Research And Application Of Intelligent Tax System Based On Neural Network

The vigorous development of the Internet has led us into the era of"Internet+",and the related big data,cloud computing,and artificial intelligence fields have ushered in a golden period of development."Internet+" penetrates into all walks of life,the relationship between the real economy and the Internet are becoming closer and closer,forming a huge data economy,the tax agency and taxpayers are generating massive transaction behavior data every day.The national tax department actively promotes the reform of "Internet+ taxation".Therefore,relying on a large amount of transaction behavior invoice data and combining other multi-source data to build an intelligent tax system,practically solve the problems in tax services and complete the intelligent reform of tax services,has great research and application value.The research topic of this thesis is the research and application of intelligent tax system based on neural network.The article first describes the overall architecture of the intelligent tax system and the technical framework of the basic platform,introduces the research focus and main contributions of each part of the system,and provides important support for subsequent research content.Secondly,in the research of tax invoice transaction behavior classification and coding model,the research focuses on the feature extraction of short text classification in invoice behavior data,and a compositional CNN-RNN text classification model based on the attention mechanism is proposed.Besides,a high-concurrency transaction behavior coding query system based on this model is constructed.In the analysis of the distribution structure of regional industries,the focus is on the relationship between industry prices and industry structure,and the differences in industry geographical distribution.In the research of knowledge graph analysis based on multi-source tax data,we build a tax enterprise knowledge graph based on the taxpayer's invoice transaction relationship,and provide variety of query functions for multiple association relationships.Finally,we summarize the research results of the thesis and look forward to potential research directions.