Using Function Call Sequence for Browser Fingerprinting Detection
Duoshi Cheng · 2022
Browser fingerprinting is a stateless tracking technique that combines browser attributes into a single identifier for tracking users. As browser fingerprinting is used by a growing number of websites, many works have focused on browser fingerprinting detection. Existing works mostly consider detection through the perspective of script, it will case false negative when script is split. In this paper, we propose BFDetector, an approach for better detecting browser fingerprinting, to solve the problem of script splitting. To this end, BFDetector uses function call sequence of web pages to detect browser fingerprinting through NLP-based one-class deep learning ensemble. We collected 7,800 function call sequences as dataset from Alexa top-10K websites’ homepage. After collecting these function call sequences, we trained a Javascript API embedding to convert each function to a vector for training deep leaning classifier. Through the test of our own dataset, the accuracy of classifier has reached 95.53%. Afterward, we analysed the characteristic of some known fingerprinting techniques’ function call sequence and found a newly fingerprinting technique that we call it media-based fingerprinting.