Aprendendo a segmentar páginas web

Abstract

Unlike traditional documents, Web pages are composed of different segments or blocks, each block has specific functions in each page. Recent work in the literature has shown that information on these segments may be useful to improve the results of numerous tasks in information retrieval and data mining areas. For this reason, there are many scientific works proposing different methods for Web pages segmentation. Generally speaking, the targeting methods found in the literature only use evidences of the page to be segmented. However, based on the observation that the pages of a site tend to have very similar layouts, we present a strategy based on machine learning that explores overall evidences of Web sites. Our method, which adopts Support Vector Machines for the learning process, and use the SOM structure (Site Object Model) to aggregate information from all pages of aWeb site, achieved good results when compared a manual segmentation approach, and with a recent approach in the literature.