Douzone Bizon said on Aug. 20 that a paper titled “Development of an AI-Based Transaction-Network Credit Scoring Model (DefaultGNN),” jointly researched with Techfin Ratings and a research team led by KAIST professor Chan-young Park (박찬영), was accepted by the ACM International Conference on Information and Knowledge Management (CIKM).

CIKM is an international academic conference in artificial intelligence, data mining and information retrieval.

According to the company, the core of the research focuses on quantitatively analyzing transaction-counterparty network information that existing corporate credit rating models could not handle, complementing existing models and strengthening synergy.

Existing models have been run mainly on independent financial information such as a company’s profitability, stability, liquidity and growth. As a result, external network information did not make a major contribution to improving model performance. DefaultGNN, by contrast, quantifies hidden risks using a range of network variables, including counterparties, transaction classifications, transaction shares, business closures, defaults and cancelled amounts.

Douzone Bizon explained that it was able to implement DefaultGNN by adding AI algorithms to enterprise resource planning (ERP) data accumulated over decades.

Ji Yong-gu (지용구), co-CEO of Douzone Bizon, said, “True corporate credit lies not in a past independent report card but in the trajectory of continuing to trade with healthy partners.” He added, “I hope this research, created at the intersection of decades of ERP data assets and the latest AI technology, becomes a starting point for inclusive finance that supplies funds to sound small and medium-sized businesses.”