Mejora de la recuperación de información en bases de datos de texto utilizando recursos lingüísticos
No Thumbnail Available
Date
2009-04
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Al convertirse la web en el mayor repositorio de conocimiento y en un medio
de publicación fácilmente accesible para todos, la Recuperación de Información ha
dejado de ser un campo exclusivo de los especialistas en Ciencias de la Información
y ha pasado a ser un campo relacionado con cualquier persona. El maximizar la
cantidad de documentos relevantes obtenidos para una consulta depende de la
destreza de este especialista para preparar una estrategia de búsqueda adecuada. Si
bien los usuarios no tienen por qué conocer técnicas de recuperación de información,
la propuesta de esta tesis es mejorar los resultados de su búsqueda por medio de un
“especialista” que implementa estas técnicas.
Se propone el refinamiento semántico de los conceptos de la consulta a fin de
mejorar la precisión de los resultados, utilizando recursos lingüísticos para construir
una estrategia de búsqueda adecuada. El refinamiento semántico propuesto consiste
en: guiar al usuario para desambiguar los conceptos ingresados por él, permitirle
seleccionar conceptos jerárquicamente relacionados a fin de precisar los documentos
a recuperar, y expandir semánticamente los conceptos a fin de aumentar la cantidad
de documentos a recuperar. Los recursos lingüísticos que pueden utilizarse son
tesauros, diccionarios, diccionarios multilinguales y ontologías. Qué recursos
utilizar, depende del área del conocimiento de la consulta y de los recursos
disponibles para ese área.
Se evalúa el refinamiento semántico, eligiendo el recurso WordNet para
consultas de dominio general y el recurso MeSH, especializado en el área salud, para
las consultas en un dominio específico. Las experiencias realizadas muestran que
aumenta la precisión de los resultados en un 19,03 % en el dominio general y en un
33,50 % en un dominio específico del conocimiento.
Otro aspecto tratado es la inmersión del refinamiento semántico en un motor
de búsqueda propio de un sitio web y en sistemas de recomendación. Los resultados
experimentales muestran que el uso del refinamiento mejora las prestaciones del
motor de búsqueda del sitio, con respecto a su uso en su forma estándar, obteniéndose un incremento cercano al 33% en la precisión y duplicando
aproximadamente la cantidad de documentos recuperados. Para la inclusión del
refinamiento en sistemas de recomendación se elige el área educación, planteando el
agregado de la personalización de los resultados utilizando metadatos del usuario y
metadatos de los documentos. De esta forma se potencia la recuperación obtenida del
refinamiento semántico porque se ordenan los resultados de distinta forma según el
usuario y el momento en que éste haya realizado la consulta.
As the Web has become one of the biggest repository of knowledge easily accessible for everyone, the Information Retrieval has stopped to be an exclusive field of information sciences specialists, and it has become a field related with any person. Although users do not have to know information retrieval techniques, the proposal of this thesis is to improve query results by using a "specialist" that implements these techniques. For this, a semantic refinement is proposed. This semantic refinement acts as a specialist in information sciences and prepares, using linguistic resources, an appropriate search strategy that represents the user’s information need. The semantic refinement consists on three steps. First, it guides the user in the disambiguation of the words submitted by him. Then, it allows the user to select concepts hierarchically related in order to reduce the amount of documents to retrieve. Finally, it expands semantically concepts to increase the amount of documents to be retrieved. Linguistic resources that can be used are thesauri, dictionaries, multilingual dictionaries and ontologies. What resources are used, depends on the domain of knowledge, and on resources available for that domain. Semantic refinement is evaluated by using WordNet as a linguistic resource for information retrieval in a general domain knowledge. Also, it was evaluated in a specific domain, by using MeSH, as a linguistic resource for health. Experimental results show that precision increases in a 19.03% in a general domain, and in a 33.50% in a specific domain of knowledge. Semantic Refinement was also applied in a website's search engine. Experimental results show that the refinement improves the performance of the site's search engine in almost 33% in precision and approximately doubles the number of documents retrieved. Another issue addressed in this thesis, is the immersion of the proposed refinement, in an recommender system of educational materials for a personalized retrieval. Personalization consists in ordering the results according to the user’s profile depending on his/her preferences. The use of the refinement inside the recommender system improves the semantic search, and because of this the recommendation will be improved.
As the Web has become one of the biggest repository of knowledge easily accessible for everyone, the Information Retrieval has stopped to be an exclusive field of information sciences specialists, and it has become a field related with any person. Although users do not have to know information retrieval techniques, the proposal of this thesis is to improve query results by using a "specialist" that implements these techniques. For this, a semantic refinement is proposed. This semantic refinement acts as a specialist in information sciences and prepares, using linguistic resources, an appropriate search strategy that represents the user’s information need. The semantic refinement consists on three steps. First, it guides the user in the disambiguation of the words submitted by him. Then, it allows the user to select concepts hierarchically related in order to reduce the amount of documents to retrieve. Finally, it expands semantically concepts to increase the amount of documents to be retrieved. Linguistic resources that can be used are thesauri, dictionaries, multilingual dictionaries and ontologies. What resources are used, depends on the domain of knowledge, and on resources available for that domain. Semantic refinement is evaluated by using WordNet as a linguistic resource for information retrieval in a general domain knowledge. Also, it was evaluated in a specific domain, by using MeSH, as a linguistic resource for health. Experimental results show that precision increases in a 19.03% in a general domain, and in a 33.50% in a specific domain of knowledge. Semantic Refinement was also applied in a website's search engine. Experimental results show that the refinement improves the performance of the site's search engine in almost 33% in precision and approximately doubles the number of documents retrieved. Another issue addressed in this thesis, is the immersion of the proposed refinement, in an recommender system of educational materials for a personalized retrieval. Personalization consists in ordering the results according to the user’s profile depending on his/her preferences. The use of the refinement inside the recommender system improves the semantic search, and because of this the recommendation will be improved.
Description
Keywords
Recuperación de información, Refinamiento semántico, Bases de datos de texto, Recursos lingüísticos