Intelligent modeling of the internet surfer behavior and application in web mining

Abstract

With the booming development of the Internet, we must sort though piles of information and misinformation in order to find knowledge. However, several technologies such as PageRank algorithm, HITS algorithm and QD-PageRank algorithm have been successfully applied to get good ranked search results in the Web. Web search engines (Aol, Ask, Bing, Yahoo, Google, ... with Google the top) have become the most important Internet tools to retrieve information. They use different ways to find relevant pages. For instance, Google search engine uses the PageRank algorithm to find relevant pages. Netscape uses web page content analysis, usage pattern information, as well as linkage analysis to find relevant pages. Highlighted in these algorithms used by different web search engines, is the inability to incorporate a simultaneous multiple-term query with a relating semantic at runtime. And yet, an ideal web search tool would be able to match the search queries to the exact context and return results within that context. To counteract this deficiency, we propose in this thesis the “Onto SQD-PageRank algorithm” that combine links, web contents and context information based on simultaneous multiple-term query. The aim of Onto SQD-PageRank is to improve the QD-PageRank algorithm by introducing a simultaneous multiple-term query and taking into account the semantic dependencies between words. The premise behind Onto SQD-PageRank algorithm is to understand what the web surfer is searching for. Experimental studies of Onto SQD-PageRank algorithm have been performed in a local database. Results show that Onto SQD-PageRank algorithm significantly outperforms the existing algorithms in the quality of the pages returned, while remaining efficient enough to be used in today's large search engines.