Ajuste de modelos del lenguaje grandes para la generación de contranarrativas en Español
Fecha
2024-05-16
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
Este trabajo Fin de Máster se centra en la investigación sobre la generación automática de contranarrativas en español para combatir mensajes de odio en redes sociales. Con dicho trabajo, pretendemos contribuir a los objetivos de desarrollo sostenible (ODS) relativos a educación de calidad (metas 4.5, 4.7 y 4.a), igualdad de género (metas 5.1,5.2 y 5.b) y reducción de desigualdades (metas 10.3 y 10.7), ya que la contranarrativa es una respuesta a los mensajes ofensivos que nos permite educar a los usuarios de redes sociales, promoviendo el respeto hacia los demás, sin juzgar su físico, raza, ideología u orientación sexual. Para ello, se plantean distintos experimentos en los que se evalúan diferentes modelos grandes del lenguaje que han sido ajustados para esta tarea. Además, se propone un nuevo marco de evaluación cuantitativo para identificar los experimentos más prometedores. Finalmente se exponen las conclusiones obtenidas la realización de este proyecto.
This Master's thesis aims to study the automatic generation of counternarratives in Spanish to combat hate crimes in social networks. With this work, we want to contribute to the Sustainable Development Goals (SDGs) related to quality education (goals 4.5, 4.7 and 4. a), gender equality (goals 5.1, 5.2 and 5.b) and reduced inequalities (goals 10.3 and 10.7), because is a response to offensive messages that allows us to educate social network users , promoting respect for others, regardless of their physical appearance, race, ideology or sexual orientation. For this purpose, we propose different experiments in which we evaluate different large models of language that have been adjusted for this task. In addition, a new quantitative evaluation framework is proposed to identify the most promising experiments. Finally, the conclusions obtained from this project are presented.
This Master's thesis aims to study the automatic generation of counternarratives in Spanish to combat hate crimes in social networks. With this work, we want to contribute to the Sustainable Development Goals (SDGs) related to quality education (goals 4.5, 4.7 and 4. a), gender equality (goals 5.1, 5.2 and 5.b) and reduced inequalities (goals 10.3 and 10.7), because is a response to offensive messages that allows us to educate social network users , promoting respect for others, regardless of their physical appearance, race, ideology or sexual orientation. For this purpose, we propose different experiments in which we evaluate different large models of language that have been adjusted for this task. In addition, a new quantitative evaluation framework is proposed to identify the most promising experiments. Finally, the conclusions obtained from this project are presented.