Classification of Transaction Behavior in Tax Invoices Using Compositional CNN-RNN Model

Transaction behavior can be recorded by the invoice in China. The invoice records several fields including transaction content, transaction code (in accordance with the Tax Classification and Coding for Commodities and Services issued by the state) and transaction unit, etc. In this paper, we propose a compositional CNN-RNN model framework with attention mechanism to classify transaction behavior collected from tax invoices to the corresponding category based on the official transaction code, which is of great importance to tax supervision and provides a new perspective to analysis the industrial structure of the region. Preliminary experiments show that the overall accuracy of classifying transaction behavior achieves 75%.