A causal graph-driven enterprise intelligent decision-making method and system

Abstract

The application discloses a kind of causal graph driven enterprise intelligent decision-making method and system, and the application belongs to enterprise management technical field.The method includes: by obtaining enterprise business process, operating index and unstructured text data, constructs enterprise-level causal graph;Based on the data to be decided data execution path reasoning, determine multi-hop causal path chain;Determine current correlation rule and decision suggestion in combination with expert rule base;Introduce the causal graph and rule label of different industries, based on structure similarity and semantic mapping rule, identify transferable causal path and correlation rule;Finally, the current and migration knowledge are fused, and enterprise decision execution scheme and causal reasoning result are generated.The scheme is based on the comprehensive application of current and migration knowledge, effectively enhances the scientificity, accuracy and interpretability of decision-making, while supporting dynamic updating and inter-industry knowledge sharing, significantly improves enterprise decision efficiency and response speed, helps enterprises to realize intelligent and fine management.

Description

Technical Field

[0001] The present application belongs to the field of enterprise management technology, and specifically relates to an enterprise intelligent decision-making method and system driven by a causal graph. Background Art

[0002] Amidst the wave of digital economy and intelligent transformation, corporate decision-making faces the dual challenges of data explosion and complex business scenarios. Traditional decision support systems, which rely heavily on statistical correlation analysis or simple rule matching, struggle to reveal the causal relationships behind the data, resulting in a lack of explainability and robustness in decision-making.

[0003] Today, knowledge graph construction technology is achieved through three key steps: information extraction, knowledge fusion, and semantic reasoning. In the information extraction stage, tools such as spaCy and LSTM-CRF are used for entity recognition, combined with remote supervision or reinforcement learning methods to extract relationships. Entity disambiguation is then resolved through pattern matching or graph embedding techniques. The knowledge fusion stage uses rule-based matching or graph embedding algorithms such as TransE to align entities across data sources, ultimately constructing a semantic network that supports both symbolic and statistical reasoning. This structured knowledge representation has been widely used in intelligent customer service, recommendation systems, and risk prevention and control. For example, in financial scenarios, related transaction data is integrated to identify fraud risk patterns.

[0004] However, in the information extraction phase, existing technologies for entity recognition based on pre-trained models (such as spaCy) are prone to missed recall when processing long-tail entities in vertical domains. Sequence labeling models such as LSTM-CRF lack support for nested entities and coreference resolution. Remote supervision methods suffer from a 15%-30% error rate in relation extraction due to noise from automatic labeling. The contextual features relied on for entity disambiguation are susceptible to interference from text abbreviations and polysemy. Graph embedding technologies (such as TransE) experience performance degradation of over 40% in low-resource scenarios. In the knowledge fusion phase, rule-based matching strategies struggle to cope with the semantic heterogeneity of cross-lingual and cross-modal data. Traditional graph embedding algorithms such as TransE have an accuracy rate of less than 60% when processing complex N-ary relationships. Entity alignment across data sources faces fusion gaps caused by data version differences (e.g., differences in statistical caliber between departments). Summary of the Invention

[0005] The embodiments of the present application provide a causal graph-driven enterprise intelligent decision-making method and system to address the existing technology problems in the information extraction stage. Entity recognition based on pre-trained models (such as spaCy) is prone to missed recall when processing long-tail entities in vertical fields. Sequence labeling models such as LSTM-CRF lack support for nested entities and coreference resolution. Remote supervision methods have an error rate of 15%-30% in relationship extraction due to automatic labeling noise. The contextual features that entity disambiguation relies on are easily affected by text abbreviations and polysemy. Graph embedding technologies (such as TransE) experience a performance drop of over 40% in low-resource scenarios. In the knowledge fusion stage, rule-based matching strategies are difficult to cope with the semantic heterogeneity of cross-language and cross-modal data. Traditional graph embedding algorithms such as TransE have an accuracy rate of less than 60% when processing N-ary complex relationships. Entity alignment across data sources faces the problem of fusion gaps caused by data version differences (such as differences in statistical calibers between different departments).

[0006] In a first aspect, an embodiment of the present application provides a causal graph-driven enterprise intelligent decision-making method, the method comprising:

[0007] Acquire enterprise business process data, operating indicator data, and unstructured text data, and construct an enterprise-level cause-and-effect graph based on the enterprise business process data and operating indicator data;

[0008] Obtaining business data to be decided, performing path reasoning in the enterprise-level causal graph based on the business data to be decided, and determining a multi-hop causal path chain;

[0009] Determining current association rules and current decision recommendations based on the multi-hop causal path chain and a preset expert rule base;

[0010] Obtain causal graph data and corresponding rule label information from different industries, determine structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determine transferable causal paths and transferable association rules based on the structural similarity and node semantic mapping rules;

[0011] Generate enterprise decision execution plans and causal reasoning results based on current association rules, current decision recommendations, transferable causal paths, and transferable association rules.

[0012] In a second aspect, an embodiment of the present application provides an enterprise intelligent decision-making system driven by a causal graph, the system comprising:

[0013] A cause-effect graph construction module is used to obtain enterprise business process data, operating indicator data, and unstructured text data, and to construct an enterprise-level cause-effect graph based on the enterprise business process data and operating indicator data;

[0014] A path reasoning module is used to obtain business data to be decided, perform path reasoning in the enterprise-level causal graph based on the business data to be decided, and determine a multi-hop causal path chain;

[0015] A rule and suggestion determination module, configured to determine the current association rule and the current decision suggestion based on the multi-hop causal path chain and a preset expert rule library;

[0016] A cross-industry migration mapping module is used to obtain causal graph data and corresponding rule label information from different industries, determine structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determine migratable causal paths and migratable association rules based on the structural similarity and node semantic mapping rules;

[0017] The decision-making plan generation module is used to generate enterprise decision execution plans and causal reasoning results based on current association rules, current decision suggestions, transferable causal paths and transferable association rules.

[0018] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the method described in the first aspect.

[0019] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0020] In an embodiment of the present application, enterprise business process data, operating indicator data and unstructured text data are obtained, and an enterprise-level causal graph is constructed based on the enterprise business process data and operating indicator data; business data to be decided is obtained, and path reasoning is performed in the enterprise-level causal graph based on the business data to be decided to determine a multi-hop causal path chain; current association rules and current decision recommendations are determined based on the multi-hop causal path chain and a preset expert rule library; causal graph data and corresponding rule label information from different industries are obtained, and structural similarity and node semantic mapping rules are determined based on the causal graph data and corresponding rule label information, and migratable causal paths and migratable association rules are determined based on structural similarity and node semantic mapping rules; enterprise decision execution plans and causal reasoning results are generated based on current association rules, current decision recommendations, migratable causal paths and migratable association rules. Through the above-mentioned enterprise intelligent decision-making method driven by causal graphs, accurate decision reasoning is achieved by integrating enterprise-level causal graphs with expert rule libraries and combining multi-hop causal path chains, and introducing cross-industry causal path migration and association rule migration mechanisms, the generalization ability and adaptability of decision models are improved. Based on the comprehensive application of current and migrated knowledge, it effectively enhances the scientificity, accuracy and interpretability of decision-making, while supporting dynamic updates and knowledge sharing among industries, significantly improving enterprise decision-making efficiency and response speed, and helping enterprises achieve intelligent and refined management. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the enterprise intelligent decision-making method driven by the causal graph provided in Example 1 of the present application;

[0022] Figure 2 This is a flow chart of the enterprise intelligent decision-making method driven by the causal graph provided in Example 2 of the present application;

[0023] Figure 3 This is a schematic diagram of the structure of the enterprise intelligent decision-making system driven by the causal graph provided in Example 3 of the present application;

[0024] Figure 4 This is a schematic diagram of the structure of the electronic device provided in Example 4 of the present application. DETAILED DESCRIPTION

[0025] To further clarify the objectives, technical solutions, and advantages of this application, specific embodiments of the present application are described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are intended only to illustrate this application and are not intended to limit it. It should also be noted that, for ease of description, the drawings only illustrate portions relevant to this application, not all of them. Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts depict the various operations (or steps) as sequential processes, many of the operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process may terminate upon completion of its operations, but may also include additional steps not shown in the accompanying drawings. The process may correspond to a method, function, procedure, subroutine, subprogram, or the like.

[0026] The following will be combined with the accompanying drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0027] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of this application can be implemented in an order other than that illustrated or described herein, and that the objects distinguished by "first," "second," and the like are generally of the same type, and do not limit the number of objects; for example, the first object can be one or more. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects connected are in an "or" relationship.

[0028] The following, in conjunction with the accompanying drawings, describes in detail a causal graph-driven enterprise intelligent decision-making method and system provided by the embodiment of the present application through specific embodiments and their application scenarios.

[0029] Example 1

[0030] Figure 1 This is a flow chart of the enterprise intelligent decision-making method driven by the causal graph provided in Example 1 of this application. Figure 1 As shown, the specific steps include:

[0031] S101, obtaining enterprise business process data, operating indicator data and unstructured text data, and constructing an enterprise-level cause-and-effect graph based on the enterprise business process data and operating indicator data.

[0032] Enterprise business process data can be structured data that reflects the various business links, operational processes and their causal relationships within the enterprise, such as: order processing process (order → review → delivery → settlement).

[0033] Operating indicator data can refer to key quantitative indicators (KPIs) that can reflect the company's operating status, performance level and business results, such as: revenue, cost, gross profit margin, ROI, inventory turnover rate; customer retention rate, complaint rate, order conversion rate, equipment utilization rate; key performance values ​​of each business department (such as sales, customer satisfaction).

[0034] Unstructured text data refers to free-format information in an enterprise that cannot be directly represented using conventional tables. This includes internal documents (meeting minutes, strategy documents, and email communication records); customer feedback comments, customer service conversation records, and social media content; and external texts such as news reports, policy interpretations, and industry analysis reports.

[0035] An enterprise-level causal graph can be a knowledge graph that expresses the causal relationship between various variables (events, operations, indicators) in the enterprise business in the form of a graph structure. It consists of nodes and directed edges: nodes: represent business events, process nodes, operating indicators, etc.; directed edges: represent the causal relationship between variables (for example, "advertising → sales growth"); each edge can contain attribute information such as causal weight, impact intensity, trigger conditions, time delay, etc.

[0036] The acquisition and modeling of enterprise business process data involves extracting data containing fields such as process definition, execution trajectory, task status, and process sequence from enterprise information systems (including ERP, MES, WMS, and CRM) through interface integration or data synchronization. Simultaneously, process log files are parsed to extract the start / end times, sequential relationships, and trigger conditions of each process node. This data is then represented using a flowchart or Petri net structure, with nodes representing events or actions and edges representing execution sequences and logical constraints. This provides a logical foundation for the subsequent construction of an enterprise-level causal graph. The extraction and modeling of operating indicator data involves extracting time-series structured KPI data (such as sales revenue, customer satisfaction, inventory turnover, and energy consumption indicators) from an enterprise data warehouse or analysis system. This data is then decomposed, normalized, and trended. Potential causal relationships between operating indicator data are identified through methods such as covariance analysis, Granger causality tests, and cross-correlation. These causal relationships are then mapped as node-to-node edges in an enterprise-level causal graph. The processing and causal extraction of unstructured text data refers to the collection of text materials such as internal management reports, meeting minutes, email correspondence, employee feedback, quality accident descriptions, as well as external news information, policy documents, user comments and other unstructured text data, using natural language processing technology for word segmentation, part-of-speech tagging, named entity recognition and dependency syntax analysis, combined with rule-based or model-based causal relationship extraction methods (such as the BERT-based relationship extraction model), to identify explicit or implicit causal expressions in the text (such as "due to raw material delays, production plans lag"), extract causal event pairs and their contextual weights, and map them into causal edges in the enterprise-level causal graph. The construction and integration of causal graph structure is to uniformly abstract the business events, process nodes, and indicator items identified in the above-mentioned enterprise business process data, operating indicator data, and unstructured text data into graph nodes, and model process dependencies, temporal causality, and text causality as directed edges in the graph. Each edge can be accompanied by attribute information such as causal strength, confidence, time delay, action period, and its source identification (such as from process, structured indicator or text); use graph databases (such as Neo4j) or graph embedding representation methods (such as GCN, TransE) for unified organization and storage, thereby constructing an enterprise-level causal graph with a unified semantic structure.

[0037] S102, obtaining business data to be decided, performing path reasoning in the enterprise-level causal graph according to the business data to be decided, and determining a multi-hop causal path chain.

[0038] Business data for decision-making refers to the data input required for intelligent decision-making in specific business scenarios. It typically includes contextual information related to the current business problem, event, or goal. Specifically, it includes the contextual state of the current business process (such as the current node, execution status, and abnormal signals); operational indicators related to the business (such as rising customer complaints and insufficient inventory); environmental constraints or external changes (such as policy changes and market fluctuations); and optional user-input goals (such as reducing costs or improving efficiency).

[0039] A multi-hop causal path chain can be defined as one or more directed paths in an enterprise-level causal graph, formed by reasoning through multiple consecutive causal edges, starting from an initial node (usually corresponding to the business data to be decided). This path represents a chain of reasoning from the "current event" to the "potential cause" or "possible consequence."

[0040] By integrating multiple data sources, including real-time business status information, process trigger conditions, and indicator anomaly signals within the enterprise's current operational environment, the system automatically extracts contextual information about pending tasks, node execution status, current business parameters, and their timestamps from internal enterprise systems (such as ERP, MES, and BI analytics platforms). Combined with user-entered goals or expectations (such as cost optimization and risk avoidance), this system constructs a set of structured business data describing the current state of the business. This data typically includes the identifier of the current triggering event, associated operating indicators (such as "increased customer complaint rate" or "abnormal production line load"), and potentially related historical process trajectory information, serving as the starting point for graph reasoning. Next, within the constructed enterprise-level causal graph, starting with the graph node corresponding to the business data to be decided, the system traverses the graph along the causal edges using a path search algorithm (such as depth-first search (DFS), breadth-first search (BFS), or heuristic search (A*) with weighted pruning). During the traversal process, the system screens causal edges based on conditional constraints, temporal attributes (such as action period and impact delay), and edge weights (indicating causal strength or confidence), eliminating branch paths with low relevance or that fail to meet logical conditions. Only paths that may impact the current business state or trigger consequences are retained. This reasoning process starts from the initial node and gradually expands to its direct consequences, indirect consequences, or potential causes, ultimately identifying a multi-hop causal chain leading from the input state to other key business variables.

[0041] S103: Determine the current association rules and the current decision suggestion according to the multi-hop causal path chain and the preset expert rule library.

[0042] The preset expert rule base can refer to a set of rules that are pre-organized and structured based on the company's internal expert knowledge, industry standards, management experience, etc., and are used to assist in the interpretation of causal graph reasoning results and the generation of decision-making recommendations. Its core content includes: rule conditions: a combination of conditions consisting of one or more business events or causal nodes, which may also include triggering numerical thresholds or logical operators (such as "if the inventory turnover rate is less than 3 and the return rate increases"); rule actions: the corresponding recommended behavior or risk response strategy when the rule conditions are met (such as "recommend increasing inventory", "triggering risk warning", "rescheduling production", etc.); applicable industry and scenario tags: each rule can be attached with metadata such as the industry to which it belongs, the business link, and the time scope of application; confidence / priority: optional, used to support multi-rule conflict processing and sorting.

[0043] Current association rules can be defined as a set of expert rules matching nodes or subpaths identified in a multi-hop causal chain. They represent the consistency between the current business state or event and the known rule conditions in the rule base. For example, if a path includes the nodes "Declining customer satisfaction" → "Increasing complaint rate," the rule base might contain the rule "If customer satisfaction decreases, customer service follow-up visits should be strengthened." This is the current association rule.

[0044] The current decision recommendation refers to the specific decision output generated by the rule action portion of the current association rule. This refers to the action recommended by the system based on the causal trigger patterns in the path and the matching rules. This may include: risk warnings (e.g., "Expected increased risk of customer churn"); operational strategies (e.g., "Recommend delaying delivery and adjusting resource allocation"); management recommendations (e.g., "Increase the frequency of quality inspections"); and automated control responses (e.g., "Automatically adjust production plans").

[0045] The system obtains a multi-hop causal path chain derived from enterprise-level causal graph reasoning. This chain consists of a series of directed, connected graph nodes, each corresponding to a business event or operational variable, representing the possible consequences or root causes of the current business state through several layers of causal influence relationships. The system then analyzes each node in the chain and its combined relationships, extracting business event pairs or event sequences that may serve as rule triggering conditions. The system then searches the expert rule library, which is stored in a "condition-action" format. The condition portion consists of event combinations, numerical thresholds, and time sequences. Based on the events extracted from the multi-hop causal path chain and the current business data context, the system logically matches the rule conditions. This includes determining whether there is an exact match between the node events, whether the threshold conditions are met, and whether the time sequence constraints defined by the rule are met (for example, a rule requires "event A to occur before event B"). Matched rules are identified as current association rules, which are triggerable in the current business context. Next, the system extracts the action portion of each current association rule—the recommended response action upon triggering—to form the current decision recommendation. For example, if a rule is defined as "If customer satisfaction decreases for two consecutive weeks and the return rate increases, then it is recommended to strengthen after-sales intervention," then, provided the conditions are met, the system will generate "strengthen after-sales intervention" as the decision recommendation. Furthermore, the system can attach metadata such as the source rule identifier, trigger node path, and matching confidence to each recommendation for subsequent sorting, explainability display, and user review. To handle situations where multiple rules are triggered simultaneously, the system supports a priority sorting mechanism (e.g., by confidence, rule level, urgency, etc.) and can also merge similar recommendations to generate a comprehensive response plan.

[0046] On the basis of the above technical solution, optionally, determining the current association rules and the current decision suggestion according to the multi-hop causal path chain and the preset expert rule library includes:

[0047] Obtaining business events corresponding to each node in the multi-hop causal path chain, and determining matching rule items from a preset expert rule library based on the business events;

[0048] Determine the triggering conditions of the rule items and the current node states of the multi-hop causal path chain, match the triggering conditions of the rule items and the current node states of the multi-hop causal path chain, and determine the current association rule that meets the triggering conditions;

[0049] A current decision suggestion for the target business is generated according to the current association rule.

[0050] In this solution, business events are observable and recordable activities, behaviors, or state changes that occur during a company's actual operations. They typically correspond to nodes in a causal graph. These events can include: operational actions within a process (e.g., "order completed," "machine failure," "supply delay"); changes in metrics (e.g., "increased customer churn," "decline in sales"); and events extracted from text (e.g., "increased complaints," "policy release").

[0051] A rule item is the basic unit in the expert rule base, representing a formalized business knowledge rule. It typically uses an "IF condition THEN action" structure, reflecting the logic between a specific business event and a recommended action. For example: IF "inventory has decreased for two consecutive periods and there is no replenishment plan," THEN "issue an inventory risk warning"; IF "complaint rate increases AND customer churn rate is abnormal," THEN "initiate a customer satisfaction improvement program."

[0052] A trigger condition is the "IF" portion of a rule, used to determine whether the rule is activated. A trigger condition typically describes the status, trend, or combination of business events, such as "order processing time > 48 hours," "the number of complaints has increased by more than 20% in the past seven days," or "the device operating temperature is above the critical value and the number of maintenance visits is zero."

[0053] The current node state can refer to the real-time or near-real-time business state data corresponding to each node (i.e., business event) in a multi-hop causal path chain within the current reasoning context. For example, if the path chain includes "device temperature → alarm → production stop", the current node state might be: "Device temperature = 85°C" (collected within the last hour); "Alarm status = Triggered"; "Production stop status = No".

[0054] The current association rules refer to the set of valid rule items identified in the current business scenario by conditionally matching the node states in the multi-hop causal path chain with the rule items in the expert rule base. These rules represent "which rules in the current causal chain are applicable," allowing for further recommendations.

[0055] A target business can be a business object or scenario currently in a decision-making state that requires recommendations or actions. This could include a business process node (e.g., "Should the shipping plan be adjusted?"); a management issue (e.g., "Should production be overtime?"); or a business goal (e.g., "Should a customer retention mechanism be triggered?").

[0056] For the multi-hop causal path chain identified in the enterprise-level causal graph, the business events corresponding to each node in the chain are retrieved. Based on the graph node identifiers, semantic labels, and contextual information, this process maps the graph nodes to event identifiers in the actual business system, such as "inventory alarm," "delayed delivery," or "increased customer churn," through an event mapping table or semantic retrieval mechanism. These events are then uniformly represented as a set of structured business events. The system then uses these business events as keywords or conditions to retrieve rule items from a pre-set expert rule library that match their semantics or attributes. This matching process can combine keyword matching, rule ontology reasoning, or semantic similarity calculation based on semantic embedding models such as BERT to identify candidate rules relevant to the current business context. Each candidate rule item typically includes at least one set of trigger conditions, such as "temperature exceeds threshold," "number of complaints increases for two consecutive days," or "inventory turnover rate falls below warning threshold." The system then retrieves real-time or near-real-time status information corresponding to each node in the multi-hop causal path chain, forming a set of current node states within that path. These states can be derived from the enterprise's real-time database, sensor acquisition systems, or historical records. The system then matches the trigger conditions of each rule item with the corresponding current node state, using a Boolean rule parser, range checker, or fuzzy matching algorithm (for imprecise conditions) to determine whether the rule's trigger conditions are met. When all trigger conditions for a rule are met, the rule is marked as the current association rule. Finally, based on the recommended actions or optimization solutions defined in the "THEN" section of the matched current association rule, combined with the current target business context (e.g., "production scheduling," "risk warning," or "marketing decision"), the system generates actionable current decision recommendations. Recommendations can include specific operational strategies, resource adjustment plans, or management interventions, and can be executed automatically or after manual review. A visual interface displays the rule source, recommendation logic, and expected impact. Rule prioritization, confidence scoring, and manual review mechanisms can be incorporated throughout the entire process to enhance the accuracy and controllability of generated recommendations.

[0057] In this solution, by deeply integrating the multi-hop causal path chain with the expert rule base, business events and trigger conditions related to the current business scenario are accurately identified, current association rules are intelligently screened out and current decision recommendations are generated, achieving highly explainable decision output under rule-driven control, significantly improving the company's response speed, decision-making accuracy and business adaptability in complex causal scenarios.

[0058] S104, obtain causal graph data and corresponding rule label information from different industries, determine structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determine migratable causal paths and migratable association rules based on the structural similarity and node semantic mapping rules.

[0059] Causal graph data refers to a collection of causal relationships represented in a graph structure, consisting of nodes and edges. Nodes represent business events, indicators, or variables within an industry (e.g., production processes, sales volume, customer complaints, etc.). Edges represent the causal influence relationships between these nodes (e.g., "Insufficient raw material supply leads to production delays"). Causal graph data is extracted and integrated from business process data, indicator data, and unstructured text data within an enterprise or industry, and exhibits structured causal logic characteristics.

[0060] The corresponding rule labels can be a collection of expert rules, policies, or constraints associated with nodes and paths in the causal graph. These rules typically describe the decision logic or response measures for specific business scenarios in an "if...then..." format (e.g., "If production delays exceed three days, initiate emergency procurement"). Rule labels can be used to identify which paths or nodes in the graph correspond to specific business decision rules, assisting the decision-making system in properly interpreting and applying causal paths.

[0061] Structural similarity can be used to measure the topological similarity between two causal graphs (or the paths within them). This includes comparisons of node connectivity, path lengths, and subgraph patterns. For example, the causal graphs of two industries both exhibit a similar structure of "supply chain disruptions leading to production bottlenecks." Even though the specific node names differ, the overall causal chain structure is similar. Structural similarity helps discover causal patterns that are comparable or transferable across industries.

[0062] Node semantic mapping rules can be a system of rules that map nodes (business events or variables) in one industry's causal graph to semantically corresponding or similar nodes in another industry's graph. Leveraging industry terminology libraries, synonym libraries, and semantic embedding models (such as BERT and Word2Vec), these rules determine the semantic equivalence or similarity of node labels in different contexts. This ensures accurate matching of business concepts that are expressed differently but are essentially the same or similar during cross-industry migration.

[0063] Transferable causal paths refer to causal chains identified in the source industry's causal graph that are similar or corresponding in structure and semantics to those in the target industry. The business causal relationships represented by these paths can be "transferred" to decision-making scenarios in another industry, helping to quickly construct or supplement the target industry's causal graph. For example, the path of "equipment failure → production stagnation → order delay" in manufacturing might be mapped to a similar path of "vehicle failure → shipping delay → customer complaint" in another industry (such as logistics).

[0064] Transferable association rules can be extracted from the source industry's expert rule base, corresponding to transferable causal paths and applicable to the target industry. After semantic mapping and format adjustment, these rules can be applied to the target industry's decision support system. For example, a rule for "equipment failure warning" in the manufacturing industry can be transferred and adjusted to become a rule for "vehicle failure warning" in the logistics industry.

[0065] To obtain causal graph data and corresponding rule label information from different industries, we first collect structured business event nodes, causal relationship edges, indicator variables, and their corresponding expert rules and decision strategies from enterprise information systems, data warehouses, and expert rule libraries across various industries through interface integration, data synchronization, or data crawling. The causal graph data includes the node topology, node labels, and causal direction and weight information of edges. The rule label information contains the rule trigger conditions, execution actions, and business semantic descriptions associated with the nodes or paths. For causal graphs and rule labels from different sources, we adopt a unified data cleaning and standardization process, including node label normalization, synonym replacement, noise filtering, and format conversion, to ensure data semantic consistency and format compatibility. Subsequently, the topological structures of different causal graphs are compared based on preset graph structure similarity algorithms (such as subgraph isomorphism detection and graph edit distance calculation) to determine their structural similarity indicators. At the same time, based on the node semantic embedding model (for example, using pre-trained language models such as BERT and Word2Vec), node labels and industry terms are vectorized and represented. By calculating the node semantic similarity, node semantic mapping rules are established. The mapping rules include synonym library support, multi-layer semantic matching, and context association discrimination to ensure the accuracy of cross-industry semantic correspondence. Finally, combining the structural similarity and node semantic mapping rules, paths with similar structures and corresponding node semantics in the source industry causal graph are screened as transferable causal paths, and rules associated with the path are extracted from the expert rule library of the source industry. After semantic mapping and format conversion, transferable association rules are formed to support causal reasoning and decision-making applications in the target industry. The above process uses a graph database (such as Neo4j) to store and manage causal graph data, uses machine learning models to automatically assist in structural and semantic similarity calculations, and improves the accuracy and practicality of mapping rules through expert manual review, ultimately achieving the effective migration and application of cross-industry causal graphs and rules.

[0066] On the basis of the above technical solution, optionally, determining the structural similarity and the node semantic mapping rules according to the causal graph data and the corresponding rule label information includes:

[0067] Determine the node topology and node label information based on the causal graph data;

[0068] Determine business variable terms and domain terms based on the corresponding rule label information, input the node label information, business variable terms and domain terms into the preset semantic embedding model, and generate node semantic similarity;

[0069] The graph structure similarity is determined according to the node topology structure, and the structure similarity and node semantic mapping rules are determined according to the graph structure similarity and the node semantic similarity.

[0070] In this solution, node topology refers to the connections, relative positions, and connection directions between nodes (i.e., business events, variables, etc.) in a causal graph. It describes the structural layout of each node in the causal chain, including information about parent-child relationships between nodes (i.e., causal directions); the number of inbound and outbound edges (indicating a node's causal or effect role); and path length and number of paths (e.g., the sequence of nodes traversed in a multi-hop path).

[0071] Node label information can refer to the semantic name or identifier carried by each node in the graph, which is used to indicate the business meaning corresponding to the node, such as "order delay", "inadequate inventory", and "customer churn".

[0072] Business variable terms can be professional variable names or causal condition terms extracted from expert rule labels, business documents, or system fields, and are used for rule modeling and matching. For example, "inventory turnover rate," "delivery cycle," and "customer satisfaction."

[0073] Domain terminology can refer to standardized expressions, term bases, or ontology terms used in a specific industry or professional field. For example, "liquidity risk" and "credit default" in the financial industry, and "process bottleneck" and "equipment utilization" in the manufacturing industry.

[0074] The preset semantic embedding model can refer to a model used to convert text terms (such as node labels, business variables, and domain terms) into vector representations. Common ones include: Word2Vec, GloVe (based on contextual co-occurrence), BERT, RoBERTa (based on deep language understanding), and dedicated knowledge graph embedding models such as TransE and ComplEx.

[0075] Node semantic similarity refers to the degree of semantic proximity between two nodes, calculated using methods such as cosine similarity after encoding their labels or terms using a semantic embedding model. For example, if the vectors for "Insufficient Inventory" and "Out-of-Stock Warning" are close, then their semantic similarity is high.

[0076] The graph structure similarity can refer to the similarity in topology between subgraphs or paths in two causal graphs in terms of connection mode, causal path depth, node number, etc. For example, if there is a three-layer structure of "raw material fluctuation → production delay → delivery exception" in both graphs, it is considered that they have structural similarity.

[0077] The node semantic mapping rule can refer to a set of matching rules for mapping node terms in the source graph to target graph terms based on node semantic similarity, business context, and domain knowledge. For example, if the similarity between "delivery delay" and "order delay" exceeds the threshold value 0.85, it is considered that the two can be mapped. Combined with the upstream and downstream causal path to confirm semantic consistency → improve mapping accuracy.

[0078] The basic structure and semantic information need to be extracted from the causal graph data of the source industry and the target industry. Specifically, the system determines the node topology structure according to the causal graph data by analyzing the directed edge set and the node set of the graph, that is, extracts the in-degree, out-degree, path level, causal connection direction, and substructure pattern (such as chain, star, ring structure, etc.) of each node, and records the structure characteristics by graph representation learning or graph traversal algorithm (such as DFS, BFS); at the same time, the identification name, label content or node metadata field of each node are extracted to determine the node label information, such as "inventory turnover rate", "customer churn" or "delivery delay", etc., as the semantic identification of the node. Subsequently, the system determines the business variable terms and domain terms according to the corresponding rule label information. Specifically, the pre-set rule label information contains the fields of "trigger condition", "business variable description", "domain term definition" and the like of the rule item, and the system automatically extracts the standard business terms (such as "KPI value below threshold", "complaint volume increase") and industry general terms (such as "customer satisfaction", "energy efficiency" and the like) by keyword extraction, part-of-speech tagging and named entity recognition technology (such as based on spaCy or BERT-NER model), which are classified into business variable terms and domain terms respectively. The node label information, business variable terms and domain terms extracted above are input into the pre-set semantic embedding model for vector coding. The model can use Word2Vec, GloVe, BERT, SBERT or cross-industry ontology knowledge enhanced embedding model to perform semantic representation learning on all terms and output high-dimensional embedding vectors. Based on these embedding vectors, the system calculates the node semantic similarity between each pair of source graph and target graph using cosine similarity and other measurement methods, and screens the candidate matching nodes in combination with the semantic threshold rule. Then, the system determines the graph structure similarity according to the node topology structure. By calculating the graph edit distance, minimum common subgraph, and graph structure pattern comparison, the similarity of the two causal graphs in local structure (such as multi-hop path shape, path depth, node connection pattern) is compared to generate a structure similarity score. Finally, the system integrates the graph structure similarity and the node semantic similarity to determine the structure similarity and the node semantic mapping rule.

[0079] The training process of the pre-set semantic embedding model is as follows:

[0080] First, a training corpus is constructed, including node label sets of causal graphs in multiple industries, business variable terms and domain term sets extracted from expert rule bases, and semantic enhanced texts from industry knowledge bases, business documents, standard term tables, policies and regulations, company reports, etc. To enhance the representation ability, the upstream and downstream field corpus (such as finance and insurance, manufacturing and logistics, etc.) can be combined, and the general language model corpus (such as Wikipedia or industry standard database) can be introduced for pre-training or fine-tuning. Next, word vector or sentence vector training methods are used for modeling. Common models include traditional word embedding algorithms such as Word2Vec (CBOW / Skip-Gram), GloVe, FastText, or context-aware pre-trained language models such as BERT, RoBERTa, ERNIE. In order to enhance the semantic distinction ability between industry terms, the above pre-trained models usually need to be fine-tuned in the industry: input the constructed sentence fragments, such as "equipment utilization rate rises leading to energy consumption decline", "credit delinquency rate rises leading to credit rating down" and the like, and use MLM (Masked Language Model) or NSP (Next Sentence Prediction) task for further training, so that the model can better capture the causal semantic relationship between terms. In order to enhance the semantic mapping ability, the graph structure comparison training strategy can be used: the nodes with similar structures but different semantics from different causal graphs are used as training pairs, and the contrast learning method (such as TripletLoss, SimCLR) is used to train the model to pull the node vectors with similar semantics closer and the node vectors with different semantics farther apart. For example, "order cancellation rate" (source industry) and "check-out rate" (target industry) are used as positive sample pairs; "inventory alarm rate" and "customer satisfaction" are used as negative sample pairs to enhance the model's discrimination ability. The semantic embedding model finally trained can map node label information, business variable terms and domain terms into semantic vectors of a unified dimension. In the use stage, the model inputs the above content to generate corresponding vector representations, and then calculates the cosine similarity, Euclidean distance, etc. to evaluate the semantic similarity between nodes, providing support for subsequent structural similarity matching, cross-industry path mapping and causal rule migration. The model can also be used in conjunction with knowledge graph embedding models (such as TransE, ComplEx, RotatE) to realize semantic and graph structure embedding fusion representation.

[0081] In this scheme, by fusing graph structure similarity and node semantic similarity, the structural similarity and semantic mapping rules between the causal graphs of the source industry and the target industry are accurately established, effectively improving the accuracy and automation of cross-industry causal path and rule migration, breaking through the industry barriers, and realizing efficient reuse and intelligent adaptation of causal knowledge.

[0082] On the basis of the above technical scheme, optionally, the migratable causal path and the migratable association rule are determined according to the structural similarity and the node semantic mapping rule, comprising:

[0083] According to the structural similarity, the causal path similar in structure in the source industry causal graph and the target industry causal graph is determined;

[0084] According to the node semantic mapping rule, the key nodes in the causal path are semantically converted to obtain mapping nodes consistent with the target industry terminology;

[0085] The mapping nodes of the causal path are determined, the association rules related to the causal path are extracted from a preset source industry expert rule library according to the causal path and the mapping nodes, and the migratable causal path and the migratable association rule are determined according to the association rules.

[0086] In the present scheme, the source industry causal graph can refer to a causal relationship graph constructed in an industry (such as manufacturing, finance) with business knowledge background, wherein the nodes represent variables, events or states, and the edges represent the causal relationship between variables (for example, “equipment aging”→“production efficiency decline”).

[0087] The target industry causal graph can refer to a causal graph constructed in another industry (such as retail, medical) with the characteristics of the industry, which has the same structure but different node semantics and business logic.

[0088] The causal path can refer to a directed path in the causal graph that starts from a business event or variable, passes through multiple intermediate nodes, and finally affects the target business result.

[0089] The key node can refer to a node in the causal path that has high industry representativeness, strong influence, and strong explainability in semantics. These nodes are usually the core trigger factors or final effect indicators of the decision logic, such as “high failure rate”, “order loss” and the like.

[0090] The target industry terminology can be a professional expression of variables or events in the target industry (such as medical, e-commerce, etc.).

[0091] The mapping node can be a new node formed by converting the key node (such as “inventory backlog”) in the source industry into a term (such as “drug oversupply”) that is semantically equivalent or similar in the target industry according to the node semantic mapping rule.

[0092] The preset source industry expert rule library can be a structured expert knowledge rule set constructed in advance for the source industry, usually in the form of “IF (condition) – THEN (suggestion)”.

[0093] An association rule may be a combination of conditions and suggestions in an expert rule associated with a causal path, which may be triggered by the current node state.

[0094] Based on the calculated structural similarity, causal paths with similar topological structures can be identified in both the source and target industry causal graphs. This process involves graph traversal and structural matching, comparing node topology (such as node degree centrality, path depth, and inbound and outbound edge patterns) with the connectivity between causal edges. Common methods include subgraph isomorphism matching, graph isomorphism embedding (Graph Matching Networks), or graph isomorphism algorithms based on Weisfeiler-Lehman trees. The system matches path structures in the source graph (e.g., "raw material delay → production disruption → delivery risk") with similarly structured paths in the target graph (e.g., "funding delay → construction stall → acceptance risk"), thereby identifying structurally transferable path pairs. Next, using trained node semantic mapping rules, the key nodes in these paths (i.e., core variables that play a leading role in causal propagation or influence the outcome node) are semantically transformed. Key nodes are selected based on causal edge weights, their centrality within the path, business impact, or expert rule priority. Using a semantic embedding model, the system calculates the semantic similarity between source industry node labels and target industry terminology. Semantically similar source node labels are replaced with semantically equivalent target industry terminology to generate corresponding mapping nodes (for example, mapping the "inventory overload" node to "material yard congestion"). Then, based on the obtained mapping nodes and their corresponding paths, the system searches the pre-set source industry expert rule library for association rules related to that path. These rules typically take the form of "If X occurs and Y meets condition Y, then the risk of Z will increase." Based on the degree of match between the business variables in the rules and the mapping nodes, the system identifies rule items with migration potential (for example, "When inventory overload reaches a threshold and the shipment rate drops by more than 10%, initiate emergency redeployment"). Finally, the matched source industry rule paths are mapped to the inference paths and rule forms in the target industry scenario, forming transferable causal paths and transferable association rules.

[0095] In this solution, by integrating graph structural similarity and node semantic similarity, dual alignment of causal graphs from different sources at the structural and semantic levels is achieved, effectively improving the ability to identify and match cross-industry causal relationships, solving the obstacles brought about by structural heterogeneity and terminology differences in knowledge transfer between traditional industries, and significantly enhancing the portability and automatic adaptation capabilities of causal paths and rules, providing a more general, efficient and explainable knowledge base for intelligent decision-making systems.

[0096] S105 , generating an enterprise decision execution plan and causal reasoning results based on the current association rules, the current decision suggestions, the transferable causal paths, and the transferable association rules.

[0097] An enterprise decision execution plan can be a specific business operation plan or action plan generated through comprehensive analysis of current association rules, current decision recommendations, transferable causal paths, and transferable association rules. It aims to guide enterprises on how to address issues in decision-making business data or optimize business processes, and includes clear execution steps, responsibility allocation, priority arrangements, and expected results. It can transform the results of causal reasoning into actionable decision instructions, supporting enterprises in achieving intelligent and scientific management and operations.

[0098] Causal reasoning results are derived by combining multi-hop causal paths and migration causal paths within an enterprise-level causal graph, and applying causal inference to decision-making business data based on current and migration association rules. This graph displays the causal chain and impact paths between business events, explaining the factors that led to the current business state. This helps explain the logical basis and causal relationships behind decision-making solutions, thereby enhancing the transparency and credibility of decisions.

[0099] First, these four types of data are uniformly structured and modeled, and a unified data representation format is used to map rule conditions, decision actions, path nodes and their weights to a unified data structure for easy subsequent processing; natural language processing and graph embedding technology are used to achieve semantic matching between rule conditions and causal path nodes, and the pre-trained industry semantic model is used to calculate the similarity between nodes and rule texts, and the business dictionary is combined to solve the problem of term heterogeneity to ensure accurate matching; then the current and migratable causal paths are weighted by causal weights and confidence, and the path priority is calculated through a weighted scoring model based on the rule triggering frequency and historical feedback data, and the key paths are screened; the current association rules are merged with the migratable association rules, and rule conflicts are detected and resolved. A priority mechanism or machine learning model is used for conflict adjustment to ensure that the rule set is consistent and reasonable, while supporting version management and expert intervention; based on the fused association rules and their matching paths, combined with the current decision recommendations, Use decision trees or decision graph models to automatically generate specific enterprise decision execution plans, which include execution actions, time arrangements, responsible departments, resource estimates and expected effects, and combine historical data and business constraints to ensure the feasibility of the plans. Build a decision causal chain diagram based on a multi-hop causal path chain, and use graph visualization tools to display causal relationships and impact paths. Automatically generate causal reasoning reports, detailing causal relationships, key factors and decision-making basis, to improve the transparency and trust of the plans. The system as a whole is based on a distributed graph database to store causal graphs and rule data, and uses a rule engine to implement rule matching and decision execution. Combined with NLP modules (such as spaCy and Transformers) for semantic processing, the decision generation module applies decision trees and optimization algorithms to complete plan optimization, and cooperates with an interactive interface to support expert adjustment and feedback, forming a closed-loop dynamic enterprise decision execution system to achieve the scientific integration and efficient implementation of current knowledge and transferred knowledge.

[0100] In this embodiment, enterprise business process data, operating indicator data and unstructured text data are obtained, and an enterprise-level causal graph is constructed according to the enterprise business process data and the operating indicator data; decision-making business data is obtained, path reasoning is performed in the enterprise-level causal graph according to the decision-making business data, and a multi-hop causal path chain is determined; a current association rule and a current decision-making suggestion are determined according to the multi-hop causal path chain and a preset expert rule base; causal graph data and corresponding rule label information from different industries are obtained, structural similarity and node semantic mapping rules are determined according to the causal graph data and the corresponding rule label information, and a migratable causal path and a migratable association rule are determined according to the structural similarity and the node semantic mapping rules; and an enterprise decision-making execution scheme and a causal reasoning result are generated according to the current association rule, the current decision-making suggestion, the migratable causal path and the migratable association rule. Through the above-mentioned enterprise intelligent decision-making method driven by the causal graph, the enterprise-level causal graph and the expert rule base are fused, precise decision-making reasoning is realized in combination with the multi-hop causal path chain, the causal path migration and the association rule migration mechanism across industries are introduced, and the generalization ability and adaptability of the decision-making model are improved. Based on the comprehensive application of current and migrated knowledge, the scientificity, accuracy and explainability of decision-making are effectively enhanced, dynamic updating and inter-industry knowledge sharing are supported, and the enterprise decision-making efficiency and response speed are significantly improved, helping enterprises to realize intelligent and fine management.

[0101] On the basis of the above technical solutions, according to the current association rule, the current decision-making suggestion, the migratable causal path and the migratable association rule, an enterprise decision-making execution scheme and a causal reasoning result are generated, which includes:

[0102] According to the current association rule and the current decision-making suggestion, an initial decision-making execution scheme corresponding to the current business problem is generated;

[0103] According to the migratable causal path and the migratable association rule, the initial decision-making execution scheme is expanded to generate an enterprise decision-making execution scheme;

[0104] The nodes and rule relationships involved in the migratable causal path are determined, and a causal reasoning result is generated according to the current association rule, the nodes and rule relationships involved in the migratable causal path.

[0105] In this scheme, the current business problem can refer to a specific business scenario or task that the enterprise currently faces and needs to respond to by making a decision, which is usually expressed through decision-making business data. These problems may involve operational bottlenecks, risk events, resource allocation, performance anomalies and other actual problems, such as "customer order cancellation rate continues to rise", "project delivery delay", "abnormally high energy consumption" and the like. It is the starting point of triggering decision-making reasoning and generating a scheme.

[0106] The initial decision execution scheme can refer to the first step operation or coping strategy set generated by the system based on the current association rule and the current decision suggestion, which directly reflects the standard response strategy of the enterprise under the current business problem. For example, if "order processing delay → customer satisfaction decline" is the reasoning path in the current causal chain, the initial scheme may be "optimize order approval process" or "temporarily increase customer service response capacity".

[0107] The enterprise decision execution scheme can be generated on the basis of the initial scheme, and the system further integrates the transferable causal path and the transferable association rule to generate a complete multi-strategy combination scheme after expanding, supplementing or optimizing the strategy. It not only contains direct measures to cope with the current business problem, but also combines the successful processing path of other industries to achieve cross-industry knowledge reinforcement. For example, the rule "fast refund → customer retention improvement" found in the financial industry is transferred to the "delayed processing of returns and exchanges" problem in the e-commerce industry as a supplementary strategy.

[0108] The nodes involved in the transferable causal path can refer to the key variables or event nodes reserved for reasoning in the structurally similar path selected from the source industry causal graph and migrated to the target industry graph through semantic mapping. These nodes represent elements with corresponding meanings in different businesses and can be operation events (such as "equipment maintenance", "order confirmation"), state indicators (such as "high inventory", "customer churn rate rise"), and abnormal trigger points (such as "abnormal fluctuations", "process delay").

[0109] The rule relationship can refer to the logical rules or conditional patterns associated with the causal connection between the above-mentioned nodes in the source industry expert rule base, usually represented in the form of "if A and B, then C". These rules define the behavior constraints or trigger logic between nodes, for example, if "supplier delay" and "insufficient safety stock", then "production line shutdown"; if "page load time > 5s", then "user bounce rate rises".

[0110] According to the identified current association rule and the current decision suggestion generated by the system, a preliminary executable strategy set can be generated in combination with the current business problem in the business context to form an initial decision execution scheme. This process includes analyzing the main problem nodes and trigger conditions reflected by the current causal chain, extracting the execution instructions under the "IF-THEN" structure in the association rule (for example: "if inventory backlog > 30%, then enable the promotion mechanism"), and generating structured operation instructions or process suggestions according to business execution parameters to ensure the operability and business adaptability of the scheme. Subsequently, the system introduces the migratable causal paths and corresponding migratable association rules that have been migrated from other industries to extend and strengthen the initial decision scheme and form a complete enterprise decision execution scheme. During the extension process, the system identifies migratable paths and strategies that are highly similar in structure or semantics to the current business problem, such as migrating the response scheme "inventory backlog → turnover rate decline → cost increase" in the manufacturing industry to the "product backlog → slow warehouse turnover → profit decline" problem scenario in the e-commerce industry, thereby introducing other industries' efficient response mechanisms for similar problems, such as adjusting the ordering strategy, optimizing the clearance channel, and activating the promotion system. The system integrates these strategies with the original initial scheme to form a strategy set optimized through cross-domain integration. In order to enhance the explainability and reasoning transparency of the scheme, the system further determines the nodes involved in the migratable causal path and the rule relationships, including the core variable nodes in the path, the causal trigger direction and threshold constraints between variables, and the premise conditions and post-decision rules bound in the rules. For example, nodes such as "production cycle extension" and "resource occupancy rate increase" and causal rules such as "if production delay is more than 3 days and equipment utilization rate > 85%, then warn about project scheduling" will be extracted as reasoning basis. Finally, based on the above current association rule, mapping nodes, and rule relationships between them, the system performs path backtracking and forward reasoning operations through the causal chain explanation engine to build a clear causal reasoning result, including the variable trend in the causal path, the key causal chain, and the strategy output chain triggered by it, thereby providing a verifiable and explainable intelligent decision suggestion chain for enterprise managers to understand, approve, and execute.

[0111] In this scheme, by integrating current association rules and cross-industry migratable knowledge, a more comprehensive enterprise decision execution scheme is constructed, and clear and explainable causal reasoning results are generated based on the nodes and rule relationships in the causal path, which not only improves the intelligence and reusability of the decision, but also enhances the operability and transparency of the scheme, helping enterprises to quickly respond to complex business problems and reduce decision risk.

[0112] Embodiment Two

[0113] Figure 2 is a flowchart of the causal graph-driven enterprise intelligent decision-making method provided by Embodiment Two of the present application. As Figure 2As shown, the specific steps include:

[0114] S201, obtaining enterprise business process data, operating indicator data and unstructured text data, and constructing an enterprise-level cause-and-effect graph based on the enterprise business process data and operating indicator data.

[0115] S202 , obtaining business data to be decided, performing path reasoning in the enterprise-level causal graph based on the business data to be decided, and determining a multi-hop causal path chain.

[0116] S203: Determine the current association rules and the current decision suggestion according to the multi-hop causal path chain and the preset expert rule library.

[0117] S204, obtain causal graph data and corresponding rule label information from different industries, determine structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determine migratable causal paths and migratable association rules based on the structural similarity and node semantic mapping rules.

[0118] S205 , generating an enterprise decision execution plan and causal reasoning results based on the current association rules, the current decision suggestions, the transferable causal paths, and the transferable association rules.

[0119] S206, obtaining task completion status data, key performance indicators, and abnormal record data during the execution process after the implementation of the enterprise decision execution plan.

[0120] Task completion status data refers to the execution results and progress of each business task node after the company executes its decision-making plan. This data is used to evaluate the effectiveness of the decision-making plan. This includes the start and completion time of each task (e.g., "supply chain restructuring task start time and completion time"); task completion percentage (e.g., "sales order completion rate 92%"); deviations between actual execution methods and the intended plan; and execution status labels (e.g., "completed," "in progress," "failed," "interrupted").

[0121] Key performance indicators (KPIs) refer to core quantitative metrics used to measure the effectiveness of decision execution, reflecting the quality of business operations and the achievement of management objectives. They are often customized based on the industry and task type. For example, financial KPIs include "operating revenue growth rate" and "net profit margin"; operational KPIs include "inventory turnover rate" and "order fulfillment cycle time"; customer KPIs include "customer satisfaction rate" and "complaint rate"; and risk KPIs include "bad debt ratio of accounts receivable" and "quality compliance rate."

[0122] The abnormal record data can refer to abnormal behaviors, unexpected events, system alarms or process deviations recorded by the system or personnel during the execution of the decision-making scheme, which are used for subsequent cause-effect chain verification and decision optimization. The abnormal record data includes system abnormalities such as system downtime, interface failure, data synchronization interruption; operation abnormalities such as tasks being skipped, process node execution sequence disorder; resource abnormalities such as raw material shortage, key post personnel absence; external interference such as policy change, supply chain interruption and other uncontrollable events.

[0123] By integrating the information systems (such as ERP system, MES system, WMS system, etc.) within the enterprise, the execution records corresponding to each task in the enterprise decision-making execution scheme are automatically collected, the task start time, actual completion time, execution personnel, operation procedure code and completion status label (such as “success”, “failure”, “delay” and the like) of each business task are extracted, and the standardized task completion status data is formed. Secondly, the key performance indicators associated with the current business scenario are called from the enterprise data warehouse or BI platform, such as “inventory turnover rate”, “purchase cycle shortening ratio” and “stockout rate” for supply chain optimization tasks, and the historical KPI baseline is combined to automatically compare and calculate the KPI fluctuation trend, so as to measure the execution effect. Then, by connecting the enterprise operation and maintenance platform, log audit system, event monitoring system and manual abnormal reporting system, the abnormal event records related to the decision implementation process are continuously collected, such as task interruption, interface timeout, data error, material delay, system downtime, etc., and are uniformly stored as structured abnormal record data, with time of occurrence, influence range, triggering condition and the like. In order to realize the consistency and traceability of data, all data fields need to be standardized modeled before collection, a unified data dictionary and business label are defined, a data synchronization strategy is established, a real-time streaming collection mechanism is supported, and a task ID or process ID is used as an associated primary key to realize the logical linkage and time alignment of various data, thereby providing a high-quality feedback basis for subsequent links such as cause-effect chain verification, model retraining and expert rule correction.

[0124] S207, according to the task completion status data, key performance indicators and abnormal record data in the execution process, a feedback evaluation index set of the enterprise decision execution scheme is determined.

[0125] The feedback evaluation metric set can refer to a set of multi-dimensional, quantifiable indicators derived from task completion status data, key performance indicators, and exception records during the implementation process, based on the actual operational results of the enterprise's decision-making execution plan. These indicators are used to comprehensively evaluate the implementation effectiveness of the plan, business achievements, and potential issues. Specifically, these indicators include the following types: Execution Completion Rate Indicators (derived from task completion status data): Task Completion Rate (number of completed tasks / total number of tasks); On-Time Completion Rate (percentage of tasks completed as planned); Average Duration of Delayed Tasks; Performance Improvement Indicators (derived from Key Performance Indicators): KPI Improvement (e.g., inventory turnover increase, cost reduction, satisfaction improvement score); Deviation from historical or expected targets; Multi-Indicator Composite Score (weighted or aggregated). Abnormal Performance Indicators (derived from abnormal performance data): Abnormal Event Frequency; Abnormal Event Impact Level (e.g., impact scope or resource utilization); Mean Time to Repair and Response Time.

[0126] First, data sources from different systems can be standardized and pre-processed, including unified field formats, timestamp alignment, and task identifier binding to ensure data relevance. Next, the system retrieves the task completion status data for each task based on the task set defined in each enterprise decision execution plan, analyzes each task's "completeness," "completion time," and "deviation rate from the scheduled duration," and calculates indicators such as task completion rate, on-time completion rate, and average task delay time to measure the implementation of the plan. At the same time, the system connects to the enterprise's data warehouse or BI analysis platform to extract key performance indicators before and after the plan is implemented, and builds a time series comparison model based on the business type, such as like-for-like analysis, month-on-month analysis, and trend fitting, to extract performance indicators such as the degree of improvement in indicators (such as "customer satisfaction increased by 12%") and target deviation values ​​(such as "the achievement rate of the operating cost reduction target was 86%"). For event-level feedback, the system aggregates exception records from the execution process, categorizing and counting exception types (system, process, and external), frequency, and severity. Combining task IDs and timestamps, the system calculates impact indicators such as the exception trigger rate, average repair time, and the number of key node outages. Ultimately, the system structures and aggregates these multi-dimensional indicators into a unified set of feedback evaluation metrics, which are then tied to the corresponding enterprise decision execution plan ID.

[0127] In this embodiment, by obtaining multi-dimensional feedback data after the implementation of the enterprise decision-making execution plan and constructing a comprehensive set of feedback evaluation indicators, it is possible to effectively quantify the implementation effect of the plan, discover potential problems and support result traceability, enhance the closed-loop management and continuous optimization capabilities of enterprise decision-making, and improve the accuracy and credibility of the intelligent decision-making system.

[0128] Based on the above technical solution, optionally, after determining the feedback evaluation indicator set of the enterprise decision execution plan, the method further includes:

[0129] According to the feedback evaluation indicator set and the preset reinforcement learning strategy, the triggering conditions of the association rules related to the enterprise decision-making plan are updated, and the recommendation strength of the current decision recommendation is dynamically adjusted.

[0130] In this solution, the pre-set reinforcement learning strategy refers to a reinforcement learning algorithm and policy model that has been pre-designed and trained within the system. This model continuously adjusts its decision-making rules based on environmental feedback (e.g., a set of feedback evaluation metrics) to optimize decision-making outcomes. The strategy evaluates the effectiveness of the execution plan through a reward function and dynamically adjusts the policy parameters based on this feedback, achieving intelligent adaptive optimization.

[0131] Association rules can be defined as a set of rules used in enterprise decision-making systems to describe the causal or logical relationships between business events or states. Association rules define the decision-making actions or responses that should be taken under specific conditions (trigger conditions), supporting automated decision execution.

[0132] A trigger condition refers to the prerequisite or judgment criteria specified in an association rule. When business data or environmental status meets the condition, the association rule is activated, triggering the corresponding decision recommendation or action. Trigger conditions are usually based on node status, indicator thresholds, or event occurrences.

[0133] Recommendation strength is a quantitative representation of the system's importance or priority for the current decision recommendation, reflecting the weight or influence of the recommendation in decision execution. Recommendation strength can be dynamically adjusted based on reinforcement learning strategy feedback, increasing or decreasing the priority of a particular recommendation and improving the flexibility and adaptability of decision-making solutions.

[0134] By integrating enterprise business systems (such as ERP, MES, and CRM) with time-series databases, task completion status data can be automatically collected to reflect the completion status of specific business tasks. Furthermore, data warehouses and data analytics techniques can be used to obtain key performance indicators (KPIs), including sales, customer satisfaction, and inventory turnover. Trends and fluctuations can be extracted through statistical analysis and time-series decomposition methods. Log parsing and anomaly detection algorithms (such as rule-based anomaly recognition or machine learning anomaly detection) can be combined to capture abnormal events and their impact during execution in real time. Based on the collected feedback evaluation metrics, a reinforcement learning model is constructed using a deep reinforcement learning algorithm (such as a deep Q-network (DQN), policy gradient, or actor-critic). The input is a multidimensional business state vector that has been normalized and feature-reduced to ensure that the environment fully reflects the current business execution feedback. The reinforcement learning action space is designed to dynamically adjust the triggering conditions and recommendation strength of association rules, specifically through continuous fine-tuning of threshold parameters and weight addition and subtraction. Trigger conditions are typically expressed as numerical thresholds or logical inferences. Automatic adjustment of threshold ranges allows for flexible changes in rule activation conditions. Recommendation strength is quantified as rule priority or weight coefficients. The reinforcement learning model optimizes action selection to enhance the execution and accuracy of recommendations. Based on a designed reward function, the system comprehensively considers indicators such as improved task completion, positive changes in KPIs, and a reduction in abnormal events to calculate an immediate reward value, which serves as feedback for the reinforcement learning model. The reward function typically takes the form of a weighted sum to ensure multi-dimensional business objectives are addressed. During reinforcement learning training, an experience replay mechanism and target network technology are employed to improve training stability and convergence speed. Parameter updates utilize gradient descent combined with momentum or soft update techniques to avoid excessive fluctuations. Rule adjustment history is logged for easy subsequent analysis and manual review.

[0135] In this solution, by combining real-time business data with reinforcement learning technology, dynamic optimization of association rule triggering conditions and recommendation strength is achieved, which improves the accuracy and adaptability of decision-making solutions, enables timely response to business changes, reduces manual intervention, and improves enterprise operational efficiency and decision-making effectiveness.

[0136] Example 3

[0137] Figure 3 This is a schematic diagram of the structure of the enterprise intelligent decision-making system driven by the cause-effect graph provided in the third embodiment of this application. Figure 3 As shown, specifically including:

[0138] A cause-effect graph construction module 301 is configured to obtain enterprise business process data, operating indicator data, and unstructured text data, and to construct an enterprise-level cause-effect graph based on the enterprise business process data and operating indicator data;

[0139] A path reasoning module 302 is configured to obtain business data to be decided, perform path reasoning in the enterprise-level causal graph based on the business data to be decided, and determine a multi-hop causal path chain;

[0140] A rule and suggestion determination module 303 is used to determine the current association rule and the current decision suggestion based on the multi-hop causal path chain and the preset expert rule library;

[0141] A cross-industry migration mapping module 304 is configured to obtain causal graph data and corresponding rule label information from different industries, determine structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determine transferable causal paths and transferable association rules based on the structural similarity and node semantic mapping rules;

[0142] The decision solution generation module 305 is used to generate an enterprise decision execution solution and causal reasoning results based on the current association rules, the current decision suggestions, the transferable causal paths and the transferable association rules.

[0143] The enterprise intelligent decision-making system driven by the causal graph provided in the embodiment of the present application can achieve Figure 1 To avoid repetition, the various processes implemented in the method embodiment are not described here.

[0144] Example 4

[0145] like Figure 4 As shown, an embodiment of the present application also provides an electronic device 400, including a processor 401, a memory 402, and a program or instruction stored in the memory 402 and executable on the processor 401. When the program or instruction is executed by the processor 401, each process of the above-mentioned causal graph-driven enterprise intelligent decision-making method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0146] It should be noted that the electronic devices in the embodiments of the present application include the mobile electronic devices and non-mobile electronic devices mentioned above.

[0147] Example 5

[0148] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, each process of the above-mentioned cable installation process based on the tension adaptive control system embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0149] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0150] It should be noted that, in this document, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or system that includes the element. In addition, it should be noted that the scope of the methods and systems of the present embodiments are not limited to performing functions in the order recited in the specification or as may be illustrated in the accompanying drawings. The methods and systems can include additional or fewer steps, or can be performed in a different order than as illustrated or discussed herein. Furthermore, features described with respect to certain examples can be combined in other examples.

[0151] From the above description of the embodiments, it is clear that the above-mentioned embodiment methods can be realized by means of software and the necessary general hardware platform, of course, they can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a plurality of instructions for making a terminal (which can be a mobile phone, computer, server, or network equipment, etc.) execute the methods described in the various embodiments of the present application.

[0152] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above specific embodiments, and the above specific embodiments are only illustrative, not restrictive. Those skilled in the art can make many forms under the inspiration of the present application without departing from the scope of the present application and the protection scope of the claims, and all of them belong to the protection scope of the present application.

[0153] The above are only preferred embodiments of the present application and the technical principles employed. The present application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that are possible for those skilled in the art will not depart from the scope of protection of the present application. Therefore, although the present application has been described in more detail through the above embodiments, the present application is not limited to the above embodiments and may include more other equivalent embodiments without departing from the concept of the present application. The scope of the present application is determined by the scope of the claims.

Claims

1. A causal graph-driven enterprise intelligent decision-making method, characterized by: The method comprises: Acquire enterprise business process data, operating indicator data, and unstructured text data, and construct an enterprise-level cause-and-effect graph based on the enterprise business process data and operating indicator data; Obtaining business data to be decided, performing path reasoning in the enterprise-level causal graph based on the business data to be decided, and determining a multi-hop causal path chain; Determining current association rules and current decision recommendations based on the multi-hop causal path chain and a preset expert rule base; Obtaining causal graph data and corresponding rule label information from different industries, determining structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determining transferable causal paths and transferable association rules based on the structural similarity and node semantic mapping rules; wherein determining structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information includes: Determine the node topology and node label information based on the causal graph data; Determine business variable terms and domain terms based on the corresponding rule label information, input the node label information, business variable terms and domain terms into the preset semantic embedding model, and generate node semantic similarity; Determining graph structure similarity based on the node topology structure, and determining structure similarity and node semantic mapping rules based on the graph structure similarity and node semantic similarity; Based on structural similarity and node semantic mapping rules, transferable causal paths and transferable association rules are determined, including: Based on the structural similarity, determining causal paths in the source industry causal graph and the target industry causal graph that have similar structures; According to the node semantic mapping rules, semantic conversion is performed on the key nodes in the causal path to obtain mapping nodes consistent with the target industry terminology; Determine a mapping node of a causal path, extract association rules related to the causal path from a preset source industry expert rule library based on the causal path and the mapping node, and determine a transferable causal path and transferable association rules based on the association rules; Generate enterprise decision execution plans and causal reasoning results based on current association rules, current decision recommendations, transferable causal paths, and transferable association rules.

2. The method according to claim 1, characterized in that in, Determining the current association rules and current decision suggestions based on the multi-hop causal path chain and the preset expert rule base includes: Obtaining business events corresponding to each node in the multi-hop causal path chain, and determining matching rule items from a preset expert rule library based on the business events; Determine the triggering conditions of the rule items and the current node states of the multi-hop causal path chain, match the triggering conditions of the rule items and the current node states of the multi-hop causal path chain, and determine the current association rule that meets the triggering conditions; A current decision suggestion for the target business is generated according to the current association rule.

3. The method according to claim 1, characterized in that in, Generate enterprise decision execution plans and causal reasoning results based on current association rules, current decision recommendations, transferable causal paths, and transferable association rules, including: Generate an initial decision execution plan corresponding to the current business problem based on the current association rules and current decision suggestions; Expanding the initial decision execution plan according to the transferable causal path and the transferable association rules to generate an enterprise decision execution plan; Determine the nodes and rule relationships involved in the transferable causal path, and generate causal reasoning results based on the current association rules, the nodes and rule relationships involved in the transferable causal path.

4. The method according to claim 1, wherein in, After generating an enterprise decision execution plan and a causal reasoning result based on the current association rules, the current decision suggestion, the transferable causal path, and the transferable association rules, the method further includes: Obtain task completion status data, key performance indicators, and exception record data during the implementation of the enterprise decision-making execution plan; A set of feedback evaluation indicators for the enterprise decision-making execution plan is determined based on the task completion status data, key performance indicators, and abnormal record data during the execution process.

5. The method according to claim 4, characterized in that in, After determining the feedback evaluation indicator set of the enterprise decision execution plan, the method further includes: According to the feedback evaluation indicator set and the preset reinforcement learning strategy, the triggering conditions of the association rules related to the enterprise decision-making plan are updated, and the recommendation strength of the current decision recommendation is dynamically adjusted.

6. A causal graph-driven enterprise intelligent decision-making system, characterized by: The system comprises: A cause-effect graph construction module is used to obtain enterprise business process data, operating indicator data, and unstructured text data, and to construct an enterprise-level cause-effect graph based on the enterprise business process data and operating indicator data; A path reasoning module is used to obtain business data to be decided, perform path reasoning in the enterprise-level causal graph based on the business data to be decided, and determine a multi-hop causal path chain; A rule and suggestion determination module, configured to determine the current association rule and the current decision suggestion based on the multi-hop causal path chain and a preset expert rule library; The cross-industry migration mapping module is used to obtain causal graph data and corresponding rule label information from different industries, determine structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information, and determine transferable causal paths and transferable association rules based on the structural similarity and node semantic mapping rules; wherein, determining structural similarity and node semantic mapping rules based on the causal graph data and corresponding rule label information includes: Determine the node topology and node label information based on the causal graph data; Determine business variable terms and domain terms based on the corresponding rule label information, input the node label information, business variable terms and domain terms into the preset semantic embedding model, and generate node semantic similarity; Determining graph structure similarity based on the node topology structure, and determining structure similarity and node semantic mapping rules based on the graph structure similarity and node semantic similarity; Based on structural similarity and node semantic mapping rules, transferable causal paths and transferable association rules are determined, including: Based on the structural similarity, determining causal paths in the source industry causal graph and the target industry causal graph that have similar structures; According to the node semantic mapping rules, semantic conversion is performed on the key nodes in the causal path to obtain mapping nodes consistent with the target industry terminology; Determine a mapping node of a causal path, extract association rules related to the causal path from a preset source industry expert rule library based on the causal path and the mapping node, and determine a transferable causal path and transferable association rules based on the association rules; The decision-making plan generation module is used to generate enterprise decision execution plans and causal reasoning results based on current association rules, current decision suggestions, transferable causal paths and transferable association rules.

7. An electronic device, characterized in that: It includes a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the enterprise intelligent decision-making method driven by a causal graph as described in any one of claims 1 to 5 are implemented.

8. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the enterprise intelligent decision-making method driven by the causal graph as described in any one of claims 1 to 5 are implemented.

Images