Learning to Rate Clinical Concepts Using Simulated Clinician Feedback

We present a user-based model for rating concepts (i.e., words and phrases) in clinical queries based on their relevance to clinical decision making. Our approach can be adopted by information retrieval systems (e.g., search engines) to identify the most important concepts in user queries in order to better understand user intent and to improve search results. In our experiments, we examine several learning algorithms and show that by using simulated user feedback, our approach can predict the ratings of the clinical concepts in newly unseen queries with high prediction accuracy.