A knowledge-based approach to scenario-specific medical free-text retrieval

This dissertation studies the challenges faced in supporting scenario-specific medical free-text retrieval with documents in online textual databases. A scenario is typically defined as a frequently-reappearing medical task. Scenario-specific queries are important because recent studies reveal that such queries prevail in clinical practice. We concern ourselves with answering scenario-specific queries using online textual databases, since such databases have become the most valuable information sources in the medical domain. Although the amount of information in such databases increases rapidly and the quality of such information is high, the utilization of these databases in clinical practice is low because of the following major challenges, which are not handled well by existing retrieval systems.;The first challenge is to automatically identify relevant databases with high accuracy. Towards this goal, we first develop a probabilistic relevancy model which estimates the relevancy of each database more accurately than traditional models. We further develop an adaptive probing technique that contacts a few databases on the fly to obtain their exact relevancy measures, and consequently to select the best databases with higher accuracy.;The second challenge in our study is to resolve query-document mismatch, mismatch between the general scenario terms in a query and specialized terms in relevant documents. We propose a knowledge-based query expansion technique which, based on a domain knowledge source, automatically identifies the specialized concepts specifically related to the original query's scenario. The technique further expands such specialized concepts to the original query, making the query a better match with relevant documents and leading to improved retrieval effectiveness.