Główne pojęcia
Identifying hateful memes and their targeted entities in low-resource languages like Bengali is crucial for understanding social dynamics and countering hate speech.
Streszczenie
Internet memes, often a source of humor, can also spread hateful content targeting individuals or communities. Existing research on high-resource languages overlooks challenges in low-resource languages like Bengali. A novel dataset, BHM (Bengali Hateful Memes), is introduced for detecting hateful memes and their targets. DORA, a multimodal deep neural network, outperforms state-of-the-art baselines on identifying hateful memes and their targets.
Statystyki
The dataset consists of 7,148 Bengali memes.
DORA outperforms several state-of-the-art rivaling baselines.
The dataset includes two sets of labels for detecting hateful memes and identifying targeted entities.
Cytaty
"Despite being the seventh most widely spoken language globally, Bengali is considered one of the notable resource-constrained languages."
"Memes have gained significant traction in social media, reaching a broad audience and influencing public sentiment."
"DORA is generalizable on other low-resource hateful meme datasets."