Aprendendo funções de ranking baseadas em blocos usando programação genética

Abstract

Today, the Internet is considered a powerful tool of communication and information. Its impact on society is increasing more and more, which means that it is becoming indispensable. In this context information searching systems are becoming increasingly important. In this paper, we propose a new search method capable of learning ranking functions that explore Web pages structure in blocks, using genetic programming. Different from previous works, our method allows combining traditional evidence in information retrieval with evidence derived from the structure of Web pages. To validate the proposed method, we use three real collections of pages (IG, CNN and BLOG). Experimental results show that our approach is able to overcome the results of a baseline of information which uses blocks information without learning machine, presenting precision benefits (MAP) of 9.38% in the IG collection, from 7.13% in CNN, and 25.87% in collection BLOG. Regarding our second baseline, which uses genetic programming out of traditional evidence in information retrieval, our method achieved benefits of 5.25% in the IG collection, 10.37% and 4.37% on CNN in collection BLOG.