Paper awarded at EXIST2025

A research team from the IC\M/T at the St. Pölten UAS achieved outstanding results at the international EXIST2025 competition, which focused on detecting sexism in social networks. The challenge went beyond simple detection – it also required understanding the intent behind posts and classifying the type of sexism.

Significant Contribution to Ethical AI Research

Using innovative AI models evaluated in English and Spanish , the results in the team ranking were impressive:

• With their explainable model (SCBM), they ranked as the fourth best team in the task understanding the intent of sexism.
• With their baseline model they ranked as the third best team in the task of detecting sexism.

Their models stood out for their transparency, efficiency, and multilingual capabilities, marking a significant contribution to ethical AI research.

The submitted paper demonstrated how discriminatory content can be automatically detected in a transparent and explainable way. Notably, the models performed well across languages – even though the descriptive adjectives used as a basis for the classification model were only available in English.

What are these “descriptive adjectives”? Together with sociologists from the University of Vienna, the team created a list of adjectives that describe typical traits of emotional and toxic statements. The model selects fitting terms like discriminatory, gender-stereotypical, patriarchal, limiting, demeaning, or inappropriate and uses them to decide whether a comment is sexist.

The results will be presented in September 2025 at the CLEF Conference in Madrid by Adrian Jaques Böck.

Authors: Roberto Labadie-Tamayo (formerly St. Pölten UAS), Adrian Jaques Böck, Djordje Slijepčević, Xihui Chen, Andreas Babic, and Matthias Zeppelzauer