Find the Fraud: Assessing Tax Non-Compliance in the IRS

Session Track: Knowledge Graphs

Session Time: 12:45 am - 01:00 am November 7

Session description

Graph Neural Networks (GNNs) are a powerful tool in graph analysis. In this this talk, Dr. Conovaloff will apply GNN techniques to fraud detection in IRS networks. GNNs not only match suspicious taxpayer network patterns but also individual taxpayer, nodal properties. This technique incorporates multi-modal data to include natural language. GNNs have an advantage over brute-force graph analysis in their ability to match patterns without complex, user-defined queries. Dr. Conovaloff will discuss basic graph convolutional network theory in a Python framework and identify risky IRS network patterns. He will then introduce a machine learning training regimen that successfully identifies graphs with suspicious properties.