An experimental exploration of drug-drug interaction extraction from biomedical texts

The First Challenge of Drug-Drug Interaction Extraction (DDIExtraction 2011) involves doing a binary DDI detection to determine whether a drug pair in a given sentence (with annotated drug names) has interaction information. This may be the first attempt at extraction of drug interaction information in wide community. In this paper we compare and evaluate the effectiveness of different strategies of example generation from texts and different feature types for drug relation extraction. The comparative results show that (1) drug interaction classification at drug entity pair level performs better than that at sentence level; (2) simple NLP output does not improve performance and more advanced way of incorporating NLP output need to be explored.