TaxAA: a Reliable Tax Auditor Assistant for Exploring Suspicious Transactions

We present TaxAA, a reliable tax auditor assistant that helps tax auditors explore suspicious transactions and get reliable evidence. We construct a Tax Audit Network(TAN) and use extended algorithms based on the semi-supervised Graph Convolutional Network(GCN) to build detection models for calculating taxpayers’ suspicion score, we choose Hierarchical Graph Convolutional Network(H-GCN) as our basic model according to the experimental results. Then the visual analytic system allows tax auditors to customize suspicious indicators to observe the suspicious relationships among taxpayers by the ”wheel” chart. Meanwhile, it can provide detailed transactions and individual information for reference. We have evaluated the assistant based on the tax data of a province, our detection model can achieve high accuracy and the visual analytic system can provide useful guidance for tax auditors.