A method for generating and evaluating procurement documents based on process mining analysis

Abstract

The application provides a procurement document generation and evaluation method based on process mining analysis, relates to the technical field of enterprise management, and comprises the following steps: collecting event log data in an enterprise procurement system, constructing a procurement process hypergraph model through a process hypergraph discovery algorithm, and generating a variation correlation topological structure; constructing a self-adaptive rule mapping network according to the procurement process hypergraph model and the variation correlation topological structure, inputting specific procurement demand data into the evolved template by utilizing the network to dynamically evolve the procurement document template, filling the file content by applying a semantic intervention filling mechanism, and generating a complete procurement document; the process hypergraph discovery algorithm is introduced to construct the procurement process hypergraph model, so that complex correlations such as multi-party interaction, group decision and high-order dependence in the procurement activity can be effectively captured, and the expression capacity of the process model is significantly improved compared with the traditional Petri net or simple directed graph model.

Description

Technical Field

[0001] This invention relates to the field of enterprise management technology, specifically to a method for generating and evaluating procurement documents based on process mining analysis. Background Technology

[0002] With the acceleration of enterprise digital transformation and the increasing complexity of supply chain management, the procurement process has become a critical link affecting operational efficiency, cost control, and risk prevention. Especially against the backdrop of globalized procurement, multi-supplier collaboration, and increasingly stringent compliance requirements, there is a greater demand for the automated generation, standardization, and intelligent evaluation of procurement documents. For example, large manufacturing enterprises need to quickly generate raw material contracts with complex terms, the service industry needs to adapt to frequently changing IT service agreements, retail enterprises need to handle flexible terms for seasonal goods procurement, and public institutions emphasize the transparency and compliance of tender documents. Currently, process mining technology and natural language processing technology are widely used in business process optimization and document automation.

[0003] However, existing procurement document generation and evaluation technologies still face the following key technical challenges in meeting the above requirements:

[0004] The procurement process model lacks expressive power. Traditional process mining often uses Petri nets or simple directed graphs, which can only capture linear sequences or basic branching relationships. They cannot effectively represent complex relationships such as multi-party interactions, group decision-making, and higher-order dependencies in procurement activities. This results in a large deviation between the model found from historical event logs and the actual procurement process, which in turn affects the accuracy and adaptability of subsequent document templates.

[0005] The lack of a dynamic matching mechanism between document generation and process model means that existing methods rely on fixed rule templates or simple keyword filling, which lacks adaptive adjustment to process variations and changes in requirements. It is impossible to achieve semantic-level intervention of clauses and comparison of historical paths, which can easily lead to clause conflicts, omissions of risk allocation content, or documents that are inconsistent with the actual execution path, seriously affecting the execution efficiency and legal validity of the contract.

[0006] The evaluation of procurement documents lacks the ability to proactively identify risks. Traditional evaluations are mainly based on post-event compliance checks or manual review, which makes it difficult to simulate path deviations under potential intervention scenarios and to quantify counterfactual risks (such as the consequences of supplier changes or delivery delays). This results in a low detection rate of potential deviation paths, and the evaluation reports lack structured intervention suggestions, making it difficult to support closed-loop optimization and continuous improvement.

[0007] Therefore, a procurement document generation and evaluation method based on process mining analysis is needed to solve the above problems. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a procurement document generation and evaluation method based on process mining analysis, which solves the problems of existing technologies.

[0010] Technical solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: a method for generating and evaluating procurement documents based on process mining analysis, comprising the following steps:

[0012] Sp1: Collect event log data from the enterprise procurement system, construct a procurement process hypergraph model through the process hypergraph discovery algorithm, and generate a mutation association topology;

[0013] Sp2: Based on the procurement process hypergraph model and mutation association topology generated by Sp1, construct an adaptive rule mapping network and use this network to dynamically evolve the procurement document template;

[0014] Sp3: Input specific procurement requirement data into the template evolved from Sp2, apply a semantic intervention filling mechanism to fill in the file content, and generate a complete procurement document;

[0015] Sp4: Using the counterfactual path checking technique of process mining, evaluate the procurement documents generated by Sp3, calculate the match between the document content and the counterfactual structure of the process model, and generate an evaluation intervention report.

[0016] Sp5: Based on the evaluation intervention report of Sp4, reconstruct the hypergraph model of the procurement process, the variation association topology, and the adaptive rule mapping network to achieve closed-loop self-organizing iteration of document generation and evaluation.

[0017] Preferably, in Sp1, procurement event sequence data is extracted from the enterprise procurement system through a multi-source event aggregation interface, a procurement process hypergraph model is constructed by using a process hypergraph discovery algorithm and embedding representation technology, and the mutated paths are converted into associated topological structures through a topology extraction component.

[0018] Preferably, in Sp2, based on the procurement process hypergraph model and variation association topology of Sp1, a graph neural network is used to capture hyperedge features, and the features are fused through the dynamic aggregation mechanism of an adaptive rule mapping network to evolve and generate a procurement document template containing branch constraints, sequence dependencies, and entity associations.

[0019] Preferably, in Sp3, the user-input procurement requirement data, including material attributes, transaction conditions, and performance terms, is mapped to the template fields of Sp2 through a semantic parsing component. The contract elements, negotiation terms, and risk allocation content are then filled in through an iterative process of generation and identification using a semantic intervention filling mechanism.

[0020] Preferably, in Sp4, the counterfactual replay technique of process mining is applied to convert the procurement documents of Sp3 into virtual event traces, which are then mapped onto the procurement process hypergraph model of Sp1. Counterfactual path matching is calculated through a structural intervention algorithm to generate an evaluation report containing deviation paths and intervention adjustments.

[0021] Preferably, in Sp5, based on the evaluation report of Sp4, a self-organizing reconstruction mechanism is used to adjust the node connections, edge weights of the mutated association topology, and cluster parameters of the adaptive rule mapping network of the procurement process hypergraph model, thereby achieving collaborative reconstruction of the model and the network.

[0022] Preferably, Sp1 further integrates a distributed consensus verification mechanism, which processes cross-system procurement event sequence data through an event verifier to ensure the integrity of the hypergraph model of the procurement process and the reliability of the variation association topology.

[0023] Preferably, in Sp2, a privacy isolation component is further embedded to encapsulate and protect the sensitive superedges of the adaptive rule mapping network, ensuring data isolation during template evolution.

[0024] Preferably, in Sp3, the procurement demand data is correlated and matched with previous virtual event traces by combining the historical intervention path comparison mechanism, thereby optimizing the structural consistency of semantic intervention filling.

[0025] Preferably, in Sp4, a hierarchical counterfactual framework is introduced, and the virtual event traces are checked for path conformity to the hypergraph model of the procurement process through a time-series intervention verification component, and the intervention adjustment path is output in a structured manner.

[0026] Beneficial effects

[0027] This invention provides a method for generating and evaluating procurement documents based on process mining analysis. It has the following beneficial effects:

[0028] 1. This invention constructs a procurement process hypergraph model by introducing a process hypergraph discovery algorithm, which can effectively capture complex relationships such as multi-party interactions, group decision-making, and higher-order dependencies in procurement activities. Compared with traditional Petri nets or simple directed graph models, it significantly improves the expressive power of the process model, making the model discovered from historical event logs closer to the actual procurement process, thereby improving the accuracy of subsequent document templates and their adaptability to various procurement scenarios.

[0029] 2. By constructing an adaptive rule mapping network and combining it with a semantic intervention filling mechanism and a historical intervention path comparison mechanism, this invention achieves dynamic matching and adaptive adjustment of document generation and process model. It can effectively cope with process variations and changes in requirements, avoid problems such as clause conflicts, omissions of risk allocation content, or inconsistencies with the actual execution path, thereby improving the execution efficiency and legal effect of procurement documents.

[0030] 3. By introducing counterfactual path checking technology, a hierarchical counterfactual framework, and a closed-loop self-organizing iterative mechanism, this invention endows the evaluation of procurement documents with the ability to identify forward-looking risks. It can simulate path deviations under potential intervention scenarios and quantify counterfactual risks, provide structured intervention suggestions, support continuous system optimization and improvement, significantly reduce the probability of potential deviation paths, and improve the risk controllability and overall reliability of the procurement process. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention;

[0032] Figure 2 This is a schematic diagram of the overall data flow structure of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0035] like Figures 1 to 2 As shown, a procurement document generation and evaluation method based on process mining analysis includes the following steps:

[0036] Sp1: Collect event log data from the enterprise procurement system, construct a procurement process hypergraph model through the process hypergraph discovery algorithm, and generate a mutation association topology;

[0037] Sp2: Based on the procurement process hypergraph model and mutation association topology generated by Sp1, construct an adaptive rule mapping network and use this network to dynamically evolve the procurement document template;

[0038] Sp3: Input specific procurement requirement data into the template evolved from Sp2, apply a semantic intervention filling mechanism to fill in the file content, and generate a complete procurement document;

[0039] Sp4: Using the counterfactual path checking technique of process mining, evaluate the procurement documents generated by Sp3, calculate the match between the document content and the counterfactual structure of the process model, and generate an evaluation intervention report.

[0040] Sp5: Based on the evaluation intervention report of Sp4, reconstruct the hypergraph model of the procurement process, the variation association topology, and the adaptive rule mapping network to achieve closed-loop self-organizing iteration of document generation and evaluation.

[0041] In Sp1, procurement event sequence data is extracted from the enterprise procurement system through a multi-source event aggregation interface. The procurement process hypergraph model is constructed by using a process hypergraph discovery algorithm and embedding representation technology. The mutated paths are converted into associated topological structures through a topology extraction component.

[0042] In Sp2, based on the hypergraph model of the procurement process and the mutated relational topology of Sp1, a graph neural network is used to capture hyperedge features. The features are then fused through the dynamic aggregation mechanism of an adaptive rule mapping network to evolve and generate a procurement document template that includes branch constraints, sequence dependencies, and entity associations.

[0043] In SP3, the procurement requirement data input by the user, including material attributes, transaction conditions and performance terms, is mapped to the template fields of SP2 through the semantic parsing component. The contract elements, negotiation terms and risk allocation content are filled in through the generation and identification iterative process of the semantic intervention filling mechanism.

[0044] In Sp4, the counterfactual replay technique of process mining is applied to convert the procurement documents of Sp3 into virtual event traces, which are then mapped onto the procurement process hypergraph model of Sp1. Counterfactual path matching is calculated through structural intervention algorithm to generate an evaluation report that includes deviation paths and intervention adjustments.

[0045] In Sp5, based on the evaluation report of Sp4, a self-organizing reconstruction mechanism is used to adjust the node connections, edge weights of the mutated association topology, and cluster parameters of the adaptive rule mapping network in the hypergraph model of the procurement process, so as to achieve collaborative reconstruction of the model and the network.

[0046] In Sp1, a distributed consensus verification mechanism is further integrated. Through event validators, cross-system procurement event sequence data is processed to ensure the integrity of the hypergraph model of the procurement process and the reliability of the mutated association topology.

[0047] In Sp2, a privacy isolation component is further embedded to encapsulate and protect the sensitive hyperedges of the adaptive rule mapping network, ensuring data isolation during template evolution.

[0048] In SP3, the historical intervention path comparison mechanism is combined to associate and match procurement demand data with previous virtual event traces, thereby optimizing the structural consistency of semantic intervention filling.

[0049] In SP4, a hierarchical counterfactual framework is introduced. The timing-based intervention verification component checks the path conformity of virtual event traces to the hypergraph model of the procurement process and outputs structured intervention adjustments to the path. Specific Implementation Example 2:

[0051] like Figures 1 to 2 As shown in this embodiment, a detailed implementation of a procurement document generation and evaluation method based on process mining analysis is provided. This method collects and processes data from an enterprise procurement system, constructs advanced models and networks, and achieves automated generation and evaluation of procurement documents, continuously optimizing through an iterative mechanism. An enterprise procurement system refers to an integrated software platform used by an enterprise to manage procurement activities, such as an enterprise resource planning system. This system records data across the entire process, including supplier management, order processing, contract signing, and payment, supporting multi-department collaboration and real-time data updates. The specific operation process of each step is described in detail below, including all extended details.

[0052] In step Sp1, event log data is first collected from the enterprise procurement system. As a comprehensive management platform, the enterprise procurement system stores various records of the procurement process, including supplier selection, price inquiries, order generation, contract review, and delivery acceptance. This event log data is automatically generated structured records, with each log entry containing a timestamp, activity name, executor identifier, and related attributes. For example, an order placement event might record the placement time, order number, supplier information, and material quantity. To collect this data, a multi-source event aggregation interface is used. This interface connects to the enterprise procurement system's database and application programming interface (API), pulling data from different modules through query and extraction mechanisms. Specifically, this interface supports the synchronous extraction of procurement event sequence data from the order management system, supplier database, and contract module. This sequence data is arranged chronologically, forming a complete event chain. Subsequently, a procurement process hypergraph model is constructed using a process hypergraph discovery algorithm. The algorithm first converts the collected event sequence data into node and hyperedge representations. Nodes represent procurement activities such as supplier selection or contract signing, while hyperedges represent complex relationships between multiple activities; for example, a hyperedge might connect supplier screening, price inquiry, and order generation to capture group interaction relationships. The algorithm iteratively scans the event sequence, identifies recurring patterns and mutation paths, and integrates embedding representation techniques to convert each node and hyperedge into a vector representation, enhancing the model's expressive power. Finally, a topology extraction component transforms mutation paths into an associative topology structure. This component analyzes anomalous paths in the model, such as delayed delivery or abnormal approvals, and maps these paths to connections in the topology graph, forming a network structure describing process mutations. Furthermore, this step integrates a distributed consensus verification mechanism, using an event validator to process cross-system procurement event sequence data. This validator coordinates across multiple data sources to ensure data consistency; for example, when data from different subsystems conflict, a voting mechanism confirms valid events, thereby guaranteeing the integrity of the procurement process hypergraph model and the reliability of the mutation-related topology structure.

[0053] In step Sp2, an adaptive rule mapping network is constructed based on the procurement process hypergraph model and variability association topology generated in step Sp1. This network is a dynamically adjusted structure used to transform the process model into a set of rules. First, key elements, such as activity sequences and branch conditions, are extracted from the hypergraph model. Then, network nodes are constructed, each corresponding to a rule, such as supplier selection rules or price negotiation rules. The procurement document template is dynamically evolved using this network. The network continuously updates rule weights through internal connections and adjusts the template framework according to the variability association topology to adapt the template to different procurement scenarios. Specifically, based on the procurement process hypergraph model and variability association topology from step Sp1, a graph neural network is used to capture hyperedge features. This network propagates information layer by layer, performs aggregation calculations on hyperedges, and extracts feature vectors representing complex relationships. Subsequently, these features are fused through the dynamic aggregation mechanism of the adaptive rule mapping network. This mechanism adjusts the aggregation method according to input changes during runtime, such as prioritizing high-frequency path features. Finally, a procurement document template containing branch constraints, sequence dependencies, and entity associations is generated. For example, the material description portion of the template is automatically linked to performance terms based on sequence dependencies. Furthermore, this step involves embedding a privacy isolation component to encapsulate and protect sensitive hyperedges in the adaptive rule mapping network. This component identifies hyperedges containing confidential information, such as supplier pricing data, during network evolution and isolates these components through an encryption layer, ensuring data isolation during template evolution and preventing the leakage of sensitive information.

[0054] In step Sp3, specific procurement requirement data is input into the template evolved in step Sp2. This requirement data is provided by the user and includes material attributes such as specifications and quantity, transaction terms such as price and payment method, and performance terms such as delivery time and quality standards. A semantic parsing component maps this data to template fields, analyzing the meaning of the data, such as identifying keywords for material attributes, and matching them to corresponding positions in the template. Subsequently, a semantic intervention filling mechanism is applied to fill in the document content. This mechanism progressively improves the content through an iterative process of generation and identification. First, a preliminary filled version is generated, then potential inconsistencies, such as conflicting terms, are identified and iteratively adjusted, ultimately producing a complete procurement document, including contractual elements such as the obligations of both parties, negotiated terms such as amendments, and risk allocation content such as liability for breach of contract. Furthermore, in this step, a historical intervention path comparison mechanism is used to correlate and match the procurement requirement data with previous virtual event traces. This mechanism extracts similar paths from historical records, such as past procurement indications of similar materials, and compares them with current data to optimize the structural consistency of the semantic intervention filling and ensure logical coherence of the filled content.

[0055] In step Sp4, the procurement documents generated in step Sp3 are evaluated using counterfactual path checking techniques from process mining. First, the procurement documents are converted into virtual event traces, which simulate the process paths after document execution, such as mapping contract terms to potential activity sequences. Then, these virtual event traces are mapped onto the procurement process hypergraph model from step Sp1. A structured intervention algorithm is used to calculate counterfactual path matching, assuming different intervention scenarios, such as changing supplier selection, and comparing the actual paths with the assumed paths to assess the degree of matching. Finally, an evaluation report containing deviation paths and intervention adjustments is generated, listing potential problems and optimization recommendations in the documents. Furthermore, a hierarchical counterfactual framework is introduced in this step, using a temporal intervention verification component to check the path conformity of the virtual event traces to the procurement process hypergraph model. This framework hierarchically analyzes paths, from basic sequences to complex branches, and outputs structured intervention adjustment paths, such as recommending modifications to specific terms to avoid deviations.

[0056] In step Sp5, based on the evaluation intervention report from step Sp4, the procurement process hypergraph model, mutation association topology, and adaptive rule mapping network are reconstructed. This reconstruction process uses a self-organizing reconstruction mechanism to adjust the node connections of the model, such as relinking active nodes to optimize paths; adjusting the edge weights of the mutation association topology, such as strengthening the weights of high-risk paths; and adjusting the cluster parameters of the adaptive rule mapping network, such as reorganizing rule clusters to adapt to new mutations. Through these adjustments, the model and network are reconstructed collaboratively, ensuring a closed-loop self-organizing iteration of document generation and evaluation. The entire process forms a feedback loop, continuously improving the accuracy and adaptability of the method. Specific Implementation Example 3:

[0058] like Figures 1 to 2 As shown, the algorithm described in Example 1 will be explained in detail below:

[0059] I. Hypergraph Discovery Algorithm in the Procurement Process (Corresponding Implementation Example—Sp1):

[0060] (a) Event log input data:

[0061] The event log collection obtained from the procurement system is denoted as:

[0062] ;

[0063] in: This represents a complete collection of procurement event logs; Indicates the number of procurement process instances contained in the log; Indicates the first Each procurement process instance corresponds to a complete or partially complete procurement execution process. Modeled as an ordered sequence of events:

[0064] ;

[0065] in: Indicates the first The number of events contained in a process instance; This indicates the first step in the process. The procurement event that occurred.

[0066] A single procurement event is defined as a triple:

[0067] ;

[0068] in: Indicates the type of procurement activity, such as requirement confirmation, supplier selection, contract review, etc. This indicates the timestamp of the procurement activity, used to depict the sequence of events; This represents the attribute vector associated with the procurement activity, including structured information such as the amount range, material category, and approval role.

[0069] The specific problem solved: By using the above modeling method, the originally messy procurement system logs are transformed into computable structured input data, providing a mathematical foundation for subsequent process mining and model building.

[0070] (ii) Formula for constructing the hypergraph of the procurement process:

[0071] The procurement process is modeled as a hypergraph structure:

[0072] ;

[0073] in: This represents a hypergraph model of the procurement process. This represents a set of nodes, where each node corresponds to a type of procurement activity. ; Represents the set of superedges; This represents the hyperedge weight function.

[0074] The definition of a hyperedge is:

[0075] ;

[0076] in: This refers to a set of multiple preceding procurement activities; Indicates the set of predecessor activities Subsequent procurement activities triggered by the same event.

[0077] This structure originates from hypergraph theory and is used to solve the problem that ordinary directed graphs can only represent "one-to-one" or "one-to-many" relationships and cannot express "multi-condition joint triggering".

[0078] (III) Calculation of Hyperedge Weights:

[0079] The superedge weight is defined as:

[0080] ;

[0081] in: Indicates the superedge The weight value; This indicates that all procurement process instances simultaneously include a set of preceding activities. The number of processes; the numerator is the number of process instances that satisfy the joint conditions; the denominator is the number of processes. This represents the total number of log processes.

[0082] This formula is derived from a frequency-based process mining model and is used to characterize the stability of joint activities in actual procurement execution.

[0083] Parameter range and meaning: The larger the value, the more stable the joint triggering relationship.

[0084] Specific problem addressed: This weight is used to distinguish between "high-frequency joint approval processes" and "occasional abnormal processes," providing a basis for subsequent template evolution.

[0085] (iv) Formula for measuring process variation:

[0086] Define process variability as:

[0087] ;

[0088] in: Indicates the first The degree of variation of a process relative to a reference process; This represents the main procurement process obtained through statistics. Indicates the sequence length or the number of elements.

[0089] Parameter value range: ;

[0090] The closer it is to 1, the more significantly the process deviates from the main process.

[0091] when:

[0092] ;

[0093] in This process was identified as a variant process.

[0094] The specific problem addressed is to identify non-routine processes such as emergency procurement and special approvals, and incorporate them into subsequent templates and rule generation, rather than simply discarding them.

[0095] II. Adaptive Rule Mapping Network Algorithm (corresponding to Sp2):

[0096] (I) Node embedding initialization formula:

[0097] ;

[0098] in: Represents a node The initial vector representation; Indicates the type of procurement activity corresponding to the node; This represents an embedding function that maps discrete activities to continuous vectors.

[0099] The specific problem to be solved: Transforming symbolic procurement activities into vector representations that can participate in mathematical calculations.

[0100] (ii) Hypergraph Convolution Update Formula:

[0101] ;

[0102] in: Indicates the first Layer node representation; Indicates the presence of nodes All superedges; This indicates the number of nodes contained in the superedge; Represents the representation of other nodes in the hyperedge; This represents a non-linear activation function.

[0103] Formula Origin: This formula originates from the weighted aggregation concept of Hypergraph Neural Networks, used to fuse the context of joint activities. Specific Problem Solved: Enabling procurement template rules to learn the real business logic of "multiple approval nodes jointly influencing the document structure."

[0104] III. Semantic Intervention Fill Algorithm (corresponding to Sp3):

[0105] (I) Semantic similarity calculation formula:

[0106] ;

[0107] in: Indicates the similarity between the semantics of the requirement and the template fields; Represents the semantic vector of procurement requirements; The numerator represents the semantic vector of the template field; the numerator represents the inner product of the vectors; and the denominator is used for normalization.

[0108] Parameter range: when ,in This triggers the fill operation.

[0109] The specific problem to be solved: ensuring that the generated content is semantically consistent with the procurement requirements fields, rather than simply filling in the positions.

[0110] IV. Counterfactual Path Checking Algorithm (corresponding to SP4):

[0111] (a) Path matching scoring formula:

[0112] ;

[0113] in: This represents a sequence of virtual events derived from the procurement documents. Indicates a legal path that can be reached in the hypergraph; This represents the probability of an event.

[0114] The specific problem to be solved: Determining whether the generated procurement documents are "truly executable" in terms of process structure. Specific Implementation Example 4:

[0116] like Figures 1 to 2 As shown, the following are specific application scenarios:

[0117] Scenario 1: Application of raw material procurement in manufacturing enterprises

[0118] A manufacturing company needs to procure a batch of key components for production line assembly. The company's procurement system has accumulated a large amount of historical event log data. The user (procurement manager) initiates this method to generate and evaluate new procurement documents.

[0119] First, the user logs into the system interface, selects "Create New Procurement Task," and enters specific procurement requirements data, including material attributes (part model, material requirements, quantity of 5000 pieces per batch), transaction terms (maximum unit price, payment method of 30-day payment period), and performance terms (delivery period not exceeding 45 days, quality standards conforming to industry norms). This data serves as the initial input for step Sp3.

[0120] The system immediately executes step Sp1: It automatically connects to the enterprise procurement system (Enterprise Resource Planning platform) database via a multi-source event aggregation interface, extracting historical procurement event sequence data from the order management, supplier management, and contract management modules. This includes records of all similar component procurements within the past two years, such as supplier screening events, inquiry records, order placements, contract signings, and delivery / acceptance. Each event contains attributes such as timestamp, activity name, executor, supplier identifier, and amount. The interface first queries the most recent 1000 relevant event sequences, then uses a process hypergraph discovery algorithm to convert these events into nodes (activities) and hyperedges (multi-activity associations), constructing a procurement process hypergraph model. This model clearly displays normal paths (e.g., screening-inquiry-order-acceptance) and mutated paths (e.g., re-inquiry due to supplier stock shortages). Simultaneously, a topology extraction component generates a mutated association topology structure, highlighting high-frequency mutated paths such as delivery delays. Finally, a distributed consensus verification mechanism is integrated to vote on event validators across departments (e.g., procurement and finance departments), ensuring the consistency and reliability of all extracted event sequence data. The output is a complete hypergraph model and topology structure, serving as the basis for subsequent steps.

[0121] Moving to step Sp2: The system takes the hypergraph model and mutated relational topology output from step Sp1 as input, and uses a graph neural network to capture hyperedge features layer by layer, such as capturing the complex dependencies of the hyperedge "supplier screening + inquiry + order". Then, these features are fused through the dynamic aggregation mechanism of an adaptive rule mapping network, and the rule weights are adjusted according to the high-frequency path priority to dynamically evolve and generate a procurement document template. This template includes standardized branch constraints (such as multi-supplier selection clauses), sequence dependencies (such as inquiry must precede order), and entity associations (such as supplier information automatically linked to performance terms). Simultaneously, a privacy isolation component is further embedded to identify and encrypt sensitive hyperedges involving supplier quotations in the template, ensuring that trade secrets are not leaked during the evolution process, and finally outputting a dedicated template for component procurement.

[0122] Step Sp3 begins: The system directly injects the user's previously input procurement requirements into the template generated in Step Sp2. First, the system analyzes the requirement text using a semantic parsing component, identifying fields such as "material requirements" as quality clauses and "45-day delivery" as performance clauses. Then, a semantic intervention-based filling mechanism is applied for iterative generation and identification: a preliminary contract draft is generated, filling in contract elements (such as the rights and obligations of both parties), negotiation clauses (such as price adjustment mechanisms), and risk allocation content (such as delay penalties); next, potential conflicts are identified (such as discrepancies between delivery cycles and historical variations), and multiple iterative adjustments are made. Simultaneously, combined with a historical intervention path comparison mechanism, historical paths for similar components are extracted from the virtual event traces in Step Sp1, and correlation matching is performed to further optimize the consistency of the filled content, such as automatically adding historically common clauses for sharing transportation risks. Finally, a complete procurement contract document (PDF format) is output.

[0123] Step Sp4: Execution Evaluation. The system converts the complete procurement document generated in Step Sp3 into a virtual event trail, simulating the sequence of activities (e.g., on-time delivery or delay path) that would occur if the contract were executed. This is then mapped onto the procurement process hypergraph model from Step Sp1, and counterfactual path matching is calculated using a structural intervention algorithm. For example, it simulates the path difference "if an alternative supplier is selected," assessing the risk. An evaluation intervention report is generated, containing deviation paths (e.g., potential delays) and intervention adjustment suggestions (e.g., increased penalties). Simultaneously, a hierarchical counterfactual framework is introduced, progressively examining the system from the basic sequence layer to the complex branch layer. Through time-series interventions, the structured output of the verification components provides specific adjustment paths, such as suggesting modifications to delivery terms to mitigate risk.

[0124] Final step Sp5: Based on the evaluation report from step Sp4, the system automatically adjusts the node connections (e.g., strengthening links to high-risk activities), edge weights of the mutated relational topology (increasing the weight of delayed paths), and cluster parameters of the adaptive rule mapping network (reorganizing rule clusters to adapt to new risks) of the procurement process hypergraph model using a self-organizing reconstruction mechanism. After completing the collaborative reconstruction, the system notifies the user that the procurement document has been optimized and saves the updated model for use in the next procurement. This application enables manufacturing companies to go from manually drafting contracts to automatically generating and evaluating them in just a few minutes, significantly reducing the risk of supply chain disruptions due to omissions in clauses.

[0125] Scenario 2: IT service procurement application in service industry companies

[0126] A service company needs to purchase cloud storage services and technical support contracts. The user (IT department head) initiates this method, logs into the system interface, selects "Create Procurement Task", and enters specific procurement requirements data, including material characteristics (cloud storage capacity of 10TB, computing resources of high-performance virtual machines), transaction specifications (annual subscription fee, phased payment), and performance conditions (service availability requirements, data backup and recovery mechanisms).

[0127] The system immediately executes step Sp1: Connecting to the company's procurement system via a multi-source event aggregation interface, it extracts historical IT service procurement event sequence data from the service management, supplier profile, and contract execution modules. This includes records related to all past cloud service subscriptions, software licenses, and technical support contracts, encompassing events such as request submission, supplier evaluation, service negotiation, contract signing, and service acceptance. Each event includes a timestamp, activity type, responsible department, and technical parameters. The interface queries the most recent 800 relevant sequences, converting them into nodes and hyperedges using a process hypergraph discovery algorithm to construct a procurement process hypergraph model. This model highlights mutation paths such as service renewals and service interruptions. Simultaneously, a topology extraction component generates a mutation-related topology structure to identify high-frequency anomalies such as service response delays. Finally, a distributed consensus verification mechanism is integrated to coordinate event validator voting across IT and legal departments, ensuring data consistency and outputting a reliable hypergraph model and topology structure.

[0128] Step Sp2: The system takes the output of Step Sp1 as input and uses a graph neural network to capture hyperedge features, such as the multi-party interaction relationship of the hyperedge "demand assessment + supplier negotiation + service level agreement signing". Features are fused through the dynamic aggregation mechanism of an adaptive rule mapping network, and rule weights are adjusted based on historical service interruption paths to dynamically evolve and generate a procurement document template. This template includes branch constraints (such as multi-cloud backup options), sequence dependencies (such as negotiations must precede payment terms), and entity associations (such as service parameters automatically linking to penalties). Simultaneously, a privacy isolation component is further embedded to identify and encrypt sensitive hyperedges involving intellectual property and technology rates, ensuring data security during the evolution process, and ultimately outputting a dedicated template for IT services.

[0129] Step Sp3 begins: The system injects the user-input procurement requirements into the template generated in Step Sp2. First, the requirements are analyzed using a semantic parsing component, for example, identifying "data backup and recovery mechanism" as a performance clause field and "phased payment" as a transaction specification field. Then, a semantic intervention filling mechanism is applied for generation and identification iteration: first, a preliminary service contract draft is generated, filling in contract components (such as service scope and obligations), negotiation elements (such as escalation clauses), and risk distribution content (such as data breach liability); then, conflicts are identified (such as inconsistencies between availability requirements and historical outages), and iterative adjustments are made. Simultaneously, combined with a historical intervention path comparison mechanism, historical paths similar to cloud services are extracted from the virtual event traces in Step Sp1, and correlation matching is performed to optimize the filled content, such as automatically adding historically common disaster recovery clauses. Finally, a complete IT service procurement contract document is output.

[0130] Step Sp4 Evaluation: The system converts the contract generated in Step Sp3 into a virtual event trace, simulating the sequence after execution (such as normal service or interruption recovery path). Mapped onto the hypergraph model of Step Sp1, counterfactual path matching is calculated using a structural intervention algorithm, for example, simulating risk scenarios such as "if service availability decreases." An evaluation intervention report is generated, containing deviation paths (such as potential interruptions) and intervention adjustment suggestions (such as adding compensation mechanisms). Simultaneously, a hierarchical counterfactual framework is introduced, checking from the basic service layer to the complex recovery layer, and verifying the structured output adjustment path of components through time-series interventions, such as suggesting strengthening data security clauses.

[0131] Final step Sp5: Based on the report from step Sp4, the system uses a self-organizing reconstruction mechanism to adjust the hypergraph model node connections (strengthening interrupted recovery links), topology edge weights (improving service risk paths), and rule-mapping network cluster parameters (reorganizing to adapt to technology upgrades). After reconstruction is complete, the user is notified that contract optimization is complete and the model should be saved. This application helps companies respond quickly to changes in technology requirements and avoid business interruptions caused by service contract vulnerabilities.

[0132] Scenario 3: Application of seasonal merchandise procurement by retail enterprises

[0133] A retail company needs to purchase seasonal promotional apparel. The user (the merchandise purchasing specialist) initiates this method, logs into the system interface, selects "Create Purchasing Task," and enters specific purchasing requirements data, including product type (summer clothing series, diverse styles), quantity (a total of 20,000 pieces, delivered in batches), transaction terms (promotional price range, bulk discount), and fulfillment terms (fast delivery cycle, return policy, and quality inspection standards).

[0134] The system immediately executes step Sp1: Connecting to the retail procurement system via a multi-source event aggregation interface, it extracts historical seasonal merchandise procurement event sequence data from merchandise management, the supplier database, and the sales module. This includes records of all past promotional apparel procurement, encompassing events such as market research, supplier selection, sample confirmation, bulk orders, order placement and production, and shelf placement. Each event includes a timestamp, event name, product identifier, quantity, and sales feedback. The interface queries the most recent 1200 relevant sequences, converting them into nodes and hyperedges using a process hypergraph discovery algorithm to construct a procurement process hypergraph model. This model highlights variability paths such as order concentration during peak demand periods and inventory backlogs. Simultaneously, a topology extraction component generates a variability association topology structure, identifying high-frequency anomalies such as abnormally high return rates or delivery delays. Finally, a distributed consensus verification mechanism is integrated to coordinate event validator voting between cross-store sales data and headquarters procurement data, ensuring data consistency and reliability, and outputting a complete hypergraph model and topology structure.

[0135] Step Sp2: The system takes the output of Step Sp1 as input and uses a graph neural network to capture hyperedge features, such as the multi-stage interaction relationship of the hyperedge "sample confirmation + bulk order + promotional terms negotiation". Features are fused through the dynamic aggregation mechanism of an adaptive rule mapping network, and rule weights are adjusted according to historical inventory backlog paths to dynamically evolve and generate a procurement document template. This template includes branch constraints (such as multi-batch delivery options), sequence dependencies (such as sample confirmation must precede production orders), and entity associations (such as automatic linking of product styles to return policies). Simultaneously, a privacy isolation component is further embedded to identify and encrypt sensitive hyperedges involving supplier production costs and discount rates, ensuring the security of trade secrets during the evolution process, and finally outputting a dedicated template for seasonal products.

[0136] Step Sp3 begins: The system injects the user-input procurement requirement data into the template generated in Step Sp2. First, the requirements are analyzed using a semantic parsing component, for example, identifying "return policy" as a risk allocation field and "fast delivery cycle" as a performance clause field. Then, a semantic intervention filling mechanism is applied for generation and identification iteration: First, a preliminary draft of the goods procurement contract is generated, filling in contract elements (such as product specifications and obligations), negotiation clauses (such as bulk discount mechanisms), and risk allocation content (such as sharing of unsold inventory); then, potential conflicts (such as delivery cycles not matching historical peak periods) are identified, and multiple iterations are performed for adjustment. Simultaneously, combined with a historical intervention path comparison mechanism, historical paths similar to promotional clothing are extracted from the virtual event traces in Step Sp1, and correlation matching is performed to further optimize the consistency of the filled content, such as automatically adding historically common unsold inventory return compensation clauses. Finally, a complete seasonal goods procurement contract document is output.

[0137] Step Sp4 Execution Evaluation: The system converts the contract generated in Step Sp3 into a virtual event trail, simulating the sequence after execution (such as normal shelf sales or inventory backlog paths). Mapped onto the hypergraph model of Step Sp1, counterfactual path matching is calculated using a structural intervention algorithm, for example, simulating an inventory risk scenario of "declining market demand." An evaluation intervention report is generated, containing deviation paths (such as potential backlog) and intervention adjustment suggestions (such as adding flexible return clauses). Simultaneously, a hierarchical counterfactual framework is introduced, progressively checking from the basic order layer to the complex sales feedback layer. The structured output of the verification component through time-series interventions provides specific adjustment paths, such as suggesting modifications to quantity batching clauses to reduce risk.

[0138] Final step Sp5: Based on the report from step Sp4, the system automatically adjusts the hypergraph model node connections (e.g., strengthening inventory feedback links), edge weights of the mutated relational topology (increasing the weight of backlog paths), and cluster parameters of the adaptive rule mapping network (reorganizing rule clusters to adapt to market demand fluctuations) using a self-organizing reconstruction mechanism. After completing the collaborative reconstruction, the system notifies the user that the procurement document has been optimized and saves the updated model for use in the next promotional season. This application helps retail enterprises effectively manage seasonal procurement, reduce inventory backlog and capital occupation, and improve the overall efficiency of promotional activities.

[0139] Scenario 4: Application of equipment bidding and procurement in public institutions

[0140] A public institution needs to purchase a batch of office equipment (such as printers and computers) and conducts open bidding. The user (the person in charge of bidding) initiates this method, logs into the system interface, selects "Create Procurement Task", and enters specific procurement requirements data, including equipment technical parameters (printer resolution, energy consumption standards, computer configuration), budget constraints (total amount limit), transaction conditions (open bidding process, fair competition requirements), and performance terms (delivery and acceptance standards, audit trail and warranty period).

[0141] The system immediately executes step Sp1: Connecting to the organization's procurement system via a multi-source event aggregation interface, it extracts historical equipment bidding and procurement event sequence data from the bidding management, bid evaluation, and contract execution modules. This includes records of all past office equipment bidding, such as bidding announcements, supplier registration, bid review, award notifications, contract signing, and equipment acceptance. Each event contains a timestamp, activity name, participant identifier, bid amount, and review comments. The interface queries the most recent 600 relevant sequences, converting them into nodes and hyperedges using a process hypergraph discovery algorithm to construct a procurement process hypergraph model. This model highlights variant paths such as strict compliance reviews and bid objections. Simultaneously, a topology extraction component generates a variant association topology structure to identify high-frequency anomalies such as insufficient bids or review disputes. Finally, a distributed consensus verification mechanism is integrated to coordinate event validator voting across regulatory and financial departments, ensuring data consistency and compliance, and outputting a complete hypergraph model and topology structure.

[0142] Step Sp2: The system takes the output of Step Sp1 as input and uses a graph neural network to capture hyperedge features, such as the multi-party participation relationship of the hyperedge "tender announcement + bid review + contract signing". Features are fused through the dynamic aggregation mechanism of an adaptive rule mapping network, and rule weights are adjusted according to historical compliance variation paths to dynamically evolve and generate a procurement document template. This template includes branch constraints (such as multi-round review options), sequence dependencies (such as the announcement must precede the bid deadline), and entity associations (such as technical parameters automatically linking to acceptance criteria). Simultaneously, a privacy isolation component is further embedded to identify and encrypt sensitive hyperedges involving bidders' business information and quotations, ensuring the evolution process meets confidentiality requirements, and finally outputting a dedicated template for public bidding.

[0143] Step Sp3 begins: The system injects the user-input procurement requirements into the template generated in Step Sp2. First, the requirements are analyzed using a semantic parsing component, identifying fields such as "fair competition requirements" as transaction conditions and "audit trail" as performance terms. Then, a semantic intervention-based filling mechanism is applied for iterative generation and identification: First, a preliminary tender document and draft contract are generated, filling in contract elements (such as equipment specifications and obligations), negotiation terms (such as post-bid details), and risk allocation content (such as liability for breach of contract); then, potential conflicts (such as budget discrepancies with historical bids) are identified, and multiple iterative adjustments are made. Simultaneously, combined with a historical intervention path comparison mechanism, similar historical bidding paths for equipment are extracted from the virtual event traces in Step Sp1, and correlation matching is performed to further optimize the consistency of the filled content, such as automatically adding historically common transparency audit clauses. Finally, a complete tender document and draft contract are output.

[0144] Step Sp4: Execution Evaluation. The system converts the files generated in Step Sp3 into virtual event traces, simulating the post-execution sequence (e.g., successful bid delivery or objection handling path). Mapped onto the hypergraph model of Step Sp1, counterfactual path matching is calculated using a structural intervention algorithm, such as simulating a compliance risk scenario of "insufficient bidding suppliers." An evaluation intervention report is generated, containing deviation paths (e.g., potential disputes) and intervention adjustment suggestions (e.g., extending the announcement period). Simultaneously, a hierarchical counterfactual framework is introduced, progressively checking from the basic announcement layer to the complex review layer. The structured output of the verification component through time-series interventions provides specific adjustment paths, such as suggesting strengthening fair review clauses to improve transparency.

[0145] Final step Sp5: Based on the report from step Sp4, the system automatically adjusts the hypergraph model node connections (e.g., strengthening review objection links), edge weights of the mutated relational topology (increasing compliance risk path weights), and cluster parameters of the adaptive rule mapping network (reorganizing rule clusters to adapt to regulatory requirements) using a self-organizing reconstruction mechanism. After completing the collaborative reconstruction, the system notifies the user that the tender document has been optimized and saves the updated model for future institutional procurement. This application helps public institutions improve procurement transparency and efficiency, ensuring compliance with regulatory standards.

[0146] As can be seen from the detailed scenarios above, this method starts with user input requirements, gradually automates the processing of historical data, generates and evaluates documents, and finally achieves closed-loop optimization, making it suitable for procurement management environments of various sizes and industries. Specific Implementation Example 5:

[0148] like Figures 1 to 2 As shown, in this embodiment, a comparative experiment was conducted to verify the effectiveness of a procurement document generation and evaluation method based on process mining analysis. The experimental environment was a simulated enterprise procurement system, with 300 procurement event logs collected as historical data, and 12 different types of procurement requirements randomly generated as test cases. The test cases were divided into three categories: raw material procurement (4 cases), IT service procurement (4 cases), and seasonal goods procurement (4 cases).

[0149] The experiment compared this method (the proposed method) with two traditional methods:

[0150] Traditional Method 1: Manual filling based on a fixed template + manual evaluation.

[0151] Traditional Method 2: Based on basic process mining (Petri net model only, without hypergraph and counterfactual evaluation) + rule template generation.

[0152] Evaluation indicators include:

[0153] File generation time (seconds): The time required from inputting the requirements to generating a complete file.

[0154] Clause Consistency Score (out of 100): Assessed by experts to assess the consistency between the document and historical processes.

[0155] Risk deviation detection rate (%): The proportion of potential risk paths detected.

[0156] Overall accuracy (%): The percentage of generated documents that pass compliance review.

[0157] The experimental results are shown in the table below (the data is the average of 12 test cases, grouped and statistically analyzed by category):

[0158] ;

[0159] Experimental data description:

[0160] Document generation time: The proposed method averages 29 seconds, which is about 90% shorter than traditional method 1 (308 seconds) and about 68% shorter than traditional method 2 (91 seconds). This is due to the adaptive rule mapping network dynamic evolution template in step Sp2 and the semantic intervention filling mechanism in step Sp3, which greatly reduces the time required for manual operation.

[0161] Clause consistency score: The proposed method averaged 92 points, an improvement of approximately 33% compared to traditional method 1 and 15% compared to traditional method 2. This is mainly attributed to the accurate modeling of complex procurement relationships by the process hypergraph model in step Sp1, and the optimization of clauses through historical intervention path comparison in step Sp3.

[0162] Risk deviation detection rate: The proposed method averages 90%, significantly higher than traditional method 1 (30%) and traditional method 2 (58%). The core advantage lies in the counterfactual path checking technology and hierarchical counterfactual framework in step Sp4, which can effectively simulate various risk scenarios (such as delivery delays, service interruptions, or inventory backlogs).

[0163] Overall accuracy: The proposed method achieves an average accuracy of 95%, and almost all generated documents pass compliance reviews, far exceeding traditional methods. This is attributed to the closed-loop self-organizing iteration in step Sp5, which ensures a high degree of alignment between the documents and the actual procurement process. Specific Implementation Example Six:

[0165] like Figures 1 to 2 As shown, the following provides supplementary explanations for terms that are not clearly defined in the above content:

[0166] Process Hypergraph Discovery Algorithm: This algorithm refers to process mining algorithms based on hypergraph theory. Unlike traditional process mining algorithms (such as Alpha or Inductive Miner), it is not limited to linear or binary branch representations; instead, it builds models by discovering higher-order associations. In procurement scenarios, this algorithm extracts multi-activity joint relationships from event logs. For example, it treats supplier selection, price inquiry, and approval as a hyperedge to capture the complexity of group decision-making, thereby improving the model's accuracy in representing actual processes.

[0167] The procurement process hypergraph model represents the procurement process as a hypergraph structure H = (V, E), where V is the set of nodes (each node represents a procurement activity, such as supplier selection or contract signing), and E is the set of hyperedges (each hyperedge connects multiple nodes, representing joint triggering relationships, such as multi-department approval). In procurement scenarios, this model is used to represent higher-order dependencies; for example, a hyperedge can connect three activities to simulate parallel negotiation, thus solving the multi-party interaction problem ignored by traditional graph models.

[0168] Variational Association Topology: Variational associations refer to abnormal patterns that deviate from the main path in the procurement process (such as delayed delivery). The topology refers to the network graph representation formed by these variational paths, including connections between nodes and edge weights. In procurement scenarios, this structure is generated by analyzing abnormal events to highlight risky paths and provide a basis for subsequent template adjustments, ensuring that the method adapts to non-standard procurement situations.

[0169] Adaptive Rule Mapping Network: This network is a dynamic neural network structure, similar to a variant of a graph neural network. "Adaptive" refers to the real-time adjustment of rule weights through a learning mechanism to respond to changes in input. In procurement scenarios, this network maps process models to a set of rules (such as clause generation rules), achieving template evolution through feature aggregation, thereby handling varying requirements without manual redesign.

[0170] Semantic intervention-based content filling mechanism: Semantic intervention refers to the introduction of semantic analysis during the content filling process to proactively adjust and optimize the content. The difference between this and ordinary keyword filling lies in its use of a generation-identification iteration to handle semantic conflicts. In procurement scenarios, this mechanism ensures that terms (such as risk allocation) conform to semantic logic, for example, iteratively correcting inconsistent performance conditions and improving the coherence and reliability of document content.

[0171] Counterfactual path checking / counterfactual replay: Both refer to the same technique, which uses a structured causal model to simulate "what if the intervention were different" scenarios. A "counterfactual path" refers to a virtual execution path under hypothetical changes (such as supplier replacement). In procurement scenarios, this technique is used to evaluate potential risks in documents, such as replaying contract execution to quantify deviations, thereby providing forward-looking recommendations.

[0172] Layered Counterfactual Framework: This framework divides counterfactual analysis into multiple layers (such as the basic sequence layer, branch intervention layer, and overall path layer), with each layer progressively assessing the impact of hypothetical scenarios. In procurement scenarios, it performs layered checks from simple clause changes to the impact of complex processes, ensuring that the evaluation report comprehensively covers risk dimensions.

[0173] Self-organizing reconstruction mechanism: Self-organization refers to a mechanism that automatically adjusts based on feedback, similar to a self-organizing map network, optimizing parameters (such as node connections, edge weights, and cluster parameters) through unsupervised learning. In procurement scenarios, this mechanism uses evaluation reports as input to dynamically reorganize the model to adapt to new variations, achieving closed-loop optimization without external intervention.

[0174] Distributed consensus verification mechanism: This mechanism uses consensus algorithms (such as simple majority voting or Raft variants) to process cross-system data and ensure consistency. In procurement scenarios, it coordinates multi-source events (such as ERP and CRM systems), filters conflicting data through event validators, and improves the reliability of model building.

[0175] Privacy Isolation Component: This component isolates sensitive information through encryption or access control methods (such as differential privacy or data anonymization). In procurement scenarios, it encapsulates sensitive hyperedges (such as supplier quotations) in the rule network to prevent data leakage without affecting the template evolution process.

[0176] Historical intervention path comparison mechanism: An intervention path refers to a virtual sequence in historical records simulating interventions. This mechanism compares current needs with historical paths using similarity calculations (such as sequence matching algorithms). In procurement scenarios, it optimizes consistency, for example, by matching similar past contracts to avoid omissions of terms.

[0177] Timing-based intervention verification component: This component uses timing logic (such as Linear Timing Logic (LTL)) to verify the path conformity of virtual event traces and examine the impact of interventions in chronological order. In procurement scenarios, it ensures that evaluations consider timing factors, such as the cascading effects of delivery delays, and outputs structured adjustment recommendations.

[0178] Virtual event trail: This term refers to a simulated sequence of events derived from procurement documents, mapping document terms to potential execution paths (such as a chain of activities from contract to delivery). In procurement scenarios, it is used to bridge document and process models, providing the input basis for counterfactual evaluation.

[0179] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0180] 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 method for generating and evaluating procurement documents based on process mining analysis, characterized in that: Includes the following steps: Sp1: Collect event log data from the enterprise procurement system, construct a procurement process hypergraph model using the process hypergraph discovery algorithm, and generate a mutation association topology; extract procurement event sequence data from the enterprise procurement system through the multi-source event aggregation interface, construct a procurement process hypergraph model using the process hypergraph discovery algorithm and embedding representation technology, and convert mutation paths into association topology structures through the topology extraction component; Sp2: Based on the procurement process hypergraph model and mutation association topology generated by Sp1, construct an adaptive rule mapping network and use this network to dynamically evolve the procurement document template; Sp3: Input procurement requirement data into the template evolved from Sp2, apply a semantic intervention filling mechanism to fill the document content, and generate a complete procurement document; map the procurement requirement data input by the user, including material attributes, transaction conditions and performance terms, to the template fields of Sp2 through the semantic parsing component, and apply the semantic intervention filling mechanism to fill the contract elements, negotiation terms and risk allocation content through the generation and identification iterative process. Sp4: Using the counterfactual path checking technique of process mining, evaluate the procurement documents generated by Sp3, calculate the match between the document content and the counterfactual structure of the process model, and generate an evaluation intervention report. Sp5: Based on the evaluation intervention report of Sp4, reconstruct the hypergraph model of the procurement process, the variation association topology, and the adaptive rule mapping network to achieve closed-loop self-organizing iteration of document generation and evaluation.

2. The procurement document generation and evaluation method based on process mining analysis according to claim 1, characterized in that: In Sp2, based on the hypergraph model of the procurement process and the mutated relational topology of Sp1, a graph neural network is used to capture hyperedge features. The features are then fused through the dynamic aggregation mechanism of an adaptive rule mapping network to evolve and generate a procurement document template that includes branch constraints, sequence dependencies, and entity associations.

3. The procurement document generation and evaluation method based on process mining analysis according to claim 1, characterized in that: In Sp4, the counterfactual replay technique of process mining is applied to convert the procurement documents of Sp3 into virtual event traces. The virtual event traces refer to the simulated event sequences derived from the procurement documents. The document terms are mapped to potential execution paths and mapped onto the procurement process hypergraph model of Sp1. The counterfactual path matching is calculated through the structural intervention algorithm to generate an evaluation report containing the deviation path and intervention adjustment.

4. The procurement document generation and evaluation method based on process mining analysis according to claim 1, characterized in that: In Sp5, based on the evaluation report of Sp4, a self-organizing reconstruction mechanism is used to adjust the node connections, edge weights of the mutated association topology, and cluster parameters of the adaptive rule mapping network in the hypergraph model of the procurement process, so as to achieve collaborative reconstruction of the model and the network.

5. The procurement document generation and evaluation method based on process mining analysis according to claim 1, characterized in that: In Sp1, a distributed consensus verification mechanism is further integrated. Through event validators, cross-system procurement event sequence data is processed to ensure the integrity of the hypergraph model of the procurement process and the reliability of the mutated association topology.

6. The procurement document generation and evaluation method based on process mining analysis according to claim 2, characterized in that: In Sp2, a privacy isolation component is further embedded to encapsulate and protect the sensitive hyperedges of the adaptive rule mapping network, ensuring data isolation during template evolution.

7. The procurement document generation and evaluation method based on process mining analysis according to claim 3, characterized in that: In SP3, the historical intervention path comparison mechanism is combined to associate and match procurement demand data with previous virtual event traces, thereby optimizing the structural consistency of semantic intervention filling.

8. The procurement document generation and evaluation method based on process mining analysis according to claim 3, characterized in that: In SP4, a hierarchical counterfactual framework is introduced. The timing-based intervention verification component checks the path conformity of virtual event traces to the hypergraph model of the procurement process and outputs structured intervention adjustments to the path.

Images