Layout-Driven Page Rank
Would block-level analysis as opposed to page-level analysis of the Web produces better search engine results? Deng Cai of the University of Illinois at Urbana-Champaign thinks so.
The algorithm sections the page using horizontal and vertical lines, and assesses blocks of content according to their position on the page. By extracting page-to-block and block-to-page relationships (the former determined by page layout analysis and the latter determined by the likelihood of a block connecting to a given page), the prototype can rank Web pages and build page and block graphs; this info is then fed to link-analysis algorithms that evaluate each page’s importance according to the type of blocks that link into it.
Definitely an intriguing idea.
[ via ACM TechNews ]