Learning to detect web spam by genetic programming

Web spam techniques enable some web pages or sites to achieve undeserved relevance and importance. They can seriously deteriorate search engine ranking results. Combating web spam has become one of the top challenges for web search. This paper proposes to learn a discriminating function to detect web spam by genetic programming. The evolution computation uses multi-populations composed of some small-scale individuals and combines the selected best individuals in every population to gain a possible best discriminating function. The experiments on WEBSPAM-UK2006 show that the approach can improve spam classification recall performance by 26%, F-measure performance by 11%, and accuracy performance by 4% compared with SVM.