Using query extension and user feedback to improve pubmed search

PubMed is a search engine that is widely used to search for medical publications. A common challenge in information retrieval, and thus also when using PubMed, is that broad search queries often result in lists of thousands of papers that are presented to the user, too narrow ones often yield small or even empty lists. To address this problem we here present a new PubMed search interface with query extension using keyword clusters generated with evolutionary algorithms to obtain more specific search results. Users can choose to add various words to their query and then rate search results; this scoring is stored in a database to enable learning from user feedback to improve keyword cluster optimization as well as query extensions. We show how users can extend PubMed queries using previously generated keyword clusters, rate query results, and use these ratings for optimizing parameters of the keyword clustering algorithms.