Enterprise tax compliance management method, system and electronic device based on knowledge graph

Abstract

The application provides a kind of enterprise tax compliance management method, system and electronic equipment based on knowledge graph, it is related to artificial intelligence and the technical field of tax management, the method comprises: first, the entity node and the relationship edge generated based on the tax-related data obtained from multiple heterogeneous data sources are used to construct an enterprise tax knowledge graph, the matching relationship is determined by performing reconciliation processing on the invoice entity and the business document entity based on the graph, and the matching relationship is updated to the above knowledge graph as new association information, then based on the updated knowledge graph, abnormal topological structure is identified by running graph algorithm, and risk prompt information is generated, thereby solving the technical problems of low accuracy of enterprise tax compliance detection and risk prevention and control lag in the prior art, to realize the efficiency, intelligentization and compliance controllability of tax management, greatly reducing the enterprise tax risk and operating cost.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and tax administration technology, and more specifically, to a knowledge graph-based enterprise tax compliance management method, system, and electronic device. Background Technology

[0002] With the rapid development of the digital economy and the continuous expansion of business scale, especially large group enterprises with long business chains, complex organizational structures, and high transaction frequency, tax management faces unprecedented challenges. Tax compliance is no longer limited to traditional tax filing operations but has evolved into a systematic project encompassing invoice management, input tax deduction, related-party transaction monitoring, and tax policy adaptation. Against this backdrop, tax authorities are continuously promoting information technology construction and upgrading the "data-driven tax administration" regulatory model, requiring enterprises to possess stronger data integration and risk identification capabilities. Simultaneously, enterprises are gradually introducing automation tools into their internal management, attempting to improve tax processing efficiency and reduce human error and compliance risks through technological means.

[0003] However, current enterprise tax management systems generally suffer from severe data silos and insufficient intelligence. Most systems are functionally limited, focusing only on one aspect of invoice recognition, tax declaration, or input tax management, lacking the ability to uniformly model and analyze the entire business data chain. In practice, cross-system document matching relies on manual intervention, making it difficult to handle complex scenarios such as inconsistent product names and multiple invoices / orders; risk control largely remains at the post-event verification stage, failing to achieve pre-event warnings and in-event control; and for frequently updated tax regulations, existing systems mostly use hard-coded rules, resulting in poor adaptability and high maintenance costs.

[0004] In summary, existing technologies suffer from technical problems such as low accuracy in cross-system document matching, lagging risk control, and poor adaptability to tax rules. Summary of the Invention

[0005] The purpose of this invention is to provide a knowledge graph-based enterprise tax compliance management method, system, and electronic device to alleviate the technical problems of low compliance detection accuracy and lagging risk prevention and control in the prior art.

[0006] In a first aspect, embodiments of the present invention provide a knowledge graph-based enterprise tax compliance management method, comprising: constructing an enterprise tax knowledge graph, wherein the enterprise tax knowledge graph includes entity nodes and relationship edges generated based on tax-related data obtained from multiple heterogeneous data sources; the entity nodes include invoice entities and business document entities; based on the enterprise tax knowledge graph, performing reconciliation processing on the invoice entities and the business document entities to determine corresponding matching relationships, and updating the enterprise tax knowledge graph as newly added association information; and based on the updated enterprise tax knowledge graph, identifying abnormal topology structures through a graph algorithm and generating tax risk indication information.

[0007] In some optional implementations, based on the aforementioned enterprise tax knowledge graph, the reconciliation processing of the aforementioned invoice entity and the aforementioned business document entity includes: using natural language processing technology to map the goods name information in the aforementioned invoice entity to standard material codes; and matching the aforementioned invoice entity with the corresponding aforementioned business document entity based on the mapping result.

[0008] In some optional implementations, the above-mentioned invoice entities are matched with the corresponding business document entities, including: when there are multiple invoices and multiple business documents that need to be matched, a mixed integer linear programming model is constructed to solve the problem in order to determine the matching scheme with the minimum total matching difference.

[0009] In some optional implementations, the graph algorithm described above includes a community detection algorithm and / or a loop detection algorithm; identifying abnormal topologies by running the graph algorithm includes: identifying related-party transaction enterprise groups in the updated enterprise tax knowledge graph using the community detection algorithm described above; and / or, detecting closed-loop paths in the cash flow or invoice flow in the updated enterprise tax knowledge graph using the loop detection algorithm described above.

[0010] In some optional implementations, identifying abnormal topologies through the running graph algorithm also includes: performing consistency analysis on the cash flow path and the invoice flow path in the updated enterprise tax knowledge graph; when the entities pointed to by the cash flow path and the invoice flow path are inconsistent, it is identified as a risk of separation of invoice and goods.

[0011] In some optional implementations, the above-mentioned identification of abnormal topological structures based on the updated corporate tax knowledge graph by running graph algorithms includes: applying a graph neural network model to the updated corporate tax knowledge graph to learn the substructure patterns of normal transaction graphs; and detecting abnormal graph substructures based on the output of the graph neural network model to identify the risk of fraudulent invoicing.

[0012] In some optional implementations, the above method further includes: obtaining a tax law knowledge graph containing executable rules based on tax law provisions; and, based on the tax law knowledge graph, determining the compliance of business attribute information associated with invoice entities that have completed automated reconciliation processing.

[0013] Secondly, embodiments of the present invention provide a knowledge graph-based enterprise tax compliance management system, comprising: a graph construction module for constructing an enterprise tax knowledge graph, wherein the enterprise tax knowledge graph includes entity nodes and relationship edges generated based on tax-related data obtained from multiple heterogeneous data sources; the entity nodes include invoice entities and business document entities; a matching and updating module for performing reconciliation processing on the invoice entities and the business document entities based on the enterprise tax knowledge graph, determining the corresponding matching relationships, and updating the matching relationships as new association information in the enterprise tax knowledge graph; and a compliance judgment module for identifying abnormal topology structures and generating tax risk indication information based on the updated enterprise tax knowledge graph using a graph algorithm.

[0014] Thirdly, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the steps of the method described in any of the first aspects above.

[0015] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing computer-executable instructions, which, when invoked and executed by a processor, cause the processor to perform the method described in any of the first aspects above.

[0016] This invention provides a knowledge graph-based enterprise tax compliance management method, system, and electronic device. The method includes: firstly, constructing an enterprise tax knowledge graph based on entity nodes and relationship edges generated from tax-related data obtained from multiple heterogeneous data sources; secondly, performing cross-referencing processing on invoice entities and business document entities based on this graph to determine matching relationships, and updating the knowledge graph as new association information based on these matching relationships; and thirdly, identifying abnormal topology structures and generating risk warning information based on the updated knowledge graph through graph algorithms. This solves the technical problems of low accuracy in enterprise tax compliance detection and lagging risk prevention and control in existing technologies, thereby achieving efficient, intelligent, and compliant tax management, and significantly reducing enterprise tax risks and operating costs. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a knowledge graph-based enterprise tax compliance management method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a knowledge graph-based enterprise tax compliance management system provided in an embodiment of the present invention; Figure 3 A flowchart illustrating another knowledge graph-based enterprise tax compliance management method provided in this embodiment of the invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Currently, tax administration is gradually building an intelligent regulatory system covering the entire business cycle of enterprises by relying on big data technology. This places higher demands on taxpayers' compliance management, especially for large group enterprises with complex organizational structures, characterized by numerous subsidiaries, diverse business models, and millions of invoices processed annually. Traditional tax compliance management relies on manual visual verification of invoices, manual tax calculation using Excel spreadsheets, and decentralized filing. This not only has significant efficiency bottlenecks but also struggles to avoid major tax risks such as "cost accounting with false invoices," "overdue input tax credits," and "non-compliant transfer pricing in related-party transactions." Compliance oversights can lead to hefty tax penalties and a downgrade in tax credit rating. Analysis reveals that current enterprise tax management systems generally suffer from the following structural defects: insufficient modularity of system functions, weak data auditing capabilities, lagging risk control mechanisms, and an imperfect tax policy update mechanism.

[0021] Based on this, the present invention provides a knowledge graph-based enterprise tax compliance management method, system, and electronic device to solve the technical problems of low accuracy in enterprise tax compliance detection and lagging risk prevention and control in the prior art, thereby achieving efficient, intelligent, and compliant tax management, and significantly reducing enterprise tax risks and operating costs.

[0022] To facilitate understanding of this embodiment, a detailed description of a knowledge graph-based enterprise tax compliance management method disclosed in this invention will be provided first. (See [link to relevant documentation]). Figure 1 The diagram illustrates a knowledge graph-based enterprise tax compliance management method, which can be executed by electronic devices and mainly includes the following steps S102 to S106: Step S102: Construct an enterprise tax knowledge graph. The enterprise tax knowledge graph includes entity nodes and relationship edges generated based on tax-related data obtained from multiple heterogeneous data sources. Entity nodes include invoice entities and business document entities. Heterogeneous data sources refer to multidimensional tax-related data sources. Different data sources typically exhibit significant differences in data modality, structural paradigm, semantic granularity, and update mechanisms, and lack unified master data identifiers and semantic mapping rules. Specifically, their heterogeneity can manifest as: format differences, fragmented business semantics, and disconnected rule semantics.

[0023] Preferably, heterogeneous data sources can include: the Golden Tax System Phase IV, ERP systems, expense control systems, and external industrial and commercial / judicial databases. Specifically, the Golden Tax System Phase IV can provide data with strong tax attributes, such as the full lifecycle status of invoices, tax bureau verification results, and tax classification codes; the ERP system can provide structured business documents such as purchase / sales orders and inbound / outbound slips, as well as standard material code master data; the expense control system can provide management semantic data such as expense reimbursement reasons, approval processes, and budget items; the financial system can provide fund flow tracking data such as bank statements, payment records, and counterparty account information; and external industrial and commercial / judicial databases can provide entity risk label data such as enterprise equity penetration relationships, abnormal operating status, and tax-related judicial cases.

[0024] The aforementioned data sources are not accessed in isolation, but rather use a tax domain ontology as a unified semantic framework to complete cross-source alignment during the knowledge extraction stage. For example, taxpayer entities are anchored through the unified social credit code, potential association anchors between business documents and invoices are established through triple cross-references of order number, invoice remarks column, and warehouse entry number, and invoice details are identified through NLP parsing of contract terms and OCR recognition, mapping their common reference objects to the same Product entity in the graph and binding their tax classification code and tax rate rules.

[0025] The enterprise tax knowledge graph constructed in this way is essentially a five-dimensional dynamic semantic network that integrates the evidence chain of business authenticity (documents), the evidence chain of tax legality (invoices), the fund flow trajectory chain (payment), the risk transmission chain of the subject (industrial and commercial / judicial) and the legal constraint chain (tax classification code). It can provide a computable topological foundation for subsequent reconciliation processing, and further provide multimodal risk analysis dimensions for abnormal topology identification.

[0026] In this embodiment, the specific implementation of constructing an enterprise tax knowledge graph may include: firstly, connecting to multi-source heterogeneous data systems to collect multimodal data covering full invoice information, structured fields of business documents, enterprise entity relationship chains, fund flow paths, and tax law texts; then, defining a unified tax graph schema based on the tax domain ontology, transforming unstructured or semi-structured data into entity nodes with clear types and semantic constraints (such as taxpayer "TaxPayer", invoice "Invoice", transaction "Transaction", product "Product", contract "Contract", regulation "Regulation") and relational edges with weights and attributes (such as issuer "IssuedBy", receiver "ReceivedBy", containing goods "Contains", matching documents "Matches", controlling relationship "ControlledBy", applicable regulations "AppliesTo").

[0027] In this context, invoice entities and business document entities are not only explicitly modeled as basic nodes, but also have potential association anchors established in the early stages of graph construction through cross-source identifier alignment (such as unified social credit code, order number, and transaction serial number), providing a computable, inferable, and scalable topological foundation for the reconciliation processing in the subsequent step S104; thus forming a dynamic knowledge graph with semantic consistency, structural evolvability, and rule embedding capability.

[0028] Step S104: Based on the enterprise tax knowledge graph, perform reconciliation processing on the invoice entity and the business document entity, determine the corresponding matching relationship, and update the matching relationship as new association information in the enterprise tax knowledge graph.

[0029] The reconciliation process is the core reasoning process that uses a knowledge graph as a unified semantic space to establish a verifiable, traceable, and evolving mapping relationship between invoice entities and business document entities to ensure business authenticity. In this embodiment, its purpose is to break through the dependence of traditional rule engines or keyword matching on strong field consistency (such as completely identical product names and strictly corresponding order numbers). It can achieve semantic alignment and logical closure of unstructured / semi-structured business documents scattered in heterogeneous systems at the graph topology level, so as to generate structured association relationships (i.e., VerifiedLink edges) with the effect of tax evidence chains, providing a reliable data foundation and reasoning premise for subsequent risk identification and compliance determination.

[0030] Preferably, in one embodiment, the above-mentioned cross-referencing process may include fuzzy matching based on semantic similarity and / or many-to-many matching based on optimization algorithms.

[0031] As a concrete example, semantic similarity-based fuzzy matching utilizes natural language processing models to vectorize and semantically align non-standardized texts such as invoice product names and business document material descriptions, mapping them to a unified standard material coding and tax classification coding system. This bridges matching gaps caused by naming differences, abbreviation ambiguities, and mixed use of industry terms at the semantic level. The matching output is a preliminary set of candidate associations with confidence between nodes, forming the initial input for dynamic graph updates.

[0032] As a concrete example, many-to-many matching based on optimization algorithms, building upon the aforementioned semantic alignment, addresses complex business scenarios where structured attributes such as amount, quantity, and time have reasonable tolerances (e.g., N invoices and M warehouse entry orders have the same total amount but different detail granularities). Using the entity attributes and relational constraints already modeled in the graph as variable boundaries, a mixed-integer linear programming model is constructed to solve for the globally optimal association scheme that minimizes the overall matching residuals (e.g., the weighted sum of amount deviations and the entropy value of quantity errors). The output of this matching is a mathematically provable deterministic matching relationship that meets the requirements of tax authenticity verification. It can be directly injected into the graph as a new association edge, driving the graph structure to evolve towards higher confidence.

[0033] In one embodiment, the reconciliation processing of invoice entities and business document entities based on the enterprise tax knowledge graph may include: using natural language processing technology to map the goods name information in the invoice entity to a standard material code; and matching the invoice entity with the corresponding business document entity based on the mapping result.

[0034] Preferably, natural language processing techniques may include a semantic understanding method jointly driven by a domain-fine-tuned pre-trained language model (such as BERT) and a tax ontology dictionary. The model is trained to adapt to the domain by using defined entity types (such as Product), relational constraints (such as Contains), and regulatory semantic tags (such as TaxTaxonomy) in the tax graph as supervision signals. This enables the model to accurately identify that expressions such as "A4 copy paper," "office paper," and "printing paper" all point to the same standard material code and corresponding tax rate rules in the tax context, thereby avoiding naming ambiguity at the semantic level and ensuring tax consistency in subsequent matching.

[0035] In this embodiment, the standard material code refers to a standardized material identifier defined in the enterprise master data system that establishes a deterministic mapping relationship with the tax classification code. It not only serves as the benchmark unit for procurement, warehousing, and cost allocation in business systems such as ERP, carrying the semantics of procurement / inventory management within the business system, but also forms a dynamic binding with the tax classification code issued by the State Taxation Administration through the mapping relationship. This ensures that the same code points to consistent tax attributes (such as applicable tax rate, deductibility, and tax exemption) throughout all stages, including business documents, invoice issuance, input tax deduction, and tax audit. As a knowledge anchor connecting the business system and tax rules, this code exists both as a key attribute of entity nodes in the graph and as a standardized bridge for cross-source document semantic alignment, making the reconciliation results naturally interpretable and verifiable.

[0036] In one embodiment, matching an invoice entity with its corresponding business document entity may include: when there are multiple invoices and multiple business documents that need to be matched, constructing a mixed-integer linear programming model to solve the problem, in order to determine the matching scheme with the minimum total matching difference.

[0037] Specifically, for complex N:M scenarios involving "multiple invoices and multiple orders" (such as issuing 10 invoices in a month, corresponding to 25 inbound orders, with the same total amount but overlapping details), it can be modeled as a mixed integer linear programming (MILP) problem to find the combination scheme that minimizes the matching residual.

[0038] The mixed-integer linear programming model can be a multi-objective optimization model that takes the structured entity attributes and relational constraints in the knowledge graph as input and aims at verifying the authenticity of tax documents. Its decision variables can represent the potential matching relationship between invoice entities and business document entities (i.e., whether to establish a VerifiedLink edge); the objective function comprehensively weights the deviation terms of various business attributes (such as the absolute value of the amount difference, the quantity error rate, and the tax rate inconsistency penalty term) to minimize the overall matching residual; the constraints can be directly derived from the domain rules already modeled in the knowledge graph, such as: the total amount of the receiving order under the same purchase order must not exceed the order amount tolerance threshold, the invoice issuance time must be within the effective time window of the corresponding business document, and the matching combination must satisfy the predefined path connectivity in the knowledge graph, etc.

[0039] Preferably, the specific implementation of the above embodiments may include: using the solution result of the MILP model as the trigger condition for graph update; only when the matching residual is lower than the preset business credibility threshold, a VerifiedLink relationship edge is solidified in the knowledge graph, and the constraint path on which the matching combination is based (such as order amount tolerance constraint and time window constraint) is marked simultaneously; if the residual exceeds the limit, the unmatched nodes and abnormal deviation dimensions (such as significant tax rate differences, quantity inversion, etc.) are injected into the compliance judgment module as risk clues to drive the subsequent graph algorithm to perform in-depth risk mining on the corresponding subgraph structure.

[0040] This mechanism enables MILP not only to perform matching calculations, but also to become the logical hub connecting the reconciliation processing in step S104 and the abnormal topology identification in step S106. The matching process itself is a lightweight graph compliance check, and its output is both the input for graph updates and the starting point for high-level risk analysis.

[0041] Step S106: Based on the updated enterprise tax knowledge graph, abnormal topology structures are identified through the running graph algorithm, and tax risk indication information is generated.

[0042] In one embodiment, the graph algorithm includes a community detection algorithm and / or a loop detection algorithm; identifying abnormal topology by running the graph algorithm may include: identifying related-party transaction groups in the updated corporate tax knowledge graph using the community detection algorithm; and / or, detecting closed-loop paths in the flow of funds or invoices in the updated corporate tax knowledge graph using the loop detection algorithm.

[0043] The community discovery algorithm can be applied to the updated enterprise tax knowledge graph. Specifically, its input node set can include invoice entities, business document entities, and corresponding taxpayer entities confirmed by the cross-referencing process in step S104. Its input edge set can include: VerifiedLink (matching relationship between invoices and business documents), IssuedBy / ReceivedBy (invoice issuance and receipt relationship), Payment (fund payment relationship), and ControlledBy (equity control relationship). The algorithm uses node degree centrality, cumulative edge weight (such as cumulative invoice amount and payment frequency), and path connectivity as clustering criteria to identify subgraph structures that meet certain conditions. For example, the number of taxpayer entities in the subgraph is not less than 3, the density of VerifiedLink edges is more than 1.5 times higher than the average density of the entire graph, and there is at least one ControlledBy-Payment-IssuedBy composite path with a length not greater than 4. This subgraph is marked as a group of related transaction enterprises, and risk warning information containing the unified social credit code of taxpayers in the group, core path topology description, and risk level identifier is generated.

[0044] The loop detection algorithm can be applied to the same updated knowledge graph. Specifically, its detection target can be a closed path composed of directed edges. The path node type can be limited to taxpayers and invoices appearing alternately, and the path edge type can be limited to one or more combinations of issuer, receiver, and fund payment relationship. For example, when the starting point and ending point of the path are detected to be the same taxpayer entity, the total path length is not less than 3, and the path contains at least one fund payment relationship edge and one issuer edge, and the tax classification codes of the invoice entities involved in the path belong to the same commodity category, it is determined to be a closed loop path of fund flow and invoice flow. This closed loop path is extracted as an abnormal topology instance, and risk warning information containing the path node sequence, edge type sequence, invoice codes involved in the closed loop, and corresponding risk types (such as: circular invoicing, fund return) is generated.

[0045] The node types, relation edge types, constraints, and output formats on which the above algorithm depends are all clearly defined by the knowledge graph schema constructed in step S102, and dynamically strengthened after the VerifiedLink edge is added in step S104; its recognition results directly serve as the data basis for generating tax risk indication information in step S106, forming a closed-loop technical logic chain of "graph construction - reconciliation update - topology recognition - risk output".

[0046] In another embodiment, identifying abnormal topology structures through a graph algorithm may further include: performing consistency analysis on the cash flow path and the invoice flow path in the updated enterprise tax knowledge graph; when the entities pointed to by the cash flow path and the invoice flow path are inconsistent, it is identified as a risk of separation of invoice and goods.

[0047] Specifically, the consistency analysis of the fund flow path and the invoice flow path can be performed using the VerifiedLink edge confirmed in step S104 of the updated enterprise tax knowledge graph as the starting point, and bidirectional path expansion along the fund payment relationship edge and the issuer / receiver relationship edge respectively: the fund flow path is defined as a directed path starting from the invoice recipient taxpayer entity, passing through the fund payment relationship edge to the payment recipient taxpayer entity; the invoice flow path can be defined as a directed path starting from the same invoice recipient taxpayer entity, passing through the recipient edge back to the issuer taxpayer entity. When the taxpayer entity sets pointed to by the two paths within the second-level jump range do not overlap, and this non-overlapping relationship cannot be explained by the principal-agent relationship carried by the contract entity already modeled in the graph (e.g., Company A issues an invoice to Company B, but Company B's funds flow to Company C, which is unrelated to Company A), it is judged as a risk of invoice-goods separation, and a risk warning information of invoice-goods separation containing the path start and end nodes, inconsistent jump level, corresponding invoice code, and risk identifier is generated.

[0048] In one embodiment, identifying abnormal topological structures based on an updated corporate tax knowledge graph by running a graph algorithm may include: applying a graph neural network model to the updated corporate tax knowledge graph to learn substructure patterns of normal transaction graphs; and detecting abnormal graph substructures based on the output of the graph neural network model to identify the risk of fraudulent invoicing.

[0049] Specifically, the graph neural network model can take the updated enterprise tax knowledge graph as input. Its node feature vector is composed of the structured attributes (including amount, quantity, timestamp, tax classification code, and standard material code) of invoice entities, business document entities, and taxpayer entities, along with semantic embedding vectors. Its edge feature vector is composed of the type identifier and weight value of the relationship types such as matching relationship between invoices and business documents, fund payment relationship, and issuer. The model learns the local substructure pattern distribution on historical normal transaction graph samples in an unsupervised manner. During the inference phase, it calculates the anomaly score for the k-hop neighborhood subgraph of each invoice entity in the current graph. When the score exceeds a preset threshold and the subgraph contains a high-confidence loop structure or a low-connectivity isolated group, it is identified as a risk of fraudulent invoicing and generates risk warning information containing the invoice code of the abnormal subgraph center, a list of neighborhood nodes, abnormal pattern type, and risk level.

[0050] In one embodiment, the method may further include: acquiring a tax law knowledge graph, which contains executable rules based on tax law provisions; and, based on the tax law knowledge graph, determining the compliance of business attribute information associated with invoice entities that have undergone automated reconciliation processing. Preferably, the compliance determination may include a determination of the compliance of input tax deduction.

[0051] Specifically, the tax regulations knowledge graph and the enterprise tax knowledge graph can establish a semantic anchoring relationship through tax classification codes; in particular, the invoice entities and commodity entities in the enterprise tax knowledge graph are explicitly associated with tax classification codes, which form a deterministic mapping with the standard material codes.

[0052] During the compliance determination process, the system calls upon the business attribute information associated with the invoice entity that has completed automated reconciliation processing. Specifically, this may include: the purpose field value of the business document entity transmitted via the VerifiedLink edge, the budget subject code, the transaction nature tag extracted from the contract text, and the tax classification code and invoice date carried by the invoice entity itself. Using the above business attribute information as input parameters, a rule matching operation is performed in the tax law knowledge graph. The tax law knowledge graph contains Regulation entities and TaxRule entities generated by the structured generation of tax law provisions, as well as the relationship edges AppliesTo, Prohibits, and Requires between them. When the tax classification code pointed to by the AppliesTo relationship of the TaxRule entity is consistent with the tax classification code associated with the current invoice entity, and its Prohibits attribute explicitly limits "input tax cannot be deducted from output tax", and its applicable conditions match the extracted purpose field value, budget subject code, or transaction nature tag, then it is determined that the invoice entity does not meet the input tax deduction conditions.

[0053] Furthermore, the generated compliance determination result can include the TaxRule entity number, the original text of the corresponding legal provision, a description of the violation, and the operation to be performed, "transfer of input tax," and is written as structured data into the updated corporate tax knowledge graph, which together with the VerifiedLink edge constitutes a dual verification basis to support the generation of risk warning information.

[0054] Based on the same inventive concept, this invention also provides a knowledge graph-based enterprise tax compliance management system, see [link to relevant documentation]. Figure 2 As shown, the system mainly includes the following parts: The graph construction module 210 is used to construct an enterprise tax knowledge graph. The enterprise tax knowledge graph includes entity nodes and relationship edges generated based on tax-related data obtained from multiple heterogeneous data sources. Entity nodes include invoice entities and business document entities. The matching and updating module 220 is used to perform reconciliation processing on invoice entities and business document entities based on the enterprise tax knowledge graph, determine the corresponding matching relationship, and update the matching relationship as new association information to the enterprise tax knowledge graph. The compliance assessment module 230 is used to identify abnormal topology structures and generate tax risk indication information based on the updated enterprise tax knowledge graph and the graph algorithm.

[0055] The knowledge graph-based enterprise tax compliance management system provided in this invention embodiment can be specific hardware on a device or software or firmware installed on the device. The system provided in this invention embodiment has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0056] To facilitate understanding, this invention also provides an application example of a knowledge graph-based enterprise tax compliance management method, which is described in detail with reference to a knowledge graph-based enterprise tax compliance management system that applies this method. In this example, the specific application scenario is the intelligent tax management of a large group enterprise, aiming to solve problems such as scattered tax data, difficulty in identifying false invoices, hidden risks in related-party transactions, low efficiency of manual reconciliation, and high costs of tax compliance. This system integrates knowledge graph, graph neural network (GNN), natural language processing (NLP), and rule engine technologies to construct a closed-loop tax management system encompassing "full invoice collection - intelligent reconciliation - graph-based risk control - compliance early warning - automatic declaration."

[0057] See Figure 3 The diagram shows another knowledge graph-based enterprise tax compliance management method. The core technology implementation process of this method mainly includes the following steps S301 to S306: Step S301: Collect multi-source heterogeneous tax data and construct a graph ontology.

[0058] The system interfaces with the Golden Tax System Phase IV, ERP systems, expense control systems, financial systems, and external industrial and commercial / judicial databases. It collects invoice data (full invoice information), business documents (contracts, orders, warehouse receipts), fund flows, enterprise industrial and commercial relationships, and tax regulations. Based on ontology, it defines a tax graph schema, constructing a heterogeneous knowledge graph containing entities such as "taxpayer, counterparty, invoice, goods, contract, funds, and regulations" and their relationships (e.g., "issued," "paid," "belongs to category").

[0059] Tax administration involves extremely fragmented and heterogeneous data, and building a unified knowledge graph is the foundation for intelligentization.

[0060] First, the data acquisition layer obtains data from multidimensional data sources through APIs, ETL tools, and RPA robots.

[0061] 1. Invoice Data: Connects to the State Taxation Administration's ledger system and OCR recognition engine to obtain VAT special / general invoices, electronic invoices, customs payment receipts, and withholding tax bills. The collected fields include not only amount, tax amount, and date, but also detailed information such as goods name, specifications, unit price, and tax rate in the detailed rows.

[0062] 2. Business Data: Synchronize purchase orders, sales orders, receiving slips, and delivery slips from ERP (SAP / Oracle); synchronize contract terms, subject matter, and settlement methods from the contract management system; and synchronize bank receipts and payment records from the funds system.

[0063] 3. Main data: Synchronize customer and supplier information from CRM / SRM; crawl business registration information, equity structure, director and supervisor relationships, list of abnormal operations, and tax bureau credit rating (A / B / M / C / D) from external data sources (Qichacha / Tianyancha).

[0064] 4. Regulatory data: Collect laws and regulations, tax incentive policies, and immediate refund catalogs issued by the State Taxation Administration and local tax bureaus.

[0065] Secondly, construct the tax domain ontology. Define entity types: `TaxPayer`, `Invoice`, `Transaction`, `Product`, `Contract`, and `Regulation`. Define relationship types: `IssuedBy`, `ReceivedBy`, `Contains`, `Matches`, `ControlledBy`, and `AppliesTo`.

[0066] The system utilizes knowledge extraction technology to structure unstructured data. For example, it uses NLP techniques to extract key information such as "transaction amount" and "time of tax liability" from contract texts, materializing them as graph nodes. Ultimately, it constructs an enterprise-level tax knowledge graph containing hundreds of millions of nodes and edges, providing a data foundation for subsequent correlation analysis.

[0067] Step S302: Intelligent invoice recognition and verification based on semantic understanding.

[0068] OCR and NLP technologies are used to extract structured data from VAT invoices (both special and general), customs payment receipts, etc. Semantic matching algorithms automatically map the "goods name" on the invoice to the enterprise's standard material code and tax taxonomy code. Simultaneously, a real-time connection to the tax bureau's verification interface is used to check the invoice status (normal, void, red-inked, out of control), and image tampering detection technology is employed to identify PS-forged invoices, ensuring the authenticity and reliability of the data source.

[0069] The difficulty in invoice processing lies in understanding unstructured information and verifying its authenticity. Therefore, the application examples of the method provided in this embodiment mainly include the implementation of the following functions: Intelligent Recognition and Structured Interpretation: Traditional OCR can only recognize text and cannot understand semantics. This system introduces a BERT-based semantic understanding model. For the "goods name" in the invoice details, the model automatically maps it to the company's "standard material code" and the tax bureau's "tax classification code" through semantic vector matching. For example, "A4 copy paper," "office paper," and "printing paper" on the invoice are uniformly mapped to "office supplies - paper," and the system automatically checks whether the tax rate matches (e.g., 13%). This solves the problem of reconciliation failure caused by "name inconsistencies."

[0070] Full invoice verification and status monitoring: The system not only verifies the four elements of an invoice (code, number, amount, and date), but also performs full invoice verification. By connecting to an invoice verification platform, it obtains all invoice information in real time and compares it with OCR recognition results to prevent tampering.

[0071] More importantly, it provides full lifecycle status monitoring. The system performs daily batch scans of the status of invoices that have not been certified for deduction. Once it is discovered that a supplier has "voided" or "reversed" (malicious reversal) an invoice that has already been delivered, or that the supplier has been listed as an "out-of-control account" or "abnormal account" by the tax authorities, the system immediately triggers a red alert, freezes the payment process for the corresponding invoice, and prevents the company from suffering financial losses (payment that cannot be deducted).

[0072] Image tampering detection: To address the risks of duplicate reimbursement of electronic invoices and Photoshop tampering, the system integrates image forensics algorithms. By analyzing the image's lighting consistency, noise distribution (ELA analysis), and metadata (Exif) tampering traces, it identifies Photoshop-forged invoices. Simultaneously, it calculates the digital fingerprint (hash) of the invoice file for duplicate checking across the entire group, preventing the same electronic invoice from being reimbursed repeatedly in subsidiaries A and B.

[0073] Step S303: Full-link automated reconciliation and three-document matching of multiple documents.

[0074] Dynamic links between invoice entities and business entities are established within the knowledge graph. The system executes an automated reconciliation algorithm to match "invoice - purchase order (PO) - receiving / acceptance slip - payment slip" across systems. For complex scenarios such as "one invoice with multiple orders," "multiple invoices with one slip," and price discrepancies, fuzzy matching and linear programming algorithms are used to find the optimal matching combination, achieving automated three / four-slip matching with clear flow and consistent amounts, thus ensuring the authenticity of input tax deduction transactions.

[0075] The core requirement for tax compliance is "consistency of three documents" (invoice, contract / order, and logistics / warehousing slip). Traditional manual reconciliation is time-consuming, labor-intensive, and prone to errors.

[0076] This system constructs an automatic cross-referencing engine based on graph path search. In the knowledge graph, invoices, orders, and inbound slips are all nodes, and cross-referencing is about finding the best matching subgraph between them.

[0077] Multi-dimensional matching algorithm: 1. Strong rule matching: Direct association based on explicit referencing relationships (such as order number being filled in the invoice remarks column).

[0078] 2. Fuzzy Matching: Matches based on supplier name, material name (semantic similarity), quantity, and amount. A tolerance range can be set (e.g., amount error < 0.05 yuan).

[0079] 3. Linear Programming Matching: For complex N:M scenarios with "multiple invoices and multiple orders" (e.g., 10 invoices issued in a month, corresponding to 25 inbound orders, with the same total amount but overlapping details), the system models it as a mixed integer linear programming (MILP) problem, solving for the combination scheme that minimizes the matching residual.

[0080] Anomaly Handling Process: The system automatically identifies discrepancies in the data. Price discrepancy: The unit price on the invoice does not match the unit price on the order. The system determines whether the discrepancy is within the allowable threshold (such as the price adjustment stipulated in the contract). If it exceeds the threshold, the purchasing agent's confirmation process is triggered.

[0081] Quantity discrepancy: The quantity on the invoice is greater than the quantity received (possibly due to fraudulent invoicing) or less than the quantity received (provisional accounting entry).

[0082] Tax rate discrepancy: If the invoice tax rate does not match the contract agreement (e.g., the agreed tax rate is 13% but the invoice is issued with a simplified tax rate of 3%), the system will automatically calculate the non-deductible tax amount and alert you to the risk.

[0083] After the reconciliation is completed, the system generates a `VerifiedLink` edge in the graph to solidify the matching relationship, which serves as the basis for the subsequent automatic generation of accounting vouchers and deduction certification.

[0084] Step S304: Risk mining of related transactions and fraudulent invoicing based on graph algorithms.

[0085] Utilize graph computing engines (such as GraphX / Neo4j) to run risk detection algorithms on the tax graph. Community detection algorithms identify hidden related-party transaction groups; cycle detection algorithms detect abnormal circular invoicing behavior (such as A->B->C->A inflating revenue); path analysis tracks the matching degree between fund flows and invoice flows, identifying characteristics of fraudulent invoicing such as "invoice-goods separation" and "invoicing under different names," and outputting risk warning reports.

[0086] Currently, corporate tax risks have shifted from single invoice risks to complex risks related to related-party transactions and business logic. This system utilizes graph neural networks (GNNs) to uncover deeper hidden risks.

[0087] Related-party transaction identification: Within the data graph, community discovery algorithms (such as Louvain) are used to identify closely related groups of companies. Combined with business registration and equity penetration paths (A controls B, B invests in C), hidden related parties are identified. The system automatically scans transaction pricing between related parties; if it significantly deviates from fair market prices (through price comparisons of similar products), a "transfer pricing risk" alert is displayed.

[0088] Detection of fraudulent invoicing gangs and circular invoicing: Fraudulent invoicing gangs usually have a characteristic topological structure.

[0089] Loop detection: Using DFS / BFS algorithms to detect loops in the flow of funds or invoices (A->B->C->A), which usually means inflated revenue without actual logistics.

[0090] Abnormal convergence / divergence: Identify a large number of newly established sole proprietorships issuing invoices to core enterprises in a short period of time ("invoice laundering" risk), or core enterprises issuing invoices to a large number of unrelated shell companies (inflating costs).

[0091] Separation of invoices and goods: Compare the consistency between "funds flow" and "invoice flow" in the data map. If company A issues an invoice to company B, but B's funds flow to company C (C is unrelated to A), it indicates the risk of "invoice and payment discrepancy".

[0092] Upstream and downstream penetration risk: The system calculates the "tax credit transmission value" for each supplier. If a supplier's upstream supplier is identified by the tax authorities as engaging in fraudulent invoicing (absconding or disappearing), the risk will be transmitted to this company along the tax network. The system will issue an early warning and recommend suspending cooperation with this high-risk supplier.

[0093] Step S305: Tax law knowledge reasoning and compliance pre-audit.

[0094] A structured tax law knowledge graph is constructed to transform ambiguous tax law provisions into actionable logical rules (such as "input tax cannot be deducted for goods purchased for collective welfare"). Before invoice authentication and deduction, the system performs compliance reasoning based on business scenarios (such as purpose and department). For non-deductible items, the system automatically calculates the input tax transfer amount, and for tax differences caused by tax rate adjustments, it automatically generates adjustment suggestions to prevent compliance risks.

[0095] Tax compliance is not just about data alignment, but also about assessing the compliance of business scenarios. The system has built a reasonable legal knowledge base.

[0096] Structuring regulations: Transforming tax law provisions into rule-based logical rules or knowledge graph reasoning paths. For example, the Provisional Regulations on Value-Added Tax stipulate that "input tax on goods purchased for collective welfare purposes shall not be deducted from output tax." In this embodiment, the system can convert this into a rule: `IF Invoice.Type == 'Purchase' AND Invoice.ItemCategory IN ['Food', 'Gift'] AND Expense.Department == 'LaborUnion' THEN Deductible = False`.

[0097] Scenario-based compliance pre-screening: Before invoices enter the "certified deduction pool," the system conducts a pre-screening based on the reconciled business background. 1. Input VAT Reversal Judgment: Identify whether the corresponding warehouse receipt for the invoice is used for tax-exempt projects, collective welfare, or personal consumption. If so, the system automatically marks the invoice as requiring "input VAT reversal" to prevent the risk of tax evasion due to erroneous deduction.

[0098] 2. Tax Incentive Applicability: Determine if the transaction meets the conditions for immediate tax refund or additional deduction (e.g., 5% additional deduction for advanced manufacturing industries). Based on the latest preferential policy database, the system automatically labels invoices that qualify for preferential treatment to prevent under-enjoyment of benefits.

[0099] 3. Invoice compliance check: Check whether the remarks column is filled in as required (e.g., construction services must fill in the project location, transportation services must fill in the origin).

[0100] Dynamic policy updates: The system monitors the State Taxation Administration's official website in real time through web crawlers. When a new policy is released (such as a tax rate reduction from 13% to 11%), the system uses NLP to automatically analyze the change points, update the rule engine, and perform retrospective simulations on historical unreported data to indicate the scope of impact.

[0101] Step S306: Automated tax base calculation and generation of tax return form.

[0102] Based on the results of reconciliation and compliance review, the system automatically collects output tax and input tax, and calculates the tax base for various taxes such as VAT and corporate income tax. It automatically generates a pre-filing form and compares it with the financial statements (invoice-form comparison, form-to-form comparison). After confirmation, the system completes the tax declaration with one click via RPA (Robotic Process Automation) or a direct connection interface, and continuously monitors fluctuations in the tax burden rate, providing a tax health diagnosis.

[0103] The ultimate goal of tax administration is accurate tax declaration.

[0104] Automatic tax base calculation: The system calculates the tax payable of all taxpayers in the entire group in real time.

[0105] Value Added Tax (VAT): `Tax Payable = Current Period Output (from Sales Invoices) - Current Period Input (from Verified and Compliant Purchase Invoices) + Input Transferred Out - Previous Period Carryforward`.

[0106] Corporate Income Tax: Tax adjustments are automatically made based on financial profit. Invoices for "Business Entertainment Expenses" and "Advertising Expenses" identified using the graph are adjusted according to the deduction ratios stipulated by tax law (e.g., entertainment expenses 60% and not exceeding 0.5% of revenue), generating a detailed tax adjustment statement.

[0107] Automatic form filling and RPA reporting: The system includes built-in tax return templates (XML / Excel format) from tax bureaus across all provinces and cities in China. Based on the calculation results, it automatically fills in the main table and dozens of supplementary tables (such as the "Detailed Statement of Input Tax Deduction Structure for the Current Period").

[0108] After completing the forms, perform a self-check by comparing the invoice data with the invoicing system data and comparing the data from the VAT form with the income tax form.

[0109] After confirming that everything is correct, the RPA robot logs into the e-Tax Bureau, automatically completes data upload, declaration submission, tax withholding, and downloads and archives the tax payment certificate. The entire process is fully automated and "unattended".

[0110] In a specific embodiment of the present invention, by deeply integrating the enterprise tax knowledge graph with multi-source heterogeneous data and coupling it with technologies such as natural language processing, mixed-integer linear programming, graph neural networks, and rule-based reasoning, the constructed tax compliance management method can achieve the following technical effects when actually deployed in large group enterprise scenarios (such as annual invoice processing volume exceeding one million, spanning more than ten ERP systems and a three-level or higher legal entity structure): In terms of risk prevention and control, the system achieves closed-loop identification and dynamic early warning of typical tax risks such as fake invoices, fraudulent invoicing groups, separation of invoices and goods, and abnormal transfer pricing in related transactions by performing structured extraction of full invoice information, semantic-level goods name mapping, and multimodal verification (including real-time status verification of tax bureau interfaces, image tampering detection, and digital fingerprint deduplication). Combined with community discovery algorithms, loop detection algorithms, and consistency analysis of fund flow and invoice flow running on the updated knowledge graph, the system integrates these with community discovery algorithms, loop detection algorithms, and consistency analysis of fund flow and invoice flow. Empirical results show that this method significantly improves the accuracy of identifying high-risk invoices through full invoice verification and graph-based risk control. The interception rate of fake invoices and illegal deductions can reach 100%, effectively avoiding the risks of supplementary tax, late payment fees, and downgrade of tax credit rating caused by false input tax.

[0111] In terms of efficiency improvement, the system achieves initial screening and matching of invoices and business documents based on standard material coding mapping and semantic similarity calculation. It also models the complex scenario of "multiple invoices and multiple orders" as a mixed integer linear programming problem to solve for the optimal reconciliation combination, thereby increasing the automated reconciliation coverage from about 40% in the traditional manual mode to 95%. On this basis, it automatically generates accounting vouchers and deduction lists based on the fixed verification relationships in the graph, and drives the direct connection of electronic tax bureau declaration through RPA, shortening the cycle of a single group-level VAT tax declaration from several days to minutes, and improving the efficiency of the financial shared service center by 3 times.

[0112] In terms of compliance enhancement, the system integrates a structured tax law knowledge graph, transforming tax law provisions into executable logical rules (such as "input tax cannot be deducted for goods purchased for collective welfare"). Combined with the business attribute information obtained through cross-checking, it conducts scenario-based compliance pre-audit, which not only automatically completes the calculation of input tax reversal and the identification of tax incentive eligibility (such as the applicability verification of the additional deduction for advanced manufacturing industries), but also conducts retrospective simulations on historical undeclared data based on the dynamic policy update mechanism. Thus, while ensuring compliance bottom line, it accurately identifies preferential policies that should be enjoyed but have not been enjoyed, and optimizes the input tax certification rhythm to improve operating cash flow.

[0113] In terms of decision support, the system constructs a group-level tax health indicator system based on a continuously updated corporate tax knowledge graph. It aggregates the tax burden rate, risk exposure index, regulatory compliance, and supply chain tax transmission risk value of each subsidiary in real time, forming a visualized panoramic view that can penetrate down to a single invoice, a single transaction, and a single related path. This provides management with an interpretable and traceable data foundation and decision-making basis for tax planning, supplier access assessment, and organizational restructuring.

[0114] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, specifically, the electronic device includes a processor and a storage device; the storage device stores a computer program, and the computer program, when run by the processor, executes the method described in any of the above embodiments.

[0115] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 400 includes: a processor 410, a memory 420, a communication interface 430, and a bus 440. The memory 420 stores machine-readable instructions that can be executed by the processor 410. When the electronic device is running, the processor 410 communicates with the memory 420 through the bus 440. The processor 410 executes the machine-readable instructions to perform the steps of the method described above.

[0116] Specifically, the memory 420 and processor 410 can be general-purpose memory and processor, without any specific limitations. When the processor 410 runs the computer program stored in the memory 420, it can execute the above method.

[0117] Processor 410 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 410 or by instructions in software form. The processor 410 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory 420, and processor 410 reads the information from memory 420 and, in conjunction with its hardware, completes the steps of the above method.

[0118] Corresponding to the above method, this embodiment of the invention also provides a computer-readable storage medium storing machine-executable instructions. When the computer-executable instructions are called and run by a processor, the computer-executable instructions cause the processor to perform the steps of the above method.

[0119] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0120] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] Furthermore, the functional modules in the various embodiments of the present invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0122] It should be noted that if the functionality is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0124] The above description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A knowledge graph-based enterprise tax compliance management method, characterized in that, The method includes: Construct an enterprise tax knowledge graph, which includes entity nodes and relationship edges generated based on tax-related data obtained from multiple heterogeneous data sources; the entity nodes include invoice entities and business document entities; Based on the enterprise tax knowledge graph, the invoice entity and the business document entity are reconciled to determine the corresponding matching relationship, and the matching relationship is updated to the enterprise tax knowledge graph as new association information; Based on the updated enterprise tax knowledge graph, abnormal topology structures are identified through a graph algorithm, and tax risk indication information is generated.

2. The method according to claim 1, characterized in that, Based on the enterprise tax knowledge graph, the reconciliation processing of the invoice entity and the business document entity includes: Using natural language processing technology, the goods name information in the invoice entity is mapped to standard material codes; Based on the mapping results, the invoice entity is matched with the corresponding business document entity.

3. The method according to claim 2, characterized in that, Matching the invoice entity with the corresponding business document entity includes: When there are multiple invoices and multiple business documents that need to be matched, a mixed-integer linear programming model is constructed to solve the problem and determine the matching scheme with the minimum total matching difference.

4. The method according to claim 1, characterized in that, The graph algorithm includes community detection algorithms and / or loop detection algorithms; identifying anomalous topology structures by running the graph algorithm includes: The community discovery algorithm is used to identify related-party transaction groups in the updated corporate tax knowledge graph. And / or, through the loop detection algorithm, detect closed-loop paths in the updated enterprise tax knowledge graph in the flow of funds or invoices.

5. The method according to claim 4, characterized in that, Identifying anomalous topologies using graph algorithms also includes: In the updated enterprise tax knowledge graph, consistency analysis is performed on the cash flow path and the invoice flow path; when the entities pointed to by the cash flow path and the invoice flow path are inconsistent, it is identified as a risk of separation of invoice and goods.

6. The method according to claim 1, characterized in that, The step of identifying abnormal topology structures based on the updated enterprise tax knowledge graph using a graph algorithm includes: The graph neural network model is applied to the updated corporate tax knowledge graph to learn the substructure patterns of the normal transaction graph; Based on the output of the graph neural network model, abnormal graph substructures are detected to identify the risk of fraudulent opening.

7. The method according to claim 1, characterized in that, The method further includes: Obtain a tax law knowledge graph, which contains executable rules based on tax law provisions; Based on the aforementioned tax law knowledge graph, compliance determination is made for the business attribute information associated with invoice entities that have completed automated reconciliation processing.

8. A knowledge graph-based enterprise tax compliance management system, characterized in that, The system includes: The graph construction module is used to construct an enterprise tax knowledge graph, which includes entity nodes and relationship edges generated based on tax-related data obtained from multiple heterogeneous data sources; the entity nodes include invoice entities and business document entities. The matching and updating module is used to perform cross-referencing processing on the invoice entity and the business document entity based on the enterprise tax knowledge graph, determine the corresponding matching relationship, and update the matching relationship as new association information to the enterprise tax knowledge graph. The compliance assessment module is used to identify abnormal topology structures and generate tax risk indication information based on the updated enterprise tax knowledge graph through a graph algorithm.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when invoked and executed by a processor, cause the processor to perform the method according to any one of claims 1 to 7.

Images