Convolutional Neural Network-Based Joint Extraction of Adverse Drug Events

Abstract

The conventional joint method for event extraction performs decoding with beam search. Excessively small beam easily leads to the local optimal solution problem., while blind beam expansion may bring too much noise. In this regard., we utilized the convolutional neural network (CNN) to first determine whether the sentences contain events., and then expanded the beam for the event-containing sentences during decoding of the joint event extraction model., which can effectively reduce the noise and improve the search probability of global optimal solution. We applied this model to the extraction of adverse drug events in the medical field and achieved good results.