An Analysis of Pseudo-News and Propaganda, Insights for Better Characterization and Detection
Abstract
Abstract—Misinformation detection, and more precisely propaganda detection, is an active field of research due to the possible benefits it can have on society. Extant approaches based on Language Models such as BERT show promise but easily overfit datasets, due to the specificity of most of them. To enhance robustness and trustworthiness, we propose a neurosymbolic approach combining text embeddings and symbolic conceptual features to improve generalization to new sources. Results show improvements over equivalent text-only methods.
Auteur(s) : Géraud Faye, Benjamin Icard, Morgane Casanova, Julien Chanson, Guillaume Gadek, Guillaume Gravier, Wassila Ouerdane, Céline Hudelot, Sylvain Gatepaille and Paul Égré
Lien vers l'article :