Measuring and Analyzing Search Engine Poisoning of Linguistic Collisions

Matthew Joslin, Neng Li, Shuang Hao, Minhui Xue, Haojin ZhuIEEE Symposium on Security and Privacy 2019 · OpenReview

Abstract

Misspelled keywords have become a lucrative target for search engine poisoning (SEP). While search engines like Google and Baidu have implemented sophisticated auto-correction mechanisms, a specific class of attack—linguistic-collision misspelling—evades these defenses. These attacks exploit typos that are legitimate words in other languages or contexts, preventing auto-correction and allowing malicious sites to rank for low-competition terms. This paper presents a large-scale analysis of 1.77 million misspelled terms across English and Chinese, revealing that approximately 1.19% of these results lead to malicious content, particularly in high-risk categories like gambling and illicit drugs.

The Rise of Linguistic-Collision Attacks

Search engines are the primary gatekeepers of the internet. To improve user experience, they automatically correct common typographical errors (e.g., changing "adoeb" to "adobe"). However, a "semantic gap" exists when a user types a word that is a misspelling in one language but a valid word in another. For instance, "idobe" is a typo for "adobe" but a legitimate word in Nigerian. In such cases, search engines often assume the input was intentional and display results for the misspelled term without correction.

Figure 4: An example of search results for "Cilis" (a misspelling of Cialis) leading to malicious online pharmacies.

Cybercriminals exploit this gap to "poison" search results. Because these linguistic collisions have significantly lower search competition than established brand names, attackers can easily secure first-page rankings for malicious links. This strategy is particularly effective for targeting mobile users, who are more prone to "fat-finger" errors on small screens.

Methodology: Identifying the Semantic Gap

To systematically study this phenomenon, the researchers developed a multi-stage pipeline designed to generate, identify, and validate linguistic-collision search terms.

Figure 5: The research pipeline for identifying and validating linguistic-collision search engine poisoning.
  1. Misspelling Generation: The team used a modified Damerau-Levenshtein edit distance to simulate realistic typos, such as adjacent key presses on a QWERTY keyboard. For Chinese, they focused on Pinyin phonetic errors, such as confusing "s" and "sh" sounds.
  2. RNN-based Prediction: To scale the analysis, they trained a Recurrent Neural Network (RNN) with Long Short-Term Memory (LSTM) units. This model calculates the "entropy" of a typo; low entropy indicates the typo resembles a legitimate word, making it a prime candidate for evading auto-correction. This approach improved the discovery rate by 2.84x compared to random sampling.
  3. Validation: The researchers crawled 1.77 million search results from Google and Baidu, using VirusTotal and manual inspection to identify malicious URLs.

Analysis and Key Findings

The study reveals that search engine poisoning via linguistic collisions is a prevalent and targeted threat.

  • Prevalence: Roughly 1.19% of non-auto-corrected misspelled terms on major search engines lead to malicious websites on the first page.
  • Category Bias: Attackers do not target all keywords equally. High-value illicit categories like "Drugs" and "Gambling" are poisoned at four times the rate of benign categories like "Food" or "Clothing." In the "Adult" category on Google, the poisoning rate reached 3.47%.
  • Mobile Vulnerability: Search volume data indicates that mobile users search for misspelled keywords at a much higher rate than desktop users. This makes the mobile demographic the primary victim of these attacks.
  • Platform Differences: The study found that Baidu's ranking algorithm is generally more susceptible to these attacks, often prioritizing linguistic-collision results over reputable domains more aggressively than Google.

Conclusion and Future Mitigation

Linguistic-collision search poisoning represents a sophisticated evolution of traditional typosquatting. By hiding in the "semantic gap" between languages, attackers bypass automated safeguards.

To mitigate this threat, search engines could integrate translation APIs to detect when a search term abruptly switches languages in a way that is inconsistent with the user's profile. Furthermore, addressing ad-policy gaps—where even legitimate companies bid on misspelled keywords—could reduce the "blueprint" available to attackers. As mobile internet usage continues to grow, securing the search experience against these subtle linguistic exploits remains a critical challenge for cybersecurity.

Summary updated

Abstract

Misspelled keywords have become an appealing target in search poisoning, since they are less competitive to promote than the correct queries and account for a considerable amount of search traffic. Search engines have adopted several countermeasure strategies, e.g., Google applies automated corrections on queried keywords and returns search results of the corrected versions directly. However, a sophisticated class of attack, which we term as linguistic-collision misspelling, can evade auto-correction and poison search results. Cybercriminals target special queries where the misspelled terms are existent words, even in other languages (e.g., "idobe", a misspelling of the English word "adobe", is a legitimate word in the Nigerian language). In this paper, we perform the first large-scale analysis on linguistic-collision search poisoning attacks. In particular, we check 1.77 million misspelled search terms on Google and Baidu and analyze both English and Chinese languages, which are the top two languages used by Internet users. We leverage edit distance operations and linguistic properties to generate misspelling candidates. To more efficiently identify linguistic-collision search terms, we design a deep learning model that can improve collection rate by 2.84x compared to random sampling. Our results show that the abuse is prevalent: around 1.19% of linguistic-collision search terms on Google and Baidu have results on the first page directing to malicious websites. We also find that cybercriminals mainly target categories of gambling, drugs, and adult content. Mobile-device users disproportionately search for misspelled keywords, presumably due to small screen for input. Our work highlights this new class of search engine poisoning and provides insights to help mitigate the threat.
BibTeX
@inproceedings{joslin2019measuring,
  author = {Matthew Joslin and Neng Li and Shuang Hao and Minhui Xue and Haojin Zhu},
  title = {{Measuring and Analyzing Search Engine Poisoning of Linguistic Collisions}},
  booktitle = {IEEE Symposium on Security and Privacy},
  year = {2019},
  url = {https://openreview.net/forum?id=DuN8FDsgcd_7},
}
Figure