Query-biased combination of evidence on the web

Recent research has shown that the hyperlink structure of the World Wide Web can be a rich source of information about the content of the environment, provided that there are effective means of understanding it. However, experience also suggests that while hyperlink structure analysis improves the overall precision of the top relevant documents for broad and generic queries, it usually fails to address the expectations of the users when the information need is precise and specific. Indeed, the authoritative measure of an information source, underpinning the hyperlink structure analysis, cannot be used as a substitute for relevance in the case of specific queries. We propose a query-biased combination of evidence mechanism that optimally merges results obtained by content and link analyses, based on the query characteristics. The combination mechanism is modelled using Dempster-Shafer’s theory of evidence, and the query characteristics are defined using a probabilistic propagation mechanism on top of the hierarchical structure of concepts provided by WordNet. Our evaluation experiments indicate that the proposed query-biased merging technique yields some interesting conclusions. 1