TB-WPRO: Title-Block Based Web Page Reorganization

2. Related Work

A variety of methods have been proposed to segment web pages into small blocks. The early way is considering the tag information in the DOM tree such as <P> (paragraph), <TABLE> (table), <UL> (list). <H1>-<H6> (heading), etc. Lin and Ho (2002) partition a page into several content blocks according to HTML tag <TABLE> and then use entropy to distinguish redundant block from the information block. Hattori et al. (2007) propose a hybrid method using content-distances and layout information to segment the webpage. However, the HTML tags do not contain any semantic information and could be misused. A FOM model is proposed by Chen et al. (2001). It treats each object in the webpage as either a basic or composite FOM which describe the object functionality. But the model does not describe the concrete segmentation method. The VIPS algorithm is proposed by Cai et al. (2003) which analyzes the webpage using the vision separator to segment the content into different areas. However, a parameter PDoC (Permitted Degree of Coherence) must be given by the user to get segmentation result. The paper does not tell how to determine proper PDoC to get blocks visually and semantically consistent with the original web page. Baluja (2006) uses a learning method to divide a webpage into 9 parts which the user could select to zoom. Chakrabarti et al. (2004) puts the nodes in DOM tree into a weighted graph, and formulates an appropriate optimization problem on it. The optimization problem is solved by a learning framework. But such machine learning method need a training process and is difficult to implement.