Intelligent management and control system for enterprise finance and taxation risks

Abstract

The invention provides an intelligent management and control system for enterprise finance and taxation risks. Invoice data, enterprise finance data and industry tax data of a target enterprise are acquired through a data acquisition module; through a data fusion module, performing distributed cleaning on abnormal data in the invoice data, the enterprise financial data and the industry tax data based on the established data mapping rule base, and performing real-time fusion to obtain a dynamic data pool; using a risk diagnosis module to extract compliance rules from enterprise finance and taxation legal provisions as conditions, using the trained random forest regression model to predict the dynamic data pool to obtain a reasonable tax threshold value, and determining a risk diagnosis result according to the reasonable tax threshold value; the risk diagnosis result is analyzed through the early warning module, the weight of each risk index is obtained and used for early warning the enterprise finance and taxation risk and generating the risk report of the target enterprise, the island problem of enterprise finance and taxation data is solved, and the finance and taxation management efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital management of corporate finance and taxation, and in particular to an intelligent management and control system for corporate finance and taxation risks. Background Art

[0002] Corporate financial and tax management is a very important part of corporate management. As the market economy becomes increasingly standardized and perfect, the role of financial and tax management in corporate competition is becoming more and more significant.

[0003] Currently, corporate financial, tax, and business data are scattered across multiple isolated systems, requiring manual collection and cross-checking, which is inefficient and error-prone. Traditional risk detection methods rely primarily on fixed threshold rules, but this approach cannot dynamically adapt to policy changes or business fluctuations. Companies often fail to adapt to new tax refund regulations in a timely manner, leading to underreporting or misreporting of risks and overpayment of taxes. Furthermore, after risk discovery, there is a lack of standardized tools and reliance on manual tracking, resulting in delayed rectification or buck-passing.

[0004] Therefore, how to carry out financial and tax management work scientifically and efficiently has become an issue that enterprises urgently need to solve. Summary of the Invention

[0005] The present invention provides an intelligent management and control system for enterprise financial and tax risks, the purpose of which is to solve the island problem of enterprise financial and tax data and improve the efficiency of financial and tax management.

[0006] To achieve the above objectives, the present invention provides an intelligent enterprise financial and tax risk management and control system, comprising:

[0007] Data acquisition module, used to obtain invoice data, corporate financial data and industry tax data of target enterprises;

[0008] The data fusion module is used to perform distributed cleaning of abnormal data in invoice data, corporate financial data, and industry tax data using the established data mapping rule library, and to fuse the cleaned data in real time to obtain a dynamic data pool;

[0009] The risk diagnosis module is used to extract compliance rules from corporate financial and tax laws and regulations. Based on the compliance rules, the module uses a trained random forest regression model to predict the dynamic data pool, obtain a reasonable tax burden threshold, and then determine the risk diagnosis result based on the reasonable tax burden threshold.

[0010] The early warning module is used to analyze the risk diagnosis results, obtain the weight of each risk indicator, issue early warnings on corporate financial and tax risks based on the weight of each risk indicator, and generate risk reports for target enterprises.

[0011] Specifically, we obtain the target company's invoice data, corporate financial data, and industry tax data, including:

[0012] Obtain the target company's invoice data through the invoice full-factor information query interface of the tax open platform;

[0013] Extract the target company's financial data from the financial system through a network interactive interface. The financial data includes contract numbers, logistics order numbers, bank statements, and tax rates.

[0014] Obtain industry tax burden data from the industry tax burden rate database through a network interactive interface.

[0015] Furthermore, based on the established data mapping rule base, abnormal data in invoice data, corporate financial data, and industry tax data are cleaned in a distributed manner, including:

[0016] Establish a standardized field mapping table based on source system fields, target system fields, and conversion rules;

[0017] Build a data mapping rule base based on the established standardized field mapping table;

[0018] Utilize the data mapping rule library to detect abnormal data in invoice data, corporate financial data, and industry tax data, and use the big data processing engine to build a cleaning pipeline to filter out all abnormal data.

[0019] Furthermore, the trained random forest regression model is used to make predictions on the dynamic data pool, including:

[0020] Analyze corporate finance and taxation laws and regulations to obtain policy impact coefficients;

[0021] The policy impact coefficient, the company's historical tax rate, and the industry benchmark tax rate are used to train the random forest regression model to obtain a trained random forest regression model;

[0022] The dynamic data pool is input into the trained random forest regression model to predict the tax burden threshold.

[0023] Furthermore, the risk diagnosis results are analyzed to obtain the weights of each risk indicator, including:

[0024] Compare the importance of each pair of risk indicators in the risk diagnosis results through expert scoring to form a judgment matrix;

[0025] Calculate the geometric mean of each risk indicator based on the judgment matrix;

[0026] The weight of each risk indicator is calculated using the geometric mean of all risk indicators.

[0027] Furthermore, before issuing early warnings for corporate fiscal and tax risks based on the weights of various risk indicators, the following also needs to be considered:

[0028] The weight of each risk indicator is checked for consistency based on the judgment matrix to determine whether the weight of each risk indicator is reasonably distributed.

[0029] Furthermore, the weight of each risk indicator is checked for consistency based on the judgment matrix, including:

[0030] Multiply the judgment matrix by the weight of each risk indicator to obtain a new vector for each risk indicator;

[0031] Calculate the maximum eigenvalue using the new vector of each risk indicator;

[0032] The consistency index is calculated based on the maximum eigenvalue;

[0033] The consistency index and the random consistency index obtained by table lookup are used to calculate and obtain the test consistency ratio;

[0034] The consistency test of each risk indicator weight is performed based on the test consistency ratio.

[0035] Furthermore, the expression for early warning of corporate fiscal and tax risks based on the weights of various risk indicators is:

[0036]

[0037] Among them, a represents the warning level, w i represents the weight of the i-th risk indicator, r i represents the i-th risk indicator, n represents the number of risk indicators, and ∈ represents the dynamic adjustment factor.

[0038] The above solution of the present invention has the following beneficial effects:

[0039] Compared with the existing technology, the present invention obtains the invoice data, enterprise financial data and industry tax burden data of the target enterprise through the data acquisition module; through the data fusion module, based on the established data mapping rule library, the abnormal data in the invoice data, enterprise financial data and industry tax burden data are distributedly cleaned, and the cleaned data are fused in real time to obtain a dynamic data pool; using the risk diagnosis module, compliance rules are extracted from the enterprise's financial and tax legal provisions, and based on the compliance rules, the dynamic data pool is predicted using the trained random forest regression model to obtain a reasonable tax burden threshold, and the risk diagnosis result is determined according to the reasonable tax burden threshold; the risk diagnosis result is analyzed by the early warning module to obtain the weight of each risk indicator, and based on the weight of each risk indicator, the enterprise's financial and tax risks are warned and a risk report for the target enterprise is generated, which solves the island problem of enterprise financial and tax data and improves the efficiency of financial and tax management.

[0040] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 2 is a system structure diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0042] To make the technical problems, technical solutions, and advantages to be solved by the present invention more clear, the following is a detailed description with reference to the accompanying drawings and specific embodiments. It is obvious that the embodiments described are only some of the embodiments of the present invention, not all of them. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0043] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0044] In response to existing problems, the present invention provides an intelligent management and control system for corporate financial and tax risks.

[0045] like Figure 1 As shown, an embodiment of the present invention provides an intelligent management and control system for corporate financial and tax risks, including:

[0046] Data acquisition module, used to obtain invoice data, corporate financial data and industry tax data of target enterprises;

[0047] The data fusion module is used to perform distributed cleaning of abnormal data in invoice data, corporate financial data, and industry tax data using the established data mapping rule library, and to fuse the cleaned data in real time to obtain a dynamic data pool;

[0048] The risk diagnosis module is used to extract compliance rules from corporate financial and tax laws and regulations. Based on the compliance rules, the module uses a trained random forest regression model to predict the dynamic data pool, obtain a reasonable tax burden threshold, and then determine the risk diagnosis result based on the reasonable tax burden threshold.

[0049] The early warning module is used to analyze the risk diagnosis results, obtain the weight of each risk indicator, issue early warnings on corporate financial and tax risks based on the weight of each risk indicator, and generate risk reports for target enterprises.

[0050] Specifically, we obtain the target company's invoice data, corporate financial data, and industry tax data, including:

[0051] Obtain the target company's invoice data through the invoice full-factor information query interface of the tax open platform;

[0052] Extract the target company's financial data from the financial system through a network interactive interface. The financial data includes contract numbers, logistics order numbers, bank statements, and tax rates.

[0053] Obtain industry tax burden data from the industry tax burden rate database through a network interactive interface.

[0054] In an embodiment of the present invention, the data acquisition module configures different interfaces for different data sources to obtain corresponding data, and the tax open platform mentioned above is the open platform of the State Administration of Taxation, and the invoice data obtained is the structured data of value-added tax invoices; the financial system refers to the website and application used by the target enterprise for financial management, such as commonly used financial software such as UFIDA / Kingdee; the industry tax burden rate database can be the quarterly report of the State Administration of Taxation, and the industry tax burden data is the industry tax burden benchmark value.

[0055] Specifically, based on the established data mapping rule base, we conduct distributed cleaning of abnormal data in invoice data, corporate financial data, and industry tax data, including:

[0056] Establish a standardized field mapping table based on source system fields, target system fields, and conversion rules;

[0057] Build a data mapping rule base based on the established standardized field mapping table;

[0058] Utilize the data mapping rule library to detect abnormal data in invoice data, corporate financial data, and industry tax data, and use the big data processing engine to build a cleaning pipeline to filter out all abnormal data.

[0059] In an embodiment of the present invention, the specific process of establishing a standardized field mapping table based on source system fields, target system fields, and conversion rules is as follows:

[0060] Identify and categorize data fields:

[0061] First, extract source system fields. For example, for structured VAT invoice data, extract fields such as the invoice code, purchaser taxpayer identification number, product name, amount, and tax rate. For financial data, extract fields such as the contract number, product name, logistics order number, bank serial number, and cost account. For industry tax burden data, extract the industry tax burden rate.

[0062] Then define the target system fields as shown in Table 1 below:

[0063] Table 1 Target system field definition

[0064] Target Field definition tax_code Tax classification and coding (such as the Goods and Services Tax Classification and Coding) contract_id Unique contract identifier (ERP contract number + year)

[0065] Create mapping rules:

[0066] First, a one-to-one matching method is used to formulate direct mapping rules. This is applicable to scenarios where the source and target fields have the same semantics and only require format conversion. The direct mapping rules are shown in Table 2 below:

[0067] Table 2 Direct mapping rules

[0068] Source system fields Target system fields Conversion rules sales_amount total_amount Keep 2 decimal places and convert the unit to "10,000 yuan" invoice_date invoice_date The date format is standardized to "YYYY-MM-DD"

[0069] Then use many-to-one or logical association to formulate complex mapping rules:

[0070] In multi-field merging scenarios, such as SQL mapping logic, tax codes are automatically assigned based on product names. If the product name contains "electronic components," the tax code is set to 1090250001. If it contains "consulting services," the tax code is set to 3040102002. All other cases are marked as "other." The complex mapping rules are shown in Table 3 below:

[0071] Table 3

[0072]

[0073] For cross-system association scenarios, the complex mapping rules are shown in Table 4 below:

[0074] Table 4

[0075]

[0076] Finally, the rule base is dynamically updated, mainly including rule version management: recording the time, modifier and effective date of each rule modification; automatic adaptation to new policies: when a new tax classification code is added, the system automatically scans the unmatched product names and generates a list to be marked for manual processing.

[0077] Develop automated mapping tools:

[0078] First, the intelligent matching engine, as follows:

[0079] NLP fuzzy matching: performs semantic analysis on unstructured fields (such as product names). A product name fuzzy matching example (simplified to keyword matching) uses keywords to determine the product tax code: a name containing "mobile phone" returns the communication equipment tax code 1090130001; a name containing "steel" returns the metal product tax code 1080302002; if no match is found, "UNKNOWN" is returned.

[0080] Then configure the rule engine as follows:

[0081] Use the Drools rule engine to implement complex logic, such as supplier risk scoring rules. Use rules to identify high-risk suppliers. If a supplier's credit score is less than 60 and there are more than three overdue orders, its risk level is marked as "red" (high risk).

[0082] Exception handling and manual intervention:

[0083] First, develop an anomaly detection mechanism, specifically:

[0084] Conflict detection: triggers an alarm when the same product name matches multiple tax codes;

[0085] Missing value processing: When a key field (such as taxpayer identification number) is empty, it is marked as "data to be completed";

[0086] Then generate the manual annotation process, specifically:

[0087] The system generates a "List of Unmapped Fields" and sends it to financial personnel for processing; the annotation results are automatically synchronized to the rule library, and subsequent similar data is automatically matched.

[0088] Mapping table version management and verification:

[0089] First, perform version control: each rule update generates a new version (such as v1.2.3), and supports historical version backtracking;

[0090] Then generate the test verification process, specifically:

[0091] Unit testing: Verify the accuracy of single field mapping (e.g., the encoding matching accuracy of 1,000 product names must be ≥ 98%);

[0092] Integration testing: simulates the entire data flow and checks whether cross-system field associations are complete.

[0093] The standardized field mapping table established in the embodiment of the present invention is shown in Table 5 below:

[0094] Table 5 Example of standardized field mapping table

[0095]

[0096] The specific implementation process of the embodiment of the present invention using the data mapping rule library to detect abnormal data in invoice data, corporate financial data, and industry tax data is as follows:

[0097] 1. Data preprocessing and standardization

[0098] First, standardize the data format. For example, for invoice data, convert fields such as the invoice code and purchaser tax number to a standard format according to the Tax Coding Specifications. For example, the invoice date "2023 / 10 / 01" is standardized to "2023-10-01," and the amount "￥150,000.00" is converted to a floating-point number "150000.00." For financial data, standardize contract amounts to "10,000 yuan," and reclassify cost items according to the Enterprise Accounting Standards. For industry data, classify tax rates by industry code (such as the National Economic Industry Classification GB / T4754) and convert them to percentage format.

[0099] Then perform field validity check:

[0100] The data format verified by regular expression is shown in Table 6:

[0101] Table 6

[0102] Field Type Regular expression rules Exception Example Invoice Code ^\d{10}$ "12345678A" (including letters) Taxpayer Identification Number ^[A-Z0-9]{15}$ "9111000" (insufficient length) tax rate ^[0-9]{1,2}(\.[0-9]{1,2})？$ "105.5%" (more than 100%)

[0103] 2. Anomaly detection based on rule base

[0104] First, perform invoice data anomaly detection

[0105] Rule 1: Four-in-one verification. The detection logic is that the invoice must match the contract, logistics, and capital flow (four-in-one). An example is:

[0106] Check whether the invoice is missing necessary related information:

[0107] There is no corresponding contract (the contract amount must be consistent with the invoice amount) or no matching logistics order number. If any of these conditions are met, it is considered "four flows missing" (such as contract flow and logistics flow missing);

[0108] Abnormal types: The invoice amount does not match the contract amount (e.g., the contract amount is 1 million yuan, and the invoice amount is 1.2 million yuan), or the logistics order number is not associated with the actual delivery record;

[0109] Rule 2: Buyer-seller risk association. The detection logic is that if the buyer or seller is in the blacklist database (such as Tianyancha's high-risk enterprise list), it is marked as an exception. Example:

[0110] Supplier X was once penalized for tax violations, and the invoices it issued automatically triggered a "high-risk supplier" warning.

[0111] Then perform anomaly detection on enterprise financial data

[0112] Rule 1: Cost item rationality. Detection logic: Based on the industry cost structure average (e.g., direct materials account for 50% to 70% in manufacturing), detect items with excessive deviations.

[0113] The cost ratio anomaly detection logic is to determine whether the material cost is reasonable based on industry standards: if the material cost ratio exceeds the industry upper limit, mark it as "abnormally high material cost"; if it is below the industry lower limit, mark it as "abnormally low material cost";

[0114] Rule 2: Matching of funds flow and invoice. The detection logic is that the bank payee must be the same as the invoice payee, and the amount error must be ≤1%. Example:

[0115] The invoice was issued to "Company A", but the payee was "Company B", which was marked as "inconsistent fund flow".

[0116] Finally, perform anomaly detection on industry tax burden data

[0117] Rule 1: Tax rate deviation warning. The detection logic is to dynamically calculate the deviation between the enterprise tax rate and the industry average, triggering a three-level warning:

[0118]

[0119] Green light: deviation rate ≤ 10%, yellow light: 10% < deviation rate ≤ 30%, red light: deviation rate > 30%;

[0120] Rule 2: Adapt to policy changes. The detection logic is to automatically update the industry average calculation rules when a new policy is released (such as adjustment of the retained tax refund ratio). Example:

[0121] In 2023, the tax rebate rate for the manufacturing industry will be increased from 60% to 100%, and the system will recalculate the reasonable tax burden range for the industry.

[0122] Next, dynamic algorithms are used to assist in abnormal judgment.

[0123] First, dynamically adjust the threshold through machine learning. The process is as follows:

[0124] Input features: historical corporate tax rate, industry volatility coefficient, and policy impact label;

[0125] Model output: Reasonable range of current tax burden rate (e.g. [6.5%, 7.8%]);

[0126] Example judgment:

[0127] If the corporate tax rate is 8.0%, but the model dynamic range is [7.0%, 8.5%], it is considered normal. If the tax rate is 6.0%, it exceeds the lower limit of the dynamic range and triggers a yellow light warning.

[0128] Then, NLP policy parsing is used to enhance the rule base. The implementation process is as follows:

[0129] Crawl policy texts from the State Administration of Taxation;

[0130] Use the BERT model to extract keywords (such as "tax rate reduction" and "audit focus");

[0131] Automatically generate new rules (for example, "the additional deduction ratio for R&D expenses is increased to 100%", which requires updating the additional deduction verification logic).

[0132] Set up exception handling and feedback mechanisms

[0133] First, abnormalities are graded and marked. The risk levels are shown in Table 7 below:

[0134] Table 7

[0135]

[0136]

[0137] Then automate the process

[0138] Scenario 1: Invoices are voided and reopened. When an "invalid invoice code" is detected, the invoicing clerk is automatically notified and the system's invoicing function is locked until the problem is resolved.

[0139] Scenario 2: Tax burden deviation adjustment. When a yellow light warning is triggered, the system pushes a "Tax Burden Optimization Suggestion Plan", including adjusting the input certification time or cost allocation ratio;

[0140] Finally, manual review and rule optimization are carried out, including:

[0141] False positive feedback: Finance personnel can mark "false positive" anomalies, and the system will automatically record and optimize model parameters;

[0142] Rule iteration: Monthly statistics on high-frequency exception types are collected and the rule base is updated (such as the addition of "cross-border e-commerce tax code mapping rules").

[0143] Specifically, the data mapping rule library is used to detect abnormal data in invoice data, corporate financial data, and industry tax data. For example, when the amount on the invoice data is detected to be a negative value, the invoice data is abnormal data, and the fields that cannot be mapped are also abnormal data. For the fields that cannot be mapped, the manual labeling process is initiated, and the data mapping rule library is automatically updated after labeling.

[0144] Specifically, we use a big data processing engine to build a cleaning pipeline to filter out all abnormal data, including:

[0145] Define a verification function to check whether the invoice code is legal, and define the invoice code as a string of 10 pure digits;

[0146] Input abnormal data into the verification function. If it returns True, it means it is legal. If it returns False, it means it is illegal, and the illegal data will be filtered out.

[0147] The embodiment of the present invention requires cleaning all accessed data. The specific process includes:

[0148] Taking manufacturing enterprises as an example, the specific cleaning steps are:

[0149] 1. Invoice data cleaning

[0150] Objective: To ensure that invoice data is legal, complete, and logically sound;

[0151] First, standardize the format: for example, unify the date format (e.g., "2023 / 10 / 01" becomes "2023-10-01"), and convert the amount unit (e.g., "￥150,000.00" becomes the floating point number "150000.00").

[0152] Then, perform validity checks: for example, use regular expressions to verify whether the invoice code is 10 digits (such as ^\d{10}$) and check whether the buyer / seller tax number complies with the tax registration number rules (15 or 18 digits / letters combination).

[0153] Next, we perform a logical check, including: verifying the invoice by calculating the total price and tax of the same invoice = amount × (1 + tax rate). For example:

[0154] To check whether the amount is calculated correctly, check whether the "Total Price and Tax" is equal to the result of amount × (1 + tax rate) (keep two decimal places). If not, mark it as "abnormal amount". Invoice cancellation status: exclude invalidated or red-reversed invoices.

[0155] 2. Enterprise financial data cleaning

[0156] Objective: To ensure that financial data complies with accounting standards and matches business reality;

[0157] First, perform account mapping, mapping the enterprise-defined account (such as "Material Cost-Miscellaneous") to the standard account (such as "Production Cost-Direct Materials").

[0158] An example mapping table is shown in Table 8 below:

[0159] Table 8

[0160] Enterprise Subjects Standard subjects Travel expenses Administrative expenses-travel expenses R&D consumables R&D expenses-material costs

[0161] Then perform numerical logic verification to balance the balance sheet: check "Assets = Liabilities + Owners' Equity" and the rationality of cost allocation: the cost of a single product should not exceed ±20% of the theoretical value of the BOM (Bill of Materials);

[0162] Finally, we conduct relevance verification to detect contracts that are not matched to bank statements, match contracts with fund statements, and identify contracts for which payments have not been received. The contract amount and payer name must be consistent with the bank statement records. If there is no matching bank statement, it is considered a "risky contract for non-repayment";

[0163] 3. Industry data cleaning

[0164] Goal: Eliminate outdated or distorted industry reference values;

[0165] First, a timeliness filter is performed, retaining only the industry tax rate data for the past three years (e.g., 2021-2023), and excluding historical data before major policy adjustments (e.g., data before the new VAT refund regulations in 2023);

[0166] Then, outliers are removed and the boxplot method is used to identify outliers:

[0167] Reasonable range = [Q1-1.5IQR, Q3+1.5IQR]

[0168] Among them, Q1 is the 25% quantile, Q3 is the 75% quantile, IQR = Q3-Q1. For example: the average tax burden rate of a certain industry is 7%. If a company reports a tax burden rate of 20%, it will be directly marked as an anomaly.

[0169] Specifically, the dynamic data pool consists of two parts. The first part is the data received by the message queue, which is used for risk prediction. The second part is the historical tax burden rate of the enterprise used to train the random forest regression model.

[0170] It should be noted that a joint index is established for high-frequency query fields in the dynamic data pool, reducing query time from minutes to milliseconds.

[0171] In an embodiment of the present invention, compliance rules are extracted from corporate finance and taxation legal provisions, which include the Value Added Tax Law and the Corporate Income Tax Law. The extracted compliance rules are key restriction conditions, such as a matching degree of purchase and sales item names ≥ 70%.

[0172] Specifically, the trained random forest regression model is used to make predictions on the dynamic data pool, including:

[0173] Analyze corporate finance and taxation laws and regulations to obtain policy impact coefficients;

[0174] The policy impact coefficient, the company's historical tax rate, and the industry benchmark tax rate are used to train the random forest regression model to obtain a trained random forest regression model;

[0175] The dynamic data pool is input into the trained random forest regression model to predict the tax burden threshold.

[0176] In an embodiment of the present invention, the corporate finance and taxation legal provisions are parsed through a natural language processing model to obtain a policy impact coefficient. The natural language processing model can be composed of a BERT model and a word segmenter. The corporate finance and taxation legal provisions are input into the natural language processing model to analyze the policy type.

[0177] In an embodiment of the present invention, the historical tax burden rate of an enterprise can be the tax burden rate of the enterprise in the past 12 months, such as 6.5%, 7.0%, ..., 7.8%. The industry benchmark tax burden rate is 7.5%. The policy impact coefficient is initially 1.0 and is currently 0.9. These data are input into the random forest regression model for prediction, and a reasonable tax burden threshold of 7.3% is obtained. The random forest regression model learns the relationship between the enterprise tax burden rate, the industry benchmark tax burden rate, the policy coefficient and the reasonable tax burden threshold.

[0178] The embodiment of the present invention determines the risk diagnosis result based on the reasonable tax burden threshold. When the tax burden rate of the target enterprise is lower than the reasonable tax burden threshold, it is determined that the target enterprise has financial and tax risks. The lower the tax burden rate of the target enterprise is, the greater the financial and tax risks of the target enterprise. The risk diagnosis result is composed of multiple risk indicators, such as invoice compliance, tax burden rationality, supplier risk, and policy matching.

[0179] Specifically, the risk diagnosis results are analyzed to obtain the weights of each risk indicator, including:

[0180] The importance of each risk indicator in the risk diagnosis results is compared through expert scoring to form a judgment matrix, which is shown in Table 9 below:

[0181] Table 9 Judgment Matrix

[0182]

[0183]

[0184] It should be noted that 1 in the above table means that both are equally important, 2 means that invoice compliance is twice as important as tax burden rationality, 3 means that invoice compliance is three times more important than supplier risk, and 5 means that invoice compliance is five times more important than policy matching.

[0185] The geometric mean of each risk indicator is calculated based on the judgment matrix. The calculation formula is as follows:

[0186]

[0187] The weight of each risk indicator is calculated using the geometric mean of all risk indicators, including:

[0188] The sum of the geometric means is calculated using the geometric means of all risk indicators. The calculation formula is:

[0189] 0.23+1.32+0.83+0.43=4.92

[0190] The weight of each risk indicator is calculated based on the sum of the geometric means, and the calculation formula is as follows:

[0191] Invoice compliance weight:

[0192] w1=2.34 / 4.92≈47.6%

[0193] Reasonable weight of tax burden:

[0194] w2=1.32 / 4.92≈26.8%

[0195] Supplier risk weight:

[0196] w3=0.83 / 4.92≈16.9%

[0197] Policy matching weight:

[0198] w4=0.43 / 4.92≈8.76%.

[0199] It is most preferred that, before issuing an early warning of the enterprise's fiscal and tax risks based on the weights of various risk indicators, the following also be included:

[0200] The weight of each risk indicator is checked for consistency based on the judgment matrix to determine whether the weight of each risk indicator is reasonably distributed.

[0201] Specifically, the weight of each risk indicator is checked for consistency based on the judgment matrix, including:

[0202] Multiply the judgment matrix by the weight of each risk indicator to obtain a new vector for each risk indicator. The calculation formula includes:

[0203] New vectors of invoice compliance:

[0204] 1×0.476+2×0.268+3×0.169+5×0.087=1.98

[0205] New vector of tax rationality:

[0206] 0.5×0.476+1×0.268+2×0.169+3×0.087=1.15

[0207] New vectors of supplier risk:

[0208] 0.33×0.476+0.5×0.268+1×0.169+2×0.087=0.63

[0209] New vector of policy matching:

[0210] 0.2×0.476+0.33×0.268+0.5×0.169+1×0.087=0.35

[0211] The maximum eigenvalue is calculated using the new vector of each risk indicator using the following formula:

[0212]

[0213] Among them, λ max represents the maximum eigenvalue;

[0214] The consistency index is calculated based on the maximum eigenvalue, and the expression is:

[0215]

[0216] Among them, CI represents consistency index;

[0217] The consistency index and the random consistency index obtained by table lookup are used to calculate the test consistency ratio, which is expressed as:

[0218]

[0219] Among them, CR represents the test consistency ratio, and RI represents the random consistency index;

[0220] The consistency test of each risk indicator weight is performed based on the test consistency ratio.

[0221] Specifically, the expression for early warning of corporate fiscal and tax risks based on the weights of various risk indicators is:

[0222]

[0223] Among them, a represents the warning level, w i represents the weight of the i-th risk indicator, r i represents the i-th risk indicator, n represents the number of risk indicators, and ∈ represents a dynamic adjustment factor, which fluctuates according to the enterprise's credit rating (e.g., a 10% risk value reduction for an AAA-rated enterprise).

[0224] Specifically, generate a risk report for the target enterprise, including:

[0225] Use Jinja2 templates to insert risk indicator numbers and specific problem descriptions, loop through and list relevant tax law provisions, and generate a risk report in HyperText Markup Language (HTML) format.

[0226] Compared with the existing technology, the present invention obtains the invoice data, enterprise financial data and industry tax burden data of the target enterprise through the data acquisition module; through the data fusion module, based on the established data mapping rule library, the abnormal data in the invoice data, enterprise financial data and industry tax burden data are distributedly cleaned, and the cleaned data are fused in real time to obtain a dynamic data pool; using the risk diagnosis module, compliance rules are extracted from the enterprise's financial and tax legal provisions, and based on the compliance rules, the dynamic data pool is predicted using the trained random forest regression model to obtain a reasonable tax burden threshold, and the risk diagnosis result is determined according to the reasonable tax burden threshold; the risk diagnosis result is analyzed by the early warning module to obtain the weight of each risk indicator, and based on the weight of each risk indicator, the enterprise's financial and tax risks are warned and a risk report for the target enterprise is generated, which solves the island problem of enterprise financial and tax data and improves the efficiency of financial and tax management.

[0227] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An intelligent management and control system for corporate finance and taxation risks, characterized by: include: Data acquisition module, used to obtain invoice data, corporate financial data and industry tax data of target enterprises; A data fusion module is used to perform distributed cleaning of abnormal data in the invoice data, the enterprise financial data, and the industry tax data using an established data mapping rule library, and to fuse the cleaned data in real time to obtain a dynamic data pool; A risk diagnosis module is used to extract compliance rules from corporate finance and tax laws and regulations, use the trained random forest regression model to predict the dynamic data pool based on the compliance rules, obtain a reasonable tax burden threshold, and determine the risk diagnosis result based on the reasonable tax burden threshold; The early warning module is used to analyze the risk diagnosis results, obtain the weight of each risk indicator, issue an early warning of the enterprise's financial and tax risks based on the weight of each risk indicator, and generate a risk report for the target enterprise.

2. The enterprise financial and tax risk intelligent management and control system according to claim 1 is characterized by: Obtain the target company's invoice data, corporate financial data, and industry tax data, including: Obtain the target company's invoice data through the invoice full-factor information query interface of the tax open platform; Extracting the target company's financial data from the financial system through a network interactive interface, including contract numbers, logistics order numbers, bank statements, and tax rates; Obtain industry tax burden data from the industry tax burden rate database through a network interactive interface.

3. The enterprise financial and tax risk intelligent management and control system according to claim 2 is characterized in that: The established data mapping rule library is used to perform distributed cleaning on abnormal data in the invoice data, the enterprise financial data, and the industry tax data, including: Establish a standardized field mapping table based on source system fields, target system fields, and conversion rules; Build a data mapping rule base based on the established standardized field mapping table; The data mapping rule library is used to detect abnormal data in the invoice data, the enterprise financial data and the industry tax data, and a big data processing engine is used to build a cleaning pipeline to filter out all abnormal data.

4. The enterprise financial and tax risk intelligent management and control system according to claim 3 is characterized in that: The trained random forest regression model is used to predict the dynamic data pool, including: Analyze corporate finance and taxation laws and regulations to obtain policy impact coefficients; The random forest regression model is trained using the policy impact coefficient, the historical corporate tax rate, and the industry benchmark tax rate to obtain a trained random forest regression model; The dynamic data pool is input into the trained random forest regression model to predict the tax burden threshold.

5. The enterprise financial and tax risk intelligent management and control system according to claim 4 is characterized in that: The risk diagnosis results are analyzed to obtain the weights of various risk indicators, including: Comparing the importance of each risk indicator in the risk diagnosis results through expert scoring to form a judgment matrix; Calculate the geometric mean of each risk indicator based on the judgment matrix; The weight of each risk indicator is calculated using the geometric mean of all risk indicators.

6. The enterprise financial and tax risk intelligent management and control system according to claim 5 is characterized in that: Before issuing early warnings for corporate fiscal and tax risks based on the weights of various risk indicators, it also includes: The weight of each risk indicator is checked for consistency based on the judgment matrix to determine whether the weight of each risk indicator is reasonably distributed.

7. The intelligent management and control system for corporate finance and taxation risks according to claim 6 is characterized in that: The weight of each risk indicator is checked for consistency based on the judgment matrix, including: Multiplying the judgment matrix by the weight of each risk indicator to obtain a new vector for each risk indicator; Calculate the maximum eigenvalue using the new vector of each risk indicator; Calculate the consistency index according to the maximum eigenvalue; Calculate the consistency index and the random consistency index obtained by table lookup to obtain the test consistency ratio; A consistency check is performed on the weight of each risk indicator based on the check consistency ratio.

8. The intelligent management and control system for corporate finance and taxation risks according to claim 6 is characterized in that: The expression for early warning of corporate fiscal and tax risks based on the weights of each risk indicator is: Among them, a represents the warning level, w i represents the weight of the i-th risk indicator, r i represents the i-th risk indicator, n represents the number of risk indicators, and ∈ represents the dynamic adjustment factor.

Images