Detección de fiabilidad en noticias en español
Fecha
2024-09-18
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
Evaluar la fiabilidad del lenguaje utilizado en la redacción de noticias es cada vez más crucial en el
panorama actual de los medios digitales. En este trabajo se realiza una investigación para desarrollar un
sistema de clasificación automática para determinar la fiabilidad de noticias escritas en español usando
Procesamiento del Lenguaje Natural. Para ello, se ha participado en la competición FLARES: FINE-
GRAINED LANGUAGE-BASED RELIABILITY DETECTION IN SPANISH NEWS del workshop IberLEF 2024, que
tiene como objetivo la detección precisa de la fiabilidad de noticias escritas en español basándose en el
lenguaje usado en ellas.
Para el desarrollo del sistema, se han probado las tres técnicas de ingeniería de prompts más usadas
(zero-shot, few-shot y CoT), empleando modelos generativos como Gemini y ChatGPT para generar las
respuestas. El rendimiento del sistema se ha evaluado exhaustivamente mediante un análisis de errores
y una matriz de confusión para identificar áreas de mejora.
Assessing the reliability of the language used in news writing is increasingly crucial in today's digital media landscape. In this work a research is carried out to develop an automatic classification system to determine the reliability of news written in Spanish using Natural Language Processing. For this purpose, we have participated in the FLARES: FINE-GRAINED LANGUAGE-BASED RELIABILITY DETECTION IN SPANISH NEWS competition of the IberLEF 2024 workshop, which aims to accurately detect the reliability of news written in Spanish based on the language used in them. For the development of the system, the three most widely used prompt engineering techniques (zero- shot, few-shot and CoT) have been tested, employing generative models such as Gemini and ChatGPT to generate the responses. The performance of the system has been thoroughly evaluated by means of an error analysis and a confusion matrix to identify areas for improvement.
Assessing the reliability of the language used in news writing is increasingly crucial in today's digital media landscape. In this work a research is carried out to develop an automatic classification system to determine the reliability of news written in Spanish using Natural Language Processing. For this purpose, we have participated in the FLARES: FINE-GRAINED LANGUAGE-BASED RELIABILITY DETECTION IN SPANISH NEWS competition of the IberLEF 2024 workshop, which aims to accurately detect the reliability of news written in Spanish based on the language used in them. For the development of the system, the three most widely used prompt engineering techniques (zero- shot, few-shot and CoT) have been tested, employing generative models such as Gemini and ChatGPT to generate the responses. The performance of the system has been thoroughly evaluated by means of an error analysis and a confusion matrix to identify areas for improvement.
Descripción
Palabras clave
Tratamiento inteligente de la Información