The purpose of this observational descriptive study is to perform secondary analysis of previously collected patient illness narratives to assess the feasibility and face validity of SA methods to characterize the sentiment of the illness narratives and explore how various SA methods and their lexicons impact the results. We analyzed de-identified patient illness narratives (n=10) at various levels of text analysis (document, paragraph, sentence), using four sentiment analysis (SA) approaches. Overall, the SA methods demonstrated a lack of polarity concordance. As we moved from document to paragraph to sentence, the variations in polarity scores increased and agreement between the SA approaches decreased.
Learning Objective: Understand the challenges and issues of using sentiment analysis to process health-related text.
Andrey Soares (Presenter)
University of Colorado Anschutz Medical Campus
Heather Coats, University of Colorado Anschutz Medical Campus
Paula Meek, University of Colorado Anschutz Medical Campus
Terrah Akard, Vanderbilt University
Ardith Doorenbos, University of Illinois
Lisa Schilling, University of Colorado Anschutz Medical Campus