Uma abordagem evolutiva para combinação de fontes de evidência de relevância em máquinas de busca

Abstract

Modern search engines use different strategies to improve the quality of their answers. An important strategy is to get an ordered list of documents based on lists produced by different sources of evidence. This work studies the use of a evolutionary technique to generate good functions of combination of three different sources of evidence: the textual content of the documents, the connecting structures between the documents in a collection and the concatenation of anchor texts pointing to each document. The functions Combination findings in this study were tested in two separate collections: the first contains queries and document a real Web search engine that contains some 12 million documents and the second is to LETOR reference collection, created to allow the fair comparison between collating functions learning methods. The experiments indicate that the studied approach here is a practical and effective alternative to combining different sources of evidence in a single list of answers. We also checked different query classes require different functions combination of sources of evidence and show that our approach is feasible to identify good features.