A new webpage classification model based on visual information using gestalt laws of grouping

Abstract

Traditional text-based webpage classification fails to handle rich-information-embedded modern webpages. Current approaches regard webpages as either trees or images. However, the former only focuses on webpage structure, and the latter ignores internal connections among different webpage features. Therefore, they are not suitable for modern webpage classification. Hence, semantic-block trees are introduced as a new representation for webpages. They are constructed by extracting visual information from webpages, integrating the visual information into render-blocks, and merging render-blocks using the Gestalt laws of grouping. The block tree edit distance is then described to evaluate both structural and visual similarity of pages. Using this distance as a metric, a classification framework is proposed to classify webpages based upon their similarity.