Biomedical Entities Impact on Rating Prediction for Psychiatric Drugs

摘要

There is a growing body of research on biomedical information extraction on social media texts, patient narratives, and scientific abstracts. Relatively less research has focused on the relationship between medical concepts mentioned in a given user review and patient satisfaction. In this work, we investigate the effect of health-related entities such as adverse drug reactions and drug indications on rating prediction. We present a method based on a supervised regression approach leveraging medical concepts of different types and their mechanisms. The experiments on a collection of reviews demonstrate that features based on biomedical entities mentioned in reviews result in performance gains of up to 8% in mean squared error. Moreover, we compute feature importance in the regression models and find that the most important features for predicting patients' negative attitudes are adverse drug reactions associated with functional problems.