网络与信息安全学报 (Aug 2022)
Research on link detection technology based on dynamic browser fingerprint
Abstract
With the emergence of user privacy leakage and online fraud, traditional detection mechanisms can no longer stop the intensified attacks and new methods are needed to assist in web security detection.As a bridge between users and network data, browsers are widely used in the interaction between users and various applications.The browser fingerprint contains various characteristic information of the user’s browser and device, and its uniqueness can greatly improve the accuracy of user identification.However, browser fingerprints may change over time, and solutions based on static fingerprints cannot meet the requirements of dynamic fingerprints detection.Motivated by this challenge, it’s necessary to improve the recognition and link detection capabilities of dynamic browser fingerprints.Besides, parameters from multi-dimensional perspective need to be obtained and screened to get fine-grained and high-discrimination features for model detecting, preventing the calculation cost caused by too many unnecessary parameters.Bidirectional Gating Recurrent Unit (BiGRU) was introduced into the fingerprint detection, and a stacked BiGRU detection model based on dynamic browser fingerprints was proposed.It solved the problem of low detection accuracy caused by ignoring the interaction of forward and backward information in the process of fingerprints detection, and it also had a certain improvement in indicators such as accuracy and linking time.Furthermore, in view of the different importance of different feature parameters, the same weight given to parameters will result in detection error.Combined with the attention mechanism to focus more stable fingerprint features, an Att-BiGRU fingerprint detection model based on dynamic browser fingerprints was proposed.It optimized the detection ability compared to BiGRU, further improved the detection accuracy of browser fingerprinting, and showed good linking ability on the comprehensive index under the benchmark model.
