摘要
自2016年“营改增”全面实施以来,与之相关的免税减税等税收优惠政策原旨在惠企助企、激发市场活力,但不法分子在巨额利润驱动下企图通过虚开增值税发票骗取出口退税、抵扣税款,严重扰乱了税收秩序。本文以“暴力”虚开发票的企业的犯罪特征为切入点,从基础征管数据和增值税发票数据中选取了24项虚开指标,构建了基于Transformer模型的虚开增值税发票识别模型,对虚开公司进行检测。实证分析表明Transformer模型对虚开增值税发票的识别召回率为0.934 7,准确率为0.986 9,AUC为0.963 9,显著优于SVM、Xgboost、MLP等传统机器学习模型,可辅助税务部门高效识别“暴力”虚开企业,节省人工筛查成本,对有效打击虚开增值税发票一类违法犯罪行为具有非常重要的实践意义。
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 intended 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 seriously 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 recognition 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 learning models such as SVM,Xgboost,and MLP.It can assist the tax department to efficiently identify"violent"falsely issued enterprises and save the cost of manual screening,which has a very important practical significance for effectively cracking down on illegal and criminal acts such as falsely issuing VAT invoices.
作者
杨慧
程建华
YANG Hui;CHENG Jianhua(School of Big Data and Statistics,Anhui University,Hefei 230039,China)
出处
《安徽工程大学学报》
CAS
2024年第1期76-85,共10页
Journal of Anhui Polytechnic University
基金
国家自然科学基金资助项目(72304001)
安徽省哲学社会科学规划一般基金资助项目(AHSKF2019D019)。
