Technique for Dark Web Tor Traffic

With the development of internet and communication technologies, life has become more convenient, but privacy and security issues have emerged as a consequence. While the Tor network protects privacy, it is often used for illegal activities due to its anonymity features, posing a threat to societal security. In Tor network traffic identification, traditional methods face challenges in feature extraction and model degradation. To address these issues, this study proposes a website fingerprinting model for Tor traffic, named CRS. The model mitigates the degradation problem of deep networks through skip connections and enhances the accuracy and flexibility of feature extraction via adaptive similarity calculation. Evaluation results show that the CRS model achieves an identification accuracy of over 98% for undefended traffic in closed environments; 91.32% under WTF-PAD defense, and 49.7% under Walkie-Talkie defense. In open environments, the precision and recall for undefended traffic both reach 98%; under WTF-PAD defense, precision is 90%, and recall is 94%. Even under both WTF-PAD and Walkie-Talkie defenses, the CRS model’s precision and recall remain significantly higher than those of other comparison models. These results validate the superior performance of the CRS model, highlight the importance of effectively addressing various defense mechanisms, and provide new insights for dark web traffic analysis and network security defense, with significant theoretical and practical value.