SubmissionNumber#=%=#69
FinalPaperTitle#=%=#SuteAlbastre at SemEval-2024 Task 4: Predicting Propaganda Techniques in Multilingual Memes using Joint Text and Vision Transformers
ShortPaperTitle#=%=#
NumberOfPages#=%=#7
CopyrightSigned#=%=#Anghelina Ion-Marian
JobTitle#==#
Organization#==#
Abstract#==#The main goal of this year's SemEval Task 4 is
detecting the presence of persuasion techniques
in various meme formats. While Subtask 1
targets text-only posts, Subtask 2, subsections
a and b tackle posts containing both images
and captions. The first 2 subtasks consist of
multi-class and multi-label classifications, in
the context of a hierarchical taxonomy of 22
different persuasion techniques.
This paper proposes a solution for persuasion
detection in both these scenarios and for vari-
ous languages of the caption text. Our team's
main approach consists of a Multimodal Learn-
ing Neural Network architecture, having Tex-
tual and Vision Transformers as its backbone.
The models that we have experimented with in-
clude EfficientNet and ViT as visual encoders
and BERT and GPT2 as textual encoders.
Author{1}{Firstname}#=%=#Ion Marian
Author{1}{Lastname}#=%=#Anghelina
Author{1}{Username}#=%=#ionanghelina
Author{1}{Email}#=%=#ion.anghelina@s.unibuc.ro
Author{1}{Affiliation}#=%=#University of Bucharest
Author{2}{Firstname}#=%=#Gabriel Sebastian
Author{2}{Lastname}#=%=#Buță
Author{2}{Username}#=%=#butasebi
Author{2}{Email}#=%=#butasebi@yahoo.com
Author{2}{Affiliation}#=%=#University of Bucharest
Author{3}{Firstname}#=%=#Alexandru
Author{3}{Lastname}#=%=#Enache
Author{3}{Username}#=%=#alexandruenache
Author{3}{Email}#=%=#enache.g.alexandru@gmail.com
Author{3}{Affiliation}#=%=#University of Bucharest

==========
èéáğö