IE University Capstone · In partnership with UNICC · 2025 By Gizela Thomas
Teaching AI to Recognize Sexism Online
A capstone collaboration with the United Nations International Computing Centre to detect sexist language in Spanish-language tweets.
The Challenge
For my capstone at IE University, I collaborated with the United Nations International Computing Centre (UNICC) to develop NLP models that detect sexist language in Spanish-language tweets.
Building the Data
A core part of the project was data cleaning and unification. We consolidated multiple datasets into a single multilingual corpus of 44,000+ tweets. This involved language detection and translation (Python’s LangDetect and Google Translate) for English and Spanish coverage; deduplication and filtering of incomplete and off-topic records; standardizing labels and metadata across sources; and consolidating 16 overlapping subcategories into 5 clearer classes, such as Sexual Insults & Objectification and Threats & Physical Harm.
The Models
We benchmarked transformer models and tested LLMs, comparing binary and multiclass classification, evaluating tradeoffs between precision and recall, and designing a pipeline that balanced accuracy with fairness.
Why It Matters
This work supports UN Sustainable Development Goal 5 (Gender Equality), specifically Target 5.2: eliminating violence against women and girls. By applying data science to online spaces, we explored how AI can help create safer digital environments.
- tweets
- 44,000+
- categories
- 16 → 5
- languages
- 2
Tools used
- Python
- NLP
- Transformers
- LLMs
- LangDetect
RelatedUN International Computing Centre (opens in a new tab)UN SDG 5: Gender Equality (opens in a new tab)IE University (opens in a new tab)
Built with teammates Sebastián Felipe Zambrano Julio, Pablo Camacho Fernandez, Juan Martin Echeverri, Yihang Li, and Alejandro Pérez. With thanks to mentor Hind Azegrouz, Elena Rivas Ruzafa of UNICC, and Dae-Jin Lee and Raquel Cabero Quiles.
