Communication Dans Un Congrès Année : 2024

EURECOM at SemEval-2024 Task 4: Hierarchical Loss and Model Ensembling in Detecting Persuasion Techniques

Youri Peskine
Paolo Papotti

Résumé

This paper describes the submission of team EURECOM at SemEval-2024 Task 4: Mul-tilingual Detection of Persuasion Techniques in Memes. We only tackled the first sub-task, consisting of detecting 20 named persuasion techniques in the textual content of memes. We trained multiple BERT-based models (BERT, RoBERTa, BERT pre-trained on harmful de-tection) using different losses (Cross Entropy, Binary Cross Entropy, Focal Loss and a custom-made hierarchical loss). The best results were obtained by leveraging the hierarchical nature of the data, by outputting ancestor classes and with a hierarchical loss. Our final submission consist of an ensembling of our top-3 best mod-els for each persuasion techniques. We obtain hierarchical F1 scores of 0.655 (English), 0.345 (Bulgarian), 0.442 (North Macedonian) and 0.177 (Arabic) on the test set.

Fichier principal
Vignette du fichier
publi-7779.semeval-1.172.pdf (172.06 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-04664496 , version 1 (30-07-2024)

Licence

Identifiants

  • HAL Id : hal-04664496 , version 1

Citer

Youri Peskine, Raphael Troncy, Paolo Papotti. EURECOM at SemEval-2024 Task 4: Hierarchical Loss and Model Ensembling in Detecting Persuasion Techniques. SEMEVAL 2024, 18th International Workshop on Semantic Evaluation, co-located with NAACL 2024, ACL, Jun 2024, Mexico, Mexico. ⟨hal-04664496⟩

Collections

126 Consultations
188 Téléchargements

Partager

  • More