Intelligent Compliance Judgment System for Procurement Business Based on Feature Vectors

By constructing a feature vector-based intelligent compliance judgment system for procurement operations, the problems of fragmented and outdated data and delayed judgment results in the procurement compliance management of group enterprises have been solved. This system enables real-time compliance control and risk identification throughout the entire procurement process, improving the accuracy and closed-loop capability of management.

CN123117785APending Publication Date: 2026-11-17SUZHOU ENERGY TECHNOLOGY DEVELOPMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611286815.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The procurement compliance management of existing group enterprises lacks unified big data resource service support. Data is scattered and fragmented, making it difficult to link and verify. Compliance judgment relies on single rule verification, which cannot combine multi-dimensional features to achieve scenario-based comprehensive judgment. The risk omission rate is high and the adaptability is poor. At the same time, the compliance judgment results are not linked enough with business processes, making it difficult to meet the compliance requirements of closed-loop management of the entire procurement process of group enterprises.

Method used

A compliance intelligent judgment system for procurement business based on feature vectors is constructed, including a big data resource service module, a feature vector construction module, a compliance intelligent judgment module, a business process linkage module, and a visualization control module. Through a unified microservice architecture, multi-source procurement business data is integrated to generate compliance judgment feature vectors. Combined with rule verification and intelligent model judgment, a comprehensive assessment and level classification of compliance risks is achieved, and the procurement business process is linked in real time.

Benefits of technology

It enables the coordinated governance and service-oriented access to data across the entire procurement chain, improving the comprehensiveness and accuracy of risk identification, meeting the closed-loop management needs of the entire group's procurement process, and reducing the cost of manual verification and audit compliance risks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117785A_ABST
    Figure CN123117785A_ABST
Patent Text Reader

Abstract

This invention discloses a feature vector-based intelligent compliance judgment system for procurement operations, relating to the field of procurement compliance judgment technology. It includes: a big data resource service module, a feature vector construction module, a compliance intelligent judgment module, a business process linkage module, and a visualization and control module. This invention integrates multi-source procurement business data by constructing a unified big data resource service module, enabling linked governance and service-oriented access to procurement data across the entire chain, thus solving the problem of fragmented and scattered data. The feature vector construction module extracts multi-dimensional procurement business features and completes vectorized mapping, providing a standardized data foundation for compliance judgment. Combined with the compliance intelligent judgment module, it achieves comprehensive judgment through the fusion of rule verification and model judgment, adapting to the differentiated compliance requirements of different procurement scenarios and improving the comprehensiveness and accuracy of risk identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of procurement business compliance discrimination, and specifically relates to an intelligent procurement business compliance discrimination system based on feature vectors. Background Art

[0002] Procurement business is a core business link for enterprises to acquire materials, projects and services in their production and operation activities, covering all process nodes such as demand submission, plan formulation, procurement execution, contract signing, performance acceptance, and financial settlement, and serves as an important foundation for enterprises to control operating costs, ensure supply chain stability and support business development. For group-type enterprises, procurement business covers multi-level organizations, multi-category targets and multi-participant entities, and its management efficiency and compliance level are directly related to the overall operational efficiency and risk prevention and control capability of the group. Intelligent procurement business compliance discrimination is a technical means that uses digital technology to automatically verify the operation standardization, subject qualification validity and transaction behavior rationality of each business node in the whole procurement process and identify risks. It can break through the limitations of traditional manual post-audit and sampling inspection, realize full-process coverage, real-time verification and pre-warning of compliance control, and has important application value for standardizing procurement operation procedures, preventing integrity risks and reducing enterprise audit compliance costs.

[0003] Feature vector technology is a core method for realizing standardized representation of multi-source heterogeneous data in the big data resource service system, which can convert multi-dimensional unstructured and semi-structured information such as scattered process data, subject data, transaction data and budget data in procurement business into quantifiable and computable structured vector features, provide a unified input standard for the compliance discrimination model, and serve as a key technical support for improving the accuracy, processing efficiency and scenario adaptability of compliance discrimination.

[0004] Existing procurement compliance management of group-type enterprises is mostly carried out relying on the approval nodes of decentralized business systems and manual post-verification methods. Procurement-related data is stored dispersedly in office automation systems, financial systems, procurement execution systems and supplier management systems, and a unified big data resource service system has not been built to realize integrated governance and linked invocation of data, making it difficult to perform cross-verification of the full-link procurement data. Compliance discrimination mostly relies on rule-based verification with fixed thresholds, and cannot combine business scenarios to extract multi-dimensional features for comprehensive discrimination, which not only has the problem of missed risk judgment caused by incomplete rule coverage, but also is difficult to adapt to the differentiated compliance requirements of different procurement categories and different procurement methods. At the same time, data such as supplier qualifications and procurement process nodes have not been subjected to feature-based unified management, and the compliance discrimination results cannot be linked with the procurement business process in real time, leading to relatively lagging risk disposal, which is difficult to meet the compliance management requirements of full-process closed-loop control for group-based procurement. Therefore, it is of great significance to develop an intelligent procurement business compliance discrimination system based on feature vectors. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a feature vector-based intelligent compliance judgment system for procurement business. It can solve the technical problems of existing procurement compliance management lacking unified big data resource service support, fragmented data making it difficult to link and verify, compliance judgment relying on single rule verification, inability to combine multi-dimensional features to achieve scenario-based comprehensive judgment, high risk omission rate and poor adaptability, and insufficient linkage between compliance judgment results and business processes, resulting in delayed risk handling and difficulty in meeting the compliance requirements of closed-loop management of the entire procurement process of a group.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a procurement business compliance intelligent judgment system based on feature vectors, the system comprising: a big data resource service module, a feature vector construction module, a compliance intelligent judgment module, a business process linkage module, and a visualization control module; The big data resource service module is used to connect to multi-source heterogeneous procurement business systems, collect and manage procurement data across the entire chain, and provide standardized data calling services for the feature vector construction module. The feature vector construction module is used to retrieve the full-dimensional related data corresponding to the procurement business to be judged, extract multi-dimensional compliance features and complete the standardization process, generate a compliance judgment feature vector and transmit it to the compliance intelligent judgment module. The compliance intelligent judgment module is used to receive compliance judgment feature vectors, combine the rule base and the judgment model to complete compliance verification and risk classification, and output the judgment results to the business process linkage module and the visualization control module respectively. The business process linkage module is used to connect with the procurement business execution node, receive the judgment result and trigger the corresponding process operation, and send the processing result back to the big data resource service module. The visualization and control module is used to receive judgment results and processing data, and to provide compliance data display and report output services.

[0007] The system is deployed using a microservice architecture. Each module achieves data interaction and business decoupling through standardized API interfaces. It supports on-demand expansion of the computing resources of a single module and can adapt to the procurement business volume and concurrent access requirements of group enterprises of different sizes. It has good scalability and deployment flexibility.

[0008] Furthermore, the big data resource service module includes a multi-source data access unit, a master data governance unit, a data storage unit, and a data service interface unit. The multi-source data access unit is equipped with standardized interfaces for connecting to office automation systems, financial control systems, transparent procurement systems, supplier management platforms, e-commerce procurement platforms, and external enterprise credit platforms, collecting procurement needs, procurement plans, supplier qualifications, bidding information, contract information, financial reconciliation, guarantee information, and entity credit data. The master data governance unit performs standardized coding and cross-system mapping processing on material master data, supplier master data, organizational structure master data, and procurement category master data. The data storage unit is equipped with a business detail storage area, a master data storage area, and a compliance result storage area, storing corresponding data hierarchically. The data service interface unit is equipped with calling interfaces categorized according to business scenarios and data dimensions.

[0009] The data governance process incorporates multi-dimensional data verification rules to automatically clean and logically complete missing and outlier values ​​collected. The master data encoding adopts a unified segmented encoding rule, corresponding to a three-segment structure of category level, attribute dimension, and serial number, ensuring the uniqueness and accuracy of cross-system data mapping and reducing data errors in subsequent compliance judgments from the data source level.

[0010] Furthermore, the multi-source data access unit is equipped with two data acquisition modes: timed incremental synchronization and real-time triggered synchronization, corresponding to the data update frequency of different business systems. The master data governance unit sets up a cross-system data mapping table to store the encoding correspondence of the same type of master data between different systems. The data storage unit sets up a historical data archiving cycle and performs archiving processing on business detail data that exceeds the time limit. The data service interface unit sets up a call identity verification mechanism and only opens the data retrieval channel with corresponding permissions to authorized modules.

[0011] Furthermore, the feature vector construction module performs the following operations when generating compliance discrimination feature vectors: The big data resource service module retrieves full-dimensional related data for the corresponding procurement business scenario, and extracts corresponding indicators for five categories of feature items: process node features, entity qualification features, transaction behavior features, budget matching features, and risk association features. Discretization and normalization are performed on all extracted feature indicators to convert qualitative features into quantitative scores and map quantitative features of different dimensions to a unified numerical range. The standardized feature indicators are arranged according to a preset feature dimension sorting rule, and a fixed-dimensional compliance discrimination feature vector is generated through a weighted mapping operation. The weighted mapping operation formula is as follows: In the formula To ensure compliance in feature vector identification, The initial feature vector after standardization. The feature weight vector is for dimension matching. ⊙ represents the Hadamard product operation. The feature weight vector is determined by combining the feature contribution of historical compliance risk samples with the analytic hierarchy process. The weight value of each dimension corresponds to the contribution ratio of this type of feature in risk identification. The sum of all weight values ​​is 1.

[0012] The generated feature vector dimensions correspond one-to-one with the procurement business scenario. Different procurement categories can directly call the exclusive feature weight configuration without reconstructing the feature extraction logic, which greatly improves the efficiency of scenario adaptation. The feature standardization process adopts the min-max normalization method to uniformly map all feature values ​​to the [0, 1] interval, completely eliminating the interference of the difference in the scale between different indicators on the subsequent judgment results, and ensuring the computability and comparability of the feature vector.

[0013] Furthermore, the compliance intelligent judgment module performs the following operations when executing compliance judgment: The built-in procurement compliance rule library is invoked to perform item-by-item matching and verification on the rule-class feature items in the input compliance judgment feature vector, outputting the pass and fail items of the rule verification and generating the rule verification score; The compliance discrimination feature vector is input into the trained intelligent discrimination model, and the model outputs the compliance probability and hidden risk points of the corresponding procurement business through model calculation, generating the model discrimination score; The comprehensive compliance risk score is calculated, and the compliance risk implementation level is classified according to the risk level classification standard. Corresponding risk descriptions and handling guidelines are then generated. The formula for calculating the comprehensive compliance risk score is as follows: Where S represents the overall compliance risk score. For rule validation score, The model's discrimination score is represented by α, where α is the rule validation weight coefficient and β is the model discrimination weight coefficient. These weight coefficients are determined through iterative optimization using a grid search method combined with historical discrimination accuracy samples, satisfying the following conditions: The initial calibration values ​​were 0.4 and 0.6, respectively.

[0014] The intelligent discrimination model is trained and constructed using the gradient boosting tree algorithm. During the training process, positive and negative samples are weighted and sampled to solve the sample imbalance problem where the proportion of compliance risk samples is low, effectively improving the accuracy of identifying hidden risks. The rule base supports configuring multi-level verification rules according to business nodes, and can automatically trigger differentiated rule verification combinations for different purchase amounts and different purchase categories, so as to achieve refined compliance control.

[0015] Furthermore, the business process linkage module performs the following operations when linking the procurement business process: Receive the judgment results and risk levels output by the compliance intelligent judgment module, and match them with the process control rules for the corresponding procurement business nodes; For high-risk business operations, process interception operations are triggered to prevent the business from entering the next process stage; for medium- and low-risk business operations, risk warning information is pushed to the relevant business personnel. Receive the risk rectification and handling results submitted by the business side, synchronize the handling data to the big data resource service module and the compliance intelligent judgment module, and trigger the secondary judgment process.

[0016] The process interception operation supports a hierarchical authorization release mechanism. Only after the corresponding level of management personnel submits a review application and passes the verification can the interception be released and the business process be promoted. All interception, warning, rectification review operations are automatically logged in full and stored synchronously in the big data resource service module, which fully records the entire compliance control process and meets the management requirements for audit traceability.

[0017] Furthermore, the compliance intelligent judgment module includes a rule verification unit, a model judgment unit, a risk grading unit, and a rule model management unit. The rule verification unit has a built-in procurement compliance rule library, which covers three categories of rule items: procurement process node specifications, entity qualification requirements, and budget control standards. The intelligent judgment model carried by the model judgment unit is trained and generated from historical procurement compliance annotation data. The risk grading unit sets three levels of risk classification standards (high, medium, and low) and corresponding judgment thresholds. The rule model management unit sets up operation entry points for adding, modifying, and disabling rules, as well as operation channels for model iteration updates and parameter adjustments.

[0018] Furthermore, the feature vector construction module extracts process node features including three indicators: completeness of approval nodes, rationality of process sequence, and completeness of business data; entity qualification features including three indicators: validity of participating entity qualifications, entity credit status, and compliance of guarantee; transaction behavior features including three indicators: matching of procurement methods, rationality of transaction prices, and matching of related transactions; budget matching features including two indicators: matching of demand and budget amount, and compliance of budget execution progress; and risk association features including two indicators: entity's historical violation records and historical risk situation of similar businesses. Risk association features are quantified through historical risk association values, calculated using the following formula: Where R is the correlation value of the subject's historical risk. Let k be the risk score of the historical risk event. Let λ be the time interval between the k-th historical risk event and the present, λ be the time decay coefficient, and m be the total number of historical risk events within the statistical range. Based on the risk traceability period of the procurement business, the value is matched with the decay rate of the risk impact over time, and the weight of historical risk events that exceed the traceability period approaches 0.

[0019] Furthermore, the visualization management module includes a compliance overview unit, a risk ledger unit, and a report generation unit. The compliance overview unit sets up an overall compliance data display area for the group, a compliance data display area for each subsidiary, and a risk distribution display area for each procurement process. The risk ledger unit sets up a risk details recording area, a handling status tracking area, and a rectification result retention area. The report generation unit sets up three types of report generation entry points: time dimension, organization dimension, and procurement category dimension, and sets up a report file export entry point.

[0020] The report generation supports custom dimension combination configuration and can automatically export standardized compliance audit reports that comply with audit standards; the risk ledger supports filtering and querying by multiple conditions such as risk level, handling status, and business node, intuitively presenting the distribution of procurement compliance risks across the entire group, and providing quantitative data support for managers to optimize procurement compliance control strategies.

[0021] Furthermore, the business process linkage module includes a process interception unit, a risk warning unit, and a handling result feedback unit. The business process linkage module connects to six types of procurement business execution nodes: procurement requirement submission, supplier access, bidding execution, contract signing, payment settlement, and guarantee verification. The process interception unit sets up process blocking trigger interfaces for corresponding business nodes and directly associates with the process flow controls of the business system. The risk warning unit sets up message push channels for corresponding personnel and sets up two push formats: in-site messages and system pending tasks. The handling result feedback unit sets up an entry point for uploading rectification materials and an entry point for submitting review applications, and receives all rectification-related data submitted by the business end.

[0022] Compared with existing technologies, this intelligent compliance judgment system for procurement operations based on feature vectors has the following advantages: This invention integrates multi-source procurement business data by constructing a unified big data resource service module, enabling the coordinated governance and service-oriented invocation of procurement data across the entire procurement chain. This solves the problem of fragmented and scattered data. A feature vector construction module extracts multi-dimensional procurement business features and completes vectorized mapping, providing a standardized data foundation for compliance assessment. Combined with a compliance intelligent assessment module, it achieves comprehensive assessment by fusing rule verification and model judgment, adapting to the differentiated compliance requirements of different procurement scenarios and improving the comprehensiveness and accuracy of risk identification. A business process linkage module enables real-time linkage between compliance assessment results and the procurement process, proactively managing risk control nodes and effectively meeting the closed-loop control needs of the entire group's procurement process, reducing manual verification costs and audit compliance risks.

[0023] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0025] Figure 1 This is a schematic diagram of the structure of a feature vector-based intelligent compliance judgment system for procurement operations. Figure 2 A flowchart illustrating the workflow of a feature vector-based intelligent compliance assessment system for procurement operations. Figure 3 This is a flowchart of the feature vector construction module generating compliance discrimination feature vectors. Detailed Implementation

[0026] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0027] This invention provides a feature vector-based intelligent compliance judgment system for procurement operations. With closed-loop compliance management across the entire procurement chain as its core objective, it constructs a five-layer architecture comprising a big data resource service module, a feature vector construction module, a compliance intelligent judgment module, a business process linkage module, and a visualization management module. Through a unified big data resource service system, it integrates multi-source heterogeneous procurement business data, achieving standardized governance and service-oriented access to data throughout the procurement process. Relying on feature vector construction technology, it transforms scattered, multi-dimensional procurement business information into uniformly quantifiable and calculable compliance judgment feature vectors. Combining rule verification and intelligent model judgment with a dual judgment mechanism, it completes a comprehensive assessment and classification of compliance risks. Simultaneously, the business process linkage module enables real-time interaction between judgment results and the procurement execution process. The accompanying visualization management module provides full-dimensional compliance data display and report output capabilities. Ultimately, it constructs a full-chain intelligent procurement compliance management system with data governance, feature extraction, intelligent judgment, process linkage, and visual management, effectively solving the problems of fragmented data, single judgment methods, lagging risk identification, and insufficient scenario adaptability in traditional procurement compliance management. This comprehensively improves the accuracy, real-time performance, and closed-loop management capabilities of procurement compliance management for group enterprises. Specific embodiments are described in detail below.

[0028] like Figure 1As shown in the figure, the feature vector-based intelligent compliance judgment system for procurement business described in this embodiment adopts a microservice architecture and is deployed on the private cloud platform of the group enterprise. Each module achieves data interaction and business decoupling through standardized application programming interfaces. The computing resources of a single module can be elastically expanded on demand according to the business concurrency, adapting to the concurrent access needs of multi-level organizations and multi-category procurement businesses within the group. The application scenario of this embodiment is the procurement compliance management scenario of a large manufacturing group. The group has 12 subsidiaries, and its procurement business covers three core procurement categories: production materials, office supplies, and engineering services. The original procurement data was scattered and stored in multiple independent business systems, and compliance management relied on manual post-event spot checks, resulting in low risk management efficiency.

[0029] The big data resource service module, serving as the data foundation of the entire system, consists of four parts: a multi-source data access unit, a main data governance unit, a data storage unit, and a data service interface unit. It undertakes the functions of collecting, governing, storing, and invoking procurement data across the entire supply chain. The multi-source data access unit has seven pre-built standardized data interface types, connecting to the group's internal office automation system, financial control system, transparent procurement system, supplier management platform, and e-commerce procurement platform, as well as external enterprise credit information disclosure platforms and third-party credit reporting platforms.

[0030] In this embodiment, differentiated data collection modes are set according to the data update characteristics of different business systems. For business data with high real-time requirements, such as procurement demand submission, bidding information release, contract signing, and payment settlement, a real-time triggered synchronization mode is adopted. After the corresponding node in the business system completes the operation, a data synchronization command is immediately triggered, and the latest business data is pushed to this module. For data with low update frequency, such as supplier master data updates, financial reconciliation data, and monthly reports, a timed incremental synchronization mode is adopted. The incremental data synchronization task is set to be executed at 2:00 AM every day, and only data entries that have changed within the previous working day are synchronized, reducing the resource consumption of data transmission on the business system.

[0031] In this embodiment, the data collected by the multi-source data access unit covers eight categories: procurement demand details, procurement plan approval records, supplier qualification and license information, bidding announcements and tender documents, contract texts and performance records, financial payment and reconciliation data, performance guarantee information, and supplier credit records. This achieves full data coverage of the entire procurement process from demand initiation to payment settlement.

[0032] The master data governance unit receives the raw business data and performs standardized governance on four core master data categories: material master data, supplier master data, organizational structure master data, and procurement category master data. In this embodiment, the master data coding adopts a three-segment unified coding rule: the first segment is the category-level code, the second segment is the attribute-dimensional code, and the third segment is the serial number code. Each segment has a fixed length to ensure the uniqueness and identifiability of the code.

[0033] During master data governance, a cross-system data mapping table is first established. This table matches the original codes of similar master data across different business systems with the standardized codes of this system. For example, supplier codes in the transparent procurement system are matched with supplier codes in the supplier management platform using the unified social credit code, mapping them to the standardized supplier master data codes of this system. The data governance process incorporates built-in data cleaning rules, automatically handling missing and outlier values ​​in the collected data. Numerical missing data is logically completed using the average value of the same category during the same period. Textual missing data is marked as pending completion and pushed to the corresponding business personnel for processing. Outlier data exceeding a reasonable value range triggers a verification alarm, which is then manually reviewed and corrected. In this embodiment, after master data governance, all similar master data across systems possess a unique standardized identifier, eliminating discrepancies caused by data silos at the data source level and providing a unified data benchmark for subsequent compliance assessments.

[0034] The data storage unit is divided into three independent storage partitions: a business detail storage area, a master data storage area, and a compliance result storage area. The business detail storage area stores the original detailed data and process operation data of all procurement transactions, stored hierarchically according to a two-level directory of procurement year and procurement category, with a 3-year online storage period. Historical business detail data exceeding the storage period is automatically archived to offline storage media, freeing up online storage resources. The master data storage area stores four types of core master data after standardized governance and cross-system mapping tables. Data updates employ a real-time synchronization mechanism to ensure the consistency and timeliness of master data. The compliance result storage area stores data such as the judgment results, risk level, and handling records output by the subsequent compliance intelligent judgment module. It establishes a correlation index with the corresponding business detail data, supporting quick querying of full-chain compliance information by business order number.

[0035] The data service interface unit sets up data call interfaces according to business scenarios and data dimensions, including multiple types of interfaces such as purchase order full data query interface, supplier main data query interface, and compliance result feedback interface. All interfaces are equipped with call identity verification mechanisms, and data call channels with corresponding permissions are only opened to authorized modules within the system. Different modules can only call the data dimensions required by their own business and cannot access data content beyond their permission scope, thus ensuring the security and compliance of data access.

[0036] like Figure 3As shown, the feature vector construction module receives standardized data from the big data resource service module and completes three core steps: compliance feature extraction, feature standardization processing, and feature vectorization mapping. Ultimately, it generates a compliance discrimination feature vector with fixed dimensions, providing standardized input for subsequent intelligent compliance discrimination. In this embodiment, when a procurement transaction enters the compliance discrimination node, the feature vector construction module first initiates a data call request to the big data resource service module to retrieve the full-dimensional associated data corresponding to that procurement transaction, covering five categories of associated data: business process information, subject qualification information, transaction execution information, budget information, and historical risk information.

[0037] The first step is multi-dimensional compliance feature extraction. This embodiment extracts five major categories of features, each containing corresponding sub-indicators, all with clearly defined quantitative calculation rules. The first category is process node features, including three indicators: approval node completeness, process sequence rationality, and business data completeness. Approval node completeness is quantified by the ratio of the number of completed approval nodes for the procurement transaction to the total number of legally mandated approval nodes required for the corresponding procurement scenario. This indicator reaches its maximum value when all required approval nodes have been completed. Process sequence rationality is quantified by verifying whether the order of each approval node conforms to the procurement business process specifications. A deduction is made for this indicator if nodes are operated in reverse order. Business data completeness is quantified by the ratio of the number of required business data uploaded for the transaction to the total number of required data. This indicator reaches its maximum value when all required data is complete.

[0038] The second category is the characteristics of the main entity's qualifications, including three indicators: the validity of the participating entity's qualifications, the entity's credit status, and the compliance of the performance guarantee. The validity of the participating entity's qualifications is quantified by verifying whether all suppliers participating in the procurement transaction possess the qualification certificates required for the corresponding procurement category, and whether all qualification certificates are valid. This indicator is set to its maximum value when all participating entity qualifications are valid. The entity's credit status is quantified by retrieving supplier credit data from external credit platforms to verify whether suppliers have records of being defaulters, administrative penalties, or abnormal business operations. This indicator is set to its maximum value when suppliers have no adverse credit records. For procurement transactions requiring a performance guarantee, this indicator is quantified by verifying whether the guarantee amount, validity period, and issuing institution meet the requirements. This indicator is set to its maximum value when the guarantee fully meets the requirements.

[0039] The third category is transaction behavior characteristics, including three indicators: matching of procurement methods, reasonableness of transaction prices, and matching of related-party transactions. Matching of procurement methods is quantified by verifying whether the procurement method matches the requirements of the corresponding procurement amount and category. This indicator reaches its maximum value when the procurement method fully complies with the regulations. Reasonableness of transaction prices is quantified by comparing the transaction price with the historical average transaction price of the same category over the past six months and calculating the degree of price deviation. This indicator reaches its maximum value when the transaction price is within a reasonable deviation range. Matching of related-party transactions is quantified by verifying whether there is a related-party relationship between the participating suppliers and the purchasing party, and whether the related-party transaction has undergone the corresponding approval procedures. This indicator reaches its maximum value when there is no related-party transaction or the related-party transaction has completed compliant approval.

[0040] The fourth category is budget matching characteristics, which includes two indicators: matching of demand and budget amount, and compliance of budget execution progress. Matching of demand and budget amount is quantified by verifying whether the requested amount for the procurement transaction is within the available budget limit for the corresponding project. This indicator takes its maximum value when the requested amount does not exceed the available budget limit. Compliance of budget execution progress is quantified by verifying whether the cumulative execution progress of the budget project meets the progress requirements of the corresponding time nodes. This indicator takes its maximum value when the execution progress is within a reasonable range.

[0041] The fifth category is risk correlation characteristics, which includes two indicators: the entity's historical violation records and the historical risk situation of similar businesses. Both indicators are quantitatively represented by historical risk correlation values. In the specific implementation of this embodiment, the formula for calculating the historical risk correlation value is as follows: Where R is the correlation value of the subject's historical risk. Let k be the risk score of the historical risk event. Let λ be the time interval between the k-th historical risk event and the present, λ be the time decay coefficient, and m be the total number of historical risk events within the statistical range.

[0042] In this embodiment, the time decay coefficient λ is determined based on the risk tracing period of the procurement business, which is set to 3 years. The value of the time decay coefficient matches the rate of decay of risk impact over time. When a historical risk event occurs 3 years ago, its corresponding weight value approaches 0 and is no longer included in the risk correlation calculation. The risk score of each historical risk event is determined according to the severity of the event; high-risk events correspond to higher risk scores, and low-risk events correspond to lower risk scores. The calculation method for historical risks of similar businesses is consistent with the entity's historical violation records, only the statistical scope is replaced with historical business risk events of the same product category and procurement method, thereby characterizing the inherent risk level of this type of procurement business.

[0043] The second step is feature standardization. In this embodiment, all extracted feature indicators are discretized and normalized. For qualitative or discrete features such as the completeness of approval nodes, discretization is first performed to convert different compliance statuses into corresponding quantitative scores, with the score range uniformly set to 0 to 1. For continuous quantitative features such as transaction price deviation, the min-max normalization method is used to map feature values ​​of different dimensions to a uniform numerical range of 0 to 1, eliminating the interference of dimensional differences between different indicators on subsequent judgment results. In this embodiment, after standardization, all feature indicators are converted into quantitative values ​​ranging from 0 to 1. Higher values ​​indicate better compliance in the corresponding dimension, while lower values ​​indicate a higher probability of compliance risk in the corresponding dimension.

[0044] The third step is to generate the compliance judgment feature vector. According to a preset feature dimension sorting rule, the standardized feature indicators are arranged in a fixed order to form an initial feature vector. Then, a weighted mapping operation is used to generate the final compliance judgment feature vector. In the specific implementation of this embodiment, the weighted mapping operation formula is as follows: In the formula To ensure compliance in feature vector identification, The initial feature vector after standardization. is the feature weight vector for dimension matching, and ⊙ represents the Hadamard product operation.

[0045] In this embodiment, the feature weight vector is calibrated using the analytic hierarchy process (AHP) combined with the feature contribution of historical compliance risk samples. The specific calibration process is as follows: First, an AHP structure is constructed, with procurement compliance risk identification as the target layer, five categories of features (process node features, entity qualification features, transaction behavior features, budget matching features, and risk correlation features) as the criterion layer, and the sub-indicators under each category as the solution layer. Second, a pairwise comparison judgment matrix is ​​constructed. Five experts in the field of procurement compliance auditing are invited to conduct pairwise comparisons and scores on the importance of compliance risk identification for the five categories of features in the criterion layer. A 1-9 scale is used to construct the judgment matrix, where 1 indicates that two elements are equally important, and 9 indicates that one element is extremely more important than the other.

[0046] Subsequently, the largest eigenvalue and corresponding eigenvector of the judgment matrix are calculated. After normalizing the eigenvectors, the initial weights for the five feature categories are obtained. Simultaneously, a consistency check is performed on the judgment matrix, calculating the consistency ratio. If the consistency ratio is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency, and the weight results are valid. If the consistency ratio does not meet the requirements, experts are asked to adjust the scores of the judgment matrix until the consistency check is passed. For each feature category's sub-indicators, the same analytic hierarchy process is used, combining the frequency and impact of risks corresponding to each indicator in historical compliance risk samples to calculate the weights of each sub-indicator.

[0047] Finally, the weights of the major category features and the weights of the sub-indicators are combined to obtain a complete feature weight vector, with the sum of all weight values ​​being 1. Each dimension's weight value corresponds to its contribution percentage in risk identification. In this embodiment, the weights of transaction behavior features and entity qualification features are relatively high, corresponding to their core role in identifying procurement compliance risks. After weighted mapping operations, a compliance discrimination feature vector with fixed dimensions is generated. The vector dimension is consistent with the total number of sub-indicators. Different procurement categories can directly call their own feature weight configurations without refactoring the feature extraction logic, significantly improving the adaptability efficiency for different business scenarios.

[0048] The compliance intelligent judgment module receives compliance judgment feature vectors transmitted by the feature vector construction module. Through a dual judgment mechanism combining rule verification and model judgment, it completes a comprehensive assessment and classification of compliance risks in procurement operations. This module consists of four parts: a rule verification unit, a model judgment unit, a risk grading unit, and a rule model management unit. First, the rule verification process is executed. The rule verification unit has a built-in procurement compliance rule library, covering three categories of rule items: procurement process nodes, specifications, main body qualification requirements, and budget control standards. Each category contains multiple specific verification rules, and all rules correspond to rule-type feature items in the feature vector. During rule verification, the rule-type feature items corresponding to the compliance judgment feature vector are matched item by item with the verification thresholds in the rule library. The number of rule items that pass verification and the number of rule items that fail verification are counted, and the rule verification score is calculated by combining the weight of each rule.

[0049] In this embodiment, the rule base supports configuring differentiated rule verification combinations based on procurement amount, procurement category, and procurement method. For example, for open bidding procurement, it triggers full rule verification, including the duration of the bidding announcement, bidder qualification requirements, and bid evaluation process specifications. For small-scale, sporadic procurement, it only triggers core rules such as budget verification and basic supplier qualification verification, achieving refined compliance control. The rule model management unit provides a visual operation entry point for adding, modifying, and deactivating rules. Administrators can dynamically adjust the rule base content according to the latest compliance management requirements without modifying the underlying system code.

[0050] Next, the model discrimination process is executed. The model discrimination unit is equipped with a trained intelligent discrimination model. The complete compliance discrimination feature vector is input into the model, and the model outputs the compliance probability and hidden risk points of the corresponding procurement business through calculation, thereby generating a model discrimination score. In this embodiment, the intelligent discrimination model is trained using the gradient boosting tree algorithm. The training samples are derived from the group's historical procurement data over the past three years, including audited compliance risk samples and normal compliance samples. During model training, to address the sample imbalance problem where the proportion of compliance risk samples is low, weighted sampling is applied to positive and negative samples, appropriately increasing the sampling weight of risk samples to improve the model's accuracy in identifying hidden risks. After model training is completed, the discrimination accuracy of the model is verified using an independent test sample set. After successful verification, it is officially deployed online. The rule model management unit provides an operation channel for model iteration updates and parameter adjustments, allowing for incremental training of the model using newly added compliance-annotated data to continuously optimize the model's discrimination performance.

[0051] After completing rule verification and model discrimination, the risk grading unit calculates the comprehensive compliance risk score and performs risk level classification based on the risk level classification standards. In the specific implementation of this embodiment, the formula for calculating the comprehensive compliance risk score is as follows: Where S represents the overall compliance risk score. For rule validation score, α is the model's discrimination score, β is the rule validation weight coefficient, and β is the model's discrimination weight coefficient.

[0052] In this embodiment, the weight coefficients are determined through grid search combined with iterative optimization using historical accuracy samples, while satisfying the constraint that α plus β equals 1. Specifically, the determination process is as follows: First, the search range and step size of the weight coefficients are defined. The value range of α is set to 0.2-0.8, the corresponding value range of β is 0.8-0.2, and the search step size is set to 0.05, thereby generating multiple different combinations of weight coefficients. Second, a historically labeled procurement business sample set is used as the test set. For each weight coefficient combination, the comprehensive score and corresponding risk level of all test samples are calculated. The judgment result is compared with the actual labeled result of the sample, and the judgment accuracy corresponding to that weight coefficient is calculated. Finally, all weight coefficient combinations are traversed, and the group with the highest judgment accuracy is selected as the final weight coefficient configuration.

[0053] In this embodiment, after grid search iterative optimization, the initial calibration value of α is 0.4, and the initial calibration value of β is 0.6, meaning the model's judgment weight is relatively higher, fully leveraging the intelligent model's ability to identify hidden risks. The risk grading unit sets three levels of risk classification standards (high, medium, and low) and corresponding judgment thresholds. A higher comprehensive score indicates better business compliance, while a lower comprehensive score indicates higher compliance risk. In this embodiment, businesses with a comprehensive score higher than the first threshold are classified as low-risk, businesses with a comprehensive score between the first and second thresholds are classified as medium-risk, and businesses with a comprehensive score lower than the second threshold are classified as high-risk. After completing the risk level classification, the system automatically generates corresponding risk descriptions and handling guidelines, clearly marking the failed rule items and the hidden risk points identified by the model, and providing corresponding rectification suggestions to provide clear guidance for subsequent business handling.

[0054] like Figure 2 As shown, the business process linkage module receives the judgment results from the compliance intelligent judgment module, realizing real-time linkage between compliance judgment and procurement business execution process, and completing a closed-loop control process of risk interception, risk warning, handling, feedback, and secondary review. This module consists of three parts: a process interception unit, a risk warning unit, and a handling result feedback unit, connecting with six core procurement business execution nodes: procurement requirement submission, supplier access, bidding, contract signing, payment settlement, and guarantee verification. The business process linkage module first receives the judgment results and risk levels output by the compliance intelligent judgment module, and matches the corresponding process control rules according to the execution node to which the business belongs. For procurement businesses judged to be of a high risk level, the process interception unit triggers a process interception operation, directly calling the business system's process flow control through the interface to prevent the business from entering the next process stage, and simultaneously generating an interception notification and pushing it to the corresponding business personnel and compliance management personnel.

[0055] In this embodiment, the process interception operation supports a tiered authorization release mechanism. Managers at the corresponding level can submit a review application, upload relevant explanatory materials and approval basis, and only after verification and approval by the compliance management department can the process interception be released and the business process proceed. All interception release operations automatically retain a full operation log and are synchronously stored in the big data resource service module. For procurement operations determined to be of medium to low risk, the risk warning unit pushes risk alerts to the corresponding business personnel, clearly informing them of the existing risk points and rectification suggestions. The push notifications include in-site messages and system pending tasks, ensuring that business personnel are promptly informed of risk information. Medium to low risk businesses are not subject to process interception, but business personnel are required to complete risk rectification within a specified time limit and upload rectification materials after completion.

[0056] The handling result feedback unit sets up an entry point for uploading rectification materials and an entry point for submitting review applications, receiving all rectification-related data submitted by the business side. After business personnel complete risk rectification and submit handling results, the handling result feedback unit synchronizes the rectification handling data to the big data resource service module for archiving, and simultaneously pushes the data to the compliance intelligent judgment module, triggering a secondary judgment process. The secondary judgment re-executes the rule verification and model judgment process, calculates the updated comprehensive risk score and risk level. If the secondary judgment result is low risk, the risk warning or process interception is lifted, allowing the business to continue; if the secondary judgment is still medium to high risk, the control status is maintained and an upgrade notification is pushed to higher-level management personnel. In this embodiment, all interception, warning, rectification, and review operations are automatically logged in full, completely recording the entire compliance control chain, meeting the management requirements for audit traceability, and achieving closed-loop management of procurement compliance control.

[0057] The visualization and control module receives the judgment results data output by the compliance intelligent judgment module and the processing data from the business process linkage module, providing comprehensive compliance data display and report output services to support management decisions. This module consists of three parts: a compliance overview unit, a risk ledger unit, and a report generation unit. The compliance overview unit has three data display areas: a group-wide overall compliance data display area, a subsidiary-wide compliance data display area, and a risk distribution display area for each procurement stage. The group-wide overall compliance data display area shows core indicators such as the total compliance rate and risk level distribution of the entire group's procurement business, intuitively presenting the overall procurement compliance status of the group.

[0058] The compliance data display area for each subsidiary displays procurement compliance data for all subsidiaries at the organizational level, supporting sorting by compliance indicators to facilitate the identification of subsidiaries with weak compliance management. The risk distribution display area for each procurement stage shows the number and percentage of risks occurring at different procurement business nodes, identifying high-risk business stages and providing a basis for targeted optimization of control strategies. The risk ledger unit includes a risk details recording area, a handling status tracking area, and a rectification result retention area. The risk details recording area records detailed information for all identified procurement compliance risks, including business order number, risk level, risk type, risk description, and involved entities. The handling status tracking area displays the real-time progress of each risky business, including statuses such as pending rectification, in progress, awaiting review, and closed loop.

[0059] The rectification results retention area stores all rectification materials and review records, establishing a link with corresponding risky business operations and supporting full-chain traceability queries. The risk ledger supports filtering and querying based on multiple conditions, such as risk level, handling status, and business node, allowing managers to quickly locate risky business operations of interest. The report generation unit has three report generation entry points: time dimension, organization dimension, and procurement category dimension, supporting custom dimension combinations and automatically generating standardized compliance audit reports that conform to audit standards. Report content includes overall compliance status, risk distribution trends, rectification completion status, and trend analysis within the period, and supports exporting to commonly used electronic document formats to meet the needs of various scenarios such as reporting, auditing, and archiving.

[0060] In summary, this embodiment, by building a complete intelligent compliance judgment system for procurement business based on feature vectors, has been successfully implemented in the procurement compliance management scenario of a large manufacturing group. It effectively integrates procurement data scattered across multiple business systems, constructing a unified and standardized data foundation and eliminating the management barriers caused by data silos. Through multi-dimensional feature extraction and vectorization mapping, scattered procurement business information is transformed into standardized feature vectors that can be quantified and calculated, providing a unified data input standard for compliance judgment and significantly improving the adaptability efficiency for different procurement scenarios. The dual judgment mechanism, integrating rule verification and intelligent model judgment, ensures both the rigid execution of explicit compliance rules and the effective identification of implicit compliance risks, comprehensively improving the comprehensiveness and accuracy of procurement compliance risk identification. The business process linkage mechanism brings compliance management nodes forward, enabling real-time interaction between compliance judgment results and the procurement execution process, changing the traditional passive management model of post-audit and effectively improving the timeliness of risk handling. The supporting visualized management system provides managers with a full-dimensional compliance data view, providing quantitative support for optimizing management strategies. The overall system has built a full-chain procurement compliance control capability, from data governance to intelligent judgment and process closure. It effectively reduces the workload of manual verification, improves the overall efficiency of group procurement compliance management, and provides a feasible implementation solution for the digital and intelligent transformation of procurement compliance in group enterprises.

[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A procurement business compliance intelligent discrimination system based on eigenvectors, characterized in that, The system includes: a big data resource service module, a feature vector construction module, a compliance intelligent judgment module, a business process linkage module, and a visual management and control module; The big data resource service module is used to connect to multi-source heterogeneous procurement business systems, collect and manage procurement data across the entire chain, and provide standardized data calling services for the feature vector construction module. The feature vector construction module is used to retrieve the full-dimensional related data corresponding to the procurement business to be judged, extract multi-dimensional compliance features and complete the standardization process, generate a compliance judgment feature vector and transmit it to the compliance intelligent judgment module. The compliance intelligent judgment module is used to receive compliance judgment feature vectors, combine the rule base and the judgment model to complete compliance verification and risk classification, and output the judgment results to the business process linkage module and the visualization control module respectively. The business process linkage module is used to connect with the procurement business execution node, receive the judgment result and trigger the corresponding process operation, and send the processing result back to the big data resource service module. The visualization and control module is used to receive judgment results and processing data, and to provide compliance data display and report output services.

2. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The big data resource service module includes a multi-source data access unit, a master data governance unit, a data storage unit, and a data service interface unit. The multi-source data access unit is equipped with standardized interfaces for connecting to office automation systems, financial control systems, transparent procurement systems, supplier management platforms, e-commerce procurement platforms, and external enterprise credit platforms. It collects procurement needs, procurement plans, supplier qualifications, bidding information, contract information, financial reconciliation, guarantee information, and entity credit data. The master data governance unit performs standardized coding and cross-system mapping processing on material master data, supplier master data, organizational structure master data, and procurement category master data. The data storage unit is equipped with a business detail storage area, a master data storage area, and a compliance result storage area, storing corresponding data hierarchically. The data service interface unit is equipped with calling interfaces categorized by business scenario and data dimension.

3. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 2, characterized in that, The multi-source data access unit is equipped with two data acquisition modes: timed incremental synchronization and real-time triggered synchronization, corresponding to the data update frequency of different business systems. The master data governance unit sets up a cross-system data mapping table to store the encoding correspondence of the same type of master data between different systems. The data storage unit sets up a historical data archiving cycle and performs archiving processing on business detail data that exceeds the time limit. The data service interface unit sets up a call identity verification mechanism and only opens the data retrieval channel with corresponding permissions to authorized modules.

4. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The feature vector construction module performs the following operations when generating compliance discrimination feature vectors: The big data resource service module retrieves full-dimensional related data for the corresponding procurement business scenario, and extracts corresponding indicators for five categories of feature items: process node features, entity qualification features, transaction behavior features, budget matching features, and risk association features. Discretization and normalization are performed on all extracted feature indicators to convert qualitative features into quantitative scores and map quantitative features of different dimensions to a unified numerical range. The standardized feature indicators are arranged according to a preset feature dimension sorting rule, and a fixed-dimensional compliance discrimination feature vector is generated through a weighted mapping operation. The weighted mapping operation formula is as follows: In the formula To ensure compliance in feature vector identification, The initial feature vector after standardization. is the feature weight vector for dimension matching, and ⊙ represents the Hadamard product operation.

5. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The compliance intelligent judgment module performs the following operations when performing compliance judgment: The built-in procurement compliance rule library is invoked to perform item-by-item matching and verification on the rule-class feature items in the input compliance judgment feature vector, outputting the pass and fail items of the rule verification and generating the rule verification score; The compliance discrimination feature vector is input into the trained intelligent discrimination model, and the model outputs the compliance probability and hidden risk points of the corresponding procurement business through model calculation, generating the model discrimination score; The comprehensive compliance risk score is calculated, and the compliance risk implementation level is classified according to the risk level classification standard. Corresponding risk descriptions and handling guidelines are then generated. The formula for calculating the comprehensive compliance risk score is as follows: Where S represents the overall compliance risk score. For rule validation score, α is the model's discrimination score, β is the rule validation weight coefficient, and β is the model's discrimination weight coefficient.

6. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The business process linkage module performs the following operations when linking the procurement business process: Receive the judgment results and risk levels output by the compliance intelligent judgment module, and match them with the process control rules for the corresponding procurement business nodes; For high-risk business operations, process interception operations are triggered to prevent the business from entering the next process stage; for medium- and low-risk business operations, risk warning information is pushed to the relevant business personnel. Receive the risk rectification and handling results submitted by the business side, synchronize the handling data to the big data resource service module and the compliance intelligent judgment module, and trigger the secondary judgment process.

7. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The compliance intelligent judgment module includes a rule verification unit, a model judgment unit, a risk classification unit, and a rule model management unit. The rule verification unit has a built-in procurement compliance rule library, which covers three categories of rule items: procurement process node specifications, entity qualification requirements, and budget control standards. The intelligent judgment model in the model judgment unit is trained and generated from historical procurement compliance annotation data. The risk classification unit sets three levels of risk classification standards (high, medium, and low) and corresponding judgment thresholds. The rule model management unit sets up operation entry points for adding, modifying, and disabling rules, as well as operation channels for model iteration updates and parameter adjustments.

8. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The feature vector construction module extracts the following process node features: completeness of approval nodes, reasonableness of process sequence, and completeness of business data. The entity qualification features include the validity of participating entity qualifications, entity credit status, and compliance of guarantee. The transaction behavior features include the matching of procurement methods, reasonableness of transaction prices, and matching of related transactions. The budget matching features include the matching of demand and budget amount and compliance of budget execution progress. The risk correlation features include the entity's historical violation records and historical risk status of similar businesses. Risk correlation features are quantified using historical risk correlation values, calculated using the following formula: Where R is the correlation value of the subject's historical risk. Let k be the risk score of the historical risk event. Let λ be the time interval between the k-th historical risk event and the present, λ be the time decay coefficient, and m be the total number of historical risk events within the statistical range.

9. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The visualization management module includes a compliance overview unit, a risk ledger unit, and a report generation unit. The compliance overview unit sets up an overall compliance data display area for the group, a compliance data display area for each subsidiary, and a risk distribution display area for each procurement process. The risk ledger unit sets up a risk details recording area, a handling status tracking area, and a rectification result retention area. The report generation unit sets up three types of report generation entry points: time dimension, organization dimension, and procurement category dimension, and sets up a report file export entry point.

10. The intelligent compliance judgment system for procurement business based on feature vectors according to claim 1, characterized in that, The business process linkage module includes a process interception unit, a risk warning unit, and a handling result feedback unit. The business process linkage module connects to six types of procurement business execution nodes: procurement requirement submission, supplier access, bidding execution, contract signing, payment settlement, and guarantee verification. The process interception unit sets up a process blocking trigger interface for the corresponding business node and directly links to the process flow control of the business system. The risk warning unit sets up a message push channel for the corresponding personnel and sets up two push formats: in-site messages and system pending tasks. The handling result feedback unit sets up an entry point for uploading rectification materials and an entry point for submitting review applications, and receives all rectification-related data submitted by the business end.