EXIST-2025: Source Intention in tweets (hard-hard)

This task aims to classify sexist tweets according to the author’s intention, allowing for a better understanding of the role social media plays in the expression and dissemination of sexist messages. In this task, a ternary classification is proposed: direct, reported, and judgemental. In this task, a soft-soft evaluation is considered, where the probability of each label predicted by the system is compared with the probability defined based on annotation disagreement in the gold standard.