BOB体育

he Department of Health Outcomes & Policy's Biomedical Informatics Seminar Series

Faculty Candidate Seminar Kirk Roberts, Ph.D., Postdoctoral Fellow, National Library of Medicine

From EHR Data to Answers: Harnessing Natural Language in Health Care

Tuesday, February 17 Clinical and Translational Research Building Room 3161/3162, 2:30-3:30 p.m.

Abstract: The transition from paper charts to electronic health records (EHR) has not reduced our dependence on natural language, despite the challenges associated with processing unstructured text. The primary reason we record this data is to answer questions, be they clinical questions to help care for the patient, administrative questions about billing, or secondary use questions for research. This talk addresses many of the necessary steps, as well as some solutions, towards answering those questions. The first step is the process of providing some structure to the unstructured text in the form of information extraction. Here, the talk focuses primarily on extracting medical concepts and events, recognizing their linguistic polarity and modality, and providing spatial grounding. Some additional experiments will also be discussed, such as the value of good data and unsupervised event structuring. The second step is the task of finding relevant data, referred to as in formation retrieval. Here, the talk focuses on contributions to the medical-related TREC tasks for searching EHRs and the biomedical literature. The final step is the task of understanding natural language questions and finding answers. Here, the talk focuses on various strategies for understanding questions, including both long, rambling, ungrammatical questions as well as short, concise, well-formed questions.