Using neural network to combat with spam pages
Moein Shahbazi, Ali Mohammad Zareh Bidoki, Meysam Shehni Birgan · 2014
Search engines resolve the most informational needs of users by indexing huge amounts of data from web pages. In this process, spam pages prevent users from reaching their desirable results. Spam pages use deceptive methods to get a higher rank than their real one in search engines. For a human expert, recognition of spam pages is an easy task, but it is too complicated for a machine. Regarding the large size of the web graph and the large number of web pages, leaving the whole task to human is impossible. As a solution to this challenge, in this paper we propose a semiautomatic method using a combinational ranking based on links between pages. At first, two valid sets of spam pages are specified by experts. Then a multilayered neural network trained by genetic algorithm will be used to calculate a global rank for the web graph. In this network the rank of spam pages is low. To evaluate the precision and performance of the proposed method, the Persian web graph corpus indexed by Parsijoo search engine is used. Experimental results show a better performance in comparison to other methods.