Enterprise risk assessment method and related device

Abstract

The embodiment of the invention discloses an enterprise risk assessment method and a related device. The method comprises the following steps: firstly, collecting tax receipt associated data of a target enterprise; secondly, preprocessing the tax receipt associated data; then, extracting risk assessment features of the pre-processed tax receipt associated data; and finally, calling an evaluation model, and evaluating the risk evaluation characteristics according to the evaluation model to obtain an evaluation result of the target enterprise. In this way, the accuracy and efficiency of risk assessment are improved through the assessment model for all-dimensional risk assessment of enterprise tax and invoice businesses.

Description

Technical Field

[0001] This application relates to the field of financial risk control technology, and in particular to a method and related apparatus for enterprise risk assessment. Background Technology

[0002] With the rapid development of the economy and society, the number of micro and small enterprises has also grown rapidly. In many application scenarios of bank micro and small enterprise loan products, the phenomenon of being attacked by enterprises with false tax invoice data is common, which has a significant impact on the asset quality of banks.

[0003] On the one hand, traditional prevention methods are proving insufficient in dealing with the current complex situation of fraud. Faced with massive amounts of data, traditional methods struggle to accurately identify tax invoice fraud. On the other hand, most existing auditing systems rely solely on simple rules and surface-level data matching, lacking the ability to effectively identify and judge complex fraud logic and constantly evolving new fraud methods.

[0004] Therefore, how to efficiently and accurately assess the risk of fraudulent corporate tax invoices is a problem that financial institutions need to solve. Summary of the Invention

[0005] In view of this, this application provides a method and related apparatus for enterprise risk assessment, aiming to solve the problems of single data sources and incomplete coverage in traditional methods, and to ensure risk assessment of all dimensions of enterprise taxation and invoice business, thereby improving the accuracy and efficiency of risk assessment.

[0006] In a first aspect, embodiments of this application provide a method for enterprise risk assessment, including: Collect tax invoice-related data of the target enterprise, wherein the tax invoice-related data is the related data generated by the target enterprise in the process of tax invoice business; The tax invoice associated data is preprocessed; Extract risk assessment features from the preprocessed tax invoice-related data; An assessment model is invoked, and the risk assessment features are evaluated based on the assessment model to obtain the assessment result of the target enterprise. The assessment result is used to characterize the risk of the tax invoice-related data of the target enterprise.

[0007] In one possible embodiment, the preprocessing of the tax invoice-related data includes: processing outliers in the tax invoice-related data, the outliers including missing values ​​and duplicate values ​​in the tax invoice-related data, the tax invoice-related data including at least one of the following: tax declaration data, invoice data, and corporate income tax data; identifying the source of the tax invoice-related data, and standardizing the tax invoice-related data from different sources to unify the format and encoding of the tax invoice-related data.

[0008] In one possible embodiment, evaluating the risk assessment features according to the evaluation model includes: constructing a basic classification model, which is a deep learning model constructed based on a neural network; inputting the current tax invoice data features and historical tax invoice features of the target enterprise into the basic classification model, and evaluating abnormal modification behavior in the tax invoice data of the target enterprise, wherein the abnormal modification behavior includes unintentional single modification behavior and intentional batch modification behavior.

[0009] In one possible embodiment, the assessment model further includes a time series analysis model; the assessment of the risk assessment features based on the assessment model includes: inputting the target enterprise's historical tax invoice data within a preset time period, and the corresponding time series change features of the historical tax invoice data, into the time series analysis model to predict the range of normal tax invoice data for the target enterprise under normal operating conditions; calculating the deviation between the current tax invoice data features and the range of normal tax invoice data to identify abnormal tax invoice data features whose deviation value is greater than a preset threshold.

[0010] In one possible embodiment, the evaluation model further includes an ensemble learning model; the evaluation of the risk assessment features based on the evaluation model includes: obtaining a first risk assessment result output by the basic classification model and a second risk assessment result output by the time series analysis model; assigning weights to the first risk assessment result and the second risk assessment result through the ensemble learning model, and outputting the evaluation result of the target enterprise; the evaluation result includes a risk level and a risk identifier corresponding to the tax invoice data features.

[0011] In one possible embodiment, the method further includes: when there are missing fields in the collected tax invoice-related data, retrieving tax invoice field matching rules pre-stored in the server; determining the missing field portion of the tax invoice-related data according to the tax invoice field matching rules; supplementing the tax invoice-related data according to the missing field portion; and inputting the supplemented tax invoice-related data into the evaluation model for evaluation.

[0012] In one possible embodiment, the risk assessment features include: basic features, advanced features, text features, image features, and network features; wherein, the basic features include invoice amount, tax paid, declared income, taxpayer type, and the industry to which the target enterprise belongs; the advanced features include the year-on-year growth rate, month-on-month growth rate, trend change rate, and related transaction identifier of declared income; the text features are the semantic features of invoice item description text and invoice remarks text; the image features are invoice code, invoice number, invoice issuer information, image texture features, and layout features; the network features are the network topology features of the transaction relationship network of the target enterprise, and the network topology features are used to identify abnormal nodes and abnormal subgraph structures in the transaction relationship network.

[0013] Secondly, this application provides an enterprise risk assessment device, comprising: a collection unit, a processing unit, an extraction unit, and an assessment unit; wherein, the collection unit is specifically used to collect tax invoice-related data of a target enterprise, the tax invoice-related data being related data generated by the target enterprise during tax issuance; the processing unit is specifically used to preprocess the tax invoice-related data; the extraction unit is specifically used to extract risk assessment features from the preprocessed tax invoice-related data; the assessment unit is specifically used to invoke an assessment model, assess the risk assessment features according to the assessment model, and obtain an assessment result for the target enterprise, the assessment result being used to characterize the risk of the tax invoice-related data of the target enterprise.

[0014] Thirdly, embodiments of this application provide an electronic device, including a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for performing the steps in the first aspect of embodiments of this application.

[0015] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of embodiments of this application.

[0016] Fifthly, embodiments of this application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of embodiments of this application. The computer program product may be a software installation package.

[0017] As can be seen, the enterprise risk assessment method and related apparatus provided in this application include the following steps: First, collecting tax invoice-related data of the target enterprise, which is related data generated by the target enterprise in the process of tax invoice business; second, preprocessing the tax invoice-related data; then, extracting risk assessment features from the preprocessed tax invoice-related data; finally, calling the assessment model, evaluating the risk assessment features according to the assessment model, and obtaining the assessment result of the target enterprise. The assessment result is used to characterize the risk of the target enterprise's tax invoice-related data. Thus, by using the assessment model to conduct risk assessment across all dimensions of enterprise taxation and invoice business, the accuracy and efficiency of risk assessment are improved. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the system architecture of an evaluation system provided in an embodiment of this application; Figure 2 This is a flowchart illustrating an enterprise risk assessment method provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the specific process of an evaluation model provided in an embodiment of this application. Figure 4 This is a schematic diagram illustrating the specific process of another evaluation model provided in this application embodiment for evaluation; Figure 5 This is a schematic diagram illustrating the specific process of another evaluation model provided in this application embodiment for evaluation; Figure 6 This is a functional unit block diagram of an enterprise risk assessment device provided in an embodiment of this application; Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0021] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0022] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article indicates that the preceding and following related objects have an "or" relationship.

[0023] In this application's embodiments, "multiple" refers to two or more. In this application's embodiments, "connection" refers to various connection methods, such as direct or indirect connections, to achieve communication between devices; this application's embodiments do not impose any limitations on this.

[0024] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] The following describes the relevant content, concepts, meanings, technical issues, technical solutions, and beneficial effects involved in the embodiments of this application.

[0026] RPA (Robotic Process Automation) is a technology that uses software to simulate human behavior on a computer system to automatically execute repetitive, rule-based, and highly standardized business processes.

[0027] With rapid economic and social development, the number of micro and small enterprises (MSEs) has also grown rapidly. In various application scenarios of bank MSE loan products, attacks using fraudulent tax invoice data by enterprises are commonplace, significantly impacting the asset quality of banks. On the one hand, traditional preventative measures are proving insufficient in dealing with the current complex landscape of fraud. Faced with massive amounts of data, traditional methods struggle to accurately identify fraudulent tax invoices. On the other hand, most existing auditing systems rely on simple rules and surface-level data matching, lacking the ability to effectively identify and judge complex fraudulent logic and constantly evolving new fraudulent methods. Therefore, how to efficiently and accurately assess the risk of fraudulent corporate tax invoices is a problem that financial institutions need to solve.

[0028] This application provides a method and related apparatus for enterprise risk assessment, aiming to solve the problems of single data sources and incomplete coverage in traditional methods, and to ensure comprehensive risk assessment for enterprise tax and invoice business, thereby improving the accuracy and efficiency of risk assessment.

[0029] First, the method in the embodiments of this application is applied to a server, combined with Figure 1 The enterprise risk assessment method in the embodiments of this application is described. Figure 1 This is a schematic diagram of the system architecture of an evaluation system provided in an embodiment of this application, such as... Figure 1 The evaluation system 100 shown includes: a server 110, a business data terminal 120 that is communicatively connected to the server 110, and a client 130.

[0030] First, server 110 collects the tax invoice-related data of the target enterprise from business data terminal 120. The tax invoice-related data is the related data generated by the target enterprise in the process of tax invoice business. Then, server 110 preprocesses the tax invoice-related data. Next, server 110 extracts the risk assessment features of the preprocessed tax invoice-related data. Finally, server 110 calls the assessment model and assesses the risk assessment features according to the assessment model to obtain the assessment result of the target enterprise. The assessment result is used to characterize the risk of the tax invoice-related data of the target enterprise.

[0031] Specifically, the business data terminal 120 communicates with the tax-related system and transmits tax-related data that needs to be collected in the technical solution, including: declared income, tax type information, tax collection data, invoice data, etc. It is the data source for the server 110 to obtain official tax invoice data from the enterprise. After authorization by the enterprise, the business data terminal 120 collects the enterprise's internal data that needs to be supplemented in the technical solution through RPA tools, such as enterprise income tax details, tax data modification records, and original scanned images of invoices.

[0032] Specifically, client 130 is the interactive platform for enterprises to initiate operations and institutions to view results. Client 130 includes an enterprise user terminal: used to support pre-loan applications and data authorization, allowing enterprises to initiate operations such as selecting their region and clicking the tax bureau authentication link to complete authorization. After authorization, the business data terminal 120 is triggered to transmit the enterprise's tax invoice data to server 110. Client 130 also includes an institutional user terminal: used to support risk result viewing and existing customer scanning, for financial institutions, such as banks, to view the pre-loan risk results output by server 110 to assist institutions in deciding whether to approve enterprise applications; and to upload an existing customer list, triggering server 110 to perform an existing risk scan to obtain the risk list of existing customers.

[0033] The following is combined Figure 2 This application describes one enterprise risk assessment method in its embodiments. Figure 2 This is a flowchart illustrating an enterprise risk assessment method provided in an embodiment of this application. The method in this embodiment is applied to, for example... Figure 1 The method for server 110 shown includes the following steps: Step S210: Collect tax invoice related data of the target enterprise.

[0034] Among them, tax invoice-related data refers to the related data generated by the target company during the tax invoice business process. Specifically, tax invoice-related data refers to all data generated by the target company throughout the entire lifecycle of tax invoice business that is related to tax compliance and transaction authenticity, including: core business data such as tax declaration data, tax collection data, invoice data, and corporate income tax data; as well as process data such as the time points and extent of modifications to tax data by the company, reflecting the dynamic changes in the data.

[0035] Step S220: Preprocess the tax invoice associated data.

[0036] In one possible embodiment, preprocessing of tax invoice-related data includes: processing outliers in the tax invoice-related data, including missing and duplicate values, wherein the tax invoice-related data includes at least one of the following: tax declaration data, invoice data, and corporate income tax data; identifying the source of the tax invoice-related data; and standardizing the tax invoice-related data from different sources to unify the format and encoding of the tax invoice-related data.

[0037] The preprocessing operations include: 1. Data cleaning, which removes missing values ​​(such as filling in missing invoice item descriptions) and duplicate data in tax invoice-related data, and may also include data that is more than 10 times the industry average; 2. Data standardization: separating data from different sources and formats; 3. Data anonymization: encrypting or replacing sensitive corporate information, such as the legal representative's ID number associated with the taxpayer identification number and contact information, to protect corporate privacy and comply with data compliance requirements.

[0038] As can be seen, in this embodiment, outliers in the tax invoice-related data are processed; the source of the tax invoice-related data is identified; and the tax invoice-related data from different sources is standardized to unify the format and encoding of the tax invoice-related data. This protects corporate privacy, avoids the risk of data leakage, reduces identification errors caused by data clutter, and ensures the quality of the acquired tax invoice-related data.

[0039] Step S230: Extract the risk assessment features of the preprocessed tax invoice related data.

[0040] Specifically, in one possible embodiment, the risk assessment features include: basic features, advanced features, text features, image features, and network features; wherein, basic features include invoice amount, tax paid, declared income, taxpayer type, and the industry to which the target enterprise belongs; advanced features include year-on-year growth rate, month-on-month growth rate, trend change rate, and related transaction identifiers of declared income; text features are the semantic features of invoice item description text and invoice remarks text; image features are invoice code, invoice number, invoice issuer information, and image texture features and layout features; network features are the network topology features of the target enterprise's transaction relationship network, which are used to identify abnormal nodes and abnormal subgraph structures in the transaction relationship network.

[0041] Specifically, the basic features include: numerical features, such as invoice amount and tax payment statistics; category features, such as the sub-industry to which the enterprise belongs and taxpayer type; advanced features include: change features, such as year-on-year revenue growth rate and month-on-month growth rate; correlation features, such as false revenue indicators and abnormal related-party transaction indicators; text features include: semantic anomalies in invoice remarks and project descriptions extracted through TF-IDF / BERT, such as keywords related to false trade; image features include: anomalies in invoice scan images identified through OCR, such as blurred invoice codes and inconsistencies in format with official documents; and network features include: topological anomalies in the enterprise transaction relationship network, such as high-frequency transactions with multiple shell companies. This transforms raw tax invoice data into risk signals that can be identified by the model, overcoming the shortcomings of traditional methods that only look at surface data and cannot uncover hidden risks.

[0042] Step S240: Invoke the assessment model, evaluate the risk assessment characteristics based on the assessment model, and obtain the assessment results of the target company.

[0043] The assessment results are used to characterize the risk associated with the tax invoices of the target enterprise. Models such as XGBoost and LSTM replace manual methods to achieve automated and accurate assessments; they can quickly process massive amounts of enterprise data and capture complex spurious logic through algorithms, ultimately outputting risk results that can be directly used for decision-making, such as risk level and risk type.

[0044] Specifically, the assessment model is not a single model. The assessment model in this application includes three core sub-models: 1. A basic classification model, based on gradient boosting tree models such as XGBoost, LightGBM, and CatBoost, or a neural network model, used to compare the company's own historical data with industry data to distinguish between unintentional errors and intentional tampering; 2. A time series analysis model, based on models such as LSTM, GRU, and Transformer, used to analyze the long-term tax invoice data trends of the company; 3. An ensemble learning model, based on model fusion frameworks such as Stacking and Blending, which integrates the results of the first two types of models to improve the reliability of the assessment and output the final risk result, such as: "high risk, and hits the income tampering feature".

[0045] For example, the process begins with the enterprise initiating an application and selecting its region. The business data terminal then calls the authentication link based on the selected region. After the enterprise completes the online authentication, the business data terminal obtains the enterprise's tax invoice-related data. Next, data cleaning is performed on the obtained tax invoice-related data. Then, for less frequently used fields pushed during the bank-tax interaction process, such as the tax payment status at the tax bureau reflecting the enterprise's actual payment results, a new cleaning standard expansion table is added. For example, the time points and magnitude of changes in the enterprise's tax data modification process reflect the frequency of changes and the degree of data value variation. Following this, indicator processing is performed. After data cleaning, the data is processed into the required indicators according to model rules. These model rules typically include dozens of single spurious indicators, which are then combined to form model variables. These variables are then used in conjunction with sample data and algorithms such as logistic regression and tree models for training and validation. The specific model rules are shown in Table 1 below. Table 1 includes examples such as "Risk number SW**1, risk characteristic is tax risk characteristic, characteristic description is risk characteristic description 1, and investigation direction is risk investigation direction 1," as well as "Risk number FP**1, risk characteristic is invoice risk characteristic, characteristic description is risk characteristic description 2, and investigation direction is risk investigation direction 2." Table 1

[0046] After the indicators are processed, the model rules are used to make judgments, and the final output results can be such as a rejection list or a strong warning list.

[0047] In this embodiment, firstly, tax invoice-related data of the target enterprise is collected. This data refers to the associated data generated by the target enterprise during its tax invoice business process. Secondly, the tax invoice-related data is preprocessed. Then, risk assessment features are extracted from the preprocessed tax invoice-related data. Finally, an assessment model is invoked to evaluate the risk assessment features, yielding the assessment result for the target enterprise. This assessment result characterizes the risk of the target enterprise's tax invoice-related data. Thus, by using the assessment model to conduct a comprehensive risk assessment of the enterprise's tax and invoice business, the accuracy and efficiency of risk assessment are improved.

[0048] In other possible embodiments, after the above steps are completed, to ensure a comprehensive assessment of the risk of fraudulent activity by enterprises, the process further includes: generating a list of enterprises to be rejected or given a strong warning based on the existing list after cleaning and processing, and conducting a comprehensive review of the existing list. The existing list refers to the list of enterprise names that have already been granted credit or approved by financial institutions, primarily from banks' existing loan clients.

[0049] Specifically, in one possible embodiment, please refer to Figure 3 , Figure 3 This is a schematic diagram illustrating the specific process of an evaluation model provided in an embodiment of this application, such as... Figure 3 As shown, the assessment of risk assessment characteristics based on the assessment model includes the following steps: S301, Construct the basic classification model.

[0050] The basic classification model is a deep learning model built based on neural networks. Specifically, the basic classification model is a risk classification tool built by the server based on neural network algorithms, such as fully connected neural networks and convolutional neural networks; its core function is to classify the nature of abnormal modification behavior of corporate tax invoice data through the logic of data feature input, algorithm calculation, and classification output.

[0051] S302, input the current tax invoice data features and historical tax invoice features from the risk assessment characteristics of the target enterprise into the basic classification model to assess the abnormal modification behavior in the tax invoice data of the target enterprise.

[0052] Among them, historical tax invoice characteristics refer to the tax invoice risk assessment characteristics of the target enterprise over a continuous period of time before the assessment date. These are usually continuous data from 6 to 12 months prior to the assessment date, including invoice amounts, revenue growth rates, and modification records for the same period in history, which are used for comparison with current tax invoice data characteristics.

[0053] Abnormal modification behavior includes unintentional single modification behavior and intentional batch modification behavior. Specifically, abnormal modification behavior refers to modifications in values, fields, and logical relationships that occur when the current tax invoice data of a target enterprise differs from the normal range reflected in historical tax invoice characteristics. These modifications exceed the industry's normal fluctuation range and are further divided into two sub-behaviors: 1. Unintentional single modification behavior refers to individual modifications made by the enterprise due to operational errors, data entry deviations, or other non-fraudulent purposes. Characteristics include: modification of a single field, small modification magnitude, and no frequent modification records in the past. For example, mistakenly entering "1.02 million yuan" instead of "1 million yuan income" during the declaration, and only this one field is modified; this is a normal error that does not require interception. 2. Intentional batch modification behavior refers to batch tampering of multiple tax invoice fields by the enterprise to conceal operational anomalies, gain trust, or other fraudulent purposes. Characteristics include: simultaneous modification of multiple fields, large modification magnitude, and multiple modification records in the past. For example, changing "0 income" to "5 million yuan income" while simultaneously modifying the corresponding tax amount and invoice data; this is a "fraudulent behavior that requires key interception."

[0054] As can be seen, in this embodiment, a basic classification model is constructed, and the current tax invoice data features and historical tax invoice features from the risk assessment characteristics of the target enterprise are input into the basic classification model to assess abnormal modification behavior in the tax invoice data of the target enterprise. In this way, complex false patterns of multi-indicator linkage anomalies can be identified, reducing the false negative rate, while ensuring that the normal business pass rate is not affected, thereby improving the accuracy and efficiency of enterprise risk assessment.

[0055] Specifically, in one possible embodiment, please refer to Figure 4 , Figure 4 This is a schematic diagram illustrating the specific process of evaluation using another evaluation model provided in this application embodiment, such as... Figure 4 As shown, the assessment model also includes a time-series analysis model, which, in assessing risk assessment characteristics based on the assessment model, includes the following steps: S401: Input the target company's historical tax invoice data within a preset time period, along with the corresponding time-series change characteristics, into the time-series analysis model to predict the range of regular tax invoice data for the target company under normal operating conditions.

[0056] Among them, the time series analysis model is a dynamic risk prediction tool built by the server based on time series algorithms such as LSTM, GRU, and Transformer; the regular tax invoice data range refers to the reasonable fluctuation range of the tax invoice data that the target enterprise should be in during the current period, such as the current month, under the normal operating conditions predicted by the time series analysis model. For example, the regular range of the declared income of a manufacturing enterprise in the current month is predicted to be between RMB 1 million and RMB 1.2 million.

[0057] Specifically, the preset time period, combined with the operational stability of micro and small enterprises, usually requires 6-12 months to reflect the trend, and this time period is generally set at 6-12 months.

[0058] S402, calculate the deviation between the current tax invoice data characteristics and the range of regular tax invoice data, and identify abnormal tax invoice data characteristics whose deviation value is greater than a preset threshold.

[0059] The preset threshold is set based on industry characteristics and historical case data. It is the critical value for judging abnormal deviations. For example, the abnormal deviation threshold for the tax burden rate of the manufacturing industry is set at 10%, and that of the wholesale industry is set at 8%. The threshold needs to be dynamically adjusted in combination with different tax invoice data types and industry subdivisions.

[0060] Specifically, common methods for calculating deviations include: 1. Absolute deviation: Current data - upper limit of the normal range, such as 150 - 120 = 300,000 yuan; 2. Relative deviation: (Current data - Upper limit of normal range) / Upper limit of normal range × 100%, such as 30 / 120 × 100% = 25%, used to quantify the degree of deviation of the current data from the normal range.

[0061] As can be seen, in this embodiment, by inputting the target company's historical tax invoice data within a preset time period, along with the corresponding time-series change characteristics, into a time-series analysis model, the range of normal tax invoice data for the target company under normal operating conditions is predicted. Then, the deviation between the current tax invoice data characteristics and the normal tax invoice data range is calculated to identify abnormal tax invoice data characteristics where the deviation value exceeds a preset threshold. This reduces the underreporting rate of long-term dynamic false information, decreases false alarms regarding normal business fluctuations, and improves the efficiency of enterprise risk identification.

[0062] Specifically, in one possible embodiment, please refer to Figure 5 , Figure 5 This is a schematic diagram illustrating the specific process of evaluation using another evaluation model provided in this application embodiment, such as... Figure 5 As shown, the evaluation model also includes an ensemble learning model, which, in assessing risk assessment features based on the evaluation model, includes the following steps: S501, obtain the first risk assessment result output by the basic classification model and the second risk assessment result output by the time series analysis model.

[0063] The ensemble learning model is a multi-model result fusion framework built by the server based on ensemble algorithms such as Stacking and Blending. Its core function is to receive independent evaluation results from the basic classification model and the time series analysis model, and output a unified and accurate final risk assessment result through preset weight allocation rules and probability fusion logic. The first risk assessment result refers to the output of the basic classification model, which mainly includes two parts: anomaly determination and static anomaly quantification value. The second risk assessment result refers to the output of the time series analysis model, which mainly includes two parts: dynamic deviation determination and dynamic deviation anomaly quantification value.

[0064] S502 uses an ensemble learning model to assign weights to the first and second risk assessment results and outputs the assessment results for the target company.

[0065] The assessment results include risk levels and corresponding risk indicators for tax invoice data characteristics. Specifically, risk levels are categorized as high, medium, and low risk. The classification is based on a weighted comprehensive risk score; for example, a comprehensive score ≥80 indicates high risk requiring immediate investigation, 50-79 indicates medium risk requiring regular monitoring, and <50 indicates low risk allowing normal processing. The risk indicators for tax invoice data characteristics correspond to specific anomaly types identified by two basic models. For example, if the first risk assessment determines intentional bulk modification, the risk indicator is: intentional bulk tampering of income and tax amount fields; if the second risk assessment determines a continuous deviation from the trend, the risk indicator is: operating income exceeding the normal trend range for six consecutive months; if both results are abnormal, the corresponding indicators are displayed simultaneously.

[0066] Specifically, weight allocation refers to the predetermined weight ratio of the results in the ensemble learning model based on the recognition accuracy of the two basic models in different scenarios. For example, in a single batch tampering scenario, if the accuracy of the basic classification model is 92% and the accuracy of the time series analysis model is 75%, then the first risk assessment result is assigned 60% weight and the second risk assessment result is assigned 40% weight. In other scenarios, if the accuracy of the time series model is 90% and the accuracy of the basic model is 78%, then the weight is adjusted to 65% for the time series result and 35% for the basic result, in order to allow the more reliable model results to play a greater role in the corresponding scenarios.

[0067] Among them, the basic classification model excels at static comparison identification, while the time-series analysis model excels at dynamic trend identification. Using either model alone will inevitably miss some false alarms or misclassify normal fluctuations due to blind spots in their capabilities. By integrating the results of both models through an ensemble learning model, full coverage of both static and dynamic anomalies can be achieved, avoiding the limitations of a single model. The ensemble learning model transforms the two types of results into a unified risk score and level through weighted fusion, allowing for direct sorting by risk level, such as prioritizing high-risk enterprises.

[0068] As can be seen, in this embodiment, the first risk assessment result output by the basic classification model and the second risk assessment result output by the time series analysis model are obtained. An ensemble learning model is then used to assign weights to the first and second risk assessment results, outputting the assessment result for the target enterprise. In this way, by combining static and dynamic anomaly signals, the risk assessment is comprehensively integrated, further reducing the false negative rate and improving the accuracy and comprehensiveness of enterprise risk assessment. Simultaneously, even in scenarios with fluctuating data quality, the identification accuracy is guaranteed, avoiding overall assessment inaccuracies due to the failure of a single model, thus improving robustness.

[0069] In one possible embodiment, the method further includes: when there are missing fields in the collected tax invoice-related data, retrieving the tax invoice field matching rules pre-stored in the server; determining the missing fields in the tax invoice-related data according to the tax invoice field matching rules; supplementing the tax invoice-related data according to the missing fields; and inputting the supplemented tax invoice-related data into the evaluation model for evaluation.

[0070] The tax invoice field matching rules refer to a set of rules pre-stored on the server, designed to address differences in fields across multiple tax bureaus, tax type audit relationships, and data deficiencies in bank-tax data. The core rules include: 1. Field mapping rules: the correspondence between synonymous fields across different tax bureaus; 2. Tax type audit supplementary rules: deriving missing fields using the logical relationships between different tax type data, such as deriving the missing taxable income field from VAT payable and tax rate; 3. Alternative data source rules: when bank-tax data fields are missing, data from the Shangyitong channel is used to complete them, such as using tax collection data collected by Shangyitong to replace the missing tax payment status from the bank-tax channel. Specifically, the missing fields refer to key fields in the tax invoice association data collected by the server that are not obtained and have a significant impact on risk assessment. These include: core business fields, such as declared income, tax paid, invoice code, and tax payment status; procedural fields, such as the time point and magnitude of changes in tax data modified by the enterprise; and fields necessary for feature extraction.

[0071] As can be seen, in this embodiment, when the collected tax invoice-related data has missing fields, the tax invoice field matching rules pre-stored on the server are retrieved; based on the tax invoice field matching rules, the missing fields in the tax invoice-related data are determined; based on the missing fields, the tax invoice-related data is supplemented; and the supplemented tax invoice-related data is input into the evaluation model for evaluation. This improves the evaluation coverage, avoids missed evaluations, ensures the accuracy of the supplemented data, and provides a reliable foundation for ensuring the effectiveness of subsequent feature extraction and model evaluation.

[0072] This application embodiment can divide the electronic device into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0073] and Figure 2 The implementation is consistent with the previous one; please refer to [link / reference]. Figure 6 , Figure 6 This is a functional unit block diagram of an enterprise risk assessment device provided in an embodiment of this application. The enterprise risk assessment device 600 is applied to, for example... Figure 1 The server 110 and enterprise risk assessment device 600 shown include: a data acquisition unit 610, a processing unit 620, an extraction unit 630, and an assessment unit 640. Specifically, the data acquisition unit 610 is used to collect tax invoice-related data of the target enterprise, which is related data generated during the tax issuance process. The processing unit 620 is used to preprocess the tax invoice-related data. The extraction unit 630 is used to extract risk assessment features from the preprocessed tax invoice-related data. The assessment unit 640 is used to call an assessment model, evaluate the risk assessment features according to the assessment model, and obtain the assessment result for the target enterprise. The assessment result is used to characterize the risk of the target enterprise's tax invoice-related data.

[0074] In one possible embodiment, in terms of preprocessing tax invoice-related data, the processing unit 620 is specifically used to: process outliers in the tax invoice-related data, including missing and duplicate values ​​in the tax invoice-related data, which includes at least one of the following: tax declaration data, invoice data, and corporate income tax data; identify the source of the tax invoice-related data; and standardize the tax invoice-related data from different sources to unify the format and encoding of the tax invoice-related data.

[0075] In one possible embodiment, in terms of evaluating risk assessment features according to the evaluation model, the evaluation unit 640 is specifically used to: construct a basic classification model, which is a deep learning model constructed based on a neural network; input the current tax invoice data features and historical tax invoice features in the risk assessment features of the target enterprise into the basic classification model, and evaluate the abnormal modification behavior existing in the tax invoice data of the target enterprise, including unintentional single modification behavior and intentional batch modification behavior.

[0076] In one possible embodiment, the assessment model further includes a time series analysis model; in terms of assessing risk assessment characteristics according to the assessment model, the assessment unit 640 is specifically used to: input the target enterprise's historical tax invoice data within a preset time period, and the time series change characteristics corresponding to the historical tax invoice data, into the time series analysis model to predict the range of regular tax invoice data of the target enterprise under normal operating conditions; calculate the deviation between the current tax invoice data characteristics and the range of regular tax invoice data, and identify abnormal tax invoice data characteristics whose deviation value is greater than a preset threshold.

[0077] In one possible embodiment, the evaluation model further includes an ensemble learning model; in terms of evaluating risk assessment features according to the evaluation model, the evaluation unit 640 is specifically used to: obtain a first risk assessment result output by the basic classification model and a second risk assessment result output by the time series analysis model; assign weights to the first risk assessment result and the second risk assessment result through the ensemble learning model, and output the evaluation result of the target enterprise; the evaluation result includes a risk level and a risk identifier corresponding to the tax invoice data features.

[0078] In one possible embodiment, the enterprise risk assessment device 600 is further configured to: when there are missing fields in the collected tax invoice-related data, retrieve the tax invoice field matching rules pre-stored in the server; determine the missing fields in the tax invoice-related data according to the tax invoice field matching rules; supplement the tax invoice-related data according to the missing fields; and input the supplemented tax invoice-related data into the assessment model for assessment.

[0079] In one possible embodiment, the risk assessment features include: basic features, advanced features, text features, image features, and network features; wherein, basic features include invoice amount, tax paid, declared income, taxpayer type, and the target company's industry; advanced features include year-on-year growth rate, month-on-month growth rate, trend change rate, and related transaction identifiers of declared income; text features are the semantic features of invoice item description text and invoice remarks text; image features are invoice code, invoice number, invoice issuer information, image texture features, and layout features; network features are the network topology features of the target company's transaction relationship network, which are used to identify abnormal nodes and abnormal subgraph structures in the transaction relationship network.

[0080] It is understood that since the method embodiments and the device embodiments are different presentations of the same technical concept, the content of the method embodiment section in this application should be adapted to the device embodiment section, and will not be repeated here.

[0081] Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application. For example... Figure 7As shown, electronic device 700 may include one or more components: a processor 701 and a memory 702 coupled to the processor 701, wherein the memory 702 may store one or more computer programs, which may be configured to implement the methods described in the examples above when executed by one or more processors 701. Electronic device 700 may be as follows: Figure 1 The server shown is 110.

[0082] Processor 701 may include one or more processing cores. Processor 701 connects to various parts within the electronic device 700 using various interfaces and lines, and performs various functions and processes data of the electronic device 700 by running or executing instructions, programs, code sets, or instruction sets stored in memory 702, and by calling data stored in memory 702. Optionally, processor 701 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). Processor 701 may integrate one or more of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. It is understood that the aforementioned modem may also not be integrated into processor 701, but may be implemented separately through a communication chip.

[0083] The memory 702 may include random access memory (RAM) or read-only memory (ROM). The memory 702 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 702 may include a program storage area and a data storage area. The program storage area may store instructions for implementing an operating system, instructions for implementing at least one function (such as touch functionality, sound playback functionality, image playback functionality, etc.), and instructions for implementing the various method examples described above. The data storage area may also store data created during the use of the electronic device 700.

[0084] It is understood that the electronic device 700 may include more or fewer structural elements than those shown in the above block diagram, such as a power module, physical buttons, WiFi (Wireless Fidelity) module, speaker, Bluetooth module, sensor, etc., without limitation.

[0085] This application also provides a computer storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements some or all of the steps of any of the methods described in the above method embodiments.

[0086] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments.

[0087] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0088] In the several embodiments provided in this application, it should be understood that the disclosed methods, apparatuses, and systems can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for example, the division of units is merely a logical functional division, and there may be other division methods in actual implementation; for example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0089] 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.

[0090] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can be physically comprised separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware or in the form of hardware plus software functional units.

[0091] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute partial steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes: a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, volatile memory, or non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM), etc., which are various media capable of storing program code.

[0092] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can easily conceive of variations or substitutions without departing from the spirit and scope of the present invention, and various modifications and alterations can be made, including combinations of the different functions and implementation steps described above, as well as software and hardware implementation methods, all of which are within the protection scope of the present invention.

Claims

1. A method of enterprise risk assessment, characterized by, The method comprises the following steps: Collecting tax invoice correlation data of a target enterprise, wherein the tax invoice correlation data is generated in a tax invoice business process of the target enterprise; Preprocessing the tax invoice correlation data; Extracting risk assessment features of the preprocessed tax invoice correlation data; Calling an evaluation model to evaluate the risk assessment features according to the evaluation model, and obtaining an evaluation result of the target enterprise, wherein the evaluation result is used to represent the risk of the tax invoice correlation data of the target enterprise.

2. The method of claim 1, wherein, The preprocessing of the tax invoice correlation data comprises the following steps: Processing outliers of the tax invoice correlation data, wherein the outliers include missing values and repeated values of the tax invoice correlation data, and the tax invoice correlation data includes at least one of the following: tax declaration data, invoice data, and enterprise income tax data; Identifying the sources of the tax invoice correlation data, and performing standardized processing on the tax invoice correlation data of different sources to unify the formats and codes of the tax invoice correlation data.

3. The method of claim 2, wherein, The evaluation of the risk assessment features according to the evaluation model comprises the following steps: Constructing a basic classification model, wherein the basic classification model is a deep learning model constructed according to a neural network; Inputting current tax invoice data features and historical tax invoice features in the risk assessment features of the target enterprise into the basic classification model to evaluate abnormal modification behaviors existing in the tax invoice data of the target enterprise, wherein the abnormal modification behaviors include non-intentional single modification behaviors and intentional batch modification behaviors.

4. The method of claim 3, wherein, The evaluation model further comprises a time series analysis model; the evaluation of the risk assessment features according to the evaluation model comprises the following steps: Inputting historical tax invoice data of the target enterprise in a preset time period and time series change features corresponding to the historical tax invoice data into the time series analysis model to predict a normal tax invoice data range of the target enterprise under a normal operating state; Performing deviation calculation on the current tax invoice data features and the normal tax invoice data range to identify abnormal tax invoice data features with a deviation value amplitude greater than a preset threshold.

5. The method of claim 4, wherein, The evaluation model further comprises an ensemble learning model; the evaluation of the risk assessment features according to the evaluation model comprises the following steps: Obtaining a first risk evaluation result output by the basic classification model and a second risk evaluation result output by the time series analysis model; Performing weight distribution on the first risk evaluation result and the second risk evaluation result through the ensemble learning model to output the evaluation result of the target enterprise, wherein the evaluation result includes a risk level and a risk identifier of corresponding tax invoice data features.

6. The method according to any one of claims 1 to 5, characterized in that, The method further comprises the following steps: When the collected tax invoice correlation data has a field missing, calling a tax invoice field matching rule pre-stored in a server; Determining a field missing part of the tax invoice correlation data according to the tax invoice field matching rule; Supplementing the tax invoice correlation data according to the field missing part; Inputting the supplemented tax invoice correlation data into the evaluation model to perform an evaluation operation.

7. The method according to any one of claims 1 to 6, characterized in that, The risk assessment features include: basic features, advanced features, text features, image features and network features; wherein the basic features include invoice amount, tax amount, declared income, taxpayer type, industry of the target enterprise; the advanced features include year-on-year growth rate, month-on-month growth rate, trend change rate, associated transaction identification of the declared income; the text features are semantic features of invoice item description text and invoice note information text; the image features are invoice code, invoice number, invoice issuer information and image texture features, layout features; the network features are network topology features of the transaction relationship network of the target enterprise, and the network topology features are used to identify abnormal nodes and abnormal subgraph structures in the transaction relationship network.

8. An enterprise risk assessment apparatus characterized by comprising: Comprise: a collection unit, a processing unit, an extraction unit, and an evaluation unit; wherein the collection unit is specifically configured to collect tax ticket association data of a target enterprise, the tax ticket association data being association data generated by the target enterprise in a tax ticket business process; the processing unit is specifically configured to preprocess the tax ticket association data; the extraction unit is specifically configured to extract risk assessment features of the preprocessed tax ticket association data; the evaluation unit is specifically configured to call an evaluation model, evaluate the risk assessment features according to the evaluation model, and obtain an evaluation result of the target enterprise, the evaluation result being used to represent the risk of the tax ticket association data of the target enterprise.

9. An electronic device, comprising: comprise a processor, a memory, the memory being used to store one or more programs and being configured to be executed by the processor, the program comprising instructions for executing steps in the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, A computer program for electronic data exchange, wherein the computer program causes a computer to execute the method of any one of claims 1-7. A computer program for electronic data exchange, wherein the computer program causes a computer to execute the method of any one of claims 1-7.

Images