A Multi-tab Webpage Fingerprinting Method Based on Multi-head Self-attention

The Tor anonymous communication network can provide Internet anonymous access function, making it challenging for network regulators to track the webpages visited by users. However, webpage fingerprinting, a commonly used passive traffic analysis technology, can identify webpages by monitoring and analyzing the near-end traffic of users. Currently, most existing webpage fingerprinting methods assume that users only open a single tab to access one webpage, which is not realistic, while multi-tab methods have limitations in utilizing mixed areas and the number of identified webpages. To solve the above limitations, this paper proposes a Multi-head Self-attention-based Multi-tab Webpage Fingerprinting (MSMWF) method on Tor, which designs a reasonable network structure according to the type of mixed webpages in multi-tab webpage traffic sequence. First, sequence embedding and block division are performed on the original multi-tab webpage traffic sequence to generate the embedded vector. Then, three sets of self-attention heads are used to extract the global features of the three types of mixed webpages, which can effectively use the correlation between different regions of the multi-tab sequences to identify the corresponding webpages. Experimental results demonstrate that MSMWF outperforms baseline methods in identifying multi-tab webpages and performs well in various experimental scenarios.