Research On The Construction Method Of Tax Law Knowledge Graph For Tax Regulation

With the continuous development of emerging technologies such as artificial intelligence,cloud computing and big data,tax,as an important part of the national economy,is facing the tasks of the times such as tax reform and modernization in the new era.The construction of smart tax came into being under this background.However,in the actual tax collection and management,experts need to interpret the regulations issued by the state and formulate rules,and computer personnel write the rules into code for tax calculation.This model relies heavily on manual intervention and cannot achieve the goal of "wisdom" of smart taxation.The tax field urgently needs a method that can automatically extract and express the rules and knowledge contained in the tax regulations.Therefore,combined with natural language processing technology,the paper studies the tax law knowledge representation and tax law knowledge extraction of tax regulations,and constructs a tax law knowledge map for tax regulations.The specific research content is divided into the following three parts:(1)Tax law knowledge representation: Before constructing the knowledge map of tax regulations,it is necessary to statistically analyze the tax law knowledge elements in tax regulations,and formulate the representation method of tax law knowledge according to the internal logic and corresponding relationship between taxes.Combined with the eight elements of tax law stipulated in the tax law regulations,this paper proposes a tax law knowledge representation method for tax regulations.In this method,the knowledge in the tax law regulations is expressed as a tax law subject faceted tree according to the internal knowledge logical structure in the tax law regulations.The tax knowledge representation in our paper highlight the semantic roles and semantic dependencies of tax subject,tax object,tax behavior and tax types.Then,the semantics of the regulations are sub packaged through the modular structure.Take the taxpayer as the source node to fuse and generate the tax law knowledge map.A structured semantic representation method of tax knowledge is formed to support the subsequent tax law knowledge extraction.(2)Tax law knowledge extraction: The automatic extraction of tax law knowledge mainly includes two parts: tax law knowledge element extraction and element relationship recognition.In view of the long text semantic dependence of tax law elements in tax regulations,this paper proposes a tax regulation element extraction model based on MRC(Machine Reading Comprehension).Using the idea of machine reading,the model can fully mine the context information in the regulations and improve the recognition performance of tax elements.In the task of relationship recognition of tax elements,the traditional relationship extraction model based on neural network is difficult to obtain the structural features of sentences.In addition,because the sentence itself contains only a small number of words,it leads to serious sparse features,which is difficult to effectively support the relationship extraction between tax law knowledge elements.Therefore,this paper proposes a tax element relationship extraction method combined with feature markers.This method constructs feature marker information in tax laws and regulations.The BERT is used to extract the abstract features,obtain the structural feature information in the tax regulations,and construct the high-order abstract semantic representation of the tax regulations relative to the tax knowledge.So as to alleviate the problem of feature sparsity in tax law element relationship recognition and improve the performance of tax law element relationship recognition.(3)Construction and verification of tax law knowledge graph for tax regulations: Based on the research of factor and factor relationship extraction,combined with the internal logical structure of tax law knowledge,this paper designs the automatic construction method of tax law knowledge map.By formulating fusion rules,the extracted tax law knowledge is fused and the tax law knowledge map is.In order to verify the knowledge map proposed in this paper,a visualization tool of tax law knowledge map is designed and implemented.At present,we have realized the automatic construction of 1000 + tax law knowledge facet tree.According to the clear regularity and logic between the knowledge elements of the tax law facet tree,the tax law facet tree is converted into executable code through automatic compilation,which is calculated on the simulated tax related transaction data to realize the automatic calculation and application of tax.Through cooperation with Xi’an Jiaotong University and Shuiyou Software Group Co.,Ltd.,the system has been deployed and passed the application test.