Intelligent approval rule modeling method oriented to process automation

Abstract

The invention discloses an intelligent approval rule modeling method oriented to process automation, and relates to the technical field of business process management, and the method comprises the following steps: S100, in a process of constructing a rule candidate set, extracting scene features, field semantic hierarchy and participation role information of each piece of historical approval data, generating a context semantic tag set, and establishing a rule candidate set; the method is used for subsequent rule difference modeling. According to the method, context semantic tags are introduced to be aligned with ternary features, so that the semantic boundary recognition capability of the rule is enhanced; constructing a rule feature matrix and a differentiation candidate set, and realizing accurate classification and processing of ambiguity rules; in combination with expression sensitivity enhancement and simulation verification, approval offset and risk are identified in advance; finally, the dynamic optimization of the rule model is realized through backtracking correction, the stability and accuracy of the rule model in multiple scenes are improved, and a closed-loop credible intelligent approval rule system is constructed.

Description

Technical Field

[0001] This invention relates to the field of business process management technology, and more specifically to an intelligent approval rule modeling method for process automation. Background Technology

[0002] "Intelligent Approval Rule Modeling for Process Automation" refers to the structural modeling and intelligent rule extraction of approval processes within an enterprise or organization by integrating process modeling techniques with artificial intelligence methods. This enables the automatic generation, optimization, and application of approval rules. In traditional approval processes, rules often rely on manual setting, resulting in rigid rules, delayed responses, and difficulty in adapting to complex business changes. This method intelligently analyzes historical approval data, business process nodes, and user behavior to automatically identify approval conditions, judgment logic, and constraint relationships. It then constructs a formalized and executable rule model, which can be embedded into a process automation system to drive the automatic judgment and execution of subsequent processes. This improves approval efficiency, accuracy, and flexibility, achieving a closed-loop intelligent process governance system from "rule design—execution—feedback—optimization."

[0003] The existing technology has the following shortcomings:

[0004] In the process of intelligent approval rule modeling for process automation, rule extraction and structured modeling are typically carried out based on a large amount of historical approval data to achieve automatic summarization and formal expression of approval conditions, judgment logic, and execution paths. However, in actual modeling, due to the large number of samples with similar structures, semantics, or partially overlapping condition combinations in historical approval records, current intelligent rule extraction algorithms often mistakenly identify rules with key detail differences as equivalent redundant items when performing redundancy compression or semantic fusion processing due to insufficient feature recognition granularity or limited context awareness, thus performing inappropriate merging or deletion operations.

[0005] This type of misjudgment mainly manifests as follows: under the same main framework, ignoring the differences in approval strategies caused by additional restrictions, boundary conditions, or contextual features, thus mistakenly aggregating multiple independent rules with fine-grained differences into a single logical path. Such erroneous modeling directly leads to the loss of differentiated constraints in the original approval rule system and a decrease in the accuracy of the approval path. Especially in process scenarios involving multi-role collaboration, sensitive element judgments, or high-risk constraints, this can result in serious consequences such as decentralization of authority, process detours, or failure of risk control mechanisms, severely impacting the execution accuracy of the intelligent approval system and the compliance stability of organizational governance.

[0006] For example, taking the corporate funding approval process as an example, the historical data contains the following two rules:

[0007] Rule A: When the purchase amount is greater than or equal to RMB 1 million and the supplier is a domestic enterprise, it must be approved by the general manager;

[0008] Rule B: When the purchase amount is greater than or equal to RMB 1 million and the supplier is an overseas company, it must be approved by the board of directors.

[0009] The two rules are highly consistent in the main condition "amount ≥ 1 million RMB", but they differ fundamentally in the additional condition "supplier type" and the corresponding approval level requirements. The existing rule modeling algorithm, during the extraction process, may only capture the common logic of the same amount threshold in the two rules, ignoring the difference in the additional constraint "domestic / overseas," thus mistakenly merging them into a single general rule: "Amount ≥ 1 million RMB requires general manager approval." This erroneous merging delegates the authority for high-value overseas procurement processes, which should be approved by the board of directors, to the general manager level, creating a mismatch in approval paths and an overreach of organizational responsibility. This could lead to serious consequences such as significant compliance risks, foreign exchange outflows, failure of international supply chain management, and ineffective internal audit accountability.

[0010] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0011] The purpose of this invention is to provide an intelligent approval rule modeling method for process automation. By introducing contextual semantic tags and ternary feature alignment, the semantic boundary recognition capability of rules is enhanced; a rule feature matrix and a differentiated candidate set are constructed to achieve accurate classification and processing of ambiguous rules; combined with expression sensitivity enhancement and simulation verification, approval deviations and risks are identified in advance; and finally, the rule model is dynamically optimized through backtracking correction to improve its stability and accuracy in multiple scenarios, thus constructing a closed-loop and reliable intelligent approval rule system to solve the problems mentioned in the background technology.

[0012] To achieve the above objectives, the present invention provides the following technical solution: an intelligent approval rule modeling method for process automation, comprising the following steps:

[0013] S100, in the process of constructing the rule candidate set, extract the scene features, field semantic hierarchy and participating role information of each historical approval data to generate a set of context semantic tags for subsequent rule difference modeling;

[0014] S200, based on the contextual semantic label set, performs feature alignment on rule expression, extracts main condition fields, contextual additional attributes and decision actions, and constructs a rule feature matrix as the basic structure for rule comparison and filtering;

[0015] S300, based on the rule feature matrix, identifies rule combinations with the same main condition field but different expression methods, extracts ambiguous additional attributes, and generates a rule differentiation candidate set;

[0016] S400, based on the rule-differentiated candidate set, embeds key additional attributes into the original rule structure, performs expression-sensitive enhancement processing, explicitly labels detailed elements that are easily ignored by compression algorithms, and generates an enhanced rule model set;

[0017] S500, based on an enhanced set of rule models, performs multi-path simulations by combining historical approval process data. It simulates process paths under different rule configurations, identifies approval deviations and logical mismatches caused by differences in rule expression, and outputs simulation comparison results and risk warning information.

[0018] Based on simulation comparison results and risk warning information, S600 determines the execution stability of the rule model in various context scenarios, initiates consistency backtracking and structural correction for detected offsets and conflicts, verifies context labels, adjusts rule expressions and semantic tag configurations, and completes closed-loop optimization of the rule model.

[0019] Preferably, step S100 includes:

[0020] During the extraction of each historical approval data, the scene characteristics, field semantic hierarchy, and participating role information corresponding to the approval behavior are collected.

[0021] Based on approval logs, business process definitions, organizational structure, business metadata and unstructured text, perform enhanced extraction of execution context data to obtain timestamps, process stages, job levels, business lines and semantic information of remarks;

[0022] The collected fields are semantically categorized, and the field types and expression levels are divided according to the preset business logic knowledge base;

[0023] Generate a structured set of contextual semantic tags, organized in the form of "tag dimension, tag key, tag value" triples, and bind them one by one with historical approval data.

[0024] Preferably, step S200 includes:

[0025] Based on the contextual semantic tag set, the logical expression of each rule is structured and parsed to identify judgment fields, status fields and role fields;

[0026] Extract three key elements: main condition field, contextual additional attributes, and decision action, and construct a unified ternary expression format;

[0027] The three types of elements are aligned according to a preset field template to construct a rule feature matrix;

[0028] Normalize the feature matrix of the rules and add semantic weights to improve the semantic discrimination ability in rule comparison.

[0029] Preferably, step S300 includes:

[0030] Semantic element recognition and structural segmentation are performed on each historical approval data, and the main condition fields, contextual additional attributes, and decision actions are identified based on the contextual semantic tag set.

[0031] The three types of key elements identified are mapped to predefined field positions to construct a rule feature matrix, and missing fields are handled with null values ​​or default unconditional handling.

[0032] Perform expression normalization processing on the rule feature matrix to standardize the expression form of semantically equivalent fields;

[0033] Based on the strength of the field's role in decision-making, semantic weight adjustment is performed to identify rule combinations that have the same main condition fields but different contextual additional attributes, extract ambiguous additional attributes, and generate a rule differentiation candidate set.

[0034] Preferably, step S400 includes:

[0035] Based on the rule-differentiated candidate set, we identify rule combinations that have the same main condition field but semantic differences in contextual additional attributes, and extract key additional attributes that are easily ignored by compression algorithms.

[0036] The extracted key additional attributes are embedded into the corresponding original rule structure, keeping the logic unchanged and explicitly expanding the content of expression;

[0037] The updated rule structure is subjected to expression-sensitive enhancement processing, and the semantic weights and saliency markers of key detail elements are annotated.

[0038] The processing results are summarized to generate an enhanced set of rule models.

[0039] Preferably, step S500 includes:

[0040] Filter representative historical approval process data and extract the input fields, process node sequence, historical decision path and context labels for each data point;

[0041] For the same approval input, the original rule model set and the enhanced rule model set are applied to perform process simulations to deduce their respective approval paths and decision results;

[0042] Compare the differences in process paths under the two rule configurations, identify approval deviations and logical mismatches, and analyze whether there is delegation of authority, overstepping of role authority, or omission of key nodes.

[0043] Output a simulation comparison report that includes path differences, offset types, and risk levels, and generate corresponding risk warning information.

[0044] Preferably, step S600 includes:

[0045] The joint analysis process simulation output identifies approval offset paths and logical conflict nodes, and identifies problem types that repeatedly occur in multiple context combination scenarios.

[0046] Based on the contextual semantic tags and semantic weight identifiers embedded in the rule model, we trace back the key fields that caused the offset and their contextual sources to determine the reasons for expression defects or insufficient modeling.

[0047] Corrections are made to rule models with structural problems, including adding contextual attributes, optimizing field nesting structures, and refining semantic tag content;

[0048] The revised rule model was re-simulated and verified to confirm that it has semantic stability and execution consistency under the same context.

[0049] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0050] This invention introduces a contextual semantic labeling mechanism and a ternary structure feature alignment strategy to ensure that rules have clear semantic boundaries and structural expressions from the initial modeling stage, significantly enhancing the ability to identify fine-grained differences. By constructing a rule feature matrix and a rule differentiation candidate set, it further enables accurate classification and processing of rule combinations with ambiguity and contextual conflicts, avoiding the misjudgment of semantic overlap as equivalent rules. Combined with an expression sensitivity enhancement mechanism and simulation verification process, potential path deviations and decision conflicts can be identified before rule execution, providing early warning of approval risks. Finally, through consistency backtracking and structural correction processes, dynamic self-optimization of the rule model is achieved, ensuring its execution stability and expression accuracy in different contextual scenarios. Overall, this scheme not only improves the accuracy and reliability of rule modeling but also constructs a closed-loop system for rule generation, verification, and optimization, providing a highly controllable and reliable rule foundation for intelligent decision-making in a process automation environment. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0052] Figure 1 This is a flowchart of the intelligent approval rule modeling method for process automation according to the present invention. Detailed Implementation

[0053] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0054] This invention provides, for example Figure 1 The intelligent approval rule modeling method for process automation shown includes the following steps:

[0055] S100, during the process of constructing the rule candidate set, when extracting each piece of historical approval data, the scene features, field semantic level and participating role information corresponding to the data are recorded, and the corresponding context semantic label set is generated for subsequent structured modeling of rule differences;

[0056] To achieve comprehensive semantic feature perception of historical approval data and avoid misjudgments in rule merging due to missing contextual information during rule modeling, a method for generating a set of contextual semantic tags is proposed. This method is used to construct a candidate set of rules while simultaneously enhancing the context of each historical approval data entry. Specifically, it includes the following steps:

[0057] While collecting historical approval record data, enhanced contextual data extraction is performed. Specifically, when parsing each approval record, not only are its core logical fields (such as conditional parameters, judgment values, and operational actions) extracted, but contextual data directly related to the approval behavior is also extracted simultaneously. This includes, but is not limited to, the business line identifier to which the approval business belongs, the current stage label of the approval process, the business timestamp of the data generation, the organizational level of the initiator and approver, and the region or area identifier where the business was initiated. This method ensures that each approval data entry not only contains executable logical elements but also carries contextual semantic clues that reflect the actual application background. The key to this step is achieving a two-way binding between approval behavior data and its contextual information, ensuring reproducibility and contextual reproducibility during subsequent rule extraction.

[0058] Contextual data enhancement extraction can be achieved in several ways, including: extracting the timestamp, geographical location, and device information of the approval action based on the log system; obtaining the process stage and link type of the node by combining the business process definition file; parsing the job level and responsibilities of the initiator and approver from the organizational structure; extracting the business line, approval item category, and applicable rule version through the business metadata system; and using natural language processing methods to perform semantic analysis on unstructured text such as approval remarks and operation instructions to identify potential contextual elements. These methods can enhance the contextual expression of approval data from multiple dimensions, improving the accuracy and discriminative ability of rule modeling.

[0059] Based on the extraction of contextual information, semantic hierarchical classification is performed on the fields involved in the approval data. This process, based on a pre-defined business logic knowledge base, categorizes fields semantically (e.g., numerical judgment fields, enumeration-type status fields, role / identity fields, etc.) and groups them according to semantic precision levels. For example, "amount" is a first-level quantitative field, "amount > 1 million yuan" is a second-level threshold expression field, and "amount > 1 million yuan and is an overseas supplier" is a third-level semantic linkage field. This semantic hierarchical classification helps to accurately identify the differences in expression granularity between rules, especially when there are subtle differences in the expression of field combinations. Hierarchical labeling can significantly improve the accuracy of semantic recognition and avoid misidentifying fine-grained differences as expression redundancy.

[0060] Based on the extracted contextual information and semantic hierarchy classification results, a unified set of contextual semantic tags is constructed for each historical approval data entry. This set is organized in a structured manner, with each tag consisting of a triple of "tag dimension, tag key, tag value." The tag dimension includes scenario dimensions (e.g., business line, time period), field dimensions (e.g., field level, field function type), and role dimensions (e.g., initiating role, approving role). For example, the contextual semantic tag set in an approval data entry might contain entries such as: "Scenario dimension: Business line = Overseas procurement," "Field dimension: Amount field level = Threshold judgment," and "Role dimension: Approver identity = General Manager." This structured tag set not only preserves the original decision-making information of the approval data but also embeds the semantic context of the decision-making process, enabling subsequent rule extraction to perform differential modeling and comparative analysis based on semantic reference.

[0061] Each historical approval data point and its generated contextual semantic tag set are bound one-to-one and stored and managed uniformly as the basic unit of the rule candidate set. Based on this, subsequent modeling steps such as rule feature comparison, difference identification, and combination optimization can utilize the corresponding semantic tag set as an auxiliary information source to determine whether there are actual contextual differences between rules, thus avoiding decision path deviations caused by relying solely on the main condition field for merging. Furthermore, this contextual semantic tag set supports dynamic expansion and feedback optimization. For example, when rule mismatch issues are discovered during simulation verification, the existing tag set can be updated in reverse to further improve the comprehensiveness and discriminative power of the tag representation, forming an evolutionary semantic-aware modeling mechanism.

[0062] During the construction of the rule candidate set, when extracting each piece of historical approval data, the associated scenario features, field semantic levels, and participating role information are recorded simultaneously, and a corresponding set of contextual semantic tags is generated. Its core function is to provide refined and structured contextual support for subsequent rule modeling. Traditional rule extraction methods often focus on the explicit logic of condition fields and result actions, neglecting the differences in the actual business environment, process stage, and organizational roles behind the approval behavior. This can easily lead to the erroneous merging of rules with similar structures but different semantics. By introducing contextual semantic tags, a reproducible and analyzable contextual reference system can be established for each approval behavior. This allows subsequent judgments on whether two rules have substantial differences to move beyond simply comparing field values, and instead to perform multi-dimensional comparisons based on their applicable business scenarios, semantic expression levels, and role participation relationships. This labeled contextual modeling approach improves the semantic granularity of rule expression, enhances the discriminative ability of rule merging and clustering, fundamentally avoids the problem of rule abstraction distortion caused by lack of context awareness, and ensures that the approval rules built in the process automation scenario are both executable and retain the necessary scenario constraints and strategy differences, significantly improving the accuracy and stability of rule modeling.

[0063] S200, based on the generated context semantic label set, performs feature alignment processing on the rule expression, extracts three key elements in each rule: main condition field, context additional attributes, and decision action, and constructs a rule feature matrix as a unified basic structure for comparison and combination filtering between rules;

[0064] To address the issues of inconsistent rule expression structures, inconsistent semantic granularity, and difficulty in accurate difference judgment and combination filtering in existing technologies, this paper proposes a rule expression feature alignment processing method based on a set of contextual semantic tags. By extracting three key elements—main condition fields, contextual additional attributes, and decision actions—a unified rule feature matrix is ​​constructed for subsequent rule comparison, aggregation, and difference analysis. The method specifically includes the following steps:

[0065] Based on the contextual semantic tag set previously generated for each historical approval record, the logical expression contained in the rule is structured and parsed. During the parsing process, all explicit condition fields in the rule are identified and categorized according to a predefined field classification dictionary. This includes judgment fields (such as amount, time, project level, etc.), status fields (such as supplier type, approval priority, etc.), and role fields (such as approver's rank, initiating department, etc.). Through this parsing process, the basic classification of all semantic components in the rule is established, laying the logical foundation for subsequent feature element segmentation.

[0066] After completing the field classification and parsing, the rule expression is extracted into three core feature elements. Specifically, fields reflecting the pre-triggered logic of the rule are extracted as main condition fields, which typically include core numerical or Boolean conditions used to trigger the judgment; conditions that accompany the main condition and are used to refine or limit the triggering scenario are defined as contextual attributes, such as role requirements, geographical restrictions, supplier background, etc.; and the final approval behavior, workflow actions, or result output determined by the rule are categorized as decision actions. Through this feature classification and extraction method, a structured semantic skeleton can be constructed from the original rule text, enabling all types of rules to be represented according to a unified ternary expression format, effectively improving the efficiency of rule comparison and filtering.

[0067] The "unified ternary expression format" refers to representing each rule as a structured expression consisting of three parts: a main condition field, contextual attributes, and a decision action. The core of this format is that regardless of the complexity or diversity of the original approval rule's expression, it must be abstracted and summarized into these three core elements: the main condition field describes the main judgment logic upon which the rule's validity depends (e.g., whether the amount exceeds a certain threshold); the contextual attributes capture the semantic context limiting the rule's applicability (e.g., supplier type, approver's position, business location, etc.); and the decision action clarifies the execution behavior after the rule is triggered (e.g., approval by a certain role, transfer to a certain node). The purpose of using the unified ternary expression format is to provide a unified semantic structure for subsequent rule alignment, comparison, clustering, and difference detection, transforming rules from "uncertain expressions at the natural language layer" into "standard entities at the structured semantic layer," thereby achieving efficient rule modeling, similarity calculation, and error merging avoidance. It not only improves the automation and comparability of rule modeling but also significantly enhances the model's ability to perceive subtle semantic differences, serving as a key foundation for building a high-precision, context-sensitive approval rule system.

[0068] After extracting the three key elements, the main condition fields, contextual attributes, and decision actions are aligned according to a preset structural template to construct a rule feature matrix. This feature matrix is ​​a two-dimensional table structure, with each row corresponding to a rule and each column corresponding to a standardized field position. Field positions include numerical judgment areas, enumeration state areas, semantic constraint areas, and execution action areas. During matrix construction, if a rule is missing a certain type of field, it is placed using "null" or "undefined" to maintain matrix structural consistency. This structured matrix not only improves the efficiency of rule data comparison but also supports subsequent vectorization processing, similarity calculation, and combined rule filtering, serving as the foundational structure for rule semantic clustering and difference detection.

[0069] After the feature matrix is ​​completed, the three types of elements of each rule are normalized and semantically weighted to improve the rule's responsiveness to fine-grained differences during comparison. For example, for the same field "amount," if the judgment values ​​of two rules are similar but not completely consistent, normalization can quantify their distance, helping to identify the judgment logic of "approximate rather than equivalent." For details in contextual attributes that are easily overlooked, such as approval roles and project levels, weighted annotations are used to increase their discriminative weight in rule comparison, preventing them from being incorrectly merged. Ultimately, the constructed rule feature matrix not only has good readability and consistency but also high efficiency and comparability, providing a unified structural foundation and semantic guarantee for subsequent operations such as rule clustering, ambiguous rule filtering, and combination judgment.

[0070] The main purpose of this step is to achieve the structural reconstruction and semantic alignment of historical approval rules, thereby providing a unified expression basis for accurate comparison, semantic difference identification, and combination screening between rules. In actual process automation scenarios, approval rules often originate from natural language descriptions, manual settings, or semi-structured records. These rules exhibit high heterogeneity in expression methods, field structures, and semantic granularity. Direct comparison or merging can easily lead to misidentification and mis-aggregation due to structural inconsistencies or semantic ambiguity. By performing feature alignment processing based on contextual semantic tag sets for rule expression, three core semantic elements can be accurately extracted from each rule: the main condition field is used to identify the core logical basis for rule triggering; contextual additional attributes are used to characterize the limited environment and business context in which the rule applies; and the decision action defines the execution path and processing behavior after the rule is triggered. These three elements constitute the skeleton of rule semantics. By uniformly extracting and mapping them to the rule feature matrix, standardized structural representation and accurate semantic comparison of rules can be achieved, avoiding the obscuring of actual differences due to expression differences or misjudging equivalence due to structural overlap. As the underlying structure for subsequent rule clustering, difference identification, and merging optimization, the feature matrix not only improves the accuracy and operability of rule modeling, but also enhances the ability to identify fine-grained semantic differences. It is a key step in achieving improved accuracy and risk control in intelligent approval rule modeling.

[0071] S300, based on the constructed rule feature matrix, identifies rule combinations that have semantic overlap but different expressions under the premise that the main condition fields are the same. It extracts the ambiguous additional attributes in the rule combination and generates a rule differentiation candidate set to distinguish rule expressions that may have differences in decision logic.

[0072] To address the difficulties in comparison, erroneous merging, and semantic ambiguity caused by inconsistent rule expressions and mixed semantic elements in existing rule modeling, a rule expression feature alignment method based on a contextual semantic tag set is proposed. By standardizing and aligning the rule expressions, key semantic elements in the rules are extracted into three unified core components, and a clear and semantically consistent rule feature matrix is ​​constructed as the foundation for rule comparison and combination selection. This method specifically includes the following steps:

[0073] Based on the contextual semantic tag set generated for each historical approval data entry, semantic element identification and structural segmentation are performed on the rule text. The core of this step is to use a semantic parsing strategy to classify the semantic roles of the content expressed in the rule, breaking it down into identifiable and categorizable independent elements. Specifically, a pre-defined semantic template and contextual tag matching mechanism are used to semantically anchor conditional judgments, business constraints, trigger statements, and behavioral actions in the rule text. For example, for the rule "When the purchase amount is greater than or equal to 1 million yuan and the supplier is an overseas company, it is approved by the board of directors," the contextual tags guide the identification of "amount greater than or equal to 1 million yuan" as the main condition, "supplier is an overseas company" as the additional constraint, and "approved by the board of directors" as the decision-making action. This step achieves the initial transformation from freely expressed text to structured semantic units, laying the semantic foundation for subsequent standardized expressions.

[0074] Based on semantic segmentation, key element extraction is performed, categorizing the semantic fragments in each rule into three core components: main condition fields, contextual attributes, and decision actions. The main condition fields refer to the core judgment conditions that trigger the rule's effectiveness, typically quantitative or qualitative logical judgments such as amount range, time frame, and approval category. Contextual attributes include semantic elements closely related to the business context, such as applicable organizational scope, participant identities, geographic tags, and risk levels, used to characterize the limited scenarios in which the rule applies. Decision actions are the specific behaviors after the rule is triggered, including the transfer recipient, approval method, and processing path. The extraction of all three types of elements relies on the semantic anchor and tag lookup table generated in the previous step, ensuring that all extracted content has clear semantic boundaries and business meaning.

[0075] After extracting the three types of elements, a unified rule feature matrix is ​​constructed using predefined field positions to ensure semantic alignment consistency across different rules. This matrix uses rules as rows and standard field positions as columns, corresponding to the main condition field set, contextual attribute set, and decision action set, respectively. Each field position defines a clear semantic dimension and standardized expression format, such as "amount judgment field," "regional label field," "approval position field," and "final approval path field." During rule mapping, if a rule lacks a certain type of field, a "null value" or "default unconditional" is used as a placeholder to maintain the integrity and alignment consistency of the matrix structure. This matrix not only supports structured storage but also facilitates semantic similarity calculation, feature weight analysis, and rule aggregation processing, serving as the core foundation for efficient and controllable rule management.

[0076] To enhance the sensitivity and accuracy of the rule feature matrix to semantic differences, a unified expression normalization and semantic weight adjustment process is performed on the constructed matrix. Expression normalization is mainly used to standardize fields that are semantically equivalent but express different meanings. For example, "amount greater than or equal to 1 million" and "amount ≥ 1 million" are normalized to the standard field "amount ≥ 1 million". Semantic weight adjustment assigns differentiated weights based on the degree of influence of fields in actual decision-making. For example, a higher recognition weight is assigned to the "supplier attribute" field in high-risk approval to avoid it being misjudged as an irrelevant condition in subsequent rule screening. Through this step, it is ensured that the rule feature matrix not only has the ability to align forms but also has the ability to sensitively respond to changes in semantic details, significantly improving the accuracy and recognition ability of rule comparison.

[0077] This step aims to provide a unified framework for the structured modeling and semantically accurate comparison of approval rules, effectively addressing inconsistencies in expression format, field hierarchy, and contextual information among existing approval rules. Since historical approval rules often originate from manual configuration, free text descriptions, or multi-source heterogeneous systems, while these rules may share similar logical intentions, they exhibit significant structural differences, making direct equivalence judgment or difference identification difficult. To address this issue, this step, based on the generated set of contextual semantic tags, performs in-depth analysis and feature alignment of the rule expression, clearly defining and extracting three key elements: main condition fields, contextual additional attributes, and decision actions. The main condition fields reflect the core basis for the rule's logical validity and serve as the main axis for rule triggering; contextual additional attributes characterize the restrictive environment in which the rule applies, covering information such as process nodes, role identities, geographical identifiers, and business scenarios; and decision actions define the operational path to be executed after the rule is triggered, such as approval role, flow direction, and processing method. Mapping these three types of elements uniformly into the rule feature matrix achieves structural consistency and semantic explicitness of rules, enabling rules with originally diverse expressions to be stored and compared isomorphically in the standard matrix. This approach not only improves the accuracy of rule comparisons and reduces the risk of misjudgments due to structural inconsistencies or semantic ambiguity, but also provides fundamental support for subsequent rule selection, combination optimization, clustering and merging, and ambiguity avoidance. It is an indispensable and crucial step in the transition from data to model in intelligent approval rule modeling.

[0078] S400, based on the rule differentiation candidate set, embeds the extracted key additional attributes into the corresponding original rule structure, performs expression sensitivity enhancement processing on the updated rule structure, and explicitly marks the detailed elements that are originally easily ignored by the rule compression algorithm in the rule model, generating an enhanced rule model set;

[0079] To address the issue of fine-grained semantic misjudgment that easily occurs during rule extraction, especially when compression algorithms mistakenly identify incomplete or low-weighted additional conditions as redundant, a rule enhancement method based on rule differentiation candidate sets is proposed. This method re-embeds key additional attributes into the rule structure and performs expression-sensitive enhancement processing, making previously easily overlooked details explicitly presented in the rule structure. This forms an enhanced rule model set, ensuring the semantic integrity of the rules and the accuracy of decision expression. Specifically, it includes the following steps:

[0080] Based on the previously constructed rule differentiation candidate set, this step identifies rule combinations with identical main condition fields but semantically divergent contextual attributes, and extracts key additional attributes that are easily overlooked by the merging algorithm. This step utilizes the previously constructed rule feature matrix to locate rule pairs with completely identical main condition fields but fine-grained differences in contextual additional attributes or decision actions, paying particular attention to additional elements with weakened semantic expression, ambiguous field content, or low weight, such as role scope restrictions, geographically applicable tags, and organizational hierarchical boundaries. Through semantic distance calculation and semantic conflict annotation mechanisms, these potentially easily lost detailed attributes are extracted, forming an additional attribute set associated with the original rule.

[0081] The extracted key additional attributes are embedded into the corresponding original rule structure to form the updated rule expression. The embedding process must maintain the original logical framework of the rule. While keeping the main condition fields unchanged, the additional attributes are explicitly presented in the rule's logical expression. Specifically, structural revisions can be achieved through methods such as nested condition statements, insertion of constraint fields, or semantic annotation expansion, ensuring consistency in the logical structure of the rules while making the expression clearer, more specific, and complete. For example, if the original rule only stated "amount ≥ 1 million yuan," the revised version adds "and the supplier is an overseas company, and the contract term exceeds 12 months" as an explicit additional condition.

[0082] The updated rule structure undergoes expression-sensitive enhancement processing to improve the rule model's ability to retain key details during extraction and compression. This processing adds semantic weight hints, field saliency markers, and contextual hint labels to the rule expression, highlighting additional attributes that are easily overlooked and strengthening their presence and incompressibility in the rule modeling algorithm. For example, weighted labels are added to conditions like "the approver must be a board member" to prevent them from being downgraded during parsing or aggregation; or explicit semantic classification is applied to expressions like "limited to overseas procurement processes" to prevent them from being misidentified as similar conditions in domestic scenarios. This mechanism ensures that the rules maintain key semantic stability throughout multiple rounds of modeling, reconstruction, and compression.

[0083] The set of rules, after undergoing key attribute embedding and expression sensitivity enhancement, is aggregated to generate an enhanced rule model set. This set retains the core logical structure of the original rules and enhances their contextual reconstruction, comparison and discrimination capabilities, and compression and anti-interference capabilities through explicit presentation of additional attributes and strengthening of semantic expression. The enhanced rule model not only provides higher discrimination accuracy in subsequent process simulation, path simulation, and difference detection, but also significantly improves the adaptability and reliability of rules in dynamic business scenarios. Furthermore, this set can serve as input for the rule feedback optimization process, enabling robust inheritance and accurate reconstruction of rule expressions across multiple rounds of strategy adjustments.

[0084] This step addresses the issue of insufficient semantic detail recognition during rule extraction and compression. By re-embedding key additional attributes into the original rule structure and performing sensitivity enhancement processing on the rule expression, a more complete and semantically clear enhanced rule model set is generated. In existing rule modeling practices, automatic extraction algorithms often aggregate and compress based on the commonalities of condition fields, lacking awareness of subtle semantic differences and contextual constraints. Especially when rules have inconsistent expressions, ambiguous boundaries of additional conditions, or low semantic weights, they are prone to misjudging fundamentally different rules as redundant, thus incorrectly merging them into a single rule, leading to mismatched approval process paths and ineffective risk control. This step reconstructs the original rules by identifying key additional attributes from the rule differentiation candidate set, embedding hidden or marginalized additional constraints into the original rule expression in an explicit and standardized structured form. Simultaneously, an expression sensitivity enhancement mechanism is introduced. Through semantic tags, high-weight annotations, and saliency hints, detailed elements that are easily ignored by modeling algorithms are given greater presence and incompressibility in the rule expression, thereby improving the semantic stability and recognition accuracy of the rule model in subsequent comparisons, clustering, and simulations. The resulting enhanced rule model set possesses higher expressive rigor and difference preservation capabilities, effectively avoiding erroneous judgments and decision deviations caused by rule abstraction distortion in process automation systems. This is a key step in achieving accurate rule modeling and reliable execution of risk control logic.

[0085] S500, based on an enhanced set of rule models, combines historical approval process data to perform multi-path process simulation, simulates process decision paths under different rule configurations, identifies approval deviations and logical mismatches caused by differences in rule expression, and outputs simulation comparison results and corresponding risk warning information.

[0086] To further verify the decision-making accuracy and behavioral consistency of the constructed enhanced rule model set in practical process applications, a multi-path process simulation method based on historical approval process data is proposed. By simulating the process flow resulting from different rule configuration schemes under the same process input conditions, the method identifies approval deviations and logical mismatches caused by differences in rule expression, and generates simulation comparison results and risk warning information, providing a basis for subsequent rule correction and model optimization. This method includes the following steps:

[0087] Prepare the input dataset for multi-path simulation. This step involves filtering archived historical approval process data to select representative approval samples as simulation input. These samples must cover multiple business types, role structures, monetary ranges, and geographical regions to ensure the comprehensiveness and coverage of the simulation results. For each sample process data, extract its original input fields, process node sequence, historical decision path, and corresponding contextual label information as the basis for subsequent simulation. The goal of this step is to construct a realistic and comparable data environment, enabling the enhanced rule model to accurately receive diverse approval input conditions during simulation.

[0088] Representative approval samples refer to historical approval records that reflect key business scenarios, diverse combinations of conditions, and typical decision-making patterns in the approval process. These samples typically include: approval processes for high-value or high-risk transactions; approval processes involving cross-departmental or multi-role collaboration; processes with special approval conditions (such as geographical restrictions, supplier classifications, contract terms, etc.); and processes that have experienced approval conflicts, abnormal paths, or rule adjustments. These samples cover a wide range of business areas, have high condition complexity, and exhibit diverse decision-making paths, which helps to comprehensively test the adaptability and accuracy of rule models in real-world business scenarios during simulations. They serve as a crucial input foundation for evaluating the effectiveness of rule modeling.

[0089] During simulation, two rule configuration schemes were applied to simulate the same set of historical approval inputs: one was the original rule model set before construction, and the other was an enhanced rule model set containing strengthened expression. During execution, based on the structure, condition judgments, and decision paths of these two rule sets, the approval path of each process under the two rule configurations was derived, including indicators such as the number of approval steps, changes in participating roles, final decision result, and time taken. By comparing the process results of the same input under different rule-driven conditions across all dimensions, changes in process paths caused by differences in rule expression were identified, with particular attention paid to changes in approval decisions resulting from strengthened expression of additional conditions.

[0090] For process path differences discovered during simulation, approval offset identification and logical mismatch analysis are performed. Offset identification mainly focuses on changes in the hierarchy of approval nodes, adjustments to the approver's identity, and process jumps; logical mismatch analysis focuses on assessing whether there are phenomena such as delegation of authority, role overreach, duplicate approvals, or omissions of key approval nodes caused by incomplete rule expression. For example, in the original rule model, a procurement process exceeding 1 million yuan without distinguishing the supplier type is approved by the general manager, while in the enhanced rule model, based on the additional condition "the supplier is an overseas company," it is approved by the board of directors. This path difference is considered a key approval offset point. This analysis process uses context labels to help determine whether the offset behavior constitutes a process logic risk.

[0091] The approval deviations and logical mismatches discovered in the simulation comparisons are organized and archived, generating a simulation comparison report that includes elements such as process number, input conditions, differences in approval paths under the two types of rule configurations, deviation type, and potential risk level, and outputs corresponding risk warning information. This risk warning information can not only prompt business personnel to review the rationality of the relevant process rule configurations, but also feed back into the rule modeling mechanism, triggering further optimization and accuracy revision of the corresponding enhanced rule model. This step achieves a closed-loop verification after rule modeling, enabling dynamic linkage between the modeling, expression, simulation, and correction of the approval process, improving the business adaptability, semantic consistency, and application reliability of the rule model.

[0092] This step aims to verify the behavior of the enhanced rule model in actual processes through multi-path process simulation, thereby identifying approval deviations and logical mismatches that may be caused by differences in rule expression, ensuring consistency, accuracy, and compliance of approval rules at the execution level. Although the completeness of the rules has been improved through contextual semantic annotation and expression enhancement during rule modeling, the actual operational effect needs to be verified by process scenarios because rule modeling itself is still an abstract semantic structure. This step matches the enhanced rule model set with real historical approval process data, simulating the approval path under different rule configurations under the same input conditions, and comparing the differences in the execution order, participating roles, processing time, and path changes of approval nodes under the original rule model and the enhanced model. Through comparative analysis, the specific impact of expression changes enhanced by additional conditions and contextual semantics on the approval logic can be identified, especially identifying high-risk behaviors such as delegation of authority, process detours, duplicate processing, or skipping of key nodes caused by improper rule expression. In addition, this step will output a simulation comparison report and risk warning information, clearly indicating the process number, triggering conditions, risk level, and suggested handling methods for deviations, providing a basis for subsequent rule optimization and process adjustment. Therefore, this step not only provides a closed-loop verification mechanism for rule models, but also ensures the logical accuracy and governance standardization of process operation through simulation, making it a key link in realizing the effective transformation of rules from modeling to practical application.

[0093] Based on simulation comparison results and risk warning information, S600 jointly determines the execution stability of the rule model in various context scenarios. For detected path deviations and decision conflicts, it initiates consistency backtracking and structural correction processes, traces and verifies the context labels corresponding to risk points, adjusts the rule structure expression and semantic tag configuration, ensures the consistency of decision-making and the accuracy of expression of the rule model in the real process, and completes the dynamic closed-loop optimization of the rule modeling process.

[0094] To ensure the approval logic built on an enhanced rule model maintains high consistency and semantic accuracy across diverse contexts, a rule consistency backtracking and structural correction method based on simulation comparison results and risk warning information is proposed. This method can perform location analysis, semantic source tracing, structural adjustment, and expression updates for approval deviations and decision conflicts, thereby achieving dynamic closed-loop optimization of the rule model from extraction to application. The method includes the following steps:

[0095] Based on the comparison results and risk warning information from the multi-path process simulation output, the execution performance of the rule model under different contextual conditions is jointly judged. This step comprehensively analyzes the approval deviation paths, logical conflict nodes, decision imbalance behaviors, and their corresponding triggering conditions in each simulation process. Statistical and cluster analysis methods are used to determine whether the problem has stability characteristics, i.e., whether it repeatedly occurs in multiple contextual combinations. If a certain type of deviation problem occurs under different business types, regional labels, and approval role structures, it is determined that the rule has expression defects or insufficient contextual modeling, and it needs to be included in the consistency backtracking process.

[0096] For rule models identified as inconsistent or semantically conflicting, a backtracking process is performed. This process leverages the contextual semantic tags and semantic weight identifiers embedded in the enhanced rule model to trace the key fields that caused the problem and their contextual origins. For example, when an issue of "approval role level offset" occurs, by analyzing the semantic tags corresponding to the "role restriction" field in the rule model, it is identified whether the algorithm ignored this element due to missing original tags, inaccurate attribute classification, or low weight configuration. Simultaneously, a horizontal comparison is performed with other rules in related rule groups that have similar expressions to determine whether the offset stems from structural issues such as misaligned expressions or inconsistent additional conditions.

[0097] Based on the source tracing results, structural corrections are performed on the problematic rule models. These corrections include: first, supplementing missing contextual attributes and explicitly incorporating them into the rule structure to ensure completeness; second, adjusting the nesting structure of the main condition fields and additional attributes to optimize the judgment order and combination of rule logic, thereby improving the discriminative ability of decision paths; and third, updating and refining the semantic tag set to enhance its recognizability and non-negligibility in the modeling algorithm. For example, the tag originally labeled "Approval Role = Senior Executive" is refined to "Approval Role = General Manager" or "Approval Role = Board Member" to avoid role identification errors caused by ambiguity. All structural corrections should maintain the original meaning of the rules, enhancing only their semantic recognizability and contextual consistency at the expression level.

[0098] The revised rule model is reintegrated into the enhanced rule model set and matched again with the original process sample for simulation verification to confirm that the revised model no longer generates approval deviations or logical conflicts in the same context. If some abnormal processes still exist, backtracking and iterative correction can continue until the rule model maintains semantic stability, path accuracy, and execution consistency in multi-scenario simulations. This step ensures that the rule modeling process has dynamic feedback capabilities and an adaptive optimization mechanism, making the rule structure not only conform to formal logical correctness but also fit the needs of real business decisions, thereby achieving closed-loop self-correction and continuous evolution of the intelligent approval rule model in actual processes.

[0099] This step aims to build a dynamic closed-loop feedback mechanism in the rule modeling process, ensuring that the rule model has stable, consistent, and reliable execution capabilities under different contextual conditions. It also enables timely identification of risk sources and correction of expression defects when problems are discovered, thereby improving the semantic accuracy and decision consistency of the rule model in real business processes. In approval process modeling, even with a relatively complete rule structure formed through context awareness and expression enhancement, semantic ambiguity, contextual conflicts, or blurred expression boundaries may still lead to rule mismatch, path deviation, and approval result conflicts in certain specific scenarios. This step, based on the simulation comparison results and generated risk warning information from the previous stage, comprehensively analyzes the behavior of the rule model in diverse contextual scenarios. By comparing whether its decision path is stable, whether the approval roles are consistent, and whether the triggering conditions are accurate, it determines whether the model has stability defects. Once a deviation path or conflicting behavior is detected, a consistency backtracking and structural correction process is initiated. The contextual semantic tags corresponding to the risk rules are extracted from the model, and the source is traced through the tags to confirm whether the cause is missing tags, attribute ambiguity, or semantic coverage errors. Furthermore, by adjusting the rule structure, reconstructing the logical order, and optimizing the semantic tag configuration, the rules are made more closely aligned with their real-world application scenarios, and structurally prevent algorithmic errors such as aggregation or misidentification. Finally, the revised rules are re-verified, forming a closed-loop optimization chain to ensure continuous evolution and robust control of rules from construction and verification to application. This mechanism not only enhances the model's adaptability to complex business differences but also strengthens the behavioral stability during rule execution, making it a crucial step in ensuring the logical correctness, compliance, and controllability of the intelligent approval system.

[0100] The aforementioned intelligent approval rule modeling method for process automation effectively addresses the problems of rule merging errors and approval logic mismatches caused by low semantic recognition granularity and lack of context awareness in existing rule extraction technologies. This method introduces a contextual semantic labeling mechanism and a ternary structure feature alignment strategy to ensure that rules possess clear semantic boundaries and structural expressions from the initial modeling stage, significantly enhancing the ability to identify fine-grained differences. By constructing a rule feature matrix and a rule differentiation candidate set, it further enables accurate classification and processing of rule combinations with ambiguity and contextual conflicts, avoiding the misjudgment of semantic overlap as equivalent rules. Combined with an expression sensitivity enhancement mechanism and simulation verification process, potential path deviations and decision conflicts can be identified before rule execution, providing early warnings of approval risks. Finally, through consistency backtracking and structural correction processes, dynamic self-optimization of the rule model is achieved, ensuring its execution stability and expression accuracy in different contextual scenarios. Overall, this scheme not only improves the accuracy and reliability of rule modeling but also constructs a closed-loop system for rule generation, verification, and optimization, providing a highly controllable and reliable rule foundation for intelligent decision-making in a process automation environment.

[0101] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0102] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0103] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does 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 one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0104] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0105] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0106] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0107] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0108] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0109] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0110] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for modeling intelligent approval rules for process automation, characterized in that, Includes the following steps: S100, in the process of constructing the rule candidate set, extract the scene features, field semantic hierarchy and participating role information of each historical approval data to generate a set of context semantic tags for subsequent rule difference modeling; S200, based on the contextual semantic label set, performs feature alignment on rule expression, extracts main condition fields, contextual additional attributes and decision actions, and constructs a rule feature matrix as the basic structure for rule comparison and filtering; S300, based on the rule feature matrix, identifies rule combinations with the same main condition field but different expression methods, extracts ambiguous additional attributes, and generates a rule differentiation candidate set; S400, based on the rule-differentiated candidate set, embeds key additional attributes into the original rule structure, performs expression-sensitive enhancement processing, explicitly labels detailed elements that are easily ignored by compression algorithms, and generates an enhanced rule model set; S500, based on an enhanced set of rule models, performs multi-path simulations by combining historical approval process data. It simulates process paths under different rule configurations, identifies approval deviations and logical mismatches caused by differences in rule expression, and outputs simulation comparison results and risk warning information. Based on simulation comparison results and risk warning information, S600 determines the execution stability of the rule model in various context scenarios, initiates consistency backtracking and structural correction for detected offsets and conflicts, verifies context labels, adjusts rule expressions and semantic tag configurations, and completes closed-loop optimization of the rule model.

2. The intelligent approval rule modeling method for process automation according to claim 1, characterized in that, Step S100 includes: During the extraction of each historical approval data, the scene characteristics, field semantic hierarchy, and participating role information corresponding to the approval behavior are collected. Based on approval logs, business process definitions, organizational structure, business metadata and unstructured text, perform enhanced extraction of execution context data to obtain timestamps, process stages, job levels, business lines and semantic information of remarks; The collected fields are semantically categorized, and the field types and expression levels are divided according to the preset business logic knowledge base; Generate a structured set of contextual semantic tags, organized in the form of "tag dimension, tag key, tag value" triples, and bind them one by one with historical approval data.

3. The intelligent approval rule modeling method for process automation according to claim 1, characterized in that, Step S200 includes: Based on the contextual semantic tag set, the logical expression of each rule is structured and parsed to identify judgment fields, status fields and role fields; Extract three key elements: main condition field, contextual additional attributes, and decision action, and construct a unified ternary expression format; The three types of elements are aligned according to a preset field template to construct a rule feature matrix; Normalize the feature matrix of the rules and add semantic weights to improve the semantic discrimination ability in rule comparison.

4. The intelligent approval rule modeling method for process automation according to claim 1, characterized in that, Step S300 includes: Semantic element recognition and structural segmentation are performed on each historical approval data, and the main condition fields, contextual additional attributes, and decision actions are identified based on the contextual semantic tag set. The three types of key elements identified are mapped to predefined field positions to construct a rule feature matrix, and missing fields are handled with null values ​​or default unconditional handling. Perform expression normalization processing on the rule feature matrix to standardize the expression form of semantically equivalent fields; Based on the strength of the field's role in decision-making, semantic weight adjustment is performed to identify rule combinations that have the same main condition fields but different contextual additional attributes, extract ambiguous additional attributes, and generate a rule differentiation candidate set.

5. The intelligent approval rule modeling method for process automation according to claim 1, characterized in that, Step S400 includes: Based on the rule-differentiated candidate set, we identify rule combinations that have the same main condition field but semantic differences in contextual additional attributes, and extract key additional attributes that are easily ignored by compression algorithms. The extracted key additional attributes are embedded into the corresponding original rule structure, keeping the logic unchanged and explicitly expanding the content of expression; The updated rule structure is subjected to expression-sensitive enhancement processing, and the semantic weights and saliency markers of key detail elements are annotated. The processing results are summarized to generate an enhanced set of rule models.

6. The intelligent approval rule modeling method for process automation according to claim 1, characterized in that, Step S500 includes: Filter representative historical approval process data and extract the input fields, process node sequence, historical decision path and context labels for each data point; For the same approval input, the original rule model set and the enhanced rule model set are applied to perform process simulations to deduce their respective approval paths and decision results; Compare the differences in process paths under the two rule configurations, identify approval deviations and logical mismatches, and analyze whether there is delegation of authority, overstepping of role authority, or omission of key nodes. Output a simulation comparison report that includes path differences, offset types, and risk levels, and generate corresponding risk warning information.

7. The intelligent approval rule modeling method for process automation according to claim 1, characterized in that, Step S600 includes: The joint analysis process simulation output identifies approval offset paths and logical conflict nodes, and identifies problem types that repeatedly occur in multiple context combination scenarios. Based on the contextual semantic tags and semantic weight identifiers embedded in the rule model, we trace back the key fields that caused the offset and their contextual sources to determine the reasons for expression defects or insufficient modeling. Corrections are made to rule models with structural problems, including adding contextual attributes, optimizing field nesting structures, and refining semantic tag content; The revised rule model was re-simulated and verified to confirm that it has semantic stability and execution consistency under the same context.

Images