Detecting Webspam Beneficiaries Using Information Collected by the Random Surfer

Search engines use several criteria to rank webpages and choose which pages to display when answering a request. Those criteria can be separated into two notion, relevance and popularity. This notion of popularity is calculated by the search engine and is related to links made to the webpage. Malicious webmasters want to artificially increase their popularity, the techniques they use are often referred to as Webspam. It can take many forms and is in constant evolution, but Webspam usually consists of building a specific dedicated structure of spam pages around a given target page. It is really important for a search engine to address the issue of Webspam otherwise it won’t be able to provide users with fair and reliable results. In this paper we propose a technique to identify webspam through the frequency language associated with random walks amongst those dedicated structures. We identify the language by calculating the frequency of appearance of k-grams on random walks launch from every node.