We have also combined some of the features used in PhishPatrol with other approaches to detect phishing Web sites. Jason Hong has been leading our group’s development of a tool called CANTINA, which analyzes the content of a Web page in combination with other heuristics to determine whether or not the page is part of a phishing site. CANTINA first employs a well-known information-retrieval algorithm to identify five terms that are important on a given Web page but relatively uncommon on the Internet overall. For example, on an eBay log-in page, this “lexical signature” might be, “eBay, user, sign, help, forgot.” If you were to search for these five terms using Google, the legitimate eBay log-in page would appear among the top search results. Phishing Web sites that have replicated the eBay log-in page are unlikely to appear because one of the criteria Google’s proprietary algorithm uses in ranking a Web page is the number of links to it from other pages on the Internet, so legitimate pages are more likely to be in the top results. This approach is not foolproof, however, especially if a legitimate site was recently created; thus, it is only one of several features that CANTINA considers in assessing a Web site.
The Evolving Threat
Read Comments (11) | Post a comment



