With the advances in the domain of NLP and NLU in recent years, such as the GPT-3 and other Large Language Models, the industry is finally mature enough to empower organisations to unlock the incredible knowledge potential hidden within omnipresent unstructured data sources. In this presentation, Dr. Vlasta Kus from GraphAware talked about the state-of-the-art technologies and complex pipelines employed with a goal of turning an archive of a major US foundation into a Knowledge Graph which enables surprise (aha-moments), massive modelling complexity and provides previously unavailable level of insights and pattern discovery.
John D. Rockefeller (1839 - 1937)
considered in his epoch the wealthiest American of all time
at peak, he controlled 90% of all oil production in the US
worried that fortune was rolling up so fast his heirs would "dissipate their inheritances or become intoxicated with power"
defined, together with other industrialists such as a steel magnate Andrew Carnegie, the modern systematic approach of targeted philanthropy through creation of foundations that had a major effect on scientific research, medicine and education
The Rockefeller Foundation
to this day 39th largest US foundation
the most prominent activities:
medical, public health and population sciences
agricultural and natural sciences
arts and humanities
social sciences
international relations
High-level professional gossip
Project goals
Potential of KGs in the digitisation of analog historical records
New insights & patterns of the grant-making process by analysing intellectual network and the evolution of research fields
Cutting-edge analytics toolkit for researchers coming to the Archive each year
Patterns - do lunches, dinners increase likelihood of grant being awarded? How much of internal discussions and developments take place within the RF before the projects are funded?
Sub-discipline trends - some declining, others rising to prominence
To humans - KGs offer the ability to explore and analyse the data and knowledge within it in a natural user-friendly way (I’ll show some of the visualisation options in the demos)
Regarding machines - well designed/built KGs are used to help complex ML solutions to leverage and learn from relational knowledge (think of GNNs in biomedical domain for drug repurposing etc.; KG embeddings leveraged in NED systems or in link prediction tasks)
Robert Yerkes - a pioneer in the study of primate intelligence and of the social behaviour of gorillas and chimpanzees
- high-level professional gossip
In the course of this project, we have:
Built a KG valuable for obtaining global view of the data as well as well as the ability to drill down in the investigations, explore, understand, hypothesize … and validate the theories
Built and analysed the intellectual network
Looked into the evolution of the research fields, observed how some rise in prominence while other fade away
Helped to identify new patterns and insights related to the grant-making process
We showed the value of KGs on the example of the archive domain, but this is easily transferable to other industries as well.