Data Entry: Please note that the research database will be replaced by UNIverse by the end of October 2023. Please enter your data into the system https://universe-intern.unibas.ch. Thanks

Login for users with Unibas email account...

Login for registered users without Unibas email account...

 
A Machine Learning Approach to Detect Differential Treatment of Anonymous Users
ConferencePaper (Artikel, die in Tagungsbänden erschienen sind)
 
ID 4649503
Author(s) Wagner, Isabel
Author(s) at UniBasel Wagner, Isabel
Year 2022
Title A Machine Learning Approach to Detect Differential Treatment of Anonymous Users
Editor(s) Atluri, Vijayalakshmi and Di Pietro, Roberto and Jensen, Christian D. and Meng, Weizhi
Book title (Conference Proceedings) Computer Security - ESORICS 2022
Place of Conference Copenhagen, Denmark
Year of Conference 2022
Publisher Springer Nature Switzerland
Pages 3-23
Abstract

Anonymous Internet use is essential to safeguard against mass surveillance and to protect privacy online. Tor Browser and the Tor anonymity network provide an effective and convenient way to browse the Internet anonymously. However, many websites make browsing inconvenient for anonymous users either by blocking access entirely, by blocking access to some functionality, or by using CAPTCHAs to make access more difficult. Prior work has relied on heuristics to study the extent to which anonymous users are treated differently. However, these heuristics either underestimated blocking or required extensive manual labeling. To address these shortcomings, here we propose a machine learning approach to detect when anonymous users are treated differently. We train binary and multi-class classifiers based on six feature sets and find that they perform very well on our test data (F1 scores 100%–94% for binary and 97%–84% for multi-class classifiers). Applying these classifiers to data collected from browsing 1,000 sites, including visits to subsites and executing search functionality, via 100 Tor exit nodes we find that 16.7% of landing pages inconvenience anonymous users, compared with 15.2% of subsites and 3.8% of search result pages. In particular, we find that websites hosted by Akamai, EdgeCast, and Cloudflare have significantly higher block rates than average, and that blocking of search results is dominated by Google which displays a block page or CAPTCHA for 39.8% of search result pages.

Series title Lecture Notes in Computer Science
Digital Object Identifier DOI 10.1007/978-3-031-17146-8_1
Document type (ISI) inproceedings
   

MCSS v5.8 PRO. 0.345 sec, queries - 0.000 sec ©Universität Basel  |  Impressum   |    
29/03/2024