Aprendizaje Automático versus Expresiones Regulares en la Detección de la Negación y la Especulación en Biomedicina

In this paper, we present a machine learning system that identify the negation and speculation signals in biomedical texts, in particular, in the BioScope corpus. The objective of this research is to compare the efficiency of this approach focused on machine learning with which it is based on regular expressions. Among the systems that follow the latter approach, we used NegEx because of its availability and popularity. The evaluation has been carried out on the three subcollections which make up Bioscope: clinical documents, scientific papers and abstracts of scientific articles. The results show the superiority of the approach based on machine learning over the use of regular expressions. In the detection of expressions of negation, the proposed system improves the F1 measure of NegEx by between 20 and 30%, depending on the collection of documents. In the speculation detection, the proposed system outperforms the F1 measure of the best system between 10 and 20%.