“Visual Summarization of Web Pages”

Binxing Jiao, Linjun Yang, Jizheng, Feng Wu
Microsoft Research Area, Beijing

Summary:
Visual Summarization is an innovative new way of representing web pages in a brief yet comprehensive manner. There are mainly two achievements from such summarization. Firstly they act as an overview for webpage retrieval systems and users find it much feasible to look at glimpse of the webpage before visiting it. Secondly, in the task of re-finding visited web pages, visual summarization, are very helpful. Google Chrome, Mozilla FireFox and safari web browsers provide a visual list of most pages order by recent visit or most visits etc.
The Research carried on in the paper proposes a new web pages summarization technique and then compares it other visual summarization techniques.
• Thumbnails
• Visual Snippet
• Internal Images
• External Images (proposed)
Thumbnails are actual resized snapshot of the whole page, where the aspect ratio is kept same. Thumbnails are useful for well formed documents specially having large images and text sizes. Visual snippet provides a dynamic technique to create a composite image runtime which includes the dominant image, title and logo of the webpage. Dominant image are the one that provide proper summarizes the whole web page, and that exists inside the webpage itself. When a summarization is only the dominant image itself it is called Internal Image. But most of the web pages do not have proper Images to represent the webpage. In that scenario, the above techniques fail to provide a proper summarization. The research proposes a new technique External Images, in which internet is searched for representative image. In the technique, first key phrases are extracted from the webpage using KEX algorithm. Various images are retrieved using the key phrases and then the images are filter using cosine string comparison technique of images titles and other details. Ranking of images are created on the basis of visual alikeness with the main page. At the end the representative is selected on the basis of high precedence.
A statistical comparison is drawn between these visual summarization techniques of web pages by taking two kinds of tasks. First, to check the proper summarization of webpage and secondly, to check which one has least error rate in re-finding tasks. Research is concluded by the comparison results, which states that different visual summarization techniques are good for different types of web pages. Thumbnails are good for simple structured web pages, where snippet view is good for the website having many internal images and textual data. Internal and external images are dependent on availability of images in the web pages and proper extraction of key phrases from the page.

References
1. Susan Dziadosz and Raman Chandrasekar. Do thumbnail previews help users make better relevance decisions about web search results? SIGIR ’02 @
2. Zhiwei Li , Shuming Shi , Lei Zhang, Improving relevance judgment of web search results with image excerpts, conference on World Wide Web, Beijing
3. Qing Yu, Shuming Shi, Zhiwei Li, Ji-Rong Wen, and Wei-Ying Ma. Improve ranking by using image information. ECIR’07, Springer-Verlag, Berlin