Un modelo para la obtención de interacciones medicamentosas mediante aprendizaje profundo sobre el corpus ’DDI EXTRACTION 2013’

Abstract

Drug-Drug interactions (DDI) constitutes essential and useful information for medical staffand patients, since they provide information on the effects of co-administered medications topatients, during therapy. This study uses a Piecewise Convolutional Neural Network (PCNN)to capture the relationship between pharmacological entities described in the biomedicaldatabases. Additionally, the model incorporates multichannel words to expand vocabularyand decrease the number of unknown words. The stochastic optimizer Adam is used forlearning the network parameters automatically, and Gaussian noise layer is added to improvethe extraction of DDI relationships. Experiments show an improved performance of thenew model, regard to the models found in the current technical literature, related to theDDIExtraction2013 Challenge, with verifiable and reproducible results.