Publicación: Búsqueda de Respuestas en foros usando modelos basados en Transformers
Cargando...
Archivos
Fecha
2022-03-11
Autores
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
Atribución-NoComercial-SinDerivadas 4.0 Internacional
info:eu-repo/semantics/openAccess
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Nacional de Educación a Distancia (España). Escuela Técnica Superior de Ingeniería Informática. Departamento de Lenguajes y Sistemas Informáticos
Resumen
Durante los últimos años se ha hecho más necesaria la búsqueda de información en internet. Debido a dicha necesidad, se han creado sitios web donde los usuarios tratan de ayudar a otros usuarios contestando sus preguntas. Estos sitios se llaman foros de respuesta colaborativa. Esto ha generado un problema: muchas de las preguntas que se plantean ya se han respondido previamente y es deseable poder recuperar respuestas a preguntas similares. Es por ello por lo que surgió la tarea de Community Question Answering (CQA), buscando permitir un acceso más cómodo a la información publicada en estos sitios web. Dentro de esta tarea han surgido varias subtareas y en este trabajo se aborda una de la más recurrentes: Dada una pregunta y un conjunto de respuestas, encontrar las respuestas que mejor encajan con la pregunta. Para ello se usarán técnicas de Deep Learning, y más en particular una nueva arquitectura de redes neuronales, los Transformers, que utilizan los mecanismos de atención. Mediante dicha técnica se ha intentado clasificar y ordenar las respuestas de distintos foros en función de la pregunta formulada. Se ha probado su validez en función de 3 colecciones distintas: SemEval 2015, 2017 y AmazonQA. Se ha concluido que los resultados extraídos con este tipo de métodos permiten superar los resultados anteriores, abriendo la posibilidad hacia una mayor implantación de estas tecnologías en sistemas reales.
During the last few years it has become more necessary to search for information on the Internet. Due to this need, websites have been created where users try to help other users by answering their questions. These sites are called collaborative response forums. This has created a problem: many of the questions asked have already been answered previously and it is desirable to be able to retrieve answers to similar questions. This is why the task of Community Question Answering (CQA) arose, seeking to allow more convenient access to the information posted on these websites. Within this task, several subtasks have emerged and this paper addresses one of the most recurrent ones: Given a question and a set of answers, find the answers that best fit the question. For this purpose, Deep Learning techniques will be used, and more in particular a new neural network architecture, the Transformers, which use the attention mechanisms. By means of this technique, an attempt has been made to classify and order the answers from different forums according to the question asked. Its effectiveness has been tested on the basis of 3 different collections: SemEval 2015, 2017 and AmazonQA. It has been concluded that the results extracted with this type of methods allow to overcome the previous results, opening the possibility towards a greater implementation of these technologies in real systems.
During the last few years it has become more necessary to search for information on the Internet. Due to this need, websites have been created where users try to help other users by answering their questions. These sites are called collaborative response forums. This has created a problem: many of the questions asked have already been answered previously and it is desirable to be able to retrieve answers to similar questions. This is why the task of Community Question Answering (CQA) arose, seeking to allow more convenient access to the information posted on these websites. Within this task, several subtasks have emerged and this paper addresses one of the most recurrent ones: Given a question and a set of answers, find the answers that best fit the question. For this purpose, Deep Learning techniques will be used, and more in particular a new neural network architecture, the Transformers, which use the attention mechanisms. By means of this technique, an attempt has been made to classify and order the answers from different forums according to the question asked. Its effectiveness has been tested on the basis of 3 different collections: SemEval 2015, 2017 and AmazonQA. It has been concluded that the results extracted with this type of methods allow to overcome the previous results, opening the possibility towards a greater implementation of these technologies in real systems.
Descripción
Categorías UNESCO
Palabras clave
Citación
Centro
Facultades y escuelas::E.T.S. de Ingeniería Informática
Departamento
Lenguajes y Sistemas Informáticos