Detección de la polaridad de las opiniones basada en nuevos recursos léxicos
Fecha
2016-06-12
Autores
Amores Fernández, Mario Alberto
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
Las opiniones son una parte importante en la vida de los seres humanos. Para extraer el sentimiento acerca de objetos, productos o servicios, se necesitan sistemas de minería de opinión automatizados. La herramienta PosNeg Opinion detecta de manera no supervisada la polaridad de las opiniones basándose en recursos léxicos, de ahí que sea sensible a la calidad de éstos, que en su mayoría están concebidos para el idioma Inglés, aquellos que fueron anotados automáticamente tienen muchos errores, los anotados manualmente recogen muy pocos términos, y el formato que presentan muchas veces limita la interacción entre ellos. Por otro lado, las opiniones generalmente presentan varios problemas que aún no son tratados eficazmente por las herramientas existentes. El objetivo de esta investigación consiste en desarrollar un sistema, a partir de las características de PosNeg Opinion, para la detección no supervisada de la polaridad de las opiniones a partir del empleo de nuevos recursos léxicos y que sea capaz de tratar la mayoría de los problemas presentes en las opiniones. Los resultados obtenidos son: la creación de los recursos SentiWordNet 4.0 y 4.1 para el idioma Inglés y SpanishSentiWordNet que es pionero en la puntuación de términos en español, los dos esquemas para la detección no supervisada de la polaridad de las opiniones, los recursos que permiten el manejo de jergas, emoticonos, palabras modificadoras y la negación, la herramienta PosNeg Opinion 3.0 que implementa el esquema finalmente propuesto auxiliándose de la biblioteca desarrollada PolarityDetection obteniendo satisfactorios valores de Exactitud, Precisión, Recall y F1.
Opinions are an important part in the life of human beings. To remove the feeling about objects, products or services, automated mining systems review are required. The PosNeg Opinion tool detects unsupervised polarity of opinions based on lexical resources, and is therefore sensitive to the quality of these, most of which are designed for the English language, those were scored automatically have many errors, annotated manually collect very few terms, and format have often limits the interaction between them. On the other hand, opinions generally have several problems that are not effectively treated by existing tools. The objective of this research is to develop a system, based on the characteristics of PosNeg Opinion for detection unsupervised polarity of opinions from the use of new lexical resources and be able to treat most problems present in the opinions. The results obtained are: the creation of SentiWordNet 4.0 and 4.1 for the English and SpanishSentiWordNet language is a pioneer in the score of terms in Spanish resources, the two schemes for detection unsupervised polarity of opinions, resources that allow finally proposed management jargons, emoticons, modifying words and denial, PosNeg Opinion 3.0 tool that implements the scheme using for this library developed with satisfactory values PolarityDetection Accuracy, Precision, Recall and F1.
Opinions are an important part in the life of human beings. To remove the feeling about objects, products or services, automated mining systems review are required. The PosNeg Opinion tool detects unsupervised polarity of opinions based on lexical resources, and is therefore sensitive to the quality of these, most of which are designed for the English language, those were scored automatically have many errors, annotated manually collect very few terms, and format have often limits the interaction between them. On the other hand, opinions generally have several problems that are not effectively treated by existing tools. The objective of this research is to develop a system, based on the characteristics of PosNeg Opinion for detection unsupervised polarity of opinions from the use of new lexical resources and be able to treat most problems present in the opinions. The results obtained are: the creation of SentiWordNet 4.0 and 4.1 for the English and SpanishSentiWordNet language is a pioneer in the score of terms in Spanish resources, the two schemes for detection unsupervised polarity of opinions, resources that allow finally proposed management jargons, emoticons, modifying words and denial, PosNeg Opinion 3.0 tool that implements the scheme using for this library developed with satisfactory values PolarityDetection Accuracy, Precision, Recall and F1.
Descripción
Palabras clave
Detección de Polaridad, Opiniones, Nuevos Recursos Léxicos, Sistema