Medical query generation by term–category correlation

Natural language descriptions are helpful for users to precisely describe medical information needs. However search engines often operate on keyword-based queries. Generating keyword-based queries from the descriptions is thus essential. Its goal lies in retrieving more relevant information that may be ranked high for easy access. In response to the goal, we present a technique MQG (Medical Query Generator) that, given an information need description, generates a query by selecting (from the description) those terms having stronger correlation to medical categories. Empirical evaluation on a medical text database OHSUMED shows that MQG greatly outperforms several state-of-the-art techniques, including those that expand queries by a complete dictionary of medical terms and their equivalence terms in retrieval. Moreover, it reduces the load incurred to the text ranker by retrieving fewer documents for ranking. It also reduces the load incurred to the search engines by using fewer terms in the queries.