Search

This is a summary of the Ph.D. thesis written by Pilar López Úbeda at Universidad de Jaén under the supervision of PhD. M. Teresa Martín Valdivia, Ph.D. L. Alfonso Ureña López and PhD. Manuel Carlos Díaz Galiano. The defense was held in Jaén on April 22, 2021. The doctoral committee was integrated by PhD. Rafael Muñoz Guillena from Universidad de Alicante, PhD. Paloma Martínez Fernández from Universidad Carlos III de Madrid, and Manuel Montes y Gómez from National Institute of Astrophysics, Optics and Electronics (Mexico). The thesis obtained the grade of Summa Cum Laude and the international mention. Este es un resumen de la tesis doctoral realizada por Pilar López Úbeda en la Universidad de Jaén bajo la dirección de los doctores Dña. M. Teresa Martín Valdivia, D. L. Alfonso Ureña López y D. Manuel Carlos Díaz Galiano. La defensa se realizó en Jaén el 22 de abril de 2021. La comisión de doctorado estuvo integrada por el PhD. Rafael Muñoz Guillena de la Universidad de Alicante, la PhD. Paloma Martínez Fernández de la Universidad Carlos III de Madrid, y Manuel Montes y Gómez del Instituto Nacional de Astrofísica, Óptica y Electrónica (México). La tesis obtuvo la calificación de Sobresaliente Cum Laude y mención de doctorado internacional. This work has been partially supported by a grant from Fondo Europeo de Desarrollo Regional (FEDER), LIVING-LANG project [RTI2018-094653-B-C21], and the Government of Andalusia [PY20_00956].

El reconocimiento de entidades con nombre (NER) es una tarea importante en el campo del Procesamiento del Lenguaje Natural que se utiliza para extraer conocimiento significativo de los documentos textuales. El objetivo de NER es identificar trozos de texto que se refieran a entidades específicas. En esta tesis pretendemos abordar la tarea de NER en el dominio biomédico y en español. En este dominio las entidades pueden referirse a nombres de fármacos, síntomas y enfermedades y ofrecen un conocimiento valioso a los expertos sanitarios. Para ello, proponemos un modelo basado en redes neuronales y empleamos una combinación de word embeddings. Además, nosotros generamos unos nuevos embeddings específicos del dominio y del idioma para comprobar su eficacia. Finalmente, demostramos que la combinación de diferentes word embeddings como entrada a la red neuronal mejora los resultados del estado de la cuestión en los escenarios aplicados. Named Entity Recognition (NER) is an important task in the field of Natural Language Processing that is used to extract meaningful knowledge from textual documents. The goal of NER is to identify text fragments that refer to specific entities. In this thesis we aim to address the task of NER in the Spanish biomedical domain. In this domain entities can refer to drug, symptom and disease names and offer valuable knowledge to health experts. For this purpose, we propose a model based on neural networks and employ a combination of word embeddings. In addition, we generate new domain- and language-specific embeddings to test their effectiveness. Finally, we show that the combination of different word embeddings as input to the neural network improves the state-of-the-art results in the applied scenarios. Tesis Univ. Jaén. Departamento de Informática. Leída el 22 abril de 2021.

Paper presented at 11th International Workshop on Spoken Dialogue Systems, IWSDS 2020; Madrid; Spain; 21 September 2020 through 23 September 2020 In this work, a Spanish corpus that was developed, within the EMPATHIC project (http://www.empathic-project.eu/) framework, is presented. It was designed for building a dialogue system capable of talking to elderly people and promoting healthy habits, through a coaching model. The corpus, that comprises audio, video an text channels, was acquired by using a Wizard of Oz strategy. It was annotated in terms of different labels according to the different models that are needed in a dialogue system, including an emotion based annotation that will be used to generate empathetic system reactions. The annotation at different levels along with the employed procedure are described and analysed.