Enterprise talent database management method based on knowledge graph

Abstract

The application discloses a knowledge graph-based enterprise talent database management method, and relates to the technical field of data processing; the method comprises the following steps: collecting talent data from various business systems, calculating the confidence coefficient of data records according to the source credibility, integrity and consistency, screening the data according to the confidence coefficient to obtain confidence data, performing entity, attribute and relationship extraction on the confidence data, and constructing a global talent knowledge graph; identifying core nodes based on the relationship edge weight and node importance; for talent query, post matching or project establishment requirements, performing semantic analysis on the requirements and mapping them into target entity nodes to construct a talent demand graph; scoring the support paths between the candidate nodes and the target entity nodes, calculating the comprehensive demand index of the candidate nodes, and outputting the recommended results. The method can realize the unified organization, semantic association, dynamic analysis and intelligent decision of enterprise talent data, and improve the talent matching accuracy and result interpretability.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a knowledge graph-based method for managing enterprise talent databases. Background Technology

[0002] As enterprises expand and their organizational structures, job systems, and talent mobility mechanisms become more complex, they need to continuously collect, integrate, and analyze multi-dimensional data on employees, including basic information, educational background, skills, project experience, training records, performance, and employment history, in their management activities such as recruitment, training, job transfer, succession, and project allocation. Currently, enterprises typically use relational databases or traditional human resources management systems to store and manage this data. These systems primarily focus on the static registration and basic querying of personnel information. Their data organization method stores various types of information independently in the form of table fields, making it difficult to uniformly model the multi-dimensional relationships between employees and entities such as positions, projects, skills, certificates, training, and performance. This data fragmentation and insufficient semantic expression capabilities prevent existing systems from accurately reflecting the formation process of talent capabilities, job fit relationships, and individual growth trajectories. Furthermore, in decision-making scenarios such as job matching, talent inventory, successor screening, and identification of trainees, existing technologies often rely on human experience or simple rule-based screening, making it difficult to comprehensively consider the semantic relationships and dynamic changes between multi-source heterogeneous data. This limits the accuracy of talent identification results, and related management decisions lack sufficient quantitative data support. Summary of the Invention

[0003] Therefore, it is necessary to provide a knowledge graph-based enterprise talent database management method to address the problems mentioned in the background technology.

[0004] The objective of this invention can be achieved through the following technical solutions: The knowledge graph-based enterprise talent database management method includes the following steps: Step 100: Collect talent data from various business systems, standardize and preliminarily clean the collected talent data, and calculate the confidence coefficient for each data point. Step 200: Filter the data bars according to their confidence coefficients to obtain confidence data. Extract entities, attributes, and relationships from the filtered confidence data. Define relationship edges and nodes based on this and construct a global talent knowledge graph. Step 300: Assign edge weights to each relation edge in the global talent knowledge graph, calculate the importance of nodes based on the edge weights, and identify core nodes; Step 400: Analyze the talent needs to construct a talent demand graph from the core nodes in the global talent knowledge graph, and output the demand results.

[0005] In some embodiments, calculating the confidence coefficient for each data point includes: The source credibility coefficient is determined based on the type of data source. The source credibility coefficient is used to characterize the reliability of the business system corresponding to the data. The integrity coefficient is calculated based on the completeness of the data fields. The integrity coefficient is used to characterize whether the key fields of the data are complete. The consistency coefficient is calculated based on the degree of consistency of corresponding fields of the same data subject in different business systems. The consistency coefficient is used to characterize the degree of consistency of data across systems. The confidence coefficient is obtained by linearly weighting and fusing the source credibility coefficient, integrity coefficient, and consistency coefficient according to preset weights. The sum of the weights corresponding to the source credibility coefficient, integrity coefficient, and consistency coefficient is one, thus obtaining the confidence coefficient of all data.

[0006] In some embodiments, calculating the integrity coefficient based on the completeness of the data field includes: For different types of talent data, a set of key fields is pre-established. The actual number of key fields filled in each data point and the total number of key fields are counted. The completeness coefficient is obtained by dividing the actual number of key fields filled in by the total number of key fields.

[0007] In some embodiments, calculating the consistency coefficient based on the consistency of corresponding fields of the same data subject in different business systems includes: A set of fields that may appear in different business systems and have cross-validation significance for the same data subject is identified. After standardizing each field in the set, the number of systems in which the same field has the same standardized value is counted in each business system participating in the comparison. The ratio of this number to the total number of systems participating in the comparison is used as the consistency degree of the field. The consistency degrees of each field are then weighted and summed according to preset weights to obtain the consistency coefficient of the data bar.

[0008] In some embodiments, the extraction of entities, attributes, and relationships from the filtered confidence data, the definition of relationship edges and nodes based on this extraction, and the construction of a global talent knowledge graph include: A talent knowledge ontology is constructed based on the business semantics of enterprise talent management, and the data bars are extracted into entities, attributes, and relationships. Entities include at least employees, departments, positions, projects, courses, certificates, skills, performance events, training events, reward and punishment events, mentors, and teams. Attributes include at least time, level, score, duration, frequency, validity period, and level. Relationships include at least being employed in, participating in, mastering, acquiring, completing, being assessed, being trained in, collaborating in, being promoted in, being transferred to, and being mentored in. For structured fields in the data bars, they are converted into entities and relationships according to field mapping rules. For unstructured fields in the data bars, talent elements are extracted through named entity recognition and relationship extraction models and converted into entities and relationships. The extracted entities are defined as nodes in the knowledge graph, the relationships between entities are defined as edges between nodes, and attributes are used as attribute information for corresponding nodes or relationship edges. A global talent knowledge graph is constructed based on nodes and relationship edges.

[0009] In some embodiments, assigning edge weights to each relation edge in the global talent knowledge graph includes: For each relation edge in the global talent knowledge graph, extract the entities at both ends of the edge. Based on the entities extracted from each data bar, use name standardization, alias merging, semantic disambiguation, and primary key association to merge the data bars pointing to the entities at both ends into supporting data bars for the same relation edge, thus obtaining the supporting data bars for all relation edges. Extract the generation time of each supporting data bar, calculate the time interval by the time difference between the current time and the generation time, and then calculate the timeliness factor d using an exponential decay method. The calculation formula is as follows:

[0010] Where Δt is the time interval, and τ is a preset attenuation constant, the value of which is determined according to the relationship type. The more stable the relationship, the smaller the value of τ; the more dynamic the relationship, the larger the value of τ. In this embodiment, the attenuation constant for the skill relationship type is 0.3, the attenuation constant for the project relationship type is 0.5, the attenuation constant for the certificate relationship type is 0.4, the attenuation constant for the training relationship type is 0.4, and the attenuation constant for the appointment or reporting relationship type is 0.05. The confidence coefficient α corresponding to the supporting data bar of the relation edge. i Time factor d i The weight of the relation edge is calculated using the following formula, where i represents the index of the supporting data strip of the relation edge.

[0011] Wherein, μ is the repeated support enhancement coefficient, used to control the degree of enhancement of the edge weight of the relation by the number of supporting data points. μ is usually set to 0, 1-0.5, and is set to 0.3 in this embodiment. By setting an enhancement factor in the form of ln(1+n), the edge weight can be monotonically increased as the number of supporting data points increases, while avoiding the infinite expansion of edge weight due to linear accumulation. Thus, the edge weight of each relation edge can be obtained and filled into the corresponding position of the global talent knowledge graph. It should be noted that the edge weight of the relation edge reflects the credibility, timeliness and validity of a single supporting data point, as well as the repeated confirmation strength of multiple supporting data points, thereby ensuring that the edge weight increases with the increase of the number of supporting data points and avoiding the distortion of edge weight due to simple linear accumulation.

[0012] In some embodiments, calculating node importance based on edge weights and identifying core nodes includes: For any node, the weights of the relation edges connected to that node are counted, forming a relation edge set Ω; each relation edge has a pre-defined relation type coefficient w. p（a） Then, weight f of all edges within the relation edge set a Relationship type coefficient w p（a） The importance of a node is obtained by performing a weighted summation; the calculation formula is as follows:

[0013] Where Ω represents the set of relation edges connected to the node, a represents the index of any relation edge connected to the node, p(a) represents the relation type of relation edge a, and w p（a） The relationship type coefficient is used to obtain the importance of each node and fill it into the corresponding position in the global talent knowledge graph. An importance threshold is preset. If the importance of a node is greater than or equal to the importance threshold, the node is marked as a core node.

[0014] In some embodiments, the step of parsing based on talent needs to construct a talent demand graph from core nodes in the global talent knowledge graph includes: S401, when the enterprise needs to search for talent, match positions, or build projects, the requirements are semantically parsed and broken down into target types, constraints, and priority conditions. Based on the talent knowledge ontology, the requirements are entity-mapped, and the keywords in the requirements are mapped to target entity types that are consistent with the global talent knowledge graph. S402, take the individual entity nodes in the target entity set of the demand as the seed nodes for retrieval in turn, and perform extended retrieval in the global talent knowledge graph to form the talent demand graph corresponding to the demand. S403. Within the talent demand graph, determine all feasible connection paths between target entity nodes and candidate nodes, and perform traversal analysis to obtain the optimal support path from the candidate node to each target entity node and the corresponding support index. For the same candidate node, weight and summarize its support index relative to the optimal support path of all target entity nodes according to the importance of the target entity node in the current demand to obtain the comprehensive demand index of the candidate node, and rank the candidate nodes accordingly, and output the demand results.

[0015] In some embodiments, the step of performing an extended search in the global talent knowledge graph to construct the talent demand graph corresponding to the demand includes: Nodes directly related to the target entity node are extracted from the global talent knowledge graph as first-order candidate nodes. If a first-order candidate node is a core node in the global talent knowledge graph, it needs to be expanded, and nodes directly connected to the first-order candidate node are used as second-order candidate nodes. That is, second-order candidate nodes are indirectly connected to the target entity node in the global talent knowledge graph. The first-order and second-order candidate nodes of the target entity node are merged into the candidate nodes of the target entity node. If a first-order candidate node is not a core node in the global talent knowledge graph, it does not need to be expanded, and the first-order candidate node is directly used as the candidate node of the target entity node. In this way, the target entity node corresponding to the requirement and the candidate nodes of each target entity node can be obtained.

[0016] In some embodiments, the traversal analysis to obtain the optimal support path from the candidate node to each target entity node and the corresponding support index includes: Determine all feasible connection paths between the target entity node k and the candidate node v, and perform traversal analysis, where k represents the index of the target entity node in the required target entity set, and v represents the index of the candidate node to which the target entity node k belongs. Specifically, for any candidate node, determine all feasible paths between it and the target entity node k. A feasible path is a directed path from the candidate node to the target entity node through several relation edges. Feasible paths serve as supporting paths for the target entity node. Count the number of relation edges contained in the supporting path q, denoted as Q, where q is the index of the supporting path. Then, the edge weight of the j-th relation edge is denoted as f. j j is a positive integer representing the index of the support path containing the relation edge, j∈Q; support index The calculation formula is as follows:

[0017] Where ρ is the path length penalty coefficient, used to suppress the phenomenon of artificially high support caused by excessively long paths; in this embodiment, it is set to 0.2. A geometric mean is used to fuse the edge weights of each relation edge on the path, ensuring that each edge contributes to the support index while avoiding the rapid decay of simple multiplication in multi-hop path scenarios. Thus, the support index of all support paths to the target entity node is obtained. The support path with the largest support index is selected as the optimal support path from the candidate node to the target entity node. Therefore, the optimal support path and corresponding support index for each target entity node can be obtained. , specifically This enables a quantitative characterization of the relationship strength between candidate nodes and target entity nodes.

[0018] Compared with the prior art, the beneficial effects of the present invention are: First, by quantifying the credibility, completeness, and consistency of data sources into a unified confidence coefficient, quality gate control is applied to multi-source talent data. This effectively eliminates duplicate, erroneous, missing, and outdated data, ensuring that the data foundation for subsequent graph construction is reliable, formatted uniformly, and semantically parsable, thus reducing the interference of noisy data on talent management results from the source. Second, through entity, attribute, and relation extraction under knowledge ontology constraints, structured fields and unstructured text originally scattered across various business systems are uniformly mapped into nodes, edges, and attributes in the graph. This breaks down the silos of traditional personnel database tables, enabling computable semantic relationships between employees, positions, projects, certificates, skills, training, and performance, thereby providing a unified expression carrier for talent profiling and relational reasoning. Then, through edge weight calculation... By calculating node importance, a hierarchical identification of core relationships and nodes in the talent knowledge graph is achieved. This improves the credibility and stability of relationships and allows subsequent analysis to prioritize high-value nodes, reducing the interference of low-relevance nodes. Finally, through demand semantic parsing, target entity mapping, and candidate expansion, a talent demand graph corresponding to the demand is extracted and constructed from the global talent knowledge graph. Based on the talent demand graph, supporting path scoring and comprehensive demand index calculation are performed. This enables the output of recommendation results with path evidence support in talent query, job matching, and project formation scenarios. This not only improves recommendation accuracy but also enhances the interpretability and traceability of the results. It realizes unified organization, semantic association, dynamic analysis, and intelligent decision-making of enterprise talent data, improving the accuracy of talent matching and the interpretability of results. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0021] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of the present invention. However, the present invention can be practiced in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0022] This embodiment provides a knowledge graph-based enterprise talent database management method, applicable to enterprises for unified management, correlation analysis, and intelligent decision-making of talent data in scenarios such as recruitment, training, job transfer, succession, project formation, and talent inventory. This method relies on the enterprise's existing HR system, performance system, training system, project management system, and office collaboration system to collect talent information from multiple sources, perform unified modeling, and manage it in a knowledge graph format, thereby solving the problems of data fragmentation, weak semantic connections, inaccurate profiles, and insufficient decision support in existing talent databases; for example... Figure 1 As shown, the enterprise talent database management method based on knowledge graphs includes the following steps: Step 100: Extract structured and unstructured talent data from various business systems, and perform standardized preliminary cleaning to remove duplicate, erroneous, and invalid records, ensuring the reliability, uniform format, and semantic parsing of the original data. The collected talent data should include at least: basic identity data, educational background data, skills data, project experience data, training data, performance evaluation data, job mobility data, organizational collaboration data, and behavioral event data. It should be noted that during data collection, a unique employee identifier, timestamp, and source identifier are attached to each data entry. During collection, a confidence coefficient is calculated for each data entry; the specific calculation process includes: The source credibility coefficient is determined based on the type of data source. Specifically, the source credibility coefficient is used to characterize the reliability of the business system corresponding to the data. Different sources are set with corresponding source credibility coefficients. Specifically, the source types include at least the master data system, human resources management system, performance appraisal system, training management system, project management system, OA approval system, third-party certificate platform, and manually entered data. In this embodiment, the basic level value of the master data system can be set to 0.95, the human resources management system to 0.93, the performance appraisal system to 0.88, the training management system to 0.85, the project management system to 0.84, the OA approval system to 0.80, the third-party certificate platform to 0.78, and the manually entered data to 0.60. A completeness coefficient is calculated based on the completeness of data fields. This coefficient characterizes whether the key fields of a data entry are complete. Specifically, a set of key fields is pre-established for different types of talent data. This set is not identical for all data but is determined according to the business type of the data: when the data is basic identity data, the pre-defined key fields include at least employee ID, name, department, job title, start date, and current job level; when the data is educational background data, the pre-defined key fields include at least education level, degree, graduating institution, major, and graduation date; when the data is project experience data, the pre-defined key fields include at least project name, project start and end dates, project role, project field, and project deliverables; when the data is performance evaluation data, the pre-defined key fields include at least... Key fields are defined to include at least the performance cycle, performance level, score, and completion rate of performance indicators. Pre-defined key fields refer to fields that reflect the core semantics of the talent data and have a direct impact on subsequent knowledge extraction, profile construction, or matching calculations. These are typically configured by the enterprise in advance based on business rules, job management rules, and data modeling rules. The actual number of pre-defined key fields filled in each data entry and the total number of key fields are counted. The completeness coefficient is obtained by dividing the actual number of filled fields by the total number of key fields. If all key fields in a data entry are filled and valid, the completeness coefficient is 1. If there are missing fields, the completeness coefficient decreases as the number of missing fields increases. The completeness coefficient quantitatively reflects whether the data is structurally complete, avoiding the misinterpretation of incomplete data with complete data. A consistency coefficient is calculated based on the degree of consistency of corresponding fields of the same data subject across different business systems. This consistency coefficient characterizes the degree of consistency of data across systems. Specifically, a set of fields requiring consistency comparison is determined. This set consists of fields that may appear in multiple business systems for the same data subject and have cross-validation significance. Specifically, when the data subject is an employee, the field set includes at least name, gender, date of birth, date of employment, department name, job title, education level, job grade, certificate name, and performance rating. When the data subject is a project, the field set includes at least project name, project number, project start and end dates, project area, and project manager. When the data subject is a certificate, the field set includes at least certificate name, certificate number, issuing authority, and validity period. Values ​​of corresponding fields for the same data subject are read from different business systems to form a field group to be compared. During the calculation process, each field is first standardized before a consistency judgment is performed. Standardization refers to unifying the different expressions of the same field in different systems into the same semantic code. For a certain field to be compared, let the field correspond to x1, x2, x3...xm in m business systems, where m represents the total number of business systems with the field to be compared. If the standardized values ​​are exactly the same in all m business systems, the consistency score of the field is recorded as 1; if there are different values, the score is calculated according to the consistency ratio. The number of systems with the same standardized value for the same field in each business system is counted, and the ratio of this number to the total number of systems participating in the comparison is used as the consistency degree of the field. Then, the consistency degrees of each field in the data bar are weighted and summed according to their preset weights to obtain the consistency coefficient of the data bar. The field weights are set according to the degree of influence of the field on talent management. Based on this, the source credibility coefficient, completeness coefficient, and consistency coefficient mentioned above will be linearly weighted and fused according to preset weights to obtain the confidence coefficient. The sum of the weights corresponding to the source credibility coefficient, completeness coefficient, and consistency coefficient is one, thus obtaining the confidence coefficient of all data. Through this confidence coefficient, a unified quality evaluation can be carried out on talent data with different sources, different completeness, different consistency, and different freshness, and a reliable data foundation can be provided for subsequent knowledge extraction, relationship construction, graph weighting, and profile generation. Step 200: Extract the confidence coefficient for each data entry and compare it with a preset confidence threshold. If the confidence coefficient is greater than or equal to the confidence threshold, it indicates that the data entry meets the core construction requirements of the knowledge graph in terms of source reliability, field completeness, cross-system consistency, and timeliness. This indicates that the data entry has high credibility and strong usability, and the data entry is considered as credible data. If the confidence coefficient is less than the confidence threshold, it indicates that the data entry has insufficient source credibility, missing fields, field conflicts, or weak timeliness. This indicates that the data entry has not yet reached the credibility standard for direct use in the construction of the core knowledge graph, and the data entry is not included in the core construction scope of the knowledge graph. This data entry is then fed back to manual verification as a data entry to be verified. Entity extraction, attribute extraction, and relation extraction are performed on the filtered data, and the extraction results are mapped to nodes, attributes, and edges in the knowledge graph to construct a talent knowledge graph. Specifically, this includes: A talent knowledge ontology is constructed based on the business semantics of enterprise talent management, and elements are extracted from data bars. Elements include entities, attributes, and relationships. Entities include at least employees, departments, positions, projects, courses, certificates, skills, performance events, training events, reward and punishment events, mentors, and teams. Attributes include at least time, level, score, duration, frequency, validity period, and level. Relationships include at least being employed in, participating in, mastering, acquiring, completing, being assessed, being trained in, collaborating in, being promoted in, being transferred to, and being mentored. In the implementation process, a combination of rule extraction and model extraction is adopted. For structured fields within data bars, they are directly converted into corresponding entities, attributes, and relationships according to preset field mapping rules. For unstructured fields, such as training summaries, project descriptions, performance evaluations, and reward and punishment explanations, talent elements are extracted through named entity recognition and relationship extraction models and mapped into entities, attributes, and relationships. After extracting entities and relationships, entities in the talent ontology are mapped to nodes in the knowledge graph, and semantic relationships between entities are mapped to edges in the knowledge graph. Time, level, score, duration, frequency, validity period, hierarchical level, and role information are used as node or edge attributes for connection. Employees, departments, positions, projects, courses, certificates, skills, performance events, training events, reward / punishment events, mentors, and teams are all considered nodes, while positions held, participated in, mastered, acquired, completed, assessed, trained, collaborated with, promoted, transferred to, and guided are all considered edges. This forms a global talent knowledge graph containing multiple types of nodes, multiple types of relationships, and multi-dimensional attributes. In this embodiment, the global talent knowledge graph uniformly stores all employees, positions, departments, projects, courses, skills, certificates, training, performance, and organizational collaboration relationships within the enterprise. Through the global talent knowledge graph, cross-departmental, cross-position, and cross-project talent relationship connectivity can be achieved, providing a unified data foundation for subsequent talent analysis, job matching, and organizational decision-making. By quantifying the credibility, completeness, and consistency of the data sources into a unified confidence coefficient, a quality gate control is first applied to multi-source talent data. This effectively eliminates duplicate, erroneous, missing, and outdated data, ensuring that the data foundation for subsequent graph construction is reliable, formatted uniformly, and semantically parsable. This reduces the interference of noisy data on talent management results from the source. Then, based on entity, attribute, and relation extraction under the constraints of knowledge ontology, the structured fields and unstructured text originally scattered in various business systems are uniformly mapped into nodes, edges, and attributes in the graph. This breaks down the table silos of traditional personnel databases, enabling computable semantic associations between employees, positions, projects, certificates, skills, training, and performance, thereby providing a unified expression carrier for talent profiling and relational reasoning.

[0023] Step 300: Assign edge weights to each edge within the global talent knowledge graph and assign importance to each node. The specific process is as follows: For each relation edge in the global talent knowledge graph, extract the entities at both ends of the edge. Based on the entities extracted from each data bar, use name standardization, alias merging, semantic disambiguation, and primary key association to merge the data bars pointing to the entities at both ends into supporting data bars for the same relation edge, thus obtaining the supporting data bars for all relation edges. Extract the generation time of each supporting data bar, and calculate the time interval by calculating the time difference between the current time and the generation time. Calculate the timeliness factor d using an exponential decay method, as shown in the following formula:

[0024] Where Δt is the time interval, and τ is a preset attenuation constant, the value of which is determined according to the relationship type. The more stable the relationship, the smaller the value of τ; the more dynamic the relationship, the larger the value of τ. In this embodiment, the attenuation constant for the skill relationship type is 0.3, the attenuation constant for the project relationship type is 0.5, the attenuation constant for the certificate relationship type is 0.4, the attenuation constant for the training relationship type is 0.4, and the attenuation constant for the appointment or reporting relationship type is 0.05. The confidence coefficient α corresponding to the supporting data bar of the relationship edge is... i Time factor d i The weight of the relation edge is calculated using the following formula, where i represents the index of the supporting data strip of the relation edge.

[0025] Wherein, μ is the repeated support enhancement coefficient, used to control the degree of enhancement of the edge weight of the relation by the number of supporting data points. μ is usually set to 0, 1-0.5, and is set to 0.3 in this embodiment. By setting an enhancement factor in the form of ln(1+n), the edge weight can be monotonically increased as the number of supporting data points increases, while avoiding the infinite expansion of edge weight due to linear accumulation. Thus, the edge weight of each relation edge can be obtained and filled into the corresponding position in the global talent knowledge graph. It should be noted that the edge weight of the relation edge reflects the credibility, timeliness and validity of a single supporting data point, as well as the repeated confirmation strength of multiple supporting data points, thereby ensuring that the edge weight increases with the increase of the number of supporting data points and avoiding the distortion of edge weight due to simple linear accumulation. For any given node, calculate the weights of the edges connecting it to other nodes, and then sum them using the relation type coefficient to obtain the node's importance. The calculation formula is as follows:

[0026] Where Ω represents the set of relation edges connected to the node, a represents the index of any relation edge connected to the node, p(a) represents the relation type of relation edge a, and w p（a） In this embodiment, the relation type coefficient is pre-configured and assigned values ​​based on the business coreness, stability, and multi-source verifiability of the relation type. Higher business coreness, stronger stability, and easier multi-source data verification result in a larger relation type coefficient. During initialization, values ​​are assigned to each relation type based on historical sample statistics, and the coefficients are fine-tuned during operation based on the verification results of newly added samples to ensure consistency between the relation type coefficients and the actual business semantics. aThe importance of a node is used to characterize its business influence and connection strength within the global talent knowledge graph, where the edge weight represents the relationship. Higher node importance indicates stronger relevance in talent organization, project collaboration, job mobility, or capability development. This allows us to determine the importance of each node and fill it into its corresponding position in the global talent knowledge graph. A preset importance threshold is used for stratified node selection, typically based on statistical analysis of historical importance samples of marked nodes in the global talent knowledge graph. This yields the importance distribution ranges for core nodes and ordinary nodes, which are then combined with the enterprise's talent management strategies for key positions, key projects, and key personnel. The importance threshold for key skills is set by taking the lower bound of the importance distribution of core nodes or the boundary between the average importance of core nodes and the average importance of ordinary nodes. This threshold can distinguish between key nodes with high connectivity, high influence, and high reusability and ordinary nodes, avoiding the omission of important nodes and suppressing the misclassification of low-connectivity nodes as core nodes. If the importance of a node is greater than or equal to the importance threshold, it means that the node has a high degree of business influence, strong relational connection ability, and high multi-source verification stability in the global talent knowledge graph. It can serve as a key anchor point in subsequent requirements analysis, path search, job matching, or project formation, and the node is marked as a core node. Step 400: When addressing a company's needs for talent search, job matching, or project formation, semantic parsing is performed on the needs, breaking them down into target type, constraints, and priority conditions. The target type indicates whether the current need pertains to talent search, job matching, or project formation; constraints represent mandatory conditions; and priority conditions represent preferred but not mandatory conditions. Needs must include at least one or more of the following: job title, professional direction, skill keywords, educational requirements, job level requirements, project field, certification requirements, performance requirements, geographical requirements, time requirements, and team role requirements. The needs are then entity-mapped based on the talent knowledge ontology. The mapping process maps keywords in the requirements to target entity types consistent with the global talent knowledge graph. Specifically, job titles are mapped to job nodes, professional directions and project domains are mapped to domain nodes or project nodes, skill keywords are mapped to skill nodes, education requirements are mapped to education attribute or educational background nodes, job level requirements are mapped to job level nodes, certificate requirements are mapped to certificate nodes, performance requirements are mapped to performance nodes, and team role requirements are mapped to role attributes or team nodes. For fuzzy terms that cannot be directly mapped, disambiguation is performed through synonym merging, word vector similarity calculation, and ontology constraint rules to obtain a set of target entities consistent with the semantics of the requirements. Using individual entity nodes from the target entity set as seed nodes, nodes directly related to the target entity nodes are extracted from the global talent knowledge graph as first-order candidate nodes. If a first-order candidate node is a core node in the global talent knowledge graph, expansion is required, and nodes directly connected to the first-order candidate node are designated as second-order candidate nodes. Second-order candidate nodes are indirectly connected to the target entity node in the global talent knowledge graph. The first-order and second-order candidate nodes of the target entity node are merged into a single candidate node. If a first-order candidate node is not a core node in the global talent knowledge graph, expansion is unnecessary, and the first-order candidate node is directly used as a candidate node for the target entity node. This process yields the corresponding requirement. The system identifies target entity nodes and candidate nodes for each target entity node. Specifically, if the demand type is talent search, the target employee or target competency tag is used as the target entity node. Centered on the target entity node, adjacent job nodes, project nodes, skill nodes, training nodes, certificate nodes, and performance nodes are extracted. If the demand type is job matching, the target job node is used as the center node, and employee nodes with a job relationship, skill association relationship, project association relationship, or performance association relationship with the job node are extracted. If the demand type is project formation, the target project node is used as the center node, and employee nodes and project nodes with a participation relationship, collaboration relationship, skill support relationship, or historical similar project relationship with the project node are extracted. After expanding the candidate nodes, the target entity nodes, candidate nodes, and related relationship edges corresponding to the requirement are collectively constructed into a talent requirement graph for that requirement. The talent requirement graph is not a simple truncated version of the global talent knowledge graph, but a local task graph formed around the target requirement. It includes target entity nodes (i.e., nodes representing requirements), candidate entity nodes, and relationship edges supporting these nodes. The talent information, job information, and project collaboration information that were originally scattered in the global graph are compressed into a local analysis space that is directly related to the current requirement. Within the talent demand graph, all feasible connection paths between target entity node k and candidate node v are determined and analyzed. Here, k represents the index of the target entity node in the set of target entities, and v represents the index of the candidate node to which target entity node k belongs. Specifically, for any candidate node, all feasible paths between it and target entity node k are determined. A feasible path is a directed path from the candidate node to the target entity node via several relation edges. Feasible paths serve as supporting paths for the target entity node. The number of relation edges contained in the supporting path q is denoted as Q, where q is the index of the supporting path. The edge weight of the j-th relation edge is denoted as f. j j is a positive integer representing the index of the support path containing the relation edge, j∈Q; support index The calculation formula is as follows:

[0027] Where ρ is the path length penalty coefficient, used to suppress the phenomenon of artificially high support caused by excessively long paths; in this embodiment, it is set to 0.2. A geometric mean is used to fuse the edge weights of each relation edge on the path, ensuring that each edge contributes to the support index while avoiding the rapid decay of simple multiplication in multi-hop path scenarios. Thus, the support index of all support paths to the target entity node is obtained. The support path with the largest support index is selected as the optimal support path from the candidate node to the target entity node. Therefore, the optimal support path and corresponding support index for each target entity node can be obtained. , specifically This enables a quantitative characterization of the relationship strength between candidate nodes and target entity nodes; For the same candidate node v, its support index relative to the optimal support path of all target entity nodes. The comprehensive demand index M of candidate node v is obtained by weighting and summing the target entity nodes according to their importance in the current demand. v The formula for calculating the weighted sum is as follows:

[0028] Where, η kThis represents the demand weight of the k-th target entity node. The demand weight is set according to the conditions that the target entity corresponding to the target entity node belongs to in the demand. In this embodiment, if the target entity corresponding to the target entity node belongs to the constraint condition in the demand, the demand weight is set to 1.0; if the target entity corresponding to the target entity node belongs to the priority condition in the demand, the demand weight is set to 0.7; if the target entity corresponding to the target entity node belongs to other conditions in the demand (neither a constraint condition nor a priority condition), the demand weight is set to 0.3. Among them, the constraint condition is used to characterize the condition that must be met, the priority condition is used to characterize the condition that is recommended but not required to be met, and the other conditions are used to characterize auxiliary descriptions or supplementary conditions. Through the above-mentioned hierarchical settings, the demand weights can accurately reflect the constraint strength and importance of different target entity nodes in the current demand, and provide a unified weight basis for subsequent path scoring and matching ranking. As can be seen from the above calculation process, the comprehensive demand index of the candidate node not only reflects the path support strength between the candidate node and the demand, but also reflects the importance of different target entity nodes to the current demand. The comprehensive demand index of each candidate node is filled into the talent demand graph corresponding to the demand, and the talent demand graph is output as the demand result to achieve the interpretability of the talent demand graph. It can transform the complex relationships in the global talent knowledge graph into the optimal support path and structured result output that is directly effective for the current demand, thereby improving the accuracy and interpretability of talent query, job matching and project formation. Furthermore, in this embodiment, all candidate nodes are sorted in descending order based on the comprehensive demand matching score, and then an interpretable result is generated by combining the optimal support path corresponding to each candidate node. Specifically, if the demand type is talent query, the talent nodes that meet the conditions and their optimal support paths are output, and path evidence such as positions, skills, projects, training, and performance supporting the talent node is given simultaneously. If the demand type is position matching, the talent nodes that best match the target position node, their corresponding optimal support paths and support indices are output, and their position suitability advantages and capability gaps are further output. If the demand type is project formation, the set of talent nodes that best match the target project node is output, and the candidates are classified into roles according to the optimal support paths and support indices corresponding to each talent node, forming recommendation results for leaders, core members, and support members. By extracting and constructing a talent demand graph corresponding to the demand from the global talent knowledge graph through demand semantic parsing, target entity mapping, and candidate expansion, and performing support path scoring and comprehensive demand index calculation based on the talent demand graph, recommendation results with path evidence can be output in talent query, job matching, and project formation scenarios. This not only improves recommendation accuracy but also enhances the interpretability and traceability of the results. It enables unified organization, semantic association, dynamic analysis, and intelligent decision-making of enterprise talent data, thereby improving the accuracy of talent matching and the interpretability of results.

[0029] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0030] The formula described above is an example of an empirical formula defined by an expert in the field. Any formula that conforms to the relationship between the parameters of this application is acceptable. The specific size of the weighting factors in the formula shall be reasonably set by a person skilled in the art based on actual use.

[0031] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A knowledge graph-based enterprise talent database management method, characterized in that, Includes the following steps: Talent data is collected from various business systems, and the collected talent data is standardized and preliminarily cleaned to calculate the confidence coefficient for each data point. Based on the confidence coefficient of the data bar, the data bars are filtered to obtain confidence data. Entities, attributes and relationships are extracted from the filtered confidence data. Based on this, relationship edges and nodes are defined, and a global talent knowledge graph is constructed. Based on the global talent knowledge graph, edge weights are assigned to each relation edge, the importance of nodes is calculated based on the edge weights, and core nodes are identified. The system analyzes the talent needs to construct a talent demand graph from the core nodes of the global talent knowledge graph and outputs the demand results.

2. The enterprise talent database management method based on knowledge graphs according to claim 1, characterized in that, The calculation of the confidence coefficient for each data point includes: The source credibility coefficient is determined based on the type of data source. The source credibility coefficient is used to characterize the reliability of the business system corresponding to the data. The integrity coefficient is calculated based on the completeness of the data fields. The integrity coefficient is used to characterize whether the key fields of the data are complete. The consistency coefficient is calculated based on the degree of consistency of corresponding fields of the same data subject in different business systems. The consistency coefficient is used to characterize the degree of consistency of data across systems. The confidence coefficient is obtained by linearly weighting and fusing the source credibility coefficient, integrity coefficient, and consistency coefficient according to preset weights. The sum of the weights corresponding to the source credibility coefficient, integrity coefficient, and consistency coefficient is one, thus obtaining the confidence coefficient of all data.

3. The enterprise talent database management method based on knowledge graphs according to claim 2, characterized in that, The calculation of the integrity coefficient based on the completeness of the data fields includes: For different types of talent data, a set of key fields is pre-established. The actual number of key fields filled in each data point and the total number of key fields are counted. The completeness coefficient is obtained by dividing the actual number of key fields filled in by the total number of key fields.

4. The enterprise talent database management method based on knowledge graphs according to claim 3, characterized in that, The calculation of the consistency coefficient based on the consistency of corresponding fields of the same data subject in different business systems includes: A set of fields that may appear in different business systems and have cross-validation significance for the same data subject is identified. After standardizing each field in the set, the number of systems in which the same field has the same standardized value is counted in each business system participating in the comparison. The ratio of this number to the total number of systems participating in the comparison is used as the consistency degree of the field. The consistency degrees of each field are then weighted and summed according to preset weights to obtain the consistency coefficient of the data bar.

5. The enterprise talent database management method based on knowledge graphs according to claim 4, characterized in that, The process of extracting entities, attributes, and relationships from the filtered confidence data, defining relationship edges and nodes based on this, and constructing a global talent knowledge graph includes: A talent knowledge ontology is constructed based on the business semantics of enterprise talent management, and entities, attributes, and relationships are extracted from data bars. For structured fields in data bars, they are transformed into entities and relationships according to field mapping rules. For unstructured fields in data bars, talent elements are extracted through named entity recognition and relationship extraction models and transformed into entities and relationships. The extracted entities are defined as nodes in the knowledge graph, the relationships between entities are defined as edges between nodes, and attributes are used as attribute information of corresponding nodes or relationship edges. A global talent knowledge graph is constructed based on nodes and relationship edges.

6. The enterprise talent database management method based on knowledge graphs according to claim 5, characterized in that, Based on the global talent knowledge graph, the edge weights are assigned to each relation edge, including: For each relation edge in the global talent knowledge graph, extract the entities at both ends of the edge. Based on the entities extracted from each data bar, use name standardization, alias merging, semantic disambiguation, and primary key association to merge the data bars pointing to the entities at both ends into supporting data bars for the same relation edge, thus obtaining the supporting data bars for all relation edges. Extract the generation time of each supporting data bar, calculate the time interval by the time difference between the current time and the generation time, and then calculate the time factor using an exponential decay method. Calculate the edge weight of the relation edge by formulating the confidence coefficient, time factor, and number of supporting data bars corresponding to the supporting data bars of the relation edge.

7. The enterprise talent database management method based on knowledge graphs according to claim 6, characterized in that, The process of calculating node importance based on edge weights and identifying core nodes includes: For any node, the weights of the edges connecting it to the node are counted to form a set of edges. Each type of edge has a predefined relationship type coefficient. The weights and coefficients of all edges in the set are then weighted and summed to obtain the importance of the node. The importance of each node is then determined and filled into the corresponding position in the global talent knowledge graph. An importance threshold is predefined. If the importance of a node is greater than or equal to the importance threshold, the node is marked as a core node.

8. The enterprise talent database management method based on knowledge graphs according to claim 7, characterized in that, The step of analyzing talent needs to construct a talent demand graph from the core nodes of the global talent knowledge graph includes: When enterprises have needs for talent search, job matching or project formation, the requirements are semantically parsed and broken down into target types, constraints and priority conditions. Based on the talent knowledge ontology, the requirements are entity mapped and the keywords in the requirements are mapped to target entity types consistent with the global talent knowledge graph. Using individual entity nodes in the target entity set of the demand as seed nodes for retrieval, and expanding the retrieval in the global talent knowledge graph to form the talent demand graph corresponding to the demand. Within the talent demand graph, all feasible connection paths between target entity nodes and candidate nodes are identified and traversed to obtain the optimal support path from the candidate node to each target entity node and the corresponding support index. For the same candidate node, its support index relative to the optimal support path of all target entity nodes is weighted and summarized according to the importance of the target entity node in the current demand to obtain the comprehensive demand index of the candidate node. Based on this, the candidate nodes are ranked and the demand results are output.

9. The enterprise talent database management method based on knowledge graphs according to claim 8, characterized in that, The process of expanding the search within the global talent knowledge graph to construct the talent demand graph corresponding to this demand includes: Nodes directly related to the target entity node are extracted from the global talent knowledge graph as first-order candidate nodes. If a first-order candidate node is a core node in the global talent knowledge graph, it needs to be expanded, and nodes directly connected to the first-order candidate node are used as second-order candidate nodes. That is, second-order candidate nodes are indirectly connected to the target entity node in the global talent knowledge graph. The first-order and second-order candidate nodes of the target entity node are merged into the candidate nodes of the target entity node. If a first-order candidate node is not a core node in the global talent knowledge graph, it does not need to be expanded, and the first-order candidate node is directly used as the candidate node of the target entity node. In this way, the target entity node corresponding to the requirement and the candidate nodes of each target entity node can be obtained.

10. The enterprise talent database management method based on knowledge graphs according to claim 9, characterized in that, The traversal analysis yields the optimal support path from the candidate node to each target entity node and the corresponding support index, including: Determine all feasible paths between candidate nodes and target entity nodes. Each feasible path is a directed path from a candidate node to the target entity node via several relation edges. These feasible paths are then used as support paths for the target entity node. Count the number of relation edges on the support paths and extract the edge weights corresponding to each relation edge in the support paths. Calculate the support index corresponding to the support path based on the edge weights of each relation edge in the support path and the path length penalty coefficient. Traverse all support paths between candidate nodes and target entity nodes, and select the support path with the largest support index as the optimal support path from candidate node to target entity node. Thus, the optimal support path from candidate node to target candidate node and the corresponding support index are obtained.

Images