An Ensemble Model of Clinical Information Extraction and Information Retrieval for Clinical Decision Support

This paper describes the participation of Mayo Clinic NLP team in the Text REtreival Conference (TREC) 2016 Clinical Decision Support track. We propose an ensemble model which combines three components: a Part-of-Speech based query term weighting model (POSBoW); a Markov Random Field model leveraging clinical information extraction (IE-MRF); and a Relevance Pseudo Feedback (RPF) model. We submitted three automatic runs and two manual runs. The experimental results show that the automatic runs outperform the median results of all participant teams for up to 76.7% of the given query topics.