The recent study titled “GazaVHR: AI-Driven Legally Grounded Conflict Harm Documentation”, co-authored by DSAI PhD students Nesibe Şebnem Paluluoğlu and Dilara Zeynep Gürer, together with our assistant professors Muhammed Furkan Akıncı and Mustafa Taha Koçyiğit, has been accepted for poster presentation at the 4th Muslims in ML (MusIML) Workshop, co-located with ICML 2025.

The paper introduces GazaVHR, a novel AI-assisted dataset developed for the structured documentation of potential human rights violations in the Gaza conflict using advanced vision-language models. The dataset pipeline begins with over 170,000 images sourced from conflict-related tweets, applies consecutive filtering and semantic clustering techniques, and aligns the output with a legally grounded reference dataset based on the Rome Statute. The final output—4,603 likely high-confidence images—demonstrates the potential of AI in scalable and systematic humanitarian documentation.

We gratefully acknowledge Adba Analytics for granting us access to the Twitter data that made this project possible.