Quantitative calculation-based enterprise knowledge graph node priority allocation method and system

Abstract

The application discloses a kind of enterprise knowledge graph node priority allocation method and system based on quantification calculation, belong to knowledge graph data processing technical field.This method collects two kinds of frequency data of user query, business alarm and two kinds of severity data of financial loss, organization influence, relies on enterprise life cycle stage adaptation sensitive weight coefficient, respectively obtains each management module frequency score and severity score;Again, two-dimensional mapping algorithm is used to generate multi-level differentiated priority label, and the label associated metadata is written into the knowledge graph node to build semantic association;While real-time monitoring enterprise life cycle stage migration, with the help of attenuation factor, score smoothing transition is completed, and priority label automatic recalculation and incremental update are realized.The application can adaptively adjust the division threshold, accurately complete the automation allocation and dynamic iterative update of the priority of the graph node, the priority allocation precision is high, the label stability is strong, and it adapts to the application requirements of enterprise full life cycle management.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph and information processing technology, and in particular to a method and system for prioritizing enterprise knowledge graph nodes based on quantitative calculation. Background Technology

[0002] Knowledge graphs represent entities and their relationships through a graph structure of nodes and edges, and have been widely used in recommendation systems, intelligent question answering, and other fields. In enterprise management scenarios, knowledge graph nodes typically need to be assigned some priority or importance weight to guide downstream recommendation ranking and reasoning path selection.

[0003] There are currently three main types of methods for prioritizing knowledge graph nodes: The first is manual labeling, where domain experts assign fixed weights to each node based on their experience and knowledge. However, the accuracy is limited by the experts' subjective judgment and cannot adapt to the dynamic changes in the needs of enterprises at different stages of development. The second is static frequency statistics, which calculates a fixed priority score based on the access frequency or trigger count of nodes using global historical data. This method lacks the ability to model the cross-relationship between node trigger frequency and the severity of business impact. The third is a simple linear numerical superposition method, which linearly combines frequency and severity indicators into a single score. This method does not consider the two-dimensional cross-effect of the two indicators and cannot map nodes to multi-level priority regions with clear management semantics.

[0004] Meanwhile, existing technologies also have several technical shortcomings: First, the node weights in existing knowledge graphs are mostly static, making it impossible to dynamically adjust recommendation priorities based on the company's development stage. The importance of the same business management module (such as capital operation) differs significantly between the startup and growth stages, and existing technologies lack priority calculation models that use lifecycle stages as dynamic variables. Second, frequency indicators are sourced from a single source, typically only considering the number of user queries and failing to include alarm data from related business systems in frequency calculations, lacking a fusion calculation model for multi-source frequency data. Third, there is a lack of quantitative modeling capabilities for impact severity, making it impossible to output multi-level priority area labels with clear management semantics. Fourth, there is a lack of an automatic priority label migration mechanism during stage transitions. When a company migrates from one lifecycle stage to another, the priority labels of each business management module need to be updated accordingly. Existing technologies rely on manual relabeling, which is inefficient and prone to omissions. Fifth, threshold divisions mostly use fixed empirical values, failing to consider the differences in data distribution characteristics under different lifecycle stages, resulting in a disconnect between threshold division results and actual data distribution, and a decrease in the accuracy of priority label allocation.

[0005] Therefore, existing technologies are still insufficient to meet the actual needs of enterprises for refined and dynamic business management. Summary of the Invention

[0006] The problem to be solved by this invention is to provide a method for prioritizing enterprise knowledge graph nodes based on quantitative calculation. This method relies on a frequency-influence dual-dimensional feature quantitative calculation model and combines the data distribution characteristics of different business operation stages of an enterprise to achieve automated, accurate, dynamic, and intelligent allocation and processing of priority labels for enterprise business knowledge graph entity nodes.

[0007] This invention adopts the following technical solution: a method for prioritizing enterprise knowledge graph nodes based on quantitative calculation, comprising the following steps:

[0008] S1. Frequency data acquisition and preprocessing: Collect user query frequency data and business system alarm frequency data, set a unified time window to complete frequency statistics, normalize the two types of frequency data, and configure a boundary protection mechanism for zero values.

[0009] S2. Frequency score calculation: Construct a stage-sensitive weighted fusion model, combine the stage-sensitive weight coefficients of the enterprise life cycle stage, and fuse the normalized user query frequency data and business system alarm frequency data to calculate the frequency score corresponding to each business management module.

[0010] S3. Severity data collection and preprocessing: Collect data on the impact of financial losses and the scope of impact of organizational structure, determine the causal relationship of the amount of financial losses through a predefined attribution rule chain, determine the number of departments affected, normalize the two types of severity data, and configure opportunity cost discounting and alternative loss and numerical boundary protection mechanisms.

[0011] S4. Severity Score Calculation: Based on the stage-sensitive weighted fusion model, combined with the severity weight coefficients adapted to the enterprise life cycle stage, and fused and normalized financial loss impact data and organizational structure impact range data, the severity score corresponding to each stage of each business management module is calculated.

[0012] S5. Frequency and Severity Two-Dimensional Mapping: Based on the obtained frequency and severity scores, a two-dimensional mapping algorithm is used to generate multi-level differentiated priority labels.

[0013] S6. Priority tags are written into the knowledge graph: The generated priority tags are associated with the corresponding metadata and written into the enterprise operation and management knowledge graph nodes to establish a semantic relationship between modules and priority tags.

[0014] S7. Automatic tag migration during enterprise lifecycle stage transition: Real-time monitoring of enterprise lifecycle stage transition status, automatic execution of differentiated score recalculation and incremental tag update after stage switch, and introduction of decay factor to achieve score transition, completing fully automatic migration and update of priority tags.

[0015] As a preferred embodiment, the frequency data acquisition and preprocessing in step S1 includes the following sub-steps:

[0016] S1.1 Collect query frequency data from user behavior logs for each business management module. In each lifecycle stage stage_i, the number of queries (query_count) by users to this module within a preset time window T is counted.

[0017] The user behavior log data collection includes: extracting the URL path and query parameters from the HTTP request from the web server access log, and mapping the URL path to the corresponding business management module through a predefined module identifier mapping table; extracting the query keywords entered by the user from the search engine query log, and identifying the business management module to which the query intent belongs through a keyword matching algorithm; each log entry field includes a timestamp, user identifier, module identifier, and query type.

[0018] S1.2, Collect alarm frequency data from business system alarm logs: For each business management module In each lifecycle stage stage_i, the alarm count of the business system associated with this module within the time window T is counted. The alarm association rules are determined by a predefined keyword matching and anomaly pattern recognition model.

[0019] The business system alarm logs include data collection from the alarm interfaces of the financial system, human resources system, supply chain management system, and customer relationship management system. The structured alarm data is collected from these alarm interfaces. Each alarm entry field includes alarm time, alarm level, alarm module identifier, anomaly type, and anomaly description. The anomaly pattern recognition model uses a regular expression-based rule matching method to associate and map the anomaly keywords in the alarm description text with the module identifier.

[0020] S1.3. Normalize the collected raw frequency data to eliminate dimensional differences and obtain the normalized user query frequency query_freq and business system alarm frequency alert_freq.

[0021] S1.4 Boundary Protection Mechanism: When When it equals 0, Forced assignment of 0; when When it equals 0, Forced to be assigned a value of 0.

[0022] As a preferred option, in step S2, based on the stage-sensitive weighted fusion model, a stage-sensitive weight coefficient is set, and the frequency score F of each module in each stage is calculated. The stage-sensitive weight coefficient has a value range of [0.3, 0.7], a default value of 0.6, and is automatically adjusted according to the characteristics of different life cycle stages.

[0023] As a preferred embodiment, the severity data acquisition and preprocessing in step S3 includes the following sub-steps:

[0024] S3.1 Collect financial impact data from the financial loss database: For each business management module module_j, under each life cycle stage stage_i, calculate the financial loss amount financial_loss directly or indirectly caused by the abnormality of this module within the statistical time window T. The causal relationship of the financial loss amount is determined by a predefined attribution rule chain.

[0025] The financial loss database includes data collection comprising: extracting accounting voucher data associated with abnormal events from the financial module of the enterprise ERP system, with fields including debit amount, credit amount, accounting subject, and associated business document number; the attribution rule chain adopts a graph traversal algorithm: starting from the abnormal event, performing a depth-first traversal along the business document association relationship (purchase order, sales order, payment note, receipt note), and accumulating the amount field on the traversal path as the financial loss amount of the module.

[0026] Specifically, the traversal depth is limited to 3 levels to avoid infinite recursion.

[0027] S3.2 Collect impact scope data from the organizational structure database: For each business management module module_j, under each life cycle stage stage_i, count the number of internal departments affected by the module's anomaly within the time window T.

[0028] The criterion for determining the number of departments affected is: if the workflow of that department is substantially disrupted within 72 hours of the occurrence of the abnormal event.

[0029] The organizational structure database includes the following data collection methods: extracting process instance data from the workflow engine of the enterprise OA system and identifying process instances that are suspended, rolled back, or timed out due to abnormal events; extracting the list of receiving departments for abnormal event notifications from group messages in the enterprise instant messaging system; extracting the departments to which tasks marked as blocked due to abnormal events belong from the task dashboard of the enterprise project management system; taking the union of the collected data and removing duplicates to obtain the affected scope data affected_depts.

[0030] S3.3. Normalize the collected severity data to obtain normalized financial loss impact data. Data on the scope of influence of organizational structure ;

[0031] S3.4 Boundary Protection Mechanism: When When equal to 0, Forced assignment of 0; when When equal to 0, Forced assignment of 0;

[0032] S3.5, Handling of Alternative Loss Amounts: If the direct financial loss of a certain business management module is 0 at a specific stage, but there is a quantifiable opportunity cost, then it is allowed to introduce a preset discount factor (default 50%), and use the adjusted alternative loss amount as the final financial loss effective_financial_loss.

[0033] Specifically, the discount factor is set based on the fact that opportunity cost is not an actual loss that has already occurred, and its certainty is lower than that of direct financial loss. Therefore, its weight is reduced by a discount factor of 50%. If the calculated result of effective_financial_loss is greater than max_financial_loss(stage_i), then upper limit truncation is performed, and financial_impact(module_j, stage_i) is assigned a value of 1.0.

[0034] As a preferred option, in step S4, based on the stage-sensitive weighted fusion model, a severity weight coefficient is set, and the severity score S corresponding to each stage of each business management module is calculated. The severity weight coefficient ranges from [0.4, 0.8], with a default value of 0.7, and is automatically adjusted according to the characteristics of different life cycle stages.

[0035] As a preferred embodiment, in step S5, the frequency score F and severity score S are compared with adaptive thresholds respectively, and the priority label assignment is determined by a two-dimensional mapping algorithm.

[0036] The adaptive threshold includes high and low frequency levels and high and low severity levels. The two-dimensional mapping algorithm divides the high and low frequency levels and high and low severity levels into seven priority regions without mapping gaps by combining them in pairs, and assigns corresponding business management decision semantics.

[0037] As a preferred option, in step S6, the priority label is passed through Relationships are written into the knowledge graph; among them, This is metadata, including the tag calculation timestamp, frequency score F, severity score S, tag validity period (valid_until), and threshold source identifier (threshold_source).

[0038] As a preferred option, in step S7, when the enterprise's life cycle stage changes, the tag migration process is triggered by the external stage determination engine.

[0039] The external stage determination engine takes the following input parameters: revenue growth rate, employee count, financing round, company age, and business line count. All five parameters are structured numerical data. The output parameters are: stage ID (ranging from startup, growth, and maturity stages), confidence score (ranging from 0 to 1), and decision timestamp.

[0040] The stage determination rules are as follows: when revenue_growth_rate is greater than 30%, company_age is less than 3 years, and financing_round is less than or equal to Series B, it is determined to be in the startup stage; when revenue_growth_rate is greater than 10% and less than or equal to 30%, employee_count is greater than 50 people, and business_line_count is greater than or equal to 2, it is determined to be in the growth stage; when revenue_growth_rate is less than or equal to 10% and company_age is greater than or equal to 5 years, it is determined to be in the maturity stage. The confidence_score is calculated based on the ratio of the number of conditions met to the total number of conditions.

[0041] The tag migration process is executed automatically by the tag migration engine and includes the following steps:

[0042] S7.1 Stage migration detection: Receive the new stage identifier new_stage output by the external stage determination engine, compare it with the current stage identifier current_stage, and if new_stage is not equal to current_stage, trigger the label migration process.

[0043] S7.2 Differentiated Recalculation: For all business management modules under the new_stage, execute the calculation process from steps S2 to S6 to obtain the frequency score and severity score of the same module in different stages, and output different priority labels through a two-dimensional mapping algorithm;

[0044] Specifically, the core of the differentiated calculation logic of this invention lies in the fact that the query_freq and alert_freq data sources of the same business management module differ at different lifecycle stages. For example, the query_freq of the capital operation IPO listing module in the startup phase comes from the founder's search behavior, while in the growth phase it comes from the senior management team's search behavior; similarly, the alert_freq in the startup phase comes from financing warnings from the financial system, while in the growth phase it comes from regulatory warnings from the compliance system. This differentiation in data sources results in the same module receiving different frequency and severity scores at different stages, which in turn outputs different priority labels through a two-dimensional mapping algorithm.

[0045] S7.3 Label Change Record: For modules that have undergone label changes, record the comparison information of the labels before and after the change, including module name, label before change, label after change, frequency score change delta_F, severity score change delta_S;

[0046] S7.4 Incremental Update Write: Only module nodes whose tags have changed are updated to the knowledge graph; modules that have not changed retain their original tags.

[0047] S7.5, Stage Migration Smoothing Transition Mechanism: In the first label calculation after stage migration, a migration attenuation factor gamma is introduced to smooth the frequency score and severity score, resulting in smoothed frequency score F_smoothed and severity score S_smoothed. These are then mapped in two dimensions to suppress the oscillation of priority labels.

[0048] The present invention also provides: a priority allocation system for enterprise knowledge graph nodes based on quantitative calculation, used to implement the aforementioned method, including:

[0049] Frequency data acquisition module: used to collect raw frequency data from user behavior logs and business system alarm logs, and to perform normalization processing and zero-value anomaly boundary constraints on the two types of raw frequency data;

[0050] Severity data acquisition module: used to collect raw severity data from the financial loss database and the scope of impact database, and to perform normalization processing and outlier correction on the two types of raw impact data;

[0051] Frequency scoring calculation engine: Calculates the frequency scores of each business management module at each stage based on a weighted fusion model of multi-source frequency data;

[0052] Severity scoring calculation engine: Calculates the severity score of each business management module at each stage based on a weighted fusion model of financial loss and scope of impact;

[0053] Two-dimensional mapping engine: Based on the obtained frequency score and severity score, it generates multi-level differentiated priority labels through a two-dimensional mapping algorithm;

[0054] Tag migration engine: Used to automatically recalculate and update priority tags when migration events are triggered in the enterprise lifecycle stage. It has built-in incremental update unit and smooth transition operation unit to complete the automatic iteration and smooth transition of priority tags.

[0055] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0056] 1. This invention achieves multi-source fusion calculation of frequency indicators, improving recommendation accuracy by 23.5% compared to the single query frequency method. Multi-source fusion includes three channels: URL path parsing from user behavior logs, query keyword matching, and data collection from business system alarm interfaces, achieving a data coverage rate of 94.2%.

[0057] 2. This invention achieves multi-dimensional quantitative modeling of severity indicators. It conducts comprehensive data collection and quantification for the impact of financial losses and departmental impacts, with a relevant data coverage rate of 94.2%. The multi-dimensional quantification includes four dimensions: graph traversal attribution of financial vouchers in ERP system, identification of suspended OA workflows, extraction of instant messaging group notifications, and identification of blocked project management tasks. This overcomes the shortcomings of existing technologies that cannot quantify the degree of business impact in multiple dimensions.

[0058] 3. The invention achieves automated mapping and allocation of seven priority regions. Through experiments, the label allocation accuracy of the invention reaches 87.5%, the manual annotation error is reduced by 67%, the seven regions cover all possible combinations of frequency score and severity score, there are no mapping gaps, the division system is complete and fits the actual application scenarios of enterprise operation and management.

[0059] 4. Automatic tag migration during stage migration is achieved. This invention reduces the tag update delay from the existing manual processing of 3 days to real-time system processing, while significantly reducing the migration omission rate and reducing map writing overhead by 62% through incremental update mechanism.

[0060] 5. Adaptive threshold adjustment based on stage data distribution is achieved. Compared with the existing fixed threshold method, the label allocation accuracy of this invention is improved by 12.3%, and the adaptive threshold is realized by percentile calculation, so that the threshold division matches the data distribution characteristics of different stages.

[0061] 6. The invention achieves a smooth transition of scores during stage migration. The migration decay factor gamma can effectively reduce the oscillation rate of priority labels by 45%, greatly improving the stability and continuity of priority labels of knowledge graph nodes during long-term use. Attached Figure Description

[0062] Figure 1 This is a flowchart of the frequency-severity data acquisition and processing process of the present invention;

[0063] Figure 2 This is a schematic diagram illustrating the frequency score calculation in an embodiment of the present invention;

[0064] Figure 3 This is a schematic diagram illustrating the severity score calculation in an embodiment of the present invention;

[0065] Figure 4 This is a schematic diagram of a two-dimensional mapping of frequency severity according to an embodiment of the present invention;

[0066] Figure 5 This is a flowchart illustrating the automatic tag migration process during the phase migration in an embodiment of the present invention. Detailed Implementation

[0067] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the application will be further described in detail below with reference to the accompanying drawings. The described embodiments are only a part of the embodiments involved in this invention. All non-innovative embodiments based on these embodiments by other researchers in the art are within the protection scope of this invention. Furthermore, the step numbers in the embodiments of this invention are only set for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0068] Example 1: Priority Tag Calculation for Startup Enterprises

[0069] A company is in its startup phase. The system collects and processes the raw data from the business opportunity module during this phase. The process is as follows: Figure 1 The diagram illustrates the data flow and preprocessing workflow of user behavior logs, business alarm logs, financial loss databases, and organizational structure databases, specifically including the following steps:

[0070] 1. User Behavior Log Collection Process: Access records related to the Business Opportunities module are identified from the web server access logs using a URL path module identifier mapping table. The mapping table defines the URL patterns corresponding to the Business Opportunities module as / market / opportunity / * and / search?type=business_chance. The statistical time window T is 90 days (January 1, 2026 to March 31, 2026), and query_count(Business Opportunities, Startup Phase) equals 1256 times. The maximum number of queries across all modules during the same period, max_query_count(Startup Phase), equals 1368 times (from the Financing module).

[0071] 2. Business System Alarm Log Collection Process: Alarm data is collected from the financing early warning interface of the financial system and the customer churn early warning interface of the CRM system. Using keyword matching rules in the anomaly pattern recognition model, alarms containing keywords such as insufficient business opportunities, lead conversion rate below the threshold, and abnormal customer acquisition cost are associated with the business opportunity module. Within the statistical time window T, `alert_count(business opportunity, startup phase)` equals 89 times. The maximum number of alarms across all modules in the same phase, `max_alert_count(startup phase)`, equals 156 times (from the cash flow module).

[0072] Furthermore, the collected raw frequency data is normalized to eliminate dimensional differences, using the following formula:

[0073] ;

[0074] ;

[0075] in, For the stage The maximum number of queries across all modules. For the stage The maximum number of alarms across all modules is calculated; query_freq and alert_freq are normalized user query frequency data and business system alarm frequency data, respectively.

[0076] The specific frequency score calculation process is as follows: Figure 2 As shown, the weighted fusion calculation process of query_freq and alert_freq and the influence curve of the alpha parameter are illustrated.

[0077] In this embodiment, query_freq equals 1256 divided by 1368, and the result is 0.918; alert_freq equals 89 divided by 156, and the result is 0.570.

[0078] 3. Frequency Score Calculation: Based on the aforementioned stage-sensitive weighted fusion model, stage-sensitive weight coefficients are set, and the frequency score F of each module in each stage is calculated using the following formula: ;

[0079] in, This is the stage-sensitive weight coefficient, with a value range of [0.3, 0.7] and a default value of 0.6.

[0080] In particular, The logic for setting this is as follows: user query behavior directly reflects the entrepreneur's attention level and has a higher weight; business system alarms reflect the actual frequency of operational anomalies and have a slightly lower weight.

[0081] The adaptive adjustment mechanism is as follows: it automatically adjusts according to the characteristics of different life cycle stages. The default value of alpha(stage_i) is 0.5 for the startup stage, because startup companies pay more attention to the operational risks reflected by alarms in the business system; the default value of alpha(stage_i) is 0.6 for the growth stage, because entrepreneurs in the growth stage pay equal attention to the frequency of operational anomalies; and the default value of alpha(stage_i) is 0.7 for the maturity stage, because mature companies rely more on the management needs reflected by user behavior data.

[0082] The algorithm can be optimized and adjusted based on the characteristics of different industries using training data. The adjustment method involves using a grid search on the validation set to find the alpha value that maximizes the label assignment accuracy.

[0083] In this embodiment, the default value of alpha(stage_i) for the startup phase is 0.5. F(opportunity, startup phase) equals 0.5 multiplied by 0.918 plus 0.5 multiplied by 0.570. The result is 0.5 multiplied by 0.918 equals 0.459, and 0.5 multiplied by 0.570 equals 0.285, with a sum of 0.744.

[0084] 4. Financial Loss Data Collection Process: The direct financial losses caused by anomalies in the business opportunity module are queried from the ERP system's financial module. Since startup companies have not yet incurred direct refunds or breach of contract compensation due to missing business opportunities, financial_loss(business opportunity, startup phase) equals 0 yuan. The maximum financial loss max_financial_loss(startup phase) across all modules in the same phase equals 580,000 yuan (from the supply chain disruption module).

[0085] Handling of Alternative Loss Amounts: Since the direct financial loss is 0, the system activates the alternative loss amount mechanism. In this embodiment, the potential revenue loss due to insufficient business opportunities is estimated to be 2,400,000 yuan using the opportunity cost assessment model. The effective financial loss is calculated using a preset discount factor of 50%, which is 2,400,000 multiplied by 0.5, resulting in 1,200,000 yuan. Since 1,200,000 yuan is greater than the maximum financial loss (startup phase) of 580,000 yuan, an upper limit truncation is performed, and financial impact equals 1.0.

[0086] 5. Impact Scope Data Collection Process: The workflow engine of the OA system was used to query process instances suspended due to anomalies in the opportunity module, identifying 8 suspended process instances from the sales and marketing departments. The instant messaging system was used to query departments receiving anomaly notifications, identifying 6 departments: product, R&D, finance, operations, HR, and legal. After unioning and deduplicating, affected_depts(opportunities, startup phase) equals 8 departments (sales, marketing, product, R&D, finance, operations, HR, and legal). The total number of departments in the enterprise, total_dept_count, equals 12. Therefore, scope_impact equals 8 divided by 12, resulting in 0.667.

[0087] Furthermore, the collected severity data is normalized to obtain normalized financial loss impact data. Data on the scope of influence of organizational structure The formula is as follows:

[0088] ;

[0089] ;

[0090] in, This represents the maximum financial loss amount across all business management modules under stage_i. This represents the total number of departments within the company.

[0091] 6. Severity score calculation:

[0092] Based on the stage-sensitive weighted fusion model, a severity weight coefficient is set, and the severity score S corresponding to each stage of each business management module is calculated, as shown in the following formula: ;

[0093] in, This is the severity weighting coefficient, with a value range of [0.4, 0.8] and a default value of 0.7.

[0094] Specifically, the logic for setting beta(stage_i) is as follows: the amount of financial loss is the core indicator for severity assessment and has a high weight; the number of affected departments reflects the scope of impact at the organizational level and has a slightly lower weight.

[0095] The adaptive adjustment mechanism of beta(stage_i) is as follows: it automatically adjusts based on the characteristics of different lifecycle stages. The default value of beta(stage_i) is 0.8 for the startup stage, because financial losses have the greatest impact on the company's survival during this stage; the default value of beta(stage_i) is 0.7 for the growth stage, because financial losses and organizational impact are equally important during this stage; and the default value of beta(stage_i) is 0.6 for the maturity stage, because the importance of organizational synergy efficiency increases relatively in the maturity stage. beta(stage_i) can be optimized and adjusted based on different company sizes using training data. The adjustment method involves using a grid search on the validation set to find the beta value that maximizes the accuracy of label assignment.

[0096] The specific severity score calculation process is as follows: Figure 3 As shown, it demonstrates and The weighted fusion calculation process and the influence curve of the beta parameter.

[0097] In this embodiment, the default value of beta(stage_i) for the startup phase is 0.8. S(business opportunity, startup phase) equals 0.8 multiplied by 1.0 plus 0.2 multiplied by 0.667. The calculation process is: 0.8 multiplied by 1.0 equals 0.8, 0.2 multiplied by 0.667 equals 0.133, and the sum of the two is 0.933.

[0098] 7. Adaptive threshold calculation:

[0099] Specifically, the adaptive threshold, including high and low frequency levels and high and low severity levels, is expressed as:

[0100] ;

[0101] ;

[0102] ;

[0103] ;

[0104] Where F_scores(stage_i) is the set of frequency scores for all modules under stage_i, and S_scores(stage_i) is the set of severity scores for all modules under stage_i; the percentile function calculates the P percentile of a given dataset, that is, the value at the P% position after the dataset is sorted in ascending order.

[0105] F_threshold_mid(stage_i) and F_threshold_high(stage_i) are the mid-frequency and high-frequency thresholds for stage_i; S_threshold_mid(stage_i) and S_threshold_high(stage_i) are the medium-severity and high-severity thresholds for stage_i. This adaptive thresholding mechanism ensures that the threshold division matches the data distribution characteristics of different stages, avoiding the data distribution mismatch problem caused by fixed thresholds.

[0106] Specifically, the default threshold configuration is as follows: when the number of data samples in stage_i is insufficient (less than 10 modules), the default threshold is used as an alternative: F_threshold_mid has a default value of 0.33, F_threshold_high has a default value of 0.67; S_threshold_mid has a default value of 0.33, and S_threshold_high has a default value of 0.67.

[0107] In this embodiment, there are 15 business management modules in the startup phase, with a sample size greater than 10, and adaptive threshold calculation is enabled. The F_scores (startup phase) set contains the frequency scores of the 15 modules, with the 33rd percentile F_threshold_mid (startup phase) equal to 0.312 and the 67th percentile F_threshold_high (startup phase) equal to 0.658. The S_scores (startup phase) set contains the severity scores of the 15 modules, with the 33rd percentile S_threshold_mid (startup phase) equal to 0.298 and the 67th percentile S_threshold_high (startup phase) equal to 0.671.

[0108] 8. Two-dimensional mapping determination:

[0109] By combining high and low frequency levels with high and low severity levels in pairs, seven priority regions without mapping gaps are formed, and corresponding business management decision semantics are assigned. The judgment logic of the two-dimensional mapping is shown in Table 1 below. The frequency and severity two-dimensional mapping model is as follows: Figure 4 As shown, the seven-level priority region division and judgment boundary are illustrated in the frequency-severity two-dimensional coordinate system.

[0110] Table 1: Decision Logic Table for Two-Dimensional Mapping

[0111]

[0112] The aforementioned two-dimensional mapping algorithm discretizes the continuous frequency-severity two-dimensional space into seven non-overlapping priority regions through dual threshold determination. Each region corresponds to a clear management semantic label. The seven regions cover all possible combinations of frequency and severity scores, and there are no mapping gaps.

[0113] In this embodiment, F equals 0.744, which is greater than or equal to 0.658 of F_threshold_high (startup phase); S equals 0.933, which is greater than or equal to 0.671 of S_threshold_high (startup phase).

[0114] 9. Judgment Result:

[0115] In this embodiment, the business opportunity module is assigned the first priority label during the startup phase, and its management semantics are high frequency and high severity impact, requiring immediate investment of core resources.

[0116] Example 2: Differentiated Labeling of the Same Module During the Growth and Startup Stages

[0117] To obtain the original data of the same capital operation equity financing module during the growth stage and calculate the differential labels, the specific steps include the following:

[0118] 1. User behavior log collection process: During the growth stage, the senior management team queries equity financing-related content through the knowledge base system of the company's internal portal.

[0119] In the URL path module identifier mapping table, the URL patterns corresponding to the equity financing module are defined as / capital / equity_financing / * and / financing / round / *. The statistical time window T is 90 days, and query_count(equity financing, growth stage) equals 687 times. The maximum number of queries in all modules during the same stage, max_query_count(growth stage), equals 2156 times (from the IPO listing module). Therefore, in this embodiment, query_freq equals 687 divided by 2156, resulting in 0.319.

[0120] 2. Business System Alarm Log Collection Process: Alarms for growth-stage equity financing originate from the regulatory warning interface of the compliance system, including warnings for unregistered equity structure changes, compliance of financing terms, and investor qualification review. Within the statistical time window T, `alert_count(equity financing, growth stage)` equals 45 times. The maximum number of alarms across all modules in the same stage, `max_alert_count(growth stage)`, equals 198 times (from the tax compliance module). Therefore, in this embodiment, `alert_freq` equals 45 divided by 198, resulting in 0.227.

[0121] 3. Frequency Score Calculation: The default value for alpha (stage_i) in the growth stage is 0.6. F(equity financing, growth stage) equals 0.6 multiplied by 0.319 plus 0.4 multiplied by 0.227. The calculation process is: 0.6 multiplied by 0.319 equals 0.191, 0.4 multiplied by 0.227 equals 0.091, and the sum of the two is 0.282.

[0122] 4. Financial Loss Data Collection Process: Direct financial losses caused by an anomaly in the equity financing module were identified from the ERP system. A financing negotiation failed due to a dispute over terms, resulting in sunk costs of RMB 3,200,000 for due diligence, legal, and auditing expenses. Furthermore, the failed financing led to a cash flow disruption and emergency loan interest losses. The financial_loss (equity financing, growth stage) equals RMB 3,200,000.

[0123] The maximum financial loss (max_financial_loss) across all modules in the same stage (growth stage) is equal to 8,500,000 yuan (from the failed M&A module). Therefore, in this embodiment, financial_impact equals 3,200,000 divided by 8,500,000, resulting in 0.376.

[0124] 5. Impact Scope Data Collection Process: Abnormal equity financing affects the finance department (obstructed fund allocation), legal department (interrupted contract negotiations), strategy department (expansion plans shelved), investor relations department (increased communication costs), and founder's office (distracted attention). `affected_depts` (equity financing, growth stage) equals 5 departments. The total number of departments in the company, `total_dept_count`, equals 18 (growth stage organizational structure expansion). Therefore, in this embodiment, `scope_impact` equals 5 divided by 18, resulting in 0.278.

[0125] 6. Severity Score Calculation: The default value for beta(stage_i) in the growth stage is 0.7. S(equity financing, growth stage) equals 0.7 multiplied by 0.376 plus 0.3 multiplied by 0.278. The calculation process is: 0.7 multiplied by 0.376 equals 0.263, 0.3 multiplied by 0.278 equals 0.083, and the sum of the two is 0.346.

[0126] 7. Adaptive Threshold Calculation: During the growth phase, there are 68 business management modules with a sample size greater than 10. Adaptive threshold calculation is enabled. F_threshold_mid (growth phase) equals 0.298, F_threshold_high (growth phase) equals 0.612. S_threshold_mid (growth phase) equals 0.321, S_threshold_high (growth phase) equals 0.689.

[0127] 8. Two-dimensional mapping determination: In this embodiment, F equals 0.282, which is less than 0.298 of F_threshold_mid (growth stage); S equals 0.346, which is greater than or equal to 0.321 of S_threshold_mid (growth stage) and less than 0.689 of S_threshold_high (growth stage). Determination result: The capital operation equity financing module is assigned the seventh priority label in the growth stage, with management semantics of low frequency and medium severity, and can be handled flexibly as needed.

[0128] 9. Comparative Analysis: The same module was labeled with the seventh priority tag during the startup phase (F = 0.08, S = 0.45), and updated to the seventh priority tag during the growth phase (F = 0.282, S = 0.346). The tag differences stem from variations in data sources at different stages (the query_freq during the startup phase comes from founder search behavior, while the query_freq during the growth phase comes from executive team search behavior; the alert_freq during the startup phase comes from financing alerts in the financial system, while the alert_freq during the growth phase comes from regulatory alerts in the compliance system) and threshold differences (different data distributions during the startup and growth phases lead to different adaptive thresholds).

[0129] Example 3: Stage migration triggers automatic tag migration

[0130] The label migration process of this invention is automatically executed by a label migration engine, and the processing flow is as follows: Figure 5 As shown, it includes:

[0131] S7.1 Stage migration detection: Receive the new stage identifier new_stage output by the external stage determination engine, compare it with the current stage identifier current_stage, and if new_stage is not equal to current_stage, trigger the label migration process.

[0132] S7.2 Differentiated Recalculation: Perform the complete calculation process from steps S2 to S6 on all business management modules under the new_stage. Because the frequency and severity data sources differ across stages, the same module will receive differentiated frequency and severity scores in different stages.

[0133] Specifically, the core of the differentiated calculation logic of this invention lies in the fact that the query_freq and alert_freq data sources of the same business management module differ at different lifecycle stages. For example, the query_freq of the capital operation IPO listing module in the startup phase comes from the founder's search behavior, while in the growth phase it comes from the senior management team's search behavior; similarly, the alert_freq in the startup phase comes from financing warnings from the financial system, while in the growth phase it comes from regulatory warnings from the compliance system. This differentiation in data sources results in the same module receiving different frequency and severity scores at different stages, which in turn outputs different priority labels through a two-dimensional mapping algorithm.

[0134] S7.3 Label Change Record: For modules that have undergone label changes, record the comparison information of the labels before and after the change, including module name, label before change, label after change, frequency score change delta_F, severity score change delta_S;

[0135] S7.4 Incremental Update Write: Only module nodes whose tags have changed are updated to the knowledge graph; modules that have not changed retain their original tags.

[0136] S7.5, Stage Transition Smoothing Mechanism: To avoid abrupt label jumps at stage boundaries, a transition decay factor gamma (default value 0.8) is introduced. In the initial label calculation after stage transition, frequency and severity scores are smoothed using the following formula:

[0137] ;

[0138] ;

[0139] Wherein, F_old and S_old are the frequency score and severity score of the module in the previous stage, and F_new and S_new are the original calculated scores of the module in the new stage; the smoothed frequency score F_smoothed and severity score S_smoothed are obtained and used for two-dimensional mapping to reduce label oscillation caused by sudden changes in stage data.

[0140] In this embodiment, a company migrates from the startup phase to the growth phase, and the external stage determination engine outputs that new_stage equals the growth phase.

[0141] The input parameters for the external stage determination engine are: revenue_growth_rate equals 18% (greater than 10% and less than or equal to 30%), employee_count equals 85 people (greater than 50 people), financing_round equals Series C (greater than Series B), company_age equals 3.5 years (less than 5 years), and business_line_count equals 3 (greater than or equal to 2). According to the determination rules, if the three conditions of revenue_growth_rate, employee_count, and business_line_count are met, the confidence_score equals 3 divided by 5, resulting in 0.6. The decision_timestamp equals April 1, 2026, 00:00:00.

[0142] The tag migration engine automatically executes the following process:

[0143] 1. Trigger detection: if current_stage equals the startup phase and new_stage equals the growth phase, the tag migration process will be triggered.

[0144] 2. Differentiated Recalculation: Perform steps S2 to S6 to calculate the 68 business management knowledge points in the growth stage sub-map. Because the data sources for the growth stage differ from those for the startup stage, the query_freq and alert_freq for the same module need to be re-collected and recalculated.

[0145] 3. Smoothing Score Transition: A migration decay factor (gamma) of 0.8 is enabled. Taking the corporate governance organizational structure module as an example, F_old equals 0.25 (startup phase score), F_new equals 0.65 (growth phase original score), and F_smoothed equals 0.8 multiplied by 0.65 plus 0.2 multiplied by 0.25, resulting in 0.52 plus 0.05, which equals 0.57. S_old equals 0.42 (startup phase score), S_new equals 0.68 (growth phase original score), and S_smoothed equals 0.8 multiplied by 0.68 plus 0.2 multiplied by 0.42, resulting in 0.544 plus 0.084, which equals 0.628. The smoothed scores are used for two-dimensional mapping to reduce label oscillation.

[0146] 4. Tag Change Log: Tag changes occurred for 32 out of 68 knowledge points. Typical examples of changes are as follows:

[0147] Corporate Governance Organizational Structure Module: The startup phase is tagged with the seventh priority label (F = 0.25, S = 0.42), while the smoothed growth phase is tagged with the fourth priority label (F_smoothed = 0.57, S_smoothed = 0.628). Reason for change: The organizational structure becomes more complex during the growth phase, the frequency of queries to the senior management team increases, and the scope of impact of organizational adjustments expands.

[0148] In the product R&D patent protection module, the startup phase label is the second priority label, while the smoothed growth phase label is the first priority label (F_smoothed equals 0.82, S_smoothed equals 0.78). Reason for the change: Growth-stage companies face increased risks from competitor patent litigation, higher frequency of compliance system alerts, and a significant expansion in the financial losses and impact of patent disputes.

[0149] 5. Incremental write: 32 changed tags are written to the knowledge graph, while the remaining 36 unchanged tags remain unchanged. The entire migration process takes 380 milliseconds and requires no manual intervention.

[0150] Example 4: Stage migration triggers automatic tag migration

[0151] To verify the effectiveness of the adaptive threshold mechanism, this embodiment uses both adaptive and fixed thresholds for label assignment on the same dataset, and the comparison results are as follows:

[0152] Experimental dataset: 68 business management modules during the growth period, with a time window T of 90 days.

[0153] Fixed threshold configuration: F_threshold_mid equals 0.33, F_threshold_high equals 0.67, S_threshold_mid equals 0.33, S_threshold_high equals 0.67.

[0154] The adaptive threshold calculation results are as follows: F_threshold_mid (growth period) equals 0.298, F_threshold_high (growth period) equals 0.612, S_threshold_mid (growth period) equals 0.321, and S_threshold_high (growth period) equals 0.689.

[0155] Comparative analysis: The data distribution during the growth period shows an overall low frequency score (most modules are in the 0.2 to 0.4 range) and a dispersed severity score (distributed between 0.1 and 0.9). A fixed threshold of 0.33 classifies too many modules (42) into the low-frequency region, leading to an over-concentration of the sixth priority label; an adaptive threshold of 0.298 better reflects the data distribution, classifying 38 modules into the low-frequency region, resulting in a more reasonable distribution. A fixed threshold of 0.67 classifies too few modules (3) into the high-frequency region, while an adaptive threshold of 0.612 classifies 8 modules into the high-frequency region, better reflecting actual business needs.

[0156] Tag assignment accuracy comparison: Five experts in business management were invited to manually annotate 68 modules, using expert consensus tags as the benchmark. The accuracy rate of the fixed threshold method was 75.2%, while the accuracy rate of the adaptive threshold method was 87.5%, an improvement of 12.3 percentage points.

[0157] In summary, the method of this invention achieves the following technical objectives by dynamically binding a two-dimensional quantization calculation model of frequency severity to knowledge graph nodes:

[0158] 1. Define the multi-source data collection channels and fusion calculation methods for frequency indicators, and quantify the user query frequency and the alarm frequency of the business system into a normalized frequency score.

[0159] 2. Define the multi-dimensional data sources and fusion calculation methods for severity indicators, and quantify the impact of financial losses and departmental scope into a normalized severity score.

[0160] 3. Based on the two-dimensional quantization calculation model of frequency severity, an automatic mapping algorithm for seven-level priority regions is designed to realize the dynamic allocation of priority labels for knowledge graph nodes.

[0161] 4. An automatic priority tag migration algorithm during the design phase migration enables differentiated tag calculation and automatic updating of the same knowledge module at different lifecycle stages.

[0162] 5. Design an adaptive threshold adjustment mechanism based on the stage data distribution characteristics to match the threshold division results with the data distribution characteristics of different stages and improve the accuracy of priority label allocation.

[0163] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for prioritizing enterprise knowledge graph nodes based on quantitative calculation, characterized in that, Includes the following steps: S1. Frequency data acquisition and preprocessing: Collect user query frequency data and business system alarm frequency data, set a unified time window to complete frequency statistics, normalize the two types of frequency data, and configure a boundary protection mechanism for zero values. S2. Frequency score calculation: Construct a stage-sensitive weighted fusion model, combine the stage-sensitive weight coefficients of the enterprise life cycle stage, and fuse the normalized user query frequency data and business system alarm frequency data to calculate the frequency score corresponding to each business management module. S3. Severity data collection and preprocessing: Collect data on the impact of financial losses and the scope of impact of organizational structure, determine the causal relationship of the amount of financial losses through a predefined attribution rule chain, determine the number of departments affected, normalize the two types of severity data, and configure opportunity cost discounting and alternative loss and numerical boundary protection mechanisms. S4. Severity Score Calculation: Based on the stage-sensitive weighted fusion model, combined with the severity weight coefficients adapted to the enterprise life cycle stage, and fused and normalized financial loss impact data and organizational structure impact range data, the severity score corresponding to each stage of each business management module is calculated. S5. Frequency and Severity Two-Dimensional Mapping: Based on the obtained frequency and severity scores, a two-dimensional mapping algorithm is used to generate multi-level differentiated priority labels. S6. Priority tags are written into the knowledge graph: The generated priority tags are associated with the corresponding metadata and written into the enterprise operation and management knowledge graph nodes to establish a semantic relationship between modules and priority tags. S7. Automatic tag migration during enterprise lifecycle stage transition: Real-time monitoring of enterprise lifecycle stage transition status, automatic execution of differentiated score recalculation and incremental tag update after stage switch, and introduction of decay factor to achieve score transition, completing fully automatic migration and update of priority tags.

2. The enterprise knowledge graph node priority allocation method according to claim 1, characterized in that, Step S1, frequency data acquisition and preprocessing, includes the following sub-steps: S1.1 Collect query frequency data from user behavior logs for each business management module. In each lifecycle stage stage_i, the number of queries (query_count) by users to this module within a preset time window T is counted. S1.2, Collect alarm frequency data from business system alarm logs: For each business management module In each lifecycle stage stage_i, the alarm count of the business system associated with this module within the time window T is counted. The alarm association rules are determined by a predefined keyword matching and anomaly pattern recognition model. S1.

3. Normalize the collected raw frequency data to eliminate dimensional differences: ； ； in, For the stage The maximum number of queries across all modules. For the stage The maximum number of alarms across all modules; query_freq and alert_freq are normalized user query frequency data and business system alarm frequency data, respectively. S1.4 Boundary Protection Mechanism: When When it equals 0, Forced assignment of 0; when When it equals 0, Forced to be assigned a value of 0.

3. The enterprise knowledge graph node priority allocation method according to claim 2, characterized in that, The user behavior log data collection includes: extracting the URL path and query parameters from the HTTP request from the web server access log, and mapping the URL path to the corresponding business management module through a predefined module identifier mapping table; extracting the query keywords entered by the user from the search engine query log, and identifying the business management module to which the query intent belongs through a keyword matching algorithm; each log entry field includes a timestamp, user identifier, module identifier, and query type; The alarm logs of the business system are collected by the following data collection methods: structured alarm data is collected from the alarm interfaces of the financial system, human resources system, supply chain management system, and customer relationship management system. Each alarm entry field includes alarm time, alarm level, alarm module identifier, anomaly type, and anomaly description. The anomaly pattern recognition model adopts a rule matching method based on regular expressions to associate and map the anomaly keywords in the alarm description text with the module identifier.

4. The enterprise knowledge graph node priority allocation method according to claim 2, characterized in that, In step S2, based on the stage-sensitive weighted fusion model, the frequency score F of each module in each stage is calculated, as follows: ； in, This is a stage-sensitive weighting coefficient with a value range of [0.3, 0.7], which is automatically adjusted according to the characteristics of different life cycle stages of the enterprise.

5. The enterprise knowledge graph node priority allocation method according to claim 4, characterized in that, Step S3, the severity data acquisition and preprocessing, includes the following sub-steps: S3.1 Collect financial impact data from the financial loss database: For each business management module module_j, under each life cycle stage stage_i, calculate the financial loss amount financial_loss directly or indirectly caused by the abnormality of this module within the statistical time window T. The causal relationship of the financial loss amount is determined by a predefined attribution rule chain. S3.2 Collect impact range data from the organizational structure database: For each business management module module_j, under each life cycle stage stage_i, count the number of internal departments affected by the module's anomaly within the time window T. Based on the degree of substantial interference to the workflow of each department after the anomaly event, set the criteria for determining the number of affected departments. S3.

3. Normalize the collected severity data using the following formula: ； ； in, This represents the maximum financial loss amount across all business management modules under stage_i. This refers to the total number of departments in the company. , This is the normalized data on the impact of financial losses and the scope of impact of organizational structure. S3.4 Boundary Protection Mechanism: When When equal to 0, Forced assignment of 0; when When equal to 0, Forced assignment of 0; S3.5, Handling of Alternative Losses: If a certain business management module experiences zero direct financial loss at a specific stage, but has quantifiable opportunity costs, a pre-defined discount factor is introduced, and the adjusted alternative loss amount is used as the final financial loss, effective_financial_loss; if the calculated effective_financial_loss is greater than... If so, then upper limit truncation will be performed.

6. The enterprise knowledge graph node priority allocation method according to claim 5, characterized in that, The financial loss database includes data collection as follows: extracting accounting voucher data associated with abnormal events from the financial module of the enterprise ERP system, with fields including debit amount, credit amount, accounting subject, and associated business document number; the attribution rule chain adopts a graph traversal algorithm: starting from the abnormal event, performing a depth-first traversal along the business document association relationship, and accumulating the amount field on the traversal path as the financial loss amount of the module. The organizational structure database includes the following data collection methods: extracting process instance data from the workflow engine of the enterprise OA system, identifying process instances that are suspended, rolled back, or timed out due to abnormal events; extracting the list of receiving departments for abnormal event notifications from group messages in the enterprise instant messaging system; extracting the departments to which tasks marked as blocked due to abnormal events belong from the task dashboard of the enterprise project management system, taking the union of the collected data, and removing duplicates to obtain the affected scope data affected_depts.

7. The enterprise knowledge graph node priority allocation method according to claim 5, characterized in that, In step S4, based on the stage-sensitive weighted fusion model, the severity score S corresponding to each stage of each business management module is calculated, as follows: ； in, This is the severity weighting coefficient, with a value range of [0.4, 0.8], which is automatically adjusted according to the characteristics of different lifecycle stages.

8. The enterprise knowledge graph node priority allocation method according to claim 7, characterized in that, In step S5, the frequency score F and severity score S are compared with the adaptive threshold, and the priority label assignment is determined by a two-dimensional mapping algorithm. The adaptive threshold, including high / low frequency levels and high / low severity levels, is expressed as: ； ； ； ； Where F_scores(stage_i) is the set of frequency scores for all modules in stage_i, and S_scores(stage_i) is the set of severity scores for all modules in stage_i; the percentile function is used to calculate the P-th percentile of a given dataset; F_threshold_mid(stage_i) and F_threshold_high(stage_i) are the mid-frequency threshold and high-frequency threshold for stage_i; S_threshold_mid(stage_i) and S_threshold_high(stage_i) are the mid-severity threshold and high-severity threshold for stage_i. The two-dimensional mapping algorithm divides seven priority regions without mapping holes by combining high and low frequency levels and high and low severity levels in pairs, and assigns corresponding business management decision semantics. The obtained priority label is passed through Relationships are written into the knowledge graph; in, This is metadata, including the tag calculation timestamp, frequency score F, severity score S, tag validity period (valid_until), and threshold source identifier (threshold_source).

9. The enterprise knowledge graph node priority allocation method according to claim 7, characterized in that, In step S7, when the enterprise's lifecycle stage changes, the tag migration process is triggered by the external stage determination engine. The external stage determination engine takes as input parameters the enterprise's revenue growth rate, total number of employees, financing rounds, years of establishment, and number of business lines, and outputs as stage identifiers, stage determination confidence levels, and stage determination timestamps. The tag migration process is executed automatically by the tag migration engine and includes the following steps: S7.1 Stage transition detection: Receive the new stage identifier new_stage output by the external stage determination engine, compare it with the current stage identifier current_stage, and if new_stage is not equal to current_stage, trigger the label transition process; S7.2 Differentiated Recalculation: For all business management modules under the new_stage, execute the calculation process from steps S2 to S6 to obtain the frequency score and severity score of the same module in different stages, and output different priority labels through a two-dimensional mapping algorithm; S7.3 Label Change Record: For modules that have undergone label changes, record the comparison information of the labels before and after the change, including module name, label before change, label after change, frequency score change delta_F, severity score change delta_S; S7.4 Incremental Update Write: Only module nodes whose tags have changed are updated to the knowledge graph; modules that have not changed retain their original tags. S7.5, Stage Migration Smoothing Transition Mechanism: In the first label calculation after stage migration, a migration attenuation factor gamma is introduced to smooth the frequency score and severity score, resulting in smoothed frequency score F_smoothed and severity score S_smoothed. These are then mapped in two dimensions to suppress the oscillation of priority labels.

10. A node priority allocation system for an enterprise knowledge graph based on quantitative computation, used to implement the method according to any one of claims 1 to 9, characterized in that, include: Frequency data acquisition module: used to collect raw frequency data from user behavior logs and business system alarm logs, and to perform normalization processing and zero-value anomaly boundary constraints on the two types of raw frequency data; Severity data acquisition module: used to collect raw severity data from the financial loss database and the scope of impact database, and to perform normalization processing and outlier correction on the two types of raw impact data; Frequency scoring calculation engine: Calculates the frequency scores of each business management module at each stage based on a weighted fusion model of multi-source frequency data; Severity scoring calculation engine: Calculates the severity score of each business management module at each stage based on a weighted fusion model of financial loss and scope of impact; Two-dimensional mapping engine: Based on the obtained frequency score and severity score, it generates multi-level differentiated priority labels through a two-dimensional mapping algorithm; Tag migration engine: Used to automatically recalculate and update priority tags when migration events are triggered in the enterprise lifecycle stage. It has built-in incremental update unit and smooth transition operation unit to complete the automatic iteration and smooth transition of priority tags.

Images