UPB @ DANKMEMES: Italian Memes Analysis - Employing Visual Models and Graph Convolutional Networks for Meme Identification and Hate Speech Detection
Fiche du document
11 mai 2021
- doi: 10.4000/books.aaccademia.7360
- eisbn: 9791280136329
https://creativecommons.org/licenses/by-nc-nd/4.0/ , info:eu-repo/semantics/openAccess
Mots-clés
EVALITA COVID-19 Infodemic linguistica computazionale Automatic Misogyny Identification Misogyny on Twitter Posts AlBERTo Convolutional Neural Network BERT Model Hate Speech Detection Multimodal Meme Detection MEME management Language Game ``La Ghigliottina'' Language & Linguistics LAN000000 CBXSujets proches
Language (New words, slang, etc.) Field guides Keys (Identification guides) Identification guides Speech, Hate Group libel Group defamation Defamation against groups Racist speech Women-hating Forensic identification Competition (Economics) Economic competition Competitiveness (Economics)Citer ce document
George-Alexandru Vlad et al., « UPB @ DANKMEMES: Italian Memes Analysis - Employing Visual Models and Graph Convolutional Networks for Meme Identification and Hate Speech Detection », Accademia University Press, ID : 10.4000/books.aaccademia.7360
Métriques
Partage / Export
Résumé
Certain events or political situations determine users from the online environment to express themselves by using different modalities. One of them is represented by Internet memes, which combine text with a representative image to entail a wide range of emotions, from humor to sarcasm and even hate. In this paper, we describe our approach for the DANKMEMES competition from EVALITA 2020 consisting of a multimodal multi-task learning architecture based on two main components. The first one is a Graph Convolutional Network combined with an Italian BERT for text encoding, while the second is varied between different image-based architectures (i.e., ResNet50, ResNet152, and VGG-16) for image representation. Our solution achieves good performance on the first two tasks of the current competition, ranking 3rd for both Task 1 (.8437 macro-F1 score) and Task 2 (.8169 macro-F1 score), while exceeding by high margins the official baselines.