Effective page segmentation combining pattern analysis and visual separators for browsing on small screens

Page segmentation plays a key role in browsing on small screens. It breaks a large page into smaller segments according to their semantic relationships. Then, various approaches such as single column adaptation and thumbnail view with zooming links can be implemented based on these page segments. However, for current flexible web pages, segmentation remains a challenging task. This paper proposes an effective automatic segmentation method which combining pattern analysis and visual separators. The basic idea is that a page's semantic structure is largely reflected by repeated continuous patterns and visual separators, which coincides with human's visual perception. The proposed method works in three steps: generating a refined tag tree from the DOM tree, recognizing and merging inexact patterns recursively, and segmenting the others by visual separators. Our experimental results show that the proposed method outperforms existing methods, especially for pages automatically generated from templates.