RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Publication
Artículo científico (JournalArticle)
Performance analysis of Particle Swarm Optimization applied to Unsupervised Categorization of Short Texts, Análisis de Prestación de Particle Swarm Optimization aplicado a Categorización no Supervisada de Textos Cortos
[ES] Existe actualmente la necesidad de acceder a información en línea tal
como resúmenes, noticias, opiniones, evaluaciones de productos, etc. Dicha información está disponible en la web, generalmente con el formato de textos cortos. Trabajos previos han demostrado la efectividad de un algoritmo discreto Particle Swarm
Optimization, llamado CLUDIPSO, para el agrupamiento de colecciones pequeñas
de textos cortos. Este artículo presenta un estudio preliminar sobre la prestación
de CLUDIPSO con colecciones más grandes. Los resultados fueron comparados con
los obtenidos con algoritmos representativos del estado del arte en el área. El trabajo experimental muestra una fuerte evidencia sobre los inconvenientes que posee
el algoritmo cuando debe agrupar colecciones de mayor tamaño. Con respecto a
este último aspecto, se discuten posibles razones del comportamiento inadecuado de
CLUDIPSO y se consideran algunas alternativas para resolver los problemas observados., [EN] Nowadays there is a need to access to on line information such as abstracts, news, opinions, evaluations of products, etc. That information is generally
available on the web as short texts. Previous works have demonstrated the e¿ectiveness of a discrete Particle Swarm Optimization algorithm, named CLUDIPSO, for
clustering small short-text corpora. This article presents a preliminary study about
the performance of CLUDIPSO on larger short-text corpora. The results were compared with those of the most representative algorithms of the state-of-the-art in the
area. The experimental work gives strong evidence about the drawbacks of this
algorithm to manage larger corpora. With respect to this last aspect, some possible
reasons about the poor behavior of CLUDIPSO with larger short texts corpora are
discussed and some alternatives in order to solve the problems observed, are considered., This research work was done in the framework of Marie Curie actions PEOPLE-IRSES 269180 WiQ-Ei: Web information Quality Evaluation initiative. The work of the first author is partially funded by the UPV program PAID-02-10-2257. The work of
the last two authors is partially funded by the MICINN research project TEXTENTERPRISE 2.0 TIN2009-13391-C04-03
(Plan I+D+i). The fourth author belongs to the VLC/CAMPUS Microcluster on Multimodal Interaction in Intelligent Systems.