Publicación: Detección de lenguaje ofensivo en redes sociales
Cargando...
Fecha
2023
Autores
Editor/a
Director/a
Tutor/a
Coordinador/a
Prologuista
Revisor/a
Ilustrador/a
Derechos de acceso
info:eu-repo/semantics/openAccess
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Desde su aparición a principios de la década de los 2000, el uso de las redes sociales se ha ido incrementando entre usuarios de todas las edades y procedencias del mundo. Este crecimiento masivo ha llevado a un aumento significativo en la interacción y la comunicación en línea. Las redes sociales son la principal herramienta de comunicación y una de las principales fuentes de información entre la mayor parte de la población mundial.
Sin embargo, a pesar de traer consigo una amplia gama de beneficios, recientemente han surgido desafíos relacionados con el lenguaje ofensivo y el discurso de odio en ellas.
Todas las redes sociales disponen de sistemas de moderación, algunos de los cuales son manuales y bastante tediosos. Hoy en día, sigue siendo un reto conseguir un sistema que sea rápido, eficaz y que aprenda igual de rápido que evoluciona el lenguaje humano. Las redes sociales serían un ecosistema mucho más seguro y sano si se consiguieran detectar este tipo de mensajes y frenar su publicación en tiempo real. En este trabajo se estudia la idoneidad de los modelos basados en Transformers, considerados el estado del arte en el campo del procesamiento de lenguaje natural, para detectar mensajes ofensivos en redes sociales. Se investigan los modelos más actuales de esta área para encontrar aquel que esté más cerca de lograr el objetivo. Se han realizado experimentos entrenando con fuentes de datos diferentes a las que fueron entrenados para probar si mejora su generalización.
Una vez realizados los experimentos que se detallan en este trabajo se ha llegado a la conclusión de que actualmente se dispone de una serie de modelos pre-entrenados que pueden suponer una muy buena base para el desarrollo de estos sistemas. Estos modelos requieren de una correcta fase de entrenamiento que permita mejorar sus métricas y generalizarlo a todas las redes sociales y usuarios del mundo.
Since its appearance in the early 2000s, the use of social networks has increased among users of all ages and backgrounds around the world. This massive growth has led to a significant increase in online interaction and communication. Social networks are the main communication tool and one of the main sources of information among most of the world's population. However, despite bringing a wide range of benefits, challenges related to offensive language and hate speech in them have recently emerged. All social networks have moderation systems, some of which are manual and quite tedious. Today, it remains a challenge to achieve a system that is fast, efficient and that learns as quickly as human language evolves. Social networks would be a much safer a nd healthier ecosystem if these types of messages could be detected and their publication stopped in real time. This project studies the suitability of Transformers based models, considered the state of the art in the field of natural language processing, to detect offensive messages on social networks. The most current models in this area are investigated to find t he one that is closest to achieving the objective. Experiments have been carried out training with data sources different from those on which they were trained to test if their generalization improves. Once the experiments detailed in this work have been carried out, it has been concluded that there is currently a series of pre trained models that can provide a very good basis for the development of these systems. These models require a correct training phase that allows them to improve their metrics and generalize them to all social networks and users in the world.
Since its appearance in the early 2000s, the use of social networks has increased among users of all ages and backgrounds around the world. This massive growth has led to a significant increase in online interaction and communication. Social networks are the main communication tool and one of the main sources of information among most of the world's population. However, despite bringing a wide range of benefits, challenges related to offensive language and hate speech in them have recently emerged. All social networks have moderation systems, some of which are manual and quite tedious. Today, it remains a challenge to achieve a system that is fast, efficient and that learns as quickly as human language evolves. Social networks would be a much safer a nd healthier ecosystem if these types of messages could be detected and their publication stopped in real time. This project studies the suitability of Transformers based models, considered the state of the art in the field of natural language processing, to detect offensive messages on social networks. The most current models in this area are investigated to find t he one that is closest to achieving the objective. Experiments have been carried out training with data sources different from those on which they were trained to test if their generalization improves. Once the experiments detailed in this work have been carried out, it has been concluded that there is currently a series of pre trained models that can provide a very good basis for the development of these systems. These models require a correct training phase that allows them to improve their metrics and generalize them to all social networks and users in the world.
Descripción
Categorías UNESCO
Palabras clave
Citación
-
Centro
Facultades y escuelas::E.T.S. de Ingeniería Informática
Departamento
Inteligencia Artificial