VIPAS: Virtual Link Powered Authority Search in the Web

摘要

With the exponential growth of the World Wide Web, looking for pages with high quality and relevance in the Web has become an important research field. There have been many keyword-based search engines built for this purpose. However, these search engines usually suffer from the problem that a relevant Web page may not contain the keyword in its page text. Algorithms exploiting the link structure of Web documents, such as HITS, have also been proposed to overcome the problems of traditional search engines. Though these algorithms perform better than keyword-based search engines, they still have some defects. Among others, one major problem is that links in Web pages are only able to reflect the view of the page authors on the topic of those pages but not that of the page readers. In this paper, we propose a new algorithm with the idea of using virtual links which are created according to what the user behaves in browsing the output list of the query result. These virtual links are then employed to identify authoritative resources in the Web. Specifically, the algorithm, referred to as algorithm VIPAS (standing for virtual link powered authority search), is divided into three phases. The first phase performs basic link analysis. The second phase collects statistics by observing the user behavior in browsing pages listed in the query result, and virtual links are then created according to what observed. In the third phase, these virtual links as well as real ones are taken together to produce an updated list of authoritative pages that will be presented to the user when the query with similar keywords is encountered next time. A Web warehouse is built and the algorithm is integrated into the system. By conducting experiments on the system, we have shown that VIPAS is not only very effective but also very adaptive in providing much more valuable information to users.