The bipartite clique-a topological paradigm for WWWeb user search customization

Web user search customization research has been fueled by the recognition that if the WWW is to attain to its optimal potential as an interactive medium the development of new and/or improved Web resource classification (page identification, referencing, indexing, etc) and retrieval/delivery systems supportive of and responsive to user preference is of prime importance. User preference, as it relates to a Web user's search agenda, entails maintaining the user as director of his search and expert as to which Web pages are relevant. In our work Web usage and Web structure mining are employed in a theoretically skillful way to produce a strongly connected virtual bipartite clique (biclique) search neighborhood of high quality pages of relevance to the Web user's search objective. Our algorithm is designed to exploit linkage data inherent in Web access logs using the Combined Log Format (C_BLF) to assemble a referer partite set of pages consistent with the user's preference and search intent (members: user's initial choice of a Web resource/page and other relevant authoritytype pages) and a request partite set (members: pages with incoming links from the referer partite). The Web user's initial page of choice becomes the first member of the referer partite and gatekeeper to the biclique neighborhood. Our algorithm uses a Web site's collective user's history (log entries) in a collaborative manner to identify and further qualify pages of relevance for membership in the appropriate partite set. Web user search customization strategically fostered by our algorithm enhances the efficiency and productivity of a Web user's activity in three ways: (1) it delivers high quality pages organized hierarchically to facilitate the user's ready assessment of the Web site's benefit to his search objective, thus minimizing time spent at an unfruitful site, (2) it facilitates ease of navigation in either breadth-first, depth-first or a combination of the and (3) it nullifies time spent locaitng and traversing path to pages hosted in much-to-large distributed search environments.