Enhancing Efficiency of Web Search Engines through Ontology Learning from Unstructured Information Sources

With the fast growth rate of information availability through the World Wide Web, search engines´ ranking become limited to deal with such enormous amount of information. Web search engines should be enriched with methodologies that enable it to understand the content of Web pages, then to align pages to the correct query category that highly match its content. In this paper, a proposed system is introduced to deal with the abundance of information by automatically understand the content of a Web page, and semantically model the ontological concepts that exist inside it. The semantic relations between ontological concepts are automatically given a score or weight based on its influence to the given query. The weighted semantic relations between ontological concepts can be viewed as a signature for the query, the highly similarity of an article to this signature, the more relevant to the query. A new relevancy measure is introduced to semantically re-rank or classify Web pages based on computing the semantic similarity of the weighted intersection ratio between ontological concepts extracted from retrieved Web pages, and ontological concepts that represents the query. Results shows that the proposed system has the highest Pearson correlation coefficient (0.890) to human judgments which outperforms semantic similarity state-of-the-art methods and Web-based methods. The proposed model, was tested to re-rank Web pages according to the semantic relevancy of the query, experiments shows that it has the highest convergence to expert ranking order of Web pages compared to other Web search engines.