Detecting Hate and Inflammatory Content in Bengali Memes: A New Multimodal Dataset and Co-Attention Framework
arXiv:2602.22391v2 Announce Type: replace Abstract: Internet memes have become a dominant form of expression on social media, including within the Bengali speaking community. While often humorous, memes can also be exploited to spread offensive, harmful, and inflammatory content targeting individuals and groups. Detecting this type of content is exceptionally challenging due to its satirical, subtle, and culturally specific nature. This problem is magnified for low-resource languages like Bengali, as existing research predominantly focuses on high-resource languages. To address this critical r
The proliferation of internet memes, combined with advancements in multimodal AI, makes the detection of harmful content in diverse linguistic and cultural contexts increasingly critical and feasible.
This research highlights the persistent challenge of content moderation in low-resource languages and directly contributes to developing more culturally nuanced and effective AI tools for combating online hate.
The creation of a new dataset and a co-attention framework specifically for Bengali memes advances the capability of AI models to detect hate speech in complex, multimodal, and low-resource linguistic environments.
- · AI researchers focused on natural language processing
- · Social media platforms operating in South Asia
- · Bengali-speaking internet users
- · Content moderation technology providers
- · Creators of hate speech and inflammatory content
- · Individuals and groups targeted by online hate
- · Manual content moderation efforts operating at scale
Improved detection capabilities will help social media platforms identify and remove harmful content more effectively in Bengali.
This effort could inspire similar research and dataset creation for other low-resource languages, fostering more equitable AI development for global content moderation.
Enhanced content moderation in specific languages may inadvertently drive purveyors of hate speech to ever more obscure or niche linguistic and multimodal expressions.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL