Use this link to cite:
http://hdl.handle.net/2183/27812 Sistema para la construcción automatizada y explotación de taxonomías
Loading...
Identifiers
Publication date
Authors
Filgueiras Rilo, Juan Luis
Advisors
Other responsabilities
Enxeñaría informática, Grao en
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen] Las taxonomías son estructuras de información que ofrecen valor jerárquico en diversas aplicaciones
basadas en conocimiento como la comprensión de consultas, la respuesta de preguntas
en lenguaje natural y la recomendación personalizada. Sin embargo, a día de hoy, la
mayoría de métodos automatizados de construcción de taxonomías son limitados respecto a
la poca expresividad que poseen las relaciones entre nodos, lo que provoca que su elaboración
se vea supeditada al esfuerzo y coste intensivos, derivados de la construcción manual y la necesidad
de conocimiento experto. Este proceso proporciona resultados de suma importancia,
pero poco adaptables a los cambios y, raramente completos. De esta situación se deriva la
incipiente necesidad de métodos de elaboración de taxonomías automatizados y de calidad,
que permitan relaciones de mayor expresividad y ofrezcan más flexibilidad ante cambios.
HiExpan es un «framework» de construcción de ontologías que deduce la relación semántica
entre nodos de la misma clase a partir de una taxonomía semilla, la cual se irá expandiendo
hasta conseguir un resultado convergente, de forma no-supervisada. El objetivo de este proyecto
es adaptar, mejorar y ofrecer una forma de plataforma que permita la construcción y
explotación de taxonomías de manera sencilla e interactiva, para poder comprobar la utilidad
de los resultados obtenidos.
Para poder lograr la consecución de los anteriores objetivos se ha adoptado una metodología
ágil con ciclos iterativos e incrementales, para adaptarse a las posibles necesidades
emergentes durante el desarrollo. De esta forma, se han alcanzado los objetivos en forma de
incrementos funcionales cada vez más expresivos, con el apoyo de herramientas para el aseguramiento
de la calidad, obteniendo un producto fiable, completo y de interés multidisciplinar,
con una diversa aplicabilidad en el ámbito de la Recuperación de la Información, e incluso
fuera del mismo.
[Abstract] Taxonomies are data structures that offer hierarchical value at diverse knowledge-based applications, such as query understanding, question answering and personalized recommendation. However, the vast majority of automatized methods are, nowadays, limited in means of poor expresivity between nodes, causing the process elaboration to be contigent on intensive effort and cost, derived from manual construction and the need of expert knowledge. This process offers results of great value, but poorly adaptative to change and, rarely complete. This situation derives the incipient need of automatized taxonomy construction methods, that allow more expresive relationships and offer more flexibility for changes. HiExpan is a framework of taxonomy construction that derives the semantic relation between nodes of same semantic class starting from a seed taxonomy, which will be iteratively expanded until having a convergent result, in a non-supervised way. The goal of this proyect is to adapt, improve and offer a platform that allows the construction and exploitation of taxonomies in a easy and interactive way, so its easy to check the usefulness of obtained results. In order to achieve the previous objectives, an agile methodology is adopted, with iterative and incremental cycles, to ensure adaptability to the possible emerging needs during development. This way, each goal is achieved in the form of a functional increment, adquiring more expresivity as it progresses through the iterations, with the help of development tools for the quality assurance, resulting in a reliable, complete product of multidisciplinar interest, with a diverse aplicability whether in the scope of Information Retrieval or elsewhere.
[Abstract] Taxonomies are data structures that offer hierarchical value at diverse knowledge-based applications, such as query understanding, question answering and personalized recommendation. However, the vast majority of automatized methods are, nowadays, limited in means of poor expresivity between nodes, causing the process elaboration to be contigent on intensive effort and cost, derived from manual construction and the need of expert knowledge. This process offers results of great value, but poorly adaptative to change and, rarely complete. This situation derives the incipient need of automatized taxonomy construction methods, that allow more expresive relationships and offer more flexibility for changes. HiExpan is a framework of taxonomy construction that derives the semantic relation between nodes of same semantic class starting from a seed taxonomy, which will be iteratively expanded until having a convergent result, in a non-supervised way. The goal of this proyect is to adapt, improve and offer a platform that allows the construction and exploitation of taxonomies in a easy and interactive way, so its easy to check the usefulness of obtained results. In order to achieve the previous objectives, an agile methodology is adopted, with iterative and incremental cycles, to ensure adaptability to the possible emerging needs during development. This way, each goal is achieved in the form of a functional increment, adquiring more expresivity as it progresses through the iterations, with the help of development tools for the quality assurance, resulting in a reliable, complete product of multidisciplinar interest, with a diverse aplicability whether in the scope of Information Retrieval or elsewhere.
Description
Editor version
Rights
Atribución-NoComercial-SinDerivadas 3.0 España







