Improved Distortion and Spam Resistance for PageRank
Year: 2018
Type: preprint
Abstract: Ranking functions are an integral component of web searching, and are thus subject to link spamming. Quite a lot of practical and theoretical knowledge has accumulated about which types of ranking functions are susceptible to which types of spam attacks, but there is no general characterization of the class of spam attacks nor of the susceptibility of a ranking function to arbitrary spam attacks.
Our first contribution is a characterization of the class of spam attacks and a definition of the spam resistance of a ranking function against all possible link-spam attacks. We show how this formal notion of spam resistance matches the intuition of practitioners. This definition allows us to evaluate novel ranking functions.
A natural way to resist spam is to design a ranking function that heavily favors trusted nodes, which are nodes known not be compromised by the spammer. However, this introduces local distortions into the ranking function. We formalize this notion of distortion and show that it matches the intuitive notion used by practitioners. We know of no ranking functions in the literature that combine high spam resistance and low distortion; we consider several well-known ranking functions (including the original form of PageRank) and show that they fail on one or both counts.
Finally, we introduce a new form of PageRank known as Min-PPR. We prove that Min-PPR has low distortion and high spam resistance. A secondary benefit is that Min-PPR comes with an explicit cost function on nodes that shows how important they are to the spammer; thus a ranker can focus their spam-detection capacity on these vulnerable nodes. Both Min-PPR and its associated cost function are straightforward to compute.
Source: arXiv (Cornell University)
Language: English
Topic: Spam and Phishing Detection
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2985086394