Anna Vissens “The Guardian: Recognising ‘bad actors’ in data leaks with AI”

At Media Party Buenos Aires 2024, Anna Vissens, Head of Data Science at The Guardian, presented a collaborative AI project alongside JournalismAI, Follow the Money, and Daily Maverick. The goal was to build an entity-linking tool that helps investigative journalists uncover corruption by correctly connecting people and events across large volumes of text, even when individuals share identical names.

The team applied spaCy’s NER models to 300,000 articles, integrating external databases like LittleSis and OpenSanctions. After manually annotating 4,000 examples across 40 hours of work, the model reached 70% accuracy, though it faced hurdles with data duplication, limited name diversity, and a slight political bias.

Vissens emphasized that high-quality data is the single most critical factor for success and that universal newsroom tools are hard to build without continuous retraining. Looking ahead, she pointed to LLMs and context augmentation as the key to pushing entity linking far beyond traditional capabilities.

Don’t miss the full presentation! Watch Anna Vissens’ talk at Media Party Buenos Aires 2024.