Knowledge Graphs in Support of Credit Risk Assessment

Abstract

In comparison with domain ontologies, knowledge graphs are less complex to build. They remove the burden of specifying boundaries for the domain and reduce completeness and consistency requirements. They have been successful in facilitating knowledge reuse and maintenance. Adding knowledge continuously, in small localised chunks, is easier than the holistic engineering required for ontologies. In this paper, we exploit this to use knowledge graphs in combination with ontologies for transfer learning in machine learning. Through the use of knowledge graphs, data is extracted and transformed from one domain to another where data is lacking. This synthesized data is then used to support machine learning overcoming the lack of data. This approach is illustrated to support transfer learning in lending risk assessment. The approach provides a template for supporting data driven innovation as a finance company explores new markets and designs new products.