Risk management and control system and method for authorized purchase
By constructing a multi-source heterogeneous knowledge base and a large-scale language model for semantic risk identification, dynamic maximum price limits and comprehensive risk indices are generated. This solves the problems of data fragmentation and static risk assessment in existing technologies, realizes the forward shift of risk control and improves approval efficiency, and establishes an adaptive and optimized procurement management system.
Patent Information
- Application Number
- CN202511721219.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-21
- Publication Date
- 2026-02-27
AI Technical Summary
The existing procurement management system suffers from data fragmentation, static risk identification and assessment, and a lack of dynamic closed-loop feedback, resulting in insufficient risk control capabilities, lengthy approval processes, and wasted resources, making it difficult to meet the needs of enterprise development.
Construct a multi-source heterogeneous knowledge base, identify semantic risks through a large-scale language model, generate dynamic maximum price limits and comprehensive risk indices, achieve real-time early warning and adaptive optimization, and manage differentiated approval processes in conjunction with a collaborative review module.
It has enabled the forward shift of risk control, improved the quality of procurement needs and the accuracy of budgets, optimized approval efficiency, established enterprise-level knowledge management and long-term capability building, and the system has the ability to self-evolve.
Smart Images

Figure CN121581635A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of computer technology and artificial intelligence, in particular to a risk management and control system and method for authorized procurement. BACKGROUND
[0002] As a key link of internal control and resource allocation of enterprises, especially large group enterprises, the management efficiency and risk control level of authorized procurement are directly related to the operating cost of the enterprise and the stability of the supply chain. In order to standardize the procurement process and improve transparency, modern enterprises have generally shifted from traditional manual and paper-based approval to management relying on enterprise resource planning systems, supplier relationship management systems and other information platforms. These systems have achieved certain results in standardizing and online procurement processes, solidifying approval nodes and providing a basis for data retention.
[0003] However, with the ever-changing market environment and the increasing complexity of business needs, the existing technical means gradually expose their inherent limitations in dealing with the risks of authorized procurement throughout the chain. The core logic of the current mainstream procurement management system often focuses on the solidification and execution of the process, but lacks the ability to identify, dynamically assess and optimize risks in depth. The defects of the existing technology are not isolated, but are interwoven, forming a systematic problem that is difficult to break through.
[0004] There is a serious data island phenomenon in the existing technology. The data generated throughout the life cycle of procurement activities is highly heterogeneous and fragmented. For example, unstructured text such as procurement demand description and technical specifications at the front end, market inquiries, contract terms at the middle end, and a large amount of process data generated during the project implementation process at the back end are often stored in different business systems, document servers, and even personal computers, lacking a unified and connected knowledge view. This fragmentation of data directly leads to the fact that all subsequent risk analysis can only be based on one-sided and static information, making it difficult to form a comprehensive and dynamic understanding.
[0005] On this basis, the existing technology for risk identification and assessment is also relatively rudimentary and passive. Due to the lack of deep semantic understanding of unstructured text, the system cannot effectively quantify the common fuzzy descriptions in procurement demand documents, and the discovery of risks is heavily dependent on the personal experience and responsibility of the approver. This makes the risk control passively move to the manual approval link, or even be discovered passively after disputes arise during project implementation. At the same time, in the process of setting procurement limits, traditional methods often rely on historical transaction prices or simple cost addition methods, making it difficult to dynamically associate the specific risks of a project with the price, resulting in a lack of scientific foresight and risk hedging ability in budgeting.
[0006] In addition, the approval process of the prior art also mostly presents rigidity and "one size fits all" characteristics. Whether it is a high-risk procurement with great value and complex technology, or a regular procurement with small amount and standard content, the same lengthy approval chain is often required. This not only causes a huge waste of approval resources and reduces operational efficiency, but also makes the energy of the approval personnel be dispersed by a large number of low-risk transactions, and it is difficult to effectively focus on the high-risk projects that really need in-depth review.
[0007] The prior art generally lacks effective closed-loop feedback and adaptive optimization mechanism. After a procurement activity is approved, the subsequent performance of the procurement activity, whether it is cost overrun, time delay or quality dispute, the relevant data cannot be systematically collected and used to optimize the procurement model and rules at the front end. The system cannot automatically learn from past failures, so the same mistakes may be repeated. This "open-loop" management mode makes the entire risk control system remain at a static level that cannot evolve itself, and it is difficult to adapt to the needs of continuous development of enterprises. SUMMARY
[0008] In view of the deficiencies of the prior art, the present application provides a risk control system and method for authorized procurement, which solves the problem that the overall risk control capability cannot evolve itself and adaptively optimize due to the lack of dynamic closed-loop feedback mechanism from the performance of the procurement activity.
[0009] To achieve the above purpose, the present application realizes the following technical scheme: a risk control system for authorized procurement, comprising: a knowledge base construction module for collecting multi-source heterogeneous data related to authorized procurement and constructing a knowledge base; an intelligent template module configured to generate service quantity list and technical parameter templates based on the knowledge base, and to perform real-time semantic risk identification and early warning when a user inputs a procurement demand text; a risk assessment and price limit module configured to receive a to-be-approved authorized procurement demand, analyze the authorized procurement demand based on the knowledge base and the results of semantic risk identification, and output a comprehensive risk index and a dynamic maximum price limit; a collaborative joint review module configured to match and execute the corresponding approval process for the authorized procurement demand according to the comprehensive risk index; an adaptive optimization module configured to collect performance data during and after the procurement project performance, quantify the performance deviation, and update the model parameters used in the risk assessment and price limit module in reverse according to the performance deviation.
[0010] Preferably, the knowledge base construction module is specifically configured to collect the multi-source heterogeneous data including at least historical procurement data, real-time market data, and the performance data of the performance process, and construct the knowledge base.
[0011] Preferably, the intelligent template module is specifically configured to complete the real-time semantic risk identification by comparing the semantic vector of the procurement demand text with the semantic vector of the historical controversial case recorded in the knowledge base.
[0012] Preferably, the risk assessment and price limiting module is configured to calculate the comprehensive risk index by weighted summation of multiple risk dimensions including at least list integrity risk, technical parameter definiteness risk, and supplier performance risk.
[0013] Preferably, the risk assessment and price limiting module is configured to calculate the dynamic maximum price by combining market price anchoring price and cost back-propagation price, and applying a risk adjustment coefficient determined by the comprehensive risk index.
[0014] Preferably, the collaborative review module is specifically configured to: compare the comprehensive risk index with a preset risk classification threshold; trigger a cross-departmental collaborative review process as the corresponding approval process when the comprehensive risk index is greater than or equal to the risk classification threshold.
[0015] Preferably, the adaptive optimization module is configured to quantify the performance deviation including at least cost deviation, time deviation, and quality deviation by analyzing the difference between the performance data and the procurement stage expectation.
[0016] Preferably, the adaptive optimization module is configured to complete the reverse update by calculating the gradient of the performance deviation with respect to the model parameters, and iteratively updating the model parameters using the gradient descent method.
[0017] Preferably, the model parameters include at least the weights of the multiple risk dimensions.
[0018] A risk management method for authorized procurement includes the following steps: S1, collecting multi-source heterogeneous data related to authorized procurement, and constructing a knowledge base; S2, generating service quantity list and technical parameter templates based on the knowledge base, and performing real-time semantic risk identification and early warning when receiving an authorized procurement demand to be approved; S3, analyzing the authorized procurement demand based on the knowledge base and the results of the semantic risk identification, and outputting a comprehensive risk index and a dynamic maximum price; S4, according to the comprehensive risk index, matching and executing the corresponding approval process for the authorized procurement demand; S5, in the procurement project implementation process and after the end, collecting the implementation process data, quantifying the implementation deviation, and updating the model parameters used in step S3 according to the implementation deviation.
[0019] The application provides a risk control system and method for authorized procurement. The application has the following beneficial effects: 1. The application constructs a multi-source heterogeneous knowledge base collecting historical procurement data, real-time market data and implementation process data, and uses a large language model to perform semantic risk identification on unstructured text. Compared with the traditional risk identification method relying on keyword matching or manual experience, the application can discover and warn potential controversy risks caused by ambiguous description and unclear definition earlier and more accurately in the procurement demand preparation stage, significantly moving the risk control point forward, and improving the quality of procurement demand from the source.
[0020] 2. The application constructs a dynamic highest limit price model that integrates market price anchoring price, cost reverse price and risk adjustment coefficient. The risk adjustment coefficient is directly linked to the multi-dimensional comprehensive risk index, so that the final output procurement limit price is no longer a static and isolated budget value, but a scientific price benchmark that can dynamically reflect market fluctuations and make forward-looking compensation for the inherent risks of the project, thereby significantly improving the rationality and accuracy of the procurement budget.
[0021] 3. The application processes and deposits all kinds of scattered and heterogeneous data generated in the procurement chain into a structured dynamic knowledge base, and converts the implicit knowledge stored in personal experience or scattered documents into enterprise-level, calculable and inheritable explicit knowledge assets. Not only does it provide a high-quality data foundation for the operation of all intelligent models in the system, but also provides effective technical support for organizational knowledge management and long-term capacity building. BRIEF DESCRIPTION OF DRAWINGS
[0022] Figure 1 The system architecture of the application is shown in the figure; Figure 2 The method flowchart of the application is shown in the figure. DETAILED DESCRIPTION
[0023] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.
[0024] Embodiments: Please refer to the accompanying Figure 1 Embodiments of the present application provide a risk management system for authorized procurement, comprising: A knowledge base construction module is configured to collect multi-source heterogeneous data related to authorized procurement and construct a knowledge base. The knowledge base construction module aims to transform scattered and heterogeneous procurement-related information into a structured, dynamic and semantically rich knowledge system, thereby providing uniform and high-quality data support and analysis basis for other modules within the system, such as intelligent template module, risk assessment and price limit module, etc. The implementation of this module lays a solid foundation for subsequent intelligent analysis, risk warning and even system self-adaptive optimization.
[0025] The function implementation of the knowledge base construction module mainly involves two closely linked technical stages of data collection and knowledge processing.
[0026] In the data collection stage, the knowledge base construction module is configured to systematically collect and integrate multi-source heterogeneous data related to the whole life cycle of authorized procurement. These data can be divided into the following main categories according to their sources and properties: First, historical procurement data, which is symbolized as This part of data constitutes the basis for the system to learn from history, and its content preferably includes but is not limited to: complete service quantity list in historical procurement projects, specific technical parameter specification documents, signed contract texts, records of highest price limit and final bid price of previous procurement, supplier's basic information archives, and post-project performance evaluation reports. The purpose of collecting historical procurement data is to provide a benchmark for the system to analyze and trace back, so that it can extract potential patterns and risk patterns from past successful or failed cases.
[0027] Second, real-time market data, which is symbolized as In order to ensure the timeliness and accuracy of system decision-making, this module actively interfaces with third-party data platforms or public market information sources through standard application program interfaces or customized network information collection technologies to obtain real-time or quasi-real-time industry-related price and cost information. These information preferably include market prices of key materials, labor cost indices of different types of work, and market average prices of similar services, etc. The introduction of real-time market data enables the subsequent dynamic price limit model to break away from the dependence on static and outdated data, thus being more close to the current market environment.
[0028] Finally, the data source that is also the core of the adaptive optimization capability of the present application is the performance process data, which is symbolized as The part of data is various process records generated in the actual implementation process of the project after the signing of the procurement contract. Both structured records and a large amount of unstructured text information are included, such as the project weekly reports submitted by the project manager regularly, the emails exchanged for clarifying technical details, the meeting minutes of the parties involved, the design change orders generated due to requirement changes, the photo records with text descriptions attached when problems are found on site, and the delivery delay reports due to causes, etc. The systematic collection of performance data The fundamental purpose is to capture the deviation between expectation and reality, and to provide the most direct and unprocessed raw input for the subsequent dynamic feedback and adaptive optimization mechanism.
[0029] After completing data collection, the system enters the knowledge processing phase. The goal of this phase is to use advanced natural language processing techniques to deeply process and intelligently organize the massive and heterogeneous data collected, and ultimately build a dynamic and computable knowledge base to represent the knowledge in symbolic .
[0030] In this embodiment, the core technology of knowledge processing is the use of large language models for entity and relationship extraction. Compared with traditional rule-based or keyword-based extraction methods, large language models have stronger context understanding and semantic generalization capabilities. This model can accurately identify predefined business entities such as project items, technical specifications, risk clauses, responsible parties, delivery nodes, and cost components from unstructured text in and , and further analyze the complex semantic associations between these entities.
[0031] After the entities and relationships are effectively extracted, the knowledge base construction module will perform the steps of knowledge fusion and storage. This step aims to align, deduplicate, and associate information from different sources to form a unified knowledge view. The module will link the supplier entities extracted from the contract text with the performance entities in the performance evaluation report and the relevant quotation information in real-time market data. The fused knowledge is preferably stored in the form of a knowledge graph or a vector database. Knowledge graphs are good at explicitly expressing complex relationships between entities, facilitating path analysis and causal reasoning; while vector databases efficiently support semantic similarity calculation by converting entities and text into high-dimensional vectors, which is crucial for subsequent semantic risk identification and other functions.
[0032] The construction and continuous updating process of the entire knowledge base can be abstractly described as a complex processing function that takes various raw data as input and outputs a highly structured and semantic knowledge system. This process can be generally expressed by the following formula: ; in: : Represents the final completed knowledge base that is continuously updated dynamically. : This represents a composite function that encompasses a range of data processing operations, including data cleaning, entity extraction, relation alignment, vectorization, knowledge fusion, and storage. : Represents the collection of historical procurement data. : Represents the collection of real-time market data. : Represents the collection of data related to the performance process.
[0033] Through the above methods, the knowledge base construction module not only completes a one-time data aggregation, but also establishes a dynamic knowledge hub that can be continuously enriched and evolved over time and with the influx of new project data. It connects previously isolated and silent data silos, transforming them into valuable assets that can serve the entire risk management system and support its core intelligent and adaptive functions.
[0034] The intelligent template module is configured to generate service quantity lists and technical parameter templates based on a knowledge base, and to perform real-time semantic risk identification and early warning when the user inputs procurement requirement text; The intelligent template module is a key technological component for achieving proactive risk prevention and source control. This module's function is not isolated but tightly coupled with the aforementioned knowledge base construction module, providing high-quality input for subsequent risk assessment and price control modules. Its core purpose is to transform the procurement requirement preparation process from a manual process reliant on personal experience and prone to errors into a standardized, interactive process with real-time intelligent feedback, thereby enabling intervention at the initial stage of risk formation.
[0035] The intelligent template module is implemented mainly through the synergistic effect of two technologies: standard template generation and real-time semantic risk identification.
[0036] First, regarding the standard template generation function. This function aims to provide a validated, structured starting point for different types of procurement projects. In this embodiment, the intelligent template module is configured to invoke and deeply analyze the knowledge base generated by the aforementioned knowledge base construction module. .
[0037] The module will use a preset algorithm to search the knowledge base. The historical high-quality procurement project documents are screened out. The high-quality here is not a subjective judgment, but based on a series of quantifiable indicators, the actual performance cost does not exceed the budget, the project delivery period meets or is better than the plan, and the number of technical disputes and changes generated during the performance process is very small. Through the induction and pattern mining of the service quantity list and technical parameter specification of these high-quality projects, the system can automatically extract and form standardized service quantity list and technical parameter templates, which are represented by symbols .
[0038] The value of the template lies in the fact that it solidifies the core service items, key technical parameters and recommended reference rules that must be included in a specific type of procurement project. This ensures the completeness and basic standardization of the procurement list at the organizational level, effectively avoiding the problem of project omission due to personal negligence, and laying a solid and consistent foundation for subsequent procurement activities.
[0039] The intelligent template module in this embodiment also has an interactive and real-time risk identification capability, which is the core innovation point that distinguishes it from traditional static templates. This function is activated during the user's actual preparation of the procurement demand document.
[0040] When the user is preparing the procurement demand document and inputting custom text descriptions, such as a functional requirement in the technical parameter section, the system's built-in semantic risk identification model (represented by symbols ) will perform an instant analysis on the input text and quantify its potential semantic risks.
[0041] The core working principle of this model is not simply keyword matching, but deep semantic understanding and comparison. The semantic risk identification model is based on a large language model. Its specific working method is as follows: first, the model will convert the user's real-time input text (represented by symbols ) into a high-dimensional semantic vector through embedding technology. At the same time, using the large number of historical dispute cases, risk clauses that have caused cost overruns, and various ambiguous descriptions stored in the aforementioned knowledge base , these risk knowledge has also been pre-converted into semantic vectors and stored in the vector database. The model calculates the similarity or distance between the semantic vector of the user's input text and the vast number of risk semantic vectors in the knowledge base to estimate the potential risk level.
[0042] This semantic risk quantification process can be generally expressed by the following formula: ; Where: : represents the calculated semantic risk score. The score is a normalized value, preferably with a value range of [0, 1], and the higher the value, the greater the semantic ambiguity, uncertainty, or potential controversy risk of the input text. : represents a semantic analysis model based on a large language model. Its implementation not only relies on the capabilities of the model itself, but more importantly, relies on a comprehensive and high-quality knowledge base as the factual basis for its risk comparison. : represents any text segment currently input by the user in the interface, which is to be analyzed. : represents the dynamic knowledge base generated by the aforementioned knowledge base construction module, which is the knowledge source for semantic risk identification.
[0043] When the calculated semantic risk score exceeds the dynamically configurable warning threshold, the system will immediately perform two coordinated actions. First, in the user interface, the system will automatically highlight the risky text, underline it, or pop up a prompt box, directly informing the user that there is a potential problem here.
[0044] Secondly, and more importantly, the system does not stop at warning. It will further use the semantic vector of the input text to perform semantic retrieval in the knowledge base , find and recommend one or more quantitative descriptions in history that have been proven to be clear, unambiguous, and effective in performance as a modification suggestion.
[0045] Through the organic combination of the above-mentioned standard template generation and real-time semantic risk identification functions, the intelligent template module greatly moves the risk prevention and control point to the most initial stage of demand preparation. It not only ensures the structural integrity of the demand, but also through real-time interaction, converts ambiguous, uncertain, and unstructured descriptions into structured and quantifiable risk signals. This not only educates and guides users to form more standardized preparation habits, but also provides more accurate and clean input for subsequent risk assessment and price limit modules, thereby effectively reducing the possibility of post-fulfillment disputes and cost out of control due to unclear procurement source descriptions.
[0046] The risk assessment and price limit module is configured to receive authorized procurement demands to be approved, analyze the authorized procurement demands based on the results of the knowledge base and semantic risk identification, and output a comprehensive risk index and a dynamic maximum price limit. The risk assessment and price limit module plays the role of the core analysis and decision support engine of the system. It should be understood that this module is a key component that connects the upstream and downstream modules. It receives the relatively standardized procurement requirements processed by the upstream intelligent template module, and converts them into quantifiable risk indexes and price benchmarks, which are then output to the downstream collaborative review module. The fundamental purpose is to transform the traditional, experience-dependent procurement evaluation process into a data-driven, model-supported, and risk-visualized scientific decision-making process.
[0047] Upon receiving a procurement requirement to be reviewed, the risk assessment and price limit module is configured to perform two core tasks in parallel: one is to build a comprehensive risk assessment model to output a comprehensive risk index, and the other is to build a dynamic maximum price limit model to output a dynamic maximum price limit.
[0048] First, regarding the implementation of the comprehensive risk assessment model. This model aims to comprehensively and quantitatively assess the potential uncertainty of a procurement task from multiple dimensions. In this embodiment, the system calculates the comprehensive risk index by weighted summation of multiple predefined risk dimensions, denoted by .
[0049] This calculation process can be generally expressed by the following formula: ; Where: : represents the final output comprehensive risk index. This index is normalized to a specific interval to facilitate subsequent risk level classification. : represents the total number of risk dimensions considered by the system. : represents the weight coefficient of the th risk dimension, and satisfies . This weight reflects the importance of different types of risk in the overall assessment, and can also be dynamically adjusted as a learnable parameter of the subsequent adaptive optimization module. : represents the normalized risk value of the th risk dimension, which has the same value range as .
[0050] In order to achieve comprehensive assessment, the risk dimensions preferably cover the main aspects of procurement activities, which are specifically described as follows: An important dimension is the list completeness risk, denoted by . This risk aims to assess whether there are missing items in the current submitted service quantity list. Its calculation method is to compare the structure of the current list with the standard template of the corresponding project type generated by the aforementioned intelligent template module, and calculate the missing item risk value based on the difference between the two structures. The structured comparison is performed to quantify the differences in terms of quantity and importance.
[0051] Another core dimension is the technical parameter explicitness risk, denoted by . This risk directly takes the output of the upstream smart template module. The semantic risk score of all technical parameter descriptions in the procurement requirement is aggregated to obtain the overall technical parameter explicitness risk . This can be done by either taking a weighted average of the scores of all parameters or simply taking the maximum value. This ensures that the ambiguity of the source requirement is directly incorporated into the final risk assessment.
[0052] In addition, the model also includes the market fluctuation risk, denoted by . The calculation of this risk relies on the real-time market data and historical procurement data stored in the aforementioned knowledge base . The volatility of the relevant material or service is measured by the coefficient of variation of the historical market price time series, where represent the standard deviation and mean of the historical market price of the service, respectively.
[0053] In the case of a designated intended supplier, the model also introduces the supplier performance risk, denoted by . The calculation of this risk value is achieved by querying the historical performance records of the specific supplier in the knowledge base and based on multi-dimensional data such as historical delivery delay rate, number of quality problems, after-sales service evaluation, etc., a pre-set scoring model is used to comprehensively evaluate and obtain the risk value.
[0054] At the same time, the risk assessment and price limit module performs its second major task, which is the construction of the dynamic maximum price model. This model aims to provide a scientific, reasonable and risk-anticipatory procurement price ceiling, rather than a static, fixed budget. In this embodiment, the calculation of the dynamic maximum price integrates market, cost and the aforementioned risk assessment results.
[0055] The calculation process of this dynamic maximum price can be generally expressed by the following formula: ; Where: : represents the final output of the recommended dynamic maximum price. : represents the market price anchor price. This price is not simply the market inquiry price, but is obtained by combining the historical transaction price and real-time market data in the knowledge base .Time series analysis or regression modeling is performed to synthesize the predicted fair market price. : represents the cost-based reverse price. This price is formed by querying and calling the industry standard cost database stored in the knowledge base according to the cost components such as materials and labor in the procurement list, accumulating each item, and adding the standard industry average profit margin. , , : respectively represent the weight coefficients of the market price method and the cost method, and satisfy These two weights are configurable, allowing the system to flexibly adjust the emphasis of the two pricing methods according to different project types or data availability. : represents the key risk adjustment coefficient, which is non-negative. The role of this coefficient is to convert intangible procurement risks into tangible price considerations.
[0056] Furthermore, the risk adjustment coefficient is directly related to the aforementioned comprehensive risk index , forming a monotonically increasing function relationship, which can be represented as This means that when the comprehensive risk index is higher, the risk adjustment coefficient also increases, thereby reserving a larger risk buffer in the calculated maximum price . This allows the system to automatically and proactively raise the price warning line when facing high-risk projects, reminding decision-makers to pay attention to potential cost overrun risks.
[0057] In the above manner, the risk assessment and pricing module successfully converts a qualitative procurement requirement into two key quantitative indicators that can guide decision-making: the comprehensive risk index representing the risk level, and the dynamic maximum price representing the price benchmark. These two highly condensed pieces of information will then be pushed together to the collaborative review module, providing a solid data foundation and scientific basis for subsequent intelligent approval routing and efficient decision-making.
[0058] The collaborative review module is configured to match and execute the corresponding approval process for authorized procurement requirements based on the comprehensive risk index. The collaborative review module is the key hub connecting automated analysis and manual decision-making. The function of this module is to convert the quantitative analysis results output by the upstream risk assessment and pricing module into specific and executable business approval processes. The core purpose of its design is to introduce a differentiated processing mechanism based on risk, accurately focusing the limited and valuable manual approval resources on procurement tasks with high uncertainty, thereby achieving dual improvement in approval efficiency and control depth.
[0059] Upon receiving a pending procurement request containing a comprehensive risk index and dynamic ceiling price , the core function of the collaborative review module is to implement a risk-based intelligent routing mechanism.
[0060] Upon receiving a procurement request, the first task of the module is to compare the comprehensive risk index calculated by the risk assessment and price limit module with the risk classification threshold preset by the system, which can be flexibly configured by the administrator, denoted by . The result of this comparison will directly determine the type of approval workflow that the procurement request will enter next.
[0061] The judgment logic of this intelligent routing can be expressed in the following formula: ; Where: : represents the specific approval workflow matched and executed by the system for the current procurement request. : represents the comprehensive risk index output by the upstream risk assessment and price limit module, which quantifies the overall risk level of the current procurement request. : represents the risk classification threshold preset by the system, which is a critical value used to distinguish low-risk and high-risk tasks.
[0062] In this embodiment, the system configures differentiated processing methods for the above two different approval processes.
[0063] When the judgment result is , it indicates that the system assesses the procurement task as low-risk. At this time, the collaborative review module will guide it into the intelligent auxiliary process. In this process, the system aims to maximize the efficiency of the review and reduce unnecessary manual intervention. The system will automatically complete a series of routine parameter verification, remind the reviewer to pay attention to the standard key nodes in the contract, and based on the analysis results of the previous module, provide a one-click pass suggestion option. The design purpose of this process is to open a fast approval channel for routine procurement tasks with clear requirements, reasonable prices, and controllable risks.
[0064] On the contrary, when the judgment result is , it indicates that the system assesses that the procurement task has a high potential risk. In this case, the collaborative review module will automatically trigger the cross-department collaborative review process. The core design of this process is to introduce multiple professional perspectives to conduct centralized consultation and decision-making on high-risk points.
[0065] When the process is triggered, the system does not simply forward the approval task, but carefully packages and pushes a data package containing rich decision support information. The data package preferably includes: the original procurement requirement document, the final value of the dynamic maximum price calculated by the risk assessment and price limit module , the comprehensive risk index
[0066] The risk attribution analysis report is the key to efficient collaborative review. It deconstructs the abstract comprehensive risk index and clearly displays the specific scores and weights of each sub-risk dimension that constitutes the total risk in visual or textual form. The report further drills down to identify the specific reasons for a high score in a certain sub-risk dimension.
[0067] Subsequently, the system automatically pushes this data package containing in-depth analysis to all relevant decision-makers in functional departments, technical experts, financial personnel, legal consultants, etc. according to pre-set rules. This approach enables each decision-maker participating in the joint review to grasp the overall picture of the procurement task and its core risk points at the first time when opening the approval task. This aims to guide the decision-makers' attention from broad, unfocused review to precise analysis and decision-making on specific risk points, thereby significantly improving the relevance and effectiveness of collaborative review.
[0068] In summary, the collaborative review module achieves intelligent and differentiated allocation of approval resources through simple and effective risk threshold judgment. It ensures that low-risk tasks can be quickly processed, while high-risk tasks can be thoroughly reviewed with professional and data-supported analysis, thereby achieving a balance between risk control and operational efficiency at the system level.
[0069] The adaptive optimization module is configured to collect performance data during and after the procurement project, quantify the performance deviation, and update the model parameters used in the risk assessment and price limit module based on the performance deviation.
[0070] The adaptive optimization module constitutes the core innovation of the invention and is the key to the transition of the entire system from a static, rule-based expert system to an intelligent system that can evolve and continuously learn. The functional logic of this module is located at the end of the entire procurement risk management chain, but its role penetrates to every front-end model of the system. Its fundamental purpose is to establish a closed-loop feedback channel from actual post-performance results to pre-analysis and prediction models, so that the system can learn from each procurement practice.
[0071] The adaptive optimization module is implemented through three logically progressive technical steps: performance deviation quantification, deviation attribution analysis, and adaptive updating of model parameters.
[0072] This module operates from the beginning of the procurement project's performance and after its completion. First, the system performs a performance deviation quantification step. This step aims to objectively and multidimensionally quantify the gap between the expectations of the procurement phase and the reality of the performance phase. To this end, the module proactively accesses the performance process data collected and stored by the aforementioned knowledge base construction module. .
[0073] The module analyzes this data, extracts the actual performance results related to the project's core objectives, and compares them with the benchmarks set during the procurement phase. This allows for the calculation of a multi-dimensional performance deviation vector, symbolized by... express.
[0074] This performance deviation vector The composition of can be summarized by the following formula: ; in: : Represents the deviation vector that comprehensively reflects the overall performance of the procurement project. This represents cost deviation. It is calculated using performance data. The actual total cost of the project calculated in the middle Subtract the dynamic maximum price limit generated for the project by the aforementioned risk assessment and price limit module. ,Right now A positive cost deviation directly reflects the situation of cost overruns. This represents the project schedule deviation. It is calculated using the actual project delivery time. Subtract the planned delivery time ,Right now . : Represents quality deviation. This value is not a simple binary judgment, but rather a comprehensive quality performance score derived from the gap between the expected quality standard and the actual quality standard. It is calculated by quantifying and weighting factors such as the number of quality problem reports, severity level, and repair costs during the performance period.
[0075] Successfully quantified performance deviations Next, the system proceeds to the crucial steps of deviation attribution and gradient calculation. Simply knowing that deviations occurred is insufficient; the key is to identify the root causes of these deviations. In this embodiment, this attribution process is transformed into a rigorous mathematical problem: calculating the performance deviation. The gradient with respect to all learnable parameters in the system.
[0076] This step aims to intelligently allocate the macroscopic deviation of performance to the specific microscopic parameters that cause it, in proportion to their contribution. This process can be represented by an abstract attribution function : ; where, : represents the deviation of performance with respect to the gradient vector of all learnable parameters within the system. Each component of this gradient vector indicates the direction and magnitude by which the corresponding model parameter should be adjusted in order to reduce the future deviation of performance. : represents the complex deviation attribution function. In specific implementations, it can be realized by combining the causal inference algorithm with the backpropagation mechanism of the model itself. Its purpose is to accurately locate those parameters that contribute most to the current deviation in the vast parameter space. : represents the set of all learnable and optimizable parameters in the system. This is a crucial set, which preferably includes but is not limited to: the weights of each risk dimension in the risk assessment model ; the market method and cost method weights in the dynamic pricing model ; the internal parameters of the semantic risk identification model in the intelligent template module ; and the internal scoring parameters of the supplier performance risk assessment model , etc. : represents the dynamic knowledge base, providing contextual information and associated paths for attribution analysis.
[0077] Finally, after calculating the gradient vector , the system performs the step of adaptive updating of model parameters, thus completing the entire learning closed loop. This embodiment uses the gradient descent method or its more advanced variants to iteratively update the parameters.
[0078] This updating process can be generally expressed by the following formula: ; where: and : represent the state of the system parameters at the th and the th iteration, respectively. : represents the learning rate. It is a hyperparameter used to control the step size of each update of the model. A reasonable setting can ensure the stability of the system in the learning process, rather than violent oscillation.
[0079] Through this step, the adaptive and self-evolution capabilities of the present invention are realized. If the system finds that a procurement project containing a certain specific ambiguous clause has a deviation vector Cost deviation If the value is always significantly positive, then through the attribution and update mechanism described above, the semantic risk model related to this ambiguous clause... The internal parameters will be automatically adjusted, resulting in a higher semantic risk score when encountering similar expressions in the future. Similarly, if a supplier's historical projects are consistently accompanied by significant timeline deviations... Its supplier performance risk model The rating parameters will also be automatically lowered.
[0080] Please see the appendix Figure 2 A risk management method for authorized procurement includes the following steps: S1. Collect multi-source heterogeneous data related to authorized procurement and build a knowledge base; S2. Based on the knowledge base, generate a service volume list and technical parameter template, and perform real-time semantic risk identification and early warning when receiving authorized procurement requests pending approval; S3. Based on the knowledge base and semantic risk identification results, analyze the authorized procurement requirements and output a comprehensive risk index and dynamic maximum price limit; S4. Based on the comprehensive risk index, match and execute the corresponding approval process for authorized procurement needs; S5. During and after the performance of the procurement project, collect performance process data, quantify the performance deviation, and update the model parameters used in step S3 in reverse based on the performance deviation.
[0081] The method in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so it will not be described again here.
[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A risk management system for authorized procurement, characterized in that, The method comprises the following steps: The knowledge base construction module is configured to collect multi-source heterogeneous data related to authorized procurement and construct a knowledge base; The intelligent template module is configured to generate service quantity list and technical parameter templates based on the knowledge base, and perform real-time semantic risk identification and early warning when a user inputs a procurement demand text; The risk assessment and price limiting module is configured to receive an authorized procurement demand to be approved, analyze the authorized procurement demand based on the knowledge base and the result of the semantic risk identification, and output a comprehensive risk index and a dynamic maximum price limit; The collaborative joint review module is configured to match and execute a corresponding approval process for the authorized procurement demand according to the comprehensive risk index; The adaptive optimization module is configured to collect performance data during and after the procurement project, quantify the performance deviation, and update the model parameters used in the risk assessment and price limiting module in a reverse direction according to the performance deviation.
2. The risk management system for authorized procurement according to claim 1, wherein, The knowledge base construction module is specifically configured to collect multi-source heterogeneous data including historical procurement data, real-time market data and performance data, and construct the knowledge base. 3.The risk management system for authorized procurement according to claim 1, wherein, The intelligent template module specifically completes the real-time semantic risk identification by comparing the semantic vector of the procurement demand text with the semantic vector of a historical controversial case recorded in the knowledge base.
4. The risk management system for authorized procurement according to claim 1, wherein, The risk assessment and price limiting module calculates the comprehensive risk index by weighted summation of multiple risk dimensions including list integrity risk, technical parameter definiteness risk and supplier performance risk.
5. The risk management system for authorized procurement according to claim 1, wherein, The risk assessment and price limiting module calculates the dynamic maximum price limit by combining market price anchoring price and cost backstepping price, and applying a risk adjustment coefficient determined by the comprehensive risk index.
6. The risk management system for authorized procurement according to claim 1, wherein, The collaborative joint review module is specifically configured to: compare the comprehensive risk index with a preset risk classification threshold value; trigger a cross-departmental collaborative joint review process as the corresponding approval process when the comprehensive risk index is greater than or equal to the risk classification threshold value.
7. The risk management system for authorized procurement according to claim 1, wherein, The adaptive optimization module quantifies the performance deviation including cost deviation, schedule deviation and quality deviation by analyzing the difference between the performance data and the procurement stage expectation.
8. The risk management system for authorized procurement according to claim 7, wherein, The adaptive optimization module completes the reverse update by calculating the gradient of the performance deviation with respect to the model parameters and iteratively updating the model parameters using the gradient descent method. 9.The risk management system for authorized procurement according to claim 8, wherein, The model parameters include at least the weights of the multiple risk dimensions.
10. A risk management method for authorized procurement, the risk management system for authorized procurement according to any one of claims 1-9, characterized in that, The method comprises the following steps: S1, collecting multi-source heterogeneous data related to authorized procurement and constructing a knowledge base; S2, generating service quantity list and technical parameter templates based on the knowledge base, and performing real-time semantic risk identification and early warning when receiving an authorized procurement demand to be approved; S3, analyzing the authorized procurement demand based on the knowledge base and the result of the semantic risk identification, and outputting a comprehensive risk index and a dynamic maximum price limit; S4, matching and executing a corresponding approval process for the authorized procurement demand according to the comprehensive risk index; S5, in the process of implementing the procurement project and after the end, collecting the implementation process data, quantifying the implementation deviation, and updating the model parameters used in step S3 according to the implementation deviation.
