Resource allocation method and system for dynamic prediction of authentication risk based on big data

Abstract

The application provides a resource allocation method and system for dynamic prediction of authentication risk based on big data, relates to the technical field of authentication data analysis, and acquires enterprise data and enterprise data packet information; credit risk feature analysis, risk prediction and model output precision analysis are performed, an enterprise risk prediction instruction is triggered, current enterprise authentication information is subjected to data prediction, credit risk feature analysis and risk prediction, current enterprise risk assessment data and model precision analysis data thereof are acquired; model precision change analysis is performed, a sample tracing instruction is triggered, current enterprise prediction data packet information is subjected to evaluation and verification analysis, data packet prediction verification results are obtained; a model updating analysis is triggered, enterprise risk assessment updating data are obtained, auditing configuration and updating are performed according to the enterprise risk assessment data, enterprise auditing frequency is obtained, and intelligent and accurate prediction and allocation of enterprise authentication are realized.

Description

Technical Field

[0001] This invention proposes a resource allocation method and system for dynamic prediction of authentication risks based on big data, which relates to the field of authentication data analysis technology, specifically to the field of dynamic prediction of authentication risks and resource allocation based on big data. Background Technology

[0002] In the certification field, enterprise compliance risk assessment is a core component of certification bodies' assessment of certification applications, compliance certification, and ongoing surveillance. Existing client risk assessment models primarily rely on historical qualification documents, monitoring records, testing reports, and past certification records provided by enterprises, as well as public credit information, industry sentiment, and publicly available enterprise data. However, these materials cannot comprehensively reflect the current operational compliance and changes in an enterprise's situation (such as violations, bankruptcy, major accidents, organizational information changes, etc.). Furthermore, some enterprises, in order to smoothly obtain certification qualifications, circumvent regulatory requirements, or maintain certification status, may adjust and optimize their application materials or, after obtaining certificates, fail to promptly report compliance status and significant information changes to the certification body, impacting the certification body's supervision and management. Therefore, certification bodies need to proactively audit enterprises. To ensure the rational use of audit resources and the effectiveness and timeliness of audit results, conducting corresponding risk assessments of enterprises is clearly essential. Based on this, this solution provides a resource allocation method for dynamic prediction of certification risks based on big data to address the aforementioned problems. Summary of the Invention

[0003] This invention provides a resource allocation method and system based on big data for dynamic prediction of authentication risks, in order to solve the above-mentioned problems: A resource allocation method based on big data for dynamic prediction of authentication risks, the method comprising: S1. Obtain the enterprise's historical enterprise authentication information and enterprise data package information through the big data platform; S2. Based on historical enterprise certification information and enterprise data package information, train and analyze the accuracy of the first enterprise certification risk prediction model to obtain current enterprise risk assessment data and accuracy analysis data of the first enterprise certification risk prediction model. Trigger enterprise risk prediction instructions based on the accuracy analysis data of the first enterprise certification risk prediction model. S3. Obtain the current enterprise prediction data packet information according to the enterprise risk prediction instruction, and train and analyze the accuracy of the second enterprise certification risk prediction model based on the current enterprise prediction data packet information to obtain the accuracy analysis data of the second enterprise certification risk prediction model. S4. Based on the accuracy analysis data of the first enterprise certification risk prediction model and the accuracy analysis data of the second enterprise certification risk prediction model, perform model accuracy change analysis, then evaluate the current enterprise prediction data package, and trigger the current enterprise prediction data package verification command. S5. Perform data verification and analysis according to the current enterprise prediction data package verification instructions, update the current enterprise prediction data package according to the obtained current enterprise prediction data package verification results, and then update the second enterprise certification risk prediction model to obtain updated enterprise risk assessment data. S6. Based on the current enterprise risk assessment data and the updated enterprise risk assessment data, classify and allocate enterprises to obtain the enterprise review frequency.

[0004] Furthermore, a resource allocation system based on big data for dynamic prediction of authentication risks, the system comprising: The data acquisition module is used to acquire the company's historical enterprise authentication information and enterprise data package information through the big data platform; The historical model analysis module is used to train and analyze the accuracy of the first enterprise certification risk prediction model based on historical enterprise certification information and enterprise data package information, obtain current enterprise risk assessment data and accuracy analysis data of the first enterprise certification risk prediction model, and trigger enterprise risk prediction instructions based on the accuracy analysis data of the first enterprise certification risk prediction model. The current model analysis module is used to obtain the current enterprise prediction data package information according to the enterprise risk prediction instruction, and to train and analyze the accuracy of the second enterprise certification risk prediction model based on the current enterprise prediction data package information, thereby obtaining the accuracy analysis data of the second enterprise certification risk prediction model. The data packet verification module is used to analyze the model accuracy change based on the accuracy analysis data of the first enterprise certification risk prediction model and the accuracy analysis data of the second enterprise certification risk prediction model, and then evaluate the current enterprise prediction data packet, thereby triggering the current enterprise prediction data packet verification instruction. The data packet update module is used to perform data verification and analysis based on the current enterprise prediction data packet verification instructions, update the current enterprise prediction data packet based on the obtained current enterprise prediction data packet verification results, and then update the second enterprise certification risk prediction model to obtain updated enterprise risk assessment data. The allocation module is used to allocate enterprises according to their risk levels based on current enterprise risk assessment data and updated enterprise risk assessment data, thereby obtaining the enterprise review frequency.

[0005] The beneficial effects of this invention are as follows: This invention realizes the full-process digitalization and dynamization of enterprise certification risk management, breaks down data silos from multiple sources, and provides comprehensive data support for risk prediction; through model accuracy analysis and sample verification, it ensures the reliability of risk assessment results; based on accurate risk assessment, it enables the rational allocation of audit resources, avoids resource waste or insufficient management, and improves the systematicness and foresight of certification body risk management. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of a resource allocation method for dynamic prediction of authentication risks based on big data. Figure 2 This is a schematic diagram of the dynamic risk prediction process. Detailed Implementation

[0007] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0008] In one embodiment of the present invention, the resource allocation method and system for dynamic prediction of authentication risks based on big data proposed in this invention includes: S1. Obtain the enterprise's historical enterprise authentication information and enterprise data package information through the big data platform; S2. Based on historical enterprise certification information and enterprise data package information, train and analyze the accuracy of the first enterprise certification risk prediction model to obtain current enterprise risk assessment data and accuracy analysis data of the first enterprise certification risk prediction model. Trigger enterprise risk prediction instructions based on the accuracy analysis data of the first enterprise certification risk prediction model. S3. Obtain the current enterprise prediction data packet information according to the enterprise risk prediction instruction, and train and analyze the accuracy of the second enterprise certification risk prediction model based on the current enterprise prediction data packet information to obtain the accuracy analysis data of the second enterprise certification risk prediction model. S4. Based on the accuracy analysis data of the first enterprise certification risk prediction model and the accuracy analysis data of the second enterprise certification risk prediction model, perform model accuracy change analysis, then evaluate the current enterprise prediction data package, and trigger the current enterprise prediction data package verification command. S5. Perform data verification and analysis according to the current enterprise prediction data package verification instructions, update the current enterprise prediction data package according to the obtained current enterprise prediction data package verification results, and then update the second enterprise certification risk prediction model to obtain updated enterprise risk assessment data. S6. Based on the current enterprise risk assessment data and the updated enterprise risk assessment data, classify and allocate enterprises to obtain the enterprise review frequency.

[0009] Combining S1 to S6, the resource allocation method and system for dynamic prediction of authentication risks based on big data proposed in this invention are specifically as follows in a preferred example: Historical enterprise certification information includes information on whether enterprise certificates have been suspended, cancelled, or revoked, as well as information on changes to certification information. The enterprise data package includes internal risk assessment data, information on changes or expiration of various administrative licenses, data on enterprises being ordered to suspend production or business for rectification by administrative regulatory departments, data on illegal and dishonest conduct, information on accidents related to quality, environment, health and safety, and information security, as well as data on changes in the enterprise's legal status, and data on changes in its historical name, address, and scope. Historical enterprise certification information is used as the input features of the model, the past certification results of enterprises are used as the judgment labels, and an appropriate algorithm is selected to train the first enterprise certification risk prediction model. Accuracy analysis was conducted on the first enterprise certification risk prediction model to obtain accuracy analysis data of the first enterprise certification risk prediction model; at the same time, the current enterprise risk assessment data of the enterprise was output to clarify the enterprise's preliminary risk level. Based on the preset model accuracy standard, if the first enterprise certification risk prediction model does not meet the requirements, the enterprise risk prediction instruction will be automatically triggered to start the model iteration and optimization process. Based on the triggered enterprise risk prediction instruction, supplementary enterprise prediction data package information is collected, including the latest annual inspection report and credit repair records related to certificates, as well as the latest regulatory notices, health and safety inspection results, and environmental monitoring data related to the enterprise. The supplementary current prediction data package information is integrated with historical data, and iterative optimization training is performed based on the framework of the first enterprise certification risk prediction model to generate the second enterprise certification risk prediction model. Accuracy analysis was conducted on the second enterprise certification risk prediction model to obtain accuracy analysis data for the second enterprise certification risk prediction model. By comparing the accuracy analysis data of the first enterprise certification risk prediction model with that of the second enterprise certification risk prediction model, the trend of accuracy change between the two model trainings is analyzed to determine the model optimization effect. Based on the results of the accuracy change analysis, the rationality and effectiveness of the current enterprise forecast data package supplemented in this update will be evaluated. Based on the evaluation results, the system automatically triggers the current enterprise's predicted data packet verification command; Based on the current enterprise forecast data packet verification instructions, a comprehensive verification and analysis is conducted on the compliance of the data packet's source, the completeness of its content, and its relevance to the enterprise's actual operations. Based on the verification results, the current enterprise forecast data package is corrected, supplemented, or improved to complete the data package update; The updated data package is used to further optimize and iterate the second enterprise certification risk prediction model, complete the model update, and output updated enterprise risk assessment data based on the updated model. By combining current enterprise risk assessment data with updated enterprise risk assessment data, the final risk level of the enterprise is determined. According to the resource allocation rules corresponding to different risk levels, the review frequency of corresponding enterprises should be determined. When determining the review frequency, factors such as the cyclical characteristics of the industry in which the enterprise operates, changes in regional regulatory intensity, and the incidence of risk events of similar enterprises should also be considered to ensure that the matching of review frequency is more dynamic and adaptable. Specifically, firstly, the cyclical characteristics of the industry in which the enterprise operates are used to characterize the magnitude of the industry's prosperity fluctuations. For example, compared to the prosperity fluctuations of the semiconductor industry, the prosperity fluctuations of the food industry are more gradual. Therefore, when the risk level of the food industry changes by the same amount, the adjustment of the audit frequency is more affected than that of the semiconductor industry. Changes in regional regulatory intensity reflect the temporal and spatial differences in the strength of policy implementation. The incidence rate of risk events of similar enterprises constitutes a horizontal reference system. Together, these form a multi-dimensional audit frequency calibration mechanism. In practice, the audit frequency adjustment coefficient is calculated in real time based on this audit frequency calibration mechanism. Then, the audit frequency adjustment coefficient is multiplied by the basic audit frequency to finally generate the initial audit frequency used for enterprise evaluation. Secondly, for certifications already obtained by enterprises, the frequency, type, and completeness of changes to the certifications since the last audit are collected to construct a dynamic stability index. This index quantifies the continued effectiveness of the certifications. The initial audit frequency is then recalibrated based on the stability index to achieve precise correction, ensuring that the final output audit frequency reflects both the current risk status of the enterprise's certifications and their evolutionary trajectory. The stability index is calculated as follows: Stability Index = 1 (Frequency of change × Type weight + Content missing rate) / 3, where the type weight is set according to the nature of the change. Generally speaking, the weight of qualification cancellation is greater than that of scope expansion, and the weight of validity renewal is less. The content missing rate is calculated by comparing the coverage of the change materials with the certification standard clauses.

[0010] The working principle and technical effects of the above technical solution are as follows: A big data platform is used to integrate multi-source enterprise data and construct a database, forming a structured enterprise data package. Combined with enterprise authentication information, credit risk characteristic analysis, risk prediction, and model accuracy assessment are conducted from both historical and current data dimensions, triggering risk prediction commands to obtain current enterprise risk assessment results and model accuracy data. Based on a comparison of historical and current model accuracy data, the trend of accuracy change is analyzed, triggering sample tracing commands to verify the current enterprise prediction data, obtaining data package prediction verification results. The model and risk assessment data are updated based on the verification results, thereby optimizing the audit configuration and generating the enterprise audit frequency, forming a closed-loop process. While existing technologies have applied big data to credit risk prediction and resource allocation, they mostly employ a linear process of single-round data training, model output, and resource allocation, without forming a complete technical chain of historical data training, current data iteration, accuracy verification, and dynamic resource allocation.

[0011] The core logic of this invention involves multi-source data integration, model training, accuracy verification, model updating, and hierarchical allocation. By linking historical and current dual-dimensional data training, adding a model accuracy comparison and verification step, and establishing a dynamic model update mechanism, it constructs an integrated technical solution for data, model, verification, updating, and allocation, which is fundamentally different from the linear process of existing technologies.

[0012] This invention addresses the technical pain points of poor adaptability and accuracy decay in static models. Existing technologies rely on historical experience to build static risk prediction models, which cannot adapt to changes in the current risk characteristics of enterprises, resulting in a decline in model accuracy over time. This invention designs a two-stage model training and accuracy comparison verification mechanism: a first enterprise certification risk prediction model is trained based on historical data and its accuracy is analyzed; then, a second enterprise certification risk prediction model is iteratively trained using current enterprise prediction data packages. By comparing the accuracy analysis data of the two models, the reliability of the prediction results is verified. Simultaneously, a prediction data package verification process is introduced, dynamically updating the second enterprise certification risk prediction model based on the verification results. This achieves an upgrade from static fixation to dynamic iteration, ensuring that risk assessment always aligns with the real-time risk characteristics of enterprises and effectively avoiding assessment bias.

[0013] This invention does not simply apply conventional technologies such as big data analysis and machine learning to enterprise authentication risk scenarios, but rather uses an original closed-loop technical process.

[0014] The resource allocation method for dynamic prediction of authentication risks based on big data proposed in this invention is implemented as follows: By acquiring various types of data from enterprises through a big data platform, storing and initially processing these data, historical enterprise authentication information and enterprise data package information can be obtained. The first enterprise certification risk prediction model is trained based on historical enterprise certification information and enterprise data package information. Specifically, the standardized historical enterprise data package is used as feature input, and the certification results of past enterprise certifications are used as labels for training. This yields the first enterprise certification risk prediction model. Based on the output of the first enterprise certification risk prediction model, historical enterprise risk assessment data is generated. The accuracy of the first enterprise certification risk prediction model is then analyzed to obtain accuracy analysis data. This can be done by calculating accuracy or recall, etc., and is executed by setting a pre-defined accuracy threshold. When the accuracy of the first enterprise certification risk prediction model is lower than the accuracy threshold, an enterprise risk prediction command is issued. The enterprise risk prediction command is triggered based on the accuracy analysis data of the first enterprise certification risk prediction model. The system triggers the first enterprise certification risk prediction model to perform data prediction based on the enterprise risk prediction instruction, obtaining the current enterprise prediction data package information. The first enterprise certification risk prediction model is then trained based on this data package. The system continuously collects the latest current enterprise prediction data package information (such as the most recent quarterly inspection report, credit repair records, and the latest regulatory notices). Using an online learning or incremental learning mechanism, this new data is used as incremental samples to rapidly iterate and train the first enterprise certification risk prediction model, optimizing model parameters and generating a second enterprise certification risk prediction model with better performance and a more accurate reflection of the enterprise's recent risk characteristics. This ensures that the second model can dynamically adapt to changes in enterprise risk characteristics, addressing the issue of accuracy decay in static models. The system then outputs current enterprise risk assessment data based on the second enterprise certification risk prediction model, performing accuracy analysis on the second enterprise certification risk prediction model to obtain accuracy analysis data.

[0015] Based on the accuracy analysis data of the second and first enterprise certification risk prediction models, a model accuracy change analysis was conducted. This analysis triggered a sample tracing instruction. The core of the sample tracing and data package evaluation mechanism lies in quantitative analysis and interference elimination. The first step was to quantify the accuracy change, specifically the ratio between the accuracy of the second and first risk certification prediction models. This determined the model accuracy change coefficient. A coefficient greater than 1 indicated a significant change in performance due to the addition of new data. Furthermore, keyword analysis was used to extract pre-defined risk keywords from the enterprise data package, such as "penalty" and "suspension," as well as non-risk keywords, such as "valid" and "compliant." The system calculates the ratio of the two quantities to determine the risk characteristic change coefficient, thereby quantifying the fluctuation of the enterprise's own risk status. By calculating the current enterprise prediction data package evaluation coefficient = model accuracy change coefficient - risk characteristic change coefficient, the impact of the enterprise's own risk changes on the model is isolated. If the evaluation coefficient still exceeds the preset threshold, it indicates that the abnormal fluctuation of the model accuracy is likely due to the quality or representativeness of the current enterprise prediction data package itself, such as the presence of noise, outliers, or sample bias, rather than the actual risk changes of the enterprise. At this time, the sample tracing instruction and the current enterprise prediction data package verification instruction will be triggered. The current enterprise prediction data package will be evaluated based on the risk characteristic change coefficient, and the current enterprise prediction data package verification instruction will be triggered based on the data evaluation information.

[0016] The current enterprise prediction data packet verification command performs an actual similarity analysis on the current enterprise prediction data packet, verifies the current enterprise prediction data packet based on the actual similarity analysis data, and obtains the data packet prediction verification result.

[0017] The current enterprise prediction data packet is updated based on the data packet prediction verification result. The second enterprise certification risk prediction model is then updated based on the updated data packet, which in turn updates and predicts the current enterprise risk assessment data to obtain updated enterprise risk assessment data.

[0018] Based on current and updated enterprise risk assessment data, enterprises are categorized and allocated accordingly to determine the frequency of enterprise audits.

[0019] The resource allocation method and system for dynamic prediction of authentication risks based on big data proposed in this invention integrates intelligent systems for model monitoring, data quality assessment, dynamic verification, and closed-loop iteration; the core logic is a verification-to-update closed loop, rather than the self-training closed loop of existing technologies.

[0020] The resource allocation method and system for dynamic prediction of authentication risk based on big data proposed in this invention uses a first enterprise authentication risk prediction model to predict the current enterprise authentication information and obtain the current enterprise prediction data package information in order to generate a preliminary risk assessment to be verified, i.e., the business prediction stage.

[0021] The resource allocation method and system for dynamic prediction of authentication risks based on big data proposed in this invention also introduces a crucial verification mechanism. This method and system collects the current actual data packets of the enterprise and compares them with the predicted data packets, updating the data packets and models based on the verification results. Based on the above, a healthy iterative process from prediction to verification to updating based on real data is formed, conforming to the basic paradigm of online or incremental learning for all machine learning models.

[0022] This invention realizes the full-process digitalization and dynamization of enterprise certification risk management, breaks down data silos from multiple sources, and provides comprehensive data support for risk prediction; through model accuracy analysis and sample verification, it ensures the reliability of risk assessment results; based on accurate risk assessment, it enables the rational allocation of audit resources, avoids resource waste or insufficient management, and improves the systematicness and foresight of certification body risk management.

[0023] Furthermore, the present invention may further include the following extended functions, forming a closed loop from risk prediction to process control, specifically: This on-site attendance and verification technology integrates blockchain and biometrics. Addressing the issues of easily forged, easily simulated, and difficult-to-trace on-site attendance data, it employs multimodal biometrics for multi-factor authentication of auditors, ensuring the attendance record is accurate and performed in real-time. It also uses GPS, base station, and Wi-Fi fusion positioning technology to record precise location and trajectory. All attendance and trajectory data are encrypted immediately after generation and uploaded to the blockchain network for verification. The immutability and timestamp properties of blockchain solidify the evidence chain. This module is linked to a risk prediction engine; if abnormal behavior is detected (such as failure to arrive at the verification location at the scheduled time or abnormal trajectory deviation), an alert will be automatically triggered, notifying management personnel. Certificate Trusted Transmission Technology Based on Digital Signatures and Attribute Encryption: Certification authorities use national cryptographic algorithms to digitally sign electronic certificates, ensuring the trustworthiness of the certificate source and the integrity of the content. In the certificate transmission process, attribute-based encryption technology is adopted. Certificate holders can independently authorize their customers, partners, and other verifiers to verify the authenticity of the certificate online. During verification, only valid or invalid verification results and limited information authorized by the enterprise (such as certificate status and validity period) are returned, without disclosing the complete certificate content or other sensitive data of the enterprise. This builds a secure, controllable, and trusted transmission chain between the institution, the enterprise, and the customer.

[0024] In one embodiment of the present invention, S1 includes: By combining big data platforms with natural language processing methods, we can obtain public credit data, industry public opinion data, and publicly available corporate data. A corporate information database is constructed and stored using public credit data, industry public opinion data, and publicly available corporate data; The significance of building a database lies in realizing centralized management and efficient utilization of multi-source heterogeneous data, improving data query and analysis efficiency, and supporting comprehensive assessment of corporate credit and risk. When storing data, a combination of structured and unstructured forms is adopted. For example, basic corporate information is stored in structured fields (a preferred example: corporate name, unified social credit code, registration date, etc.). The raw data in the database is converted according to the preset enterprise data format specifications; For example, converting date fields from different sources to a format such as YYYY-MM-DD, and converting amounts to a uniform unit; The converted data is integrated based on the enterprise's main information, and various types of data from the same enterprise are linked together. The integrated data is packaged by enterprise to generate corresponding enterprise data package information and historical enterprise authentication information.

[0025] The working principle and technical effects of the above technical solution are as follows: Data on certificate status changes, certification information changes, and publicly available enterprise data are collected through a big data platform, with all data originating from publicly available information; a database is constructed based on data type and enterprise dimension to classify and store multi-source data, ensuring data storage security and scalability; the enterprise data in the database is uniformly converted and integrated into a package, generating a unique, structured enterprise data package and historical enterprise certification information for each enterprise. In this embodiment of the application, the certificate status change data may include at least: the status of the enterprise certificate being suspended, cancelled, or revoked; Changes to certification information may include: adjustments to the certification scope, changes in the frequency of surveillance audits, and changes in the renewal timeline. Publicly available data from enterprises may include at least: internal risk assessment data, changes or expiration of various administrative licenses and qualifications, data on enterprises being ordered to suspend production or business for rectification by administrative regulatory authorities, information on accidents related to quality, environment, health and safety, and information security, as well as data on changes in the enterprise's legal status, and data on changes in the enterprise's historical name, address, and scope. It should be noted that, in this embodiment of the application, the unified format conversion of enterprise data in the database can be achieved by converting data of different formats into the same type of standardized format.

[0026] The resource allocation method for dynamic prediction of authentication risks based on big data proposed in this invention solves the problems of inconsistent formats and scattered storage of multi-source data, and realizes centralized and standardized management of enterprise data; the storage architecture ensures storage efficiency and access convenience of massive enterprise data; the integrated enterprise data package provides a structured data foundation for subsequent risk feature analysis and model training, avoiding analysis bias caused by data chaos.

[0027] In one embodiment of the present invention, S2 includes: Credit risk characteristic analysis is performed on historical enterprise authentication information and enterprise data package information to obtain historical enterprise risk characteristic analysis data; The first enterprise authentication risk prediction model was trained based on historical enterprise authentication information and enterprise data package information. The first enterprise certification risk prediction model outputs historical enterprise risk assessment data. Based on the historical enterprise risk characteristic analysis data and historical enterprise risk assessment data, the model output accuracy analysis is performed to obtain the accuracy analysis data of the first enterprise certification risk prediction model. The accuracy analysis data of the first enterprise certification risk prediction model is compared with the preset model accuracy analysis threshold to obtain the accuracy comparison result of the first enterprise certification risk prediction model. The enterprise risk prediction instruction is triggered based on the accuracy comparison results of the first enterprise certification risk prediction model.

[0028] In a preferred example: For the target company, collect historical enterprise authentication information and enterprise data package information that meet the requirements of step S1. Based on the two types of information, conduct corresponding credit risk characteristic analysis, extract the core characteristic dimensions related to enterprise credit, and form the target company risk characteristic vector.

[0029] Historical enterprise certification information and enterprise data package information are used as input samples for the model. The past certification results of enterprises are used as judgment labels. Appropriate machine learning algorithms are selected to construct and train the first enterprise certification risk prediction model.

[0030] The historical information and data of the target enterprise are input into the trained first enterprise certification risk prediction model. Based on the built-in feature association rules and risk judgment logic, the model outputs the corresponding historical enterprise risk assessment data, which clarifies the credit risk level of the enterprise in the historical stage. The built-in feature association rules are set by the certification party according to actual needs, and the accuracy analysis of the output of the first enterprise certification risk prediction model is performed based on these rules, forming the accuracy analysis data of the first enterprise certification risk prediction model.

[0031] The accuracy analysis data of the first enterprise certification risk prediction model was compared with the bank's preset model accuracy analysis threshold to obtain the accuracy comparison results of the first enterprise certification risk prediction model.

[0032] Based on the accuracy comparison results of the first enterprise certification risk prediction model, the next step is to proceed as follows: if the model accuracy reaches the preset threshold, it is determined that the model can be directly used for the current enterprise risk prediction; if it does not reach the threshold, the enterprise risk prediction instruction is automatically triggered, and the model iteration and optimization process is started.

[0033] The working principle and technical effect of the above technical solution are as follows: Based on historical data, features related to credit risk are mined through a preset method. Using historical enterprise certification information as labels and enterprise data packages as input, a first enterprise certification risk prediction model is trained. The model outputs historical enterprise risk assessment data, and then, combined with historical risk feature analysis data, the model output accuracy analysis is carried out to obtain the accuracy data of the first enterprise certification risk prediction model. The accuracy data of the first enterprise certification risk prediction model is compared with a preset threshold. If the accuracy requirement is met, an enterprise risk prediction instruction is triggered. If the requirement is not met, the model training process needs to be optimized. The risk prediction model built on historical data provides basic algorithmic support for risk assessment; risk feature analysis clarifies the risk correlation factors in historical data, improving the relevance of model training; the model accuracy analysis and threshold comparison mechanism ensure the reliability of the initial model, avoids using an inaccurate model for actual risk prediction, and guarantees the effectiveness of risk assessment.

[0034] In one embodiment of the present invention, S3 includes: The prediction function of the first enterprise certification risk prediction model is triggered by the enterprise risk prediction command; The first enterprise certification risk prediction model is used to predict the current enterprise certification information to obtain the current enterprise prediction data package information. Credit risk characteristic analysis is performed based on current enterprise certification information and current enterprise prediction data package information to obtain current enterprise risk characteristic analysis data. The first enterprise authentication risk prediction model is trained based on the current enterprise authentication information and the current enterprise prediction data package information to obtain the second enterprise authentication risk prediction model. The current enterprise risk assessment data is output based on the second enterprise certification risk prediction model, such as... Figure 2 As shown; Based on the current enterprise risk characteristic analysis data and the current enterprise risk assessment data, the accuracy of the model output is analyzed to obtain the accuracy analysis data of the second enterprise certification risk prediction model.

[0035] In a preferred example: Based on the target company's past audit results, issue a risk prediction instruction to the company and activate the pre-set first enterprise certification risk prediction model; The first enterprise certification risk prediction model completes calculations based on the certification information of the target enterprise and outputs a risk trend prediction data package for the corresponding enterprise. By combining the target company's certification information and prediction data package, we can conduct qualification risk characteristic analysis and extract the company's core risk characteristics and evolution logic. Using the target company's certification information and prediction data package as new samples, the first company certification risk prediction model is iteratively optimized to generate the second company certification risk prediction model. The risk assessment of the target enterprise is completed through the second enterprise certification risk prediction model, and the corresponding risk assessment conclusions are output. By combining enterprise risk characteristic analysis data and risk assessment conclusions, we conduct model output accuracy analysis to form the accuracy analysis conclusions of the second enterprise certification risk prediction model.

[0036] The working principle and technical effect of the above technical solution are as follows: After triggering the enterprise risk prediction instruction, the first enterprise certification risk prediction model is called to predict the current enterprise certification information and generate the current enterprise prediction data package; combining the current enterprise's actual certification information and the prediction data package, the qualification risk feature analysis process is repeated to extract the current enterprise's risk feature data; using the current enterprise's certification information and the prediction data package as new training data, the first enterprise certification risk prediction model is iteratively trained to optimize the model parameters and obtain the second enterprise certification risk prediction model; using the second enterprise certification risk prediction model to output the current enterprise risk assessment data, and then combining it with the current risk feature analysis data to conduct model accuracy analysis, generating the second enterprise certification risk prediction model accuracy data, thus completing the upgrade from the historical model to the current optimized model.

[0037] By iteratively training the model with current data, the model can adapt to changes in the risk characteristics of the current enterprise, avoiding the problem of the model becoming outdated due to long-term use of historical data. The generation of current enterprise risk assessment data and second enterprise certification risk prediction model accuracy data provides current dimension data support for model accuracy change analysis and sample verification, improving the timeliness and accuracy of risk prediction.

[0038] In one embodiment of the present invention, the qualification risk characteristic analysis includes: Based on preset risk characteristic judgment information, qualification risk keywords and non-qualification risk keywords are obtained from enterprise certification information and enterprise data package information. The preset risk characteristic judgment information includes preset risk characteristic information and preset non-risk characteristic information, used to distinguish non-risk characteristic information and avoid non-risk characteristic information being mistaken for risk characteristic information, leading to inaccurate sample data. The enterprise certification information and enterprise data package information here can be historical or current enterprise predicted data package information, etc. Obtain the ratio of non-qualification risk keywords to qualification risk keywords to obtain the keyword analysis coefficient; The keyword analysis coefficients are compared with preset keyword analysis thresholds to obtain keyword comparison results; Based on the keyword comparison results, a sample bias warning is triggered, and the keyword analysis coefficients are used to generate enterprise risk characteristic analysis data. Enterprise risk characteristic analysis data consists of keyword analysis coefficients exceeding a threshold. The role of the keyword analysis coefficients is to analyze pseudo-risks in the model training samples, thereby determining whether the training samples are accurate and effective.

[0039] In a preferred example: Pre-define risk-related information (such as certificate suspension, continuous changes, etc.) and non-risk-related information (such as certificate validity status); Clearly define the criteria for distinguishing between the two types of information to avoid misclassifying non-risk information as risk information; Based on the above rules, qualification risk keywords and non-qualification risk keywords are extracted from the enterprise's historical and current certification information and prediction data packages. Qualification risk keywords include things like certificate suspension, cancellation, revocation, frequent adjustments to certification scope, and multiple changes of auditors. Non-qualification risk keywords include things like certificate continued validity, complete audit records, stable certification scope, and long-term auditor service. The number of keywords in the two categories is counted, the ratio of non-qualification risk keywords to qualification risk keywords is calculated, and the keyword analysis coefficient is obtained. The calculated keyword analysis coefficients are compared with preset thresholds to obtain keyword comparison results; if the coefficients exceed the thresholds, a sample bias warning is triggered. Keyword analysis coefficients exceeding the threshold are directly used as enterprise risk characteristic analysis data to determine whether there are false risk interferences in the model training samples and whether they are accurate and usable.

[0040] In a preferred embodiment, the setting of the threshold and the judgment based on the threshold are as follows: Calculate the coefficient values ​​for all samples and divide them into intervals based on the actual qualification and risk situation of the enterprises: For companies that have far more non-risk keywords than risky keywords, they are generally classified as low-risk companies. For companies with slightly more non-risk keywords than risky keywords, there may be potential risks, and they will be classified as medium-risk companies. For companies with far fewer non-risk keywords than risky keywords, they will be identified as high-risk companies, and their review frequency will be increased accordingly. The working principle and technical effect of the above technical solution are as follows: Based on the preset risk feature judgment information, qualification risk keywords and non-qualification risk keywords are screened from enterprise certification information and data packages; the ratio of the number of non-qualification risk keywords to qualification risk keywords is calculated to obtain the keyword analysis coefficient; the coefficient is compared with the preset keyword analysis threshold, and if the coefficient is too high, a sample bias warning is triggered; the keyword analysis coefficient is used as enterprise risk feature analysis data to determine whether there is false risk information in the training sample, so as to ensure the accuracy of the sample.

[0041] The following is an example of calculating the ratio of non-qualification risk keywords to qualification risk keywords: Keyword analysis coefficient = Number of non-qualification risk keywords / Number of qualification risk keywords; This measure aims to assess the relative proportion of positive / neutral information to negative / risk information in a company's data. A very low coefficient indicates that the data consists almost entirely of risky keywords, suggesting a high level of risk for the company.

[0042] An excessively high coefficient indicates that there are very few risky keywords, while non-risky keywords make up the vast majority.

[0043] The information may be incomplete, deliberately embellished, or the sample may be too one-sided, failing to fully reflect the true situation of the enterprise, thus triggering a sample bias warning.

[0044] In this embodiment of the application, the preset risk feature judgment information may include: preset risk features and non-risk features; Pre-defined risk characteristics include certificate suspension, cancellation, revocation, and frequent adjustments to the scope of certification; Pre-defined non-risk characteristics include certificates that remain valid and within their validity period, a stable certification scope over a long period, continuous and complete audit records, and a fixed auditor team; In the embodiments of this application, an excessively high coefficient indicates that the proportion of non-risk features is too high, and there may be non-risk features mixed into the sample.

[0045] By using keyword classification and coefficient calculation, the system effectively distinguishes between risk and non-risk characteristics of enterprises, preventing non-risk information from being misjudged as risk information and reducing the interference of pseudo-risk samples on model training. The sample bias warning mechanism can promptly detect deviations in sample data, providing a basis for sample optimization, improving the accuracy of risk characteristic analysis, and thus ensuring the quality of model training samples.

[0046] In one embodiment of the present invention, S4 includes: The ratio of the accuracy analysis data of the second enterprise certification risk prediction model to the accuracy analysis data of the first enterprise certification risk prediction model is obtained to obtain the model accuracy variation coefficient; the model accuracy variation coefficient is used to analyze the changes in model accuracy. The model accuracy variation coefficient is compared with the preset model accuracy variation threshold to obtain the model accuracy comparison results; The sample tracing instruction is triggered based on the model accuracy comparison results; if the model accuracy change coefficient is greater than the preset model accuracy change threshold, it indicates that there is a difference in input information between the two models, triggering the sample tracing instruction to perform sample tracing analysis.

[0047] Based on the sample tracing instructions, obtain the ratio of the current enterprise risk characteristic analysis data to the historical enterprise risk characteristic analysis data, and obtain the risk characteristic change coefficient; The difference between the model accuracy variation coefficient and the risk characteristic variation coefficient is calculated to obtain the evaluation coefficient of the current enterprise prediction data package; by obtaining the difference, the influence of sample characteristics is eliminated, thereby realizing the prediction accuracy analysis of the current enterprise prediction data package; The current enterprise predicted data packet evaluation coefficient is compared with the preset current data packet evaluation threshold to obtain the current data packet evaluation comparison result; The current enterprise predicted data packet verification instruction is triggered based on the current data packet evaluation and comparison results.

[0048] The working principle and technical effect of the above technical solution are as follows: The ratio of the accuracy analysis data of the second enterprise certification risk prediction model to the accuracy analysis data of the first enterprise certification risk prediction model is calculated to obtain the model accuracy change coefficient, thereby quantifying the trend of model accuracy change; this coefficient is compared with a preset model accuracy change threshold. If the coefficient exceeds the threshold (indicating abnormal model accuracy change, possibly due to differences in input samples), a sample tracing instruction is triggered; under the sample tracing instruction, the ratio of the current enterprise risk characteristic analysis data to the historical enterprise risk characteristic analysis data is calculated to obtain the risk characteristic change coefficient; by calculating the difference between the model accuracy change coefficient and the risk characteristic change coefficient, the influence of natural changes in sample characteristics on model accuracy is eliminated, and the evaluation coefficient of the current enterprise prediction data package is obtained; this evaluation coefficient is compared with a preset threshold, and a verification instruction for the current enterprise prediction data package is triggered based on the result.

[0049] The model accuracy variation coefficient is used to quantify accuracy fluctuations and promptly detect abnormal changes in the model; the calculation of risk feature variation coefficient and evaluation coefficient effectively distinguishes the impact of sample feature changes and prediction data deviations on model accuracy, avoiding misjudgments; the sample traceability and verification command triggering mechanism provides a clear basis for the reliability verification of prediction data and ensures the credibility of model output results.

[0050] In one embodiment of the present invention, S5 includes: Collect the actual data packets of the current enterprise based on the current enterprise predicted data packet verification instructions; Perform a similarity analysis between the current enterprise's actual data packets and the current enterprise's preset data packets to obtain data packet similarity analysis data; The data packet similarity analysis data is compared with a preset similarity analysis threshold to obtain the similarity comparison result; The current enterprise's predicted data packets are verified based on the similarity comparison results to obtain the data packet prediction verification results.

[0051] The working principle and technical effect of the above technical solution are as follows: After triggering the current enterprise's predicted data packet verification command, the actual data packet of the current enterprise is collected; through methods such as text similarity matching and numerical deviation analysis, the similarity between the actual data packet and the predicted data packet of the current enterprise is compared to generate data packet similarity analysis data; the similarity analysis data is compared with a preset similarity analysis threshold. If the similarity is higher than the threshold, it means that the predicted data packet matches the actual situation and the verification is successful; if the similarity is lower than the threshold, it is determined that the predicted data packet has a deviation and the verification fails; the data packet prediction verification result is generated based on the comparison results.

[0052] In this embodiment of the application, the actual data package of the current enterprise may include: the latest certification certificate status of the enterprise (such as real-time status such as valid, suspended, or cancelled), the actual certification scope change record (including change time, change content and approval results), key conclusions in recent audit reports (such as audit findings, non-compliance items, etc.). By directly comparing actual data with predicted data, the accuracy of the predicted data package is intuitively verified, avoiding subsequent risk assessments based on erroneous predicted data. Similarity analysis and threshold judgment mechanisms provide standardized basis for verification results, reducing the subjectivity of manual verification. The verification results provide key references for model updates and risk assessment data correction, ensuring the reliability of decision-making.

[0053] In a preferred example, the step of updating the current enterprise predicted data packet based on the data packet prediction verification result, and thereby obtaining updated enterprise risk assessment data, includes: Based on the data packet prediction verification results, the current enterprise's actual data packets are used to update the current enterprise's predicted data packets, and the updated data packets for the current enterprise are obtained. The second enterprise certification risk prediction model is updated by using the current enterprise certification information and the current enterprise update data package to obtain the updated enterprise certification risk prediction model. Based on the updated enterprise certification risk prediction model, the current enterprise risk assessment data is updated and predicted to obtain updated enterprise risk assessment data.

[0054] The working principle and technical effect of the above technical solution are as follows: Based on the data packet prediction verification results, if there is a deviation in the predicted data packet, the actual data packet of the current enterprise is used to replace the predicted data packet to generate an updated data packet for the current enterprise; the current enterprise authentication information and the updated data packet are used as new training data to adjust the parameters and retrain the second enterprise authentication risk prediction model, optimize the model's adaptability to the risk characteristics of the current enterprise, and obtain an updated enterprise authentication risk prediction model; the updated model is used to re-predict the risk level of the current enterprise, generate updated enterprise risk assessment data, and replace the original risk assessment results. It enables dynamic iterative updates of the model, allowing the model to be corrected in a timely manner based on actual data deviations, avoiding the long-term use of biased models; the updated risk assessment data is more in line with the actual situation of enterprises, improving the accuracy of risk assessment results; it provides a more reliable risk basis for audit configuration optimization and resource allocation, reducing control errors caused by risk assessment deviations.

[0055] In one embodiment of the present invention, S6 includes: By combining the company's historical risk assessment records, industry risk benchmarks, and factors such as the company's current business scale and certification coverage, a multi-dimensional resource allocation assessment index system is constructed. Among them, risk level is the core indicator, which directly determines the priority of resource allocation. Industry risk benchmarks can help judge the company's relative risk level within the industry. Companies with risks higher than the industry average are given special attention. Business scale and certification coverage affect the specific scale of resource allocation. Companies with larger scale or wider certification scope involve more audit stages, and the required audit resources (such as the number of auditors and audit time) also increase accordingly.

[0056] The specific process for determining the review frequency has been described above and will not be repeated here.

[0057] According to one embodiment of the present invention, the system includes: The data acquisition module is used to acquire the company's historical enterprise authentication information and enterprise data package information through the big data platform; The historical model analysis module is used to train and analyze the accuracy of the first enterprise certification risk prediction model based on historical enterprise certification information and enterprise data package information, obtain current enterprise risk assessment data and accuracy analysis data of the first enterprise certification risk prediction model, and trigger enterprise risk prediction instructions based on the accuracy analysis data of the first enterprise certification risk prediction model. The current model analysis module is used to obtain the current enterprise prediction data package information according to the enterprise risk prediction instruction, and to train and analyze the accuracy of the second enterprise certification risk prediction model based on the current enterprise prediction data package information, thereby obtaining the accuracy analysis data of the second enterprise certification risk prediction model. The data packet verification module is used to analyze the model accuracy change based on the accuracy analysis data of the first enterprise certification risk prediction model and the accuracy analysis data of the second enterprise certification risk prediction model, and then evaluate the current enterprise prediction data packet, thereby triggering the current enterprise prediction data packet verification instruction. The data packet update module is used to perform data verification and analysis based on the current enterprise prediction data packet verification instructions, update the current enterprise prediction data packet based on the obtained current enterprise prediction data packet verification results, and then update the second enterprise certification risk prediction model to obtain updated enterprise risk assessment data. The allocation module is used to allocate enterprises according to their risk levels based on current enterprise risk assessment data and updated enterprise risk assessment data, thereby obtaining the enterprise review frequency.

[0058] The working principle and technical effects of the above technical solution are as follows: A big data platform is used to integrate multi-source enterprise data and construct a database, forming a structured enterprise data package. Combined with enterprise authentication information, credit risk characteristic analysis, risk prediction, and model accuracy assessment are conducted from both historical and current data dimensions, triggering risk prediction commands to obtain current enterprise risk assessment results and model accuracy data. Based on a comparison of historical and current model accuracy data, the trend of accuracy change is analyzed, triggering sample tracing commands to verify the current enterprise prediction data, obtaining data package prediction verification results. The model and risk assessment data are updated based on the verification results, thereby optimizing the audit configuration and generating the enterprise audit frequency, forming a closed-loop process. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims and equivalents of this invention, this invention also intends to include these modifications and variations.

Claims

1. A resource allocation method for dynamic prediction of authentication risks based on big data, characterized in that, The method includes: S1. Obtain the enterprise's historical enterprise authentication information and enterprise data package information through the big data platform; S2. Based on historical enterprise certification information and enterprise data package information, train and analyze the accuracy of the first enterprise certification risk prediction model to obtain current enterprise risk assessment data and accuracy analysis data of the first enterprise certification risk prediction model. Trigger enterprise risk prediction instructions based on the accuracy analysis data of the first enterprise certification risk prediction model. S3. Obtain the current enterprise prediction data packet information according to the enterprise risk prediction instruction, and train and analyze the accuracy of the second enterprise certification risk prediction model based on the current enterprise prediction data packet information to obtain the accuracy analysis data of the second enterprise certification risk prediction model. S4. Evaluate the current enterprise prediction data package based on the accuracy analysis data of the first enterprise certification risk prediction model and the accuracy analysis data of the second enterprise certification risk prediction model, and determine whether to trigger the current enterprise prediction data package verification instruction based on the evaluation results. S5. Perform data verification analysis based on the current enterprise prediction data package verification instructions, update the current enterprise prediction data package based on the obtained current enterprise prediction data package verification results, update the second enterprise certification risk prediction model, and obtain updated enterprise risk assessment data. S6. Based on the current enterprise risk assessment data and the updated enterprise risk assessment data, classify and allocate enterprises to obtain the enterprise review frequency; S2 further includes: Credit risk characteristic analysis is performed on historical enterprise authentication information and enterprise data package information to obtain historical enterprise risk characteristic analysis data; The credit risk characteristic analysis includes: Based on preset risk characteristics, obtain qualification risk keywords and non-qualification risk keywords from enterprise certification information and enterprise data package information; Obtain the ratio of non-qualification risk keywords to qualification risk keywords to obtain the keyword analysis coefficient; The keyword analysis coefficients are compared with preset keyword analysis thresholds to obtain keyword comparison results; Based on the keyword comparison results, a sample bias warning is triggered, and the keyword analysis coefficients are used to generate enterprise risk characteristic analysis data. S4 includes: The ratio of the accuracy analysis data of the second enterprise certification risk prediction model to the accuracy analysis data of the first enterprise certification risk prediction model is obtained to obtain the model accuracy variation coefficient. The model accuracy variation coefficient is compared with the preset model accuracy variation threshold to obtain the model accuracy comparison result; a sample tracing instruction is triggered based on the model accuracy comparison result. Based on the sample tracing instructions, obtain the ratio of the current enterprise risk characteristic analysis data to the historical enterprise risk characteristic analysis data, and obtain the risk characteristic change coefficient; The difference between the coefficient of change in model accuracy and the coefficient of change in risk characteristics is used to obtain the evaluation coefficient of the current enterprise prediction data package; The current enterprise predicted data packet evaluation coefficient is compared with the preset current data packet evaluation threshold to obtain the current data packet evaluation comparison result; Trigger the current enterprise predicted data packet verification instruction based on the current data packet evaluation and comparison results; S5 includes: Collect the actual data packets of the current enterprise based on the current enterprise predicted data packet verification instructions; Perform a similarity analysis between the current enterprise's actual data packets and the current enterprise's predicted data packets to obtain data packet similarity analysis data; The data packet similarity analysis data is compared with a preset similarity analysis threshold to obtain the similarity comparison result; The current enterprise's predicted data packets are verified based on the similarity comparison results to obtain the data packet prediction verification results; Based on the data packet prediction verification results, update the current enterprise prediction data packet to obtain updated enterprise risk assessment data; The step of updating the current enterprise prediction data packet based on the data packet prediction verification results to obtain updated enterprise risk assessment data includes: Based on the data packet prediction verification results, the current enterprise's actual data packets are used to update the current enterprise's predicted data packets, and the updated data packets for the current enterprise are obtained. The second enterprise certification risk prediction model is updated by using the current enterprise certification information and the current enterprise update data package to obtain the updated enterprise certification risk prediction model. Based on the updated enterprise certification risk prediction model, the current enterprise risk assessment data is updated and predicted to obtain updated enterprise risk assessment data.

2. The resource allocation method for dynamic prediction of authentication risk based on big data as described in claim 1, characterized in that, S1 includes: By combining big data platforms with natural language processing methods, we can obtain data on certificate status changes, authentication information changes, and publicly available enterprise data. A corporate information database is constructed and stored using certificate status change data, certification information change data, and publicly available corporate data. The raw data in the enterprise information database is converted according to the preset enterprise data format specifications; The converted data is integrated based on the enterprise's main information, and various types of data from the same enterprise are linked together. The integrated data is packaged by enterprise to generate corresponding enterprise data package information and historical enterprise authentication information.

3. The resource allocation method for dynamic prediction of authentication risk based on big data as described in claim 1, characterized in that, S2 includes: The first enterprise authentication risk prediction model was trained based on historical enterprise authentication information and enterprise data package information. The first enterprise certification risk prediction model outputs historical enterprise risk assessment data. Based on the historical enterprise risk characteristic analysis data and historical enterprise risk assessment data, the model output accuracy analysis is performed to obtain the accuracy analysis data of the first enterprise certification risk prediction model. The accuracy analysis data of the first enterprise certification risk prediction model is compared with the preset model accuracy analysis threshold to obtain the accuracy comparison result of the first enterprise certification risk prediction model. The enterprise risk prediction instruction is triggered based on the accuracy comparison results of the first enterprise certification risk prediction model.

4. The resource allocation method for dynamic prediction of authentication risk based on big data as described in claim 3, characterized in that, The S3 includes: triggering the prediction function of the first enterprise certification risk prediction model through an enterprise risk prediction instruction; According to the enterprise risk prediction instruction, the current enterprise certification information is predicted using the first enterprise certification risk prediction model to obtain the current enterprise prediction data package information. Credit risk characteristic analysis is performed based on current enterprise certification information and current enterprise prediction data package information to obtain current enterprise risk characteristic analysis data. The first enterprise authentication risk prediction model is trained based on the current enterprise authentication information and the current enterprise prediction data package information to obtain the second enterprise authentication risk prediction model. Output the current enterprise risk assessment data based on the second enterprise certification risk prediction model; Based on the current enterprise risk characteristic analysis data and the current enterprise risk assessment data, the accuracy analysis of the model output is performed to obtain the accuracy analysis data of the second enterprise certification risk prediction model.

5. The resource allocation method for dynamic prediction of authentication risk based on big data as described in claim 1, characterized in that, S6 includes: Based on the current enterprise risk assessment data, perform preliminary configuration of enterprise certification to obtain preliminary configuration data; Based on the updated enterprise risk assessment data, perform in-depth configuration of enterprise certification to obtain in-depth configuration data; The initial configuration data is adjusted based on the deep configuration data to obtain the enterprise review frequency; Based on current and updated enterprise risk assessment data, enterprises are categorized and allocated accordingly to determine the frequency of enterprise audits.

6. A resource allocation system for dynamic prediction of authentication risks based on big data, characterized in that, The resource allocation method for dynamic prediction of authentication risks based on big data, as described in any one of claims 1 to 5, comprises: The data acquisition module is used to acquire the enterprise's historical enterprise authentication information and enterprise data package information through the big data platform; The historical model analysis module is used to train and analyze the accuracy of the first enterprise certification risk prediction model based on historical enterprise certification information and enterprise data package information, obtain current enterprise risk assessment data and accuracy analysis data of the first enterprise certification risk prediction model, and trigger enterprise risk prediction instructions based on the accuracy analysis data of the first enterprise certification risk prediction model. The current model analysis module is used to obtain the current enterprise prediction data package information according to the enterprise risk prediction instruction, and to train and analyze the accuracy of the second enterprise certification risk prediction model based on the current enterprise prediction data package information, thereby obtaining the accuracy analysis data of the second enterprise certification risk prediction model. The data packet verification module is used to perform model accuracy change analysis based on the accuracy analysis data of the first enterprise certification risk prediction model and the accuracy analysis data of the second enterprise certification risk prediction model, and to evaluate the current enterprise prediction data packet, thereby triggering the current enterprise prediction data packet verification instruction. The data packet update module is used to perform data verification analysis based on the current enterprise prediction data packet verification instructions, update the current enterprise prediction data packet based on the obtained current enterprise prediction data packet verification results, update the second enterprise certification risk prediction model, and obtain updated enterprise risk assessment data. The allocation module is used to allocate enterprises according to their risk levels based on current enterprise risk assessment data and updated enterprise risk assessment data, thereby obtaining the enterprise review frequency.

Images