DIMEMEX-2025

Social networks play a crucial role in people’s lives by transforming the dynamics of communication and information sharing. Analyzing the content from these platforms has become a hot research topic for the computational linguistics community. However, despite the notable advances made in recent years, there are still open challenges that merit additional research for better treatment or deeper understanding. One such challenge is the detection of abusive content, which includes aspects like hate speech, aggression, offensive language, and other related phenomena.

Given the multimodal nature of social media platforms, we aim to promote the research and development of multimodal computational models for the detection of abusive content in Mexican Spanish, particularly hate, offensive, and vulgar memes. Memes are well known for providing predominantly a humorous or ironic meaning based on the conjunction of text and images. So, the absence of either text or image may alter its interpretation.

As a way to promote the development of innovative solutions with societal impact as well as to tackle multimodality challenges, DIMEMEX comprises three subtasks:

a) A three-way classification: hate speech, inappropriate content, and neither. Participants are free to use any approach of their choice.

b) A finer-grained classification distinguishing instances containing hate speech into different categories such as classism, sexism, racism, and others. 

c) A three-way classification: hate speech, inappropriate content, and neither. Unlike a) participants are restricted to focusing exclusively on leveraging LLMs to detect the specified categories.