Intelligent Research And Judgment Of New Electronic Invoice Abnormal Behavior

Electronic invoices have been born with the development of Internet technology and user needs.The issuance of electronic invoices will improve the efficiency of national tax management and control,save corporate costs,and bring more convenient services to consumers.Electronic invoices have the advantages of strong anti-counterfeiting,easy storage,efficient circulation,data networking,and low-carbon environmental protection,etc.However,in the electronic invoice system,there are some abnormal behaviors,such as repeated reimbursement,illegal change of invoice status,over limit invoicing and so on.According to the circulation process of electronic invoices in the electronic invoice service system,this paper proposes possible abnormal behaviors and provides corresponding investigation procedures.In particular,based on the quantified and multi-sourced characteristics of invoice data at the invoice issuing system,such as the abnormal behavior of a large number of large-scale invoices issued by a company in a short time,a method of feature research and judgment based on multi-layer perceptron is given.In addition,for the abnormal situation of enterprises /users hidden in various types of abnormal behavior data,this paper also proposes an abnormal behavior association analysis method based on K-Means and Skip-gram.Both methods train and judge models based on historical invoice data and abnormal data,and perform real-time research and judgment on new data.Among them,the accuracy of research and judgment models based on machine learning for abnormal behaviors can reach more than 99%,and the accuracy of association analysis models between enterprises / users is also about 80%.From the perspective of training results,both methods have strong practicability and can provide reliable security guarantee for the electronic invoice service system.