Building the world's largest news knowledge graph

Join us in our last Grenoble Data Science 2024 meetup.

Elin and Robert will share their experience with LLM and knowledge graphs applied to news data.

Looking forward to seeing you!

Abstract
Wrangling millions of new entities and relationships every day, it ain't cheap or easy. That's why we fine-tuned Phi-3-mini-4k to exceed Claude Sonnet 3.5 for graph extraction quality by up to 20% and to reduce cost by orders of magnitude. This required the development of creative training data set engineering and novel loss metrics. Now our analysts are deriving real-time insights from the world's largest news knowledge graph. Our talk will discuss the process from start to finish - including engineering the data set, fine-tuning Phi-3, and generating the ever-evolving news knowledge graph.

Short Bio
Elin Törnquist, PhD is a co-founder of Emergent Methods and AskNews and acts as Director of Transparency, spearheading research and ensuring diversity and transparency. She is an expert in design and management of interdisciplinary research projects.
Robert Caulk, PhD is the founder and CEO of Emergent Methods and AskNews, where he directs development and coordinates research projects. He is a highly cited researcher with more than 12 years in open-source software development.

Special mention
We address a special thank you to the Cowork In Grenoble and Minimistan teams for their continued support in hosting us.