A knowledge graph-based professional competence evaluation system

Abstract

The application discloses a kind of professional ability evaluation systems based on knowledge graph, comprising: ability bottom account module, for obtaining the professional ability data of target personnel, generates ability bottom account;Graph construction module is used to construct function graph;Interaction generation module is used to generate ability interaction diagram, constructs function spectrum pulse;Representation generation module is used to input the ability interaction diagram and function spectrum pulse into improved LightGCN model, introduce function spectrum pulse through mechanism, generate personnel representation and post representation;Combination generation module is used to construct ability transaction table, and function combination is generated using Eclat algorithm introducing vertical item cluster trace extraction mechanism;Evaluation generation module is used to generate ability evaluation result.The application realizes the fine modeling of professional ability structure and the high-precision evaluation of post adaptation relationship, improves the accuracy and application reliability of ability evaluation result.

Description

Technical Field

[0001] This invention relates to the field of data processing and intelligent assessment technology, and in particular to a knowledge graph-based vocational ability assessment system. Background Technology

[0002] With the continuous improvement of the digitalization and intelligence of human resource management, the demand for technologies related to personnel competency modeling, job matching analysis, and career development assessment is constantly growing. Knowledge graph-based career competency assessment technology is gradually becoming a key research and application area. Existing career competency assessment methods mainly rely on rule matching, statistical analysis, or simple vector similarity calculation for competency assessment and job recommendation. However, these methods generally suffer from the following problems in practical applications: The diverse sources and heterogeneous structures of occupational competency data lead to both information gaps and data redundancy during the competency baseline construction process, affecting the accuracy of subsequent modeling. Knowledge graph construction relies heavily on static relation extraction, lacking effective modeling of hierarchical and dependency relationships between competencies. This results in the underutilization of structural information between competencies, making it difficult to support the refined expression of complex job competency requirements. Graph-based competency representation methods typically employ unordered neighborhood propagation during node representation learning, ignoring sequential dependencies and hierarchical evolution relationships between competencies. This leads to semantic aliasing and path distortion during competency information propagation, reducing the discriminative power of personnel and job representations. In competency combination mining, traditional frequent itemset mining methods rely solely on transaction co-occurrence relationships for combination generation, lacking comprehensive consideration of competency cluster consistency, job consistency, and path order constraints. The generated competency combinations fail to reflect the true occupational competency structure. In the evaluation phase, most methods rely on a single matching index for calculation, failing to simultaneously characterize competency coverage and competency synergy, resulting in insufficient stability and weak interpretability of job suitability evaluation results.

[0003] Therefore, how to provide a knowledge graph-based professional competence assessment system is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a knowledge graph-based vocational competency assessment system. This invention employs a method of constructing a competency ledger, constructing a functional graph, generating a competency interaction graph, constructing a functional spectrum, generating representations using an improved LightGCN model, and mining functional combinations using the Eclat algorithm with a vertical term cluster trace extraction mechanism. This method completes the structured modeling of vocational competency data, the expression of competency relationships, the learning of node representations, and the analysis of competency combinations. Based on personnel representations, job representations, and functional combinations, it calculates competency coverage, competency synergy, and job suitability. It has the advantages of accurate expression of competency relationships, controllable competency propagation process, reasonable competency combination structure, and high stability of assessment results.

[0005] A knowledge graph-based vocational competency assessment system according to an embodiment of the present invention includes: The competency ledger module is used to acquire the professional competency data of the target personnel, perform data cleaning and field unification processing on the professional competency data, and generate a competency ledger. The graph construction module is used to extract personnel entities, capability entities, and job entities based on the capability ledger, establish the mastery relationship between personnel and capabilities, and the demand relationship between capabilities and jobs, and construct a functional graph. The interaction generation module is used to map personnel entities, capability entities, and job entities as nodes based on the functional graph, and to map the mastery relationship between personnel and capability, as well as the requirement relationship between capability and job, as interaction edges to generate a capability interaction graph, and to construct a functional spectrum based on the hierarchical relationship and dependency relationship between capability entities. The representation generation module is used to input the capability interaction graph and functional spectrum into the improved LightGCN model, introduce a functional spectrum connection mechanism in the lightweight graph propagation layer to perform neighborhood information propagation, and generate personnel representation and job representation. The combination generation module is used to construct a capability transaction table based on the functional map, and uses the Eclat algorithm with the introduction of the vertical item cluster trace disjunction mechanism to perform vertical itemset mining and generate functional combinations; The assessment generation module is used to calculate capability coverage, capability synergy, and job fit based on the personnel representation, job representation, and function combination, and generate capability assessment results.

[0006] Optionally, the capability ledger module specifically comprises: Obtain the professional competence data of the target personnel, including personnel competence records, learning records, task records, job records, and evaluation records; The professional skills data is processed by deleting duplicate records, filling in empty records, removing invalid fields, and unifying field names to generate standard records. The personnel identifiers, skill names, job titles, recording times, and data sources in the standard records are formatted uniformly to generate standardized records; The standardized records are merged according to personnel identification, capability name, and job title to generate a capability ledger.

[0007] Optionally, the atlas construction module specifically comprises: Read the personnel identifier field, capability name field, and job name field from the capability ledger, and generate the corresponding personnel entity, capability entity, and job entity respectively; The capability ledger establishes a relationship between personnel and capabilities based on the correspondence between personnel identification and capability name. Establish the relationship between capabilities and job requirements in the capability ledger by recording the correspondence between capability names and job titles; The personnel entities, capability entities, and job entities, as well as the relationships between personnel and capability and between capability and job requirements, are linked and organized to generate a functional map.

[0008] Optionally, the step of mapping personnel entities, capability entities, and job entities to nodes based on the functional graph, and mapping the mastery relationship between personnel and capabilities, as well as the requirement relationship between capabilities and job requirements, to generate a capability interaction graph, specifically: In the functional graph, locate personnel entities, capability entities, and job entities, and generate corresponding personnel nodes, capability nodes, and job nodes respectively; Write the personnel identifier in the personnel node, the ability name in the ability node, and the job name in the job node; In the functional graph, the first interaction edge between personnel nodes and capability nodes is established based on the relationship between personnel and capability mastery. In the functional graph, a second interaction edge is established between capability nodes and job nodes based on the relationship between capability and job requirements. The personnel nodes, capability nodes, job nodes, first interaction edges, and second interaction edges are combined to generate a capability interaction graph.

[0009] Optionally, the construction of the functional spectrum based on the hierarchical and dependency relationships between capability entities specifically includes: Match the hierarchical relationships and dependencies between capability entities in the functional graph; The higher-level and lower-level capability entities in the hierarchical relationship are respectively defined as the hierarchical start point and hierarchical end point to form the hierarchical direction; The preceding and following capability entities in the dependency relationship are determined as the dependency start point and dependency end point, respectively, forming the dependency direction; Connect capability entities according to the hierarchical and dependency directions to form an ordered capability path; Record the path location and path source for the capability entities in the ordered capability path to generate a functional spectrum.

[0010] Optionally, the improved LightGCN model specifically includes a graph input layer, a spectral pulse mapping layer, a lightweight graph propagation layer, and a representation output layer; The graph input layer is based on personnel nodes, capability nodes, job nodes, first interaction edges and second interaction edges in the capability interaction graph. It establishes connection records between personnel nodes and capability nodes according to the first interaction edge, establishes connection records between capability nodes and job nodes according to the second interaction edge, and generates a node adjacency table from the connection records. The spectrum mapping layer, based on the ordered capability paths, path positions, and path sources in the functional spectrum, arranges the capability nodes in the ordered capability paths according to their path positions and writes the path sources into the corresponding capability nodes to generate a spectrum path table. The lightweight graph propagation layer introduces a functional spectrum connection mechanism, which includes: matching capability nodes in the node adjacency table with capability nodes in the spectrum path table, retaining capability nodes with path location and path source to form a spectrum neighborhood; determining the propagation order of capability nodes within the spectrum neighborhood according to the path location in the spectrum path table; passing the node representation of the personnel node to the capability node connected by the first interaction edge, and passing it level by level between capability nodes according to the propagation order, to the job node connected by the second interaction edge, generating a node propagation representation; The representation output layer extracts the personnel representation corresponding to the personnel node and the job representation corresponding to the job node from the node propagation representation.

[0011] Optionally, the construction of the capability transaction table based on the functional graph specifically includes: Locate the mastery relationship between personnel and capabilities in the aforementioned functional map, and obtain the personnel identifiers and capability names corresponding to the mastery relationship; Locate the relationship between capabilities and job requirements in the job function map, and obtain the job titles corresponding to the capability names; Group capability names according to the same job title and generate capability cluster identifiers; Locate the path position and path source corresponding to the capability name in the functional spectrum; Write the personnel identifier, capability name, job title, capability cluster identifier, path location, and path source into the same transaction record to generate a capability transaction table.

[0012] Optionally, the Eclat algorithm, which incorporates a vertical item cluster trace extraction mechanism, is used to perform vertical itemset mining and generate functional combinations, specifically as follows: Based on the capability transaction table, capability names are used as capability items, and personnel identifiers are used as transaction identifiers, and a vertical index table between capability items and transaction identifiers is established. Write the corresponding capability cluster identifier, job name, path location and path source for each capability item in the vertical index table to form an extended capability item; The vertical cluster trace extraction mechanism includes: performing cross-calculation on the transaction identifier columns corresponding to any two extended capability items to obtain cross-transaction identifiers; comparing the capability cluster identifiers of the two extended capability items for consistency during the cross-calculation process; comparing the job names of the two extended capability items for consistency during the consistency comparison process; and performing sequential matching on the path positions of the two extended capability items during the consistency comparison process; retaining only capability item combinations that have the same capability cluster identifier, the same job name, and whose path positions satisfy the sequential matching relationship. Associate the retained capability items with the corresponding cross-transaction identifiers to generate candidate function combinations; Repeatedly perform cross-calculation and consistency comparison on candidate function combinations to expand the number of capability items and generate function combinations.

[0013] Optionally, the step of repeatedly performing cross-calculation and consistency comparison on candidate function combinations to expand the number of capability items and generate function combinations specifically involves: Using the combination of capabilities in the candidate function combination as the basic combination, determine the cross-transaction identifier corresponding to the basic combination; Select capability items that are not included in the basic combination from the extended capability items, and determine the transaction identifier column corresponding to the capability items; The cross transaction identifier is cross-operated with the transaction identifier column to obtain the extended cross transaction identifier; Perform capability cluster identifier consistency comparison, job name consistency comparison, and path position order matching on the capability items in the basic combination and the capability items not included in the basic combination; Under the conditions that the capability cluster identifiers are consistent, the job names are consistent, and the path positions satisfy the order matching relationship, the capability items not included in the basic combination are incorporated into the basic combination to form an extended function combination; The extended function combination is associated with the extended cross-transaction identifier, and the extended function combination is used as the new basic combination. Cross-operation and consistency comparison are repeatedly performed in a way that one capability item is added in each round until there are no capability items that meet the requirements of consistent capability cluster identifier, consistent job name and path position order matching relationship, and a function combination is generated.

[0014] Optionally, the evaluation generation module specifically comprises: Based on personnel representation, determine the capability node representation corresponding to the target personnel; based on job representation, determine the job node representation corresponding to the target job. Calculate the matching value between the capability node representation and the job node representation to generate capability coverage; Based on the combination of capability items in the functional combination, the number of identical capability cluster identifiers and the number of consecutive path positions corresponding to the capability item combination are counted to generate capability synergy. Job suitability is generated based on the aforementioned capability coverage and capability synergy. The capability coverage, capability synergy, and job suitability are written into the same assessment record to generate capability assessment results.

[0015] The beneficial effects of this invention are: In the interactive generation module, a capability interaction graph is constructed and a functional spectrum is introduced. An ordered capability path is formed through hierarchical and dependency relationships. A functional spectrum connection mechanism is introduced in the lightweight graph propagation layer of the improved LightGCN model to limit the propagation range of nodes and determine the propagation order, so as to realize the orderly transmission of capability information along the structural path and improve the accuracy of personnel representation and job representation in expressing the capability structure. In the combination generation module, based on the capability transaction table and introducing the vertical item cluster trace extraction mechanism, the capability item combination is constrained in three dimensions: capability cluster identifier, job name, and path location. During the vertical item set mining process, capability item combinations that conform to the structural relationship are screened to obtain functional combinations with strong structural consistency, thereby improving the rationality and interpretability of capability combinations. In the assessment generation module, based on personnel representation, job representation, and functional combination, the ability coverage, ability synergy, and job suitability are calculated respectively. The overall ability matching degree and ability synergy relationship are comprehensively characterized to form the ability assessment results, thereby improving the stability, consistency, and refinement of professional ability assessment. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a knowledge graph-based vocational ability assessment system proposed in this invention. Figure 2 This is a data flow diagram of a knowledge graph-based vocational competency assessment system proposed in this invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0018] refer to Figures 1-2 A knowledge graph-based vocational competency assessment system includes: The competency ledger module is used to acquire the professional competency data of the target personnel, perform data cleaning and field unification processing on the professional competency data, and generate a competency ledger. The graph construction module is used to extract personnel entities, capability entities, and job entities based on the capability ledger, establish the mastery relationship between personnel and capabilities, and the demand relationship between capabilities and jobs, and construct a functional graph. The interaction generation module is used to map personnel entities, capability entities, and job entities as nodes based on the functional graph, and to map the mastery relationship between personnel and capability, as well as the requirement relationship between capability and job, as interaction edges to generate a capability interaction graph, and to construct a functional spectrum based on the hierarchical relationship and dependency relationship between capability entities. The representation generation module is used to input the capability interaction graph and the functional spectrum into the improved LightGCN model, and introduce the functional spectrum connection mechanism in the lightweight graph propagation layer to perform neighborhood information propagation and generate personnel representation and job representation. The combination generation module is used to construct a capability transaction table based on the functional map, and uses the Eclat algorithm with the introduction of the vertical item cluster trace disjunction mechanism to perform vertical itemset mining and generate functional combinations; The assessment generation module is used to calculate capability coverage, capability synergy, and job fit based on personnel representation, job representation, and function combination, and generate capability assessment results.

[0019] In this embodiment, the capability ledger module specifically includes: Obtain the professional competence data of the target personnel, which includes personnel competence records, learning records, task records, job records, and evaluation records; Perform duplicate record deletion, null value filling, invalid field removal, and field name standardization on professional ability data to generate standard records; Standardize the format of personnel identification, skill name, job title, recording time, and data source in standard records to generate standardized records; The standardized records are merged according to personnel identification, capability name, and job title to generate a capability ledger.

[0020] In this embodiment, the map construction module specifically comprises: Retrieve the personnel identifier field, ability name field, and job title field from the capability ledger. The personnel identifier field uses the unique number of the target personnel in the professional capability data. The ability name field uses the ability name after unified field processing in the standard record. The job title field uses the job title after unified field processing in the standard record. Duplicate personnel identifiers in the personnel identifier field are deduplicated. Each deduplicated personnel identifier generates a personnel entity. The data content of the personnel entity includes the personnel identifier and the corresponding data source. Duplicate capability names are removed from the capability name field. Each deduplicated capability name generates a capability entity. The data content of the capability entity includes the capability name and the corresponding record time. Duplicate job names are removed from the job name field. Each deduplicated job name generates a job entity. The data content of the job entity includes the job name and the corresponding data source. Retrieve records in the capability ledger that contain both personnel identifiers and capability names, map the personnel identifiers in the same record to personnel entities, and map the capability names in the same record to capability entities, thus establishing the relationship between personnel and capability control. Retrieve records in the competency ledger that contain both competency names and job titles, map competency names in the same record to competency entities, and map job titles in the same record to job entities, thus establishing a competency-job requirement relationship. Personnel entities, capability entities, and job entities are used as entity content in the function graph, and the mastery relationship between personnel and capabilities and the requirement relationship between capabilities and job requirements are used as relationship content in the function graph. The functional map is generated by associating and organizing the entities into personnel, capability, and job entities, as well as the relationships between personnel and capability and between capability and job requirements. The functional map includes entity content and relationship content. The entity content includes personnel, capability, and job entities, while the relationship content includes the relationships between personnel and capability and between capability and job requirements.

[0021] In this implementation, personnel entities, capability entities, and job entities are mapped as nodes based on the functional graph, and the mastery relationship between personnel and capabilities, as well as the requirement relationship between capabilities and job requirements, are mapped as interaction edges to generate a capability interaction graph, specifically: Retrieve personnel entities, capability entities, and job entities from the entity content of the functional graph. Personnel entities generate personnel nodes, capability entities generate capability nodes, and job entities generate job nodes. Write the personnel identifier from the personnel entity to the personnel node, write the capability name from the capability entity to the capability node, and write the job name from the job entity to the job node. Extract the mastery relationship between personnel and capabilities from the relational content of the functional graph, map the personnel entities in the mastery relationship to personnel nodes, map the capability entities in the mastery relationship to capability nodes, and establish the first interaction edge between personnel nodes and capability nodes. Extract the relationship between capabilities and job requirements from the functional graph, map capability entities in the requirement relationship to capability nodes, map job entities in the requirement relationship to job nodes, and establish a second interaction edge between capability nodes and job nodes. The data content of the first interaction edge includes personnel identifier, ability name, and mastery relationship marker; the data content of the second interaction edge includes ability name, job name, and requirement relationship marker. The capability interaction graph is generated by associating and organizing the nodes according to the connection correspondence between personnel nodes, capability nodes, job nodes, first interaction edges, and second interaction edges. The capability interaction graph includes node content and interaction edge content. The node content includes personnel nodes, capability nodes, and job nodes, and the interaction edge content includes the first interaction edge and the second interaction edge.

[0022] In this implementation, a functional hierarchy is constructed based on the hierarchical and dependency relationships between capability entities, specifically as follows: Capability entities are retrieved from the entity content of the functional map, and hierarchical and dependency relationships between capability entities are retrieved from the relation content of the functional map. Hierarchical relationships are generated from records in the capability ledger that indicate the hierarchy of capability categories, the hierarchy of capability levels, or the stratification of job capabilities. The data content of hierarchical relationships includes superior capability entities, subordinate capability entities, and hierarchical relationship markers. Dependency relationships are generated from records in the capability ledger that indicate the sequence of capability learning, prerequisites for task completion, or job capability succession relationships. The data content of dependency relationships includes prerequisite capability entities, successor capability entities, and dependency relationship markers. In a hierarchical relationship, the higher-level capability entity is defined as the starting point of the hierarchy, and the lower-level capability entity is defined as the ending point of the hierarchy, forming a hierarchical direction from the higher-level capability entity to the lower-level capability entity; in a dependency relationship, the preceding capability entity is defined as the starting point of the dependency, and the following capability entity is defined as the ending point of the dependency, forming a dependency direction from the preceding capability entity to the following capability entity. Connect capability entities with hierarchical relationships according to their hierarchical direction, and connect capability entities with dependent relationships according to their dependency direction to obtain the directional connection relationship between capability entities; in the directional connection relationship, the capability entity that does not have a preceding capability entity or a superior capability entity is used as the starting capability entity of the path, and the capability entity that does not have a following capability entity or a subordinate capability entity is used as the ending capability entity of the path. The starting capability entity is connected sequentially along the hierarchical and dependency directions to the ending capability entity, forming an ordered capability path. The path positions of the capability entities in the ordered capability path are recorded according to the order of connection. The path position of the starting capability entity is recorded as 1, and the path position of the next adjacent capability entity is increased by 1 based on the path position of the previous capability entity. Record the path source for the capability entities in the ordered capability path. The path source includes the source of hierarchical relationship and the source of dependency relationship. Associate and organize the ordered capability path, path location and path source to generate a functional spectrum.

[0023] In this embodiment, the improved LightGCN model specifically includes a graph input layer, a spectral pulse mapping layer, a lightweight graph propagation layer, and a representation output layer; The graph input layer receives personnel nodes, capability nodes, job nodes, first interaction edges, and second interaction edges from the capability interaction graph. It writes personnel identifiers for personnel nodes, capability names for capability nodes, and job names for job nodes. It records the connection correspondence between personnel nodes and capability nodes according to the first interaction edge and the connection correspondence between capability nodes and job nodes according to the second interaction edge, forming a connection record. Using personnel nodes, capability nodes, and job nodes as adjacency objects and the first and second interaction edges as adjacency criteria, it writes the adjacent nodes corresponding to each node into the same record row, generating a node adjacency table. The spectral mapping layer receives the ordered capability paths, path locations, and path sources from the functional spectral path, maps the capability entities in the ordered capability paths to capability nodes in the capability interaction graph, arranges the capability nodes in ascending order of path location, and writes the path source into the corresponding capability node to generate the spectral path table. The lightweight graph propagation layer configures initial node representations for personnel nodes, capability nodes, and job nodes. The initial node representation is represented by the numerical vector corresponding to the node identifier. The capability nodes in the node adjacency list are compared with the capability nodes in the spectral path list one by one. Capability nodes that exist in both the node adjacency list and the spectral path list are retained to form a spectral neighborhood. The functional spectrum connection mechanism limits the propagation range of the lightweight graph propagation layer through the spectrum neighborhood, limits the propagation order of the lightweight graph propagation layer through the path position in the spectrum path table, and records the source of capability nodes in hierarchical and dependency relationships through the path source. Starting from the personnel node, the initial node representation of the personnel node is passed to the capability node connected by the first interaction edge to obtain the first layer representation of the capability node; according to the path position in the spectral path table, the node representation of the capability node at the previous position is passed to the capability node at the next position, and the node representation received by the capability node at the next position is added to the existing representation of the capability node at the next position to obtain the spectral propagation representation of the capability node. When a capability node has multiple neighboring nodes, the node representations transmitted by the multiple neighboring nodes are added together, and the average of the added results is taken according to the number of neighboring nodes to obtain the neighborhood propagation representation of the capability node. The spectral propagation representation and the neighborhood propagation representation of the capability node are added together to obtain the node propagation representation of the capability node; the node propagation representation of the capability node is passed to the post node connected by the second interaction edge, and the average is taken according to the number of capability nodes received by the post node to obtain the node propagation representation of the post node. The representation output layer selects the node propagation representation corresponding to the personnel node from the node propagation representation to generate the personnel representation; it also selects the node propagation representation corresponding to the job node from the node propagation representation to generate the job representation.

[0024] In this embodiment, the improved LightGCN model maintains the same overall structure as the original LightGCN model. Both models learn node embedding representations based on graph structure data, use node adjacency relationships as the basis for information propagation, and acquire node representations through multi-layer propagation. In terms of input, both models use a graph structure composed of nodes and interaction edges as the computational object. During propagation, both models update node representations by accumulating neighborhood node information. In the output stage, both models extract the embedding vector of the target node as the representation result. In terms of improvement, a functional spectral path connection mechanism is introduced, adding a spectral path path table in addition to the node adjacency table as a source of propagation constraints. This method transforms the hierarchical and dependency relationships between capability entities into ordered capability paths, and limits the propagation order between capability nodes by path location. The path source is embedded as a propagation source marker in the node representation process. In terms of propagation scope, it changes from full-neighborhood propagation based on adjacency to spectral-neighborhood-limited propagation, retaining only capability nodes that simultaneously satisfy node adjacency and spectral-path constraints to participate in propagation, thereby reducing interference from irrelevant nodes. In terms of propagation order, it changes from unordered neighborhood propagation to ordered propagation based on path location, allowing node information to be transmitted hierarchically along the capability structure, avoiding cross-level information mixing. Regarding propagation content, it... The path source in the spectral path table is embedded with node propagation representations, enabling node representations to simultaneously include structural relationship information and path source information. Regarding node representation updates, a superposition process of spectral propagation representations and neighborhood propagation representations is introduced. By accumulating and averaging node representations from different path locations, the stability of node representations is enhanced. In the job node generation stage, by aggregating and averaging the propagation representations of multiple capability nodes, the job node representations can reflect multi-capability collaborative relationships. In terms of technical effectiveness, by introducing a functional spectral path integration mechanism, capability information propagation is given structural and sequential constraints, improving personnel representation and... The job representation enhances the ability to express hierarchical and dependency relationships of abilities, improves the accuracy of ability coverage calculation and ability synergy identification in the ability assessment process, reduces the interference of irrelevant ability information on the results, and enhances the stability and consistency of the assessment results. By limiting the propagation path and propagation order, the node propagation process conforms to the actual evolutionary relationship of occupational abilities, improving the model's expressive and generalization abilities in complex ability relationship scenarios. By integrating spectral path information and neighborhood structure information, the improved LightGCN model enhances its structural perception ability while maintaining lightweight computational characteristics, achieving refined modeling of occupational ability structure.

[0025] In this implementation, a capability transaction table is constructed based on the functional graph, specifically as follows: From the relationship content of the job function map, locate the mastery relationship between personnel and capabilities, and extract the personnel identifiers and capability names from the mastery relationship; from the relationship content of the job function map, locate the requirement relationship between capabilities and positions, and extract the capability names and position names from the requirement relationship; Using the ability name in the mastery relationship as the matching object, find the job name corresponding to the same ability name in the demand relationship, and form a corresponding record consisting of personnel identifier, ability name and job name; Merge competency names with the same job title, group competency names under the same job title into the same competency cluster, and write the same competency cluster identifier for the same competency cluster; Find the capability entity that matches the capability name in the functional spectrum, and obtain the path location and path source corresponding to the capability entity; Write the personnel identifier, capability name, job title, capability cluster identifier, path location, and path source into the same transaction record; the personnel identifier in the transaction record is the transaction affiliation of the target personnel, the capability name is the transaction content, the job title is the job direction, the capability cluster identifier is the capability classification mark, the path location is the position mark of the capability in the ordered capability path, and the path source is the source of the hierarchical relationship or dependency relationship corresponding to the capability. Multiple transaction records are grouped according to personnel identifiers. Transaction records under the same personnel identifier constitute the corresponding capability transactions for that personnel. Multiple capability transactions are arranged in a tabular format to form a capability transaction table. The capability transaction table includes personnel identifier, capability name, job title, capability cluster identifier, path location, and path source fields. Each row in the capability transaction table corresponds to one transaction record, and each transaction record corresponds to one capability content of the target personnel under a job position.

[0026] In this embodiment, the Eclat algorithm, which incorporates a vertical item cluster trace disjunction mechanism, is used to perform vertical itemset mining and generate functional combinations, specifically: Based on the capability transaction table, capability names and personnel identifiers are extracted. The capability name is determined as the capability item, and the personnel identifier is determined as the transaction identifier. Transaction identifiers corresponding to the same capability item are grouped into the same transaction identifier column, forming a correspondence between capability items and transaction identifier columns. The capability item, transaction identifier column, capability cluster identifier, job name, path location, and path source are written into the same index record to form a vertical index table. In the vertical index table, capability items containing capability cluster identifier, job name, path location, and path source are determined as extended capability items. The vertical item cluster trace extraction mechanism includes vertical item cross-intersection, cluster item comparison, job comparison, and path extraction. Vertical item cross-intersection involves selecting the transaction identifier columns corresponding to any two extended capability items, retaining the transaction identifiers that exist in both transaction identifier columns simultaneously, and obtaining the cross-transaction identifier. Cluster item comparison involves comparing the capability cluster identifiers corresponding to two extended capability items. If the capability cluster identifiers are the same, the corresponding capability item combination is retained; if the capability cluster identifiers are different, the corresponding capability item combination is deleted. Job comparison involves comparing the job names corresponding to two extended capability items. If the job names are the same, the corresponding capability item combination is retained; if the job names are different, the corresponding capability item combination is deleted. Path extraction involves comparing the path positions corresponding to two extended capability items. If the two path positions are adjacent in the same ordered capability path, the corresponding capability item combination is retained; if the two path positions are not adjacent, the corresponding capability item combination is deleted. Ability item combinations that simultaneously satisfy the following conditions are identified as valid ability item combinations: having the same ability cluster identifier, the same job name, and adjacent path positions. Valid ability item combinations are associated with cross-transaction identifiers to generate candidate function combinations. When adding extended ability items that are not included to a candidate function combination, the cross-transaction identifiers corresponding to the candidate function combination are cross-operated with the transaction identifier columns corresponding to the extended ability items that are not included to obtain extended cross-transaction identifiers. For the capability items in the candidate function combination and the extended capability items not included, perform capability cluster identification comparison, job name comparison, and path location comparison; if the capability cluster identification is the same, the job name is the same, and the path location is adjacent in the same ordered capability path, then the extended capability items not included are merged into the candidate function combination to form an extended function combination; perform cross calculation and consistency comparison repeatedly on the candidate function combination to count the number of extended capability items and generate the function combination.

[0027] In this embodiment, cross-calculation and consistency comparison are repeatedly performed on candidate function combinations to expand the number of capability items and generate function combinations, specifically as follows: The capability combination in the candidate function combination is used as the basic combination. The basic combination includes capability item, cross-task identifier, capability cluster identifier, job name, path location and path source. The cross-task identifier is obtained from the basic combination. The cross-task identifier is composed of the personnel identifier that corresponds to all capability items in the basic combination. Filter out extended capability items not included in the basic combination from the vertical index table, and obtain the transaction identifier column, capability cluster identifier, job name, path location and path source corresponding to the extended capability item; Perform cross-operation between the cross-transaction identifier corresponding to the basic combination and the transaction identifier column corresponding to the extended capability item, retaining the personnel identifier that exists in both the cross-transaction identifier and the transaction identifier column to obtain the extended cross-transaction identifier; perform a consistency comparison between the capability cluster identifier corresponding to the basic combination and the capability cluster identifier corresponding to the extended capability item. If they match, retain the extended capability item; otherwise, delete the extended capability item; perform a consistency comparison between the job name corresponding to the basic combination and the job name corresponding to the extended capability item. If they match, retain the extended capability item; otherwise, delete the extended capability item. The maximum path position in the basic combination is sequentially matched with the path position corresponding to the extended capability item. If the path position corresponding to the extended capability item is equal to the maximum path position in the basic combination plus 1, the extended capability item is retained. If the path position corresponding to the extended capability item is not equal to the maximum path position in the basic combination plus 1, the extended capability item is deleted. The retained extended capability items are incorporated into the basic combination to form an extended function combination. The extended function combination includes the capability items in the basic combination, the incorporated extended capability items, the extended cross-transaction identifier, the capability cluster identifier, the job title, the path location, and the path source. The extended function combination is used as the new basic combination. In each round, one extended capability item that meets the requirements of consistent capability cluster identifier, consistent job name, and continuous path position is merged from the vertical index table. If there is no extended capability item that meets the requirements of consistent capability cluster identifier, consistent job name, and continuous path position in the vertical index table, the extension is stopped, and the extended function combination obtained when the extension is stopped is determined as the function combination. The function combination includes capability item combination, cross-transaction identifier, capability cluster identifier, job name, path position, and path source.

[0028] In this embodiment, the evaluation generation module specifically comprises: Select the personnel node representation corresponding to the target personnel from the personnel representation, and obtain the capability node representation corresponding to the capability node propagation source from the personnel node representation; select the job node representation corresponding to the target job from the job representation; multiply the capability node representation and the job node representation by each element along the same dimension, and add the results of the element-wise multiplication to obtain the capability matching value; divide the capability matching value by the number of non-zero dimensions in the job node representation to obtain the capability coverage. Obtain capability item combinations, capability cluster identifiers, and path positions from the functional combinations; count the number of capability items with the same capability cluster identifier within the same functional combination to obtain the number of capability clusters that are consistent; arrange the capability items within the same functional combination in ascending order of path position, and count the number of times the path position difference between adjacent capability items is equal to 1 to obtain the number of consecutive path positions. Add the number of consistent capability clusters and the number of consecutive path locations, and divide by the number of capability items within the function combination to obtain the capability synergy; add the capability coverage and capability synergy and divide by 2 to obtain the job fit. Write the personnel identifier corresponding to the target personnel, the job name corresponding to the target position, the capability coverage, capability synergy, and job suitability into the same assessment record to generate capability assessment results.

[0029] Example 1: To verify the feasibility of this invention in practice, it was applied to a job competency assessment scenario in a large equipment manufacturing enterprise. This enterprise had positions in R&D design, process development, quality inspection, equipment maintenance, and project management, with a total of 186 personnel participating in the assessment. Data collected included 12,840 personnel competency records, 3,560 learning records, 6,290 task records, 420 job records, and 2,180 evaluation records. Traditional methods mainly rely on manual scoring and item-by-item comparison with job competency tables, which suffers from inconsistent competency names, difficulty in quantifying project experience, and a lack of structured correlation between job requirements and personnel competencies. For example, the same competency might be recorded as "process optimization," "process improvement," or "process enhancement" in different departments, easily splitting into different competency items during manual assessment, leading to an underestimation of the true competency coverage of personnel.

[0030] In application of this invention, the competency ledger module first cleanses and unifies the fields of the occupational competency data, unifying duplicate records, null records, and invalid fields to generate a competency ledger. The competency graph construction module extracts personnel entities, competency entities, and job entities from the competency ledger, establishing the mastery relationship between personnel and competencies, as well as the demand relationship between competencies and jobs, thus constructing a functional graph. The interaction generation module further transforms the functional graph into a competency interaction graph and constructs a functional hierarchy based on the hierarchical and dependency relationships between competency entities. For example, "mechanical drawing" is a basic competency, "process route design" is a professional competency, and "complex assembly process development" is a job competency; these three form an ordered competency path through the functional hierarchy.

[0031] The representation generation module inputs the capability interaction graph and functional spectrum into the improved LightGCN model, introducing a functional spectrum connection mechanism in the lightweight graph propagation layer to ensure that personnel capability information is transmitted along an ordered capability path, generating personnel and job representations. The combination generation module constructs a capability transaction table based on the functional graph and uses the Eclat algorithm with a vertical item cluster trace extraction mechanism to perform vertical itemset mining, generating functional combinations. The evaluation generation module calculates capability coverage, capability synergy, and job fit based on personnel representation, job representation, and functional combinations, generating capability evaluation results. During implementation, the company selected 60 employees to complete job fit assessments using both traditional methods and this invention, and compared the evaluation results with actual performance, job competency feedback, and project completion status three months later.

[0032] Table 1 Comparison of the effects of vocational ability data processing and map construction

[0033] As shown in Table 1, this invention has significant advantages in vocational competency data processing and functional map construction. Traditional methods rely on manual compilation, which easily leads to inconsistent competency names and missing relationships when dealing with records from multiple departments, positions, and sources. This invention, through its competency ledger module, completes data cleaning and field unification, achieving a competency name consistency rate of 96.8% and effectively reducing the duplication of synonymous competencies. The map construction module associates and organizes personnel entities, competency entities, and position entities, making the relationships between personnel and competency mastery, and between competency and position requirements, more complete. The number of effective relationships in the functional map increases from 4260 to 6885, indicating that the system can discover competency relationships that are not stably expressed in manually generated tables. The average generation time for a single competency ledger is reduced from 12.6 minutes to 3.8 minutes, further demonstrating that this invention can reduce the workload of manual compilation and improve the efficiency of front-end data preparation for vocational competency assessment.

[0034] Table 2 Comparison of Professional Competency Assessment Results and Actual Job Performance

[0035] As shown in Table 2, the competency assessment results generated by this invention have a higher consistency with actual job performance. Traditional methods mainly rely on human experience and static indicators, making it difficult to accurately reflect the hierarchical and synergistic relationships between competencies. Therefore, the correlation coefficient between job fit and actual performance is only 0.62. This invention generates personnel and job representations with functional spectrum constraints by improving the LightGCN model, and generates functional combinations by introducing the Eclat algorithm with a vertical term cluster trace extraction mechanism, so that the assessment results simultaneously reflect competency coverage and competency synergy. The competency gap identification accuracy reaches 91.3%, indicating that the system can more accurately identify competency items in job requirements that are not fully covered by the target personnel. The job recommendation hit rate and project task competency prediction accuracy are also significantly improved, indicating that this invention can not only be used for static competency scoring, but also provide a more reliable assessment basis for job allocation, training plans, and project personnel arrangements.

[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A knowledge graph-based professional competence evaluation system, characterized in that, include: The competency ledger module is used to acquire the professional competency data of the target personnel, perform data cleaning and field unification processing on the professional competency data, and generate a competency ledger. The graph construction module is used to extract personnel entities, capability entities, and job entities based on the capability ledger, establish the mastery relationship between personnel and capabilities, and the demand relationship between capabilities and jobs, and construct a functional graph. The interaction generation module is used to map personnel entities, capability entities, and job entities as nodes based on the functional graph, and to map the mastery relationship between personnel and capability, as well as the requirement relationship between capability and job, as interaction edges to generate a capability interaction graph, and to construct a functional spectrum based on the hierarchical relationship and dependency relationship between capability entities. The representation generation module is used to input the capability interaction graph and functional spectrum into the improved LightGCN model, introduce a functional spectrum connection mechanism in the lightweight graph propagation layer to perform neighborhood information propagation, and generate personnel representation and job representation. The combination generation module is used to construct a capability transaction table based on the functional map, and uses the Eclat algorithm with the introduction of the vertical item cluster trace disjunction mechanism to perform vertical itemset mining and generate functional combinations; The assessment generation module is used to calculate capability coverage, capability synergy, and job fit based on the personnel representation, job representation, and function combination, and generate capability assessment results.

2. The knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The capability ledger module specifically includes: Obtain the professional competence data of the target personnel, including personnel competence records, learning records, task records, job records, and evaluation records; The professional skills data is processed by deleting duplicate records, filling in empty records, removing invalid fields, and unifying field names to generate standard records. The personnel identifiers, skill names, job titles, recording times, and data sources in the standard records are formatted uniformly to generate standardized records; The standardized records are merged according to personnel identification, capability name, and job title to generate a capability ledger.

3. The knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The map construction module is specifically as follows: Read the personnel identifier field, capability name field, and job name field from the capability ledger, and generate the corresponding personnel entity, capability entity, and job entity respectively; The capability ledger establishes a relationship between personnel and capabilities based on the correspondence between personnel identification and capability name. Establish the relationship between capabilities and job requirements in the capability ledger by recording the correspondence between capability names and job titles; The personnel entities, capability entities, and job entities, as well as the relationships between personnel and capability and between capability and job requirements, are linked and organized to generate a functional map.

4. The knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The process involves mapping personnel entities, capability entities, and job entities to nodes based on a functional graph, and mapping the mastery relationship between personnel and capabilities, as well as the requirement relationship between capabilities and job positions, to generate a capability interaction graph. Specifically: In the functional graph, locate personnel entities, capability entities, and job entities, and generate corresponding personnel nodes, capability nodes, and job nodes respectively; Write the personnel identifier in the personnel node, the ability name in the ability node, and the job name in the job node; In the functional graph, the first interaction edge between personnel nodes and capability nodes is established based on the relationship between personnel and capability mastery. In the functional graph, a second interaction edge is established between capability nodes and job nodes based on the relationship between capability and job requirements. The personnel nodes, capability nodes, job nodes, first interaction edges, and second interaction edges are combined to generate a capability interaction graph.

5. A knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The construction of the functional spectrum based on the hierarchical and dependency relationships between capability entities is specifically as follows: Match the hierarchical relationships and dependencies between capability entities in the functional graph; The higher-level and lower-level capability entities in the hierarchical relationship are respectively defined as the hierarchical start point and hierarchical end point to form the hierarchical direction; The preceding and following capability entities in the dependency relationship are determined as the dependency start point and dependency end point, respectively, forming the dependency direction; Connect capability entities according to the hierarchical and dependency directions to form an ordered capability path; Record the path location and path source for the capability entities in the ordered capability path to generate a functional spectrum.

6. The knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The improved LightGCN model specifically includes a graph input layer, a spectral pulse mapping layer, a lightweight graph propagation layer, and a representation output layer; The graph input layer is based on personnel nodes, capability nodes, job nodes, first interaction edges and second interaction edges in the capability interaction graph. It establishes connection records between personnel nodes and capability nodes according to the first interaction edge, establishes connection records between capability nodes and job nodes according to the second interaction edge, and generates a node adjacency table from the connection records. The spectrum mapping layer, based on the ordered capability paths, path positions, and path sources in the functional spectrum, arranges the capability nodes in the ordered capability paths according to their path positions and writes the path sources into the corresponding capability nodes to generate a spectrum path table. The lightweight graph propagation layer introduces a functional spectrum connection mechanism, which includes: matching capability nodes in the node adjacency table with capability nodes in the spectrum path table, retaining capability nodes with path location and path source, and forming a spectrum neighborhood; The propagation order of capability nodes in the spectral neighborhood is determined according to the path position in the spectral path table; the node representation of the personnel node is passed to the capability node connected by the first interaction edge, and then passed level by level between capability nodes according to the propagation order, and then to the job node connected by the second interaction edge to generate the node propagation representation. The representation output layer extracts the personnel representation corresponding to the personnel node and the job representation corresponding to the job node from the node propagation representation.

7. The knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The construction of the capability transaction table based on the functional graph is as follows: Locate the mastery relationship between personnel and capabilities in the aforementioned functional map, and obtain the personnel identifiers and capability names corresponding to the mastery relationships; Locate the relationship between capabilities and job requirements in the job function map, and obtain the job titles corresponding to the capability names; Group capability names according to the same job title and generate capability cluster identifiers; Locate the path position and path source corresponding to the capability name in the functional spectrum; Write the personnel identifier, capability name, job title, capability cluster identifier, path location, and path source into the same transaction record to generate a capability transaction table.

8. The knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The Eclat algorithm, which incorporates a vertical item cluster trace extraction mechanism, is used to perform vertical itemset mining and generate functional combinations, specifically as follows: Based on the capability transaction table, capability names are used as capability items, and personnel identifiers are used as transaction identifiers, and a vertical index table between capability items and transaction identifiers is established. Write the corresponding capability cluster identifier, job name, path location and path source for each capability item in the vertical index table to form an extended capability item; The vertical cluster trace extraction mechanism includes: performing cross calculations on the transaction identifier columns corresponding to any two extended capability items to obtain the cross transaction identifier; During the cross-calculation process, the capability cluster identifiers of the two extended capability items are compared for consistency; during the consistency comparison process, the job names of the two extended capability items are compared for consistency; during the consistency comparison process, the path positions of the two extended capability items are matched in order; only capability item combinations with consistent capability cluster identifiers, consistent job names, and path positions that satisfy the order matching relationship are retained. Associate the retained capability items with the corresponding cross-transaction identifiers to generate candidate function combinations; Repeatedly perform cross-calculation and consistency comparison on candidate function combinations to expand the number of capability items and generate function combinations.

9. A knowledge graph-based vocational ability assessment system according to claim 8, characterized in that, The process of repeatedly performing cross-calculation and consistency comparison on candidate function combinations to expand the number of capability items and generate function combinations specifically involves: Using the combination of capability items in the candidate function combination as the basic combination, determine the cross-transaction identifier corresponding to the basic combination; Select capability items that are not included in the basic combination from the extended capability items, and determine the transaction identifier column corresponding to the capability items; The cross transaction identifier is cross-operated with the transaction identifier column to obtain the extended cross transaction identifier; Perform capability cluster identifier consistency comparison, job name consistency comparison, and path position order matching on the capability items in the basic combination and the capability items not included in the basic combination; Under the conditions that the capability cluster identifiers are consistent, the job names are consistent, and the path positions satisfy the order matching relationship, the capability items not included in the basic combination are incorporated into the basic combination to form an extended function combination; The extended function combination is associated with the extended cross-transaction identifier, and the extended function combination is used as the new basic combination. Cross-operation and consistency comparison are repeatedly performed in a way that one capability item is added in each round until there are no capability items that meet the requirements of consistent capability cluster identifier, consistent job name and path position order matching relationship, and a function combination is generated.

10. A knowledge graph-based vocational ability assessment system according to claim 1, characterized in that, The evaluation generation module is specifically as follows: Based on personnel representation, determine the capability node representation corresponding to the target personnel; based on job representation, determine the job node representation corresponding to the target job. Calculate the matching value between the capability node representation and the job node representation to generate capability coverage; Based on the combination of capability items in the functional combination, the number of identical capability cluster identifiers and the number of consecutive path positions corresponding to the capability item combination are counted to generate capability synergy. Job suitability is generated based on the aforementioned capability coverage and capability synergy. The capability coverage, capability synergy, and job suitability are written into the same assessment record to generate capability assessment results.

Images