Link spam detection based on genetic programming

Link spam refers to unfairly gaining a high ranking on search engines for a web page by means of trickily manipulating the link graph to confuse the hyper-link structure analysis algorithms. It seriously affects the quality of the search engine query results. Detecting link spam has become a big challenge for web search. This paper proposes to learn a discriminant function to detect link spam by genetic programming. In this article, the representation of individuals, the genetic operators and the fitness function are studied. The experiments on WEBSPAM-UK2006 are carried out to find the preferable parameters and evaluate the validity of genetic programming. The experimental results show that this method can improve spam classification recall by 27.5%, F-measure by 12.1% and accuracy by 4.6% compared with SVM.