Measuring and Analyzing Search Engine Poisoning of Linguistic Collisions

Explore a comprehensive analysis of linguistic-collision search poisoning attacks in this 18-minute IEEE conference talk. Delve into the first large-scale study examining 1.77 million misspelled search terms on Google and Baidu, focusing on English and Chinese languages. Learn about the sophisticated attack method targeting queries where misspelled terms are legitimate words in other languages, evading auto-correction mechanisms. Discover how a deep learning model improves the collection rate of linguistic-collision search terms by 2.84x compared to random sampling. Gain insights into the prevalence of abuse, with 1.19% of linguistic-collision search terms on major search engines leading to malicious websites. Understand the main target categories for cybercriminals, including gambling, drugs, and adult content. Examine the disproportionate impact on mobile device users and explore potential mitigation strategies for this emerging threat in search engine poisoning.

Introduction
Why Search is Important
Spelling Mistakes
Snickers
Auto Correction
Linguistic Collisions
English
Categories
English misspellings
Chinese misspellings
Character level recurrent neural network
Pinyin input
Fuzzy Pinyin input
Google and Baidu
Google English
Google Chinese
Alexa
Devices
Languages
Conclusion
Questions
Question from FTC
Promoting Poison URLs