Research Paper Accepted Into the Australasian Conference on Information Systems
Collaborative paper on knowledge graphs to further enable transfer learning in credit risk accepted into ACIS
SYDNEY, Australia | October 2020
A collaborative research paper on the use of knowledge graphs to further enable transfer learning concepts in credit risk assessment has been accepted into the Australasian Conference on Information Systems (ACIS).
ACIS is the premier conference in Australasia for Information Systems academics, covering technical, organisational, business and social issues in the application of IT. The research paper titled ‘Knowledge Graphs in Support of Credit Risk Assessment’ shows that the use of a constrained knowledge graph could help solve the problem of trying to build machine learning (ML) models where there is limited or no historical data.
The paper was co-authored by Rich Data Corp’s Chief Data Scientist, Dr Hendra Suryanto and Professor Ghassan Beydoun from the University of Technology Sydney and builds on findings from their previous research into transfer learning.
The paper demonstrates that knowledge graphs are a potential solution to link concepts from two different but related ontologies, such as two ontologies describing different lending products, leveraging human expert knowledge. If the first ontology can extract sufficient data to build an ML model but the second ontology can’t, a knowledge graph can be used to establish links between the ontologies, which can then be used to map higher order functions, or concepts.
Once the mapping function between the two ontologies has been established, the mapping can be used to create a virtual data set from the existing data extracted on which an ML model can be trained. This solves a common business problem where a lender is looking to launch an innovative new lending product but lacks the historical data to train a model specific to that use case.
The paper has been published as research in progress and is subject to rigorous peer review. The Australasian Conference on Information Systems takes place virtually on 1-4 December 2020 from Wellington, New Zealand and will host over 200 global academic delegates.