EdAttack

Sonja Häffner and Simon Polichinel von der Maase presented their paper at ISA 2026, discussing a semi-automated pipeline for constructing event datasets on rare attacks in education. Their approach combines large language models, synthetic data, and active learning to enhance efficiency, accuracy, and scalability in data collection and annotation.

On March 19, Simon and Sonja presented at the University of Pittsburgh’s HAIL seminar, discussing their work on a semi-automated pipeline for detecting rare events from text. Their method uses large language models, synthetic data, and active learning to efficiently build structured datasets, focusing on underreported attacks on education.