Risk intelligent identification method and system for recruitment system of fusion agent

Abstract

The application discloses a bidding system risk intelligent identification method and system of fusion intelligent agent, belongs to the technical field of intelligent risk management and control, and the method comprises the following steps: key features of bidding business are extracted through multi-agent cooperation to generate a tenderer risk feature library; key risk indexes are extracted, a tenderer risk matrix is constructed, a sensitive comprehensive value is calculated to determine a sensitive level, a field risk entropy, a project risk entropy and a robustness are calculated, a comprehensive risk coefficient is generated, a tenderer candidate pool is generated after entropy weight evaluation; three-level risk verification rules are constructed, and risk verification is carried out level by level based on the integrity of qualification, data security adaptability and historical risk credit; the risk audit score is calculated, the risk audit is determined, if not passed, the audit deviation degree is calculated to determine the abnormal risk level, trigger hierarchical management and control, and generate a risk dynamic monitoring report; the intelligentization of risk identification management is realized, and the reliability of bidding risk prevention and control is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent risk management technology, specifically to a method and system for intelligent risk identification in a procurement system that integrates intelligent agents. Background Technology

[0002] In procurement management, the bidder is a core link, and its risk management directly determines the stability and operational efficiency of the procurement project. However, existing bidder risk management has some problems, including: relying on manual experience to assign weights to multi-dimensional evaluation indicators, resulting in large subjective biases and difficulty in objectively reflecting the actual importance of the indicators; bidder data is scattered across various business systems, forming information silos, making it impossible to integrate dynamic risk signals from multiple levels of bidders in real time; and most existing digital systems are single-point risk detection systems, lacking a global risk perception and hierarchical verification mechanism through multi-agent collaboration.

[0003] A patent application with publication number CN121860389A discloses a digital and intelligent procurement management method and system. The method includes: acquiring and segmenting procurement process data to obtain data for multiple procurement process stages; performing anomaly analysis on the data of each procurement process stage to determine abnormal risk indicators, and identifying abnormal procurement process stage data based on these indicators; performing data analysis on the abnormal risk indicators to determine abnormal characteristics and risk types, and assessing the risk of each abnormal procurement process stage data based on these indicators; determining the risk assessment value of the procurement process based on the risk assessment results of each abnormal procurement process stage data, and determining the risk level of the procurement process based on this value; and determining a risk emergency management strategy based on the risk level to conduct emergency management of the risks in the procurement process. This technical solution constructs a fully intelligent system from risk perception, identification, assessment to handling, improving the risk prevention and control capabilities, operational efficiency, and compliance level of procurement operations, providing a solid technical guarantee for the sound operation of enterprises.

[0004] Although there is an existing digital and intelligent procurement management method and system that enables automatic risk perception, identification, quantification, and closed-loop control throughout the procurement process, the lack of a multi-agent collaborative risk identification mechanism in procurement business scenarios leads to the system distributing projects to unqualified bidders without multi-level risk verification, thereby causing procurement compliance risks and data security vulnerabilities. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention aims to provide a method and system for intelligent risk identification in procurement systems that integrates intelligent agents. By constructing a multi-agent conflict disambiguation mechanism coordinated by a master intelligent agent, it combines weighted voting and secondary verification to adjudicate conflicting labels of the same feature in multi-source data, thereby improving the data quality of the bidder risk feature database. Furthermore, by introducing a dynamic weight adaptation mechanism driven by sensitivity levels, it adjusts the influence weights of key risk indicators in real time according to the quantified values ​​of confidentiality levels and privacy protection requirements, ensuring that risk assessments match the actual risk exposure of the project. By constructing three-level risk verification rules and setting independent verification thresholds and rectification mechanisms for each dimension, it achieves dimensional positioning and targeted handling of qualification, data security, and historical credit dimensions, avoiding the inefficiency of terminating the entire process due to non-compliance in a single dimension. Simultaneously, by achieving automated collaboration among various stages through multiple intelligent agents, it enhances the intelligence level of risk identification in procurement systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A risk intelligent identification method for procurement systems that integrates intelligent agents includes:

[0008] By leveraging multi-agent collaboration, key features of the procurement process are extracted, generating a risk feature database for bidders.

[0009] Extract key risk indicators, construct a risk matrix for bidders, calculate domain risk entropy, quantify the data sensitivity of target procurement projects, calculate the comprehensive sensitivity value to determine the sensitivity level, calculate project risk entropy and robustness, combine domain risk entropy to generate a comprehensive risk coefficient, conduct entropy weight assessment, and generate a candidate pool of bidders.

[0010] A three-tiered risk verification rule is established, which verifies the risk level by level based on the integrity of qualifications, data security compatibility and historical risk credit, and selects candidate bidders who comply with the three-tiered risk verification.

[0011] For candidate bidders who meet the Level 3 risk verification, a risk audit score is calculated, and a risk audit judgment is made. If the audit fails, the audit deviation is calculated, the abnormal risk level is determined, and hierarchical control is triggered. All intelligent agents work together to generate a dynamic risk monitoring report.

[0012] Specifically, the steps for determining the sensitivity level include:

[0013] Extract key risk indicators from the target area and calculate risk coefficients;

[0014] Key risk indicators with risk coefficients greater than a preset risk threshold are selected, and after normalization, a set of key risk indicators is formed.

[0015] Construct a risk matrix for bidders, obtain risk weighting factors, extract quantitative values ​​of key risk indicators for each bidder, and calculate domain risk entropy.

[0016] In the top ranking The average domain risk entropy of the risk compliance bidders is the benchmark entropy. If the domain risk entropy is greater than the benchmark entropy, the corresponding bidder is eliminated.

[0017] The classification level is mapped to a classification quantification value, and the privacy protection requirement is mapped to a privacy protection quantification value. A weighted comprehensive sensitivity value is calculated, and the sensitivity level is determined.

[0018] If the overall sensitivity value is greater than the preset sensitivity threshold, it is determined to be at a high sensitivity level; otherwise, it is determined to be at a low sensitivity level.

[0019] Specifically, the steps for generating the pool of bidder candidates include:

[0020] The impact weights of each key risk indicator are dynamically adjusted based on the sensitivity level.

[0021] Based on the bidder's risk matrix, the risk impact of each key risk indicator is calculated, and the project risk entropy is calculated by weighting the sensitivity comprehensive value.

[0022] Obtain sandbox simulation records, extract features from the simulated scene and calculate the first similarity between the simulated scene features and the target scene features;

[0023] When the first similarity is greater than the preset similarity threshold, the risk compliance rate, risk anomaly recall rate, and assessment stability rate are selected, and the robustness is calculated by weighting and performing a robustness assessment.

[0024] If the robustness is greater than or equal to the preset pass threshold, the robustness is deemed to be qualified, and the qualified bidder is retained; otherwise, the robustness is deemed to be unqualified, and the unqualified bidder is eliminated.

[0025] Specifically, the steps for generating the pool of bidder candidates also include:

[0026] Set the basic weights for domain risk entropy, project risk entropy, and robustness respectively;

[0027] The NSGA-II multi-objective optimization algorithm is used to determine the Pareto optimal weights for domain risk entropy, project risk entropy, and robustness, respectively.

[0028] The risk entropy of the domain, the risk entropy of the project, and the robustness are weighted and integrated to generate a comprehensive risk coefficient for each bidder, and then an entropy weight assessment is performed.

[0029] If the overall risk coefficient is greater than the preset risk matching threshold, it is assessed as high risk and the corresponding bidder is eliminated.

[0030] Otherwise, if the risk is assessed as low, the corresponding bidder is retained and sorted in ascending order according to the comprehensive risk coefficient to generate a candidate pool of bidders.

[0031] Specifically, the steps for constructing the three-level risk verification rules include:

[0032] Based on the compliance standards of the target field and the data security level of the procurement project, a three-level risk verification rule is constructed, including the first level verifying the integrity of the candidate bidder's qualifications, the second level verifying the data security compatibility of the candidate bidder, and the third level verifying the candidate bidder's historical risk credit.

[0033] Set initial verification thresholds for each level, including the first-level initial verification threshold, the second-level initial verification threshold, and the third-level initial verification threshold.

[0034] Specifically, the steps for conducting step-by-step risk verification include:

[0035] Calculate the qualification validity score, security adaptation score, and historical credit score;

[0036] The initial verification thresholds at each level are corrected in real time to obtain the first-level verification threshold, the second-level verification threshold, and the third-level verification threshold, and risk verification is performed step by step.

[0037] When the valid qualification score is greater than or equal to the first-level verification threshold, the first-level verification is deemed compliant, candidate bidders compliant with the first-level verification are screened, and the second-level risk verification is triggered.

[0038] For candidate bidders who meet the Level 1 verification requirements;

[0039] When the security compatibility score is greater than or equal to the level 2 verification threshold, the level 2 verification is deemed compliant, candidate bidders who meet the level 2 verification are selected, and level 3 risk verification is triggered.

[0040] For candidate bidders who meet the Level 2 verification requirements;

[0041] When the historical credit score is greater than or equal to the Level 3 verification threshold, the Level 3 risk verification is deemed compliant. Candidate bidders who meet the Level 3 risk verification criteria are selected and ranked in ascending order of comprehensive risk coefficient to determine the first candidate bidder list.

[0042] Specifically, the steps for conducting a risk audit assessment include:

[0043] Obtain basic risk information of bidders and extract historical feature data;

[0044] By associating key information from the first candidate bidder list, basic risk information of bidders, and historical characteristic data, a bidder risk information set is formed, and itemized scoring and calibration are performed.

[0045] In terms of qualifications, obtain real-time valid qualification scores, and calculate and calibrate qualification scores based on the valid qualification scores.

[0046] In terms of data security, the security adaptation score is adjusted according to the sensitivity level to obtain the calibrated security score.

[0047] In terms of historical credit, the historical credit score is used as the benchmark value, and the calibrated credit score is obtained by deducting the corresponding violation record points.

[0048] Based on calibration qualification score, calibration safety score and calibration credit score, a risk audit score is calculated by weighting and a risk audit judgment is made.

[0049] If the risk audit score is greater than or equal to the preset risk audit threshold, the risk audit is deemed to have passed, and a second list of candidate bidders is obtained; otherwise, the risk audit is deemed to have failed, and the bidder is marked as an abnormal bidder.

[0050] Specifically, the steps for determining the level of abnormal risk include:

[0051] Iterate through the second-tier bidder candidate list, filtering by risk audit score in ascending order. The list of candidate bidders;

[0052] Integrate the content and risk constraints of the target procurement projects to generate targeted bidding instructions and push them to specific parties;

[0053] For bidders with unusual behavior, the audit deviation is calculated based on the risk audit score and the risk audit threshold. ;

[0054] Set risk deviation threshold , ,and Determine the level of abnormal risk;

[0055] like If it is, it is judged as a low-risk anomaly; if If so, it is judged as a medium-risk abnormality; if If so, it is judged as a high-risk anomaly.

[0056] Specifically, the steps for generating a dynamic risk monitoring report include:

[0057] Based on the level of abnormal risk, tiered control measures are triggered.

[0058] For low-risk anomalies, record the warning status in the bidder's risk characteristic database and push risk alerts;

[0059] For projects with abnormal medium-risk conditions, project awarding will be suspended, and rectification materials will be submitted.

[0060] For high-risk anomalies, remove candidate bidders from the second-tier candidate list and finally determine the bidder list;

[0061] In the bidder risk feature database, add risk audit pass status, related project award records, update abnormal risk level and control measures to obtain the updated bidder risk feature database;

[0062] Integrate risk audit findings, abnormal risk levels, hierarchical control implementation records, bidder list, and updated bidder risk characteristic database;

[0063] Generate dynamic risk monitoring reports.

[0064] The intelligent risk identification system for the procurement system integrating intelligent agents includes: a risk feature construction module, a risk entropy weight assessment module, a risk verification module, and a risk audit module;

[0065] The risk feature construction module is used to extract key features of the bidding and procurement business through multi-agent collaboration and generate a risk feature library of bidders.

[0066] The risk entropy weight assessment module is used to extract key risk indicators, construct a risk matrix for bidders, calculate domain risk entropy, quantify the data sensitivity of target procurement projects, calculate a comprehensive sensitivity value to determine the sensitivity level, calculate project risk entropy and robustness, generate a comprehensive risk coefficient by combining domain risk entropy, conduct entropy weight assessment, and generate a candidate pool of bidders.

[0067] The risk verification module is used to construct a three-level risk verification rule, and to perform step-by-step risk verification based on the integrity of qualifications, data security adaptability and historical risk credit, and to screen candidate bidders who comply with the three-level risk verification.

[0068] The risk audit module is used to calculate the risk audit score for candidate bidders who meet the three-level risk verification, make a risk audit judgment, and calculate the audit deviation degree if they fail to pass the audit, determine the abnormal risk level and trigger hierarchical control. All intelligent agents work together to summarize and generate a dynamic risk monitoring report.

[0069] The beneficial effects of this invention are:

[0070] 1. This invention coordinates multiple sub-agents to process multi-source heterogeneous data in parallel through a master intelligent agent. It employs a weighted voting rule based on confidence level and data source authority to disambiguate conflicting labels, improving the consistency and accuracy of risk feature extraction and overcoming the information silos and noise interference problems caused by single data sources or manual extraction. It innovatively introduces domain risk entropy to quantify industry risk distribution, combining the confidentiality level and privacy requirements of target procurement projects to quantify the comprehensive sensitivity value. The invention dynamically adjusts the weights of indicators to calculate the project risk entropy, making risk assessment more closely aligned with project scenarios and providing industry-leading insights. The average domain risk entropy of each risk-compliant bidder is used as the benchmark entropy. Robustness is calculated by combining sandbox simulation records. The comprehensive risk coefficient is calculated by weighting the domain risk entropy and the project risk entropy and conducting entropy weight evaluation. Bidders with comprehensive risk coefficients lower than the risk matching threshold are selected to generate a candidate pool of bidders. This solves the problems of not considering risk assessment and insufficient scenario adaptation in traditional management and control, and more intuitively reflects the comprehensive risk level of bidders and the degree of adaptation to project scenarios.

[0071] 2. This invention constructs a three-tiered risk verification rule based on three core dimensions: qualification, data security, and historical credit. It combines the sensitivity level and comprehensive risk coefficient of the target procurement project to dynamically adjust the initial verification thresholds at each level, conducting step-by-step risk verification. It selects compliant bidders from the bidder pool, and implements targeted rectification mechanisms for non-compliant bidders, thereby improving verification efficiency. In the risk audit phase, it achieves hierarchical control of abnormal risks through itemized scoring calibration and audit deviation calculation. It simultaneously updates the bidder risk characteristic database and generates dynamic risk monitoring reports, forming a cycle of screening, control, feedback, and optimization. This invention solves the problems of incomplete and lagging traditional compliance control processes, improving the efficiency and quality of bidder screening and project awarding, and strengthening risk prediction capabilities through dynamic updates and hierarchical control. Attached Figure Description

[0072] Figure 1 A schematic diagram of a risk intelligent identification method for a procurement system that integrates intelligent agents;

[0073] Figure 2 This is a flowchart illustrating the process of generating the candidate pool of bidders in this invention;

[0074] Figure 3 This is a flowchart illustrating the step-by-step risk verification process in this invention;

[0075] Figure 4 This is a flowchart for determining the level of abnormal risk in this invention;

[0076] Figure 5 Structure diagram of a risk intelligent identification system for a procurement system that integrates intelligent agents. Detailed Implementation

[0077] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0078] Example 1

[0079] refer to Figures 1 to 5 As shown in the figure, this embodiment introduces a risk intelligent identification method for a procurement system with integrated intelligent agents, including the following steps:

[0080] Multi-source heterogeneous data is acquired from industry regulatory platforms, bidder registration systems, and internal enterprise risk compliance databases. This data includes risk regulatory standards for the target field, bidder historical risk records, and bidding qualification documents for the target field. The acquired multi-source heterogeneous data undergoes preprocessing, such as deduplication, format standardization, and missing value completion. The integrity of the multi-source heterogeneous data is then verified. Through multi-agent collaboration, key features of the procurement business are extracted from the preprocessed and integrity-verified multi-source heterogeneous data to generate a bidder risk feature database. The multi-agent collaboration process includes: scheduling the preprocessing agent to perform format standardization and missing value completion, the feature extraction agent to extract key fields, and the verification agent to complete integrity verification and correlation. The risk mining agent extracts the equity relationships of bidders, historical joint bidding records, overlapping contact information, and temporal features of bidding behavior. After parallel processing by each agent, intermediate results are sent back to the master agent. Conflicting annotations of the same feature in different data sources are resolved to form a unified key feature output, generating a risk feature library for bidders. This library includes risk regulatory standards such as data privacy regulations and industry qualification requirements, historical risk records of bidders such as penalties and rectification records, and bidding qualification documents in the target field such as data processing licenses and industry access certificates. Key features include, but are not limited to, qualification validity, target field access suitability, frequency of risk penalties, overlap of related entities, frequency of historical joint bidding, and price deviation.

[0081] The disambiguation decision-making process includes: employing a weighted voting rule based on confidence level and data source authority. The master agent collects multiple candidate annotation results and confidence scores returned by each sub-agent for the same feature. Simultaneously, it determines the authority weight based on the historical accuracy of the data source, calculates the weighted confidence score of each candidate annotation result, and selects the annotation result with the highest weighted confidence score as the final output. If the difference between the highest and second-highest weighted confidence scores is less than the disambiguation threshold, the master agent is triggered to reschedule the verification agent to extract the original context information for the same feature. A second verification is performed, and the consistency of associated features is cross-validated by the associated risk mining intelligent agent. The matching degree of associated features is used as the auxiliary basis for disambiguation, and the result of the second verification is used as the final decision. This disambiguation mechanism, which combines weighted voting and second verification, solves the problem of inconsistent labeling of the same feature in multi-source heterogeneous data and improves the data quality of the bidder's risk feature library. The disambiguation threshold is determined by comprehensively setting the tolerance requirements of the bidder's risk feature library for data consistency and the historical verification accuracy. It can be set to a value of 0.1, which is used to trigger the second verification to ensure the reliability of the labeling when the discrimination is insufficient.

[0082] Key risk indicators, such as risk suitability in the target domain, are extracted from the bidder risk feature database to construct a bidder risk matrix. The domain risk entropy of key risk indicators is calculated using a risk weighting method to quantify the risk level of bidders in the target domain. Combined with the data sensitivity of the target procurement project, such as confidentiality level and privacy protection requirements, a comprehensive sensitivity value is calculated using a weighted quantification method to determine the sensitivity level of the target procurement project. The influence weight of each key risk indicator is dynamically adjusted, and the project risk entropy is calculated using a weighted quantification method. The risk simulation records of bidders in the sandbox are retrieved, such as the risk compliance rate of the target domain procurement project, to calculate robustness. The domain risk entropy, project risk entropy, and robustness are weighted and fused using a multi-objective optimization algorithm to generate a comprehensive risk coefficient. A risk matching threshold is set to conduct entropy weight evaluation. Bidders with a comprehensive risk coefficient lower than the risk matching threshold are selected to generate a bidder candidate pool.

[0083] Based on the risk specifications of the target domain, the data security level of the procurement project, and the comprehensive risk coefficient, a three-level risk verification rule is constructed. The first level verifies the completeness of the candidate bidder's qualifications, the second level verifies the data security suitability of the candidate bidder, and the third level verifies the candidate bidder's historical risk credit. Initial verification thresholds are set for each level. Based on the correlation analysis between the sensitivity level of the target procurement project and the comprehensive risk coefficient, the initial verification thresholds at each level are adjusted in real time. For example, for high-risk and high-sensitivity target procurement projects, the second-level initial verification threshold is automatically increased. Combining the three-level risk verification rule with the adjusted verification thresholds at each level, risk verification is carried out step by step. Candidate bidders that comply with the three-level risk verification are selected from the candidate bidder pool. For candidate bidders that do not comply with the verification at any level, a rectification mechanism is implemented.

[0084] For candidate bidders who pass the Level 3 risk verification, a risk audit score is calculated, and a risk audit is conducted. If the risk audit score is greater than or equal to the risk deviation threshold, the risk audit is deemed to have passed, a project-oriented award instruction is generated, and the target procurement project is distributed to the candidate bidder. Otherwise, the risk audit is deemed to have failed, the bidder is marked as an abnormal bidder, and the audit deviation is calculated to determine the abnormal risk level and trigger tiered control measures. For example, low-risk abnormalities trigger early warning prompts, medium-risk abnormalities suspend project awarding, and high-risk abnormalities remove candidate bidders. The bidder risk feature database is updated simultaneously, and a dynamic risk monitoring report containing audit results, control measures, and risk warnings is generated.

[0085] Specifically, the steps for generating the pool of bidder candidates include:

[0086] Key risk indicators for the target domain are extracted from the bidder risk feature database. Correlation analysis is used to calculate the correlation between each key risk indicator and the target domain risk indicators, yielding risk coefficients. The correlation analysis process includes: using the Pearson correlation coefficient algorithm to quantify the linear correlation between each key risk indicator and the target domain risk indicators, and normalizing the correlation coefficient to the range of 0-1 as the risk coefficient; setting a risk threshold based on the distribution of historical risk anomaly samples, with the value being the correlation coefficient at the 70th quantile of the historical risk anomaly sample distribution, to reduce the dimensionality of risk assessment calculations; and selecting key risk indicators with risk coefficients greater than the risk threshold to form an initial set of key risk indicators. The target domain risk indicators are obtained from historical risk audit reports, which are constructed by uniformly compiling historical bidder three-level risk verification results, anomaly classification and control records, and audit score data.

[0087] The specific calculation process of the Pearson correlation coefficient algorithm includes: extracting all observations of each key risk indicator in the historical risk sample, and all observations of the target domain risk indicator at the corresponding time within the same historical period, forming two sets of paired observation sequences; firstly, calculating the average value of each set of observation sequences, then subtracting the average value of each set of observation sequences from each observation value to obtain two sets of deviation value sequences; multiplying the two deviation values ​​at the same time in the two sets of deviation value sequences, and summing the products at all times to obtain the numerator; then squaring each deviation value in the two sets of deviation value sequences, summing them to obtain two sums of squares, multiplying the two sums of squares and taking the square root to obtain the denominator; finally, calculating the ratio of the numerator to the denominator to obtain the Pearson correlation coefficient between each key risk indicator and the target domain risk indicator, with a value range between -1 and 1; taking the absolute value of the Pearson correlation coefficient and linearly mapping it to the 0-1 interval to obtain the normalized correlation degree, which is used as the risk coefficient;

[0088] Based on the characteristics of different key risk indicators, the initial set of key risk indicators is normalized. For example, the validity of qualifications is a positive indicator, and the frequency of penalties is a negative indicator, thus forming a quantified set of key risk indicators.

[0089] Construct a risk matrix for bidders, with bidders as rows and key risk indicators as columns;

[0090] Based on information entropy theory, risk weighting factors are determined by acquiring the contribution of each key risk indicator in historical risk incident data. The process for determining the risk weighting factor is as follows: the contribution is a quantitative assessment value of the responsibility of each key risk indicator for causing the accident in the historical risk incident analysis; the historical risk incident data is traversed, the contribution of each key risk indicator is extracted, and the contribution of each key risk indicator is divided by the sum of the contributions of all key risk indicators. The resulting ratio is the risk weighting factor corresponding to each key risk indicator; the quantitative values ​​of key risk indicators of each bidder are extracted from the bidder risk matrix, and the domain risk entropy is calculated to quantify the bidder's risk in the target market. The risk level calculation process for the bidding area includes: first, calculating the proportion of each key risk indicator's value distribution among all bidders; substituting this proportion into the information entropy formula to obtain the single-indicator entropy value; the determination process for the single-indicator entropy value is as follows: for each key risk indicator, calculate the product of the proportion of its value distribution for each value interval and the logarithm of that proportion (base 2); sum the products corresponding to all value intervals of each key risk indicator and take the negative number to obtain the single-indicator entropy value for each key risk indicator; then multiply each single-indicator entropy value by the corresponding risk weighting factor and sum them to obtain the area risk entropy for each bidder; and finally, retrieve the top-ranked companies in the industry... The average domain risk entropy of the risk compliance bidders is defined as the benchmark entropy. If the domain risk entropy is greater than the benchmark entropy, the corresponding bidder is eliminated; otherwise, the corresponding bidder is retained. Determined based on the total number of bidders in the target field;

[0091] The system maps classification levels to quantitative classification values ​​and privacy protection requirements to quantitative privacy protection values ​​using expert scoring rules. The specific mapping process includes: security experts pre-determining a tiered scoring mapping rule based on risk and compliance requirements. Classification levels are mapped from low to high leakage risk to quantitative classification values ​​ranging from 0 to 1, and privacy protection requirements are mapped from low to high control intensity to quantitative privacy protection values ​​ranging from 0 to 1. During processing, the conversion from classification level to quantitative value is directly performed according to the tiered scoring mapping rule. The quantitative classification values ​​and quantitative privacy protection values ​​are then weighted to obtain a comprehensive sensitivity value. The weighting calculation process includes: allocating weights based on security control priority. Since the unauthorized leakage of classified data will trigger more severe compliance accountability and security impacts, classification control has a higher priority than privacy protection requirements. Therefore, the weight of the quantitative classification value is 0.6, and the weight of the quantitative privacy protection value is 0.4, with the sum of the two weights being 1. Finally, the quantitative classification value and the quantitative privacy protection value are multiplied by their respective weights and summed to obtain the comprehensive sensitivity value.

[0092] The sensitivity threshold is determined based on the distribution of sensitivity comprehensive values ​​of historical procurement projects. It is set at the 60th percentile of the distribution of sensitivity comprehensive values ​​of historical procurement projects and is used as the critical benchmark for dividing high sensitivity level and low sensitivity level in each sensitivity level determination to ensure that the determination result is consistent with the characteristics of historical risk distribution. The sensitivity level of the target procurement project is determined. If the sensitivity comprehensive value is greater than the sensitivity threshold, it is determined to be high sensitivity level; otherwise, it is determined to be low sensitivity level.

[0093] Based on the sensitivity level of the target procurement project, the influence weights of each key risk indicator are dynamically adjusted. Specifically, the weight coefficient corresponding to a high sensitivity level is used as the baseline weight for each key risk indicator. The weights of risk suitability and qualification validity are increased to 1.5 times the baseline weight, while the weight of penalty frequency is decreased to 0.5 times the baseline weight. For projects with low sensitivity levels, the baseline weights of each key risk indicator remain unchanged. Combining the bidder risk matrix, for the target procurement project, the normalized indicator values ​​of the bidders on each key risk indicator are extracted from the bidder risk matrix. Each normalized indicator value is multiplied by its corresponding influence weight after dynamic adjustment based on the sensitivity level to obtain the single-indicator risk contribution of each key risk indicator, which is the risk impact degree of each key risk indicator. Combined with the comprehensive sensitivity value, the project risk entropy is calculated using a weighted average. The weighted calculation process includes multiplying the risk impact degree of each key risk indicator by its corresponding influence weight after dynamic adjustment based on the sensitivity level, summing all products, and then multiplying by the comprehensive sensitivity value to obtain the project risk entropy. The larger the project risk entropy value, the higher the overall risk level of the target procurement project.

[0094] Based on the target procurement project scenario, such as data processing or qualification licensing business in the target field, sandbox simulation records are obtained from the industry risk sandbox shared database, and simulation scenario features are extracted. Target scenario features are extracted from the target procurement project scenario, and the first similarity between the simulation scenario features and the target scenario features is calculated. The calculation process includes extracting key feature tags from the target scenario, including business type, regulatory level, data sensitivity, and risk review frequency. Each key feature tag is compared item by item with the scenario feature tags of each sandbox simulation record, and the proportion of matching tags to the total number of tags is counted to obtain the first similarity. The industry risk sandbox shared database collects sandbox test cases, regulatory penalty notices, and risk review files from historical procurement projects in various industries. After anonymization, these are stored according to business fields and a scenario feature index is built.

[0095] A similarity threshold is set based on the risk assessment tolerance requirements. The lower the risk assessment tolerance requirements, the higher the strictness of scenario matching, and the larger the corresponding similarity threshold. Therefore, the similarity threshold is set to 0.7, which means that the simulated scenario and the target scenario need to reach more than 70% feature similarity before the corresponding sandbox verification conclusion can be adopted. When the first similarity is greater than the similarity threshold, the risk compliance compliance rate, risk anomaly recall rate, and assessment stability rate of the target domain procurement projects are selected from the sandbox simulation records.

[0096] The process for determining the risk compliance compliance rate is as follows: The number of test cases in the sandbox simulation records where all key risk indicators of the target domain's procurement projects are greater than or equal to the compliance threshold during the simulation test is statistically analyzed. The number of test cases is then divided by the total number of cases participating in the simulation test in the target domain, and the resulting ratio is the risk compliance compliance rate. The compliance threshold is determined based on the statistical results of the risk indicator compliance baseline of similar procurement projects in historical risk audit reports. Specifically, all procurement projects in the target domain that have passed risk audits within the past three years are extracted, and the actual values ​​of their key risk indicators at the audit nodes are calculated. The median of these actual values ​​is then taken as the compliance threshold.

[0097] The process for determining the risk anomaly recall rate is as follows: count the number of cases correctly identified as having risk anomalies in the sandbox simulation records, divide the number of cases by the total number of cases actually marked as having risk anomalies in the target domain simulation test data, and the resulting ratio is the risk anomaly recall rate.

[0098] The process for determining the stability rate is as follows: count the number of cases in the sandbox simulation records where the risk assessment conclusions of the target domain procurement projects remain consistent in multiple consecutive simulation tests, divide the number of cases by the total number of cases in the target domain participating in multiple simulation tests, and the resulting ratio is the stability rate.

[0099] The weights for risk compliance compliance rate, risk anomaly recall rate, and assessment stability rate are set according to the importance of the scenario. Specifically, if the scenario belongs to a heavily regulated area, such as data processing or licensing, risk compliance compliance is a fundamental requirement for participating in procurement, so the risk compliance compliance rate is given the highest weight (0.5). Risk anomaly recall reflects the completeness of risk event identification in the sandbox simulation, so its weight is next (0.3). Assessment stability rate reflects the consistency of results from multiple sandbox tests and has the lowest weight (0.2). If the scenario belongs to a high-dynamic-risk area, such as emerging businesses or cross-border services, risk anomaly recall reflects the ability to capture unknown risks, so its weight is given the highest weight (0.5). The weights of compliance rate and assessment stability rate decrease sequentially, taking values ​​of 0.3 and 0.2 respectively. If the scenario belongs to a regular business area, the importance of risk compliance rate, risk anomaly recall rate and assessment stability rate are relatively balanced, and the weight allocation can be one-third for each, and the sum of the weights of risk compliance rate, risk anomaly recall rate and assessment stability rate is always 1. The robustness is calculated by weighting the risk compliance rate, risk anomaly recall rate and assessment stability rate. The weighting calculation process is as follows: multiply the risk compliance rate by the risk compliance rate weight, the risk anomaly recall rate by the risk anomaly recall rate weight, and the assessment stability rate by the assessment stability rate weight. The three products are added together to obtain the robustness. The higher the robustness value, the more reliable the risk assessment conclusion under the current scenario based on historical sandbox verification.

[0100] Based on the robustness distribution statistics of the qualified samples in historical procurement projects, the 10th percentile of the robustness distribution is taken as the minimum acceptable benchmark for robustness and a qualified threshold of 0.6 is set for robustness assessment. If the robustness is greater than or equal to the qualified threshold, the robustness is deemed qualified and qualified bidders are retained; otherwise, the robustness is deemed unqualified and unqualified bidders are eliminated.

[0101] Based on the sensitivity level of the target procurement project, basic weights for domain risk entropy, project risk entropy, and robustness are set. At high sensitivity levels, the basic weights for domain risk entropy and project risk entropy are higher than the basic weight for robustness, emphasizing risk control as a priority; at low sensitivity levels, the basic weight for robustness is higher than the basic weights for domain risk entropy and project risk entropy, emphasizing efficiency and cost priority. The NSGA-II multi-objective optimization algorithm is adopted, with the risk optimization objective of minimizing overall risk and maximizing robustness. Pareto optimal weights for domain risk entropy, project risk entropy, and robustness are determined separately, and then weighted and fused together. The risk entropy and robustness are used to generate a comprehensive risk coefficient for each bidder. The Pareto optimal weight of the domain risk entropy is determined based on the historical proportion of domain risk incidents in the industry, and can be set to 0.4, representing a statistical result that historical procurement domain risk incidents account for 40% of all risk incidents. The Pareto optimal weight of the project risk entropy is determined based on the number of past risk anomaly records for the current project, and can be set to 0.3, representing a statistical result that single-project risk anomalies account for 30%. Since the sum of the Pareto optimal weights of the domain risk entropy, project risk entropy, and robustness is 1, the Pareto optimal weight of robustness is set to 0.3.

[0102] The computation process of the NSGA-II multi-objective optimization algorithm includes: initializing multiple sets of weight populations, iteratively performing selection, crossover, and mutation operations, calculating the comprehensive risk coefficient and robustness for each set of weight populations, eliminating inferior weight schemes through non-dominated stratification and crowding, and continuously iterating until the maximum number of iterations is reached, finally outputting a Pareto optimal weight combination without superiority or inferiority conflicts; the maximum number of iterations is determined according to the risk assessment accuracy requirements, the higher the risk assessment accuracy, the more iterations are required, and the value is 200.

[0103] A risk matching threshold is set based on the project's sensitivity level and the tenderer's risk appetite. The value is determined through a combination of expert review and statistical analysis of risk data from similar historical projects: for projects with high sensitivity levels, the risk matching threshold is set lower, for example, 0.4; for projects with low sensitivity levels, the risk matching threshold is set higher, for example, 0.7. The specific value of the risk matching threshold is specified in advance in the tendering plan. An entropy weight assessment is performed in conjunction with a comprehensive risk coefficient. If the comprehensive risk coefficient is greater than the risk matching threshold, the project is assessed as high risk and the corresponding bidder is eliminated; otherwise, it is assessed as low risk, the corresponding bidder is retained, and the bidders are sorted in ascending order of comprehensive risk coefficient to generate a candidate pool of bidders.

[0104] Specifically, the steps for conducting step-by-step risk verification include:

[0105] Based on the risk specifications of the target field and the data security level of the procurement project, a three-level risk verification rule is constructed. The first level verifies the completeness of the candidate bidder's qualifications, such as whether core qualifications are complete; the second level verifies the data security suitability of the candidate bidder, such as whether the security level matches the sensitivity level of the target procurement project; and the third level verifies the candidate bidder's historical risk credit, such as whether there are any industry-related records of dishonesty. Differentiated initial verification thresholds are set for each level, including a first-level initial verification threshold based on the qualification requirements stipulated in the procurement business, which is the minimum score for meeting the qualification completeness standard; a second-level initial verification threshold based on the distribution of historical security non-compliance cases, which is the minimum acceptable score for security compliance; and a third-level initial verification threshold based on the risk distribution of historical dishonest bidders, which is the highest acceptable threshold for dishonesty records.

[0106] The remaining validity period and total validity period of the qualification are calculated using a nonlinear decay function to determine the qualification validity score. The calculation process includes: first, calculating the ratio of the remaining validity period to the total validity period to obtain the remaining proportion; then, according to the principle that "the higher the remaining proportion, the higher the qualification validity score, and the qualification validity score declines rapidly as the expiration date approaches," the remaining proportion is mapped to the 0-100 score range to obtain the qualification validity score. Specifically, an exponential decay rule is used for mapping, so that when the remaining validity period is less than 20% of the total validity period, the qualification validity score drops sharply to reflect the risk that the qualification is about to expire.

[0107] The data security level of candidate bidders is obtained from the bidder security filing database. Combined with the sensitivity level of the target procurement project, a gain coefficient is set based on the difference between the quantified data security level and the quantified sensitivity level. The smaller the difference, the larger the gain coefficient, with a selectable value range of 0.6 to 1.2. A weighted security fit score is calculated. The weighted calculation process includes: quantifying the bidder's data security level into a basic security score, calculating the product of the basic security score and the gain coefficient, and obtaining the security fit score. The bidder security filing database is constructed by uniformly collecting and classifying the risk assessment reports, data compliance filing materials, and security qualification certificate information of each bidder.

[0108] Penalty records are extracted from the bidder risk characteristic database and quantified according to severity, such as a warning deducting 1 point and a fine deducting 3 points. The results are then used to compile statistics on candidate bidders' penalties. The frequency of penalties within a given time period is used to calculate historical credit scores using a negative exponential function. The calculation process includes: summing the weighted total deductions from all penalty records; using this weighted total deduction as the exponential input; and mapping the weighted total deduction to a credit score range of 0-100 points using a negative exponential function to obtain the historical credit score. Fewer deductions result in a historical credit score closer to full marks, while more deductions lead to a decreasing historical credit score. In this embodiment, the following settings are provided: Months;

[0109] The sensitivity level of target procurement projects is quantified into sensitivity level coefficients using industry sensitivity level quantification standards, including high sensitivity level coefficients and low sensitivity level coefficients. Since projects with different sensitivity levels have fundamentally different risk tolerances for bidders' qualifications, security, and creditworthiness, and the overall project risk coefficient can further amplify or weaken the actual impact of the sensitivity level, it is necessary to adjust the initial verification thresholds at each level. Based on the sensitivity level coefficient and the overall risk coefficient, the initial verification thresholds at each level are adjusted in real time. The specific process of real-time adjustment includes: for the first-level initial verification threshold, multiplying the sensitivity level coefficient by the overall risk coefficient to obtain the project's overall risk multiplier; dividing the project's overall risk multiplier by the product of the baseline sensitivity level coefficient and the baseline risk coefficient to obtain the risk adjustment factor; and calculating the product of the first-level initial verification threshold and the risk adjustment factor to obtain the adjusted first-level verification threshold. The minimum score for qualification integrity is adjusted in the same direction as the project risk level. The benchmark sensitivity level coefficient is determined based on the value corresponding to the medium sensitivity level in the industry's general data sensitivity level classification standard, and the benchmark risk coefficient is determined based on the historical risk level statistical average of similar conventional procurement projects. For the secondary initial verification threshold, the corrected secondary verification threshold is obtained by multiplying it by the risk adjustment factor. For the tertiary initial verification threshold, since it represents the highest acceptable threshold for breaches of trust, a reverse correction is adopted. This is achieved by calculating the ratio of the tertiary initial verification threshold to the risk adjustment factor. Based on the qualification validity score, security fit score, and historical credit score, combined with the tertiary risk verification rules, a step-by-step risk verification is conducted.

[0110] If the valid qualification score is greater than or equal to the Level 1 verification threshold, the Level 1 verification is deemed compliant, and candidate bidders compliant with Level 1 verification are selected from the bidder candidate pool, triggering Level 2 risk verification; otherwise, the Level 1 verification is deemed non-compliant, and the qualification supplementation and rectification mechanism is implemented, meaning the candidate bidder... The bidder must submit valid qualification documents within a specified time; failure to do so will result in removal from the bidder candidate pool. This embodiment specifies... 2 business days;

[0111] For candidate bidders who pass the Level 1 verification, if their security compatibility score is greater than or equal to the Level 2 verification threshold, they are deemed compliant with Level 2 verification. Candidate bidders compliant with Level 2 verification are then selected from the candidate bidder pool, and Level 3 risk verification is triggered. Otherwise, they are deemed non-compliant with Level 2 verification, and the data security remediation mechanism is implemented, meaning the candidate bidder... The security protection upgrade is completed within a specified time, such as access permission optimization. After the rectification is completed, the application for secondary risk verification is resubmitted. This embodiment sets... 5 business days;

[0112] For candidate bidders who meet the Level 2 verification, if their historical credit score is greater than or equal to the Level 3 verification threshold, they are deemed to meet the Level 3 risk verification. Candidate bidders who meet the Level 3 risk verification are selected from the candidate bidder pool and ranked in ascending order according to their comprehensive risk coefficient to determine the first candidate bidder list.

[0113] Otherwise, the Level 3 verification will be deemed non-compliant, and the corresponding rectification mechanism will be implemented. This will involve organizing online risk-specific training and completing the assessment, recalculating the historical credit score, and determining whether the Level 3 risk verification is compliant. If it continues... If the Level 3 risk verification fails to meet the requirements, the corresponding candidate bidder will be disqualified from participating and added to the risk blacklist in the target field; among them, The specific size can be set by those skilled in the art according to actual needs, and this embodiment does not limit it.

[0114] Specifically, the steps for generating a dynamic risk monitoring report include:

[0115] Basic risk information of bidders is obtained from the bidder qualification filing system and security risk reporting platform. Historical feature data is extracted from the bidder risk feature database. Based on the first candidate bidder list, and using the candidate bidder ID as the index, the key information in the first candidate bidder list is associated with the basic risk information and historical feature data of the bidders to form a bidder risk information set. The key information in the first candidate bidder list includes, but is not limited to, the name of the candidate bidder and its business scope.

[0116] Based on the bidder's risk information set, a segmented scoring calibration is performed. In the qualification dimension, based on the valid qualification score, real-time valid qualification scores are obtained from the industry qualification verification platform. Adding a new core qualification adds 2 points, resulting in a calibrated qualification score. In the data security dimension, based on the security adaptation score and adjusted according to the sensitivity level of the target procurement project, 6 points are deducted if security protection is not implemented for a highly sensitive target procurement project, and 4 points are added for completing encryption algorithm iterations, resulting in a calibrated security score. In the historical credit dimension, using the historical credit score as a baseline, a calibrated credit score is obtained by deducting points from corresponding violation records. For example, deducting... The score is calculated based on newly added violation records within a given time period. The risk audit score is calculated by weighted summation based on calibration qualification score, calibration security score, and calibration credit score. The weighted summation process includes: allocating dimension weights according to the risk control priority of the procurement project, with data security having the highest priority, followed by qualification and historical credit dimensions. Therefore, the weight of the data security dimension is set to 0.4, and the weights of the qualification and historical credit dimensions are both set to 0.3. The calibration qualification score is multiplied by its weight, the calibration security score by its weight, and the calibration credit score by its weight, and then summed to obtain the risk audit score. In this embodiment, the following settings are used: Months;

[0117] Risk audit thresholds are set based on the sensitivity level and risk control requirements of the target procurement projects. For high-sensitivity projects involving core business data, the impact of risk leakage and violations is wide-ranging and profound; therefore, the risk audit threshold for high-sensitivity projects is set at 75 points. For low-sensitivity projects with lower risk and relatively relaxed risk control requirements, the risk audit threshold is set at 50 points. A risk audit is then conducted. If the risk audit score is greater than or equal to the risk audit threshold, the risk audit is considered passed, and a second-tier candidate list of bidders is obtained; otherwise, the risk audit is considered failed, and the bidder is marked as an abnormal bidder.

[0118] Iterate through the second-tier bidder candidate list, filtering by risk audit score in ascending order. The system identifies candidate bidders, integrates the target procurement project content and risk constraints (such as data encryption requirements), and generates a project-specific award instruction. This instruction includes, but is not limited to, the target procurement project ID, candidate bidder ID, and risk accountability clauses. The instruction is then pushed to these candidates through a risk management platform. The number of bidders required for the target procurement project is determined by this number. For example, if the target procurement project requires two bidders, then... ;

[0119] For bidders with unusual behavior, the audit deviation is calculated based on the risk audit score and the risk audit threshold. The calculation process includes: calculating the absolute difference between the risk audit score and the risk audit threshold, and dividing the absolute difference by the risk audit threshold to obtain the audit deviation. Set a risk deviation threshold based on the ratio of the risk audit score to the risk audit threshold. , ,and Determine the level of abnormal risk; if If it is, it is judged as a low-risk anomaly; if If so, it is judged as a medium-risk abnormality; if If so, it is judged as a high-risk anomaly;

[0120] Based on the level of abnormal risk, tiered control is triggered. For low-risk abnormalities, the warning status is recorded in the bidder risk feature database and a risk alert is pushed. For medium-risk abnormalities, the project award is suspended and rectification materials are submitted, such as supplementary qualification certificates and security protection upgrade plans. For high-risk abnormalities, the candidate bidder is removed from the second list of candidate bidders, and the final list of bidders is determined.

[0121] In the bidder risk feature database, add risk audit pass status, related project award records, update abnormal risk level and control measures to obtain the updated bidder risk feature database;

[0122] Integrate risk audit findings, abnormal risk levels, tiered control implementation records, bidder lists, and updated bidder risk characteristic databases to generate dynamic risk monitoring reports that include audit results, control measures, and risk warnings.

[0123] Example 2

[0124] Please see Figure 5 Another embodiment of the present invention provides a risk intelligent identification system for a procurement system that integrates intelligent agents, comprising: a risk feature construction module, a risk entropy weight assessment module, a risk verification module, and a risk audit module;

[0125] The risk feature construction module is used to acquire multi-source heterogeneous data and extract key features of the bidding and procurement business from the pre-processed and integrity-verified multi-source heterogeneous data through multi-agent collaboration to generate a risk feature library for bidders.

[0126] The risk entropy weight assessment module is used to extract key risk indicators from the bidder risk feature library, construct the bidder risk matrix, calculate the domain risk entropy of key risk indicators, combine the data sensitivity quantification of the target procurement project to calculate the sensitivity comprehensive value, determine the sensitivity level of the target procurement project, dynamically adjust the influence weight of each key risk indicator to calculate the project risk entropy, retrieve the risk simulation records of the bidder in the sandbox to calculate robustness, and use a multi-objective optimization algorithm to weight and fuse the domain risk entropy, project risk entropy and robustness to generate a comprehensive risk coefficient for entropy weight assessment and generate a bidder candidate pool.

[0127] The risk verification module is used to construct three levels of risk verification rules. The first level verifies the integrity of qualifications, the second level verifies the data security adaptability, and the third level verifies the historical risk credit. It sets the initial verification thresholds for each level, and adjusts the initial verification thresholds for each level in real time according to the sensitivity level and comprehensive risk coefficient. It performs risk verification step by step and selects candidate bidders that meet the three-level risk verification from the candidate bidder pool.

[0128] The risk audit module is used to calculate risk audit scores for candidate bidders who meet the three-level risk verification criteria, and to determine the risk level of the audit. If the risk audit score is greater than or equal to the risk deviation threshold, the risk audit is deemed to have passed, a project-oriented award instruction is generated, and the target procurement project is distributed to the candidate bidder. Otherwise, the risk audit is deemed to have failed, the bidder is marked as an abnormal bidder, the audit deviation is calculated to determine the abnormal risk level, the hierarchical control is triggered, the bidder risk feature database is updated synchronously, and a dynamic risk monitoring report is generated.

[0129] Working principle and effects:

[0130] The system acquires heterogeneous data from multiple sources. The master control agent breaks down the data into structured, semi-structured, and unstructured sub-tasks. Preprocessing agents are scheduled to perform noise filtering, format standardization and missing value completion, feature extraction agents to extract key fields, and verification agents to perform integrity verification. After parallel processing, the agents return intermediate results. The master control agent identifies conflicting annotations of the same feature in different data sources and makes a preliminary decision using a weighted voting rule based on confidence and data source authority. If the difference between the highest and second-highest votes is less than the disambiguation threshold, the verification agent is triggered for secondary verification, and the associated risk mining agent is linked for cross-validation to eliminate discrepancies in multi-source annotations and ensure the basic data quality of the feature library. Finally, key features such as qualification validity, target domain access suitability, and risk penalty frequency are extracted to generate a risk feature library for bidders. This multi-agent collaborative mechanism replaces the traditional manual extraction and single data source processing methods, eliminating data silos, format differences, and information noise interference.

[0131] Key risk indicators are extracted from the bidder risk feature database, and a bidder risk matrix is ​​constructed. Domain risk entropy is calculated to quantify the bidder's compliance risk level in the target domain. Combined with the confidentiality level and privacy protection requirements of the target procurement project, a weighted comprehensive sensitivity value is calculated and the sensitivity level is determined. Based on this, the influence weight of each key risk indicator is dynamically adjusted and the project risk entropy is calculated, so that the risk assessment results adapt to the sensitivity of the project. The robustness of the bidder's risk simulation records in the sandbox is calculated. The NSGA-II multi-objective optimization algorithm is used to weight and fuse the domain risk entropy, project risk entropy, and robustness to generate a comprehensive risk coefficient and perform entropy weight evaluation to form a bidder candidate pool. This breaks through the subjective limitations of traditional static weighting and achieves dynamic adaptation of risk assessment and procurement scenarios.

[0132] Based on the risk specifications of the target field and the data security level of the procurement project, and combined with the comprehensive risk coefficient, a three-level risk verification rule is constructed, which includes qualification integrity, data security adaptability, and historical risk credit. Initial verification thresholds are set for each level, and the initial verification thresholds for each level are adjusted in real time according to the sensitivity level and the comprehensive risk coefficient. The three levels of compliant candidate bidders are verified and screened step by step, and differentiated rectification measures such as qualification supplementation, security upgrade or credit training are triggered for non-compliance at each level. This ensures that any bidder without compliant qualifications cannot pass the complete verification process, and avoids the termination of the entire process due to a single dimension problem.

[0133] For candidate bidders who pass the three-level verification, a risk audit score is calculated by weighting the scores from the dimensions of qualification, data security, and historical credit. This score is then compared with the risk deviation threshold to determine whether the audit passes. If the audit passes, a project-oriented award instruction is generated and distributed to the corresponding bidder. If the audit fails, the bidder is marked as an abnormal bidder, and the audit deviation is calculated to determine the abnormal risk level. This triggers hierarchical control and updates the bidder risk feature database simultaneously, generating a dynamic risk monitoring report. This forms a closed-loop system with multi-agent collaborative identification, dynamic entropy weight assessment, three-level progressive verification, and hierarchical audit control. This ensures that no bidder that has not passed the three-level risk verification can be awarded a project, avoiding procurement compliance risks and data security vulnerabilities caused by distributing projects to unqualified bidders without passing the three-level risk verification.

[0134] Overall, through a four-layer modular architecture of risk feature construction, risk entropy weight assessment, risk verification, and risk auditing, and with a multi-agent collaborative mechanism running through the entire process of data acquisition, feature extraction, risk assessment, risk compliance verification, and audit control, this approach combines multi-source data fusion, dynamic risk assessment, hierarchical risk compliance verification, risk audit judgment, and graded control. This enables proactive identification and prevention of bidder risks throughout the entire procurement process, solving the technical problems of traditional methods such as strong subjectivity in risk quantification, incomplete risk compliance verification processes, lagging monitoring, and poor scenario adaptability. In particular, by integrating a multi-agent collaborative risk identification mechanism with the rigid constraints of multi-level risk verification, it solves the problem of the system distributing projects to unqualified bidders without multi-level risk verification.

[0135] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A risk intelligent identification method for a procurement system integrating intelligent agents, characterized in that, include: By leveraging multi-agent collaboration, key features of the procurement process are extracted, generating a risk feature database for bidders. Extract key risk indicators, construct a risk matrix for bidders, calculate domain risk entropy, quantify the data sensitivity of target procurement projects, calculate the comprehensive sensitivity value to determine the sensitivity level, calculate project risk entropy and robustness, combine domain risk entropy to generate a comprehensive risk coefficient, conduct entropy weight assessment, and generate a candidate pool of bidders. A three-tiered risk verification rule is established, which verifies the risk level by level based on the integrity of qualifications, data security compatibility and historical risk credit, and selects candidate bidders who comply with the three-tiered risk verification. For candidate bidders who meet the Level 3 risk verification, a risk audit score is calculated, and a risk audit judgment is made. If the audit fails, the audit deviation is calculated, the abnormal risk level is determined, and hierarchical control is triggered. All intelligent agents work together to generate a dynamic risk monitoring report.

2. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 1, characterized in that, The specific steps for determining the sensitivity level include: Extract key risk indicators from the target area and calculate risk coefficients; Key risk indicators with risk coefficients greater than a preset risk threshold are selected, and after normalization, a set of key risk indicators is formed. Construct a risk matrix for bidders, obtain risk weighting factors, extract quantitative values ​​of key risk indicators for each bidder, and calculate domain risk entropy. In the top ranking The average domain risk entropy of the risk compliance bidders is the benchmark entropy. If the domain risk entropy is greater than the benchmark entropy, the corresponding bidder is eliminated. The classification level is mapped to a classification quantification value, and the privacy protection requirement is mapped to a privacy protection quantification value. A weighted comprehensive sensitivity value is calculated, and the sensitivity level is determined. If the overall sensitivity value is greater than the preset sensitivity threshold, it is determined to be at a high sensitivity level; otherwise, it is determined to be at a low sensitivity level.

3. The intelligent risk identification method for a procurement system based on a fusion of intelligent agents according to claim 2, characterized in that, The specific steps for generating the bidder candidate pool include: The impact weights of each key risk indicator are dynamically adjusted based on the sensitivity level. Based on the bidder's risk matrix, the risk impact of each key risk indicator is calculated, and the project risk entropy is calculated by weighting the sensitivity comprehensive value. Obtain sandbox simulation records, extract features from the simulated scene and calculate the first similarity between the simulated scene features and the target scene features; When the first similarity is greater than the preset similarity threshold, the risk compliance rate, risk anomaly recall rate, and assessment stability rate are selected, and the robustness is calculated by weighting and performing a robustness assessment. If the robustness is greater than or equal to the preset pass threshold, the robustness is deemed to be qualified, and the qualified bidder is retained; otherwise, the robustness is deemed to be unqualified, and the unqualified bidder is eliminated.

4. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 3, characterized in that, The specific steps for generating the bidder candidate pool also include: Set the basic weights for domain risk entropy, project risk entropy, and robustness respectively; The NSGA-II multi-objective optimization algorithm is used to determine the Pareto optimal weights for domain risk entropy, project risk entropy, and robustness, respectively. The risk entropy of the domain, the risk entropy of the project, and the robustness are weighted and integrated to generate a comprehensive risk coefficient for each bidder, and then an entropy weight assessment is performed. If the overall risk coefficient is greater than the preset risk matching threshold, it is assessed as high risk and the corresponding bidder is eliminated. Otherwise, if the risk is assessed as low, the corresponding bidder is retained and sorted in ascending order according to the comprehensive risk coefficient to generate a candidate pool of bidders.

5. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 4, characterized in that, The specific steps for constructing a three-level risk verification rule include: Based on the compliance standards of the target field and the data security level of the procurement project, a three-level risk verification rule is constructed, including the first level verifying the integrity of the candidate bidder's qualifications, the second level verifying the data security compatibility of the candidate bidder, and the third level verifying the candidate bidder's historical risk credit. Set initial verification thresholds for each level, including the first-level initial verification threshold, the second-level initial verification threshold, and the third-level initial verification threshold.

6. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 5, characterized in that, The specific steps for conducting step-by-step risk verification include: Calculate the qualification validity score, security adaptation score, and historical credit score; The initial verification thresholds at each level are corrected in real time to obtain the first-level verification threshold, the second-level verification threshold, and the third-level verification threshold, and risk verification is performed step by step. When the valid qualification score is greater than or equal to the first-level verification threshold, the first-level verification is deemed compliant, candidate bidders compliant with the first-level verification are screened, and the second-level risk verification is triggered. For candidate bidders who meet the Level 1 verification requirements; When the security compatibility score is greater than or equal to the level 2 verification threshold, the level 2 verification is deemed compliant, candidate bidders who meet the level 2 verification are selected, and level 3 risk verification is triggered. For candidate bidders who meet the Level 2 verification requirements; When the historical credit score is greater than or equal to the Level 3 verification threshold, the Level 3 risk verification is deemed compliant. Candidate bidders who meet the Level 3 risk verification criteria are then selected and ranked in ascending order of comprehensive risk coefficient to determine the first candidate bidder list.

7. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 6, characterized in that, The specific steps for conducting a risk audit assessment include: Obtain basic risk information of bidders and extract historical feature data; By associating key information from the first candidate bidder list, basic risk information of bidders, and historical characteristic data, a bidder risk information set is formed, and itemized scoring and calibration are performed. In terms of qualifications, obtain real-time valid qualification scores, and calculate and calibrate qualification scores based on the valid qualification scores. In terms of data security, the security adaptation score is adjusted according to the sensitivity level to obtain the calibrated security score. In terms of historical credit, the historical credit score is used as the benchmark value, and the calibrated credit score is obtained by deducting the corresponding violation record points. Based on calibration qualification score, calibration safety score and calibration credit score, a risk audit score is calculated by weighting and a risk audit judgment is made. If the risk audit score is greater than or equal to the preset risk audit threshold, the risk audit is deemed to have passed, and a second list of candidate bidders is obtained; otherwise, the risk audit is deemed to have failed, and the bidder is marked as an abnormal bidder.

8. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 7, characterized in that, The specific steps for determining the level of abnormal risk include: Iterate through the second-tier bidder candidate list, filtering by risk audit score in ascending order. The list of candidate bidders; Integrate the content and risk constraints of the target procurement projects to generate targeted bidding instructions and push them to specific parties; For bidders with unusual behavior, the audit deviation is calculated based on the risk audit score and the risk audit threshold. ; Set risk deviation threshold , ,and Determine the level of abnormal risk; like If it is, it is judged as a low-risk anomaly; if If so, it is judged as a medium-risk abnormality; if If so, it is judged as a high-risk anomaly.

9. The intelligent risk identification method for a procurement system based on a fusion intelligent agent according to claim 8, characterized in that, The specific steps for generating a dynamic risk monitoring report include: Based on the level of abnormal risk, tiered control measures are triggered. For low-risk anomalies, record the warning status in the bidder's risk characteristic database and push risk alerts; For projects with abnormal medium-risk conditions, project awarding will be suspended, and rectification materials will be submitted. For high-risk anomalies, remove candidate bidders from the second-tier candidate list and finally determine the bidder list; In the bidder risk feature database, add risk audit pass status, related project award records, update abnormal risk level and control measures to obtain the updated bidder risk feature database; Integrate risk audit findings, abnormal risk levels, hierarchical control implementation records, bidder list, and updated bidder risk characteristic database; Generate dynamic risk monitoring reports.

10. A risk intelligent identification system for a procurement system integrating intelligent agents, used to implement the risk intelligent identification method for a procurement system integrating intelligent agents as described in any one of claims 1-9, characterized in that, include: The system includes a risk characteristic construction module, a risk entropy weight assessment module, a risk verification module, and a risk audit module. The risk feature construction module is used to extract key features of the bidding and procurement business through multi-agent collaboration and generate a risk feature library of bidders. The risk entropy weight assessment module is used to extract key risk indicators, construct a risk matrix for bidders, calculate domain risk entropy, quantify the data sensitivity of target procurement projects, calculate a comprehensive sensitivity value to determine the sensitivity level, calculate project risk entropy and robustness, generate a comprehensive risk coefficient by combining domain risk entropy, conduct entropy weight assessment, and generate a candidate pool of bidders. The risk verification module is used to construct a three-level risk verification rule, and to perform step-by-step risk verification based on the integrity of qualifications, data security adaptability and historical risk credit, and to screen candidate bidders who comply with the three-level risk verification. The risk audit module is used to calculate the risk audit score for candidate bidders who meet the three-level risk verification, make a risk audit judgment, and calculate the audit deviation degree if they fail to pass the audit, determine the abnormal risk level and trigger hierarchical control. All intelligent agents work together to summarize and generate a dynamic risk monitoring report.

Images