Marginal information identification and decision support method and system based on management line

Abstract

The invention provides a marginal information identification and decision support method and system based on a management line, and relates to the field of enterprise management digitization, and the method comprises the following steps: obtaining multi-source business data and flow data generated in an enterprise management process, and combining with the management line comprising a longitudinal management path and a transverse collaborative network, performing association and mapping on the multi-source business data and the process data to obtain an association and mapping result; carrying out key node identification on the multi-source business data and the process data based on the management line in combination with the association and mapping result, and determining a target node influencing a management decision; identifying marginal information which changes relative to the historical state; performing management influence evaluation on the identified marginal information, and judging the influence degree of the marginal information on the current management decision; and generating corresponding decision support information and outputting the decision support information to the management decision interface. The method improves the precision and efficiency of management decision, adapts to the dynamic business scene of an enterprise, and continuously optimizes the decision support effect.

Description

Technical Field

[0001] This invention relates to the field of digital enterprise management, and more specifically, to a method and system for identifying and supporting decision-making based on management line marginal information. Background Technology

[0002] In modern enterprise management, with the advancement of digital transformation, enterprises generate massive amounts of multi-source business and process data in various aspects such as procurement, production, sales, and control. This data forms the core foundation for supporting enterprise management decisions. Currently, widely used enterprise management systems such as OA, ERP, and BI mainly perform data recording, statistical summarization, and basic report output functions. They can only meet the basic needs of information retention and post-event review in daily management and have significant technical shortcomings in supporting refined and forward-looking management decisions, making it difficult to adapt to the control needs of enterprises as they grow in scale.

[0003] The existing enterprise management system mainly solves the problem of information recording and statistics, and has the following shortcomings: (1) Management data is scattered and it is difficult to focus on key management nodes, resulting in a blurred correspondence between data and management nodes. Massive amounts of data become ineffective and redundant, and it is difficult to transform them into effective basis for supporting node control and decision-making; (2) Report results are hindsight biased and cannot identify risks in advance. It is difficult to issue early warnings at the risk incubation stage, which means that managers can only passively deal with problems that have occurred and cannot achieve forward-looking decision-making; (3) Managers need to manually judge which information really affects decision-making. There is a lack of quantitative marginal effect analysis and scientific screening criteria, which not only makes decision-making inefficient, but also easily leads to control errors due to subjective bias and makes it impossible to quickly lock in key information that has a substantial impact on decision-making.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for identifying and supporting decision-making based on management lines, in order to solve the problems mentioned above.

[0006] To solve the above problems, the specific technical solution adopted by the present invention is as follows: According to one aspect of the present invention, a method for identifying and supporting decision-making based on management lines is provided, comprising the following steps: S1. Acquire multi-source business data and process data generated during enterprise management, and combine them with management lines that include vertical management paths and horizontal collaborative networks to associate and map the multi-source business data and process data, and obtain the association and mapping results. S2. Based on the management line and combined with the correlation and mapping results, identify key nodes in multi-source business data and process data to determine the target nodes that affect management decisions. S3. At the target node, perform marginal change analysis on real-time multi-source business data and process data related to node status changes, and identify marginal information that has changed relative to historical status. S4. Conduct a management impact assessment on the identified marginal information to determine its degree of impact on current management decisions; S5. When the degree of impact meets the preset decision triggering conditions, generate corresponding decision support information and output it to the management decision interface.

[0007] Preferably, the step of acquiring multi-source business data and process data generated during enterprise management, and combining them with management lines that include vertical management paths and horizontal collaborative networks, to correlate and map the multi-source business data and process data to obtain the correlation and mapping results includes the following steps: S11. Based on the company's organizational structure and business processes, identify the company's vertical management path and horizontal collaboration network, and generate a management line to characterize the company's management decision-making and execution process based on the vertical management path and horizontal collaboration network. S12. Collect multi-source business data and process data generated during enterprise management, and obtain standardized multi-source business data and process data by cleaning and standardizing the multi-source business data and process data. S13. Based on the multi-source business data and process data of the management line and the standard, determine the dynamic correlation and mapping results between the multi-source business data and process data and the management line nodes by analyzing the causal chain between nodes.

[0008] Preferably, the step of identifying the enterprise's vertical management path and horizontal collaboration network based on the enterprise's organizational structure and business processes, and generating a management line to characterize the enterprise's management decision-making and execution network based on the vertical management path and horizontal collaboration network, includes the following steps: S111. Obtain the company's organizational structure and identify the company's decision-making, management, and execution levels to obtain the identification results; S112. Based on the identification results, determine the decision nodes and execution nodes in the vertical management path to form the vertical management path; S113. Obtain the enterprise's business processes and identify collaborative nodes and rules by analyzing the interaction and collaboration relationships between various functional departments; S114. Based on collaborative nodes and collaborative rules, nodes scattered across different functional departments are connected according to their business coupling and information dependency relationships to form a horizontal collaborative network. S115. The vertical management path and the horizontal collaborative network are structurally integrated to obtain the management line used to represent the enterprise's management decision-making and execution process.

[0009] Preferably, the step of determining the dynamic correlation and mapping results between the multi-source business data and process data and the management line nodes by analyzing the causal chains between nodes, based on the multi-source business data and process data of the management line and the standard, includes the following steps: S131. Classify and align standard multi-source business data and process data according to the attributes of management line nodes to form an initial data-node association pool; S132. Using a pre-set expert experience base, filter out invalid data that is irrelevant to the management line node control requirements in the initial data-node association pool to form a node association candidate dataset. S133. Perform structured transformation and feature extraction on the candidate dataset associated with nodes, and construct a causal chain by identifying the causal relationship rules between business data patterns and node state changes; S134. Based on the causal chain and real-time data stream, construct and dynamically update the mapping relationship library between business data patterns and node states, and output the dynamic association mapping results.

[0010] Preferably, the step of performing structured transformation and feature extraction on the node-associated candidate dataset, and constructing a causal chain by identifying the causal association rules between business data patterns and node state changes, includes the following steps: S1331. Transform the business events and process states in the node-associated candidate dataset into standardized transactions with a preset time window as the unit, and extract features related to node decisions for each standardized transaction to form a structured transaction set. S1332. Based on the structured transaction set, calculate the support and attention indicators for the combination of business data pattern-node state change, and when the preset constraints are met, include the combination of business data pattern-node state change in the condition frequent itemset. S1333. Construct a causal chain with causal direction based on conditional frequent itemsets.

[0011] Preferably, constructing a causal chain with causal direction based on conditional frequent itemsets includes the following steps: S13331. Based on conditional frequent itemsets, generate candidate association rules with business data patterns as antecedents and node state changes as consequents. S13332. Calculate the confidence level of each candidate association rule and filter candidate association rules with a confidence level greater than the preset minimum confidence level threshold. S13333. Perform time series analysis on the filtered candidate association rules and verify whether the occurrence time of the antecedent event is earlier than that of the consequent event. If so, the candidate association rule is determined as a causal chain with causal direction; otherwise, the candidate association rule is excluded.

[0012] Preferably, the process of identifying key nodes in multi-source business data and process data based on management lines and combining association and mapping results to determine target nodes affecting management decisions includes the following steps: S21. Based on the association and mapping results, obtain the marginal indicator data of each node in the management line, and construct the time series sequence of marginal information of each node in the management line according to the marginal indicator data. S22. Based on the time series of marginal information of each management line node, calculate the marginal effect of each management line node, identify key decision points, and generate a marginal effect analysis report. S23. Based on the marginal effect analysis report, and combined with the preset decision impact threshold and target node screening rules, identify the target nodes that have an impact on the current management decision from the key decision points.

[0013] Preferably, the step of calculating the marginal effect of each management line node based on the time series sequence of marginal information of each management line node, identifying key decision points, and forming a marginal effect analysis report includes the following steps: S221. Based on the time sequence of marginal information of each management line node, and combined with the hierarchical attributes of the management line nodes, the management line nodes are divided into intervention group nodes and control group nodes. S222. Examine whether the intervention group nodes meet the preset parallel trend assumption within the time window before the marginal change occurs, and adjust the intervention group nodes that do not meet the parallel trend assumption until the preset parallel trend assumption is met. S223. For intervention group nodes that satisfy the parallel trend hypothesis, construct a counterfactual time series sequence in which the intervention group nodes do not undergo marginal changes by combining them with the control group nodes that match them. S224. Based on the counterfactual time series where no marginal change occurred in the intervention group nodes, extract the actual observation results of the intervention group nodes after the marginal change occurred, calculate the difference between the actual observation results and the corresponding time point values ​​of the counterfactual time series, and obtain the instantaneous value of the marginal effect at each time point. S225. Calculate the average treatment effect based on the instantaneous value of the marginal effect, and use it as the initial marginal effect estimate of the intervention group nodes; then, combine the support and confidence of the association rule to correct the initial marginal effect estimate and obtain the final marginal effect estimate. S226. Calculate the statistical significance of the final marginal effect estimate, and screen out the intervention group nodes that meet the preset judgment criteria. Determine the screened intervention group nodes as key decision points, and generate a marginal effect analysis report that includes the magnitude and direction of the marginal effect through integration.

[0014] Preferably, the step of dividing the management line nodes into intervention group nodes and control group nodes based on the time sequence of marginal information of each management line node and in combination with the hierarchical attributes of the management line nodes includes the following steps: S2211. Based on the vertical management path and horizontal collaborative network of the management line, extract the hierarchical attributes of each management line node; S2212. Based on the time series of marginal information of each node, identify the marginal change trend of each node within a preset time window, and detect whether there are marginal change events that meet the preset criteria. S2213. The management line nodes that detect marginal change events are marked as intervention candidate nodes, and the remaining management line nodes that do not detect significant changes are used as control nodes.

[0015] According to another aspect of the present invention, a marginal information identification and decision support system based on management lines is provided, the system comprising: The association mapping module is used to acquire multi-source business data and process data generated during enterprise management, and combine them with management lines that include vertical management paths and horizontal collaborative networks to associate and map the multi-source business data and process data, and obtain the association and mapping results. The target node determination module is used to identify key nodes in multi-source business data and process data based on management lines and in combination with association and mapping results, and to determine the target nodes that affect management decisions. The marginal information identification module is used to perform marginal change analysis on real-time multi-source business data and process data related to node state changes at the target node, and to identify marginal information that has changed relative to the historical state. The impact assessment module is used to evaluate the management impact of identified marginal information and determine its degree of impact on current management decisions. The decision support output module is used to generate corresponding decision support information and output it to the management decision interface when the marginal information meets the preset decision triggering conditions.

[0016] The beneficial effects of this invention are as follows: 1. This invention achieves the transformation from massive data to effective decision-making information by associating and mapping multi-source business and process data, identifying target nodes that affect decision-making, focusing on target nodes to capture marginal information, assessing the impact of marginal information on decision-making, and triggering targeted decision support. At the same time, it can match the needs of enterprise management decision-making, help decision-makers quickly grasp key changes, avoid blind decision-making, improve the accuracy and efficiency of management decisions, adapt to dynamic business scenarios of enterprises, and continuously optimize the effect of decision support.

[0017] 2. This invention achieves deep integration of multi-source business and process data with the enterprise management framework through a progressive process of hierarchical construction of management lines, precise processing of multi-source data, rigorous establishment of causal chains, and dynamic updating of mapping relationships. The dynamic mapping relationship library built based on causal chains and real-time data streams can be updated in real time with business changes, always maintaining the timeliness and accuracy of data and node associations, providing a high-quality, traceable, and dynamically adaptable core data foundation for subsequent key node identification and decision support.

[0018] 3. This invention constructs a time-series sequence of node marginal information based on the association mapping results, enabling the quantification and traceability of changes in node operating status; it divides intervention and control groups by combining node hierarchical attributes, and ensures the effectiveness of grouping through parallel trend testing, eliminating interference from historical trends and laying a scientific comparative foundation for marginal effect calculation; it further improves the accuracy of quantitative results by correcting the effect estimate by combining the support and confidence of causal association rules; through statistical significance testing and preset threshold screening, the target nodes finally locked have both significant decision-making influence and fit the characteristics of enterprise management levels, effectively eliminating redundant nodes. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart of a marginal information identification and decision support method based on management lines according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a boundary information identification and decision support system based on management lines according to an embodiment of the present invention.

[0020] In the picture: 1. Association mapping module; 2. Target node determination module; 3. Marginal information identification module; 4. Impact degree judgment module; 5. Decision support output module. Detailed Implementation

[0021] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0022] According to embodiments of the present invention, a method and system for identifying and supporting decision-making based on management lines are provided.

[0023] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, according to an embodiment of the present invention, a method for identifying and supporting decision-making based on management lines is provided, comprising the following steps: S1. Acquire multi-source business data and process data generated during enterprise management, and combine them with management lines that include vertical management paths and horizontal collaborative networks to associate and map the multi-source business data and process data, and obtain the association and mapping results. As a preferred implementation, the step of acquiring multi-source business data and process data generated during enterprise management, and combining them with management lines that include vertical management paths and horizontal collaborative networks, to correlate and map the multi-source business data and process data to obtain the correlation and mapping results includes the following steps: S11. Based on the company's organizational structure and business processes, identify the company's vertical management path and horizontal collaboration network, and generate a management line to characterize the company's management decision-making and execution process based on the vertical management path and horizontal collaboration network. As a preferred embodiment, the step of identifying the enterprise's vertical management path and horizontal collaboration network based on the enterprise's organizational structure and business processes, and generating a management line to characterize the enterprise's management decision-making and execution context based on the vertical management path and horizontal collaboration network, includes the following steps: S111. Obtain the company's organizational structure and identify the company's decision-making, management, and execution levels to obtain the identification results; It should be noted that an enterprise's organizational structure can be obtained by connecting to its internal OA system, organizational structure manuals, and other sources. This allows for the acquisition of core information about the organization's structure, such as departmental setup, job levels, and allocation of authority and responsibility. By combining this information with the core authority and responsibility of each department and position, the organizational structure can be hierarchically broken down, identifying three core levels: the decision-making level, the management level, and the execution level. The decision-making level comprises core authority and responsibility for corporate strategy formulation and approval of major matters, corresponding to senior management positions and decision-making committees. The management level comprises authority and responsibility for strategy implementation, departmental control, and process coordination, corresponding to middle management positions in various functional departments. The execution level comprises authority and responsibility for specific business operations and process implementation, corresponding to frontline operational positions and frontline personnel. The final result includes the identification of departments, positions, and core authority and responsibility for each level.

[0024] S112. Based on the identification results, determine the decision nodes and execution nodes in the vertical management path to form the vertical management path; Specifically, based on the obtained hierarchical identification results, the core work content and business nodes of each level are further broken down to determine the decision-making nodes and execution nodes in the vertical management path. Among them, the decision-making nodes correspond to the specific positions or meeting nodes in the decision-making layer that are responsible for strategy formulation, approval of major matters, and determination of core rules. Each decision-making node has a clearly defined decision-making scope, decision-making authority, and output results. The execution nodes correspond to the specific positions or operation nodes in the execution layer that are responsible for specific business operations, process implementation, and data reporting. The execution scope, operation standards, and input and output requirements of each execution node are clearly defined.

[0025] In this way, by combining the management's control functions, control connection nodes are built between decision-making nodes and execution nodes, clarifying how the instructions of the decision-making nodes are transmitted from top to bottom to the execution nodes, and how the feedback from the execution nodes is reported from bottom to top to the decision-making nodes, forming a vertical management path of top-down, closed-loop transmission from decision-making nodes, management control nodes, and execution nodes.

[0026] S113. Obtain the enterprise's business processes and identify collaborative nodes and rules by analyzing the interaction and collaboration relationships between various functional departments; Specifically, by identifying the company's core business processes, including procurement, production, sales, and cost control processes, and through methods such as connecting to business system logs, creating business process diagrams, and interviewing business leaders in various departments, information such as the specific flow steps, participating departments, operating standards, and flow timelines of each business process is obtained. Furthermore, the frequency of interactions, collaborative content, and information transmission paths between functional departments are analyzed in detail to identify collaborative nodes and rules in cross-departmental collaboration.

[0027] Among them, a collaborative node refers to a node that requires the joint participation and collaborative operation of multiple functional departments to complete the corresponding business process. Each collaborative node clearly defines its lead department, participating departments, and the collaborative responsibilities of each department. Collaboration rules refer to the collaboration norms that must be followed during the operation of collaborative nodes, including the collaboration sequence of each department, information transmission standards, division of responsibilities, conflict resolution mechanisms, etc.

[0028] S114. Based on collaborative nodes and collaborative rules, nodes scattered across different functional departments are connected according to their business coupling and information dependency relationships to form a horizontal collaborative network. It should be noted that, based on the identified collaborative nodes and collaborative rules, the execution nodes and control nodes scattered across different functional departments are classified and sorted by analyzing the business coupling and information dependency relationships between each node.

[0029] Specifically, business coupling primarily assesses the degree of business overlap and collaboration between nodes; higher coupling equates to higher connection priority. Information dependency primarily assesses the input-output relationship between nodes, i.e., whether the output of one node is the input of another, clarifying the direction of information transmission and the strength of dependency between nodes. Following the principle of prioritizing business coupling from high to low and information dependency from strong to weak, and in conjunction with collaboration rules, nodes from different functional departments are horizontally connected, determining the collaboration partners, collaboration methods, and information transmission paths for each node. This ultimately forms a horizontal collaboration network covering all cross-departmental collaboration nodes, connecting various functional departments, and adhering to collaboration rules.

[0030] S115. The vertical management path and the horizontal collaborative network are structurally integrated to obtain the management line used to represent the enterprise's management decision-making and execution process.

[0031] It's important to note that structured integration uses the vertical management path as its framework, embedding collaborative and related nodes from the horizontal collaboration network into the corresponding levels of the vertical management path. The responsibilities and business scope of the horizontal collaboration nodes are matched with those of the vertical nodes. For example, cross-departmental collaboration nodes are embedded below management control nodes and above execution nodes, enabling vertical decision-making directives to control horizontal collaboration and facilitating feedback of horizontal collaboration results to the vertical path. The integrated node network is then structurally analyzed to clarify the positioning of each node in the vertical hierarchy and its role in horizontal collaboration, streamlining the vertical transmission and horizontal collaboration relationships between nodes and eliminating redundancy and connection chaos. Simultaneously, by defining the overall operational logic of the management line, vertical decisions can be efficiently implemented through horizontal collaboration, and problems encountered during horizontal collaboration can be promptly fed back to the decision-making level through the vertical path. Ultimately, this forms a structured, closed-loop, and traceable management line, fully representing the entire process of enterprise management decision-making, transmission, execution, and collaboration.

[0032] S12. Collect multi-source business data and process data generated during enterprise management, and obtain standardized multi-source business data and process data by cleaning and standardizing the multi-source business data and process data. It's important to note that data cleaning involves systematically checking and addressing various anomalies in collected multi-source data. This includes removing data entries with excessively high missing values, using interpolation to supplement a small number of critical missing data entries, identifying and deleting duplicate data to avoid redundancy, and correcting data entry errors, formatting errors, and other abnormal data. Data standardization cleaning unifies the formats, units of measurement, coding rules, and naming conventions of various data types. For example, it adjusts inconsistent date and monetary data formats from different business systems to a unified format, converts quantitative data with different units of measurement into a unified measurement standard, and encodes various data entries according to preset rules, ensuring comparability and compatibility of data from different sources and of different types.

[0033] S13. Based on the multi-source business data and process data of the management line and the standard, determine the dynamic correlation and mapping results between the multi-source business data and process data and the management line nodes by analyzing the causal chain between nodes.

[0034] As a preferred embodiment, the step of determining the dynamic correlation and mapping results between the multi-source business data and process data and the management line nodes by analyzing the causal chains between nodes based on the multi-source business data and process data of the management line and the standard includes the following steps: S131. Classify and align standard multi-source business data and process data according to the attributes of management line nodes to form an initial data-node association pool; Specifically, the core attributes of each node in the management line include the node's hierarchical attributes, functional attributes, scope of authority and responsibility, and corresponding business processes. The hierarchical attributes are divided into decision-making level, management level, and execution level, while the functional attributes are divided into decision-making nodes, execution nodes, and collaboration nodes.

[0035] In this process, standard multi-source business data and process data are classified and sorted out. According to the business links and control dimensions corresponding to the data, different types of data are divided into management line nodes with corresponding attributes. During the classification process, the corresponding node identifier and the relationship between data and node are marked for each type of data. At the same time, the original data collection information and standardized format are retained. All the data and node correspondences after classification are integrated and summarized to form an initial data-node association pool.

[0036] S132. Using a pre-set expert experience base, filter out invalid data that is irrelevant to the management line node control requirements in the initial data-node association pool to form a node association candidate dataset. It should be noted that the preset expert experience base includes node control requirements, data association standards, and invalid data judgment rules extracted by experts in various fields of enterprise management. It also determines the core control indicators, required data range, and data types and judgment conditions unrelated to node control for each management line node.

[0037] This process involves using an expert experience database to compare each type of data in the initial data-node association pool with the corresponding node's management requirements and data association standards, determining whether each piece of data is necessary for node operation, decision-making, and execution. Invalid data primarily includes data unrelated to node management requirements, redundant data with extremely low value that cannot support node decision-making, and data that has no overlap with the node's business scope. Invalid data is then filtered in batches to remove invalid data entries from the initial data-node association pool, while filtering records are retained, clearly identifying the reason for removal and the corresponding node for each type of invalid data. The remaining valid data after filtering is then reviewed and verified to ensure that all remaining data is highly relevant to the management requirements of the corresponding management line nodes and can support the node's decision-making and execution. The association relationships between these valid data and their corresponding nodes are then integrated to form a node association candidate dataset.

[0038] S133. Perform structured transformation and feature extraction on the candidate dataset associated with nodes, and construct a causal chain by identifying the causal relationship rules between business data patterns and node state changes; In a preferred embodiment, the step of performing structured transformation and feature extraction on the node-associated candidate dataset, and constructing a causal chain by identifying the causal association rules between business data patterns and node state changes, includes the following steps: S1331. Transform the business events and process states in the node-associated candidate dataset into standardized transactions with a preset time window as the unit, and extract features related to node decisions for each standardized transaction to form a structured transaction set. It should be noted that the size of the preset time window is dynamically adjusted according to the business type. For example, a weekly time window is set for high-frequency businesses such as procurement and sales, while a monthly time window is set for low-frequency businesses such as strategic control and cost accounting. This ensures that each time window can fully cover the cycle of a single business link, while avoiding data redundancy due to an excessively large window and data fragmentation due to an excessively small window.

[0039] Furthermore, by splitting and integrating business events and process states in the node-associated candidate dataset, business events and process states occurring within the same time window and falling under the control of the same management line node are integrated into a standardized transaction. Each standardized transaction contains core information such as a transaction identifier, corresponding time window, associated node identifier, business event details, process state parameters, and original data entries, ensuring uniformity in format and content dimensions across different transactions, and providing comparability and analyzability. Then, by performing feature extraction related to node decisions, the core information in each standardized transaction that influences node decisions and reflects changes in node state is extracted. The extracted features are divided into three categories: numerical features, categorical features, and relational features.

[0040] Numerical features include quantifiable indicators such as the fluctuation range, mean, extreme values, and differences from historical benchmarks of business data; categorical features include qualitative indicators such as process flow status, node operation status, and business event types; and relational features include dependencies between different business data and the strength of the correlation between data and node status. All extracted features are bound to corresponding standardized transactions and related nodes, integrating all standardized transactions and corresponding features to form a structured transaction set.

[0041] S1332. Based on the structured transaction set, calculate the support and attention indicators for the combination of business data pattern-node state change, and when the preset constraints are met, include the combination of business data pattern-node state change in the condition frequent itemset. Specifically, from the structured transaction set, all combinations of business data patterns and node status changes can be filtered out. Business data patterns refer to a set of regular characteristics exhibited by the same type of business data, such as continuously rising procurement costs or a sharp drop in sales orders. Node status changes refer to the switching of management line nodes from one operational state to another, such as switching from normal control to warning state, or from execution to completion state. Then, the support and attention metrics for each combination are calculated. The support metric measures the frequency of the combination's occurrence in all standardized transactions, calculated by dividing the number of transactions in which the combination appears by the total number of transactions in the structured transaction set. The attention metric measures the relevance of the combination to node decisions. Combining the weights of node control needs in a pre-set expert experience base, the weight of the combination's influence on node decisions is calculated; the higher the attention, the greater the reference value of the combination for node decisions.

[0042] The preset constraints are set based on an expert experience base, specifying minimum thresholds for support and attention, such as support no less than 5% and attention no less than 0.6. Simultaneously, the business data patterns corresponding to the combinations and the node state changes are required to have a reasonable business logic relationship. The support and attention metrics for each combination are compared with the constraints. If both support and attention are met and a business logic relationship is established, the combination of business data pattern and node state change is included in the frequent itemset. If any constraint is not met, the combination is removed and not included in the frequent itemset. Finally, all combinations that meet the constraints are integrated to form the frequent itemset.

[0043] S1333. Construct a causal chain with causal direction based on conditional frequent itemsets.

[0044] As a preferred embodiment, constructing a causal chain with causal direction based on conditional frequent itemsets includes the following steps: S13331. Based on conditional frequent itemsets, generate candidate association rules with business data patterns as antecedents and node state changes as consequents. It should be noted that by traversing the resulting frequent itemsets, all valid combinations of business data patterns and node state changes are extracted. The core components of each combination are determined: one business data pattern corresponds to one or more node state changes, or multiple business data patterns jointly correspond to one node state change. Simultaneously, when generating candidate association rules, a fixed logic is followed, using the business data pattern in each combination as the antecedent of the rule and the node state change as the consequent. This ensures that each rule reflects the potential triggering relationship between the business data pattern and the node state change, laying the foundation for subsequent causal direction verification.

[0045] Specifically, for combinations of single business data patterns corresponding to single node state changes, association rules for single predecessors and single successors are directly generated; for combinations of multiple business data patterns corresponding to a single node state change, association rules for multiple predecessors and single successors are generated to clarify the combination relationship between multiple predecessors; for combinations of one business data pattern corresponding to multiple node state changes, multiple association rules for single predecessors and single successors are generated separately to avoid confusion of state changes of different nodes.

[0046] In addition, a unique identifier is assigned to each candidate association rule, which is associated with its corresponding frequent itemset combination. The management node identifier, support and attention index of the rule are recorded synchronously. Finally, all generated rules are integrated to form a complete set of candidate association rules.

[0047] S13332. Calculate the confidence level of each candidate association rule and filter candidate association rules with a confidence level greater than the preset minimum confidence level threshold. Specifically, confidence is used to measure the reliability of candidate association rules, that is, the probability that when the business data pattern corresponding to the antecedent occurs, the node state change corresponding to the consequent will also occur. It is calculated by dividing the number of times the combination corresponding to the candidate association rule occurs by the total number of times all combinations containing the antecedent of the rule occur. The calculation result ranges from 0 to 1. The higher the value, the stronger the reliability of the antecedent triggering the consequent, and the closer the association between the two.

[0048] Next, by accessing a pre-defined experience database, a preset minimum confidence threshold is obtained. This threshold is set differently based on the hierarchical and functional attributes of the management nodes, aligning with the decision-making rigor requirements of different nodes. For example, the rule threshold for decision-making level nodes is set to 0.8, for management level nodes to 0.7, and for execution level nodes to 0.6. This ensures that the threshold settings match the node control needs and avoids rule selection bias caused by uniform thresholds. Then, the confidence of each rule in the candidate association rule set is calculated one by one. Combining this with the quantitative data of frequent itemsets of the conditions mentioned earlier, the total number of combinations corresponding to each antecedent and the number of combinations of the corresponding rule are counted and substituted into the formula to complete the calculation. The calculation process and results for each rule are recorded for subsequent traceability and verification. After the calculation is completed, the confidence of each rule is compared with the preset minimum confidence threshold of the corresponding node. Candidate association rules with confidence greater than the threshold are selected, while rules with confidence lower than or equal to the threshold are removed. These removed rules are deemed unreliable and cannot be used as the basis for causal association. Finally, all rules that meet the confidence threshold are integrated to form the selected candidate association rule set.

[0049] S13333. Perform time series analysis on the filtered candidate association rules and verify whether the occurrence time of the antecedent event is earlier than that of the consequent event. If so, the candidate association rule is determined as a causal chain with causal direction; otherwise, the candidate association rule is excluded.

[0050] It should be noted that by filtering candidate association rules, all corresponding standardized transactions can be extracted. From each transaction, two key time points are extracted: the time when the business data pattern corresponding to the preceding event occurs, and the time when the node state change corresponding to the following event occurs. This ensures that both time points accurately correspond to specific moments within a preset time window, while retaining timestamp information to avoid verification bias caused by time ambiguity. Then, time series analysis is performed on each candidate association rule, comparing the occurrence times of the preceding and following events. Specifically, this includes verifying whether the preceding event occurs earlier than the following event, and verifying whether the time interval between the two is within a reasonable range. A reasonable time interval is set to not exceed a preset time window, ensuring a direct temporal correlation between the preceding and following events, rather than accidental simultaneous occurrences or unrelated sequential occurrences.

[0051] During the verification process, if the antecedent occurs earlier than the consequent and the time interval is within a reasonable range, it indicates that the association rule has a clear causal direction, meaning that the occurrence of the business data pattern does indeed trigger a change in the node state. This rule is then identified as a valid causal association rule and included in the causal chain construction. If the verification fails, meaning the antecedent occurs later than the consequent, or the time interval exceeds a reasonable range, it indicates that the rule is merely a data-level co-occurrence without an actual causal relationship and cannot demonstrate a causal direction; therefore, it is directly excluded. After verification, all valid causal association rules that pass the verification are linked together according to the transmission relationship between management line nodes. For example, the state change of an upstream node is used as a supplementary antecedent to the association rule of a downstream node, forming a complete causal chain with a clear causal direction.

[0052] S134. Based on the causal chain and real-time data stream, construct and dynamically update the mapping relationship library between business data patterns and node states, and output the dynamic association mapping results.

[0053] Specifically, a mapping relationship library between business data patterns and node states is built, using a causal chain with a clear causal direction as the core framework. This mapping relationship library needs to cover multiple functional modules to ensure the integrity, traceability, and updability of data and node associations. The core modules include a data pattern module, a node state module, a causal association module, an association strength module, and an update log module. The data pattern module stores the characteristics, identifiers, and corresponding data sources of all valid business data patterns. The node state module stores all possible operating states and state switching conditions of each node in the management line. The causal association module stores verified causal association rules and complete causal chains. The association strength module stores the support, attention, and confidence indicators corresponding to each mapping relationship. The update log module records the time, content, and reason for all subsequent update operations.

[0054] After establishing the mapping relationship library framework, initial data import and association construction are performed. The resulting complete causal chains, effective combinations of frequent conditional itemsets, and processed standard multi-source business data and process data are imported in batches into the corresponding modules of the mapping relationship library. Based on the triggering logic of the causal chains, a one-to-one or many-to-one mapping relationship is established between business data patterns and node states and management line nodes. The node state change type, associated management line node identifier, and causal transmission path corresponding to each business data pattern are determined. At the same time, association strength indicators such as support, attention, and confidence are bound to the corresponding mapping relationship to form the initial mapping relationship library. After the initial mapping relationship library is formed, it connects to various internal business systems and process management platforms to collect real-time data streams and monitor changes in business data patterns, node state transitions, and new data patterns and node states appearing in new business scenarios. When a new business data pattern appears, existing business data patterns fluctuate significantly, nodes exhibit new operating states or state transition types, or combinations in the real-time data stream that meet the causal chain triggering conditions but are not included in the existing mapping relationship are detected, the mapping relationship library update mechanism is automatically triggered. During the update process, the support, attention, and confidence indices of the new data pattern and node state combination are recalculated. Referring to the constraints in S1332 and the causal direction verification logic in S1333, the business logic association and temporal sequence relationship are verified. If all requirements are met, it is included in the mapping relationship library, and the corresponding content of the data pattern module, node state module, causal association module, and association strength module are updated. Simultaneously, business data patterns that no longer appear, invalid node state transitions, and mapping relationships with association strength indices below a preset threshold are deleted in batches, ensuring the simplicity and effectiveness of the mapping relationship library. Finally, it can output dynamically associated mapping results based on the dynamically updated mapping relationship library.

[0055] S2. Based on the management line and combined with the correlation and mapping results, identify key nodes in multi-source business data and process data to determine the target nodes that affect management decisions. As a preferred implementation, the process of identifying key nodes in multi-source business data and process data based on management lines and combining association and mapping results to determine target nodes affecting management decisions includes the following steps: S21. Based on the association and mapping results, obtain the marginal indicator data of each node in the management line, and construct the time series sequence of marginal information of each node in the management line according to the marginal indicator data. S22. Based on the time series of marginal information of each management line node, calculate the marginal effect of each management line node, identify key decision points, and generate a marginal effect analysis report. In a preferred embodiment, the step of calculating the marginal effect of each management line node based on the time series sequence of marginal information of each management line node, identifying key decision points, and forming a marginal effect analysis report includes the following steps: S221. Based on the time sequence of marginal information of each management line node, and combined with the hierarchical attributes of the management line nodes, the management line nodes are divided into intervention group nodes and control group nodes. As a preferred embodiment, the step of dividing the management line nodes into intervention group nodes and control group nodes based on the temporal sequence of the marginal information of each management line node and in combination with the hierarchical attributes of the management line nodes includes the following steps: S2211. Based on the vertical management path and horizontal collaborative network of the management line, extract the hierarchical attributes of each management line node; Specifically, the management line is the result of a structured fusion of vertical management paths and horizontal collaborative networks, encompassing the hierarchical positioning and functional attributes of each node. Based on the vertical management path of the management line, the hierarchical attributes of each node are extracted, clarifying whether each node belongs to the decision-making, management, or execution level. Simultaneously, the functional attributes of the nodes are associated, such as decision-making nodes, execution nodes, and collaborative nodes, forming a node hierarchy-functional attribute mapping table. During the extraction process, the position of each node in the vertical management path must be verified one by one to confirm its hierarchical affiliation. For example, decision-making level nodes correspond to the top of the vertical path, responsible for strategy formulation and major decisions; management level nodes correspond to the middle level of the vertical path, responsible for strategy implementation and departmental control; and execution level nodes correspond to the bottom of the vertical path, responsible for specific business execution. Simultaneously, combined with the horizontal collaborative network, the hierarchical attributes of collaborative nodes are labeled, ensuring that the hierarchical attributes of all management line nodes are accurately extracted without omission or confusion.

[0056] S2212. Based on the time series of marginal information of each node, identify the marginal change trend of each node within a preset time window, and detect whether there are marginal change events that meet the preset criteria. Specifically, the preset standards are based on the expert experience base and the differentiated settings of node level attributes. The core is to clarify the judgment thresholds for the fluctuation range and duration of marginal indicators. For example, the preset standard for marginal changes of decision-making level nodes is that the fluctuation range of marginal indicators exceeds the historical average by ±15% and the duration is not less than one time window. For management level nodes, it is that the fluctuation range exceeds the historical average by ±20% and the duration is not less than one time window. For execution level nodes, it is that the fluctuation range exceeds the historical average by ±25% and the duration is not less than one time window. At the same time, the judgment logic for the direction of fluctuation is determined.

[0057] Subsequently, trend analysis is performed on the time series of marginal information for each node. A time series trend recognition algorithm is used to determine the changing patterns of the node's marginal indicators within a preset time window, identifying different trends such as rising, falling, and stable. Simultaneously, the algorithm focuses on detecting whether there are marginal change events in the time series that meet preset criteria. Specifically, it determines whether any data point in the time series experiences fluctuations exceeding the preset threshold for the corresponding level, and whether this fluctuation persists for the preset time window. If such a event exists, the node is determined to have experienced a marginal change event, and information such as the time of occurrence, fluctuation amplitude, and direction of change is recorded. If no such event exists, the node is determined not to have experienced a significant marginal change, and its marginal indicators are in a stable operating state. After the detection is completed, a marginal change detection report for each node is generated.

[0058] S2213. The management line nodes that detect marginal change events are marked as intervention candidate nodes, and the remaining management line nodes that do not detect significant changes are used as control nodes.

[0059] Specifically, based on the marginal change detection results, all management line nodes can be grouped and labeled. Nodes that detect marginal change events meeting preset criteria are uniformly labeled as intervention candidate nodes. The core characteristic of these nodes is that their marginal indicators fluctuate significantly within a preset time window, indicating a clear marginal change. Simultaneously, management line nodes that do not detect significant marginal change events and whose marginal indicators remain stable within the preset time window are all designated as control nodes. The selection of control nodes requires no additional screening; it only needs to ensure that they have not experienced significant marginal changes and are potentially comparable to the hierarchical and functional attributes of the intervention candidate nodes. This provides a reference benchmark for subsequent parallel trend testing and counterfactual time series construction, isolating the impact of other interfering factors besides marginal changes on the node's operational status. After labeling, separate lists of intervention candidate nodes and control nodes are compiled, clearly specifying node identification, hierarchical attributes, functional attributes, and whether marginal changes have occurred, ensuring the grouping results are clear and traceable.

[0060] S222. Examine whether the intervention group nodes meet the preset parallel trend assumption within the time window before the marginal change occurs, and adjust the intervention group nodes that do not meet the parallel trend assumption until the preset parallel trend assumption is met. Specifically, the purpose of the parallel trend hypothesis is to verify whether the operating trends of the intervention group nodes and the matched control group nodes are consistent before the marginal change occurs, thereby ruling out the possibility that the marginal effect of the intervention group nodes is caused by their own historical trends. Before the test, the test time window is determined, which is a preset time period before the marginal change occurs, usually 3-5 time windows set in the previous text, consistent with the time dimension of the marginal information time series. The marginal information time series of all intervention group nodes and their matched control group nodes within the time window are extracted, focusing on the marginal indicator change trends of both during this time period.

[0061] In the testing process, a combination of trend comparison and statistical analysis was used. On the one hand, time-series trend graphs of marginal indicators of the intervention group and the control group were plotted to visually compare whether the slope and fluctuation range of the two trends were consistent. On the other hand, the trend coefficients of the marginal indicators of the two groups were calculated through regression analysis to test whether there was a significant difference between the two coefficients. The criteria for determining the parallel trend hypothesis were that there was no significant difference between the trend coefficients of the two groups and the goodness of fit of the trend graph was ≥0.8, and the statistical significance was p>0.05.

[0062] If the test results meet the preset criteria, it indicates that the intervention group node and the control group node have the same trend before the marginal change, and the next step can proceed. If the parallel trend hypothesis is not met, the intervention group node needs to be adjusted. The adjustment methods include the following three: First, rematch the intervention group node with the control group node, prioritizing control nodes of the same level, function, and with more similar historical trends; second, narrow the test time window before the marginal change occurs, eliminating data from time periods with abnormal trend fluctuations; third, if there is no suitable control node for the intervention group node, temporarily remove it from the intervention group and reselect alternative nodes from the intervention candidate nodes. After adjustment, the parallel trend test needs to be performed again, and this process is repeated until all retained intervention group nodes meet the preset parallel trend hypothesis.

[0063] S223. For intervention group nodes that satisfy the parallel trend hypothesis, construct a counterfactual time series sequence in which the intervention group nodes do not undergo marginal changes by combining them with the control group nodes that match them. It should be noted that by fitting the time series of marginal information of the control group nodes, the operational state of the intervention group nodes when no marginal change occurred is obtained. This transforms the unobservable counterfactual state into a calculable and comparable time series, providing a benchmark for subsequent marginal effect calculations. Specifically: First, the matching control group nodes corresponding to each intervention group node satisfying the parallel trend assumption are determined. One intervention group node corresponds to 1-3 control group nodes. The matching principles are same level, same function, and consistent historical trends. The marginal information time series sequences of these control group nodes are extracted over the entire time window. The mean or fitted value of the marginal indicators of the control group nodes is calculated to form a comprehensive control group time series sequence, ensuring that this sequence can stably reflect the node operation trend under the state of no marginal change. Then, a fitting algorithm is used, based on the comprehensive control group time series sequence and combined with the marginal information time series sequence of the intervention group nodes before the marginal change, to fit the marginal indicator data of the intervention group nodes after the marginal change but before the marginal change occurred. The fitted virtual data of the intervention group nodes after the marginal change is connected with the actual marginal information time series sequence before the change, forming a complete counterfactual time series sequence without marginal change. Each counterfactual time series corresponds one-to-one with the actual marginal information time series of the corresponding intervention group node, containing the same time dimension and marginal indicator type. The only difference is in the data of the time period after the marginal change occurs, clearly presenting a comparison of the two states of the intervention group node with and without marginal change.

[0064] S224. Based on the counterfactual time series where no marginal change occurred in the intervention group nodes, extract the actual observation results of the intervention group nodes after the marginal change occurred, calculate the difference between the actual observation results and the corresponding time point values ​​of the counterfactual time series, and obtain the instantaneous value of the marginal effect at each time point. It should be noted that, from the actual marginal information time series of the intervention group node, all actual observation data after the marginal change occurred are extracted, and the time point corresponding to each data point is labeled to ensure the accuracy of the time points; at the same time, virtual fitting data at the same time point are extracted from the counterfactual time series corresponding to the intervention group node to form two sets of one-to-one time point data, and then the difference between the two sets of data is calculated one by one according to the time point. The calculation logic is to subtract the corresponding value of the counterfactual time series from the actual observation result at the same time point to obtain the instantaneous value of the marginal effect at that time point.

[0065] In this system, the sign of the instantaneous value represents the direction of the marginal effect. A positive value indicates that the marginal change has a positive impact on the node's operation, while a negative value indicates a negative impact. The absolute value of the instantaneous value indicates the strength of the marginal change's impact at that point in time. After the calculation is completed, the instantaneous marginal effect values ​​of each intervention group node are organized, sorted in chronological order, and labeled with the corresponding time point and marginal indicator type, forming a sequence of instantaneous marginal effect values ​​for that node.

[0066] S225. Calculate the average treatment effect based on the instantaneous value of the marginal effect, and use it as the initial marginal effect estimate of the intervention group nodes; then, combine the support and confidence of the association rule to correct the initial marginal effect estimate and obtain the final marginal effect estimate. It should be noted that the initial marginal effect estimate of the intervention group node is calculated by averaging the obtained instantaneous marginal effect value sequence, i.e., the average treatment effect. The calculation method is the arithmetic mean of the instantaneous marginal effect values ​​at all time points after the marginal change occurred. This average value can comprehensively reflect the overall impact of the marginal change on the intervention group node and is used as the initial marginal effect estimate.

[0067] Since the initial estimate only considered the difference between the actual observation and the counterfactual, without considering the causal relationship between the data and the node state changes, it is necessary to combine the constructed causal relationship rules and extract the support and confidence indices of the relationship rules corresponding to the nodes in the intervention group to revise the initial marginal effect estimate and improve its accuracy. Specifically: the product of support and confidence is used as the causal relationship strength coefficient. This coefficient ranges from 0 to 1. The larger the coefficient, the stronger the causal relationship between the business data pattern and the node state changes, and the higher the reliability of the marginal effect; the smaller the coefficient, the weaker the causal relationship, and the lower the weight of the initial estimate needs to be. The revised calculation formula is as follows: Final marginal effect estimate = Initial marginal effect estimate × Causal correlation strength coefficient; If a node in an intervention group corresponds to multiple association rules, the average of the multiple coefficients is taken as the final correction coefficient. After correction, the final marginal effect estimate for each intervention group node is obtained, and the initial estimate, correction coefficient, final estimate, and calculation process are recorded.

[0068] S226. Calculate the statistical significance of the final marginal effect estimate, and screen out the intervention group nodes that meet the preset judgment criteria. Determine the screened intervention group nodes as key decision points, and generate a marginal effect analysis report that includes the magnitude and direction of the marginal effect through integration.

[0069] Specifically, a statistical significance test is performed on the final marginal effect estimate for each intervention group node. The t-test is used to calculate the statistical significance p-value of the estimate, verifying whether the marginal effect is caused by accidental factors rather than the true impact of marginal change. The smaller the p-value, the higher the statistical significance of the final marginal effect estimate, and the more reliable the result. This is further combined with pre-defined judgment criteria. These criteria are based on an expert experience database and differentiated according to the node's hierarchical attributes. The core criteria include two requirements: first, a statistical significance p < 0.05, indicating that the marginal effect is a true impact; second, the absolute value of the final marginal effect estimate is greater than the pre-defined effect threshold for the corresponding node level, for example, ≥ 0.2 for decision-making level nodes, ≥ 0.15 for management level nodes, and ≥ 0.1 for execution level nodes. These threshold settings align with the rigorous requirements of node control.

[0070] Specifically, by comparing the statistical significance p-value and the final estimated marginal effect for each intervention group node, intervention group nodes that simultaneously meet both preset criteria are selected and identified as key decision points. The core characteristic of these nodes is that marginal changes have a significant and reliable impact on their operational status, and this impact is sufficient to support management decision adjustments; these nodes are the focus of subsequent attention. After selection, all analysis results are integrated to generate a standardized marginal effect analysis report. The report covers core information for each intervention group node, including its identifier, hierarchical attributes, functional attributes, final estimated marginal effect, direction of marginal effect, statistical significance p-value, causal correlation strength coefficient, instantaneous marginal effect value sequence, key decision point identifier, and selection criteria. It also supplements the report with an overall analysis of node marginal effects and explanations of abnormal nodes, ensuring data completeness and logical clarity.

[0071] For example, as shown in Table 1, only nodes 001 and 003 simultaneously met the requirements of statistical significance and effect strength, and were identified as key decision points.

[0072] Table 1. Example of screening key decision points for intervention group nodes

[0073] S23. Based on the marginal effect analysis report, and combined with the preset decision impact threshold and target node screening rules, identify the target nodes that have an impact on the current management decision from the key decision points.

[0074] It should be noted that the target node selection rules are formulated based on node characteristics and decision-making needs. The rules include three aspects: First, the absolute value of the final marginal effect estimate of the key decision point must exceed the decision impact threshold of its corresponding level, and the statistical significance p < 0.05, to ensure that the marginal effect of the node is real and significant and can have a substantial impact on the decision. Second, the functional attributes of the node are highly matched with the core focus of the current management decision, that is, the causal transmission path of the node can directly cover the control scope of the current decision. For example, if the current decision focus is to improve sales performance, then key decision points related to sales control will be selected first, and nodes that are irrelevant to the current decision will be eliminated. Third, redundant nodes will be eliminated. For multiple key decision points with overlapping functions and consistent causal transmission paths, only the node with the largest marginal effect and the most direct impact on the decision will be retained to avoid node duplication leading to low efficiency in subsequent marginal information identification.

[0075] S3. At the target node, perform marginal change analysis on real-time multi-source business data and process data related to node status changes, and identify marginal information that has changed relative to historical status. Specifically, during marginal change analysis, business data and process data corresponding to the target node are collected in real time. The real-time data is synchronized and organized according to a preset time window, and a timestamp is marked for each real-time data point. The real-time data is compared one by one with the historical state benchmark. Methods such as difference analysis and trend comparison are used to determine whether the real-time data has changed significantly relative to the historical state. The change judgment standard is consistent with the preset marginal change standard of S2212, that is, the fluctuation amplitude exceeds the threshold corresponding to the historical benchmark, and the duration reaches a time window.

[0076] In identifying marginal information that has changed relative to historical states, significant changes are detected and, in conjunction with the functional attributes of the target node and the causal transmission path, information reflecting abnormal node operation and potentially affecting management decisions is selected as marginal information. Each piece of marginal information must be labeled with the corresponding target node identifier, data type, direction of change, fluctuation amplitude, occurrence time, and corresponding historical benchmark, and compiled into a marginal information list.

[0077] S4. Conduct a management impact assessment on the identified marginal information to determine its degree of impact on current management decisions; Specifically, before conducting a management impact assessment on the identified marginal information, the information is categorized and organized according to the hierarchical attributes, functional attributes, and change types of the target nodes. For example, strategic marginal information corresponds to decision-making level target nodes, control-related marginal information corresponds to management level target nodes, and operational marginal information corresponds to execution level target nodes. The direction of change, fluctuation range, and duration of each piece of marginal information are also marked. Simultaneously, core assessment criteria are configured, including the core priorities of current corporate management decisions, the magnitude of the marginal effect of the target nodes, the strength of the causal transmission path corresponding to the marginal information, the support and confidence of association rules, and pre-set impact assessment standards. The assessment standards are conducted from four dimensions: first, the scope of impact, i.e., the number of business links and functional departments affected by the marginal information; second, the intensity of impact, i.e., the fluctuation range and magnitude of the marginal effect corresponding to the marginal information; third, timeliness, i.e., whether the impact of the marginal information on current decisions is urgent and whether immediate adjustments to decisions are needed; and fourth, relevance, i.e., the degree of alignment between the marginal information and the core priorities of current management decisions.

[0078] Furthermore, the evaluation process employs a combination of quantitative and qualitative methods, setting scoring criteria for each dimension and calculating a comprehensive evaluation score for each piece of marginal information. Based on the score, the degree of impact is categorized into three levels: high, medium, and low. A high impact level corresponds to a comprehensive score ≥80 points, indicating that the marginal information exhibits large fluctuations, a wide impact range, is highly relevant to current decisions, and is time-sensitive, potentially leading to deviations in decision implementation or significant operational risks. A medium impact level corresponds to 60-79 points, indicating that the marginal information has some fluctuations, a limited impact range, and is relevant to decisions, requiring attention and timely adjustments to decisions. A low impact level corresponds to <60 points, indicating that the marginal information has small fluctuations, a narrow impact range, low relevance to current decisions, and no substantial impact on decision implementation. Upon completion of the evaluation, a marginal information management impact assessment report must be generated, including the target node corresponding to each piece of marginal information, the comprehensive evaluation score, the degree of impact, the scope of impact, and potential risks, while also indicating the priority order for processing each piece of marginal information.

[0079] S5. When the degree of impact meets the preset decision triggering conditions, generate corresponding decision support information and output it to the management decision interface.

[0080] Specifically, based on the assessment results of the management impact of marginal information, after determining the impact level (high, medium, and low) of each piece of marginal information, and in conjunction with preset decision triggering conditions, differentiated triggering judgments are made for marginal information of different levels. This ensures that decision triggering aligns with management needs, takes into account timeliness and risk control. The decision triggering conditions are specifically divided into three categories, which precisely correspond to the impact assessment results: First, marginal information with a high impact level directly meets the triggering conditions because it has a significant impact on current management decisions and is highly timely, requiring no additional waiting and immediately initiating the decision support information generation process. Second, marginal information with a medium impact level, considering its certain persistence, meets the triggering conditions if the duration of this level of marginal information reaches a preset threshold, avoiding erroneous decision triggering due to short-term fluctuations. Third, marginal information with a low impact level, where a single piece of low-impact marginal information has no substantial impact on decision-making, but to avoid potential cumulative risks, if multiple pieces of the same type of low-impact marginal information appear simultaneously, and the overall impact level after aggregation reaches the medium-impact level standard, the triggering conditions are met, enabling early detection and response to risks.

[0081] The triggering judgment process needs to be based on the management impact assessment results of marginal information. All marginal information that has completed the management impact assessment is compared with the above three types of decision triggering conditions one by one. Marginal information that meets the triggering conditions is selected and sorted according to priority. The sorting principle is that marginal information with high impact level is prioritized over medium impact level, and marginal information with medium impact level is prioritized over low impact level. Marginal information of the same level is sorted according to timeliness to ensure that the marginal information with the greatest impact on decision-making and the most urgent need for resolution is processed first.

[0082] For each piece of marginal information that meets the triggering conditions, it is necessary to combine its corresponding target node attributes, the direction of the marginal effect, the scope of influence, and the causal transmission path and correlation mapping results constructed above to generate targeted and actionable decision support information, ensuring that the decision recommendations align with node control needs and actual business scenarios. For example, if the target node is a procurement control node, and the marginal information is a continuous increase in procurement costs with a negative marginal effect that inhibits procurement control efficiency, then the decision support information includes specific cost control recommendations, directions for selecting qualified alternative suppliers, procurement batch and cycle adjustment plans, and corresponding cost calculation references. If the target node is a sales control node, and the marginal information is a positive surge in sales order volume with a positive marginal effect that promotes sales performance, then the decision support information includes temporary capacity adjustment recommendations, emergency inventory allocation plans, sales channel optimization directions, and order delivery priority arrangements, etc.

[0083] After decision support information is generated, it needs to be standardized. This involves unifying the information format, identifying core implementation recommendations, labeling corresponding marginal information sources, assessing the level of impact, and determining implementation priorities. This avoids information clutter and ensures that decision-makers can quickly grasp the core content and implement it directly. Simultaneously, it connects to the enterprise's internal management decision-making interface, establishing a standardized information output interface to achieve seamless integration between decision support information and the management decision-making interface. The standardized decision support information is then output to the decision-making interface according to a preset priority order.

[0084] like Figure 2 As shown, according to an embodiment of the present invention, a marginal information identification and decision support system based on management lines is provided, the system comprising: The association mapping module 1 is used to acquire multi-source business data and process data generated during enterprise management, and combine them with management lines that include vertical management paths and horizontal collaborative networks to associate and map the multi-source business data and process data to obtain association and mapping results. Target node determination module 2 is used to identify key nodes in multi-source business data and process data based on management lines and in combination with association and mapping results, and to determine target nodes that affect management decisions. The marginal information identification module 3 is used to perform marginal change analysis on real-time multi-source business data and process data related to node state changes at the target node, and identify marginal information that has changed relative to the historical state. The impact assessment module 4 is used to evaluate the management impact of the identified marginal information and determine its degree of impact on current management decisions. The decision support output module 5 is used to generate corresponding decision support information and output it to the management decision interface when the marginal information meets the preset decision triggering conditions.

[0085] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0086] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying and supporting decision-making based on marginal information of management lines, characterized in that, Includes the following steps: S1. Acquire multi-source business data and process data generated during enterprise management, and combine them with management lines that include vertical management paths and horizontal collaborative networks to associate and map the multi-source business data and process data, and obtain the association and mapping results. S2. Based on the management line and combined with the correlation and mapping results, identify key nodes in multi-source business data and process data to determine the target nodes that affect management decisions. S3. At the target node, perform marginal change analysis on real-time multi-source business data and process data related to node status changes, and identify marginal information that has changed relative to historical status. S4. Conduct a management impact assessment on the identified marginal information to determine its degree of impact on current management decisions; S5. When the degree of impact meets the preset decision triggering conditions, generate corresponding decision support information and output it to the management decision interface.

2. The marginal information identification and decision support method based on management lines according to claim 1, characterized in that, The process of acquiring multi-source business data and process data generated during enterprise management, and combining this with management lines that include vertical management paths and horizontal collaborative networks, to correlate and map the multi-source business data and process data to obtain the correlation and mapping results includes the following steps: S11. Based on the company's organizational structure and business processes, identify the company's vertical management path and horizontal collaboration network, and generate a management line to characterize the company's management decision-making and execution process based on the vertical management path and horizontal collaboration network. S12. Collect multi-source business data and process data generated during enterprise management, and obtain standardized multi-source business data and process data by cleaning and standardizing the multi-source business data and process data. S13. Based on the multi-source business data and process data of the management line and the standard, determine the dynamic correlation and mapping results between the multi-source business data and process data and the management line nodes by analyzing the causal chain between nodes.

3. The marginal information identification and decision support method based on management lines according to claim 2, characterized in that, The process of identifying the enterprise's vertical management path and horizontal collaboration network based on its organizational structure and business processes, and generating a management line to characterize the enterprise's management decision-making and execution network based on the vertical management path and horizontal collaboration network, includes the following steps: S111. Obtain the company's organizational structure and identify the company's decision-making, management, and execution levels to obtain the identification results; S112. Based on the identification results, determine the decision nodes and execution nodes in the vertical management path to form the vertical management path; S113. Obtain the enterprise's business processes and identify collaborative nodes and rules by analyzing the interaction and collaboration relationships between various functional departments; S114. Based on collaborative nodes and collaborative rules, nodes scattered across different functional departments are connected according to their business coupling and information dependency relationships to form a horizontal collaborative network. S115. The vertical management path and the horizontal collaborative network are structurally integrated to obtain the management line used to represent the enterprise's management decision-making and execution process.

4. The marginal information identification and decision support method based on management lines according to claim 2, characterized in that, The process of determining the dynamic correlation and mapping results between multi-source business data and process data and management line nodes by analyzing the causal chains between nodes, based on multi-source business data and process data and standards, includes the following steps: S131. Classify and align standard multi-source business data and process data according to the attributes of management line nodes to form an initial data-node association pool; S132. Using a pre-set expert experience base, filter out invalid data that is irrelevant to the management line node control requirements in the initial data-node association pool to form a node association candidate dataset. S133. Perform structured transformation and feature extraction on the candidate dataset associated with nodes, and construct a causal chain by identifying the causal relationship rules between business data patterns and node state changes; S134. Based on the causal chain and real-time data stream, construct and dynamically update the mapping relationship library between business data patterns and node states, and output the dynamic association mapping results.

5. The marginal information identification and decision support method based on management lines according to claim 4, characterized in that, The process of performing structured transformation and feature extraction on the candidate dataset associated with nodes, and constructing a causal chain by identifying the causal relationship rules between business data patterns and node state changes, includes the following steps: S1331. Transform the business events and process states in the node-associated candidate dataset into standardized transactions with a preset time window as the unit, and extract features related to node decisions for each standardized transaction to form a structured transaction set. S1332. Based on the structured transaction set, calculate the support and attention indicators for the combination of business data pattern-node state change, and when the preset constraints are met, include the combination of business data pattern-node state change in the condition frequent itemset. S1333. Construct a causal chain with causal direction based on conditional frequent itemsets.

6. The marginal information identification and decision support method based on management lines according to claim 5, characterized in that, The construction of a causal chain with causal direction based on conditional frequent itemsets includes the following steps: S13331. Based on conditional frequent itemsets, generate candidate association rules with business data patterns as antecedents and node state changes as consequents. S13332. Calculate the confidence level of each candidate association rule and filter candidate association rules with a confidence level greater than the preset minimum confidence level threshold. S13333. Perform time series analysis on the filtered candidate association rules and verify whether the occurrence time of the antecedent event is earlier than that of the consequent event. If so, the candidate association rule is determined as a causal chain with causal direction; otherwise, the candidate association rule is excluded.

7. The marginal information identification and decision support method based on management lines according to claim 1, characterized in that, The process of identifying key nodes in multi-source business data and process data based on management lines and combining correlation and mapping results to determine target nodes affecting management decisions includes the following steps: S21. Based on the association and mapping results, obtain the marginal indicator data of each node in the management line, and construct the time series sequence of marginal information of each node in the management line according to the marginal indicator data. S22. Based on the time series of marginal information of each management line node, calculate the marginal effect of each management line node, identify key decision points, and generate a marginal effect analysis report. S23. Based on the marginal effect analysis report, and combined with the preset decision impact threshold and target node screening rules, identify the target nodes that have an impact on the current management decision from the key decision points.

8. The marginal information identification and decision support method based on management lines according to claim 7, characterized in that, The process of calculating the marginal effect of each management line node based on the time series of marginal information of each management line node, identifying key decision points, and generating a marginal effect analysis report includes the following steps: S221. Based on the time sequence of marginal information of each management line node, and combined with the hierarchical attributes of the management line nodes, the management line nodes are divided into intervention group nodes and control group nodes. S222. Examine whether the intervention group nodes meet the preset parallel trend assumption within the time window before the marginal change occurs, and adjust the intervention group nodes that do not meet the parallel trend assumption until the preset parallel trend assumption is met. S223. For intervention group nodes that satisfy the parallel trend hypothesis, construct a counterfactual time series sequence in which the intervention group nodes do not undergo marginal changes by combining them with the control group nodes that match them. S224. Based on the counterfactual time series where no marginal change occurred in the intervention group nodes, extract the actual observation results of the intervention group nodes after the marginal change occurred, calculate the difference between the actual observation results and the corresponding time point values ​​of the counterfactual time series, and obtain the instantaneous value of the marginal effect at each time point. S225. Calculate the average treatment effect based on the instantaneous value of the marginal effect, and use it as the initial marginal effect estimate of the intervention group nodes; then, combine the support and confidence of the association rule to correct the initial marginal effect estimate and obtain the final marginal effect estimate. S226. Calculate the statistical significance of the final marginal effect estimate, and screen out the intervention group nodes that meet the preset judgment criteria. Determine the screened intervention group nodes as key decision points, and generate a marginal effect analysis report that includes the magnitude and direction of the marginal effect through integration.

9. The marginal information identification and decision support method based on management lines according to claim 8, characterized in that, The process of dividing management line nodes into intervention group nodes and control group nodes based on the time sequence of marginal information of each management line node and the hierarchical attributes of the management line nodes includes the following steps: S2211. Based on the vertical management path and horizontal collaborative network of the management line, extract the hierarchical attributes of each management line node; S2212. Based on the time series of marginal information of each node, identify the marginal change trend of each node within a preset time window, and detect whether there are marginal change events that meet the preset criteria. S2213. The management line nodes that detect marginal change events are marked as intervention candidate nodes, and the remaining management line nodes that do not detect significant changes are used as control nodes.

10. A marginal information identification and decision support system based on management lines, used to implement the marginal information identification and decision support method based on management lines as described in any one of claims 1-9, characterized in that, The system includes: The association mapping module is used to acquire multi-source business data and process data generated during enterprise management, and combine them with management lines that include vertical management paths and horizontal collaborative networks to associate and map the multi-source business data and process data, and obtain the association and mapping results. The target node determination module is used to identify key nodes in multi-source business data and process data based on management lines and in combination with association and mapping results, and to determine the target nodes that affect management decisions. The marginal information identification module is used to perform marginal change analysis on real-time multi-source business data and process data related to node state changes at the target node, and to identify marginal information that has changed relative to the historical state. The impact assessment module is used to evaluate the management impact of identified marginal information and determine its degree of impact on current management decisions. The decision support output module is used to generate corresponding decision support information and output it to the management decision interface when the marginal information meets the preset decision triggering conditions.

Images