Fecha
2024-09-14
Derechos de acceso
info:eu-repo/semantics/closedAccess
Título de la revista
ISSN de la revista
Título del volumen
Editorial
Springer
Resumen
Touché 2023 task 4 evaluated stance detection in a Multilingual multi-target setting with a reduced annotated dataset. This is why we have tested different approaches focused on increasing training data by (1) including new samples from back-translating original training data and (2) adding samples from unlabeled data using label propagation. The results showed that back-translation was successful, while label propagation worsened the performance. We obtained the best results with a transformer-based model fine-tuned in two steps: the first on a related, more extensive dataset and the second on the development data. This shows the usefulness of including related data in our approach and suggests additional research based on taking advantage of other datasets and data augmentation. Besides, given that the current results were close to a 0.35 f1 score, there is still room for improvement in this task.
Descripción
Categorías UNESCO
Palabras clave
Data augmentation, Label propagation, Multilingual Stance detection
Citación
Avila, J., Rodrigo, Á., Centeno, R. (2024). Best of Touché 2023 Task 4: Testing Data Augmentation and Label Propagation for Multilingual Multi-target Stance Detection. In: Goeuriot, L., et al. Experimental IR Meets Multilinguality, Multimodality, and Interaction. CLEF 2024. Lecture Notes in Computer Science, vol 14958. Springer, Cham. https://doi.org/10.1007/978-3-031-71736-9_13
Centro
E.T.S. de Ingeniería Informática
Departamento
Lenguajes y Sistemas Informáticos
Grupo de investigación
Natural language processing & information retrieval group (NLP&IR)



