Improving Information Access by Relevance and Topical Feedback

One of the main bottle-necks in providing more effective information access is the poverty of the query end. With an average query length of about two terms, users provide only a highly ambiguous statement of the, often complex, underlying information need. Implicit and explicit feedback can provide us with additional information that can help disambiguate the query and provide more focused search results. We investigate the effects of using different types of feedback. Retrieval results of pseudo-relevance, explicit relevance and topical feedback are compared. Although on average explicit relevance feedback in combination with pseudo relevance feedback works best, for individual queries results are unpredictable. There is a large potential for improvement if we can predict which type of feedback will perform best for a query. Since we are dealing with feedback potentially provided by users standard evaluation measures are not sufficient to evaluate feedback techniques, and the quality of user interaction should also be taken into account.