Graph-based AI infers business conduct risk from sparse data

Researchers have developed a graph-based framework to infer business conduct risk, addressing the challenge of sparse and visibility-biased data. Their approach uses a Graph Convolutional Neural Network (GCNII) combined with Positive--Unlabeled learning to account for potential contamination in unlabeled data. In forward-looking evaluations, this method demonstrated superior performance compared to non-graph and simple graph-based benchmarks, particularly for firms with limited prior incident records, highlighting the value of inter-firm relationships in risk prioritization. AI

IMPACT This research could improve risk assessment in finance by leveraging relational data, potentially leading to more accurate predictions for firms with limited traditional data.

RANK_REASON Academic paper on a novel graph-based inference method for risk management. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• GCNII
• No Data Is Not No Risk: Visibility Aware Graph-Based Inference of Business Conduct Risk

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →