Method and electronic apparatus for matching a user with a job position using credential data

The use of Verifiable Credentials to generate standardized skill data and semantic matching addresses the subjectivity of resume-based evaluations, ensuring fair and transparent recruitment by objectively matching candidates with jobs, reducing biases and improving explainability.

KR103045213B1Active Publication Date: 2026-11-06BLOCK SPOON CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
KR1020260032865
Authority / Receiving Office
KR · KR
Patent Type
Patents
Current Assignee / Owner
Priority Date
2026-02-13
Filing Date
2026-02-23
Publication Date
2026-11-06
Estimated Expiration
2046-02-23

AI Technical Summary

Technical Problem

Existing recruitment matching systems rely heavily on resumes and cover letters, which are subjective and vary in evaluation based on writing style, leading to inconsistencies and lack of explainability, particularly disadvantaging international students and entry-level talent, and creating biases in candidate exposure and job matching.

Method used

A method utilizing Verifiable Credentials (VCs) to generate standardized skill data, including competency identifiers and proficiency levels, and a semantic matching system to evaluate candidates fairly and explainably by correlating candidate and job information.

Benefits of technology

Enables fair and transparent recruitment matching by objectively evaluating candidates' competencies, reducing biases, and providing clear explanations for recommendation results, thus enhancing trust and accuracy in candidate-job matching.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 112026021758511-PAT00033_ABST
    Figure 112026021758511-PAT00033_ABST
Patent Text Reader

Abstract

The present invention relates to a method for matching a user with a job using credential data. This method acquires credential data including issuing agency and achievement information for one or more candidates, and based on this data, generates standard skill data regarding the candidate including a competency identifier, proficiency level, and evidence strength according to a standard competency classification system. Based on the generated standard skill data and job posting data, a candidate group semantically related to job information is derived, and a candidate ranking is determined by calculating the suitability of each candidate. Subsequently, the method includes the step of generating a final ranking by adjusting the candidate ranking to satisfy pre-established fairness criteria, and generating recommendation evidence information for the final ranking to generate a candidate recommendation result report.
Need to check novelty before this filing date? Find Prior Art

Description

Technology Field

[0001] The present disclosure relates to a method and electronic device for matching a user with a job using credential data. Background Technology

[0002] Existing recruitment matching systems primarily operate based on resumes, cover letters, and the text of job postings. While this approach may be effective to a certain extent for large-scale recommendations, it has limitations in reliably comparing and evaluating candidates' actual capabilities and explaining the rationale behind those assessments. Since resumes and cover letters rely heavily on expressiveness and language proficiency, evaluations can vary even among candidates with identical capabilities depending on their writing style. International students and entry-level talent are particularly at a disadvantage in this structure. Furthermore, recommendations centered on click and application logs concentrate exposure on existing popular postings and candidates, leading to issues such as "cold starts" and popularity bias regarding new candidates and postings. In addition, because it is difficult to clearly explain "why a particular candidate is suitable" for recommendation results, there are limitations in securing the trust of companies and organizations. To address these problems, a new recruitment matching method utilizing verifiable credential data is required. This offers the possibility to evaluate candidates' capabilities in a more objective and standardized manner, while laying the foundation to enhance fairness and explainability. The problem to be solved

[0003] The purpose of one embodiment is to generate standard skill data including competency identifiers, proficiency levels, and evidence strength based on credential data including issuing agency and achievement information, and to provide fair and explainable recruitment matching by evaluating the semantic relevance between a candidate and a job through this. means of solving the problem

[0004] A method for matching a user with a job using credential data according to one embodiment involves acquiring credential data including issuing agency and achievement information for one or more candidates, generating standard skill data regarding the candidate including a competency identifier, proficiency level, and evidence strength according to a pre-established standard competency classification system based on the credential data, and deriving a candidate group semantically related to job information among one or more candidates based on the standard skill data and job posting data. Additionally, the suitability of each candidate is calculated based on the standard skill data and job posting data of the candidate group, the candidates within the candidate group are sorted according to the suitability of each candidate to determine the candidate ranking, and the candidate ranking is adjusted based on the candidate ranking to satisfy pre-established fairness criteria to generate a final ranking. Furthermore, recommendation evidence information for the final ranking is generated by correlating the standard skill data of the candidate group with the job posting data, and a candidate recommendation result report is generated by combining the final ranking and the recommendation evidence information. Effects of the invention

[0005] The recruitment matching method and system according to the present invention can perform matching between candidates and jobs more fairly and explainably by generating standard skill data based on credential data including issuing agency and achievement information. According to one embodiment, by representing candidates and jobs in the same semantic space through natural language processing-based competency extraction and embedding similarity calculation, candidates semantically suitable for a job can be quickly selected from a large pool of candidates. Furthermore, through a fairness correction procedure, deviations in recommendation and exposure between groups can be mitigated, thereby generating a fair final ranking that is not biased toward a specific group. Moreover, by enhancing the explainability of recommendation results, a basis for rational decision-making can be provided to both companies and candidates. The effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art to which the present invention belongs. Brief explanation of the drawing

[0006] Figure 1 schematically illustrates the overall processing structure of a candidate-job matching system using credential data according to one embodiment. FIG. 2 is a diagram illustrating the internal processing structure of a data standardization layer according to one embodiment. FIG. 3a is a diagram illustrating the internal processing structure of a matching layer according to one embodiment. FIG. 3b is a drawing for explaining in detail the operation of a matching layer according to one embodiment. FIG. 4a is a diagram illustrating the internal processing structure of an output layer according to one embodiment. FIG. 4b is a diagram for explaining in detail the operation of an output layer according to one embodiment. FIG. 5 shows an example of an electronic device according to one embodiment. Specific details for implementing the invention

[0007] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Accordingly, actual implementations are not limited to the specific embodiments disclosed, and the scope of this specification includes modifications, equivalents, or substitutions included in the technical concept described by the embodiments.

[0008] Terms such as "first" or "second" may be used to describe various components, but these terms should be interpreted solely for the purpose of distinguishing one component from another. For example, the first component may be named the second component, and similarly, the second component may be named the first component.

[0009] When it is stated that a component is "connected" to another component, it should be understood that it may be directly connected to or joined to that other component, or that there may be other components in between.

[0010] Singular expressions include plural expressions unless the context clearly indicates otherwise. In this document, phrases such as “A or B,” “at least one of A and B,” “at least one of A or B,” “A, B or C,” “at least one of A, B and C,” and “at least one of A, B, or C” may each include any one of the items listed together with the corresponding phrase, or all possible combinations thereof. In this specification, terms such as “comprising” or “having” are intended to designate the existence of the described feature, number, step, action, component, part, or combination thereof, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0011] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this specification.

[0012] Hereinafter, embodiments will be described in detail with reference to the attached drawings. In the description with reference to the attached drawings, identical components are given the same reference numeral regardless of the drawing number, and redundant descriptions thereof will be omitted.

[0014] Figure 1 schematically illustrates the overall processing structure of a candidate-job matching system using credential data according to one embodiment.

[0015] In one embodiment, the present invention can solve technical challenges aimed at overcoming the structural limitations of existing resume and cover letter-centered recruitment matching methods, such as expression deviation, lack of verification, and inexplicability. Existing resume and cover letter-centered recruitment matching methods have the limitation that it is difficult to objectively verify a candidate's actual capabilities. In existing methods, even candidates with the same capabilities may have different evaluation results depending on their writing style or language ability. Existing methods fail to provide clear explanations for recommendation results, which may limit the ability to secure the trust of companies and institutions. The present invention can utilize competency data based on Verifiable Credentials (VCs) as core inputs, wherein the issuing institution, recipient, achievement criteria, and verification basis are specified. The present invention can convert VC-based competency data into standardized competency records using natural language processing and large-scale language models. The present invention can compare candidates and job roles within the same semantic space using these standardized competency records. The present invention can enhance the accuracy and reliability of candidate-job matching through embedding similarity calculation and interaction-based precise re-ranking. The present invention can ensure the reliability and transparency of recommendation results by correcting for fairness and generating explanatory grounds for matching results. Based on verifiable competency data, the present invention can realize fair and explainable recruitment matching.

[0016] In one embodiment, the present invention can reliably evaluate a candidate's actual competency by utilizing objective competency data based on Verifiable Credentials (VCs) that specify the issuing organization and achievement information. The VC data may include objective information such as the issuing organization, achievement standards, and grounds for verification. The VC data can record the candidate's competency in a verifiable manner. Unlike the subjective descriptions found in existing resumes and self-introductions, the VC data can provide competency information in a standardized structure. The VC data can be converted into standard competency identifiers, proficiency levels, and evidence strength through natural language processing and large-scale language models. An evaluation method based on VC data can reliably compare and analyze a candidate's actual competency. The VC data can integrate and utilize multiple competency information generated by various issuing organizations. Since the VC data enables evaluation according to consistent standards for each candidate, evaluation deviations can be minimized.

[0017] In at least one embodiment of the present disclosure, the method may obtain Verifiable Credential (VC) data including issuing authority and achievement information for one or more candidates. The method may collect VC data from an external issuing authority or an internal system. The method may parse the issuing authority information and achievement information included in the VC data. The method may separate and store VC data corresponding to each of a plurality of candidates. The method may verify the metadata of the VC data to select only valid data.

[0018] In one embodiment, VC data may correspond to each of one or more candidates. VC data may be uniquely mapped for each candidate. VC data may be managed in conjunction with a candidate identification key. VC data may be designed to enable parallel processing for multiple candidates. VC data may include detailed information for each candidate, such as the issuing authority, achievement standards, and verification basis.

[0019] In at least one embodiment of the present disclosure, the electronic device (100) may include a data standardization layer (110), a matching layer (120), and an output layer (130). The data standardization layer (110) may convert input credential data into standardized competency data. The matching layer (120) may perform candidate-job matching using standardized competency data and job posting data. The output layer (130) may generate final ranking and descriptive information regarding the matching results. Each layer operates independently, and data transfer between layers may be performed through a standardized interface. The system structure may be designed to facilitate scalability and modularization.

[0020] In one embodiment, the electronic device (100) can collect multiple data necessary for matching processing. The electronic device (100) can collect various input data, such as credential (VC) data, job posting data, and standard skill data. The electronic device (100) can acquire data by linking with an external system or an internal database. When collecting data, the electronic device (100) can verify the data format and specifications. The electronic device (100) can preprocess the collected data to match the inter-layer transfer format.

[0021] In at least one embodiment of the present disclosure, the data standardization layer (110) may perform the role of converting credential data into structured standard competency data. The data standardization layer (110) may extract competency candidates from natural language-based text included in the VC data. The data standardization layer (110) may map the extracted competency candidates to competency identifiers of a standard competency classification system. The data standardization layer (110) may calculate the proficiency level and evidence strength corresponding to the competency identifiers. The data standardization layer (110) may generate a standard skill record (SSR) including competency identifiers, proficiency levels, evidence strength, and source information. The data standardization layer (110) may store the generated SSR data in a repository and transmit it to a subsequent matching layer.

[0022] In one embodiment, the matching layer (120) can perform candidate-job matching using standardized SSR data and job posting data. The matching layer (120) can convert the SSR data and job posting data into vectors in a shared semantic space. The matching layer (120) can derive a semantically related group of candidates by calculating similarity between the vectors. The matching layer (120) can calculate the suitability of each candidate by applying an interaction-based model to the derived group of candidates. When calculating the suitability, the matching layer (120) can reflect whether essential competencies are met and the level of proficiency. The matching layer (120) can determine the ranking of candidates based on the calculated suitability. The matching layer (120) can transmit the ranking results and supporting information to the output layer.

[0023] In at least one embodiment of the present disclosure, the output layer (130) may generate a final ranking and descriptive information thereto. The output layer (130) may perform fairness correction processing on the ranking results provided by the matching layer. The output layer (130) may classify candidates included in the candidate ranking according to group criteria and calculate a recommendation rate or exposure rate by group. The output layer (130) may adjust scores or order so that the deviation between groups satisfies the fairness criteria. The output layer (130) may generate recommendation basis information by correlating SSR data with job posting data. The output layer (130) may generate a candidate recommendation result report by combining the final ranking and recommendation basis information. The output layer (130) may provide the result report through an enterprise dashboard, an external system, or an API.

[0024] In one embodiment, the electronic device (100) may collect multiple data necessary for matching processing. The electronic device (100) may include different types of data, such as credential (VC) data, job posting data, and standard skill data (standard competency classification system). Depending on the source and characteristics of each data, the electronic device (100) may collect data from various data sources, such as external issuing agencies, recruitment platforms, and standard classification system repositories. When collecting data, the electronic device (100) may verify the format and structure of each data according to predefined specifications. The electronic device (100) may manage data quality by checking for omissions, errors, or duplicates in the collected data. After collecting data, the electronic device (100) may convert it into a standardized data format for transmission to a subsequent data standardization layer. During the data conversion process, the electronic device (100) may perform consistency verification procedures, such as essential fields, data types, and referential integrity. The electronic device (100) may transmit the data converted into a standardized data format to a subsequent layer through an interface. The data collection and preprocessing flow of the input layer can have a direct impact on the accuracy and reliability of the entire matching system.

[0025] In at least one embodiment of the present disclosure, the electronic device (100) may include credential (VC) data, job posting data, and standard skill data (standard competency classification system). The electronic device (100) may collect each of the credential (VC) data, job posting data, and standard skill data independently. The electronic device (100) may apply separate collection policies and verification procedures for each data type. In the case of credential (VC) data, the electronic device (100) may collect data in a structure that includes detailed items such as the issuing organization, achievement standards, and verification grounds. In the case of job posting data, the electronic device (100) may collect data in a structure that includes items necessary for job matching, such as job information, required competencies, essential conditions, and preferential conditions. In the case of standard skill data, the electronic device (100) may collect data in a structure that includes standardized classification system information, such as competency identifiers, classification information, proficiency standards, and ground strength.

[0026] In at least one embodiment of the present disclosure, the electronic device (100) may collect data through an appropriate interface according to the characteristics of each data. In the case of credential (VC) data, the electronic device (100) may collect data through an external issuing agency API or an internal authentication system. In the case of job posting data, the electronic device (100) may collect data through various interfaces such as a recruitment platform API, a corporate internal recruitment system, or a file upload method. In the case of standard skill data, the electronic device (100) may collect data through a standard classification system repository, a database of an external standard organization, or a periodic data synchronization interface. The electronic device (100) may perform security and compatibility verification procedures, such as authentication, access rights, and data format conversion, for each interface.

[0027] In at least one embodiment of the present disclosure, the electronic device (100) can convert collected data into a standardized format for transmission to a subsequent layer. The electronic device (100) can map the field names of each collected data to a standardized data schema. The electronic device (100) can ensure consistency by applying constraints of the data schema, such as data types, required input values, and reference relationships. The electronic device (100) can correct errors by detecting missing fields or abnormal values ​​during the data conversion process. The electronic device (100) can transmit the data converted into a standardized data format to a data standardization layer in various ways, such as APIs, message queues, and file transfers.

[0028] In at least one embodiment of the present disclosure, the electronic device (100) may include credential (VC) data, job posting data, and standard skill data (standard competency classification system). The electronic device (100) may collect each of the credential (VC) data, job posting data, and standard skill data from separate data sources. The electronic device (100) may define detailed collection policies, such as collection cycles, data formats, and authentication methods, for each data type. In the case of credential (VC) data, the electronic device (100) may convert data structures that differ by issuing agency to conform to a standard schema. In the case of job posting data, the electronic device (100) may integrate data formats provided by various recruitment platforms to convert them into a consistent structure. In the case of standard skill data, the electronic device (100) may manage version control and update history of the standard classification system together.

[0029] In at least one embodiment of the present disclosure, the electronic device (100) can collect data through various interfaces, such as an external issuing agency, an internal system, a recruitment platform, and a standard classification system repository, depending on the characteristics of each data. The electronic device (100) can collect credential (VC) data in real time through an external issuing agency API. The electronic device (100) can automatically collect internally issued VC data through integration with an internal authentication system. The electronic device (100) can periodically collect job posting data through integration with a recruitment platform API or a data feed. The electronic device (100) can keep standard skill data up to date through a data synchronization interface with a standard classification system repository. The electronic device (100) can apply security policies, such as data access rights, authentication tokens, and data encryption, for each interface.

[0030] In one embodiment, the electronic device (100) can convert each collected data into a structured data format to transmit it to a subsequent data standardization layer. The electronic device (100) can convert the original format of each data into a standardized structured data format, such as JSON, XML, or CSV. When converting data, the electronic device (100) can apply a standardized data schema, such as field names, data types, and reference relationships. The electronic device (100) can transmit the data converted into a structured data format to the data standardization layer in various ways, such as API calls, file transfers, or message queues. The electronic device (100) can perform data integrity verification and error detection procedures during the data transmission process.

[0031] In at least one embodiment of the present disclosure, the electronic device (100) may apply a standardized data schema to maintain data consistency and consistency. The electronic device (100) may perform data consistency verification, such as required fields, data types, and referential integrity, according to the standardized data schema. The electronic device (100) may automatically correct data that violates constraints defined in the data schema or treat it as an error. The electronic device (100) may version control the standardized data schema to maintain compatibility when the data structure is changed. The electronic device (100) may utilize a schema mapping table for interaction with the data standardization layer.

[0032] In one embodiment, the electronic device (100) can collect credential (VC) data from an external issuing agency or an internal system. The electronic device (100) can collect VC data in real time through an API of an external issuing agency. The electronic device (100) can automatically collect internally issued VC data through integration with an internal authentication system. When collecting VC data, the electronic device (100) can perform authentication procedures and data format conversion for each issuing agency. The electronic device (100) can record metadata such as the time of collection, source, and authentication status of the VC data.

[0033] In one embodiment, the electronic device (100) can parse the issuing agency information and achievement information included in the VC data. The electronic device (100) can extract detailed information such as the agency name, agency identifier, and issuance date from the issuing agency field of the VC data. The electronic device (100) can parse detailed items such as achievement criteria, achievement date, and verification basis from the achievement information field of the VC data. The electronic device (100) can store the parsed issuing agency information and achievement information according to a standardized data schema. The electronic device (100) can manage data quality by detecting omissions, errors, and inconsistencies during the parsing process.

[0034] In at least one embodiment of the present disclosure, credential data may correspond to each of one or more candidates. Credential data may be uniquely mapped for each candidate. Credential data may be managed in conjunction with a candidate identification key. Credential data may be designed to enable parallel processing for multiple candidates.

[0035] In one embodiment, the credential (VC) data may include detailed information for each candidate, such as the issuing organization, achievement standards, and verification basis. The credential (VC) data may include detailed items for each candidate, such as issuing organization information, achievement standards, verification basis, issuance date, and certification status. The credential (VC) data may be stored by separating each detailed item into standardized fields. Multiple credential (VC) data may exist for each candidate, and each VC data may have a unique identifier.

[0036] In one embodiment, the electronic device (100) can separate and store VC data corresponding to each of a plurality of candidates. The electronic device (100) can separate and store data based on a candidate identification key within the VC data. The electronic device (100) can independently perform the processes of collecting, verifying, and storing VC data for each candidate. The electronic device (100) can manage the change history and status information of the VC data for each candidate together.

[0037] In one embodiment, the electronic device (100) can manage VC data and a candidate identification key in conjunction. The electronic device (100) can link with candidate records in a database based on the candidate identification field of the VC data. The electronic device (100) can ensure data integrity by mapping the candidate identification key to a unique identifier of the VC data. The electronic device (100) can support parallel processing for multiple candidates by utilizing the candidate identification key.

[0038] In one embodiment, the electronic device (100) can design VC data to enable parallel processing for multiple candidates. The electronic device (100) can design the structure of the VC data to be suitable for parallel processing, thereby enabling simultaneous data collection and verification for a large set of candidates. The electronic device (100) can apply a synchronization mechanism to prevent data conflicts and duplication for each candidate during parallel processing. The electronic device (100) can integrate the results of parallel processing and transmit them collectively to a subsequent data standardization layer.

[0039] In at least one embodiment of the present disclosure, an electronic device (100) may collect job posting data from a recruitment platform or a corporate system. The electronic device (100) may collect job posting data in real-time or periodically through a recruitment platform API. The electronic device (100) may automatically collect internal job posting data through integration with an internal corporate recruitment system. When collecting job posting data, the electronic device (100) may perform data format conversion and field mapping. The electronic device (100) may record metadata such as the time of collection, source, and job posting identifier of the job posting data.

[0040] In at least one embodiment of the present disclosure, an electronic device (100) can parse job information, required competencies, mandatory conditions, preferential conditions, etc. included in job posting data. The electronic device (100) can extract detailed information such as job title, job code, and department name from the job information field of the job posting data. The electronic device (100) can parse each item from the required competencies, mandatory conditions, and preferential conditions fields of the job posting data and convert them into structured data. The electronic device (100) can store each parsed item according to a standardized data schema. The electronic device (100) can manage data quality by detecting omissions, errors, and inconsistencies during the parsing process.

[0041] In at least one embodiment of the present disclosure, job posting data may include competencies required for the job, mandatory conditions, preferred conditions, etc. Job posting data may be stored in a structure that includes multiple items, such as required competencies, mandatory conditions, and preferred conditions, for each job. Job posting data may include additional information for each item, such as detailed competency names, proficiency standards, and whether they are preferred.

[0042] In one embodiment, an electronic device (100) can identify and extract required competencies, essential competencies, and preferred competency items from job posting data. The electronic device (100) can automatically identify required competencies, essential competencies, and preferred competency items from text or structured fields of job posting data. The electronic device (100) can map each identified competency item to a standardized competency identifier or classification system. The electronic device (100) can manage data quality by verifying whether there is duplication, omission, or inconsistency in the extracted competency items.

[0043] In one embodiment, the electronic device (100) can convert the extracted competency items into query information that can be utilized in a subsequent matching step. The electronic device (100) can convert the extracted competency items to fit a standardized query schema. The electronic device (100) can generate query information for each competency item that includes additional information such as a query key, value, weight, and whether it is required. The electronic device (100) can transmit the generated query information to a matching layer to be utilized for candidate-job matching.

[0044] In one embodiment, the electronic device (100) may configure metadata for job posting data to include information such as requirements, weights, and whether it is mandatory. The electronic device (100) may configure detailed information such as requirements, weights, whether it is mandatory, and whether it is preferred as metadata for each competency item in the job posting data. The electronic device (100) may store the metadata in accordance with a standardized data schema. The electronic device (100) may utilize the metadata to provide importance and filtering criteria for each competency in a subsequent matching stage.

[0045] In one embodiment, the electronic device (100) can collect standard skill data (standard competency classification system) from a standard classification system repository or an external standard organization. The electronic device (100) can periodically collect standard skill data through a data synchronization interface with the standard classification system repository. The electronic device (100) can query standard skill data in real time from the database of an external standard organization. The electronic device (100) can manage the version, update history, and source information of the standard skill data together. When collecting standard skill data, the electronic device (100) can perform data format conversion and field mapping.

[0046] In at least one embodiment of the present disclosure, the electronic device (100) may periodically update data to maintain the timeliness and consistency of the standard skill data. The electronic device (100) may perform data synchronization operations periodically by pre-defining the update cycle of the standard skill data. The electronic device (100) may maintain data consistency by recording the change history and version information of the standard skill data. When updating the standard skill data, the electronic device (100) may reflect changes by comparing the differences with existing data. By maintaining the latest state of the standard skill data, the electronic device (100) may utilize the latest competency classification system in the subsequent matching stage.

[0047] In one embodiment, the standard skill data may be a standard competency classification system including a competency identifier system and classification information. The standard skill data may include a unique identifier, classification code, parent / subordinate classification information, etc., for each competency. The standard skill data may include additional information such as definitions, descriptions, and related job functions for each competency. The standard skill data may be stored in a hierarchical structure according to a standardized classification system.

[0048] In at least one embodiment of the present disclosure, standard skill data may include proficiency levels and evidence strength criteria corresponding to each competency identifier. Standard skill data may define evaluation criteria, such as proficiency levels and evidence strength, for each competency. Standard skill data may include achievement standards, verification grounds, evaluation methods, etc., required for each proficiency level. Standard skill data may include detailed criteria, such as the type, reliability, and source of supporting materials, for each evidence strength.

[0049] In one embodiment, the electronic device (100) can utilize proficiency and basis level information included in standard skill data to provide consistent evaluation criteria in subsequent data standardization and matching stages. The electronic device (100) can store the proficiency and basis level information of standard skill data in accordance with a standardized data schema. The electronic device (100) can utilize the proficiency and basis level information to consistently apply competency evaluation criteria to credential (VC) data and job posting data. The electronic device (100) can transmit the evaluation criteria of standard skill data to subsequent data standardization and matching stages to increase the reliability and consistency of candidate-job matching.

[0050] Let's put the following content into a flowchart.

[0051] In at least one embodiment of the present disclosure, the data standardization layer (110) can convert each of the credential (VC) data, job posting data, and standard skill data (standard competency classification system) collected from the input layer into a standardized data format. The data standardization layer (110) can analyze the original structure and field composition of each data transmitted from the input layer. In the case of credential (VC) data, the data standardization layer (110) can extract key fields such as the issuing organization, achievement standards, and verification basis and convert them into standardized field names and data types. In the case of job posting data, the data standardization layer (110) can convert items necessary for matching, such as job information, required competencies, essential conditions, and preferential conditions, according to a standard schema. In the case of standard skill data, the data standardization layer (110) can convert standardized classification system information, such as competency identifiers, classification information, proficiency standards, and evidence strength, into a consistent format. The data standardization layer (110) can define conversion rules for each data type and detect and correct items of omission, error, or inconsistency during data conversion. The data standardization layer (110) can serialize the converted data into a standard data format such as structured JSON, XML, or Parquet. By converting the data into a standardized data format, the data standardization layer (110) can ensure consistency in data interpretation and processing in subsequent layers.

[0052] In at least one embodiment of the present disclosure, the data standardization layer (110) can ensure consistency by applying constraints of the data schema, such as field names, data types, and reference relationships of each data. The data standardization layer (110) can apply a mapping table that matches the field names of each data according to a predefined standard schema. In the case of data types, the data standardization layer (110) can perform conversion and verification procedures for each data type, such as strings, integers, dates, and booleans. In the case of reference relationships, the data standardization layer (110) can verify the referential integrity of competency identifiers between credential (VC) data and standard skill data. The data standardization layer (110) can ensure data consistency and integrity by applying constraints of the data schema (e.g., required input values, uniqueness, foreign key constraints, etc.). The data standardization layer (110) can perform automatic correction or exception handling for data where errors occurred during the schema verification process. By ensuring consistency, the data standardization layer (110) can prevent data interpretation errors in subsequent matching and output stages.

[0053] In at least one embodiment of the present disclosure, the data standardization layer (110) may apply an integrated standard for transmitting the converted data to the subsequent matching layer (120) and output layer (130). The data standardization layer (110) may predefine the input data format required by the matching layer (120) and output layer (130). The data standardization layer (110) may serialize the converted data according to the integrated standard (e.g., common field names, data types, reference structures, etc.). The data standardization layer (110) may transmit the data by clearly distinguishing the required fields, optional fields, and reference relationships according to the integrated standard. By applying the integrated standard, the data standardization layer (110) can ensure compatibility and scalability when transmitting data between layers. Due to the transmission of data conforming to the integrated standard, the data standardization layer (110) can achieve the effect of being able to process the data immediately without additional data conversion at the matching and output layers.

[0054] In one embodiment, the data flow between layers may be configured in the order of input, transformation, matching, and output. The electronic device (100) may collect credential (VC) data, job posting data, and standard skill data from an external or internal data source. The data standardization layer (110) may convert the data transmitted from the input layer into a standardized format. The matching layer (120) may receive the standardized data as input and perform candidate-job matching and suitability calculation. The output layer (130) may receive the matching results as input and perform fairness correction and generate explanatory information. Each layer may clearly distinguish and process the input, transformation, matching, and output stages according to the order of the data flow. This stepwise data flow configuration can improve the modularity and maintainability of the entire system.

[0055] In at least one embodiment of the present disclosure, each layer may include step-by-step flow management logic, such as verifying the consistency of input data, processing transformations, and providing a subsequent layer interface. Upon receiving input data, each layer may perform consistency verification procedures, such as checking for the existence of required fields, data type matching, and referential integrity. For data that has passed consistency verification, each layer may perform data transformation processing according to predefined transformation rules. Each layer may provide an interface (API, message queue, file transfer, etc.) for transmitting the transformed data to a subsequent layer. Each layer may include exception handling and error log recording functions when an error occurs at each stage of the data flow. Each layer may ensure the reliability and consistency of data transmission through the flow management logic.

[0056] In one embodiment, the data standardization layer (110) can provide the generated standard skill record (SSR) data as input to the matching layer (120). After generating the SSR data, the data standardization layer (110) can serialize the structure and fields of the SSR data to match the input format required by the matching layer (120). The data standardization layer (110) can transmit the SSR data to the matching layer (120) in various ways, such as API calls, message queues, or file transfers. When transmitting the SSR data, the data standardization layer (110) can verify data integrity and check whether the transmission was successful. Once the transmission of the SSR data is complete, the data standardization layer (110) can immediately utilize it for candidate-job matching in the subsequent matching stage.

[0057] In at least one embodiment of the present disclosure, the data standardization layer (110) can extract competency candidates from natural language-based text data included in the credential (VC) data. The data standardization layer (110) can separate and preprocess natural language text fields such as achievement descriptions, activity history, and verification grounds of the VC data. The data standardization layer (110) can input the preprocessed text into a large-scale language model or a natural language processing (NLP) engine to extract competency candidates. The data standardization layer (110) can generate a list of competency candidates by applying a few-shot extract method or predefined keyword-based rules. The data standardization layer (110) can evaluate the quality of the extracted competency candidates and remove unnecessary noise or duplicate items. The data standardization layer (110) can convert the competency candidate extraction results into structured data and utilize it in a subsequent normalization step.

[0058] In at least one embodiment of the present disclosure, the data standardization layer (110) can separate and refine unstructured text, such as achievement description text and evidence sentences, and utilize it for competency extraction. The data standardization layer (110) can divide achievement description text and evidence sentences within the VC data into sentence units. The data standardization layer (110) can identify sentences directly related to competency among the divided sentences. The data standardization layer (110) can perform refinement processing on the identified sentences, such as removing stop words, converting standardized terms, and correcting typos. The data standardization layer (110) can input the refined text into a competency extraction model to increase the accuracy of competency candidates. The data standardization layer (110) can obtain the effects of structuring and standardizing unstructured text through the refinement and extraction process.

[0059] In at least one embodiment of the present disclosure, the data standardization layer (110) can normalize competency candidates and map them to competency identifiers of a standard competency classification system. The data standardization layer (110) can apply normalization rules to the extracted competency candidates to integrate synonyms, similar expressions, typos, etc. The data standardization layer (110) can map the normalized competency candidates to competency identifiers defined in a standard competency classification system. The data standardization layer (110) can determine the final competency identifiers through alias-based search, similarity-based candidate search, rule verification, etc. The data standardization layer (110) can store the competency identifier mapping results as core fields of SSR data. The data standardization layer (110) can ensure consistency and comparability of competency representations through normalization and identifier mapping.

[0060] In one embodiment, the data standardization layer (110) can calculate the proficiency level and evidence strength corresponding to the competency identifier. The data standardization layer (110) can analyze relevant information for each competency identifier, such as achievement standards, activity periods, and verification grounds within the VC data. The data standardization layer (110) can apply predefined rules for calculating the proficiency level based on the analyzed information. The data standardization layer (110) can apply criteria for calculating evidence strength, such as the number and quality of supporting documents and the reliability of the issuing institution. The data standardization layer (110) can store the calculated proficiency level and evidence strength as fields in the SSR data. The data standardization layer (110) can increase the reliability and objectivity of the candidate's competency through the calculation of proficiency and evidence strength.

[0061] In one embodiment, the data standardization layer (110) can generate Standardized Skill Record (SSR) data regarding a candidate, including a competency identifier, proficiency level, and evidence strength, based on VC data and according to a pre-established standard competency classification system. The data standardization layer (110) can map each item of the VC data to fit the structure of the standard competency classification system. The data standardization layer (110) can generate SSR data including the mapped competency identifier, calculated proficiency level, evidence strength, issuing agency information, etc. The data standardization layer (110) can serialize and store the generated SSR data in a standardized data format. The data standardization layer (110) can provide a basis for competency comparison and subsequent matching for each candidate through the generation of SSR data.

[0062] In at least one embodiment of the present disclosure, the matching layer (120) can convert SSR data and job posting data into vectors in a shared semantic space. The matching layer (120) can input SSR data and job posting data into a pre-trained embedding model. The matching layer (120) can convert text or structured fields of each data into a vector form. The matching layer (120) can adjust embedding parameters so that SSR data and job posting data can be compared in the same embedding space. The matching layer (120) can store the embedded vectors by classifying them into candidate vectors, job vectors, etc. The matching layer (120) can establish a basis for calculating semantic similarity between candidate and job through the conversion of the shared semantic space.

[0063] In at least one embodiment of the present disclosure, the matching layer (120) can calculate similarity between transformed vectors. The matching layer (120) can calculate cosine similarity or inner product distance between the SSR data embedding vector and the job posting data embedding vector. The matching layer (120) can evaluate semantic relevance between candidate and job based on the calculated similarity values. The matching layer (120) can select the top N candidates based on the similarity calculation results. The matching layer (120) can efficiently derive a group of candidates from a large set of candidates through similarity calculation.

[0064] In one embodiment, the matching layer (120) may select some of one or more candidates as a candidate group based on similarity. The matching layer (120) may select candidates with high similarity scores as a candidate group based on a threshold or top N criteria. The matching layer (120) may quickly search for candidates in a large dataset by applying an approximate nearest neighbor (ANN) search algorithm. The matching layer (120) may pass the candidate selection results to a subsequent re-ranking step.

[0065] In at least one embodiment of the present disclosure, the matching layer (120) can derive a candidate group among one or more candidates that is semantically related to job information based on SSR data and job posting data. The matching layer (120) can select candidates with high semantic relevance by comparing the embedding vectors of the SSR data and the job posting data. The matching layer (120) can derive a candidate group by considering detailed items such as competency identifiers, proficiency levels, and evidence strength of the job information and SSR data together. The matching layer (120) can quickly select candidates whose actual competencies match the job through semantic-based candidate group derivation.

[0066] In one embodiment, the matching layer (120) can calculate the suitability of each candidate based on the SSR data of the candidate group and the job posting data. The matching layer (120) can compare the SSR data and the job posting data for each candidate group. The matching layer (120) can calculate a suitability score by synthesizing various factors such as the degree of competency match, proficiency level, strength of evidence, and whether essential competencies are met. The matching layer (120) can store the calculated suitability score for each candidate and utilize it in a subsequent ranking determination step.

[0067] In one embodiment, the matching layer (120) can calculate a first suitability value for each candidate by an interaction-based model using the candidate group's SSR data and job posting data together as inputs. The matching layer (120) can input the candidate group's SSR data and job posting data into an interaction-based cross-encoder model in pairs. The matching layer (120) can store the output value of the cross-encoder model as a first suitability value for each candidate. The matching layer (120) can transmit the first suitability calculation result to the evaluation of whether essential competencies are met and the final suitability calculation stage.

[0068] In one embodiment, the matching layer (120) can adjust the first suitability value based on the first suitability value to reflect whether the essential competencies included in the job posting data are satisfied. The matching layer (120) can evaluate whether each candidate's SSR data satisfies the essential competencies of the job posting data. If the essential competencies are not satisfied, the matching layer (120) can apply a penalty to the first suitability value or adjust the score. The matching layer (120) can transmit the result of reflecting whether the essential competencies are satisfied to the final suitability calculation stage.

[0069] In at least one embodiment of the present disclosure, the matching layer (120) can calculate the final fit based on the adjusted primary fit value, reflecting the proficiency level included in the SSR data. The matching layer (120) can calculate the final fit score by multiplying the adjusted primary fit value by a weighting factor for each proficiency level of the SSR data. The matching layer (120) can give greater contribution to the final fit score as the proficiency level for each competency increases. The matching layer (120) can utilize the result of the final fit calculation in the candidate ranking and output layer delivery steps.

[0070] In at least one embodiment of the present disclosure, the output layer (130) can determine candidate rankings by sorting candidates within a candidate group according to their suitability for each candidate. The output layer (130) can sort the final suitability scores for each candidate calculated by the matching layer (120) in descending order. The output layer (130) can generate candidate rankings based on the sorted results. The output layer (130) can transmit the generated candidate rankings to a subsequent fairness correction and explanation information generation step.

[0071] In one embodiment, the output layer (130) can classify candidates included in the candidate ranking into multiple groups according to predefined group criteria. The output layer (130) can classify candidates according to predefined group criteria such as gender, age, region, and experience. The output layer (130) can extract group attribute information for each candidate and group candidates belonging to the same group. The output layer (130) can utilize the group classification results in the fairness correction step.

[0072] In one embodiment, the output layer (130) can calculate a recommendation rate or exposure rate for each group. The output layer (130) can aggregate the number of candidates included in the recommendation list for each group. The output layer (130) can calculate the recommendation rate (RR) or exposure rate (ER) for each group as a ratio relative to the total number of candidates. The output layer (130) can use the calculated recommendation rate or exposure rate to calculate the deviation between groups.

[0073] In at least one embodiment of the present disclosure, the output layer (130) can generate a final ranking by adjusting the score of the candidate ranking so that the deviation of the recommendation rate or exposure rate between groups satisfies a preset fairness standard. The output layer (130) can generate a final ranking by adjusting the deviation of the recommendation rate or exposure rate by group ( ) can be calculated. The output layer (130) can calculate the fairness threshold ( Constraints such as ≤ ε) and a performance degradation tolerance limit (loss ≤ δ) can be set. The output layer (130) can adjust the score or order of the candidate rankings by applying a correction function (g). The output layer (130) can perform score recalibration or limited order adjustment according to the principle of least change. The output layer (130) can generate a corrected final ranking that satisfies fairness criteria.

[0074] In at least one embodiment of the present disclosure, the output layer (130) can generate a final ranking by adjusting the candidate rankings based on the candidate rankings to satisfy pre-set fairness criteria. The output layer (130) can calculate the final ranking by combining the candidate rankings with group-specific recommendation rates, exposure rates, and deviation information. The output layer (130) can transmit the result of calculating the final ranking to the recommendation basis information generation and result report output stage.

[0075] In at least one embodiment of the present disclosure, the output layer (130) can match the competency identifier included in the candidate group's SSR data with the required competency included in the job posting data. The output layer (130) can match the competency identifier of the SSR data with the required competency of the job posting data through a mapping table or a similarity-based matching algorithm. The output layer (130) can store the matching results in a structured form for each candidate. The output layer (130) can use the competency mapping results as input for generating recommendation basis information.

[0076] In one embodiment, the output layer (130) can identify at least one of a proficiency level or a strength of evidence for a corresponding competency identifier. The output layer (130) can extract the proficiency level and strength of evidence values ​​included in the SSR data for each mapped competency identifier. The output layer (130) can identify and record at least one of the proficiency level or strength of evidence for each competency. The output layer (130) can include the identified information in the candidate-specific recommendation evidence information.

[0077] In at least one embodiment of the present disclosure, the output layer (130) may generate recommendation basis information by correlating identified information with each candidate included in the final ranking. The output layer (130) may combine identified information, such as competency mapping results, proficiency levels, and evidence strength, for each candidate included in the final ranking. The output layer (130) may generate the combined information as a recommendation basis information package. The output layer (130) may include the recommendation basis information package in the final result report.

[0078] In at least one embodiment of the present disclosure, the output layer (130) may generate recommendation basis information for the final ranking by correlating the candidate group's SSR data with the job posting data. The output layer (130) may generate recommendation basis information by synthesizing various information such as competency mapping between the SSR data and the job posting data, proficiency level, strength of basis, and whether essential competencies are met. The output layer (130) may structure and store the generated recommendation basis information according to the final ranking. The output layer (130) may enhance the explainability of the recommendation result through the generation of recommendation basis information.

[0079] In at least one embodiment of the present disclosure, the output layer (130) may generate a candidate recommendation result report by combining the final ranking and recommendation basis information. The output layer (130) may combine the corrected final ranking and recommendation basis information for each candidate. The output layer (130) may output the combined results in various ways, such as an enterprise dashboard, an external system API, or a feedback interface. The output layer (130) may include key information in the candidate recommendation result report, such as the ranking, score, group deviation summary, and recommendation basis package. By generating the final result report, the output layer (130) may provide reasonable and reliable decision-making grounds to both the enterprise and the candidate.

[0081] FIG. 2 is a diagram illustrating the internal processing structure of a data standardization layer according to one embodiment.

[0082] In one embodiment, the data standardization layer (110) can perform the role of converting credential data into structured standard competency data. The data standardization layer (110) can receive credential (VC) data collected from the input layer. The data standardization layer (110) can identify key metadata included in the credential data, such as the issuing authority, achievement standards, and verification basis. The data standardization layer (110) can separate natural language-based achievement description text included in the credential data. The data standardization layer (110) can preprocess the separated text to remove unnecessary symbols, stop words, and noise. The data standardization layer (110) can input the preprocessed text into a large-scale language model to extract competency candidates. The data standardization layer (110) can perform normalization processing to integrate synonyms and similar expressions for each extracted competency candidate. The data standardization layer (110) can map the normalized competency candidates to competency identifiers defined in a standard competency classification system. The data standardization layer (110) can calculate the proficiency level and evidence strength corresponding to the mapped competency identifier. The data standardization layer (110) can generate a standard skill record (SSR) containing the competency identifier, proficiency level, evidence strength, and source information. The data standardization layer (110) can store the generated SSR data in a standard skill record repository. The data standardization layer (110) can provide the stored SSR data as input data to a subsequent matching layer (120). Due to the step-by-step processing described above, the data standardization layer (110) can achieve the effect of consistently converting the unstructured text of the credential data into structured standard competency data.

[0083] In one embodiment, the data standardization layer (110) may include a process of extracting competency candidates from natural language-based text data included in credential data. The data standardization layer (110) may divide the achievement description text into sentences. The data standardization layer (110) may search for competency-related keywords or patterns for each divided sentence. The data standardization layer (110) may generate a list of competency candidates using a few-shot extract method by utilizing a large-scale language model. The data standardization layer (110) may separately identify proof sentences and achievement standard sentences during the competency candidate extraction process. The data standardization layer (110) may evaluate the quality of the extracted competency candidates and filter out candidates with low reliability. The data standardization layer (110) may provide the competency candidate extraction results as input to a subsequent normalization step.

[0084] In one embodiment, the data standardization layer (110) can normalize the extracted competency candidates and map them to competency identifiers of the standard competency classification system. The data standardization layer (110) can perform rule-based normalization processing that integrates synonyms and similar expressions for each competency candidate. The data standardization layer (110) can improve the consistency of competency expressions by removing stop words and noise. The data standardization layer (110) can search for mapping candidates by comparing the normalized competency candidates with competency identifiers of the standard competency classification system. The data standardization layer (110) can determine the final competency identifier through alias-based matching or similarity-based search. The data standardization layer (110) can record the determined competency identifier as a field of the SSR data. The data standardization layer (110) can verify the latest version of the classification system and referential integrity during the competency identifier mapping process.

[0085] In at least one embodiment of the present disclosure, the data standardization layer (110) may include a process for calculating the proficiency level and evidence strength corresponding to the competency identifier. The data standardization layer (110) may analyze the expression of evidence within the achievement description for each competency identifier. The data standardization layer (110) may calculate the proficiency level by evaluating quantitative and qualitative indicators such as achievement standards, activity periods, and performance levels. The data standardization layer (110) may calculate the evidence strength by evaluating the reliability of the issuing institution and the number and quality of supporting materials. The data standardization layer (110) may record the results of calculating the proficiency level and evidence strength as fields in the SSR data. The data standardization layer (110) may perform correction or exception processing for items that do not meet the standards during the process of calculating the proficiency level and evidence strength.

[0086] In at least one embodiment of the present disclosure, the data standardization layer (110) may generate Standardized Skill Record (SSR) data for each candidate, including a competency identifier, a proficiency level, and a strength of evidence. The data standardization layer (110) may generate a list of extracted and normalized competency identifiers for each of one or more candidates. The data standardization layer (110) may complete the SSR data structure by combining the proficiency level and strength of evidence corresponding to each competency identifier. The data standardization layer (110) may additionally include source information, such as an issuing organization, achievement standards, and supporting materials, in the SSR data. The data standardization layer (110) may store the generated SSR data in a Standardized Skill Record repository. The data standardization layer (110) may provide the stored SSR data as input data to a subsequent matching layer (120) and an output layer (130). Due to the structuring of the SSR data, the data standardization layer (110) can ensure consistency in competency comparison and evaluation for each candidate.

[0087] In one embodiment, the data standardization layer (110) converts credential data existing in various formats into structured SSR data aligned with a standard competency classification system, thereby enabling comparison and evaluation of candidates based on consistent criteria in the subsequent matching stage. The data standardization layer (110) can unify differences in the description style, language, and expression of credential data according to standardization rules. The data standardization layer (110) can record competency identifiers, proficiency levels, and evidence strength in a consistent format based on the standard competency classification system. Due to the structured SSR data, the data standardization layer (110) enables the matching layer (120) to compare competencies of each candidate based on the same criteria. The data standardization layer (110) can minimize evaluation deviations and interpretation errors among candidates due to the provision of consistent SSR data. The data standardization layer (110) can improve matching accuracy and reliability through standardization processing.

[0088] In one embodiment, the data standardization layer (110) can provide the generated SSR data as input to the subsequent matching layer (120). The data standardization layer (110) can serialize the SSR data to match the input format required by the matching layer (120). The data standardization layer (110) can convert the field names, data types, and reference structures of the SSR data to match the specifications predefined with the matching layer (120). When transmitting SSR data, the data standardization layer (110) can prevent data interpretation errors by verifying essential fields and referential integrity. Due to the consistency and structuring of the SSR data, the data standardization layer (110) can achieve the effect of enabling immediate processing by the matching layer (120) without additional data conversion. By providing SSR data, the data standardization layer (110) can improve the candidate-job matching accuracy and processing efficiency of the matching layer (120).

[0089] In at least one embodiment of the present disclosure, at step 210, the data standardization layer (110) may collect VC data corresponding to each of one or more candidates. The data standardization layer (110) may receive VC data issued for each candidate from the input layer. The data standardization layer (110) may identify key metadata items included in the VC data, such as issuing institution information, achievement information, award recipient, achievement criteria, and verification basis. The data standardization layer (110) may check for omissions by comparing the metadata items of each VC data with a predefined list of required items. The data standardization layer (110) may convert the metadata of the collected VC data into a standardized data format. The data standardization layer (110) may provide the converted VC data as input to a subsequent consistency verification step. Through the process of collecting and verifying VC data, the data standardization layer (110) may obtain the effect of securing reliable competency data for each candidate.

[0090] In at least one embodiment of the present disclosure, the data standardization layer (110) can verify the consistency of issuing agency information, achievement information, and metadata included in the collected VC data. The data standardization layer (110) can verify whether the issuing agency information of the VC data matches a pre-registered list of trusted agencies. The data standardization layer (110) can verify whether the achievement information logically matches the achievement criteria and verification grounds within the VC data. The data standardization layer (110) can check for consistency and omission of values ​​through cross-referencing between metadata items. The data standardization layer (110) can separately identify VC data in which errors or inconsistencies are found as a result of the consistency verification. The data standardization layer (110) can ensure that untrusted VC data is not reflected in subsequent processing through consistency verification.

[0091] In at least one embodiment of the present disclosure, the data standardization layer (110) can identify invalid VC data during the metadata verification process. The data standardization layer (110) can determine that VC data is invalid if essential metadata items are missing or if the issuing authority information does not meet the trust criteria. The data standardization layer (110) can record invalid VC data in a separate error handling list. The data standardization layer (110) can record the error type and cause in detail for the invalid VC data. The data standardization layer (110) can block invalid VC data from being transmitted to the standard competency record creation stage.

[0092] In at least one embodiment of the present disclosure, the data standardization layer (110) can correct VC data containing errors. The data standardization layer (110) can automatically classify the error types of the VC data containing errors. The data standardization layer (110) can apply predefined correction rules according to the error types. For example, if the issuing institution name is a typographical error, the data standardization layer (110) can automatically correct it to the standard institution name. If a format error in the achievement information is found, the data standardization layer (110) can apply a regular expression-based automatic correction logic. The data standardization layer (110) can verify the correction effect by performing re-verification on the corrected VC data. The data standardization layer (110) can obtain the effect of improving data quality through the error correction process.

[0093] In at least one embodiment of the present disclosure, the data standardization layer (110) can automatically supplement missing metadata items. The data standardization layer (110) can automatically fill in missing required items within the VC data using predefined default values ​​or an external reference database. For example, if the issuance date is missing, the data standardization layer (110) can supplement it by looking up the relevant information in the official records of the issuing agency. After supplementing the missing items, the data standardization layer (110) can re-verify the consistency of the entire metadata. The data standardization layer (110) can record the source of supplementation and the method of supplementation for the automatically supplemented items.

[0094] In one embodiment, the data standardization layer (110) may provide corrected VC data to a subsequent processing step. The data standardization layer (110) may transmit the VC data, with errors and omissions corrected, as input data to a standard competency record generation step. The data standardization layer (110) may perform a series of standardization processes on the corrected VC data in a subsequent step, such as competency extraction, normalization, and identifier mapping. The data standardization layer (110) may obtain the effect of improving the reliability and consistency of the entire matching system due to the use of the corrected VC data.

[0095] In one embodiment, the data standardization layer (110) can verify the consistency of the issuing agency information and achievement information included in the VC data. The data standardization layer (110) can verify whether the issuing agency information and achievement information are logically connected within the same VC. The data standardization layer (110) can determine, for example, that there is a discrepancy if achievement items that cannot be certified by a specific issuing agency are included. The data standardization layer (110) can perform consistency verification by referring to the list of achievement information allowed by each issuing agency. The data standardization layer (110) can record the results of the consistency verification in the metadata status field.

[0096] In one embodiment, the data standardization layer (110) can perform consistency verification through cross-referencing between metadata items. The data standardization layer (110) can cross-reference multiple metadata items within the VC data, such as the issuing organization, achievement criteria, verification basis, and award recipient. The data standardization layer (110) can, for example, compare whether the achievement criteria match the official criteria of the issuing organization. The data standardization layer (110) can verify whether the verification basis is logically connected to the achievement information. If a discrepancy or omission is found as a result of cross-referencing, the data standardization layer (110) can set an error flag.

[0097] In one embodiment, the data standardization layer (110) can determine the validity of the metadata. The data standardization layer (110) can verify whether the metadata items of each VC data meet predefined validity criteria. The data standardization layer (110) can, for example, verify whether the issuing authority code is in a valid format and whether the achievement information is within an allowed value range. The data standardization layer (110) can record the result of the validity determination in the metadata status field.

[0098] In at least one embodiment of the present disclosure, the data standardization layer (110) may apply automatic correction logic based on the result of a validity judgment. The data standardization layer (110) may apply predefined automatic correction rules in stages to items where an error is found as a result of a validity judgment. For example, if an error in the issuing authority code is found, the data standardization layer (110) may automatically correct it to the correct code through a similar code recommendation algorithm. If an error in the value of achievement information is found, the data standardization layer (110) may automatically replace it by referring to a standard value table. After automatic correction, the data standardization layer (110) may re-verify the validity of the entire metadata.

[0099] In one embodiment, the data standardization layer (110) can input the corrected metadata into the standard competency record generation step. The data standardization layer (110) can transmit the metadata, which has been automatically corrected, to the standard competency record (SSR) generation module. The data standardization layer (110) can perform subsequent processing, such as competency extraction, normalization, and identifier mapping, based on the corrected metadata. The data standardization layer (110) can achieve the effect of improving the reliability and consistency of the standard competency record through the use of the corrected metadata.

[0100] In one embodiment, at step 220, the data standardization layer (110) can extract competency candidates from natural language-based text data included in the credential (VC) data and normalize the extracted competency candidates to map them to competency identifiers of a standard competency classification system. The data standardization layer (110) can receive natural language text included in the VC data as input. The data standardization layer (110) can separate information related to achievement descriptions and proofs from the input natural language text on a sentence-by-sentence basis. The data standardization layer (110) can apply a large-scale language model-based competency candidate extraction model to each separated sentence. The data standardization layer (110) can apply normalization rules that integrate synonyms and similar expressions to the extracted competency candidates. The data standardization layer (110) can map the normalized competency candidates by comparing them with competency identifiers defined in a standard competency classification system. The data standardization layer (110) can determine the final competency identifier through alias matching or similarity-based search logic. The data standardization layer (110) can provide the confirmed competency identifier as input to the subsequent proficiency and evidence strength calculation step. Due to the step-by-step processing described above, the data standardization layer (110) can achieve the effect of converting VC text of various expressions into consistent standard competency data.

[0101] In at least one embodiment of the present disclosure, the data standardization layer (110) can calculate the proficiency level and the strength of evidence corresponding to the competency identifier. The data standardization layer (110) can analyze the expression of evidence within the achievement description and the evidence sentence for the determined competency identifier. The data standardization layer (110) can extract quantitative and qualitative information, such as achievement criteria, activity period, evaluation grade, and award history, included in the evidence sentence. The data standardization layer (110) can apply a predefined rule for calculating the proficiency level to the extracted information. For example, the data standardization layer (110) can assign a high proficiency level if the activity period is above a certain standard or if the evaluation grade corresponds to a higher grade. The data standardization layer (110) can calculate the strength of evidence by combining multiple factors, such as the number of supporting documents, the reliability of the issuing institution, and the specificity of the evidence. The data standardization layer (110) can record the calculated proficiency level and the strength of evidence as structured data together with the competency identifier. The data standardization layer (110) can achieve the effect of providing reliable evaluation indicators for each capability through such step-by-step calculation processes.

[0102] In at least one embodiment of the present disclosure, the data standardization layer (110) may generate Standardized Skill Record (SSR) data for each candidate, including a competency identifier, a proficiency level, and a strength of evidence. The data standardization layer (110) may integrate the confirmed competency identifier, the calculated proficiency level, and the strength of evidence for each candidate into a single record. The data standardization layer (110) may additionally include metadata such as issuing agency information, source of supporting materials, and date of production in the Standardized Skill Record (SSR). The data standardization layer (110) may store the generated SSR data in a standardized database or SSR repository. The data standardization layer (110) may provide the stored SSR data as input data for a subsequent matching layer. Through the process of generating and storing SSRs, the data standardization layer (110) can achieve the effect of consistently managing competencies for each candidate in a structured form.

[0103] In at least one embodiment of the present disclosure, the data standardization layer (110) can extract natural language-based text data and related metadata included in the VC data. The data standardization layer (110) can separate and extract text fields and metadata fields within the VC data. The data standardization layer (110) can remove noise, such as unnecessary special characters, HTML tags, and emojis, from the text fields. The data standardization layer (110) can separately extract structured information, such as the issuance date, institution name, and authentication code, from the metadata fields. The data standardization layer (110) can provide the extracted text and metadata as input to subsequent sentence separation and feature extraction steps. Through the preprocessing of the text and metadata, the data standardization layer (110) can achieve the effect of ensuring data quality and processing consistency.

[0104] In at least one embodiment of the present disclosure, the data standardization layer (110) can separate the achievement description text and identify sentences related to evidence or criteria to refine them into a form usable in subsequent processing steps. The data standardization layer (110) can divide the achievement description text into sentence units. For each divided sentence, the data standardization layer (110) can search for keywords related to evidence or criteria (e.g., "award," "evaluation," "certification," "project completion," etc.). The data standardization layer (110) can distinguish between evidence sentences and general description sentences based on the keyword search results. The data standardization layer (110) can refine the evidence sentences by removing unnecessary conjunctions, redundant expressions, stop words, etc. included in them. The data standardization layer (110) can provide the refined evidence sentences as input to subsequent proficiency and evidence strength calculation steps. Through the process of separating and refining achievement description and evidence sentences, the data standardization layer (110) can obtain the effect of securing high-quality text data usable for competency evaluation.

[0105] In at least one embodiment of the present disclosure, the data standardization layer (110) can select features that can be utilized as competency candidates from the extracted text. The data standardization layer (110) can search for competency-related keywords or patterns for each divided sentence. The data standardization layer (110) can check whether there is a match with terms defined in a standard competency classification system, such as, for example, "development," "design," "operation," and "analysis." The data standardization layer (110) can temporarily store the matching sentences or phrases as competency candidates. The data standardization layer (110) can provide the sentences selected as competency candidates as input for subsequent normalization and mapping steps. Through the feature selection process, the data standardization layer (110) can prevent the mixing of unnecessary information and achieve the effect of increasing the accuracy of competency extraction.

[0106] In one embodiment, the data standardization layer (110) can select features based on criteria such as the semantic clarity of the competency candidate, the possibility of mapping to a standard classification system, and the reliability of supporting information. The data standardization layer (110) can determine whether the competency candidate has a concrete and clear meaning rather than an ambiguous expression. The data standardization layer (110) can verify whether the competency candidate can be directly mapped to an identifier defined in a standard competency classification system. The data standardization layer (110) can verify whether the supporting information corresponding to the competency candidate is based on a reliable issuing institution or official data. The data standardization layer (110) can select final features by combining multiple criteria such as semantic clarity, mapping possibility, and reliability of supporting information. Through the application of such multi-stage criteria, the data standardization layer (110) can achieve the effect of selecting only high-quality competency features suitable for standardization and matching.

[0107] In at least one embodiment of the present disclosure, the data standardization layer (110) can process natural language text included in VC data by separating it into detailed components. The data standardization layer (110) can classify text within the VC data into detailed categories such as achievement descriptions, evidence sentences, activity history, and evaluation results. The data standardization layer (110) can apply separate sentence separation and refinement rules for each category. The data standardization layer (110) can provide each separated component as input for subsequent competency extraction, proficiency calculation, and evidence strength evaluation stages. By processing the separation by detailed components, the data standardization layer (110) can achieve the effect of increasing the accuracy and efficiency of each processing stage.

[0108] In at least one embodiment of the present disclosure, the data standardization layer (110) can divide the achievement description text into sentences. The data standardization layer (110) can separate the achievement description text into individual sentences by applying a natural language processing-based sentence splitting algorithm. The data standardization layer (110) can search for competency-related keywords and patterns for each separated sentence. The data standardization layer (110) can provide the results of the sentence splitting and keyword search as input to the subsequent competency extraction and normalization steps. Through the sentence-unit splitting process, the data standardization layer (110) can achieve the effect of systematically extracting competency information even from complex descriptive text.

[0109] In one embodiment, the data standardization layer (110) can perform competency extraction and normalization processing for each divided sentence. The data standardization layer (110) can apply a competency extraction algorithm based on a large-scale language model to each divided sentence. The data standardization layer (110) can apply normalization rules, such as synonym integration, stop word removal, and expression unification, to the extracted competency candidates in stages. The data standardization layer (110) can map the normalized competency candidates by comparing them with identifiers of a standard competency classification system. The data standardization layer (110) can provide the mapping results as input to the SSR generation stage. Through competency extraction and normalization processing, the data standardization layer (110) can achieve the effect of converting the achievement descriptions of various expressions into standardized competency data.

[0110] In one embodiment, the data standardization layer (110) can identify sentences related to evidence or standards and refine them into a form that can be utilized in subsequent processing steps. The data standardization layer (110) can search for the presence of evidence-related keywords such as "certification," "award," "evaluation," and "project completion" within the sentences. The data standardization layer (110) can classify candidate evidence sentences separately based on the keyword search results. The data standardization layer (110) can refine the evidence sentences by removing unnecessary conjunctions, redundant expressions, stop words, etc. included in the evidence sentences. The data standardization layer (110) can provide the refined evidence sentences as input to the proficiency level and evidence strength calculation step. Through the process of identifying and refining evidence sentences, the data standardization layer (110) can achieve the effect of increasing the reliability of the competency evaluation.

[0111] In at least one embodiment of the present disclosure, at step 230, the data standardization layer (110) may perform the role of extracting skill candidates (or competency candidates) from natural language-based text data included in the credential (VC) data. The data standardization layer (110) may receive natural language text, such as achievement descriptions, activity history, and award history within the VC data, as input. The data standardization layer (110) may divide the input text into sentence units or semantic units. The data standardization layer (110) may apply a competency extraction model to each divided sentence to identify competency-related expressions. The data standardization layer (110) may extract keywords or phrases considered as competency candidates from among the identified expressions. The data standardization layer (110) may store the extracted competency candidates in the form of a temporary list. Due to the step-by-step processing described above, the data standardization layer (110) can achieve the effect of deriving competency candidates from various formats of VC text according to consistent criteria. In the present disclosure, skill candidates may also be referred to as competency candidates.

[0112] In at least one embodiment of the present disclosure, the data standardization layer (110) can normalize the extracted competency candidates and map them to competency identifiers of a standard competency classification system. The data standardization layer (110) can apply normalization rules that integrate synonyms and similar expressions to each of the extracted competency candidates. The data standardization layer (110) can remove noise such as stop words, unnecessary modifiers, and typos. The data standardization layer (110) can map the normalized competency candidates by comparing them with competency identifiers defined in a standard competency classification system. The data standardization layer (110) can determine the final competency identifier through logic such as alias matching, similarity-based search, and rule verification. The data standardization layer (110) can provide the determined competency identifier as input to the subsequent proficiency and evidence strength calculation step. Through the normalization and mapping process described above, the data standardization layer (110) can achieve the effect of consistently converting competency candidates of various expressions into standardized identifiers.

[0113] In at least one embodiment of the present disclosure, the data standardization layer (110) can convert VC text of various expressions into consistent standard competency data through a competency candidate extraction and normalization process. The data standardization layer (110) can apply mapping logic to match various expressions of the same meaning within the VC text to a standard competency identifier. The data standardization layer (110) can record the normalization and mapping results as structured standard competency data (SSR). The data standardization layer (110) may include fields such as competency identifier, proficiency level, evidence strength, and source information in the standard competency data. The data standardization layer (110) can provide the generated standard competency data as input data to a subsequent matching layer. Through such processing, the data standardization layer (110) can achieve the effect of removing expression deviations in the VC text and comparing and evaluating candidate competencies according to consistent criteria.

[0114] In one embodiment, the data standardization layer (110) may set up a model for extracting competency candidates. The data standardization layer (110) may select a natural language processing-based competency extraction model. The data standardization layer (110) may apply at least one of a large-scale language model, a rule-based model, or a hybrid model. When setting up the model, the data standardization layer (110) may consider the domain characteristics of the VC data, text length, expression diversity, etc. The data standardization layer (110) may pre-define the input format, output format, parameters, etc. of the model. The data standardization layer (110) may apply the results of the model setting to the competency candidate extraction step.

[0115] In one embodiment, the data standardization layer (110) may apply a few-shot extract-style processing method using a large-scale language model. The data standardization layer (110) may input a few examples (samples) into the large-scale language model to extract competency candidates with similar patterns. The data standardization layer (110) may configure an extract-style prompt to selectively extract only competency-related phrases within the text. The data standardization layer (110) may optimize model performance by reviewing the diversity, representativeness, and domain suitability of the few-shot examples. The data standardization layer (110) may convert the inference results of the large-scale language model into a list of competency candidates. Due to the application of the few-shot extract method as described above, the data standardization layer (110) can achieve the effect of effectively extracting competency candidates from various VC texts even with a small amount of training data.

[0116] In one embodiment, the data standardization layer (110) can manage the quality of the extracted competency candidates. The data standardization layer (110) can calculate quality evaluation indicators for the results of extracting competency candidates. The data standardization layer (110) can evaluate, for example, the accuracy, recall, duplication rate, and inconsistency rate of the candidates. The data standardization layer (110) can filter the candidate list or perform additional verification steps based on the quality evaluation results. The data standardization layer (110) can remove unnecessary candidates or error candidates through a quality management procedure. The data standardization layer (110) can log the quality management results and utilize them for subsequent analysis and improvement.

[0117] In at least one embodiment of the present disclosure, the data standardization layer (110) may apply quality control procedures, such as stop word removal, synonym integration, and noise reduction, during the normalization and mapping process. The data standardization layer (110) may remove stop words and meaningless words within competency candidates according to a predefined list. The data standardization layer (110) may apply a mapping table that integrates synonyms or similar expressions into standard expressions. The data standardization layer (110) may automatically detect and remove noise, such as typos, duplicates, and unnecessary modifiers. Through quality control procedures, the data standardization layer (110) may increase the consistency of normalization and mapping and ensure the reliability of standard competency data.

[0118] In one embodiment, the data standardization layer (110) can extract competency candidates by utilizing a large-scale language model. The data standardization layer (110) can perform competency candidate extraction considering contextual meaning by inputting VC text into the large-scale language model. The data standardization layer (110) can effectively identify competencies of various expressions by combining the pre-learned knowledge of the language model with domain-specific examples. The data standardization layer (110) can convert the inference results of the language model into a structured list of competency candidates. By applying a large-scale language model-based extraction method, the data standardization layer (110) can secure high flexibility and scalability compared to existing rule-based methods.

[0119] In at least one embodiment of the present disclosure, the data standardization layer (110) can extract competency candidates using the extracted text as input. The data standardization layer (110) can use text, such as achievement descriptions and activity history separated from VC data, as input values ​​for a large-scale language model. The data standardization layer (110) can extract competency candidates of similar patterns by including few-shot examples in the prompt along with the input text. The data standardization layer (110) can select only competency-related phrases from the output result of the language model and organize them into a list. The data standardization layer (110) can provide the extracted competency candidate list as input to a subsequent normalization and mapping step.

[0120] In one embodiment, the method can generate a list of skill candidates by performing Extract-style processing according to a few-shot method using a large-scale language model. The data standardization layer (110) can configure an Extract-style prompt to selectively extract only competency-related keywords or phrases within the text. The data standardization layer (110) can optimize model performance by reviewing the diversity, representativeness, and domain suitability of the few-shot examples. The data standardization layer (110) can convert the inference results of the large-scale language model into a list of skill candidates. Due to the application of the few-shot Extract method as described above, the data standardization layer (110) can achieve the effect of effectively extracting skill candidates from various VC texts even with a small amount of training data.

[0121] In at least one embodiment of the present disclosure, the data standardization layer (110) can generate a skill candidate list based on results derived using a few-shot extract method. The data standardization layer (110) can extract competency-related phrases from the output results of a large-scale language model and organize the skill candidate list into a structured form. The data standardization layer (110) can apply post-processing procedures, such as removing duplicates, sorting, and filtering, to each item in the skill candidate list. The data standardization layer (110) can provide the generated skill candidate list as input to subsequent normalization and identifier mapping steps. Through the process of generating the skill candidate list, the data standardization layer (110) can establish a foundation for converting various representations of VC data into standardized competency data.

[0122] In at least one embodiment of the present disclosure, the data standardization layer (110) may provide the generated skill candidate list as input to the subsequent normalization and identifier mapping step. The data standardization layer (110) may map each item in the skill candidate list by comparing it with a competency identifier of the standard competency classification system. The data standardization layer (110) may structure and store the mapping result as standard competency data (SSR). The data standardization layer (110) may provide the generated SSR data as input data to the subsequent matching layer. Due to the above step-by-step processing, the data standardization layer (110) can achieve the effect of consistently comparing and evaluating competencies for each candidate.

[0123] In at least one embodiment of the present disclosure, in steps 240 and 250, the data standardization layer (110) may perform the role of normalizing competency candidates extracted from credential (VC) data and mapping them to competency identifiers of a standard competency classification system. The data standardization layer (110) may apply normalization rules step by step to each competency candidate extracted from the VC data. The data standardization layer (110) may integrate synonyms and similar expressions into standard expressions. The data standardization layer (110) may remove noise such as stop words, unnecessary modifiers, and typos. The data standardization layer (110) may map the normalized competency candidates by comparing them with competency identifiers defined in a standard competency classification system. The data standardization layer (110) may determine the final competency identifier through logic such as alias matching, similarity-based search, and rule verification. The data standardization layer (110) may provide the determined competency identifier as input to the subsequent proficiency and evidence strength calculation steps. The data standardization layer (110) can achieve the effect of consistently converting various representations of competency candidates into standardized identifiers through the normalization and mapping process described above.

[0124] In one embodiment, the data standardization layer (110) can enable fair comparison and evaluation between candidates in the subsequent matching stage by standardizing competency candidates of various expressions according to consistent criteria. The data standardization layer (110) can apply mapping logic that matches various expressions of the same meaning to a single standard competency identifier. The data standardization layer (110) can record the normalization and mapping results as structured standard competency data (SSR). The data standardization layer (110) can include fields such as competency identifier, proficiency level, evidence strength, and source information in the standard competency data. The data standardization layer (110) can provide the generated standard competency data as input data for a subsequent matching layer. Through such processing, the data standardization layer (110) can achieve the effect of removing expression deviations in VC text and comparing and evaluating candidate competencies according to consistent criteria.

[0125] In at least one embodiment of the present disclosure, the data standardization layer (110) may apply a principle of integrating various expressions of the same meaning into a single standard expression in order to minimize expression deviations of competency candidates. The data standardization layer (110) may identify expressions of the same meaning by referring to a predefined synonym dictionary or a list of similar expressions for competency candidates. The data standardization layer (110) may convert the identified synonyms and similar expressions into a standard expression in bulk. The data standardization layer (110) may record the results of integration into a standard expression for each competency candidate. Through the application of such expression unification principles, the data standardization layer (110) can achieve the effect of comparing competencies described differently for each candidate based on the same standard.

[0126] In at least one embodiment of the present disclosure, the data standardization layer (110) may apply a strategy of linking a normalized competency candidate with a competency identifier defined in a standard competency classification system. The data standardization layer (110) may preload a list of identifiers of the standard competency classification system. The data standardization layer (110) may first check whether there is a match between the normalized competency candidate and the standard identifier. If there is no match, the data standardization layer (110) may apply alias matching or similarity-based search logic. Finally, the data standardization layer (110) may determine the mapping between the competency candidate and the standard identifier through rule verification. Through such a classification system linkage strategy, the data standardization layer (110) can achieve the effect of accurately linking competency candidates of various expressions to standardized identifiers.

[0127] In at least one embodiment of the present disclosure, at step 240, the data standardization layer (110) may apply normalization rules to each of the extracted competency candidates. The data standardization layer (110) may sequentially apply detailed normalization rules, such as synonym integration, stop word removal, and typo correction, for each competency candidate. The data standardization layer (110) may store the results of applying the normalization rules in a temporary list. The data standardization layer (110) may provide the normalized competency candidates as input to a subsequent mapping step. Through the application of such rules, the data standardization layer (110) can achieve the effect of improving the quality of the input data and increasing the matching accuracy.

[0128] In at least one embodiment of the present disclosure, the data standardization layer (110) may apply a normalization rule that integrates synonyms and similar expressions for each extracted competency candidate. The data standardization layer (110) may convert various expressions of competency candidates into standard expressions by referring to a synonym dictionary or a similar expression mapping table. The data standardization layer (110) may, for example, integrate expressions with similar meanings such as "programming," "coding," and "software development" into "programming." The data standardization layer (110) may record the integrated standard expressions for each competency candidate. Through such integration of synonyms and similar expressions, the data standardization layer (110) can achieve the effect of preventing comparison errors caused by expression deviations.

[0129] In one embodiment, the data standardization layer (110) can remove noise such as stop words, unnecessary modifiers, and typos. The data standardization layer (110) can delete unnecessary words within the competency candidate by referring to a list of stop words. The data standardization layer (110) can correct typos within the input text by applying a typo correction algorithm. The data standardization layer (110) can remove unnecessary modifiers or redundant expressions to leave only the core meaning of the competency candidate. Through such noise reduction processing, the data standardization layer (110) can achieve the effect of improving the quality of the normalized competency candidate and enhancing mapping accuracy.

[0130] In at least one embodiment of the present disclosure, at step 250, the data standardization layer (110) can map a normalized competency candidate by comparing it with a competency identifier defined in a standard competency classification system. The data standardization layer (110) can load a list of identifiers of the standard competency classification system into a database or memory. The data standardization layer (110) can first check whether there is a direct match between the normalized competency candidate and the standard identifier. If there is no direct match, the data standardization layer (110) can apply alias matching or similarity-based search logic. The data standardization layer (110) can record the mapping results in a temporary mapping table. Through the search and verification process described above, the data standardization layer (110) can achieve the effect of accurately connecting the competency candidate to the standard identifier.

[0131] In at least one embodiment of the present disclosure, the data standardization layer (110) can determine the final competency identifier through logic such as alias matching, similarity-based search, and rule verification. The data standardization layer (110) can identify alternative expressions of competency candidates by referring to a list of aliases registered in a standard competency classification system. If an alias exists, the data standardization layer (110) can map it to a standard identifier corresponding to that alias. If no alias exists, the data standardization layer (110) can search for the most similar standard identifier by applying similarity-based search logic. The data standardization layer (110) can verify the accuracy of the mapping through rule verification. Through the alias matching logic as described above, the data standardization layer (110) can achieve the effect of consistently connecting competency candidates of various expressions to a standard identifier.

[0132] In one embodiment, the method can normalize a competency candidate and map it to a competency identifier of a standard competency classification system. The data standardization layer (110) can calculate text similarity or embedding similarity between the normalized competency candidate and the standard identifier. The data standardization layer (110) can select a standard identifier whose similarity score is above a threshold as a candidate. The data standardization layer (110) can determine the identifier with the highest similarity among multiple candidates as the final mapping target. The data standardization layer (110) can record the determined identifier in the competency identifier field of the SSR data. Through such similarity-based search, the data standardization layer (110) can achieve the effect of effectively connecting expressions that do not directly match to the standard identifier.

[0133] In at least one embodiment of the present disclosure, at step 260, the data standardization layer (110) may perform the role of calculating the proficiency level and the strength of evidence for competency candidates extracted from the credential (VC) data. The data standardization layer (110) may separately calculate the proficiency level and the strength of evidence for each competency candidate extracted from the VC data. The data standardization layer (110) may analyze various input data for each competency candidate, such as descriptions of achievement, supporting materials, and issuing agency information. The data standardization layer (110) may calculate the proficiency level in the form of a numerical value or a grade based on the analyzed input data. The data standardization layer (110) may calculate the strength of evidence by reflecting multiple factors such as the reliability, quantity, and quality level of the supporting materials. The data standardization layer (110) may record the calculated proficiency level and strength of evidence in a standard skill record (SSR). Through the above calculation process, the data standardization layer (110) can achieve the effect of generating objective and consistent evaluation indicators for competencies for each candidate.

[0134] In one embodiment, the data standardization layer (110) can calculate the proficiency level and evidence strength corresponding to the determined competency identifier. The data standardization layer (110) can calculate the proficiency level and evidence strength for each competency identifier mapped to a standard competency classification system. The data standardization layer (110) can determine the proficiency level for each competency by analyzing the evidence materials and achievement descriptions entered for each competency identifier. The data standardization layer (110) can calculate the evidence strength by evaluating the issuing agency information and the number and quality of evidence materials for each competency identifier. The data standardization layer (110) can record the calculated proficiency level and evidence strength in the corresponding fields of the SSR data. Through such processing, the data standardization layer (110) can provide reliable evaluation indicators for each competency identifier.

[0135] In one embodiment, the data standardization layer (110) can generate a standard skill record (SSR) that includes a competency identifier, a proficiency level, evidence strength, and issuing organization or source information. The data standardization layer (110) can define an SSR data structure that includes multiple fields such as a competency identifier, a proficiency level, evidence strength, issuing organization information, and source information. The data standardization layer (110) can include the proficiency level and evidence strength calculated for each competency in the SSR data. The data standardization layer (110) can store the SSR data in a standard skill record repository. The data standardization layer (110) can provide the stored SSR data as input data to a subsequent matching layer. Through the SSR generation process described above, the data standardization layer (110) can enhance the consistency and usability of the structured competency data.

[0136] In one embodiment, the data standardization layer (110) may define proficiency level and evidence strength indicators included in the standard skill record (SSR). The data standardization layer (110) may define the proficiency level indicator by expressing the degree of achievement of competence as a numerical value or grade. The data standardization layer (110) may define the evidence strength indicator by reflecting multiple factors such as the reliability, quantity, and quality level of the evidence. The data standardization layer (110) may clearly distinguish and include the proficiency level and evidence strength fields within the SSR data structure. Through such indicator definitions, the data standardization layer (110) can clarify the evaluation criteria for the SSR data.

[0137] In one embodiment, the data standardization layer (110) may express the proficiency level as a numerical value or grade representing the degree of achievement of competence. The data standardization layer (110) may express the proficiency level on various scales, such as grades from 1 to 5 or numerical values ​​from 0 to 100. The data standardization layer (110) may determine the proficiency level by analyzing keywords, achievement criteria, and whether there is official certification within the achievement description. The data standardization layer (110) may record the result of calculating the proficiency level in the Level field of the SSR data. Through such numerical or grade expressions, the data standardization layer (110) may quantitatively compare the relative degree of achievement of competence.

[0138] In at least one embodiment of the present disclosure, the data standardization layer (110) can calculate the evidence strength as an indicator reflecting the reliability, number, and quality level of the supporting materials. To calculate the evidence strength, the data standardization layer (110) may comprehensively consider multiple evaluation factors, such as the reliability grade of the issuing institution, the number of supporting materials, formality, and detail. The data standardization layer (110) can calculate the evidence strength score by assigning weights to each evaluation factor. The data standardization layer (110) can record the calculated evidence strength score in the Evidence Strength field of the SSR data. Through the above method of calculating evidence strength, the data standardization layer (110) can quantitatively express the reliability and verification level of the competency.

[0139] In one embodiment, the data standardization layer (110) may apply logic to calculate the proficiency level and the strength of evidence by analyzing the achievement description, supporting sentences, and issuing agency information included in the VC data. The data standardization layer (110) may separate the text of the achievement description from the VC data and use it as input data. The data standardization layer (110) may extract supporting sentences and related metadata to obtain information necessary for calculating the proficiency level. The data standardization layer (110) may evaluate the reliability, quantity, and quality level of the issuing agency information and supporting materials and use them to calculate the strength of evidence. The data standardization layer (110) may calculate the proficiency level and the strength of evidence by performing the above input data analysis and evaluation process step by step. The data standardization layer (110) may generate structured competency evaluation information by recording the calculated results in the SSR data.

[0140] In at least one embodiment of the present disclosure, the data standardization layer (110) can record the calculated proficiency level and evidence strength in a standard skill record (SSR). The data standardization layer (110) can accurately input the results of the calculation of the proficiency level and evidence strength into each field of the SSR data. The data standardization layer (110) can utilize the proficiency level and evidence strength recorded in the SSR data in subsequent matching and evaluation stages. Through such a recording process, the data standardization layer (110) can increase the reliability and usability of the SSR data.

[0141] In one embodiment, the data standardization layer (110) can analyze the evidence expressions and achievement standards of the VC data to calculate the proficiency level. The data standardization layer (110) can extract evidence expressions from natural language-based achievement description text within the VC data. The data standardization layer (110) can use the extracted evidence expressions to calculate the proficiency level by comparing them with predefined achievement standards. The data standardization layer (110) can determine the proficiency level by evaluating the presence and detail of evidence expressions that meet the achievement standards. Through such analysis of evidence expressions and achievement standards, the data standardization layer (110) can enable the calculation of an objective and consistent proficiency level.

[0142] In one embodiment, the data standardization layer (110) can extract natural language-based text data and related metadata included in the VC data. The data standardization layer (110) can separate and extract various text and metadata, such as achievement descriptions, evidence sentences, and issuing agency information, from the VC data. The data standardization layer (110) can preprocess the extracted text data to remove unnecessary modifiers, stop words, typos, etc. The data standardization layer (110) can use the refined text data as input for the subsequent evidence expression analysis and achievement standard mapping steps. Through the text and metadata extraction process described above, the data standardization layer (110) can secure basic data for calculating proficiency levels and evidence strength.

[0143] In one embodiment, the data standardization layer (110) can separate the achievement description text and identify sentences related to evidence or standards, and refine them into a form that can be utilized in subsequent processing steps. The data standardization layer (110) can identify sentences directly related to evidence or achievement standards within the achievement description text. The data standardization layer (110) can separate the identified sentences from unnecessary information and refine them into a form suitable for calculating proficiency levels. The data standardization layer (110) can provide the refined evidence sentences as input for subsequent achievement standard mapping and evaluation steps. Through the sentence separation and refinement process described above, the data standardization layer (110) can increase the accuracy of proficiency level calculation.

[0144] In one embodiment, the data standardization layer (110) can identify expressions corresponding to achievement standards within the extracted evidence sentences. The data standardization layer (110) can search for predefined achievement standard keywords or patterns within the evidence sentences. The data standardization layer (110) can record the searched achievement standard expressions separately and use them as a basis for calculating the proficiency level. Through the achievement standard identification process described above, the data standardization layer (110) can enable objective evaluation of the evidence sentences.

[0145] In one embodiment, the data standardization layer (110) can calculate a proficiency level by mapping the identified achievement criteria to a predefined proficiency level system. The data standardization layer (110) can predefine a proficiency level corresponding to each achievement criterion. The data standardization layer (110) can calculate a proficiency level for each competency by mapping the identified achievement criteria to the corresponding level system. The data standardization layer (110) can record the calculated proficiency level in the Level field of the SSR data. Through the mapping process described above, the data standardization layer (110) can enable consistent calculation of proficiency levels without subjective interpretation.

[0146] In at least one embodiment of the present disclosure, the data standardization layer (110) can evaluate the reliability of the issuing institution and the number and quality of supporting materials to calculate the strength of evidence. The data standardization layer (110) can pre-define the reliability grade of the issuing institution and compare the issuing institution information of each VC data with the corresponding grade. The data standardization layer (110) can comprehensively analyze multiple evaluation factors, such as the number, formality, and detail of supporting materials. The data standardization layer (110) can calculate the strength of evidence score by assigning weights to each evaluation factor. The data standardization layer (110) can record the calculated strength of evidence score in the Evidence Strength field of the SSR data. Through the evaluation and calculation process described above, the data standardization layer (110) can quantitatively express the reliability and verification level of the capability.

[0147] In at least one embodiment of the present disclosure, the data standardization layer (110) can verify the consistency of the issuing agency information, achievement information, and metadata included in the VC data. The data standardization layer (110) can verify the consistency and validity of the issuing agency information, achievement information, and metadata within the VC data. The data standardization layer (110) can use the results of the consistency verification as input data for calculating the strength of the evidence. Through the consistency verification process described above, the data standardization layer (110) can enable a reliable evaluation of the strength of the evidence.

[0148] In at least one embodiment of the present disclosure, the data standardization layer (110) may refer to the reliability grade of the issuing institution and reflect it in the calculation of the evidence strength. The data standardization layer (110) may refer to a reliability grade table predefined for each issuing institution. The data standardization layer (110) may calculate a reliability score by comparing the issuing institution information of the VC data with the corresponding grade table. The data standardization layer (110) may reflect the calculated reliability score as a weight in the evaluation of the evidence strength. By reflecting the reliability grade as described above, the data standardization layer (110) can increase the objectivity of the calculation of the evidence strength.

[0149] In at least one embodiment of the present disclosure, the data standardization layer (110) can evaluate the number and quality level of supporting documents included in the VC data. The data standardization layer (110) can aggregate the number of supporting documents within the VC data. The data standardization layer (110) can evaluate the quality level, such as formality, detail, and timeliness, for each supporting document. The data standardization layer (110) can use the results of the number and quality level evaluation as input data for calculating the strength of the evidence. Through such an evaluation process, the data standardization layer (110) can increase the reliability of the calculation of the strength of the evidence.

[0150] In one embodiment, the data standardization layer (110) can calculate the strength of evidence by comprehensively considering various factors such as the number, detail, and formality of the evidence materials. The data standardization layer (110) can assign a basic score based on the number of evidence materials. The data standardization layer (110) can apply additional weights based on the detail and formality of the evidence materials. The data standardization layer (110) can calculate the strength of evidence score as a result of the weighted sum of multiple factors such as the number, detail, and formality. The data standardization layer (110) can record the calculated strength of evidence score in the SSR data. Through such a comprehensive evaluation method, the data standardization layer (110) can ensure objectivity and consistency in the calculation of the strength of evidence.

[0151] In at least one embodiment of the present disclosure, the data standardization layer (110) may receive credential (VC) data and standard skill data as inputs and generate a Standardized Skill Record (SSR) regarding a candidate. The data standardization layer (110) may simultaneously receive VC data and standard skill data collected from an input layer. The data standardization layer (110) may extract metadata, such as issuing organization information, descriptions of achievement, and supporting materials, included in the VC data. The data standardization layer (110) may convert the extracted text and metadata into competency candidates through natural language processing and analysis based on a large-scale language model. The data standardization layer (110) may normalize the competency candidates into competency identifiers defined in a standard competency classification system. The data standardization layer (110) may calculate the proficiency level and the strength of evidence for each competency identifier. The data standardization layer (110) may generate SSR data including the calculated competency identifier, proficiency level, strength of evidence, and issuing organization or source information. The data standardization layer (110) can consistently convert VC data of various formats into structured SSR data through the above step-by-step processing.

[0152] In at least one embodiment of the present disclosure, the data standardization layer (110) may store the generated SSR data in a standard skill record repository. After generating the SSR data, the data standardization layer (110) may store it in the standard skill record repository in accordance with the data structure. The data standardization layer (110) may perform indexing and duplicate verification of the SSR data within the repository. The data standardization layer (110) may periodically check the integrity and consistency of the stored SSR data. The data standardization layer (110) may apply an optimized storage method so that the SSR data within the repository can be efficiently searched by a subsequent matching layer.

[0153] In at least one embodiment of the present disclosure, the data standardization layer (110) may provide stored SSR data as input data to a subsequent matching layer. The data standardization layer (110) may retrieve data stored in the SSR storage and transmit it to the matching layer (120) in real-time or in batch mode. The data standardization layer (110) may ensure data consistency in subsequent processing steps by providing the SSR data to the matching layer while maintaining the field-specific structure of the SSR data. When providing SSR data, the data standardization layer (110) may support various search conditions, such as by candidate, by competency, or by period.

[0154] In one embodiment, the data standardization layer (110) can enhance the consistency and usability of structured competency data through the SSR generation process described above. The data standardization layer (110) can ensure the comparability of competency data for each candidate by standardizing the SSR generation and storage process. The data standardization layer (110) can maintain data quality so that structured SSR data can be utilized for various subsequent processes such as matching, evaluation, and explanation. Due to the standardization of SSR data, the data standardization layer (110) can provide objective evidence that companies and institutions can trust.

[0155] In at least one embodiment of the present disclosure, the data standardization layer (110) may generate a standard skill record (SSR) including a competency identifier, proficiency level, evidence strength, and issuing organization or source information. The data standardization layer (110) may implement a linkage with a standard competency classification system through a competency identifier field within the SSR data model. The data standardization layer (110) may record the degree of achievement of each competency as a numerical value or grade through a proficiency level field. The data standardization layer (110) may include evaluation results, such as the reliability, quantity, and quality level of supporting materials, through an evidence strength field. The data standardization layer (110) may clearly record the reliability evidence of each competency through an issuing organization or source information field. Through such field configuration, the data standardization layer (110) may ensure structural consistency and scalability of the SSR data.

[0156] In one embodiment, the data standardization layer (110) may include each field within the SSR data structure by clearly distinguishing them. The data standardization layer (110) may define the SSR data structure in various formats, such as JSON, XML, or RDB tables. The data standardization layer (110) may clearly specify the data type, length, and whether each field is required. The data standardization layer (110) may apply data validation logic to prevent duplication or omission between fields. Through the clear distinction of the SSR data structure, the data standardization layer (110) can increase the efficiency of subsequent system integration and data migration.

[0157] In at least one embodiment of the present disclosure, the data standardization layer (110) can manage candidate-specific competency information in a standardized format through an SSR data model. The data standardization layer (110) can store candidate-specific competency information in a consistent structure based on the SSR data model. The data standardization layer (110) can manage the standardized SSR data so that it can be utilized for competency comparison and statistical analysis among candidates. Due to the standardization of the SSR data model, the data standardization layer (110) can obtain the effect of facilitating data linkage and expansion with various external systems.

[0158] In at least one embodiment of the present disclosure, the data standardization layer (110) may store the generated SSR data in a standard skill record repository. When storing the SSR data, the data standardization layer (110) may perform field-specific indexing and duplicate checks. The data standardization layer (110) may periodically verify the integrity and consistency of the SSR data within the repository. The data standardization layer (110) may apply an optimized storage structure so that the SSR data within the repository can be quickly searched even in a large-scale environment.

[0159] In at least one embodiment of the present disclosure, the data standardization layer (110) may provide stored SSR data as input data to a subsequent matching layer. The data standardization layer (110) may transmit data stored in the SSR storage to the matching layer (120) in real-time or in batch mode. The data standardization layer (110) may ensure data consistency by providing the SSR data to the matching layer while maintaining the field-specific structure of the SSR data. When providing SSR data, the data standardization layer (110) may support various search conditions, such as by candidate, by competency, or by period.

[0160] In one embodiment, the data standardization layer (110) can apply a consistent data structure and an efficient approach during the storage and retrieval process of SSR data. When storing SSR data, the data standardization layer (110) can maximize indexing and search efficiency based on a standardized data structure. When searching for SSR data, the data standardization layer (110) can provide various search functions such as field-specific conditional search, complex query, and sorting. Through an efficient approach, the data standardization layer (110) can enable rapid SSR data retrieval even in a large-scale data environment.

[0161] In one embodiment, the data standardization layer (110) may define an SSR data structure that includes multiple fields such as a competency identifier, proficiency level, evidence strength, issuing organization information, and source information within the SSR data structure. The data standardization layer (110) may clearly define the name, data type, and input rules of each field within the SSR data structure. The data standardization layer (110) may record an identifier of the standard competency classification system in the competency identifier field. The data standardization layer (110) may record the degree of achievement in the form of a numerical value or grade in the proficiency level field. The data standardization layer (110) may record evaluation results such as the reliability, quantity, and quality level of supporting materials in the evidence strength field. The data standardization layer (110) may record the reliability evidence, such as the issuer and certification path of the VC data, in the issuing organization information and source information fields. Through the above field definitions, the data standardization layer (110) can ensure structural consistency and scalability of the SSR data.

[0162] In one embodiment, the data standardization layer (110) may include the proficiency level and evidence strength calculated for each competency in the SSR data. The data standardization layer (110) may record the proficiency level calculated for each competency in the corresponding field of the SSR data. The data standardization layer (110) may record the evidence strength calculated for each competency in the corresponding field of the SSR data. Through the recording of the proficiency level and evidence strength for each competency, the data standardization layer (110) may quantitatively express the reliability and degree of achievement of each candidate's competency.

[0163] In at least one embodiment of the present disclosure, the data standardization layer (110) may include competency identifiers defined in a standard competency classification system as fields in the SSR data. The data standardization layer (110) may record competency candidates in the SSR data by mapping them to identifiers of the standard competency classification system. The data standardization layer (110) may record information about each competency by analyzing the supporting materials and achievement descriptions entered for each competency identifier. In one embodiment, the data standardization layer (110) may record information about each competency by analyzing the supporting materials and achievement descriptions entered for each competency identifier. The data standardization layer (110) may analyze the achievement descriptions, supporting materials, and issuing agency information of the VC data for each competency identifier. Based on the analysis results, the data standardization layer (110) may record the reliability and degree of achievement for each competency identifier in the SSR data.

[0164] In one embodiment, the data standardization layer (110) can calculate a proficiency level corresponding to the determined competency identifier. The data standardization layer (110) can calculate the proficiency level by analyzing the achievement description, supporting materials, issuing agency information, etc. entered for each competency identifier. The data standardization layer (110) can record the calculated proficiency level in the corresponding field of the SSR data. In at least one embodiment of the present disclosure, the data standardization layer (110) can record the calculated proficiency level in the corresponding field of the SSR data. When calculating the proficiency level, the data standardization layer (110) can record it in the form of a grade (e.g., Beginner, Intermediate, Advanced) or a numerical value (e.g., levels 1 to 5). Through the recording of the proficiency level, the data standardization layer (110) can quantitatively compare the degree of competency achievement for each candidate.

[0165] In one embodiment, the data standardization layer (110) can calculate the strength of evidence corresponding to the determined capability identifier. The data standardization layer (110) can calculate the strength of evidence by analyzing the reliability, number, quality level, issuing agency information, etc. of the evidence materials entered for each capability identifier. The data standardization layer (110) can record the calculated strength of evidence in the corresponding field of the SSR data. In at least one embodiment of the present disclosure, the data standardization layer (110) can record the calculated strength of evidence in the corresponding field of the SSR data. When calculating the strength of evidence, the data standardization layer (110) can record it in the form of a grade (e.g., High, Medium, Low) or a numerical value (e.g., levels 1 to 3). Through the recording of the strength of evidence, the data standardization layer (110) can quantitatively provide a reliability evaluation for each capability.

[0166] In one embodiment, the data standardization layer (110) may include issuing organization information or source information as fields in the SSR data. The data standardization layer (110) may record source information, such as the issuing organization, certification path, and issuance date included in the VC data, in the SSR data. The data standardization layer (110) may provide grounds for evaluating the reliability of each capability through the source information fields. In one embodiment, the data standardization layer (110) may provide grounds for evaluating the reliability of each capability through the source information fields. The data standardization layer (110) may enable companies and organizations to objectively verify the reliability of the SSR data through the recording of source information.

[0167] In one embodiment, at step 270, the data standardization layer (110) can store the generated SSR data in a standard skill record repository. When storing the SSR data, the data standardization layer (110) can perform field-specific indexing and duplicate checks. The data standardization layer (110) can periodically verify the integrity and consistency of the SSR data within the repository. The data standardization layer (110) can apply an optimized storage structure so that the SSR data within the repository can be quickly retrieved even in a large-scale environment.

[0168] In at least one embodiment of the present disclosure, the data standardization layer (110) may provide stored SSR data as input data to a subsequent matching layer. The data standardization layer (110) may transmit data stored in the SSR storage to the matching layer (120) in real-time or in batch mode. The data standardization layer (110) may ensure data consistency by providing the SSR data to the matching layer while maintaining the field-specific structure of the SSR data. When providing SSR data, the data standardization layer (110) may support various search conditions, such as by candidate, by competency, or by period.

[0169] In one embodiment, the data standardization layer (110) can apply an efficient indexing and search structure within the storage of SSR data. When storing SSR data, the data standardization layer (110) can create an index for each major field, such as a competency identifier, candidate ID, and issuing organization. The data standardization layer (110) can enable fast searching and access to SSR data through the indexing structure. The data standardization layer (110) can apply various indexing methods, such as B-Tree and hash index, to prevent a decrease in search speed even in a large-scale data environment.

[0170] In at least one embodiment of the present disclosure, the data standardization layer (110) can enable fast searching and access through field-by-field indexing of SSR data. The data standardization layer (110) can support searching by various conditions, such as by candidate, by capability, and by period, through field-by-field indexing of SSR data. The data standardization layer (110) can utilize an indexing structure to quickly provide SSR data even for large data requests from subsequent matching layers.

[0171] In one embodiment, the data standardization layer (110) can provide the stored SSR data as input data to a subsequent matching layer. The data standardization layer (110) can transmit the data stored in the SSR storage to the matching layer (120) in real-time or in batch mode. The data standardization layer (110) can ensure data consistency by providing the data to the matching layer while maintaining the field-specific structure of the SSR data.

[0172] In at least one embodiment of the present disclosure, the data standardization layer (110) may provide a call interface for data linkage between the SSR storage and the matching layer. The data standardization layer (110) may link SSR data to the matching layer through various call interfaces, such as APIs, message queues, and data pipelines. The data standardization layer (110) may clearly define linkage policies, such as data formats, authentication, and access rights, in the call interface.

[0173] In one embodiment, the data standardization layer (110) can enable real-time provision and utilization of SSR data through a subsequent layer call interface. The data standardization layer (110) can immediately provide SSR data to the matching layer in response to a request through a real-time call interface. The data standardization layer (110) can periodically transmit a large amount of SSR data to the matching layer through a batch processing interface. By ensuring the scalability and stability of the call interface, the data standardization layer (110) can increase the overall system integration efficiency.

[0175] FIG. 3a is a drawing for explaining the internal processing structure of a matching layer according to one embodiment, and FIG. 3b is a drawing for explaining in detail the operation of a matching layer according to one embodiment.

[0176] Referring to FIGS. 3a and 3b, in at least one embodiment of the present disclosure, a matching layer (120) can perform candidate-job matching using standardized SSR data and job posting data. The matching layer (120) can simultaneously receive SSR data generated by the data standardization layer (110) and job posting data collected from the input layer. The matching layer (120) can utilize structured information, such as competency identifiers, proficiency levels, and evidence strength, included in the SSR data. The matching layer (120) can analyze required competencies, mandatory conditions, preferential conditions, and weighting information included in the job posting data. The matching layer (120) can embed the input SSR data and job posting data into vectors of the same semantic space. The matching layer (120) can evaluate the relationship between the candidate and the job by calculating the semantic similarity between the embedded vectors. The matching layer (120) can produce a candidate-job matching result based on the similarity evaluation result. The matching layer (120) can transmit the generated matching results to subsequent ranking, fairness correction, and explanation generation stages.

[0177] In at least one embodiment of the present disclosure, the matching layer (120) may operate in a two-stage structure of a first-stage candidate search (e.g., step 320) and a second-stage precise re-ranking (e.g., step 330). In the first stage, the matching layer (120) can rapidly select a group of candidates that are semantically related to the job from a large set of candidates. In the second stage, the matching layer (120) can calculate the suitability of the candidates derived in the first stage by precisely reflecting contextual relationships and job requirements. By combining high-speed search and precise evaluation through the two-stage structure, the matching layer (120) can simultaneously secure speed and accuracy. The matching layer (120) can manage input, processing, and output results separately for each stage.

[0178] In one embodiment, the first-stage candidate search can serve to rapidly select candidates semantically related to the job from a large set of candidates. The first-stage candidate search can transform SSR data and job posting data into the same embedding space. The first-stage candidate search can calculate the similarity between the embedded candidate vector and the job vector. Based on the calculated similarity, the first-stage candidate search can select the top N candidates as a candidate group. The first-stage candidate search can perform semantic-based filtering at high speed on a large dataset. The first-stage candidate search can improve processing speed by applying efficient search techniques such as ANN (Application Neighbor Network) search. The first-stage candidate search can reduce the computational burden of the subsequent detailed evaluation stage due to the rapid selection of candidates.

[0179] In one embodiment, the matching layer (120) can embed SSR data and job posting data into vectors in a shared semantic space. The matching layer (120) can apply fields such as competency identifiers, proficiency, and evidence strength of the SSR data to an embedding model as input values. The matching layer (120) can input required competencies, mandatory conditions, preferred conditions, and weight information of the job posting data into the embedding model. The matching layer (120) can utilize a pre-trained language model or a domain-specific embedding model. The matching layer (120) can generate candidate-specific vectors and job-specific vectors as embedding results. The embedded vectors can be normalized so that they can be compared in the same semantic space.

[0180] In at least one embodiment of the present disclosure, the matching layer (120) can calculate semantic similarity between a candidate vector and a job posting vector based on a dual-encoder structure. The dual-encoder structure can embed candidate data and job data independently, respectively. The dual-encoder structure can calculate cosine similarity or inner product distance between two embedding vectors. Since the dual-encoder structure enables parallel computation on a large set of candidates, the search speed can be significantly improved. The dual-encoder structure can apply various matching criteria by adjusting the number of dimensions of the embedding vectors, the normalization method, and the similarity calculation function.

[0181] In at least one embodiment of the present disclosure, the method can calculate similarity between vectors. The matching layer (120) can perform vector operations on each of the candidate vector and the job vector. The matching layer (120) can apply various similarity indicators such as cosine similarity, inner product, and Euclidean distance. The matching layer (120) can quantify the semantic relevance to the job for each candidate based on the calculated similarity value. The matching layer (120) can filter the candidate group based on a similarity threshold or the top N criteria.

[0182] In at least one embodiment of the present disclosure, the matching layer (120) can evaluate the semantic relevance between a candidate and a job by calculating similarity between embedded vectors. The matching layer (120) can consider a candidate with a high similarity value as a candidate semantically suitable for the job. The matching layer (120) can evaluate the diversity or concentration of the candidate group by analyzing the similarity distribution. The matching layer (120) can transmit the semantic relevance evaluation results to a subsequent ANN search and re-ranking step.

[0183] In at least one embodiment of the present disclosure, the matching layer (120) can quickly search for the top N candidates based on similarity using Approximate Nearest Neighbor (ANN) search. The matching layer (120) may apply an ANN search algorithm (e.g., HNSW, FAISS, ScaNN, etc.). The matching layer (120) may store embedding vectors in an ANN index and search for candidates based on their proximity to the query vector. ANN search can significantly improve search speed by utilizing an index structure without sequentially comparing the entire set of candidates. ANN search can quickly return the top N candidates based on similarity or candidates above a threshold. ANN search enables high-speed search in real-time or batch processing, even in large datasets.

[0184] In at least one embodiment of the present disclosure, the matching layer (120) can derive a candidate group at high speed from a large dataset through ANN search. The matching layer (120) can pass the ANN search results as input candidates for two-stage precise re-ranking. ANN search can adjust the trade-off between search speed and accuracy. ANN search can optimize system resource usage during the candidate generation process. ANN search can meet real-time matching requirements in a large-scale recruitment environment.

[0185] In at least one embodiment of the present disclosure, the two-stage precise re-ranking can serve to calculate a suitability score for the candidate group derived in the first stage by precisely reflecting contextual relationships and job requirements. The two-stage precise re-ranking can evaluate the interaction with job posting data for each candidate group selected in the first stage. Beyond simple embedding similarity comparison, the two-stage precise re-ranking can comprehensively consider the contextual relationship between the candidate and the job, whether essential competencies are met, and skill levels. The two-stage precise re-ranking can calculate a suitability score for each candidate by applying an interaction-based model (e.g., a cross-encoder). The two-stage precise re-ranking can readjust the ranking of candidates within the candidate group based on the calculated suitability score. The two-stage precise re-ranking can enhance the reliability and explainability of the final matching result through precise evaluation.

[0186] In at least one embodiment of the present disclosure, the matching layer (120) can perform a precise comparison between a candidate and a job posting by applying an interaction-based cross-encoder model. The matching layer (120) can input candidate SSR data and job posting data into the cross-encoder model in the form of pairs. The cross-encoder model can integrally encode the input candidate-job pairs to reflect contextual interaction information. The cross-encoder model can output a fit score or probability value for each pair. The cross-encoder model can perform domain-specific fit evaluation by fine-tuning a pre-trained language model. The cross-encoder model can perform parallel inference on the entire candidate pool in batch units.

[0187] In one embodiment, the matching layer (120) can calculate a primary suitability value for each candidate through a cross-encoder. The matching layer (120) can calculate a primary suitability score for each candidate by using candidate-specific SSR data and job posting data as input values. The primary suitability value can comprehensively reflect the contextual consistency between the candidate and the job, the degree of competency fulfillment, and the level of proficiency. The primary suitability value can be used as basic data for reflecting subsequent essential competencies and proficiency and for calculating the final ranking.

[0188] In one embodiment, the method can calculate the suitability of each candidate based on the SSR data of the candidate group and the job posting data. The matching layer (120) can compare the competency identifier, proficiency, and strength of evidence of the SSR data with the required competency and essential conditions of the job posting data for each candidate group. The matching layer (120) can calculate a suitability score for each candidate based on the comparison results. The matching layer (120) can determine the ranking of candidates within the candidate group based on the calculated suitability score.

[0189] In at least one embodiment of the present disclosure, the matching layer (120) may normalize the output value of the cross-encoder model and convert it into a fitness score. The matching layer (120) may normalize the raw output value of the cross-encoder to a range of 0 to 1 or a specified scale. The normalized fitness score allows for an intuitive comparison of relative fitness for each candidate. The normalization process prevents distortion of the score distribution and can provide a consistent standard in subsequent fairness correction and description generation steps.

[0190] In at least one embodiment of the present disclosure, the matching layer (120) can adjust the suitability by reflecting whether the required competencies included in the job posting data are met. The matching layer (120) can compare the competency identifiers included in the candidate-specific SSR data with the list of required competencies in the job posting data. If the required competencies are missing, the matching layer (120) can apply a penalty to the suitability score of the candidate. The matching layer (120) can adjust the weight for each candidate based on whether the required competencies are met. The process of reflecting required competencies enables the strict application of minimum standards for job requirements.

[0191] In at least one embodiment of the present disclosure, the matching layer (120) can calculate the final suitability by reflecting the proficiency level included in the SSR data. The matching layer (120) can verify the proficiency level (Level) for each competency as a numerical value or grade. The matching layer (120) can apply a weighting factor to the suitability score according to the proficiency level. The matching layer (120) can calculate the final suitability score for each candidate by summing the contributions for each competency. Reflecting the proficiency level can contribute to quantitatively evaluating the feasibility of actual job performance.

[0192] In one embodiment, the matching layer (120) may impose a penalty when essential competencies are not met, or adjust weights according to the level of proficiency. The matching layer (120) may apply a score reduction or a rank drop to candidates who do not meet essential competencies. The matching layer (120) may assign a higher weight to competencies with a high level of proficiency. The matching layer (120) may reflect the results of the essential competencies and proficiency in the final suitability calculation and candidate ranking determination. These step-by-step adjustments can increase the reliability of the matching results and the objectivity of the job suitability evaluation.

[0193] In at least one embodiment of the present disclosure, at step 310, the matching layer (120) may analyze job posting data to construct a query to be used for matching. The matching layer (120) may receive job posting data collected from the input layer as input values. The matching layer (120) may parse text and metadata included in the job posting data. The matching layer (120) may extract information necessary for matching, such as required competencies, mandatory conditions, preferential conditions, and weights, from the parsed data. The matching layer (120) may convert the extracted information into a structured query form. The matching layer (120) may provide the converted query information as input values ​​for subsequent embedding and matching steps. The query construction process may ensure a consistent input format corresponding to various representation methods of the job posting. The result of the query construction may directly affect matching accuracy and system consistency.

[0194] In one embodiment, the matching layer (120) can identify required competencies, weights, mandatory conditions, and preferential conditions included in a job posting and convert them into query information that can be utilized in a subsequent matching step. The matching layer (120) can automatically identify required competencies, weights, mandatory conditions, and preferential conditions by applying a text analysis model based on natural language processing. The matching layer (120) can map each identified item to a field of the query schema. The matching layer (120) can assign weight and condition values ​​to each item in the query information. The matching layer (120) can remove unnecessary noise during the structuring process of the query information and leave only standardized input values. The matching layer (120) can directly utilize the converted query information in the vectorization and embedding steps. The structuring and conversion process of the query information can increase the efficiency and accuracy of the subsequent matching step.

[0195] In at least one embodiment of the present disclosure, the matching layer (120) may include a flow for analyzing the requirements of the job posting data. The matching layer (120) may divide the text of the job posting data into sentences. The matching layer (120) may detect keywords and patterns related to the requirements for each divided sentence. The matching layer (120) may classify each item, such as required competencies, mandatory conditions, preferential conditions, and weights, based on the detected keywords and patterns. The matching layer (120) may refine the classified items into input information required for query configuration. The requirement analysis flow may be consistently performed through an automated text processing pipeline. The results of the requirement analysis may be utilized as input values ​​for the query schema design and vectorization steps.

[0196] In one embodiment, the matching layer (120) can derive input information necessary for query configuration through requirement analysis. The matching layer (120) can assign the input information derived as a result of the requirement analysis to each field of the query schema. The matching layer (120) can apply predefined rules or verification logic to verify the accuracy of the derived input information. The matching layer (120) can improve the structural completeness of the query based on the verified input information. Deriving input information through requirement analysis can strengthen the reliability and consistency of the matching system.

[0197] In one embodiment, the matching layer (120) can design the identified required competencies, weights, mandatory conditions, and preferential conditions into a structured query schema. The matching layer (120) can define each field of the query schema by classifying them into required competencies, weights, mandatory conditions, preferential conditions, etc. The matching layer (120) can clearly specify the data type and input format corresponding to each field. The matching layer (120) can consider compatibility with a standard competency classification system when designing the query schema. The matching layer (120) can include mandatory input values ​​and optional input values ​​separately in the query schema. The query schema design can ensure data consistency in subsequent embedding and matching stages. The query schema can be designed to be extensible to accommodate various job posting formats.

[0198] In one embodiment, the matching layer (120) can provide a consistent input format in a subsequent matching step through a query schema. The matching layer (120) can normalize the input query information according to the query schema. The matching layer (120) can convert the normalized query information into an input value for an embedding model. Providing a consistent input format through a query schema can increase the reliability and reproducibility of the matching results. A query schema-based input format can improve the standardization and maintenance efficiency of the entire system.

[0199] In at least one embodiment of the present disclosure, the matching layer (120) can analyze the required competencies included in the job posting data. The matching layer (120) can extract phrases related to the required competencies from the text of the job posting data. The matching layer (120) can match the extracted phrases by comparing them with a standard competency classification system. The matching layer (120) can normalize expression differences or synonyms of the required competencies and convert them into consistent identifiers. The results of the required competency analysis can be entered into the required competency field of the query schema. The required competency analysis process can increase the matching accuracy and the reliability of candidate evaluation.

[0200] In at least one embodiment of the present disclosure, the matching layer (120) can identify essential competencies in job posting data. The matching layer (120) can separately classify items corresponding to essential conditions among the required competencies. The matching layer (120) can apply a predefined keyword or pattern matching logic to identify essential competencies. The result of identifying essential competencies can be stored in the essential competency field within the query information. The identification of essential competencies can be directly utilized for filtering and suitability evaluation of candidate-job matching.

[0201] In one embodiment, the matching layer (120) may include the results of identifying essential competencies in the query information. The matching layer (120) may use the essential competency fields within the query information as input values ​​for the embedding and matching steps. The essential competency information may be reflected preferentially in the subsequent suitability calculation and filtering process. The process of identifying essential competencies and reflecting query information can increase the accuracy and reliability of the matching system.

[0202] In one embodiment, the matching layer (120) may assign weights and conditions to each required capability. The matching layer (120) may assign weight values ​​by quantifying importance or priority for each required capability. The matching layer (120) may set additional conditions, such as fulfillment conditions or preferential conditions, for each required capability. The assignment of weights and conditions may be explicitly recorded in each field of the query information. The weight and condition information may be used to adjust the contribution of each capability during the embedding and fit calculation stages.

[0203] In at least one embodiment of the present disclosure, the matching layer (120) may reflect weight and condition information in the query schema. The matching layer (120) may include weight and condition values ​​in each required competency field of the query schema. The weight and condition information reflected in the query schema enables a precise suitability evaluation in a subsequent matching step. The process of reflecting weights and conditions can contribute to deriving customized matching results for each job function.

[0204] In at least one embodiment of the present disclosure, the matching layer (120) can convert query information into a vector of a shared semantic space. The matching layer (120) can convert each field of the query information into an input value of an embedding model. The matching layer (120) can encode the query information into a high-dimensional vector through the embedding model. The query vector can be normalized so that it can be compared with the SSR data in the same semantic space. The query vectorization process can serve as the basis for calculating semantic similarity between candidate and job. The query vector can be input into a subsequent dual encoder-based candidate search and re-ranking step.

[0205] In one embodiment, the matching layer (120) can map the required competency within the query to a standard competency identifier. The matching layer (120) can apply mapping logic that matches various expressions of the required competency with identifiers defined in the standard competency classification system. Standard competency identifier mapping can derive a consistent identifier by normalizing synonyms, similar expressions, typos, etc. The result of the standard competency identifier mapping can be used as an input value for generating a query vector.

[0206] In at least one embodiment of the present disclosure, the matching layer (120) can utilize the standard identifier mapping results to generate a query vector. The matching layer (120) can convert the standard identifier mapping results into input values ​​for an embedding model. The standard identifier-based query vector can ensure consistency in semantic comparison with SSR data. The standard identifier mapping and vector generation process can increase the accuracy and reliability of the matching system.

[0207] In at least one embodiment of the present disclosure, the matching layer (120) may include context information in the query vector. The matching layer (120) may extract context information, such as the job title, industry group, experience requirements, and work type of the job posting, as a separate field. The matching layer (120) may encode the extracted context information as an additional dimension or feature value of the query vector. The query vector containing context information may reflect contextual relevance in comparison with the candidate SSR vector.

[0208] In at least one embodiment of the present disclosure, the matching layer (120) may input a query vector with context information reflected therein into a subsequent embedding and matching step. The matching layer (120) may input the query vector with context information reflected therein into a dual encoder and cross encoder-based matching model. The query vector with context information reflected therein can accurately reflect job-specific characteristics and requirements. The process of including and utilizing context information can increase the precision and explainability of the matching result.

[0209] In at least one embodiment of the present disclosure, at step 320, the matching layer (120) can perform candidate-job matching using standardized SSR data and job posting data. The matching layer (120) can receive SSR data generated from the data standardization layer (110) as input values. The matching layer (120) can receive job posting data collected from the input layer as input values. The matching layer (120) can feed the SSR data and job posting data into the same processing pipeline. The matching layer (120) can preprocess the text and metadata of the SSR data and job posting data, respectively. The matching layer (120) can convert the preprocessed data into input values ​​for an embedding model. The matching layer (120) can convert the SSR data and job posting data into vectors through the embedding model. The matching layer (120) can compare the converted vectors in a shared semantic space. The matching layer (120) can produce a candidate-job matching result based on the result of calculating similarity between the vectors. The matching layer (120) can transmit the matching results to subsequent ranking and output layers. The method of embedding and comparing SSR data and job posting data can reduce evaluation errors caused by representational deviations and linguistic diversity. Embedding-based matching can simultaneously achieve rapid search for a large set of candidates and precise suitability evaluation.

[0210] In at least one embodiment of the present disclosure, the matching layer (120) can convert SSR data and job posting data into vectors in a shared semantic space. The matching layer (120) can use each field of the SSR data (competency identifier, proficiency level, strength of evidence, etc.) as input values ​​for an embedding model. The matching layer (120) can also use the required competencies, mandatory conditions, preferential conditions, weights, etc. of the job posting data as input values ​​for an embedding model. The matching layer (120) can apply a pre-trained language model or a domain-specific embedding model as an embedding model. The matching layer (120) can feed input data into an embedding model after undergoing preprocessing steps such as tokenization, normalization, and padding. The embedding model can convert the input data into high-dimensional vectors. The converted vectors are located in the same semantic space, making it possible to compare semantic similarity. By mapping the SSR data and job posting data into the same embedding space, consistent comparison becomes possible despite various representation methods. The embedding vector transformation process can directly affect matching accuracy and system consistency.

[0211] In at least one embodiment of the present disclosure, at step 321, the matching layer (120) calculates similarity between vectors and may select some of one or more candidates as a candidate group based on said similarity. The matching layer (120) may calculate similarity between candidate vectors generated from SSR data and job vectors generated from job posting data. The matching layer (120) may apply numerical indicators such as cosine similarity, inner product distance, and Euclidean distance as a similarity calculation method. The matching layer (120) may calculate a suitability score for each candidate based on the calculated similarity value. The matching layer (120) may select candidates whose suitability score is above a threshold as a candidate group. The matching layer (120) may apply an approximate nearest neighbor (ANN) search algorithm to quickly search for the top N candidates from a large set of candidates. The similarity-based candidate group selection process enables the efficient selection of candidates who are semantically relevant to the job. The results of deriving the candidate pool can be used as input values ​​for the subsequent precise re-ranking and final recommendation stages.

[0212] In at least one embodiment of the present disclosure, the matching layer (120) can convert SSR data and job posting data into the same embedding space to enable semantic-based comparison. The matching layer (120) can select a model based on the structure of the embedding model and a pre-trained dataset. The matching layer (120) can prioritize embedding models specialized in skill and job domains. The matching layer (120) can analyze the representation characteristics of the SSR data and job posting data to select a model that provides an embedding space with high semantic consistency. When selecting a model, the matching layer (120) can evaluate the distribution of semantic similarity between vectors, expressiveness, computational efficiency, etc. Converting to the same embedding space overcomes representational differences between various data sources and enables consistent semantic comparison. The model selection criteria may affect the overall performance and scalability of the matching system.

[0213] In at least one embodiment of the present disclosure, the matching layer (120) can calculate semantic similarity between a candidate vector and a job posting vector based on a dual-encoder structure. The matching layer (120) can independently embed SSR data and job posting data by applying a dual-encoder structure. The dual-encoder structure can generate vectors using separate encoders for each input data. The matching layer (120) can compare the generated candidate vector and job vector in the same semantic space. The dual-encoder structure enables parallel embedding and high-speed similarity calculation for large datasets. Dual-encoder-based similarity calculation can maximize search efficiency by combining with ANN (Around Next Neighbor) search. The application of the dual-encoder structure can increase the scalability and real-time processing performance of the matching system.

[0214] In one embodiment, the matching layer (120) can construct a domain-specific semantic space through fine-tuning of the embedding model. The matching layer (120) can further train a pre-trained embedding model with a domain-specific dataset. During the fine-tuning process, the actual distribution and characteristics of SSR data and job posting data can be reflected. Fine-tuning can internalize differences in representation within the domain, term inconsistencies, and specific characteristics of the competency structure into the model. The fine-tuned embedding model can improve the accuracy of job-competency matching compared to a general embedding model. The construction of a domain-specific semantic space enables customized matching by job and industry.

[0215] In one embodiment, the matching layer (120) can reflect the characteristics of SSR data and job posting data during the fine-tuning process. The matching layer (120) may include various fields of SSR data, such as competency identifiers, proficiency levels, and evidence strength, in the fine-tuning dataset. The matching layer (120) may also include required competencies, mandatory conditions, preferential conditions, and weights of job posting data in the fine-tuning dataset. During the fine-tuning process, the importance of each data field may be assigned as a weight. The fine-tuned model can form a semantic space similar to the data distribution in an actual matching situation. The fine-tuning strategy can improve the domain fit and real-world performance of the matching system.

[0216] In one embodiment, the matching layer (120) can embed SSR data and job posting data by utilizing a pre-trained embedding model. The matching layer (120) can apply a publicly available pre-trained language model or vector embedding model to the embedding pipeline. The pre-trained model may include parameters pre-trained with large-scale plain text or domain-specific text. The matching layer (120) can embed SSR data and job posting data by using the parameters of the pre-trained model as initial values. The use of a pre-trained model can ensure stable embedding quality even in situations of data scarcity. The pre-trained model can further secure domain suitability through fine-tuning.

[0217] In at least one embodiment of the present disclosure, the matching layer (120) may apply an embedding model specialized for skill and job domains. The matching layer (120) may select an embedding model that has been pre-trained with a dataset related to skills and jobs. A domain-specialized embedding model may have higher precision in job-competency representation compared to a general language model. By applying a domain-specialized model, the matching layer (120) can precisely reflect the semantic difference between job-specific required competencies and candidate competencies. Skill / job domain-specialized embeddings can increase the accuracy of customized matching by industry and job.

[0218] In one embodiment, the matching layer (120) can improve domain fit by freezing or fine-tuning some parameters of the embedding model. The matching layer (120) can freeze the lower layer parameters of the pre-trained model and fine-tune only the upper layer. Parameter freezing allows for additional learning of only domain-specific information while maintaining the generalization performance of the existing model. Fine-tuning allows for the incorporation of representations specialized for domain data into the model. The parameter freezing and fine-tuning strategies can prevent overfitting and improve learning efficiency. A combination of freezing and fine-tuning can achieve a balance between embedding quality and domain fit.

[0219] In one embodiment, the matching layer (120) can embed SSR data and job posting data to compare them in a shared semantic space. The matching layer (120) can convert each field of the SSR data and each item of the job posting data into input values ​​for an embedding model. The embedding model can convert the input values ​​into high-dimensional vectors so that the SSR data and job posting data are located in the same semantic space. The embedded vectors can be compared through distance or similarity calculations between the vectors. The SSR and job posting data embedding process overcomes various representation methods and linguistic differences, enabling semantic-based matching. The embedding results can be used as input values ​​for subsequent similarity calculations, candidate group derivation, and suitability evaluation stages.

[0220] In at least one embodiment of the present disclosure, the matching layer (120) can generate candidate vectors from SSR data. The matching layer (120) can use each field of the SSR data, such as competency identifiers, proficiency levels, and evidence strength, as input values ​​for an embedding model. The embedding model can convert the input SSR data into high-dimensional candidate vectors. The candidate vectors can reflect competency structures and proficiency information inherent in the SSR data. The candidate vectors can be directly compared with job vectors in a shared semantic space. The candidate vector generation process can reflect quantitative and qualitative characteristics of competencies for each candidate in the embedding. The candidate vectors can be used as input values ​​for subsequent similarity calculation and candidate group derivation steps.

[0221] In one embodiment, the matching layer (120) can generate a job vector from job posting data. The matching layer (120) can use each item of the job posting data, such as required competencies, mandatory conditions, preferred conditions, and weights, as an input value for an embedding model. The embedding model can convert the input job posting data into a high-dimensional job vector. The job vector can reflect job requirements and condition information inherent in the job posting. The job vector can be directly compared with a candidate vector in a shared semantic space. The job vector generation process can reflect the structural characteristics of the required competencies for each job in the embedding. The job vector can be used as an input value for subsequent similarity calculation and candidate group derivation steps.

[0222] In at least one embodiment of the present disclosure, at step 322, the matching layer (120) can perform candidate-job matching using standardized SSR data and job posting data. The matching layer (120) can receive SSR data generated by the data standardization layer (110) as input. The matching layer (120) can receive job posting data collected from the input layer as input. The matching layer (120) can feed the SSR data and job posting data into the same processing pipeline. The matching layer (120) can preprocess the text and metadata of the SSR data and job posting data, respectively. The matching layer (120) can convert the preprocessed data into input values ​​for an embedding model. The matching layer (120) can convert the SSR data and job posting data into vectors through the embedding model. The matching layer (120) can compare the converted vectors in a shared semantic space. The matching layer (120) can produce a candidate-job matching result based on the result of calculating similarity between the vectors. The matching layer (120) can transmit the matching results to subsequent ranking and output layers. The method of embedding and comparing SSR data and job posting data can reduce evaluation errors caused by representational deviations and linguistic diversity. Embedding-based matching can simultaneously achieve rapid search for a large set of candidates and precise suitability evaluation.

[0223] In at least one embodiment of the present disclosure, the matching layer (120) can convert SSR data and job posting data into vectors in a shared semantic space. The matching layer (120) can use each field of the SSR data (competency identifier, proficiency level, strength of evidence, etc.) as input values ​​for an embedding model. The matching layer (120) can also use the required competencies, mandatory conditions, preferential conditions, weights, etc. of the job posting data as input values ​​for an embedding model. The matching layer (120) can apply a pre-trained language model or a domain-specific embedding model as an embedding model. The matching layer (120) can feed input data into an embedding model after undergoing preprocessing steps such as tokenization, normalization, and padding. The embedding model can convert the input data into high-dimensional vectors. The converted vectors are located in the same semantic space, making it possible to compare semantic similarity. By mapping the SSR data and job posting data into the same embedding space, consistent comparison becomes possible despite various representation methods. The embedding vector transformation process can directly affect matching accuracy and system consistency.

[0224] In one embodiment, the matching layer (120) calculates similarity between vectors and can select some of one or more candidates as a candidate group based on said similarity. The matching layer (120) can calculate similarity between candidate vectors generated from SSR data and job vectors generated from job posting data. The matching layer (120) can apply numerical indicators such as cosine similarity, inner product distance, and Euclidean distance as a similarity calculation method. The matching layer (120) can calculate a suitability score for each candidate based on the calculated similarity value. The matching layer (120) can select candidates whose suitability score is above a threshold as a candidate group. The matching layer (120) can quickly search for the top N candidates from a large set of candidates by applying an Approximately Nearest Neighbor (ANN) search algorithm. The similarity-based candidate group selection process enables the efficient selection of candidates who are semantically relevant to the job. The results of the candidate group derivation can be used as input values ​​for subsequent precise re-ranking and final recommendation stages.

[0225] In at least one embodiment of the present disclosure, the matching layer (120) can calculate the similarity between a candidate vector generated from SSR data and a job vector generated from job posting data. The matching layer (120) can match the embedding vector for each candidate in the SSR data with the embedding vector of the job posting data in a 1:1 or 1:N manner. The matching layer (120) can perform an inner product operation or a cosine similarity operation between the embedding vectors. The matching layer (120) can convert the similarity calculation result into a suitability score for each candidate. The similarity calculation pipeline may include sequential steps of input data preprocessing, embedding, vector operation, and score calculation. Each step may be optimized independently within the pipeline. The structure of the similarity calculation pipeline can ensure parallel processing and scalability for large datasets.

[0226] In one embodiment, the matching layer (120) can calculate a suitability score for each candidate based on the result of a similarity calculation. The matching layer (120) can normalize the calculated similarity value into a score in the range of 0 to 1 or -1 to 1. The matching layer (120) can store the normalized suitability score for each candidate. The matching layer (120) can generate a ranking for each candidate based on the suitability score. The suitability score calculation process can reflect the numerical similarity of the embedding vector and the degree of agreement with the job requirements. The suitability score can be used as an input value for subsequent filtering, re-ranking, and fairness correction steps.

[0227] In at least one embodiment of the present disclosure, the matching layer (120) may derive a candidate group based on a suitability score. The matching layer (120) may classify candidates whose suitability score is greater than or equal to a preset threshold as a candidate group. The matching layer (120) may select the top N candidates as a candidate group. The criteria for deriving the candidate group may be set by combining various rules, such as a threshold, the top N, and the fulfillment of essential competencies. The results of deriving the candidate group may be classified and stored by job, group, and condition.

[0228] In at least one embodiment of the present disclosure, the matching layer (120) can rapidly search for the top N candidates in a large set of candidates by applying an approximate nearest neighbor (ANN) search algorithm. The matching layer (120) can apply various implementation methods such as HNSW, FAISS, and LSH as the ANN search algorithm. The matching layer (120) can enable high-speed search by storing embedding vectors in an indexing structure. ANN search can significantly reduce the amount of computation compared to a complete search of the entire set of candidates. The ANN search algorithm can efficiently search for candidate vectors closest to the job vector within a vector space. ANN-based search can contribute to real-time matching and the provision of scalable services in large datasets.

[0229] In one embodiment, the matching layer (120) can optimize search efficiency through ANN search. The matching layer (120) can balance search speed and accuracy by adjusting parameters of the ANN search (search depth, index size, etc.). The matching layer (120) can pass the ANN search results to a subsequent suitability calculation and re-ranking stage. Optimizing search efficiency can directly contribute to reducing the overall processing time of the system and processing large-scale concurrent requests.

[0230] In at least one embodiment of the present disclosure, the matching layer (120) may apply a numerical similarity index between embedding vectors. The matching layer (120) may selectively apply various numerical indices, such as cosine similarity, inner product distance, and Euclidean distance. The matching layer (120) may determine an optimal matching method by comparing similarity values ​​calculated for each index. The numerical similarity index may reflect various characteristics of the embedding vectors, such as directionality, magnitude, and distance. The selection of the similarity index may be adjusted according to job characteristics, data distribution, and system requirements.

[0231] In at least one embodiment of the present disclosure, the matching layer (120) may calculate a suitability score for each candidate based on the calculated similarity value. The matching layer (120) may apply post-processing steps, such as normalization, weighting, and threshold filtering, to the similarity value. The matching layer (120) may store the suitability score for each candidate and select the top N candidates or candidates above the threshold. The suitability score calculation process may be performed independently by job, group, and condition.

[0232] In one embodiment, the matching layer (120) may apply numerical indicators such as cosine similarity, inner product distance, and Euclidean distance as a similarity calculation method. The matching layer (120) may evaluate the directional similarity of two vectors by applying cosine similarity. The matching layer (120) may simultaneously reflect the magnitude and direction of two vectors through the calculation of inner product distance. The matching layer (120) may measure the absolute distance between vectors by applying Euclidean distance. Each similarity indicator may be selectively applied depending on data characteristics and the purpose of matching. The selection of similarity indicators may affect matching accuracy, processing speed, and system consistency.

[0233] In one embodiment, the matching layer (120) can select a candidate group by rapidly searching for the top N candidates based on the similarity using an approximate nearest neighbor (ANN) search. The matching layer (120) can respond to real-time search requests by pre-constructing an ANN search index. The matching layer (120) can search for nearest neighbors in the candidate vector index using a job vector as a query. ANN search can significantly reduce the amount of computation compared to a complete search of the entire set of candidates. The results of the ANN search can be returned as a list of the top N candidates. The application of ANN search can contribute to real-time matching and the provision of scalable services in large datasets.

[0234] In at least one embodiment of the present disclosure, the matching layer (120) may select the top N candidates as a candidate group based on calculated similarity values. The matching layer (120) may sort the candidate list in descending order based on similarity scores. The matching layer (120) may designate the top N candidates as a candidate group. The criteria for selecting the top N candidates may be adjusted according to system settings, job requirements, and service policies. The list of top N candidates may be used as input values ​​for subsequent re-ranking and fairness correction steps.

[0235] In at least one embodiment of the present disclosure, the matching layer (120) may select candidates whose suitability score is above a threshold as a candidate group. The matching layer (120) may filter candidates based on a predefined threshold value. The matching layer (120) may exclude candidates below the threshold from the candidate group. The threshold setting may be adjusted independently by job, group, or service. Threshold-based filtering can improve the quality and consistency of the candidate group.

[0236] In one embodiment, the matching layer (120) can refine the candidate group by applying thresholds and filter rules. The matching layer (120) can apply additional filter rules, such as whether essential competencies are met, proficiency level, and strength of evidence. The matching layer (120) can refine the candidate group in multiple stages by combining multiple filter rules. The result of applying filter rules can improve the job suitability and reliability of the candidate group.

[0237] In one embodiment, the matching layer (120) can perform the derivation of candidate groups for multiple jobs in parallel. The matching layer (120) can process independent embeddings and similarity calculations for multiple jobs in parallel. The matching layer (120) can generate a separate list of candidate groups for each job. The parallel processing structure can improve the system's throughput and response speed. The results of the multiple job processing enable the provision of simultaneous matching and recommendation services for various jobs.

[0238] In one embodiment, the matching layer (120) can generate independent candidate groups for each job. The matching layer (120) can independently apply job-specific requirements, thresholds, and filter rules. The matching layer (120) can separately store and manage job-specific candidate group lists. Generating independent candidate groups for each job can increase the precision of customized recommendations and job-specific matching.

[0239] In one embodiment, at step 330, the matching layer (120) may perform a two-stage re-ranking on the derived candidate group. The matching layer (120) may receive as input a set of the top N candidates selected based on embedding similarity in the first stage. The matching layer (120) may apply a cross-encoder-based precision re-ranking model to reflect interaction information between each candidate and the job posting. In the two-stage re-ranking process, the matching layer (120) may combine candidate-specific SSR data and job posting data and use them as input. Through the two-stage re-ranking, the matching layer (120) can precisely evaluate contextual differences that are not distinguished by simple embedding similarity alone and whether essential competencies are met. The two-stage re-ranking structure may comprehensively reflect various factors such as detailed competency matching between candidate and job, proficiency level, and strength of evidence. By applying a two-stage re-ranking, the matching layer (120) can improve the accuracy and reliability of the suitability evaluation for each candidate.

[0240] In one embodiment, at step 331, the matching layer (120) can perform a precise comparison between a candidate and a job posting by applying an interaction-based cross-encoder model. The matching layer (120) can combine candidate-specific SSR data and job posting data in a pair form as input values ​​to the cross-encoder model. For each input candidate-job pair, the cross-encoder model can calculate a suitability score by considering the interaction of text and metadata. The cross-encoder model can process various information integrally, such as the competency identifier, proficiency level, and evidence strength of the candidate SSR, and the required competencies, mandatory conditions, and preferred conditions of the job posting. The cross-encoder model can apply an attention mechanism to each input pair to precisely evaluate the interaction between contextual relationships and competencies. Compared to a simple vector similarity-based evaluation, the cross-encoder-based precise comparison can more accurately reflect the detailed match between actual job requirements and candidate competencies. The application of cross-encoder models can increase the explainability and evaluation reliability of candidate-job matching.

[0241] In one embodiment, at step 332, the matching layer (120) can perform a precise suitability evaluation that considers contextual relationships beyond simple similarity comparison. The matching layer (120) can analyze the relationship between the semantic context within a sentence and competencies by integrating the text and metadata of the candidate SSR data and the job posting data. The matching layer (120) can reflect various evaluation factors, such as whether essential competencies are met, proficiency level, and strength of evidence, by utilizing the internal attention weights and feature interaction results of the cross-encoder model. A suitability evaluation that considers contextual relationships enables precise differentiation of detailed alignment with job-specific requirements, even if the same competencies are possessed. Through context-based precise evaluation, the matching layer (120) can overcome the limitations of existing embedding similarity-based evaluations and produce highly reliable matching results based on actual job suitability.

[0242] In at least one embodiment of the present disclosure, the matching layer (120) can perform cross-encoder-based re-ranking for each of the top N candidates. The matching layer (120) can receive the list of top N candidates derived in Step 1 as input. The matching layer (120) can generate input pairs for a cross-encoder model by combining SSR data and job posting data for each candidate. For each generated candidate-job input pair, the matching layer (120) can calculate a suitability score through the cross-encoder model. The cross-encoder-based re-ranking structure can comprehensively evaluate various information, such as the competency identifier, proficiency level, and evidence strength of the SSR for each candidate, as well as the required competencies, mandatory conditions, and preferential conditions of the job posting. By performing a detailed evaluation individually for each of the top N candidates, the matching layer (120) can reflect fine differences in suitability scores for each candidate. Cross-encoder-based re-ranking can increase the reliability of the final ranking calculation based on the actual suitability between the candidate and the job.

[0243] In at least one embodiment of the present disclosure, the matching layer (120) can efficiently perform re-ranking operations by inputting a plurality of candidate-job pairs in batch units. The matching layer (120) can collectively generate input pairs combining SSR data and job posting data for each of the top N candidates. The matching layer (120) can feed the generated candidate-job input pairs into a cross-encoder model in batch units. Batch unit input processing enables simultaneous performance of fit inference for a plurality of candidate-job pairs by utilizing parallel computing resources such as GPUs. The batch processing method can significantly improve the processing speed of re-ranking operations for a large group of candidates. Efficient batch processing can ensure the responsiveness and scalability of the system in real-time or large-scale recruitment matching environments.

[0244] In one embodiment, the matching layer (120) can generate candidate-job pairs through cross-encoder input formatting. The matching layer (120) can extract key fields such as competency identifiers, proficiency levels, and evidence strengths from the SSR data for each candidate. The matching layer (120) can extract key fields such as required competencies, mandatory conditions, preferred conditions, and weights from the job posting data. The matching layer (120) can combine the extracted candidate SSR data and job posting data into a single input sequence or template and convert them to fit the input format of the cross-encoder model. During the input formatting process, the separator, tag, order, etc., of each field can be clearly specified so that the model can recognize the structure of the information. Cross-encoder input formatting can improve the accuracy of detailed competency matching and contextual relationship analysis between candidate and job.

[0245] In one embodiment, the matching layer (120) can generate candidate-job pairs by combining candidate-specific SSR data and job posting data. The matching layer (120) can extract information such as competency identifiers, proficiency levels, and evidence strength from the SSR data for each candidate. The matching layer (120) can extract information such as required competencies, mandatory conditions, and preferred conditions from the job posting data. The matching layer (120) can combine the extracted candidate SSR data and job posting data into a single input pair. The candidate-job pair generation process provides an input structure that can evaluate whether there is a detailed match with the job requirements for each candidate. The candidate-job pair is subsequently used as an input value for a cross-encoder model, enabling a precise suitability evaluation.

[0246] In at least one embodiment of the present disclosure, the matching layer (120) can input the generated candidate-job pairs into a cross-encoder in batch units to perform fitness inference in parallel. The matching layer (120) can group multiple candidate-job pairs into a single batch and input them simultaneously into the cross-encoder model. Batch unit input enables the simultaneous calculation of fitness scores for multiple candidate-job pairs by utilizing parallel computing resources such as GPUs. Parallel inference processing can significantly improve the processing speed of re-ranking operations for a large pool of candidates. Batch unit inference can ensure the processing efficiency and scalability of the system in real-time recruitment matching or large-scale data environments.

[0247] In at least one embodiment of the present disclosure, the matching layer (120) can calculate the suitability for each candidate by reflecting whether essential competencies are satisfied, the level of proficiency, the strength of evidence, and contextual information. The matching layer (120) can calculate a primary suitability score for each candidate-job pair using the output value of a cross-encoder model. The matching layer (120) can apply a penalty if essential competencies are not satisfied by reflecting whether essential competencies are satisfied in the calculated suitability score. The matching layer (120) can adjust the suitability score upward for competencies with higher proficiency by applying the level of proficiency included in the SSR data as a weight. The matching layer (120) can further adjust the suitability score for competencies with strong evidence of verification by reflecting the strength of evidence in the SSR data. The matching layer (120) evaluates the semantic alignment between candidate SSR data and job posting data, and can reflect the detailed suitability with actual job requirements in the score. The first suitability calculation process can increase the reliability and explainability of the job suitability evaluation for each candidate by integratively reflecting various evaluation elements.

[0248] In at least one embodiment of the present disclosure, the matching layer (120) may apply scaling processing to the calculated suitability scores to ensure consistency in the score distribution. The matching layer (120) may calculate statistical indicators such as the minimum value, maximum value, and average value of the primary suitability scores. The matching layer (120) may normalize the calculated suitability scores to a predefined range, such as 0 to 1 or 0 to 100. The score scaling process may maintain a consistent distribution of suitability scores for each candidate, thereby increasing comparability in subsequent ranking and fairness correction stages. Score scaling may standardize suitability evaluation results between various jobs or job postings, thereby ensuring evaluation consistency throughout the entire system.

[0249] In at least one embodiment of the present disclosure, the matching layer (120) can normalize the output value of the model and utilize it for subsequent fit adjustment and final ranking calculation. The matching layer (120) can input the raw output value (raw score) of the cross-encoder model into a normalization function (e.g., softmax, minimax normalization, etc.). The normalized output value is a fit score for each candidate and can be consistently utilized for reflecting essential competencies in subsequent stages, proficiency weighting, fairness correction, etc. Normalizing the model output can increase the interpretability and comparability of the fit score, thereby improving the reliability of the final candidate ranking calculation. The normalized fit score can also be utilized for generating the final recommendation result report and explanatory information.

[0250] In one embodiment, the matching layer (120) can calculate a final score by adjusting the results of the suitability calculation for each candidate. The matching layer (120) can receive the first suitability score for each candidate, calculated through a two-stage re-ranking, as an input value. The matching layer (120) can adjust the input first suitability score by reflecting additional evaluation factors, such as whether essential competencies are met and the level of proficiency. The matching layer (120) can calculate the final suitability score for each candidate based on the adjusted score. The process of calculating the final suitability score may include step-by-step operations such as applying weights for each evaluation factor, imposing penalties, and summing contributions. The final suitability score can be directly used for determining candidate rankings and outputting recommendation results.

[0251] In one embodiment, the matching layer (120) can adjust the suitability score by reflecting whether the required competencies are met and the proficiency level. The matching layer (120) can evaluate whether the required competencies are met by comparing the list of required competencies included in the job posting data with the competency identifiers in the candidate SSR data. If the required competencies are not met, the matching layer (120) can adjust the score by applying a penalty to the suitability score. The matching layer (120) can check the proficiency level included in the SSR data for each competency and correct the score by applying a weighting factor according to the proficiency level. The reflection of whether the required competencies are met and the proficiency level can be performed independently in stages, and the results of each stage can be cumulatively reflected in the calculation of the final suitability score.

[0252] In at least one embodiment of the present disclosure, the matching layer (120) may reflect whether the required competencies included in the job posting data are satisfied in the suitability calculation process. The matching layer (120) may extract required competency items from the job posting data. The matching layer (120) may determine whether each candidate satisfies the required competencies by comparing the list of competency identifiers in the candidate-specific SSR data with the required competency items. Whether the required competencies are satisfied may be evaluated according to logical operations or matching rules. The evaluation result regarding whether the required competencies are satisfied may be used as an input value for the suitability score adjustment step. Various score adjustment policies may be applied, such as imposing a penalty when the required competencies are not satisfied or maintaining the weight when they are satisfied.

[0253] In at least one embodiment of the present disclosure, the matching layer (120) can adjust the suitability score by utilizing the proficiency level included in the SSR data as a weight. The matching layer (120) can identify the proficiency level for each competency within the SSR data. The matching layer (120) can assign a predefined weight coefficient according to the proficiency level. The matching layer (120) can calculate a score for each competency by multiplying the proficiency weight coefficient by the suitability contribution for each competency. The proficiency weight coefficient can be set, for example, to beginner (1.0), intermediate (1.2), advanced (1.5), etc. The result of applying the proficiency weight can be cumulatively reflected in the calculation of the final suitability score.

[0254] In at least one embodiment of the present disclosure, the matching layer (120) can evaluate whether the required competencies are met for each candidate. The matching layer (120) can compare the list of required competencies in the job posting data with the list of competency identifiers in the candidate SSR data using a 1:1 matching method or a set operation method. The matching layer (120) can logically determine whether all required competencies are included in the candidate SSR. The evaluation of whether the required competencies are met can be performed independently for each candidate. The evaluation of whether the required competencies are met can be used as an input value for subsequent score adjustment and penalty imposition steps.

[0255] In at least one embodiment of the present disclosure, the matching layer (120) may apply a penalty to the suitability score for candidates who do not meet the required competencies. For candidates who do not meet the required competencies, the matching layer (120) may multiply or subtract a predefined penalty coefficient (e.g., -0.5, -1.0, etc.) from the suitability score. The method of applying the penalty may be set differently depending on the number of unmet competencies, importance, job-specific policies, etc. The application of the penalty may cause a decrease in the absolute value of the suitability score or a drop in ranking. The result of applying the penalty may be directly reflected in the calculation of the final score and the determination of the candidate ranking.

[0256] In one embodiment, the matching layer (120) may readjust the weights used to calculate the suitability based on whether the required competencies are met. The matching layer (120) may apply a basic weight or a corrected weight to a candidate who meets all the required competencies. For a candidate who does not meet the required competencies, the matching layer (120) may lower the weight coefficient or set the contribution of a specific competency to zero. The weight readjustment may be performed independently by competency, by candidate, and by job. The results of the weight readjustment may be cumulatively reflected in each candidate's final suitability score.

[0257] In one embodiment, the matching layer (120) can reflect the proficiency level included in the SSR data in the calculation of the suitability. The matching layer (120) can identify the proficiency level (Level) for each competency within the SSR data. The matching layer (120) can assign a predefined level weighting coefficient according to the proficiency level. The matching layer (120) can calculate a score for each competency by multiplying the level weighting coefficient by the suitability contribution for each competency. The reflection of proficiency levels can be performed independently for each candidate and for each competency. The results of the proficiency level reflection can be cumulatively reflected in the calculation of the final suitability score.

[0258] In one embodiment, the matching layer (120) can calculate a suitability score by applying a level weighting factor according to the proficiency level. The matching layer (120) checks the proficiency level for each competency in the SSR data and can assign a level weighting factor, for example, beginner (1.0), intermediate (1.2), advanced (1.5), etc. The matching layer (120) can calculate a score for each competency by multiplying the suitability contribution for each competency by the corresponding level weighting factor. The application of the level weighting factor can cause the effect of assigning a relatively high suitability score to candidates with high proficiency. The result of applying the level weighting factor can be cumulatively reflected in the calculation of the final suitability score.

[0259] In one embodiment, the method can determine the candidate rankings by sorting candidates within a candidate group according to the suitability of each candidate. The matching layer (120) can sort the candidates within the candidate group in descending order based on the final suitability score calculated for each candidate. The candidate ranking determination can be automatically calculated based on the magnitude of the final suitability score. The ranking determination result can be used as an input value for subsequent fairness correction and result output steps.

[0260] In at least one embodiment of the present disclosure, the matching layer (120) may calculate a final suitability score by summing the contributions for each competency. The matching layer (120) may calculate the suitability contribution for each competency within the SSR data, and then sum all the contributions for each competency to calculate the final suitability score for each candidate. The summing of contributions for each competency may be performed by reflecting various factors, such as essential competencies, preferred competencies, proficiency levels, and evidence strength, as weights. The final suitability score may provide the effect of increasing the precision and reliability of candidate-job matching.

[0261] In one embodiment, the matching layer (120) can generate a ranking result based on the result of calculating the suitability for each candidate. The matching layer (120) can receive the final suitability score for each candidate, calculated through a two-stage re-ranking, as an input value. The matching layer (120) can execute a ranking logic that calculates the rank of each candidate based on the input suitability score. The process of generating the ranking result may include step-by-step operations such as sorting suitability scores in descending order, handling ties, and applying a cut-off line. The ranking result can be used as an input value for subsequent fairness correction and descriptive information combination steps. The matching layer (120) can also generate skill matching basis information for each candidate included in the ranking result. The generation of the ranking result can be used as core data for the candidate recommendation result report.

[0262] In at least one embodiment of the present disclosure, the matching layer (120) may generate a candidate recommendation result report by combining the final ranking and recommendation basis information. The matching layer (120) may combine the calculated final ranking list and recommendation basis information for each candidate into a data structure. The recommendation basis information may include detailed items such as top matching capabilities, vector similarity values, proficiency levels, and strength of the basis. The candidate recommendation result report may be provided through various output channels, such as a dashboard for companies or institutions, an external system API, or a feedback interface. The result report may be composed by integrating multiple pieces of information, such as the final ranking, score, basis information, and a summary of deviations by group. The process of generating the result report may include step-by-step processing such as data combination, format conversion, and interface integration.

[0263] In one embodiment, the matching layer (120) can determine the ranking of each candidate based on the calculated suitability score. The matching layer (120) can collect the suitability score calculated for each candidate. The matching layer (120) can assign a ranking index by sorting the suitability scores in descending order. The ranking calculation process may include additional rules, such as handling ties and applying a cut-off line. The ranking calculation results may be stored as a final ranking and score map for each candidate.

[0264] In at least one embodiment of the present disclosure, the matching layer (120) may generate supporting information related to skill matching for each candidate in the form of a draft. The matching layer (120) may derive a matching result by comparing the competency identifiers of the candidate SSR data and the job posting data. The matching layer (120) may extract detailed items such as top matching competency, vector similarity, proficiency level, and strength of evidence to form supporting information. The draft supporting information may be generated independently for each candidate. The draft supporting information may subsequently be utilized in the explanation generation and report combination steps of the output layer.

[0265] In at least one embodiment of the present disclosure, the matching layer (120) may determine the ranking of candidates based on the calculated final suitability score. The matching layer (120) may receive a list of final suitability scores as input and perform sorting in descending order. The matching layer (120) may determine the ranking of each candidate by assigning a ranking index to the sorted list. The ranking determination process may apply additional rules in stages, such as tie-breaking and the application of a cut-off line. The determined ranking information may be used as input for the recommendation result report and the fairness correction step.

[0266] In one embodiment, the matching layer (120) may assign ranks by sorting the suitability scores of each candidate in descending order. The matching layer (120) may sort the array of suitability scores using a descending sorting algorithm (e.g., Quicksort, Mergesort, etc.). The sorted list of scores may be arranged in order from the highest-ranked candidate to the lowest-ranked candidate. The descending sorting results may be stored for each candidate along with a rank index. The sorting process may be optimized so that it can be performed efficiently on a large set of candidates.

[0267] In at least one embodiment of the present disclosure, the matching layer (120) may determine the ranking by applying a predefined cutoff line or additional criteria to candidates with the same score. The matching layer (120) may apply tie-breaking rules to multiple candidates with the same suitability score. Tie-breaking rules may be implemented, for example, by sequentially comparing additional evaluation items (e.g., proficiency level, strength of evidence, time of support, etc.). The matching layer (120) may determine whether to include candidates in the final ranking by applying a predefined cutoff line (e.g., Top-K, score threshold, etc.). As a result of applying the cutoff line, only some of the candidates with the same score may be included in the final recommendation list. The tie-breaking and cutoff line application process can ensure consistency and fairness in ranking calculation.

[0268] In one embodiment, the matching layer (120) can generate recommendation basis information for each candidate. The matching layer (120) can generate basis information based on the matching results of competency identifiers between candidate SSR data and job posting data. The basis information may include multiple items such as top matching competency, vector similarity, proficiency level, and basis strength. The process of generating basis information can be performed independently for each item. The generated basis information can be used as explanatory information for a recommendation result report.

[0269] In at least one embodiment of the present disclosure, the matching layer (120) may record top matching competencies for each candidate. The matching layer (120) may select competencies with high matching or similarity by comparing the competency identifier of the candidate SSR data with the required competency of the job posting data. The matching layer (120) may generate a list of top matching competencies based on the top N (e.g., 3, 5, etc.) among the selected competencies. Top matching competencies may be recorded along with similarity scores by competency, proficiency levels, strength of evidence, etc. The record of top matching competencies may be included as a core item of recommendation evidence information.

[0270] In at least one embodiment of the present disclosure, the matching layer (120) may include a vector similarity value for each candidate in the supporting information. The matching layer (120) may calculate numerical values, such as cosine similarity or inner product distance, between the embedding vectors of the candidate SSR data and the job posting data. The calculated vector similarity value may be included in the supporting information for each candidate. The vector similarity value may be used as a basis to numerically explain the semantic fit between the candidate and the job. Including the vector similarity value may enhance the explainability of the recommendation results.

[0272] FIG. 4a is a drawing for explaining the internal processing structure of an output layer according to one embodiment, and FIG. 4b is a drawing for explaining in detail the operation of an output layer according to one embodiment.

[0273] Referring to FIGS. 4a and 4b, in at least one embodiment of the present disclosure, an output layer (130) may receive a ranking result generated by a matching layer (120) as input, perform fairness correction and description generation processing, and finally output a candidate recommendation result. The output layer (130) may receive ranking result data from the matching layer (120) as input. The output layer (130) may perform a fairness correction (post-processing) procedure step by step on the input ranking result. After the fairness correction is completed, the output layer (130) may execute a description information generation logic. The output layer (130) may implement the fairness correction processing and the description information generation as separate modules or functions. The output layer (130) may generate a final candidate recommendation result in a data structure that combines fairness correction and description information. The output layer (130) may provide the generated candidate recommendation result through various output channels, such as an enterprise dashboard, an external system API, or a feedback interface. The output layer (130) may include multiple pieces of information in the final output result, such as rankings, scores, evidence information, and a summary of deviations by group. The output layer (130) may record the processing results of each step as a log or metadata and utilize them for system operation and verification.

[0274] In at least one embodiment of the present disclosure, the output layer (130) may generate a final ranking and descriptive information thereto. The output layer (130) may produce a final ranking list to which fairness correction has been applied. The output layer (130) may generate descriptive information for each candidate corresponding to the final ranking. The descriptive information may include detailed items such as SSR-based competency mapping, proficiency level, evidence strength, and matching score. The output layer (130) may combine the final ranking and descriptive information to generate a consistent result structure. The output layer (130) may output the generated result structure to various interfaces such as a report, dashboard, and API.

[0275] In one embodiment, at step 410, the output layer (130) may perform a correction operation to enhance the fairness of the candidate recommendation results. The output layer (130) may classify the candidates included in the ranking results according to predefined group criteria (e.g., gender, age, region, etc.). The output layer (130) may perform a statistical operation to calculate the recommendation rate or exposure rate by group. The output layer (130) may use a formula to measure the deviation of the recommendation rate or exposure rate between groups (e.g., = |RR(g1) - RR(g2)|) can be applied. The output layer (130) can determine whether the measured deviation is below a predefined fairness threshold (ε). If the deviation exceeds the threshold, the output layer (130) can execute a score or order adjustment algorithm. The output layer (130) can generate a corrected ranking that satisfies the fairness criteria through score rescaling or limited order change. The output layer (130) can apply the minimal change principle so that the degradation of the existing ranking performance is minimized during the correction process. The output layer (130) can record metadata such as group deviations, adjustment history, and performance changes related to the corrected ranking.

[0276] In at least one embodiment of the present disclosure, the output layer (130) may classify candidates included in the candidate ranking according to predefined group criteria and calculate a recommendation rate or exposure rate by group. The output layer (130) may identify group attributes (e.g., gender, age, region, etc.) for each candidate as metadata. The output layer (130) may calculate a recommendation rate (RR(g)) or exposure rate (ER(g)) by aggregating the number of candidates included in Top-K by group. The output layer (130) may organize the recommendation rate or exposure rate by group into a table to enable comparison between groups.

[0277] In at least one embodiment of the present disclosure, the output layer (130) calculates a recommendation rate (RR(g)) or exposure rate (ER(g)) for each group, and a deviation between groups ( ) can be calculated. The output layer (130) can arithmetically calculate the difference in recommendation rate or exposure rate by group. The output layer (130) = max_g(RR(g)) - min_g(RR(g)) or A formula such as = max_g(ER(g)) - min_g(ER(g)) can be applied. The output layer (130) is the calculated deviation ( It is possible to determine whether ) meets fairness standards.

[0278] In one embodiment, the output layer (130) has a processability threshold ( Constraints such as (≤ ε) and a performance degradation tolerance limit (loss ≤ δ) can be set. The output layer (130) can store the fairness threshold (ε) and the performance degradation limit (δ) as system parameters. The output layer (130) deviation ( If ) exceeds the threshold (ε), a correction algorithm can be applied to keep the performance degradation (loss) within the limit (δ). The output layer (130) can verify step-by-step whether the constraint is satisfied.

[0279] In at least one embodiment of the present disclosure, the output layer (130) may apply a correction function (g) to adjust the score or order of the candidate rankings to satisfy the constraints. The output layer (130) may call the correction function (g) to perform score rescaling or order change. The output layer (130) may pass existing scores, group recommendation rates, thresholds, performance degradation limits, etc., as input values ​​to the correction function. The output layer (130) may receive a list of adjusted scores or orders as the output value of the correction function. The output layer (130) may finally verify whether the adjustment result satisfies both the fairness criteria and the performance limits.

[0280] In at least one embodiment of the present disclosure, the output layer (130) can generate a corrected ranking that satisfies fairness criteria while maintaining existing ranking performance as much as possible by performing score rescaling or limited order adjustment according to the minimal change principle. When rescaling scores, the output layer (130) may apply an algorithm that preserves the existing score distribution as much as possible (e.g., least squares error minimization). When adjusting order, the output layer (130) may minimize overall ranking changes by changing the order only within Top-K in a limited manner. The output layer (130) may log rank changes, score changes, and group recommendation rate changes before and after correction. The output layer (130) may include metadata related to the corrected ranking in the final output result.

[0281] In one embodiment, the output layer (130) may include a fairness correction procedure that retrospectively adjusts the final ranking without retraining the model. The output layer (130) may directly utilize the ranking results calculated from the existing matching layer (120) as input values. The output layer (130) may retrospectively adjust only the score or order without changing or retraining the model parameters. The output layer (130) can increase system operational efficiency through a post-processing method. Since the output layer (130) does not require model retraining, rapid correction is possible in a real-time or batch processing environment. The output layer (130) may record the results of the post-processing as a separate log or report.

[0282] In one embodiment, at step 420, the output layer (130) may perform an operation to ensure explainability for the recommendation result. The output layer (130) may execute explanation information generation logic for each candidate included in the final ranking. The output layer (130) may derive competency mapping results by comparing SSR data and job posting data. The output layer (130) may extract detailed items such as proficiency level, evidence strength, and matching score for each competency. The output layer (130) may modularize the explanation information generation process and output it in various explanation formats (e.g., tables, graphs, text, etc.). The output layer (130) may combine the generated explanation information into the final result structure.

[0283] In one embodiment, at step 430, the output layer (130) can generate a candidate recommendation result report by combining the final ranking and recommendation basis information. The output layer (130) can combine the final ranking list with fairness correction applied and the recommendation basis information for each candidate into a data structure. The output layer (130) may include detailed items in the recommendation basis information, such as top matching capabilities, vector similarity values, proficiency levels, and strength of the basis. The output layer (130) can provide the candidate recommendation result report through various output channels, such as a dashboard for companies or institutions, an external system API, or a feedback interface. The output layer (130) may integrate and include multiple pieces of information in the result report, such as the final ranking, score, basis information, and a summary of deviations by group. The output layer (130) can perform step-by-step processing, such as data combination, format conversion, and interface integration, during the process of generating the result report.

[0284] In at least one embodiment of the present disclosure, the output layer (130) can generate recommendation basis information by matching the candidate group's SSR data with the job posting data. The output layer (130) can derive a matching result by comparing the competency identifier included in the SSR data with the required competency of the job posting data. The output layer (130) can extract detailed information such as proficiency level, basis strength, and vector similarity for each matched competency. The output layer (130) can structure and store the derived basis information for each candidate. The output layer (130) can convert the basis information structure into various formats such as a table, list, or graph.

[0285] In at least one embodiment of the present disclosure, the output layer (130) maps the candidate's competency identifier to the required competency of the job posting based on SSR and can identify the Evidence Strength and Level of proficiency for each competency. The output layer (130) can verify whether there is a match between the competency identifier of the SSR data and the required competency of the job posting. The output layer (130) can extract the Level of proficiency value recorded in the SSR for the matched competency. The output layer (130) can verify the Evidence Strength value for each competency and use it as a reliability indicator. The output layer (130) can organize the matching results, proficiency, and Evidence Strength information by competency for each candidate.

[0286] In one embodiment, the output layer (130) may select the top M core competencies for each candidate and configure an evidence package including competencies, evidence strength, and proficiency levels. The output layer (130) may select the top M competencies with high matching scores from the SSR data for each candidate. The output layer (130) may group detailed information, such as proficiency levels, evidence strength, and matching scores, into a package for each selected competency. The output layer (130) may structure the evidence packages for each candidate and include them in a result report. The output layer (130) may convert the format of the evidence packages into various forms, such as tables, lists, and graphs.

[0287] In at least one embodiment of the present disclosure, the output layer (130) can provide fair and explainable candidate recommendation results by performing fairness correction and explanation generation processing separately and then combining them, thereby enabling users and companies to trust and utilize the recommendation results. The output layer (130) can independently execute fairness correction processing and explanation information generation. The output layer (130) can combine the two processing results to create a consistent final recommendation result structure. The output layer (130) can integrate detailed items such as fairness correction history, explanation information, and evidence packages and include them in the result report. The output layer (130) can provide the final recommendation results to various stakeholders, such as companies, institutions, and candidates. The output layer (130) can increase the trust of users and companies through the results combining fairness correction and explanation information. The output layer (130) can record the usage history of the recommendation results, feedback, and trust indicators as separate logs.

[0288] In at least one embodiment of the present disclosure, the output layer (130) may receive a suitability score for each candidate calculated by the matching layer (120) as input and perform a candidate ranking initialization operation. The output layer (130) may receive a list of suitability scores for each candidate from the matching layer (120) as input. The output layer (130) may store the input suitability score list in an internal data structure. The output layer (130) may execute a candidate ranking initialization process step by step based on the stored suitability score list. The output layer (130) may verify the reliability and consistency of the input suitability score during the candidate ranking initialization process. The output layer (130) may calculate an initial ranking for each candidate based on the verified suitability score. The output layer (130) may transmit the results of the initial ranking calculation to subsequent sorting and Top-K setting steps.

[0289] In one embodiment, the output layer (130) can execute a series of operations in stages, including initial sorting criteria and Top-K settings, during the candidate ranking initialization process. The output layer (130) can sort the candidate list by first applying the initial sorting criteria. The output layer (130) can check the K value for Top-K settings in the sorted candidate list. The output layer (130) can select the top K candidates based on the K value. The output layer (130) can store the selected Top-K candidate list in a separate data structure. The output layer (130) can provide the Top-K candidate list as an input value to a subsequent final fit-based sorting step.

[0290] In one embodiment, the output layer (130) can sort the list of candidates based on the suitability score calculated for each candidate. The output layer (130) can evaluate the suitability score corresponding to each candidate through a comparison operation. The output layer (130) can sort the list of candidates in descending order of suitability scores. The output layer (130) can store the sorting results in the form of a temporary list or array. The output layer (130) can assign a rank of the sorted list to each candidate.

[0291] In at least one embodiment of the present disclosure, the output layer (130) may apply a descending order of suitability scores or a predefined priority rule as a sorting criterion. The output layer (130) may retrieve the predefined priority rule from system parameters or a policy table. The output layer (130) may apply priority rules (e.g., experience, qualifications, group attributes, etc.) as additional sorting criteria in addition to the suitability score. If multiple sorting criteria are applied, the output layer (130) may perform multi-level sorting according to priority. The output layer (130) may record the results of application for each sorting criterion as a log or metadata.

[0292] In at least one embodiment of the present disclosure, the output layer (130) may set the top K candidates from a sorted list of candidates as the Top-K candidate group. The output layer (130) may extract the Top-K candidates by referencing the top K indices of the sorted list. The output layer (130) may map the identifier of each Top-K candidate to a suitability score and store it in a separate data structure. The output layer (130) may transmit the result of setting the Top-K candidate group to a subsequent processing step.

[0293] In one embodiment, the output layer (130) can dynamically determine the Top-K value (K) based on system parameters or user input. The output layer (130) can query the K value from a system parameter table or configuration. The output layer (130) can reflect the K value entered through the user interface in real time. The output layer (130) can reset the Top-K candidate group when the K value changes. The output layer (130) can record the history of K value determination as a log or metadata.

[0294] In at least one embodiment of the present disclosure, the output layer (130) may perform a final fit-based sorting operation for the Top-K candidate group. The output layer (130) may sort by comparing the final fit scores of each candidate within the Top-K candidate group. The output layer (130) may generate a final ranking list from the sorted Top-K candidate list. The output layer (130) may transmit the final ranking list to a subsequent fairness correction and explanation information generation step.

[0295] In at least one embodiment of the present disclosure, the output layer (130) can re-rank each candidate according to the final suitability score. The output layer (130) can check for ties in the final suitability score of each candidate. In the event of a tie, the output layer (130) can re-rank by applying predefined additional rules (e.g., experience, qualification, random, etc.). The output layer (130) can store the re-ranked results in a data structure. The output layer (130) can record the re-ranking history as a log or metadata.

[0296] In one embodiment, the output layer (130) may select candidates from the Top-K candidate group in order of highest final suitability score. The output layer (130) may compare the suitability scores of each Top-K candidate and sort them in descending order. The output layer (130) may assign a final rank to each candidate according to the sorted order. The output layer (130) may transmit the list of candidates included in the final rank to a subsequent output stage.

[0297] In one embodiment, the output layer (130) may apply additional selection rules, such as tie-breaking and cut-off lines. The output layer (130) may apply a predefined cut-off score or priority criteria to candidates with a tie. The output layer (130) may exclude candidates below the cut-off line from the Top-K candidate group. The output layer (130) may store the results of tie-breaking and cut-off line application in a data structure. The output layer (130) may record the history of selection rule application as a log or metadata.

[0298] In at least one embodiment of the present disclosure, the output layer (130) may generate and maintain a fitness score map corresponding to each Top-K candidate. The output layer (130) may store the fitness score for each Top-K candidate as map data in the form of key-value pairs. The output layer (130) may update the score map in real time or version control the change history. The output layer (130) may maintain the score map in memory or a database so that it can be referenced in a subsequent processing step.

[0299] In at least one embodiment of the present disclosure, the output layer (130) can be utilized for subsequent fairness correction and descriptive information generation operations based on the score map. The output layer (130) can provide the score map as an input value for a fairness correction algorithm. The output layer (130) can link the score map with candidate-specific evidence data in the descriptive information generation logic. The output layer (130) can record the history of score map utilization as a log or metadata. The output layer (130) can ensure the reliability and consistency of the final candidate recommendation results based on the score map.

[0300] In at least one embodiment of the present disclosure, at step 410, the output layer (130) may perform post-processing on the ranking results provided by the matching layer (120). The output layer (130) may use the scores and ranks for each candidate included in the ranking results as input values. The output layer (130) may refer to predefined group criteria and correction constraints for the post-processing operation. The output layer (130) may sequentially execute a series of post-processing steps, such as calculating recommendation rates or exposure rates by group, calculating deviations between groups, and applying a correction function. The output layer (130) may combine the post-processing results with the existing ranking results to generate a final recommendation result.

[0301] In at least one embodiment of the present disclosure, at step 411, the output layer (130) may classify candidates included in the candidate ranking according to predefined group criteria and calculate a recommendation rate or exposure rate by group. The output layer (130) may identify the group attributes (e.g., gender, age, region, etc.) of each candidate in the candidate ranking list. The output layer (130) may divide the candidate list into multiple groups according to the group criteria. The output layer (130) may calculate a recommendation rate (RR) or exposure rate (ER) for each divided group. Based on the calculated recommendation rate or exposure rate by group, the output layer (130) may calculate the deviation between groups ( ) can be calculated. The output layer (130) calculates the fairness threshold ( between groups deviation ( If it exceeds ≤ ε, it is possible to determine whether to apply a correction function.

[0302] In one embodiment, the output layer (130) may pre-set a group definition policy for a fairness correction operation. The output layer (130) may retrieve the group definition policy from system parameters or a policy table. The output layer (130) may check the group criteria (e.g., gender, age, region, etc.) included in the group definition policy. The output layer (130) may dynamically select the group criteria to classify the candidate list according to the group definition policy. When the group definition policy is changed, the output layer (130) may re-execute the group classification and correction operation.

[0303] In at least one embodiment of the present disclosure, the output layer (130) may classify candidates included in the candidate ranking according to predefined group criteria. The output layer (130) may classify candidates by group criteria by referring to group attribute information stored for each candidate. The output layer (130) may store a list of candidates for each classified group in a separate data structure. The output layer (130) may provide the list of candidates for each group as an input value to a subsequent recommendation rate or exposure rate calculation step.

[0304] In at least one embodiment of the present disclosure, the output layer (130) may utilize various attributes such as gender, age, and region as group criteria according to a group definition policy. The output layer (130) may apply multiple attributes (e.g., gender, age, region, etc.) specified in the group definition policy as group classification criteria. The output layer (130) may divide the candidate list by each attribute or generate subdivided groups according to a combination of multiple attributes. The output layer (130) may record the classification results by group criteria as a log or metadata.

[0305] In one embodiment, at step 412, the output layer (130) can set a correction constraint in the fairness correction operation. The output layer (130) can retrieve the correction constraint from system parameters or a policy table. The output layer (130) includes a fairness threshold ( Values ​​such as ≤ ε) and performance degradation allowance limit (loss ≤ δ) can be checked. The output layer (130) can determine whether to apply a correction function and the correction strength based on correction constraints.

[0306] In one embodiment, the output layer (130) has a processability threshold ( Constraints such as ≤ ε) and a performance degradation tolerance limit (loss ≤ δ) can be set. The output layer (130) has a process threshold ( By setting ≤ ε), it can be determined whether the deviation between groups is within the allowable range. The output layer (130) can set a performance degradation allowable limit (loss ≤ δ) to restrict the existing ranking performance degradation from exceeding the allowable limit during the correction process. The output layer (130) can use the results of setting the fairness threshold and the performance degradation allowable limit as reference values ​​for applying the correction function.

[0307] In one embodiment, at step 413, the output layer (130) can determine whether to apply a correction function based on set constraints. The output layer (130) determines whether to apply the deviation between groups ( The output layer (130) can calculate the ) and performance degradation (loss) values. The output layer (130) can perform a comparison operation to determine whether the calculated value satisfies the constraint. If the output layer (130) does not satisfy the constraint, it can determine to apply a correction function (g). If the output layer (130) satisfies the constraint, it can maintain the existing ranking result.

[0308] In one embodiment, the output layer (130) can perform a group classification operation that classifies candidates included in the candidate ranking according to group criteria. The output layer (130) can divide the candidate list by group criteria by checking group attribute information for each candidate. The output layer (130) can store the candidate list for each divided group in a separate data structure. The output layer (130) can provide the group classification results as input values ​​to a subsequent recommendation rate or exposure rate calculation step.

[0309] In at least one embodiment of the present disclosure, the output layer (130) may apply predefined criteria such as gender, age, and region in a group classification operation. The output layer (130) may select multiple attributes such as gender, age, and region as group classification criteria. The output layer (130) may divide the candidate list by each attribute and organize data by group. The output layer (130) may generate basic data for calculating recommendation rates or exposure rates for each divided group.

[0310] In one embodiment, the output layer (130) can divide the candidate list by group criteria to generate basic data for calculating the recommendation rate or exposure rate for each group. The output layer (130) can verify the size and members of the candidate list divided by group criteria. The output layer (130) can extract the number of candidates and ranking information required for calculating the recommendation rate (RR) or exposure rate (ER) for each group. The output layer (130) can provide the generated basic data as input values ​​to a subsequent recommendation rate or exposure rate calculation step.

[0311] In one embodiment, the output layer (130) can generate multiple groups by simultaneously considering multiple attributes (e.g., gender, age, region, etc.) in a group classification operation. The output layer (130) can subdivide candidate groups according to combinations of multiple attributes (e.g., gender + age, gender + region, etc.). The output layer (130) can store a list of candidates for each subdivided group in a separate data structure. The output layer (130) can provide the list of groups by combination of multiple attributes as input values ​​for a subsequent deviation measurement and correction operation.

[0312] In one embodiment, the output layer (130) can subdivide a candidate group according to a combination of multiple attributes and perform deviation measurement and correction operations for each group. The output layer (130) can calculate a recommendation rate (RR) or exposure rate (ER) for each subdivided group. Based on the recommendation rate or exposure rate for each group, the output layer (130) calculates the deviation between groups ( ) can be calculated. The output layer (130) can use the group deviation measurement results as input values ​​for applying a correction function.

[0313] In at least one embodiment of the present disclosure, the output layer (130) calculates a recommendation rate or exposure rate by group and a deviation between groups ( A deviation measurement operation can be performed to calculate ). The output layer (130) compares the results of the group recommendation rate (RR) or exposure rate (ER) calculation to calculate the deviation between groups ( ) can be calculated. The output layer (130) calculates the deviation ( ) fairness threshold( It can be determined whether correction is necessary by comparing with ≤ ε). The output layer (130) can provide the deviation measurement result as an input value to the score / order adjustment step.

[0314] In one embodiment, the output layer (130) can calculate the recommendation rate (RR(g)) for each group. The output layer (130) can calculate the recommendation rate (RR(g)) by dividing the number of people included in the Top-K candidates for each group by the total number of people in the group. The output layer (130) can store the calculated recommendation rate (RR(g)) values ​​in a separate data structure for each group. The output layer (130) can use the result of calculating the recommendation rate as an input value for calculating the deviation between groups.

[0315] In one embodiment, the output layer (130) can use the recommendation rate calculation result as an input value for calculating the deviation between groups. The output layer (130) performs a comparison operation on the recommendation rate (RR(g)) values ​​for each group to calculate the maximum and minimum recommendation rate difference between groups ( ) can be calculated. The output layer (130) is the calculated The value can be compared with a fairness threshold (ε) to determine whether correction is necessary. The output layer (130) can record the results of the recommendation rate calculation and deviation calculation as a log or metadata.

[0316] In at least one embodiment of the present disclosure, the output layer (130) can calculate an exposure rate (ER(g)) for each group. The output layer (130) can calculate the exposure rate (ER(g)) by calculating the exposure frequency of each group member within the Top-K candidate list for each group. The output layer (130) can store the calculated exposure rate (ER(g)) values ​​in a separate data structure for each group. The output layer (130) can calculate the exposure rate calculation results as a deviation between groups ( It can be used as an input value for calculation and correction operations.

[0317] In one embodiment, the output layer (130) calculates the exposure rate result based on the deviation between groups ( ) can be used as an input value for calculation and correction operations. The output layer (130) calculates the exposure rate (ER(g)) values ​​for each group and calculates the difference between the maximum and minimum exposure rates ( ) can be calculated. The output layer (130) is the calculated The value can be compared with a fairness threshold (ε) to determine whether correction is necessary. The output layer (130) can record the exposure rate calculation and deviation calculation results as a log or metadata.

[0318] In one embodiment, the output layer (130) may perform a score / order adjustment operation to adjust the score or order of the candidate rankings based on the deviation measurement results and correction constraints. The output layer (130) the deviation measurement results ( The output layer (130) can determine whether to apply a correction function by comparing the correction constraints (fairness threshold, performance degradation allowance limit, etc.) with the correction constraints. The output layer (130) can perform score rescaling or limited order adjustment of the candidate rankings by applying the correction function (g). The output layer (130) can generate a corrected final ranking by combining the score and order adjustment results with the existing ranking results.

[0319] In one embodiment, the output layer (130) has a processability threshold ( It is possible to set (≤ ε). The output layer (130) can retrieve a fairness threshold (ε) value from system parameters or a policy table. The output layer (130) can set the fairness threshold (ε) setting result as a deviation between groups ( It can determine whether correction is necessary by comparing with ). The output layer (130) can use the result of setting the fairness threshold (ε) as a reference value for applying the correction function.

[0320] In at least one embodiment of the present disclosure, the output layer (130) may use the result of setting a fairness threshold as a reference value for applying a correction function. The output layer (130) uses the fairness threshold (ε) and the deviation between groups ( Compare with In the case of > ε, the application of a correction function (g) can be determined. The output layer (130) is If ≤ ε, the existing ranking results can be maintained. The output layer (130) can record the fairness threshold setting and comparison results as a log or metadata.

[0321] In at least one embodiment of the present disclosure, the output layer (130) may adjust the score or order of the candidate rankings by applying a correction function (g) to satisfy the constraints. The output layer (130) may prioritize the principle of minimal change when applying the correction function (g). The output layer (130) may minimize the degradation of existing ranking performance through score rescaling or limited order adjustment. The output layer (130) may generate a corrected final ranking by combining the result of applying the correction function with the existing ranking result.

[0322] In at least one embodiment of the present disclosure, the output layer (130) can generate a corrected ranking that satisfies fairness criteria while maintaining existing ranking performance as much as possible by performing score rescaling or limited order adjustment according to the minimal change principle. When rescaling, the output layer (130) can minimize the difference between each candidate's existing suitability score and the corrected score. When performing limited order adjustment, the output layer (130) can allow only minimal rank changes within Top-K candidates. The output layer (130) can record the results of applying the minimal change principle as a log or metadata. The output layer (130) can verify that the performance loss (loss) is within an allowable limit (δ) by comparing the corrected final ranking with the existing ranking result. The output layer (130) can output the verified correction result as a final candidate recommendation result.

[0323] In one embodiment, the output layer (130) can generate a final ranking and descriptive information thereto. The output layer (130) can receive the final ranking data provided by the matching layer (120) as input. The output layer (130) can refer to SSR data and job posting data for each candidate included in the input final ranking. The output layer (130) can generate descriptive information by comparing the competency identifier included in the SSR data with the required competency included in the job posting data. The output layer (130) can combine the generated descriptive information with the final ranking data to form candidate-specific recommendation basis information. The output layer (130) can include detailed information such as proficiency level, strength of the basis, and match in the recommendation basis information. The output layer (130) can output the final result containing the recommendation basis information through an enterprise dashboard, an external system API, or a user interface.

[0324] In one embodiment, the output layer (130) can match competency identifiers included in the candidate group's SSR data with required competencies included in the job posting data. The output layer (130) can compare each competency identifier in the SSR data with each required competency in the job posting data in a 1:1 or 1:N mapping manner. The output layer (130) can determine the correspondence relationship by applying synonyms, similar expressions, and standard classification system mapping rules between competency identifiers. The output layer (130) can store the matched competency identifier pairs in a separate mapping table. Based on the mapping table, the output layer (130) can check the status of required competency fulfillment for each candidate.

[0325] In at least one embodiment of the present disclosure, the output layer (130) can identify at least one of a proficiency level or a strength of evidence for a corresponding competency identifier. The output layer (130) can check the proficiency level field value included in the SSR data. The output layer (130) can check the strength of evidence field value included in the SSR data. The output layer (130) can extract at least one of a proficiency level or a strength of evidence for each corresponding competency identifier. The output layer (130) can include the extracted proficiency level and strength of evidence information in the recommended evidence information for each candidate.

[0326] In one embodiment, the output layer (130) can generate recommendation basis information by matching the identified information to each candidate included in the final ranking. The output layer (130) can refer to the mapping results of SSR data and job posting data for each candidate in the final ranking list. The output layer (130) can generate recommendation basis information by combining detailed information such as the competency identifier, proficiency level, strength of evidence, and match status corresponding to each candidate. The output layer (130) can combine the generated recommendation basis information with the final ranking data and output it as a structured result package.

[0327] In one embodiment, the output layer (130) may apply a mapping policy that matches competency identifiers included in the candidate group's SSR data with required competencies included in the job posting data. The output layer (130) may define matching criteria between the competency identifiers of the SSR data and the required competencies of the job posting data according to the mapping policy. The output layer (130) may include synonym integration, allowance of similar expressions, and mapping rules based on a standard classification system in the mapping policy. The output layer (130) may apply similarity thresholds or matching rules between competency identifiers according to the mapping policy. The output layer (130) may support various mapping structures such as 1:1, 1:N, and N:1 according to the mapping policy. When the mapping policy is changed, the output layer (130) may recalculate the existing mapping results.

[0328] In one embodiment, at step 420, the output layer (130) identifies at least one of a proficiency level or evidence strength for a corresponding competency identifier and may organize the identified information into a structured evidence package. The output layer (130) may define an evidence package structure for each corresponding competency identifier that includes detailed fields such as proficiency level, evidence strength, match status, and source information. The output layer (130) may generate the evidence package structure in various data formats such as JSON, XML, and tables. The output layer (130) may standardize the notation method of each field within the evidence package (e.g., proficiency level, evidence strength grade, match / mismatch indication, etc.). The output layer (130) may consistently output recommended evidence information for each candidate according to the evidence package structure. The output layer (130) may convert the evidence package structure into a form suitable for external system integration or dashboard visualization.

[0329] In one embodiment, at step 421, the output layer (130) can perform competency mapping between SSR data and job posting data. The output layer (130) can derive mapping candidate pairs by comparing each competency identifier of the SSR data with each required competency of the job posting data. The output layer (130) can determine the final mapping result by applying synonyms, similar expressions, and standard classification system mapping rules between competency identifiers. The output layer (130) can store the mapping result in a separate mapping table or data structure. Based on the mapping result, the output layer (130) can check the status of required competency fulfillment for each candidate.

[0330] In at least one embodiment of the present disclosure, the output layer (130) may correspond the competency identifiers included in the candidate group's SSR data with the required competencies included in the job posting data. The output layer (130) may compare the competency identifiers of the SSR data with the required competencies of the job posting data in a 1:1 or 1:N manner. The output layer (130) may apply identifier matching of a standard classification system, synonym mapping, similarity thresholds, etc., as a correspondence criterion. The output layer (130) may store the corresponding competency identifier pairs in a mapping table. The output layer (130) may check whether the required competencies are met for each candidate based on the mapping table.

[0331] In one embodiment, the output layer (130) may indicate whether there is a match or a mismatch for the corresponding competency identifier. The output layer (130) may check the mapping results between each competency identifier in the SSR data and each required competency in the job posting data. The output layer (130) may indicate 'Match' if the mapping results match, and 'Mismatch' if they do not match. The output layer (130) may include the match / mismatch indication information in the recommendation basis information. The output layer (130) may indicate the match / mismatch indication in various ways, such as visual icons, text, or colors.

[0332] In at least one embodiment of the present disclosure, at step 422, the output layer (130) may configure the recommended evidence information in the form of an evidence package. The output layer (130) may include detailed fields such as a competency identifier, proficiency level, evidence strength, match status, and source information corresponding to the evidence package. The output layer (130) may standardize the evidence package structure to generate a consistent data format. The output layer (130) may clearly define the notation method and data type of each field within the evidence package. The output layer (130) may combine the recommended evidence information configured in the form of an evidence package with the final result data.

[0333] In at least one embodiment of the present disclosure, the output layer (130) may indicate a proficiency level for a corresponding competency identifier. The output layer (130) may check the proficiency level field value of the SSR data. The output layer (130) may extract a proficiency level for each corresponding competency identifier. The output layer (130) may indicate the extracted proficiency level in the corresponding competency identifier field within the basis package. The output layer (130) may indicate the proficiency level in various ways, such as grade, score, or level. The output layer (130) may provide consistent results by standardizing the method of indicating the proficiency level.

[0334] In at least one embodiment of the present disclosure, the output layer (130) may indicate the strength of evidence for a corresponding competency identifier. The output layer (130) may check the field value of the strength of evidence field in the SSR data. The output layer (130) may extract the strength of evidence for each corresponding competency identifier. The output layer (130) may indicate the extracted strength of evidence in the corresponding competency identifier field within the evidence package. The output layer (130) may indicate the strength of evidence in various ways, such as grade, score, or reliability. The output layer (130) may standardize the method of indicating the strength of evidence to provide consistent results.

[0335] In at least one embodiment of the present disclosure, at step 430, the output layer (130) may generate a final ranking and descriptive information thereto. The output layer (130) may receive the ranking result calculated by the matching layer (120) as an input value. The output layer (130) may refer to SSR data and job posting data for each candidate included in the input ranking result. The output layer (130) may generate descriptive information by comparing the competency identifier included in the SSR data with the required competency included in the job posting data. The output layer (130) may combine the generated descriptive information with the final ranking data to form recommendation basis information for each candidate. The output layer (130) may include detailed information such as proficiency level, strength of the basis, and whether it matches in the recommendation basis information. The output layer (130) may output the final result including the recommendation basis information through an enterprise dashboard, an external system API, or a user interface.

[0336] In one embodiment, the output layer (130) receives the ranking results generated by the matching layer (120) as input, performs fairness correction and description generation processing, and finally outputs the candidate recommendation results. After receiving the ranking results, the output layer (130) can classify candidates according to predefined group criteria. The output layer (130) can calculate the recommendation rate or exposure rate for each group. The output layer (130) can adjust the score or ranking so that the deviation of the recommendation rate or exposure rate between groups is below the fairness threshold. The output layer (130) can generate description information by mapping SSR data and job posting data to the final ranking to which fairness correction has been applied. The output layer (130) can output the candidate recommendation results by combining the generated description information and the corrected final ranking.

[0337] In one embodiment, the output layer (130) can match competency identifiers included in the candidate group's SSR data with required competencies included in the job posting data. The output layer (130) can compare each competency identifier in the SSR data with each required competency in the job posting data in a 1:1, 1:N, or N:1 mapping manner. The output layer (130) can determine the correspondence relationship by applying synonyms, similar expressions, and standard classification system mapping rules between competency identifiers. The output layer (130) can store the matched competency identifier pairs in a separate mapping table. Based on the mapping table, the output layer (130) can check the status of required competency fulfillment for each candidate.

[0338] In one embodiment, the output layer (130) can identify at least one of a proficiency level or a strength of evidence for a corresponding competency identifier. The output layer (130) can check the proficiency level field value included in the SSR data. The output layer (130) can check the strength of evidence field value included in the SSR data. The output layer (130) can extract at least one of a proficiency level or a strength of evidence for each corresponding competency identifier. The output layer (130) can include the extracted proficiency level and strength of evidence information in the recommended evidence information for each candidate.

[0339] In one embodiment, the output layer (130) can generate recommendation basis information by matching the identified information to each candidate included in the final ranking. The output layer (130) can refer to the mapping results of SSR data and job posting data for each candidate in the final ranking list. The output layer (130) can generate recommendation basis information by combining detailed information such as the competency identifier, proficiency level, strength of evidence, and match status corresponding to each candidate. The output layer (130) can combine the generated recommendation basis information with the final ranking data and output it as a structured result package.

[0340] In at least one embodiment of the present disclosure, the output layer (130) can generate a candidate recommendation result report by combining the final ranking and recommendation basis information. The output layer (130) can combine the final ranking with fairness correction applied and the recommendation basis information into a single result structure. The output layer (130) may include candidate-specific scores, rankings, recommendation basis packages, and group deviation summary information in the combined result structure. The output layer (130) can convert the combined result structure into a report format.

[0341] In at least one embodiment of the present disclosure, the output layer (130) may provide the candidate recommendation result report through an enterprise dashboard, an external system, or an application programming interface (API). The output layer (130) may display the generated candidate recommendation result report on an enterprise or institutional dashboard. The output layer (130) may link the generated candidate recommendation result report through an external system API. The output layer (130) may provide the generated candidate recommendation result report to a user interface or a feedback system.

[0342] In at least one embodiment of the present disclosure, the output layer (130) may combine the final ranking with fairness correction applied and the generated recommendation basis information. The output layer (130) may combine the final ranking list with fairness correction applied and the recommendation basis information for each candidate into a single data structure. The output layer (130) may include candidate scores, rankings, recommendation basis packages, and group deviation summary information in the combined data structure. The output layer (130) may convert the combined data structure into a report format.

[0343] In at least one embodiment of the present disclosure, the output layer (130) can generate a consistent result structure by combining each candidate's score, rank, and corresponding grounds information with the corrected Top-K candidate list. When generating the Top-K candidate list, the output layer (130) may refer to the mapping results of SSR data and job posting data for each candidate. The output layer (130) can generate a consistent result structure including each candidate's score, rank, recommendation grounds package, and group deviation summary information. The output layer (130) can provide the generated result structure through an enterprise dashboard, an external system API, or a user interface.

[0344] In at least one embodiment of the present disclosure, the output layer (130) may generate candidate recommendation results including a recommendation list, scores and rankings for each candidate, and a recommendation basis package. The output layer (130) may refer to the mapping results between SSR data and job posting data for each candidate included in the recommendation list. The output layer (130) may generate a result structure including scores, rankings, and a recommendation basis package (competency identifier, proficiency level, strength of basis, match, etc.) for each candidate. The output layer (130) may convert the generated result structure into a report format and provide it to an enterprise dashboard, an external system API, or a user interface.

[0345] In at least one embodiment of the present disclosure, the output layer (130) may generate candidate recommendation results including group deviation summary information. The output layer (130) may calculate group recommendation rates or exposure rates based on group classification criteria (gender, age, region, etc.) for each candidate. The output layer (130) may calculate the deviation of the group recommendation rates or exposure rates and generate summary information. The output layer (130) may include group deviation summary information in the candidate recommendation results. The output layer (130) may provide results including group deviation summary information to an enterprise dashboard, an external system API, or a user interface.

[0346] In at least one embodiment of the present disclosure, at step 440, the output layer (130) may provide candidate recommendation results through a recommendation API, a dashboard for an enterprise or institution, or a candidate feedback interface. The output layer (130) may link candidate recommendation results in real time to an external system through a recommendation API. The output layer (130) may provide candidate recommendation results visualized on a dashboard for an enterprise or institution. The output layer (130) may provide recommendation results and supporting information to individual candidates through a candidate feedback interface.

[0347] In at least one embodiment of the present disclosure, the output layer (130) may provide candidate recommendation results through a dashboard for a company or institution. The output layer (130) may visually display the final ranking, recommendation basis information, group deviation summary information, etc., on the dashboard for a company or institution. The output layer (130) may provide functions for viewing detailed information by candidate, filtering, and comparison on the dashboard. The output layer (130) may provide functions for downloading recommendation results from the dashboard or linking to an external system.

[0348] In at least one embodiment of the present disclosure, the output layer (130) can link candidate recommendation results through an external system API. The output layer (130) can transmit recommendation result data (final ranking, recommendation basis, summary of deviation by group, etc.) in a standardized data format through an external system API. When linking with an external system API, the output layer (130) can apply authentication and access control policies. The output layer (130) can provide recommendation results in real-time or in batch mode through an external system API.

[0350] FIG. 5 shows an example of an electronic device according to one embodiment.

[0351] Referring to FIG. 5, the electronic device (100) may include memory (510) and a processor (530).

[0352] The memory (510) can store instructions (e.g., programs) executable by the processor (530). For example, the instructions may include instructions for executing the operation of the processor (530) and / or the operation of each component of the processor (530).

[0353] The memory (510) can be implemented as a volatile memory device or a non-volatile memory device.

[0354] Volatile memory devices can be implemented as DRAM (dynamic random access memory), SRAM (static random access memory), T-RAM (thyristor RAM), Z-RAM (zero capacitor RAM), or TTRAM (Twin Transistor RAM).

[0355] Non-volatile memory devices can be implemented as EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, MRAM (Magnetic RAM), Spin-Transfer Torque (STT)-MRAM, Conductive Bridging RAM (CBRAM), FeRAM (Ferroelectric RAM), PRAM (Phase change RAM), Resistive RAM (RRAM), Nanotube RRAM, Polymer RAM (PoRAM), Nano Floating Gate Memory (NFGM), holographic memory, Molecular Electronic Memory Device, or Insulator Resistance Change Memory.

[0356] The processor (530) can process data stored in memory (510). The processor (530) can execute computer-readable code (e.g., software) stored in memory (510) and instructions triggered by the processor (530).

[0357] The processor (530) may be a data processing device implemented in hardware having a circuit having a physical structure for executing desired operations. For example, the desired operations may include code or instructions included in a program.

[0358] For example, a data processing device implemented in hardware may include a microprocessor, a central processing unit, a processor core, a multi-core processor, a multiprocessor, an Application-Specific Integrated Circuit (ASIC), and a Field Programmable Gate Array (FPGA).

[0359] The processor (530) can cause the electronic device (100) to perform one or more operations by executing code and / or instructions stored in memory (510). The operations performed by the electronic device (100) have been described in detail with reference to FIGS. 1 to 4b, so redundant descriptions will be omitted below.

[0361] The embodiments described above may be implemented as hardware components, software components, and / or combinations of hardware and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. In addition, other processing configurations, such as parallel processors, are also possible.

[0362] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or instruct the processing unit independently or collectively. Software and / or data may be stored on any type of machine, component, physical device, virtual equipment, computer storage medium, or device so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on computer-readable recording media.

[0363] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may store program instructions, data files, data structures, etc., either individually or in combination, and the program instructions recorded on the medium may be those specifically designed and configured for the embodiment or those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc.

[0364] The hardware device described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

[0365] Although the embodiments have been described above with reference to the limited drawings, those skilled in the art can apply various technical modifications and variations based thereon. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or if the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0366] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

Claim 1 A method for matching a user with a job using credential data, performed by an electronic device comprising one or more processors and memory, comprising: a step of acquiring Verifiable Credential (VC) data including issuing authority and achievement information for one or more candidates—wherein the VC data corresponds to each of the one or more candidates; a step of generating Standardized Skill Record (SSR) data regarding the candidate, including a competency identifier, proficiency level, and evidence strength according to a pre-established standard competency classification system based on the VC data; a step of deriving a candidate group semantically related to the job information among the one or more candidates based on the SSR data and job posting data; a step of calculating the suitability of each candidate based on the SSR data of the candidate group and the job posting data; a step of determining the candidate ranking by sorting the candidates within the candidate group according to the suitability of each candidate; a step of generating a final ranking by adjusting the candidate ranking based on the candidate ranking to satisfy a pre-established fairness standard; and a step of generating recommendation basis information for the final ranking by correlating the SSR data of the candidate group with the job posting data. A method comprising the step of generating a candidate recommendation result report by combining the above-mentioned final ranking and the above-mentioned recommendation basis information. Claim 2 The method according to claim 1, wherein the step of generating the SSR data comprises: a step of extracting competency candidates from natural language-based text data included in the VC data; a step of normalizing the competency candidates and mapping them to the competency identifiers of the standard competency classification system; and a step of calculating the proficiency level and evidence strength corresponding to the competency identifiers. Claim 3 A method according to claim 1, wherein the step of deriving the candidate group comprises: converting the SSR data and the recruitment announcement data into vectors of a shared semantic space; calculating similarity between the vectors; and selecting some of the one or more candidates as the candidate group based on the similarity. Claim 4 The method according to claim 1, wherein the step of calculating the suitability for each candidate comprises: a step of calculating a first suitability value for each candidate by an interaction-based model using the SSR data of the candidate group and the job posting data together as inputs; a step of adjusting the first suitability value based on the first suitability value by reflecting whether the essential competencies included in the job posting data are satisfied; and a step of calculating a final suitability based on the adjusted first suitability value by reflecting the proficiency level included in the SSR data. Claim 5 A method according to claim 1, wherein the step of adjusting the candidate ranking to generate the final ranking comprises: classifying the candidates included in the candidate ranking into a plurality of groups according to a predefined group criterion; calculating a recommendation rate or exposure rate for each of the groups; and adjusting the score of the candidate ranking to generate the final ranking so that the deviation of the recommendation rate or exposure rate between the groups satisfies a pre-set fairness criterion. Claim 6 A method according to claim 1, wherein the step of generating the recommendation basis information comprises: a step of matching a competency identifier included in the SSR data of the candidate group with a required competency included in the job posting data; a step of identifying at least one of a proficiency level or a strength of basis for the matched competency identifier; and a step of matching the identified information to each candidate included in the final ranking to generate the recommendation basis information.

Citation Information

Patent Citations

  • Job recommendation system using my data and job recommendation method using it

    KR1020230127044A

  • Method and apparatus for providing job competence information based on job posting

    KR1020230150226A