An intelligent process decision processing method and device based on a knowledge graph

Abstract

The application provides a kind of intelligent process decision processing method and device fused knowledge graph, it is related to the technical field of knowledge graph, method includes: by pre-processing to multiple source business data and constructing standard knowledge source data set, generate decision context set in combination with real-time business context, carry out candidate starting node matching, multi-hop path traversal and path constraint analysis based on knowledge graph, calculate path confidence, stability and adaptability and carry out sorting screening, to determine target reasoning path, and drive process instance into target execution branch according to the business action in path.The application can comprehensively analyze multiple source data under complex business scenarios, improve the processing efficiency of process engine.

Description

Technical Field

[0001] This invention relates to the technical field of knowledge graphs, and specifically to an intelligent process decision-making method and apparatus that integrates knowledge graphs. Background Technology

[0002] As a core infrastructure within an enterprise's information system, business process management systems typically rely on process engines to schedule and control various business nodes. At decision nodes involving approval routing, task assignment, or strategy judgment, mainstream technical solutions generally employ decision mechanisms based on rule tables or conditional expressions, using pre-configured logical rules to select process paths. These solutions typically process structured data, completing matching judgments by setting explicit field conditions, threshold ranges, and logical combinations. While some systems introduce simple rule engines or decision tables to enhance flexibility, the overall approach remains static rule-driven. Furthermore, as enterprise business complexity increases, data across multiple systems becomes increasingly decentralized and heterogeneous. The information relied upon for process decisions includes not only current node data but also historical execution records, business documents, and domain knowledge. Existing technologies typically aggregate this data through interface integration or data warehouses, but lack unified semantic modeling and relational expression methods, making it difficult to efficiently utilize this data during the decision-making process.

[0003] Under the existing technological framework, a significant technical deficiency lies in the lack of a unified understanding and dynamic reasoning capability for multi-source business semantics in the decision-making process. Specifically, when the process reaches a decision node, it can only perform precise matching and judgment based on the current structured fields, failing to comprehensively analyze the implicit semantic relationships in the real-time business context, cross-system knowledge associations, and historical experience. This makes it difficult to accurately depict the true decision-making logic in complex business scenarios. Especially when fuzzy conditions, multi-factor coupling, or cross-domain knowledge collaboration are involved, traditional rule mechanisms require extensive manual configuration and frequent adjustments. Furthermore, knowledge between different systems cannot form effective linkages, further resulting in a lack of adaptability and scalability in decision path selection, leading to low efficiency. Summary of the Invention

[0004] This invention provides an intelligent process decision processing method and apparatus that integrates knowledge graphs, which can comprehensively analyze multi-source data in complex business scenarios and improve the processing efficiency of the process engine.

[0005] A first aspect of the present invention provides an intelligent process decision processing method integrating knowledge graphs, the method comprising: Multi-source business data from multiple systems is acquired and preprocessed to form a standard knowledge source data set. When the process instance runs to the target decision node, fusion processing is performed based on the acquired real-time business context data to form a decision context set. Similarity matching is performed based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, wherein the process knowledge graph is constructed based on the standard knowledge source data set; Starting from the set of candidate starting nodes, perform multi-hop path traversal in the process knowledge graph, and generate a set of candidate reasoning paths based on path constraints, node association strength, and historical hit frequency. Graph structure analysis is performed on the candidate inference path set to form a path analysis result set that includes path confidence, path stability, and path fitness. Based on the path analysis result set, a sorting and filtering process is performed to determine the target inference path, and the process instance is driven to enter the target execution branch according to the business action identifier corresponding to the target inference path.

[0006] In a second aspect of the invention, an intelligent process decision processing apparatus integrating knowledge graphs is provided. The apparatus is used to execute an intelligent process decision processing method integrating knowledge graphs as described above. The apparatus includes an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire multi-source business data from multiple systems, perform preprocessing to form a standard knowledge source data set, and when the process instance runs to the target decision node, perform fusion processing based on the acquired real-time business context data to form a decision context set. The processing module is used to perform similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, wherein the process knowledge graph is constructed based on the standard knowledge source data set; The processing module is used to perform multi-hop path traversal in the process knowledge graph starting from the candidate starting node set, and generate a candidate reasoning path set based on path constraints, node association strength and historical hit frequency. The processing module is used to perform graph structure analysis processing on the candidate inference path set to form a path analysis result set including path confidence, path stability and path fitness. The output module is used to perform sorting and filtering processing based on the path analysis result set, determine the target inference path, and drive the process instance to enter the target execution branch according to the business action identifier corresponding to the target inference path.

[0007] In a third aspect of the invention, an electronic device is provided, including a processor and a memory, the memory having a stored computer program, wherein the computer program, when executed by the processor, implements an intelligent process decision processing method integrating knowledge graphs as described in any of the preceding claims.

[0008] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium storing instructions that, when executed, perform an intelligent process decision processing method integrating knowledge graphs as described in any of the preceding claims.

[0009] In summary, one or more technical solutions provided in the embodiments of the present invention have at least the following technical effects or advantages: This invention achieves unified semantic modeling and relational expression of multi-system and heterogeneous data by uniformly preprocessing multi-source business data and constructing a standard knowledge source data set, and then forming a decision context set based on real-time business context data. It combines process knowledge graphs to perform similarity matching, multi-hop path traversal and path scoring analysis on candidate starting nodes, and incorporates node contribution, relationship contribution, path confidence, stability and adaptability into a comprehensive evaluation. The target reasoning path is determined by sorting and filtering, thereby enabling comprehensive analysis of multi-source data in complex business scenarios, reducing repeated queries and redundant calculations, and improving the decision efficiency and execution reliability of the process engine under second-level response conditions. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating an intelligent process decision-making method integrating knowledge graphs disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of a module of an intelligent process decision processing device that integrates knowledge graphs, as disclosed in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention.

[0011] Explanation of reference numerals in the attached drawings: 201, acquisition module; 202, processing module; 203, output module; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification 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.

[0013] In the description of the embodiments of the present invention, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0014] In the description of the embodiments of the present invention, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0015] Existing business process management systems generally rely on static rule tables or conditional expressions for path selection at decision nodes. The main objects they process are structured data, and they lack unified semantic modeling and relational expression of multi-source business data. This makes it impossible to comprehensively analyze real-time context, historical experience and cross-system knowledge in complex business scenarios. The decision path lacks adaptability and scalability, and it is difficult to achieve structured traceability of decision basis, thus limiting the improvement of the intelligent and refined decision-making capabilities of the process engine.

[0016] This invention discloses an intelligent process decision-making method integrating knowledge graphs, which is applied to a server. The server includes, but is not limited to, electronic devices such as mobile phones, tablets, wearable devices, and PCs (Personal Computers), and can also be a backend server running an intelligent process decision-making method integrating knowledge graphs. The server can be implemented using a standalone server or a server cluster composed of multiple servers.

[0017] This embodiment discloses an intelligent process decision-making method that integrates knowledge graphs, referring to... Figure 1 It includes the following steps: S110: Obtain multi-source business data from multiple systems, perform preprocessing to form a standard knowledge source data set, and when the process instance runs to the target decision node, perform fusion processing based on the obtained real-time business context data to form a decision context set.

[0018] S120: Perform similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set.

[0019] S130: Starting from the set of candidate starting nodes, perform multi-hop path traversal in the process knowledge graph, and generate a set of candidate reasoning paths based on path constraints, node association strength, and historical hit frequency.

[0020] S140, perform graph structure analysis on the candidate inference path set to form a path analysis result set including path confidence, path stability and path fitness.

[0021] S150: Based on the path analysis result set, perform sorting and filtering processing to determine the target inference path, and drive the process instance to enter the target execution branch according to the business action identifier corresponding to the target inference path.

[0022] In practice, the first step is to perform unified field mapping and data format conversion on multi-source business data from various systems, including business systems, rule systems, historical log systems, and document systems. This maps similar fields from different systems to a unified field naming system. For example, "Applicant Number," "User ID," or "Operator ID" from different systems are uniformly mapped to "Business Entity Identifier," ensuring that subsequent processing can consistently identify the same entity. Next, semantic cleaning is performed on text, numeric, and time-type fields, including removing invalid characters, filling in missing values, standardizing time formats, and unifying unit representations. This ensures that the multi-source data achieves semantic and representational consistency, thus forming standard data that can be directly used for knowledge modeling.

[0023] Subsequently, time alignment processing is performed on the multi-source business data that has undergone field unification and semantic cleaning. Records from different systems are sorted and synchronized according to a global time base or process time sequence to ensure that cross-system events accurately reflect the same decision-making moment. At the same time, an event identifier sequence is generated to associate the business operation trajectories of different systems. This step also includes identifier normalization processing, which assigns a unique business object identifier and node identifier to each data record to distinguish different entities and their positions in the process, ensuring that entity nodes do not duplicate or conflict when constructing the knowledge graph.

[0024] When a process instance reaches the target decision node, real-time business context data associated with that node is collected, including current business parameters, application object attributes, resource status, approval status, task priority, and operator attributes. This data is integrated with historical process status information, such as previous node outputs, historical approval records, and historical operation paths, to achieve complete capture of the real-time context. Subsequently, the real-time context data is fused with a standard knowledge source dataset. Using vectorized representation or feature encoding, data from different sources and of different types are uniformly mapped to the same high-dimensional feature space, forming a decision context set. This allows subsequent decision reasoning to consider the current state, historical experience, and cross-system knowledge associations based on a unified representation.

[0025] Among them, the standard knowledge source data set refers to the data set that can be directly used for knowledge graph construction after field unification, semantic cleaning, time alignment and identifier normalization; real-time business context data refers to the business status, resource status and operator attribute data related to the decision node collected in real time at the decision node of the process instance; fusion processing refers to the feature integration, vector encoding or multimodal fusion of real-time context data and standard knowledge source data, so that it can be directly used for candidate node matching and path reasoning in a unified representation space, ensuring that the decision logic can integrate historical experience and current business status.

[0026] Furthermore, in scenarios involving rapid transitions to combat readiness, the core challenge for enterprises is the instantaneous changes in business constraints and resource conditions when switching from normal production to combat readiness support. Traditional rule-driven process engines struggle to respond quickly. In normal production, production priorities, resource allocation rules, quality inspection tolerances, and supply chain response mechanisms are all executed according to predetermined processes and standards. When entering combat readiness, these rules need to be readjusted. For example, specific model tasks or critical resources may require priority scheduling, quality inspection standards may be temporarily relaxed or strengthened, and supply chain response paths and delivery priorities may dynamically change. Hard-coded rules cannot support rapid refactoring under such instantaneous adjustments, making it difficult for process engines to generate effective decisions within a second-level response window.

[0027] Knowledge graphs can provide dynamic decision support in this scenario, expressing historical experience, business constraints, and real-time status through multi-level associations of nodes and relationships. However, when there are too many associated nodes and deep relationship paths in the knowledge graph, multi-hop inference requires traversing a large number of nodes and edges. The scoring and path selection process involves multi-dimensional calculations such as node contribution value, relationship contribution value, conflict propagation value, path confidence, path stability, and path suitability. These calculations consume significant time in real-time combat readiness scenarios. Although the inference algorithm can accurately evaluate the priority and feasibility of each candidate path, the generated results may lag behind the actual execution window of the combat readiness task, resulting in "correct but invalid" decisions. That is, the decision path conforms to logic and business constraints, but the optimal scheduling time has been missed, failing to meet the second-level response requirements.

[0028] This lag stems primarily from three factors: First, the sheer size of the knowledge graph and the increased number of nodes and edges lead to increased complexity in multi-hop traversal and scoring calculations. Second, the node and relation attributes involve numerous real-time parameters, such as resource availability, inventory levels, equipment health status, and supply chain accessibility, requiring frequent reading and computation. Third, path evaluation requires a comprehensive consideration of historical experience and real-time constraints, including node retention values, relation retention values, bottleneck contributions, conflict propagation, and accessibility verification. Multi-dimensional calculations rely on linear or non-linear weighting, increasing computational latency. Therefore, even with accurate decision-making logic, the target reasoning path generated by the knowledge graph may not be able to immediately drive process instances into the target execution branch in rapid combat readiness switching scenarios, impacting overall combat readiness support efficiency.

[0029] In one possible implementation, before performing similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, the method further includes: performing hierarchical parsing on the standard knowledge source data set and binding business object identifiers, status time identifiers, process identifiers, and combat readiness stage identifiers to form a combat readiness decision semantic space; performing entity recognition and entity merging processing on the combat readiness decision semantic space to form a business entity node set; performing attribute mapping processing on the business entity node set to write static attributes, dynamic attributes, and combat readiness attributes into the entity attribute set; and performing relation extraction processing based on the combat readiness decision semantic space to form a business relation edge set containing ordinary business relations and combat readiness constraint relations, wherein the business relation edge set... The system performs constraint association processing, binding the war preparedness phase identifier, response time limit, confidence source, scope of application, failure condition, and conflict level to the corresponding business relationship edge, and setting a delayed inference identifier for non-real-time business relationship edges; it calculates node retention values ​​and relationship retention values ​​based on the set of business entity nodes and the set of business relationship edges, and filters them to form a war preparedness graph subgraph; it establishes a two-layer index structure for the process knowledge graph, including a real-time scheduling index and an audit traceability index; it sets state gating identifiers for business entity nodes and business relationship edges, and performs relationship switching control between normal production state and war preparedness support state based on the war preparedness switching signal; and it forms a process knowledge graph based on business entity nodes, business relationship edges, entity attribute sets, war preparedness graph subgraphs, two-layer index structure, and state gating identifiers.

[0030] Specifically, firstly, a hierarchical parsing process is performed on the standard knowledge source data set, classifying and managing the data according to the source system, business type, and system attributes. Each record is bound to a business object identifier, a status time identifier, a process identifier, and a readiness stage identifier. The business object identifier is used to uniquely identify the business entity in the process, such as a task object, resource object, or execution branch object; the status time identifier is used to indicate the time node or process stage corresponding to the data, so that time series information can be referenced during path reasoning; the process identifier is used to indicate the specific business process to which the data belongs, ensuring that data from different business processes are not confused; the readiness stage identifier is used to distinguish between peacetime production status and readiness support status, so that subsequent reasoning and path selection can be constrained and referenced based on the readiness status.

[0031] Entity identification and merging are performed within the constructed semantic space for strategic decision-making. Key business entity nodes are identified, including task objects, resource objects, process objects, quality inspection objects, supply objects, permission objects, and execution branch objects. For multiple representations of the same business object appearing in different systems or data sources, entity merging is performed by matching business object identifiers, system codes, and semantic aliases. This ensures that each business entity corresponds to a unique node in the process knowledge graph, avoiding redundant calculations and path scoring conflicts, and improving the efficiency of multi-hop reasoning.

[0032] Attribute mapping is performed on the set of business entity nodes, writing the static, dynamic, and standby attributes of each node into the entity attribute set. Static attributes include business object type, department, fixed constraints, and confidentiality level, describing the constant characteristics of the business entity. Dynamic attributes include current availability, remaining capacity, inventory quantity, and online status, reflecting resource availability and business execution capabilities in real time. Standby attributes include standby priority, response time limit, emergency availability identifier, and degraded usage conditions, describing the scheduling priority and executable capability of nodes in standby status. This entity attribute set enables subsequent path scoring to quickly access node characteristics without cross-system queries, reducing decision-making latency.

[0033] Based on the semantic space of combat readiness decision-making, an execution relationship extraction process is performed to form a set of business relationship edges, including ordinary business relationships and combat readiness constraint relationships. Ordinary business relationships include belonging relationships, calling relationships, approval relationships, supply relationships, and execution relationships, used to describe regular process dependencies and operational relationships between nodes. Combat readiness constraint relationships include priority guarantee relationships, resource mutual exclusion relationships, alternative adaptation relationships, time limit constraint relationships, confidentiality isolation relationships, and tolerance switching relationships, used to describe operational constraints and resource limitations under combat readiness conditions. Subsequently, the execution constraint association processing of business relationship edges is performed, binding the combat readiness stage identifier, response time limit, confidence source, scope of application, failure condition, and conflict level to the corresponding relationship edge; a delayed inference identifier is set for non-real-time business relationship edges to ensure that they do not affect real-time path inference in combat readiness rapid switching scenarios, reducing the computational burden.

[0034] The node retention value and relationship retention value are calculated based on the set of business entity nodes and the set of business relationship edges, and a strategic preparedness graph subgraph is formed by filtering according to thresholds. The calculation of node retention value and relationship retention value takes into account strategic preparedness priority, response time limit, availability, path carrying capacity, conflict level and historical hit frequency, so as to quantify the contribution of each node and relationship edge to the real-time decision path, and determine which nodes and relationship edges are included in the scope of rapid strategic preparedness inference, thereby compressing the size of the knowledge graph and improving the second-level response capability.

[0035] The formula for calculating the node retention value is:

[0036] in, This represents the retention value of the i-th business entity node. The larger the value, the more it should be retained in the strategic preparedness map subgraph. Indicates the combat readiness priority, with a value range of 0 to 1, and is normalized by the combat readiness command level, model mission level, or support level; This indicates the maximum allowed response time; a smaller value indicates that faster scheduling is required. This represents the maximum allowable response time for the current target decision node. For stable parameters, the range of values ​​is... to ; Indicates the current availability status. 1 indicates full availability, 0 indicates unavailability, and partial availability is mapped to 0 to 1 based on remaining capacity, inventory, or online status. This represents the normalized value of historical hit frequency, ranging from 0 to 1; This represents the real-time risk value, ranging from 0 to 1, and is derived from the risk of chain disruption, equipment malfunction, permission conflict, and unreachable transportation. to The weighting coefficients range from 0 to 1, and their sum is 1. The formula ensures that critical nodes are prioritized for retention by increasing the retention tendency of nodes with high priority, fast response, strong availability, and high historical hit rate, while suppressing high-risk nodes.

[0037] The formula for calculating the relation retention value is:

[0038] in, This represents the preserved value indicating the relationship between the i-th business entity node and the j-th node; Indicates the strength of the relationship, ranging from 0 to 1, and is a fusion of rule confidence, log co-occurrence frequency, and expert confirmation. Indicates the degree of matching with the current combat readiness stage, ranging from 0 to 1; This represents the normalized value of historical hit frequency, ranging from 0 to 1. This indicates the expected decision depth, calculated from the shortest path depth or the average number of inference layers. Indicates the maximum allowed inference depth; Indicates the conflict level, from 0 to 1, normalized by resource exclusivity, confidentiality isolation, tolerance conflict, and supply chain disruption; to These are weighting coefficients, and their sum is 1. This formula enhances strong associations, high matching degrees, high historical hits, and low-depth relationships, while weakening high-conflict relationships, reducing long-link reasoning and conflict path traversal.

[0039] A two-layer index structure is established for the process knowledge graph. The real-time scheduling index is used to quickly locate key nodes and relationship edges related to the current decision node and the target execution branch, supporting second-level reasoning access. The audit traceability index is used to record historical versions, rule sources and decision paths, so as to perform source tracing analysis and closed-loop optimization after the decision is completed.

[0040] Set status gating identifiers for business entity nodes and business relationship edges, including identifiers for normal availability, combat readiness availability, switchover freeze, and manual review. Activate or freeze nodes and relationship edges according to combat readiness switchover signals to ensure that only the corresponding nodes and relationship edges are enabled in different combat readiness stages, and avoid conflicts or redundancy caused by mixing normal rules and combat readiness rules during path reasoning.

[0041] Finally, the business entity nodes, business relationship edges, entity attribute sets, strategic preparedness graph subgraphs, two-layer index structures, and state gating identifiers are integrated to form a complete process knowledge graph. This provides a unified data structure and semantic constraints for subsequent similarity matching, candidate starting node selection, and path reasoning based on decision context sets and graph node vectors. A process knowledge graph is a knowledge representation model that uses a graph structure to represent various entities and their relationships in an enterprise's business processes. Nodes represent business entities, such as task objects, resources, processes, quality inspection items, suppliers, or access control points. Edges represent business relationships or constraints between entities, including belonging relationships, calling relationships, approval relationships, and strategic constraints such as priority assurance, resource mutual exclusion, substitution adaptation, and time limit constraints. Nodes and edges simultaneously contain static attributes, dynamic attributes, and strategic attributes to describe the type, status, availability, priority, and risk information of business objects. Through index structures and status gating identifiers, it enables rapid access to real-time scheduling paths, dynamic matching of decision contexts, and status control under strategic switching conditions. This supports unified semantic modeling, dynamic path reasoning, and interpretable intelligent decision-making in complex, multi-source, and heterogeneous data environments.

[0042] In one possible implementation, before calculating node retention values ​​and relationship retention values ​​based on the set of business entity nodes and the set of business relationship edges, and filtering to form a strategic preparedness graph subgraph, the method further includes: obtaining the strategic preparedness stage identifier, target execution branch set, response time limit, inference depth, and candidate starting node set corresponding to the target decision node, and obtaining the set of business entity nodes associated with the candidate starting node set; obtaining the node retention value, strategic preparedness priority, response time limit, current availability status, real-time risk value, and business type corresponding to each business entity node in the set of business entity nodes; calculating the real-time compression coefficient based on the response time limit, inference depth, and graph congestion; calculating the initial node retention threshold based on the distribution statistics of node retention values ​​and the real-time compression coefficient; performing type balancing correction on the initial node retention threshold based on the business type to form a local node retention threshold corresponding to each business type; performing a first correction on the local node retention threshold in conjunction with the state gating identifier, and performing a second correction based on the actual inference time and the number of candidate inference paths to form a node retention threshold, wherein the node retention threshold is used to perform a first filtering process on the node retention values, so that the set of business entity nodes after the first filtering process is used to form a strategic preparedness graph subgraph.

[0043] Specifically, firstly, the system obtains the preparedness phase identifier, target execution branch set, response time limit, maximum inference depth, and candidate starting node set corresponding to the target decision node. Then, it reads the set of business entity nodes associated with the candidate starting node set from the process knowledge graph. The preparedness phase identifier distinguishes between the normal production state and the preparedness support state of the current decision. The target execution branch set represents the business path that the decision node can trigger. The response time limit refers to the maximum time window allowed from triggering the decision to completing path inference. The maximum inference depth limits the level of path search. The candidate starting node set is the set of business entity nodes that can serve as the starting point of the path, obtained based on the decision context matching. The set of business entity nodes associated with the candidate starting node set is used for subsequent calculation of node retention values ​​and filtering, ensuring that path inference is performed within the executable range.

[0044] Subsequently, for each node in the business entity node set, its corresponding node retention value, readiness priority, response time limit, current availability status, real-time risk value, and business type are obtained. The node retention value is calculated by combining previous path contribution, historical hit frequency, and node attributes, and is used to quantify the importance of the node in path reasoning; the readiness priority reflects the priority order in which the node is scheduled in the readiness state; the response time limit indicates the time required for the node to complete the action from triggering; the current availability status indicates whether the node is schedulable at this moment, including capacity, inventory, or personnel online status; the real-time risk value comprehensively considers factors such as potential chain breakage risks, equipment anomalies, permission conflicts, and supply unavailability; the business type is used to achieve balanced retention of different types of nodes in the generation of node retention thresholds, avoiding excessive pruning of certain types of nodes.

[0045] The real-time compression coefficient is calculated based on the response time limit, maximum inference depth, and current graph congestion of the target decision node. The real-time compression coefficient reflects the compression requirements of the current combat readiness switch on the graph inference scale and number of nodes. The tighter the response time limit, the greater the inference depth, and the higher the graph congestion, the larger the compression coefficient, thus guiding the adjustment of subsequent node retention thresholds to meet second-level response requirements. Graph congestion can be calculated as the ratio of the number of reachable paths from the candidate starting node to the preset node capacity, reflecting the density of the current graph near the decision node.

[0046] The formula for calculating the real-time compression factor is as follows:

[0047] in, This represents the real-time compression factor; a larger value indicates higher requirements for node selection and graph compression. The response time limit for the current target decision node. For reference, the maximum allowable response time; d represents the maximum inference depth. The reference inference depth; K is the number of paths reachable from the current set of candidate starting nodes. For reference, the number of reachable paths; To stabilize parameters and avoid the denominator being zero; , , The weighting coefficients range from 0 to 1, and the sum of the three is 1. This formula adjusts the compression intensity by comprehensively considering the response time, inference depth, and reachable path density to ensure that node selection meets the second-level response requirements for combat readiness.

[0048] Subsequently, the initial node retention threshold is calculated based on the statistical distribution of node retention values ​​and the real-time compression coefficient. The statistical distribution of node retention values ​​may include the mean, standard deviation, and quantiles. By combining these with the real-time compression coefficient to form the initial threshold, the screening criteria can be automatically adjusted when node retention values ​​are concentrated or discrete, thereby ensuring that key nodes are retained while low-value or high-risk nodes are compressed, thus optimizing the size of the strategic preparedness map submap.

[0049] The formula for the initial node retention threshold is:

[0050] in, Reserve a threshold for the initial node; The mean value is retained for each node; Standard deviation; This is the discrete adjustment coefficient, ranging from 0 to 2; These are quantile values. The quantile ratio is typically between 0.5 and 0.95. This formula primarily relies on the mean and standard deviation when combat readiness pressure is low, but shifts to quantile selection when pressure is high, to balance node coverage and response speed.

[0051] Subsequently, the initial node retention thresholds are adjusted for type balancing based on business type, resulting in local node retention thresholds corresponding to each business type. For example, the thresholds for task-type, resource-type, process-type, quality inspection-type, supply-type, and permission-type nodes are adjusted separately to ensure that various key nodes are reasonably retained in the rapid graph, preventing certain types of nodes from being pruned due to excessively high or low overall thresholds, thus ensuring path reachability and business integrity.

[0052] The first correction is performed in conjunction with the status gating identifiers. The status gating identifiers include those for peacetime availability, combat readiness availability, handover freeze, and manual review. The threshold correction appropriately raises the screening threshold for nodes in the frozen or manually reviewed state or transfers them to the manual review set, ensuring that only nodes in the corresponding status are retained in different combat readiness phases.

[0053] Subsequently, a second correction is performed based on the actual inference time and the number of candidate inference paths to form the final node retention threshold. If the inference time of the previous round exceeds the allowable time limit, the threshold is raised to reduce the node size; if there are insufficient candidate paths, the threshold is lowered to replenish nodes, ensuring path reachability and combat readiness response speed. The final node retention threshold is used to perform the first screening process on the node retention values, thereby forming a set of business entity nodes used to generate the combat readiness map subgraph.

[0054] In one possible implementation, before calculating node retention values ​​and relationship retention values ​​based on the set of business entity nodes and the set of business relationship edges, and filtering to form a strategic preparedness graph subgraph, the method further includes: obtaining the strategic preparedness stage identifier corresponding to the target decision node, the target execution branch set, the candidate starting node set, the response time limit, the inference depth, and the candidate range of the graph subgraph; and obtaining the set of business relationship edges located between the candidate starting node set and the target execution branch set; generating hierarchical relationship retention thresholds for each relationship level identifier based on the relationship timeliness pressure coefficient, the relationship bottleneck contribution value, and the relationship conflict propagation value, wherein the thresholds are determined according to the response time limit and the triggering delay of the business relationship edge set. The relationship timeliness pressure coefficient is calculated based on the distribution and the number of currently reachable paths. The relationship bottleneck contribution value is calculated based on the number of occurrences of paths in the effective inference path, the number of connected target execution branches, and the shortest path depth. The relationship conflict propagation value is calculated based on the conflict level of each business relationship edge, the number of reachable conflicting relationships within the preset inference depth, and the number of manually reviewed relationships. The hierarchical relationship retention threshold is modified based on the actual inference time, the actual trigger success rate, and the proportion of manual review to form the relationship retention threshold. The relationship retention threshold is used to perform a second screening process on the relationship retention value, so that the set of business relationship edges after the second screening process can be used to form the strategic reserve graph subgraph.

[0055] Specifically, firstly, the system obtains the preparedness phase identifier, target execution branch set, candidate starting node set, response time limit, maximum inference depth, and candidate range of the graph subgraph corresponding to the target decision node. Then, it reads the set of business relationship edges located between the candidate starting node set and the target execution branch set from the process knowledge graph. The preparedness phase identifier distinguishes whether the current process is in normal production or preparedness support status. The target execution branch set represents the set of business paths that can be triggered by this decision node. The candidate starting node set is the set of business entity nodes that can serve as the starting point of a path, obtained through decision context matching. The response time limit represents the maximum allowed time window from triggering the decision to completing path inference. The maximum inference depth limits the traversal level of multi-hop paths. The candidate range of the graph subgraph limits the subset of nodes and relationship edges that can participate in second-level inference to ensure real-time performance and path reachability. By obtaining the above information, the system can determine the set of key relationship edges that should be prioritized during preparedness switching, providing a data foundation for subsequent relationship retention threshold calculations.

[0056] Subsequently, based on the response time limit, the trigger latency distribution of the business relationship edge set, and the number of currently reachable paths, the relationship timeliness pressure coefficient is calculated. The relationship timeliness pressure coefficient quantifies the time pressure experienced by the current path inference under rapid switching conditions. When the response time limit is tight, the trigger latency distribution is high, or the number of reachable paths is insufficient, the coefficient value increases, indicating that the threshold needs to be increased during relationship screening to compress high-latency or low-value relationship edges. The trigger latency distribution refers to the statistical delay time from the candidate starting node to the target execution branch that can be supported for each business relationship edge; the number of reachable paths refers to the number of effective paths that can be formed from the candidate starting node to the target execution branch, used to assess the current load and redundancy of the graph.

[0057] Based on the frequency of path occurrences, the number of connected target execution branches, and the shortest path depth in the effective inference paths, the bottleneck contribution value of each business relationship edge is calculated. The frequency of path occurrences indicates the frequency with which the relationship edge is used in historical or current candidate paths, reflecting its importance in path inference; the number of connected target execution branches measures the number of target execution branches that the relationship edge can cover; the shortest path depth indicates how close the edge is to the start and end points in the path, with a smaller depth indicating a key bridging role in the path. By combining these indicators, key relationship edges can be identified, preventing the accidental deletion of important connections during the screening process, thereby ensuring the connectivity and path integrity of the strategic preparedness map subgraph.

[0058] Subsequently, based on the conflict level of each business relationship edge, the number of achievable conflicting relationships within the preset inference depth, and the number of manually reviewed relationships, a relationship conflict propagation value is calculated. The conflict level reflects the degree of risk that the relationship edge is subject to resource exclusivity, confidentiality isolation, tolerance conflict, or supply chain disruption; the number of achievable conflicting relationships represents the cumulative number of potential conflicts along the path extending from the edge; and the number of manually reviewed relationships represents the number of paths involving the edge that require manual confirmation or review. The relationship conflict propagation value is used to measure the conflict risk that may be introduced after retaining the relationship edge, so as to raise the screening threshold for high-risk relationship edges when generating the relationship retention threshold, and reduce failures or manual intervention caused by path conflicts in second-level inference.

[0059] Based on the relationship timeliness pressure coefficient, relationship bottleneck contribution value, and relationship conflict propagation value, hierarchical relationship retention thresholds are generated for each relationship level identifier. Relationship level identifiers are used to distinguish the functional categories of business relationship edges in the graph, such as entry-level receiving relationship edges, resource guarantee relationship edges, process connection relationship edges, quality inspection switching relationship edges, supply response relationship edges, permission constraint relationship edges, and exit-level trigger relationship edges. The hierarchical relationship retention thresholds are adjusted based on the pressure coefficient to raise the threshold for high-latency edges, based on the bottleneck contribution to lower the threshold for key bridging edges, and based on conflict propagation to raise the threshold for high-conflict edges. This achieves a dynamic filtering strategy for different relationship types, ensuring that key relationship edges are retained while reducing the entry of high-latency and high-risk edges into the rapid readiness graph subgraph.

[0060] The formula for calculating the hierarchical relationship retention threshold is as follows:

[0061] in, This represents the threshold for retaining hierarchical relationships corresponding to the r-th relation level identifier; This represents the average relation retention value of the edges at this level; This represents the standard deviation of the retained values. This is the discrete adjustment coefficient, with a value ranging from 0 to 2; This refers to the time-dependent pressure coefficient of the relationship. This is the timeliness weighting coefficient, with a value ranging from 0 to 1; Average bottleneck contribution value; This is the bottleneck protection coefficient, with a value ranging from 0 to 1; This represents the average value of conflict propagation in the relationship. This is the conflict suppression coefficient, with a value between 0 and 1; and These represent the upper and lower limits of the threshold, respectively. This formula achieves dynamic screening threshold settings for different relationship levels by comprehensively considering the timeliness, criticality, and conflict risk of the relationship edges.

[0062] Subsequently, a closed-loop adjustment is performed on the hierarchical relationship retention threshold based on the actual inference time, actual trigger success rate, and the proportion of manual review, forming the final relationship retention threshold. If the actual inference time exceeds the allowable response time limit, the closed-loop adjustment will raise the threshold to compress low-value or high-latency relationship edges; if the trigger success rate decreases or the target execution branch coverage is insufficient, the closed-loop adjustment will lower the threshold to replenish key relationship edges; if the proportion of manual review increases, the threshold for high-conflict relationship edges will be increased accordingly to reduce manual intervention paths. The final relationship retention threshold is used to perform a second filtering process on the relationship retention values. The filtered set of business relationship edges can be directly used to generate a strategic reserve graph subgraph, thereby ensuring that the graph has reachable paths, complete key relationships, and controllable conflicts with a second-level response time.

[0063] The formula for calculating the relationship retention threshold after closed-loop correction is as follows:

[0064] in, This represents the relation retention threshold after closed-loop correction corresponding to the r-th relation level identifier; A threshold is reserved for the initial hierarchical relationship; L represents the actual reasoning time of the previous round; Indicates the maximum allowed response time; For stabilizing parameters; W is the percentage of manual review in the previous round, ranging from 0 to 1; V is the number of target execution branches covered by the current set of pre-reserved relation edges; Minimum coverage quantity; , , These are the time consumption correction coefficient, the verification correction coefficient, and the branch coverage correction coefficient, respectively, with values ​​ranging from 0 to 0.2. This formula dynamically adjusts the relationship retention threshold by comprehensively considering real-time performance, path coverage, and manual verification feedback to ensure that the paths generated from the sub-graphs of the preparedness map are both reliable and efficient at a second-level response time.

[0065] In one possible implementation, similarity matching is performed based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set. Specifically, this includes: performing structured parsing on the decision context set to form context entity fields, context intent fields, context constraint fields, and context state fields; generating a context joint vector based on the context entity set, intent vector, constraint vector, and state gating identifier, wherein object recognition processing is performed on the context entity fields to form the context entity set, intent recognition processing is performed on the context intent fields to form the intent vector, and constraint extraction processing is performed on the context constraint fields to form the constraint vector. The system performs state gating mapping on the context state field to form a state gating identifier; it performs multi-dimensional similarity calculation based on the context joint vector and the vectors corresponding to each candidate graph node read from the process knowledge graph to generate a comprehensive similarity for each candidate graph node; it performs preliminary sorting on the candidate graph nodes based on the comprehensive similarity and selects candidate graph nodes with a comprehensive similarity greater than the similarity threshold to form an initial matching node set; it then performs reachability verification, conflict checking, and deduplication and merging on the branch set in sequence on the initial matching node set, and selects representative nodes that meet the preset minimum and maximum number of starting nodes to form a candidate starting node set.

[0066] Specifically, firstly, a structured parsing process is performed on the decision context set. The original multi-source business data, historical process states, operator attributes, and current task parameters are divided into context entity fields, context intent fields, context constraint fields, and context state fields according to fields and semantics. Context entity fields describe the business objects involved, such as tasks, resources, processes, quality inspection items, or permission nodes. Context intent fields describe the type of business action required by the target decision node, such as production priority adjustment, resource allocation, or approval triggering. Context constraint fields describe path and operational constraints, including response time limits, resource occupancy status, confidentiality isolation, and tolerance requirements. Context state fields indicate the status of the current node, such as available during normal operation, available for emergency use, frozen during switchover, or requiring manual review. Through structured parsing, multi-source information is mapped into a unified field system, ensuring that subsequent feature extraction can be processed based on standardized vectors.

[0067] Subsequently, object recognition processing is performed on the context entity field to form a context entity set. This processing identifies business objects with the same descriptions in different systems or data sources as a single entity node by matching business object identifiers, system codes, and historical aliases, avoiding redundant modeling. Intent recognition processing is performed on the context intent field to form an intent vector. This is typically achieved by converting text descriptions, state signals, or task types into high-dimensional vector representations through rule mapping, word vectors, or deep learning methods, so as to reflect business action requirements in similarity calculations. Constraint extraction processing is performed on the context constraint field to form a constraint vector, transforming response time limits, resource constraints, confidentiality levels, and path conditions into vectorized features. State gating mapping is performed on the context state field to form a state gating identifier, which is used to limit the availability of candidate graph nodes during matching. For example, nodes that are ready for combat use are given priority in matching, while nodes that are switched off or manually reviewed are given lower priority or excluded.

[0068] A context joint vector is generated by fusing the context entity set, intent vector, constraint vector, and state gating identifier to represent the comprehensive semantic and constraint information of the current decision context. This context joint vector not only includes business object information but also incorporates intent and constraint characteristics. It can simultaneously perform multi-dimensional similarity calculations with the graph node vector, node attribute vector, state gating vector, and neighborhood structure vector of each candidate node in the process knowledge graph. The graph node vector represents the semantic features of the node, the node attribute vector represents the node's static and dynamic attributes as well as its readiness attributes, the state gating vector indicates whether the node can participate in the current state reasoning, and the neighborhood structure vector represents the structural relationship between the node and surrounding key nodes, resources, and target branches.

[0069] Multidimensional similarity calculation typically employs a weighted combination of cosine similarity or Euclidean distance, with the following formula:

[0070] in, This represents the overall similarity of the i-th candidate graph node. The larger the value, the more suitable the node is as a candidate starting node. The semantic part of the context joint vector is obtained by fusing the context entity and the intent vector. The graph node vector of the candidate node; The constraint portion of the context joint vector is obtained by fusing constraint vectors. The candidate node attribute vector; This is the reachability value of the node's neighborhood, reflecting the reachability of the node to the target execution branch and key resources; This is the node state gating matching value; This is the node's service handling capacity value, representing its ability to handle critical resources and target branches within the path. This is a real-time risk value, reflecting the potential anomalies or conflicts that may exist in the node; to The weighting coefficients range from 0 to 1, with a total value of 1. This formula comprehensively evaluates candidate nodes based on semantic matching, constraint matching, neighborhood reachability, state adaptation, and business acceptance, and uses real-time risk suppression to prevent unexecutable or highly conflicting nodes from entering the candidate set.

[0071] Candidate graph nodes are initially sorted based on comprehensive similarity, and nodes with similarity greater than a threshold are selected to form an initial set of matching nodes. When the number of initial matching nodes is less than the preset minimum number of starting nodes, an extended recall strategy is used to supplement the graph with nodes that are adjacent to the context entity set and match the state gating, ensuring that subsequent path traversal has sufficient starting point coverage.

[0072] Subsequently, reachability verification is performed on the initial matching node set around the target execution branch set. The number of reachable paths, the shortest path depth, and the path triggering delay of each initial matching node to each target branch are counted. Nodes that cannot form a valid path, exceed the maximum inference depth, or exceed the response time limit are removed and moved to the backup starting node set to ensure that the nodes in the candidate starting node set can support the actual decision path.

[0073] Then, conflict checking is performed on the initial set of matching nodes to detect possible resource mutual exclusion, confidentiality isolation, tolerance conflict, supply chain disruption and manual review requirements in the node association path. Nodes with high conflict and no alternative path are transferred to the candidate range for manual review, while nodes with low conflict and alternative path are retained as backup starting nodes after reducing the comprehensive similarity, so as to optimize the reliability and executability of candidate nodes.

[0074] After conflict verification, the node set is deduplicated and merged. Multiple nodes pointing to the same business object, system code or graph master node are merged into a representative node. The node with the highest comprehensive similarity, the most reachable paths, and the lowest triggering delay is retained as the representative node, thereby avoiding repeated path traversal and redundant calculations.

[0075] Finally, representative nodes are sorted based on comprehensive similarity, number of reachable paths, trigger delay, and conflict verification results. Representative nodes that meet the preset minimum and maximum number of starting nodes are selected to form a candidate starting node set, which provides an executable entry point for subsequent multi-hop path traversal and target reasoning in the process knowledge graph starting from this set.

[0076] Furthermore, starting with the set of candidate starting nodes, the process knowledge graph is traversed, generating all possible paths using a multi-hop path traversal method. Multi-hop path traversal refers to starting from the starting node and sequentially visiting subsequent nodes along the business relationship edges between nodes until the target execution branch or the preset maximum inference depth is reached. Each jump considers not only the topological structure of the nodes in the graph but also node attributes, relationship attributes, state gating identifiers, and business constraints to determine validity, thereby ensuring that the generated path conforms to the requirements of the current operational readiness stage and task context.

[0077] During the traversal, path constraints are applied to each path for filtering. These constraints include maximum inference depth, node state availability, combat readiness phase matching, response time limits, resource consumption conflicts, and security level constraints. For any node or relation edge that does not meet any path constraints, the current path expansion is terminated, and the path is marked as invalid, thus preventing invalid nodes from entering the candidate inference path set. In this way, each candidate path meets the time, resource, and state requirements of the combat readiness scenario at the time of generation.

[0078] Simultaneously with path generation, the association strength of nodes and relationships within each path is calculated. Node association strength reflects the business relevance, logical dependency, and historical cooperation frequency between adjacent nodes in the path, while relationship edge association strength reflects the call frequency, approval dependency, and robustness of supply relationships between business objects. Combining node association strength and relationship edge strength for weighted path scoring ensures that key nodes and key relationships are reflected in path priority ranking, thereby guaranteeing that path traversal results are biased towards important nodes and high-value relationship edges.

[0079] Furthermore, the paths are weighted based on historical hit frequency. Historical hit frequency refers to the number of times nodes and relation edges in a path have been successfully triggered in past combat readiness exercises or actual executions, reflecting the feasibility and reliability of the path in actual business operations. For paths with high historical hit frequency, their weight in the candidate inference path set is increased, ensuring that the final candidate path set prioritizes paths that have been verified to be feasible, thereby improving the accuracy and stability of decision-making.

[0080] When generating a set of candidate inference paths, each path is recorded as a structured object containing a node sequence, a relation edge sequence, node and relation attributes, a state gating identifier, a path trigger delay, and historical hit information. The node sequence stores the business object identifier and business attributes of each node in the path; the relation edge sequence stores the business relationship type, constraints, and weight information between nodes; the path trigger delay records the response time that may be consumed during path execution; and the historical hit information records the hit status of the path in past combat readiness or exercises. With this information, each path can not only be used for subsequent scoring and ranking but also for interpretable decision analysis in rapid combat readiness switching scenarios.

[0081] In practice, a combination of breadth-first or depth-first traversal algorithms and pruning strategies can be used. Pruning strategies include: immediately stopping path expansion when any node or relationship in the path does not meet state gating criteria, response time limits, or resource constraints; terminating further expansion when the path length reaches the maximum inference depth; and filtering or marking paths as backup paths based on thresholds when they contain edges with high-conflict relationships or low-history-historical-hit relationships. These steps effectively control the number of paths and reduce the computational burden on second-level response times due to the scale of the graph.

[0082] The final generated candidate inference path set contains multiple paths that meet path constraints, have high node correlation strength, and have reliable historical hit frequency. This provides a complete input data structure for subsequent path scoring, sorting and filtering, and target inference path determination, ensuring that the war preparedness map subgraph can generate an executable, reliable, and high-value path set in a second-level decision-making scenario.

[0083] In one possible implementation, graph structure analysis is performed on the candidate inference path set to form a path analysis result set including path confidence, path stability, and path fit. Specifically, this includes: calculating the reachability contribution value, path triggering efficiency contribution value, and risk-weighted contribution value for each business entity node in the candidate inference path to obtain the node contribution value; calculating the path acceptance contribution value, relationship timeliness contribution value, conflict impact value, and bottleneck weighted value for each business relationship edge in the candidate inference path to obtain the relationship contribution value; calculating the path confidence, path stability, and path fit for each path based on the node contribution value and relationship contribution value to form a three-dimensional score; calculating a comprehensive similarity value by combining the node contribution value, relationship contribution value, and three-dimensional score, and selecting matching nodes according to a threshold to obtain the path analysis result set.

[0084] Specifically, firstly, for each business entity node in the candidate inference path, a reachability contribution value, a path triggering efficiency contribution value, and a risk-weighted contribution value are calculated to obtain the node contribution value. The reachability contribution value quantifies the node's bridging role and criticality in the path from the candidate starting node to the target execution branch, reflecting whether the node can ensure the target execution branch is reachable. The path triggering efficiency contribution value measures the proportion of the time required for the node to participate in path execution to the total path triggering latency, ensuring the path is executable within a second-level response window. The risk-weighted contribution value considers the node's real-time risks, including equipment malfunction risks, supply chain disruption risks, permission conflict risks, and transportation unavailability risks, reducing the positive impact of high-risk nodes on the path score through risk weighting. The node contribution value is a comprehensive score obtained by weighting the above three indicators, serving as an important input for subsequent path confidence and stability calculations.

[0085] Subsequently, for each business relationship edge in the candidate inference path, the path acceptance contribution value, relationship timeliness contribution value, conflict impact value, and bottleneck weighting value are calculated to obtain the relationship contribution value. The path acceptance contribution value reflects the ability of the relationship edge to connect candidate starting nodes, key resource nodes, and target execution branches, measuring its role in path integrity; the relationship timeliness contribution value reflects the impact of the delay required for triggering the business relationship edge on the completion time of the entire path, ensuring that the path is executable within the response time limit; the conflict impact value is used to quantify the conflict risks that the relationship edge may introduce, such as resource mutual exclusion, confidentiality isolation, tolerance conflict, and supply chain disruption; the bottleneck weighting value is used to emphasize irreplaceable key connections in the path, ensuring that key bridging relationship edges are given priority in the scoring. The relationship contribution value is obtained by combining the above four indicators, providing an important reference for the overall path scoring.

[0086] After calculating the node contribution value and relationship contribution value, these are used to calculate the path confidence, path stability, and path fit for each path, forming a three-dimensional score. Path confidence assesses the probability of a path successfully completing its target execution branch, calculated by weighted summation of node and relationship contributions. Path stability assesses the frequency with which a path is successfully triggered repeatedly in historical readiness procedures or exercises, reflecting path reliability. Path fit assesses whether a path meets the requirements of the current readiness stage, response time limit, state gating, and branch coverage, evaluated by comparing path trigger latency, node availability, and relationship conflict propagation values ​​with current context constraints. The three-dimensional score quantifies the value of each path from three dimensions: reachability, stability, and fit.

[0087] Path confidence, path stability, and path fit are combined with node contribution values ​​and relationship contribution values ​​to calculate a comprehensive similarity value, which is used to determine whether nodes and relationships in the path meet the matching criteria. The comprehensive similarity value can be expressed as a weighted sum or normalized cosine similarity, uniformly mapping the matching degree of nodes and relationships in semantics, structure, and constraint characteristics to a single numerical value. Matching nodes that meet the conditions are selected according to a preset threshold, while nodes with low similarity values ​​or high risks are removed, thereby ensuring that the paths in the final path analysis result set are reachable, reliable, and adapted to the needs of combat readiness scenarios.

[0088] The formula for calculating the comprehensive similarity value is as follows:

[0089] in, This represents the overall path similarity value; The sum of the contribution values ​​of the nodes in the path; The sum of the contributions of the relationships in the path; , , These are path confidence, path stability, and path fit, respectively. This represents the sum of path risks. to The weighting coefficients range from 0 to 1, and their sum is 1. This formula allows for the unified quantification of node and relationship contributions, path 3D scoring, and risk information. It is used to filter and match nodes and form the final path analysis result set, ensuring that the path analysis results reflect both business logic and reliability and real-time performance in rapid combat readiness switching scenarios.

[0090] In one possible implementation, a sorting and filtering process is performed based on the path analysis result set to determine the target inference path. The process instance is then driven to enter the target execution branch according to the business action identifier corresponding to the target inference path. Specifically, this includes: normalizing the path confidence, path stability, and path suitability to form a comprehensive analysis value; performing a preliminary sorting of candidate inference paths based on the comprehensive analysis value to form a candidate path priority list; selecting a preset number of paths with the highest comprehensive scores from the candidate path priority list as the target inference path according to the target selection strategy; and parsing the business action identifiers of each node in the selected target inference path to form an executable business action sequence in path order.

[0091] Specifically, firstly, the path confidence, path stability, and path fit of each path in the path analysis result set are normalized. The purpose of normalization is to unify the scoring indicators with different dimensions and value ranges to the range of 0 to 1, so that subsequent comprehensive analysis can compare the merits of each path on the same scale. In practice, the min-max normalization method can be used for each indicator:

[0092]

[0093] Where X represents the original score, and These are the minimum and maximum values ​​of the indicator in the candidate path set, respectively. To avoid stable parameters with a denominator of zero, after normalization, the three indicators are weighted and summed according to preset weights to form a comprehensive analysis value. This is used for subsequent path sorting. The weights can be dynamically adjusted based on the combat readiness stage, response time limit, and task priority.

[0094] Subsequently, the candidate inference paths are initially ranked based on the comprehensive analysis values, forming a priority list of candidate paths. This initial ranking prioritizes paths with higher scores, using the comprehensive analysis value as the criterion. Simultaneously, the comprehensive analysis values ​​can be fine-tuned during the ranking process by incorporating path length, number of nodes, trigger latency, and historical hit frequency. This ensures that the ranking results not only reflect path scores but also consider real-time executability and reliability, guaranteeing that critical paths are prioritized for subsequent processing.

[0095] Based on the target selection strategy, the highest-scoring paths from the candidate path priority list are selected as the target inference paths. The target selection strategy includes the minimum number of retained paths, the maximum allowable inference latency, and the minimum path fit threshold. When there are insufficient high-scoring paths or the trigger latency is close to the allowable limit, the strategy allows a fallback to an alternative or suboptimal path to ensure that the target execution branch is reachable and the path execution is stable.

[0096] Then, the business action identifiers of each node in the selected target inference path are parsed to form an executable sequence of business actions arranged in the path order. Business action identifiers include production priority adjustment actions, resource allocation actions, quality inspection tolerance switching actions, supply chain response actions, and permission verification actions. During parsing, node status gating identifiers and node availability status are combined to eliminate node actions in switch-frozen or manually reviewed states, ensuring that the formed business action sequence can be implemented in actual execution and avoiding path execution failures due to node unavailability.

[0097] Ultimately, the target inference path and its sequence of business actions are used to drive process instances into the target execution branch. At the same time, the trigger time, node status and execution result of each action are recorded, providing a complete basis for subsequent path optimization, node weight updates and decision interpretability, thereby achieving second-level response and reliable decision-making in the scenario of rapid switching of combat readiness.

[0098] This embodiment also discloses an intelligent process decision processing device that integrates knowledge graphs, referring to... Figure 2 The device includes an acquisition module 201, a processing module 202, and an output module 203. It is used to execute any of the above-described intelligent process decision-making methods that integrate knowledge graphs, wherein: The acquisition module 201 is used to acquire multi-source business data from multiple systems, perform preprocessing to form a standard knowledge source data set, and when the process instance runs to the target decision node, perform fusion processing based on the acquired real-time business context data to form a decision context set. Processing module 202 is used to perform similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, wherein the process knowledge graph is constructed based on the standard knowledge source data set; The processing module 202 is used to perform multi-hop path traversal in the process knowledge graph starting from the candidate starting node set, and generate a candidate reasoning path set based on path constraints, node association strength and historical hit frequency. The processing module 202 is used to perform graph structure analysis processing on the candidate inference path set to form a path analysis result set including path confidence, path stability and path fitness. The output module 203 is used to perform sorting and filtering processing based on the path analysis result set, determine the target inference path, and drive the process instance to enter the target execution branch according to the business action identifier corresponding to the target inference path.

[0099] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0100] This embodiment also discloses an electronic device, as shown in the reference. Figure 3 The electronic device may include: at least one processor 301, at least one communication bus 302, user interface 303, network interface 304, and at least one memory 305.

[0101] The communication bus 302 is used to enable communication between these components.

[0102] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0103] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0104] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0105] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory may include a non-transitory computer-readable storage medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. As a computer storage medium, the memory 305 may include an operating system, a network communication module, a user interface 303 module, and an application program for an intelligent process decision processing method integrating knowledge graphs.

[0106] exist Figure 3 In the electronic device shown, the user interface 303 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 301 can be used to call an application program stored in the memory 305 that integrates a knowledge graph intelligent process decision processing method. When executed by one or more processors 301, the electronic device executes one or more methods as described in the above embodiments.

[0107] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0108] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0109] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0110] 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.

[0111] Furthermore, the functional units in the various embodiments of the present invention 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. The integrated unit can be implemented in hardware or as a software functional unit.

[0112] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 305 and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned memory 305 includes various media capable of storing program code, such as a USB flash drive, external hard drive, magnetic disk, or optical disk.

[0113] The present invention also discloses a non-transitory computer-readable storage medium storing instructions. When executed by one or more processors 301, these instructions cause an electronic device to perform one or more methods as described in the above embodiments.

[0114] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of other embodiments of this disclosure upon considering the specification and the disclosure of practical truths. This invention is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for intelligent process decision processing that integrates knowledge graphs, characterized in that, The method includes: Multi-source business data from multiple systems is acquired and preprocessed to form a standard knowledge source data set. When the process instance runs to the target decision node, fusion processing is performed based on the acquired real-time business context data to form a decision context set. Similarity matching is performed based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, wherein the process knowledge graph is constructed based on the standard knowledge source data set; Starting from the set of candidate starting nodes, perform multi-hop path traversal in the process knowledge graph, and generate a set of candidate reasoning paths based on path constraints, node association strength, and historical hit frequency. Graph structure analysis is performed on the candidate inference path set to form a path analysis result set that includes path confidence, path stability, and path fitness. Based on the path analysis result set, a sorting and filtering process is performed to determine the target inference path, and the process instance is driven to enter the target execution branch according to the business action identifier corresponding to the target inference path.

2. The intelligent process decision processing method integrating knowledge graphs according to claim 1, characterized in that, Before performing similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, the method further includes: The standard knowledge source data set is subjected to hierarchical parsing, and business object identifiers, status time identifiers, process identifiers, and combat readiness stage identifiers are bound to form a combat readiness decision semantic space. Entity identification and entity merging processes are performed on the aforementioned combat readiness decision semantic space to form a set of business entity nodes; Perform attribute mapping processing on the set of business entity nodes, and write static attributes, dynamic attributes, and combat readiness attributes into the entity attribute set; Based on the semantic space of the combat readiness decision, a set of business relationship edges containing ordinary business relationships and combat readiness constraint relationships is formed. Constraint association processing is performed on the set of business relationship edges to bind the combat readiness stage identifier, response time limit, confidence source, applicable scope, failure condition and conflict level to the corresponding business relationship edge, and a delayed inference identifier is set for non-real-time business relationship edges. Calculate node retention values ​​and relationship retention values ​​based on the set of business entity nodes and the set of business relationship edges, and filter them to form a strategic preparedness graph subgraph; A two-layer index structure, including a real-time scheduling index and an audit traceability index, is established for the process knowledge graph. Set status gating identifiers for the business entity nodes and the business relationship edges, and perform switching control of the relationship between normal production status and combat readiness support status according to the combat readiness switching signal; The process knowledge graph is formed based on the business entity nodes, the business relationship edges, the entity attribute set, the strategic preparedness graph subgraph, the two-layer index structure, and the state gating identifier.

3. The intelligent process decision processing method integrating knowledge graphs according to claim 2, characterized in that, Before calculating node retention values ​​and relationship retention values ​​based on the set of business entity nodes and the set of business relationship edges, and filtering to form a strategic preparedness graph subgraph, the method further includes: Obtain the combat readiness stage identifier, target execution branch set, response time limit, inference depth, and candidate starting node set corresponding to the target decision node, and obtain the set of business entity nodes associated with the candidate starting node set; Obtain the node retention value, combat readiness priority, response time limit, current availability status, real-time risk value, and business type of each business entity node in the set of business entity nodes; The real-time compression coefficient is calculated based on the response time limit, the inference depth, and the graph congestion. The initial node retention threshold is calculated based on the distribution statistics of the node retention values ​​and the real-time compression coefficient. Based on the business type, the initial node retention threshold is adjusted by type balancing to form the local node retention threshold corresponding to each business type; The local node retention threshold is modified by combining the state gating identifier and modified by the second modification based on the actual inference time and the number of candidate inference paths to form a node retention threshold. The node retention threshold is used to perform a first filtering process on the node retention value, so that the set of business entity nodes after the first filtering process is used to form the strategic preparedness map subgraph.

4. The intelligent process decision processing method integrating knowledge graphs according to claim 2, characterized in that, Before calculating node retention values ​​and relationship retention values ​​based on the set of business entity nodes and the set of business relationship edges, and filtering to form a strategic preparedness graph subgraph, the method further includes: Obtain the combat readiness stage identifier, target execution branch set, candidate starting node set, response time limit, inference depth, and candidate range of the graph subgraph corresponding to the target decision node, and obtain the set of business relationship edges located between the candidate starting node set and the target execution branch set; Based on the relationship timeliness pressure coefficient, the relationship bottleneck contribution value, and the relationship conflict propagation value, a hierarchical relationship retention threshold is generated for each relationship level identifier. The relationship timeliness pressure coefficient is calculated based on the response time limit, the trigger delay distribution of the business relationship edge set, and the number of currently reachable paths. The relationship bottleneck contribution value is calculated based on the number of times the path appears in the effective inference path, the number of connected target execution branches, and the shortest path depth. The relationship conflict propagation value is calculated based on the conflict level of each business relationship edge, the number of reachable conflicting relationships within the preset inference depth, and the number of manually reviewed relationships. The hierarchical relationship retention threshold is modified based on the actual reasoning time, actual trigger success rate, and proportion of manual review to form a relationship retention threshold. The relationship retention threshold is used to perform a second filtering process on the relationship retention value, so that the set of business relationship edges after the second filtering process is used to form the strategic reserve graph subgraph.

5. The intelligent process decision processing method integrating knowledge graphs according to claim 1, characterized in that, The step of performing similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set specifically includes: The decision context set is subjected to structured parsing to form context entity fields, context intent fields, context constraint fields, and context state fields; A context joint vector is generated based on the context entity set, intent vector, constraint vector, and state gating identifier. The context entity set is formed by performing object recognition processing on the context entity field, the intent vector is formed by performing intent recognition processing on the context intent field, the constraint vector is formed by performing constraint extraction processing on the context constraint field, and the state gating identifier is formed by performing state gating mapping on the context state field. Multidimensional similarity calculation is performed based on the context joint vector and the vectors corresponding to each candidate graph node read from the process knowledge graph to generate a comprehensive similarity for each candidate graph node. The candidate graph nodes are initially sorted based on the comprehensive similarity, and candidate graph nodes with a comprehensive similarity greater than the similarity threshold are selected to form an initial set of matching nodes. After performing reachability verification, conflict checking, and deduplication and merging on the initial matching node set in sequence, representative nodes that meet the preset minimum and maximum number of starting nodes are selected to form the candidate starting node set.

6. The intelligent process decision processing method integrating knowledge graphs according to claim 1, characterized in that, Graph structure analysis is performed on the candidate inference path set to generate a path analysis result set containing path confidence, path stability, and path fitness, specifically including: For each business entity node in the candidate inference path, calculate the reachability contribution value, the path triggering efficiency contribution value, and the risk-weighted contribution value to obtain the node contribution value; For each business relationship edge in the candidate reasoning path, calculate the path acceptance contribution value, relationship timeliness contribution value, conflict impact value, and bottleneck weighting value to obtain the relationship contribution value; Based on the node contribution value and the relationship contribution value, the path confidence, path stability and path fitness of each path are calculated to form a three-dimensional score; The node contribution value, the relationship contribution value, and the three-dimensional score are combined to calculate a similarity value, and matching nodes are selected according to a threshold to obtain the path analysis result set.

7. The intelligent process decision processing method integrating knowledge graphs according to claim 1, characterized in that, The process of sorting and filtering based on the path analysis result set to determine the target inference path, and then driving the process instance to enter the target execution branch according to the business action identifier corresponding to the target inference path, specifically includes: The path confidence, path stability, and path fitness are normalized to form a comprehensive analysis value; Based on the comprehensive analysis values, the candidate reasoning paths are initially sorted to form a priority list of candidate paths; Based on the target selection strategy, a preset number of paths with the highest comprehensive scores are selected from the candidate path priority list as the target inference paths; Parse the business action identifiers of each node in the selected target inference path to form an executable business action sequence in the order of the path.

8. An intelligent process decision processing device integrating knowledge graphs, characterized in that, The apparatus is used to execute an intelligent process decision processing method integrating knowledge graphs as described in any one of claims 1-7, the apparatus comprising an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire multi-source business data from multiple systems, perform preprocessing to form a standard knowledge source data set, and when the process instance runs to the target decision node, perform fusion processing based on the acquired real-time business context data to form a decision context set. The processing module is used to perform similarity matching processing based on the decision context set and the graph node vectors in the process knowledge graph to determine the candidate starting node set, wherein the process knowledge graph is constructed based on the standard knowledge source data set; The processing module is used to perform multi-hop path traversal in the process knowledge graph starting from the candidate starting node set, and generate a candidate reasoning path set based on path constraints, node association strength and historical hit frequency. The processing module is used to perform graph structure analysis processing on the candidate inference path set to form a path analysis result set including path confidence, path stability and path fitness. The output module is used to perform sorting and filtering processing based on the path analysis result set, determine the target inference path, and drive the process instance to enter the target execution branch according to the business action identifier corresponding to the target inference path.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing a computer program, wherein the computer program, when executed by the processor, implements an intelligent process decision processing method that integrates knowledge graphs as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform an intelligent process decision-making method integrating knowledge graphs as described in any one of claims 1-7.

Images