Etudiant : EL HAIDAMI LAILA

Filière : Master Web Intelligence et Sciences des Données (WISD)

Encadrant : Pr. SABRI ABDELOUAHED

Annèe : 2023

Résumé : Social media has become a prominent medium for sharing thoughts and information in both everyday life and during natural crises. Humanitarian organizations can leverage these platforms to save lives and alleviate the suffering of affected individuals. In this master’s report, we introduce a novel multimodal deep learning architecture based on captions to classify natural crisis tweets. Furthermore, we introduce an annotation methodology that enables the construction of a French multimodal dataset comprising crisis-related tweets. This French dataset enhances our proposed multimodal approach, enabling it to classify not only English tweets but also French tweets. This expanded capability enhances the applicability and effectiveness of the approach in crisis management and response scenarios. This paves the way for a multilingual approach, allowing for the handling of multiple languages, which can be considered as a future goal.