Risk Identification of Violent False Issuing Invoice Based on Transformer Model
Since the full implementation of the policy of"replacement of business tax with value-added tax"in 2016,the related preferential tax policies such as tax exemption and reduction were originally in-tended to benefit enterprises and stimulate market vitality.However,driven by huge profits,criminals attempted to defraud export tax rebates and tax deductions by falsely issuing VAT invoices,which seri-ously disrupted the taxation order.This paper takes the criminal characteristics of"violent"false issuing of enterprises as the starting point,and selects 24 false issuing indicators from the basic collection and management data and value-added tax invoice data,and constructs a falsely issued VAT invoice recogni-tion model based on the Transformer model to detect the false companies.Empirical analysis shows that the Transformer model has a recognition recall rate of 0.9347 for falsely issued VAT invoices,with an accuracy of 0.9869 and an AUC of 0.9639,which is significantly better than traditional machine learn-ing models such as SVM,Xgboost,and MLP.It can assist the tax department to efficiently identify"vio-lent"falsely issued enterprises and save the cost of manual screening,which has a very important practi-cal significance for effectively cracking down on illegal and criminal acts such as falsely issuing VAT in-voices.
万方数据
