Mining Twitter Data for Potential Drug Effects

摘要

Adverse drug reactions have become one of the top causes of deaths. For surveillance of adverse drug events, patients have gradually become involved in reporting their experiences with medications through the use of dedicated and structured systems. The emerging of social networking provides a way for patients to describe their drug experiences online in less-structured free text format. We developed a computational approach that collects, processes and analyzes Twitter data for drug effects. Our approach uses a machine-learning-based classifier to classify personal experience tweets, and use NLM's MetaMap software to recognize and extract word phrases that belong to drug effects. Our results on 5 medications demonstrate the validity of our approach, and its ability to correctly extract potential drug effects from the Twitter data. It is conceivable that social media data can serve to complement and/or supplement traditional time-consuming and costly surveillance methods.