Complex Event Extraction using DRUM

Complex mechanisms, such as cell-signaling pathways, consist of many highly interconnected components, yet they are often described in disconnected fragmentary ways. The goal of DRUM (Deep Reader for Understanding Mechanisms) is to develop a system that can read papers and combine results of individual studies into a comprehensive explanatory model. A first step is to automatically extract relevant events and event relationships from the literature. This paper describes initial steps in extending an existing general deep language understanding system, TRIPS, to read biomedical papers. In a preliminary evaluation, our system was the best performing system among the participants, achieving results close to human expert performance. These results suggested that our system is viable for complex event extraction and, ultimately, understanding complex systems and mechanisms.