Detección de ironía en textos cortos enfocada a la minería de opinión
Fecha
2017-06-25
Autores
Sotolongo Peña, Anakarla
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Las opiniones textuales imponen grandes retos a las aplicaciones de minería de opinión ya que varios problemas están presentes; entre ellos: la escritura de opiniones de manera irónica o sarcástica. Existen varias propuestas que permiten detectar ironía; sin embargo, la efectividad de la clasificación es baja, sobre todo debido a que aún no son suficientemente expresivos los rasgos identificados para discriminar entre oraciones irónicas y no irónicas. Es por ello que el objetivo general de esta investigación consiste en detectar de manera efectiva la ironía en textos cortos a partir de la identificación de nuevos rasgos lingüísticos y la aplicación de algoritmos de aprendizaje supervisado. Los principales resultados obtenidos son: (1) El estudio y consecuente identificación de las características de los rasgos utilizados en la clasificación de la ironía, así como los principales algoritmos de aprendizaje supervisado utilizados con tales propósitos. (2) La creación de nuevos rasgos que permiten caracterizar los textos irónicos y por tanto discriminar entre clases, principalmente basados en patrones específicos y generales, la ambigüedad, el desbalance contextual, las estadísticas de aparición de las palabras y el contraste en el texto. (3) Un módulo que permite detectar de manera efectiva la ironía a partir de los rasgos creados y seleccionados de la literatura, mediante el uso de varias técnicas de aprendizaje supervisado. La combinación de los rasgos creados, así como los algoritmos de clasificación empleados, permitieron obtener valores de calidad en la detección de la ironía superiores a los reportados previamente en la literatura.
Textual opinions impose great challenges on opinion mining applications because several problems are present; such as: the opinion could be writing in an ironic or sarcastic way. There are several proposals to detect irony; however, the irony classification effectiveness is low, mainly because the features identified to discriminate between ironic and non-ironic sentences are not enough expressive. For that reason, this research is focus on the irony detect in short texts by identifying new linguistic features and applying supervised learning algorithms. The main results obtained are: (1) The study and consequent identification of the principal characteristics of features used in irony classification, as well as the main supervised learning algorithms used for such purposes. (2) The creation of new features that allow characterizing ironic texts and therefore discriminate between classes, mainly based on specific and general patterns, ambiguity, contextual imbalance, statistics on word appearances and contrast in the text. (3) A module which allows the effective irony detection from the created features and others selected from the literature, by applying supervised learning techniques. The combination of the created features, as well as the selected classification algorithms, allowed to obtain quality values in the irony detection better than those previously reported in the literature.
Textual opinions impose great challenges on opinion mining applications because several problems are present; such as: the opinion could be writing in an ironic or sarcastic way. There are several proposals to detect irony; however, the irony classification effectiveness is low, mainly because the features identified to discriminate between ironic and non-ironic sentences are not enough expressive. For that reason, this research is focus on the irony detect in short texts by identifying new linguistic features and applying supervised learning algorithms. The main results obtained are: (1) The study and consequent identification of the principal characteristics of features used in irony classification, as well as the main supervised learning algorithms used for such purposes. (2) The creation of new features that allow characterizing ironic texts and therefore discriminate between classes, mainly based on specific and general patterns, ambiguity, contextual imbalance, statistics on word appearances and contrast in the text. (3) A module which allows the effective irony detection from the created features and others selected from the literature, by applying supervised learning techniques. The combination of the created features, as well as the selected classification algorithms, allowed to obtain quality values in the irony detection better than those previously reported in the literature.
Descripción
Palabras clave
Minería de Opinión, Ironía en Textos Cortos, Clasificación, Aprendizaje Automático Supervisado