Filtrado inteligente de noticias basado en Deep Learning

Lopera Martínez, Ana Isabel (2018). Filtrado inteligente de noticias basado en Deep Learning. Thesis (Master thesis), E.T.S. de Ingenieros Informáticos (UPM).

Description

Title: Filtrado inteligente de noticias basado en Deep Learning
Author/s:
  • Lopera Martínez, Ana Isabel
Contributor/s:
Item Type: Thesis (Master thesis)
Masters title: Ingeniería Informática
Date: July 2018
Subjects:
Faculty: E.T.S. de Ingenieros Informáticos (UPM)
Department: Inteligencia Artificial
Creative Commons Licenses: Recognition - No derivative works - Non commercial

Full text

[thumbnail of TFM_ANA_ISABEL_LOPERA_MARTINEZ.pdf]
Preview
PDF - Requires a PDF viewer, such as GSview, Xpdf or Adobe Acrobat Reader
Download (2MB) | Preview

Abstract

El análisis de posicionamiento de noticias consiste en identificar la disposición
de una noticia con respecto a un hecho o entidad, y supone una tarea fundamental
para determinar la veracidad de ésta. Este trabajo presenta una solución sobre el
problema presentado en el desaf´ıo de Fake News Challenge, proponiendo un servicio
web de filtrado de noticias basado en redes neuronales hacia delante. La herramienta
está orientada a usuarios finales y les permitirá obtener la posición de una noticia
concreta respecto del titular. Para el desarrollo del clasificador se han comparado
representaciones basadas en el método de bag of words y vectores de medias de
embedding de word2vec. La representación del titular y el cuerpo de la noticia
alimenta un modelo de red hacia delante que realiza una clasificación multiclase.
El modelo final consiste en una red neuronal optimizada con tres capas ocultas
aplicando regularización L2, tomando como entrada la representaci´on de vectores
de medias. Se consigue un resultado de 87% en valor predictivo positivo y 86% en
términos de recall.---ABSTRACT---Analysis of news stance consists on addressing the position on a report towards a
fact or entity, and takes an important role in the assessment of veracity of the report
itself. This project proposes a solution relating to the Fake News Challenge problem,
and proposes a web service for news filtering, based on feedforward neural networks.
This tool is conceived to end-users and will let them obtain the stance of a text with
respect a claim. To develop the classifier, text representations based on bag-of-words
and vector embedding mean have been considered. Text representations of headline
and article body are used as input for a feedforward neural network trained to make
a multiclass classification. The final model proposed consists on a hand-tuned 3-layer feedforward network applying L2 regularization, taking as input two vectors of
means of word embeddings. This model achieves a 87% in terms on precision and a
86% in terms on recall.

More information

Item ID: 51620
DC Identifier: https://oa.upm.es/51620/
OAI Identifier: oai:oa.upm.es:51620
Deposited by: Biblioteca Facultad de Informatica
Deposited on: 12 Jul 2018 11:18
Last Modified: 12 Jul 2018 11:18
  • Logo InvestigaM (UPM)
  • Logo GEOUP4
  • Logo Open Access
  • Open Access
  • Logo Sherpa/Romeo
    Check whether the anglo-saxon journal in which you have published an article allows you to also publish it under open access.
  • Logo Dulcinea
    Check whether the spanish journal in which you have published an article allows you to also publish it under open access.
  • Logo de Recolecta
  • Logo del Observatorio I+D+i UPM
  • Logo de OpenCourseWare UPM