Segmentación y detección de tópicos enfocado a la minería de opinión
Fecha
2016-05-15
Autores
Torres López, Carmen
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Central “Marta Abreu” de Las Villas
Resumen
La detección y seguimiento de tópicos es un área que ha sido aplicada en disímiles campos, entre ellos, en el análisis de sentimientos. PosNeg Opinion es una herramienta desarrollada en el CEI-UCLV que detecta de manera no supervisada la polaridad de opiniones mostrando excelentes valores de precisión y exactitud; sin embargo, solo es capaz de clasificar las opiniones expresadas en una oración o la opinión que expresa el texto en su totalidad, y por tanto no realiza un análisis del sentimiento por tópicos. El objetivo de la presente investigación consiste en desarrollar una herramienta que permita de manera no supervisada y efectiva descubrir tópicos en las opiniones. Los resultados obtenidos son: el diseño de un esquema para la segmentación y detección de tópicos en opiniones que propone identificar unidades textuales, pre-procesar el texto, representar las unidades textuales, aplicar técnicas de segmentación, representar la colección de segmentos, agrupar los segmentos y etiquetar los grupos de segmentos; el marco de trabajo OpinionTopicDetection que aplica el esquema propuesto para el descubrimiento de tópicos en opiniones; la instanciación del marco de trabajo con la aplicación OpinionTD para el descubrimiento de tópicos; un analizador léxico para pre-procesar los textos de opiniones y un algoritmo para etiquetar los grupos de tópicos basado en la identificación de sustantivos y sus similitudes semánticas. La evaluación realizada de la propuesta arrojó una precisión de 0.74 y exhaustividad 0.86, mostrando un resultado satisfactorio de la investigación desarrollada en el descubrimiento de tópicos en opiniones.
Topic detection and tracking is a research field that has been applied in several fields, one of them is sentiment analysis. PosNeg Opinion is a tool developed in the CEI-UCLV that detects unsupervised polarity of opinions showing excellent values of precision and recall; however, it is only able to classify the opinions expressed in a sentence or the opinion that text expresses as a whole, and therefore does not perform sentiment analysis by topic. The objective of this research is to develop a tool that allows unsupervised and effective discovering of topics in opinions texts. The main results are: the design of a scheme for segmentation and detection of topics in opinions that aims to identify textual units, text pre-processing, textual units representation, applying segmentation techniques, represent the collection of segments, segments clustering and labeling the groups of segments; OpinionTopicDetection framework which applies the proposed scheme for discovering topics in opinions; instantiation of the framework with OpinionTD desktop application for topics discovery; a lexical analyzer to pre-process the opinions texts and an algorithm for labeling groups of topics based on identifying nouns and their semantic similarities. The assessment of the proposal showed precision of 0.74 and recall of 0.86, showing a successful outcome of the research conducted in the topics discovery in opinions.
Topic detection and tracking is a research field that has been applied in several fields, one of them is sentiment analysis. PosNeg Opinion is a tool developed in the CEI-UCLV that detects unsupervised polarity of opinions showing excellent values of precision and recall; however, it is only able to classify the opinions expressed in a sentence or the opinion that text expresses as a whole, and therefore does not perform sentiment analysis by topic. The objective of this research is to develop a tool that allows unsupervised and effective discovering of topics in opinions texts. The main results are: the design of a scheme for segmentation and detection of topics in opinions that aims to identify textual units, text pre-processing, textual units representation, applying segmentation techniques, represent the collection of segments, segments clustering and labeling the groups of segments; OpinionTopicDetection framework which applies the proposed scheme for discovering topics in opinions; instantiation of the framework with OpinionTD desktop application for topics discovery; a lexical analyzer to pre-process the opinions texts and an algorithm for labeling groups of topics based on identifying nouns and their semantic similarities. The assessment of the proposal showed precision of 0.74 and recall of 0.86, showing a successful outcome of the research conducted in the topics discovery in opinions.
Descripción
Palabras clave
Segmentación, Detección de Tópicos, Minería de Opinión, Desarrollo de Herramienta