Education and research · La Rioja
Cyberbullying detection in student texts with NLP
The challenge
A university needed to analyze texts written by students to predict their involvement in cyberbullying situations and each one's role.
What we did
We trained language models with fine-tuning on the texts and, on top of their outputs, applied a classic classification layer with voting (ensemble) to predict involvement in cyberbullying and the role: cyberbully, cybervictim or cyberbystander.
Technologies
Python · PyTorch · Transformers · Fine-tuning · NLP · Ensemble
Result
Automatic classification of the role in cyberbullying situations, supporting research and early detection.
“They understood our research problem and gave us a reliable model to identify the roles; real support to detect cyberbullying in time.”