Drug-Drug Interactions discovery based on CRFs, SVMs and rule-based methods

Information about medications is critical in improving the patients’ safety and quality of care. Most adverse drug events are predictable from the known pharmacology of the drugs and many represent known interactions and are, therefore, likely to be preventable. However, most of this information is locked in free-text and, as such, cannot be actively accessed and elaborated by computerized applications. In this work, we propose three different approaches to the problem of automatic recognition of drug-drug interactions that we have developed within the “First Challenge Task: Drug-Drug Interaction Extraction” competition. Our approaches learn to discriminate between semantically interesting and uninteresting content in a structured prediction framework as well as a rule-based one. The systems are trained using the DrugDDI corpus provided by the challenge organizers. An empirical analysis of the three approaches on this dataset shows that the inclusion of rule-based methods is indeed advantageous.