System and Method for Generating Customized Cover Letter using Corporate Analysis Report and Experience Data Embedding
The career data infrastructure system addresses the inefficiencies in structuring unstructured user experiences and recruitment data, providing personalized career analysis and recruitment matching by mapping to a domain-specific ontology, enhancing job preparation and matching efficiency.
Patent Information
- Application Number
- KR1020260154276
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2026-01-27
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-11
- Estimated Expiration
- 2046-08-18
AI Technical Summary
Conventional job support systems fail to structure users' unstructured experience data effectively, leading to incomplete demonstration of capabilities, and lack of mid-to-long-term career path recommendations due to inadequate mapping of industry domains and company requirements, resulting in inefficient job preparation and mismatched recruitment outcomes.
A career data infrastructure system that structures unstructured career and recruitment data using an AI-based framework, identifies missing data through the STARR framework, and maps it to a domain-specific ontology for personalized career analysis and recruitment matching, generating customized self-introduction letters through prompt engineering.
Minimizes user cognitive load by structuring unstructured experiences into sophisticated datasets, analyzes competency gaps, and visualizes career paths, enabling efficient job preparation and precise matching of user competencies with company requirements.
Smart Images

Figure 112026100780424-PAT00013_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a career data infrastructure system that collects unstructured career data from job seekers and unstructured recruitment data from companies, respectively, through a user experience-based input structure or an interview-based collection structure, extracts and structures quantitative and qualitative data fields while preserving original utterances or original input data, and maps the data fields and the relationships between them to utilize for document generation, career analysis, candidate evaluation, recruitment matching, job suitability analysis, and career path recommendation.
[0002] The present invention relates to the fields of AI-based customized career roadmap design, conversational experience data structuring utilizing natural language processing (NLP), and an intelligent document automatic generation system linked with a domain ontology knowledge graph.
[0003] More specifically, it identifies missing data based on the Context, Task, Behavior, Result, Reflection (STARR) framework in users' unstructured experience texts and presents customized follow-up questions to structure job experiences into a precise dataset, and maps this to nodes on a static ontology where job, task, and knowledge elements of a specific industry domain are defined hierarchically and associatively.
[0004] This invention relates to an intelligent career agent system that visually recommends a step-by-step career transition path including logical connectivity with target job postings and intermediate roles, and performs prompt engineering by combining optimally selected core experience data based on embedding similarity with company-specific reports extracted by crawling public disclosure data and job postings of target companies, thereby automatically generating a customized self-introduction letter with a precise match between the user's capabilities and the company's requirements, and supporting editing through interactive communication. Background Technology
[0005] The recent recruitment market is rapidly shifting toward rolling recruitment that prioritizes job suitability and practical skills, and the importance of systematic career management and customized job preparation to align with this trend is increasing.
[0006] Conventional job support systems had the inconvenience of requiring users to structure their own experiences. Users often listed their experiences as unstructured text, omitting essential elements such as situations, tasks, actions, and results, which limited their ability to properly demonstrate their capabilities.
[0007] Existing recruitment platforms caused a problem by mapping users' resumes and job postings based primarily on keywords, failing to grasp the complex hierarchical structure of industry domains or technical relationships, and recommending only postings that matched immediate job titles, thereby failing to present mid-to-long-term career paths.
[0008] This method had limitations in that the process of users individually analyzing a company's latest disclosures or job requirements to write a customized self-introduction letter by connecting them to the most suitable material from their numerous experiences consumed time and effort, thereby lowering the efficiency of job preparation.
[0009] Furthermore, if all user experience data was indiscriminately reflected to improve the quality of the writing, strategic weighting based on the company's actual business goals or the intent of each self-introduction question was not achieved, resulting in awkward context and making it difficult to consistently provide high-quality final results that precisely matched the user's capabilities with the company's requirements. The problem to be solved
[0010] The present invention has the primary objective of establishing a career data infrastructure that supports both the generation of personalized AI results and explainable B2B recruitment matching by subdividing, collecting, and structuring career data on the job seeker side and recruitment data on the company side, respectively, and connecting the data from both sides on an ontology or matching model. The present invention has the primary objective of intelligently identifying missing data on the Context, Task, Behavior, Result, and Reflection (STARR) framework from fragmentary and unstructured experience texts of users, and structuring said experience into a sophisticated and objectified data set through dynamic priority-based customized follow-up questions that minimize the user's cognitive load.
[0011] In addition, the present invention has another objective of analyzing the gap between a user's current competency level and a target job by mapping the structured experience data on a semantic basis to a static ontology knowledge graph that reflects job, task, and knowledge elements of a specific industry domain, rather than simple keyword matching, and visualizing and providing the intermediate job-based step-by-step career transition path required to overcome this gap and the basis for its recommendation.
[0012] Furthermore, another objective of the present invention is to generate a customized report for a company that encompasses macroscopic business objectives and microscopic required competencies by crawling public disclosure data, social issues, and job descriptions of the target company selected by the user from various angles, and to select and combine the optimal core experience data with the highest similarity using vector embedding techniques, thereby finally producing a differentiated, customized self-introduction letter in which the user's core competencies match the requirements of the target company according to the detailed types of application questions. means of solving the problem
[0013] An experience organization system for customized job search using questions according to an embodiment of the present invention includes a memory for storing instructions and a processor. When the instructions are executed by the processor, the system receives career-related initial input data from a user terminal, classifies the initial input data into one of a plurality of scenarios including a performance-oriented type and a process-oriented type, loads a reference data structure including at least one of predefined experience information, competency information, goal information, preference information, compensation information, organizational culture information, and matching judgment information according to the scenario, identifies missing data among the initial input data that is not mapped to the reference data structure, generates customized follow-up questions by applying a prompt engineering algorithm that reflects the content of the scenario and the initial input data, provides the follow-up questions to the user terminal, receives answer data input from the user terminal regarding the follow-up questions and updates the missing data, and if the missing data exists, repeats the generation and provision of the follow-up questions, and if the reference data structure is satisfied, converts the updated reference data into structured career data that can be used for subsequent AI result generation, career analysis, recruitment matching, or recommendation models and stores it.
[0014] The above experience organization system is not limited to organizing experience materials for writing a user's self-introduction letter, but also derives career-related information existing in the user's tacit knowledge through a user experience-based input structure. During the derivation process, at least one of multiple experiences, episodes, facts of work performance, achievements, competencies, reasons for judgment, work methods, collaboration methods, preferred conditions, and career goals is extracted and stored as structured career data that can be utilized for generating subsequent AI results, career analysis, recruitment matching, and recommendation models.
[0015] The above experience information field may include at least one of a situation, task, action, result, and reflection, and in one embodiment may be configured based on the STARR framework, but the scope of the invention is not limited thereto.
[0016] The above experience information further includes task data, and the above instructions can control the generation of a question asking about the key performance indicators of the task or problems faced during the execution of the task as the above follow-up question if the above missing data is the above task data.
[0017] The above experience information further includes behavioral data, and the above instructions can control the generation of a question asking the user about the tool used or the strategy applied as the above follow-up question if the above missing data is the above behavioral data.
[0018] The above instructions can control the generation of the problem-solving question as the follow-up question when the initial input data is classified as process-oriented and problem-solving content is absent among the behavior data.
[0019] The above experience information further includes behavioral data and result data, and the above instructions can control the generation of a question asking for quantitative data representing the change before and after the user's action performance as the above follow-up question when the above missing data is the above result data.
[0020] The above instructions can control the generation of the following question asking for the numerical indicator when the initial input data is classified as performance-oriented and when the result data lacks a numerical indicator.
[0021] The above scenario further includes a composite type that combines the performance-oriented type and the process-oriented type, and the instructions can control the generation of a connectivity-enhancing question asking about the causal relationship as a tail question when the initial input data is classified as a composite type and there is no causal relationship in which the behavior data leads to the result data.
[0022] The above experience information further includes reflection data, and the above instructions can control the generation of a question asking for lessons or improvements learned through the result data as a follow-up question if the above missing data is the reflection data.
[0023] The above instructions can control the processor to determine that it is in a state of insufficient information when it is unable to grasp the context of the initial input data, and to output a guide message requesting the user terminal to re-enter the experience.
[0024] The above instructions identify the user's input purpose before receiving the above career-related initial input data and load the above standard data structure according to the input purpose, wherein the input purpose may include at least one of writing a self-introduction letter, preparing for an interview, writing a resume, writing a career description, job search, preparing for a job change, career counseling, job transition, salary negotiation, performance evaluation, performance review, recruitment matching, and career path recommendation.
[0025] The above missing data includes at least one of a missing state in which information corresponding to a specific field of the reference data structure is absent, a deficient state in which information exists in the field but the reliability is below a preset threshold, an uncertain state in which the possibility of interpreting the extracted information is multiple or the basis for inference is weak, and a conflict state in which information extracted from multiple utterances of the user is mutually contradictory, and the instructions can be controlled to generate a customized tail question corresponding to the identified data state.
[0026] The above instructions can control the provision of the following questions sequentially according to the calculated priority by considering at least one of the following factors when there are multiple identified missing data or multiple data fields requiring supplementation, the narrative development order of the reference data structure, whether it is a core field according to the scenario type, the degree of relevance to subsequent functions, and the user's fatigue or the possibility of disengaging from the conversation.
[0027] The above instructions can store the above answer data as original conversation data and store the career data field extracted from the original conversation data by connecting it to the corresponding original utterance section.
[0028] The above instructions may assign at least one reliability type among explicit utterance-based, inference-based, uncertainty, and conflict to the career data field.
[0029] The above instructions can identify multiple experience candidates or multiple episodes from one of the above initial input data or conversation sessions, extract career data fields for each of the multiple experience candidates or episodes, and extract common competencies, recurring patterns, strengths, or career directions among multiple experiences of the same user and store them as a structured career data set for each user.
[0030] The above instructions can map relationships between multiple extracted career data fields, and the relationships may include at least one of the following: a relationship between experience and competency, a relationship between behavior and performance, a relationship between tools used and job competency, a relationship between problem-solving methods and work styles, a relationship between preferred conditions and organizational culture fit, or a basis relationship between experience data and subsequent generated results.An ontology-based career path recommendation system according to an embodiment of the present invention comprises a memory for storing instructions and a processor, wherein, when the instructions are executed by the processor, the system constructs and stores a static ontology in which standard job duties, tasks, industries, or required knowledge within a specific industry domain are each defined as ontology nodes, and associations, hierarchies, utilizations, and presuppositional relationships between the ontology nodes are defined as ontology edges, and receives user-side structured career data and corporate-side unstructured data, wherein the corporate-side unstructured data includes at least one of job posting data and corporate insider interview-based unstructured conversation data, stores the corporate insider interview-based unstructured conversation data as original utterance data, extracts corporate-side data fields from the original utterance data, extracts user-side data fields from the user-side structured career data, normalizes the user-side data fields and the corporate-side data fields, respectively, into standard terms defined in the static ontology to create user nodes and job posting nodes, maps them to the ontology nodes, and the static On the ontology, the ontology edge connected from the user node to the job posting node is searched to calculate a career path that can be moved between jobs, and a path suitability score is calculated based on the connection strength or distance of the ontology edge. Along with the career path and the path suitability score, at least one of job suitability, job transitionability, lack of competency, organizational culture suitability, compensation condition suitability, and items requiring verification or grounds for recommendation is calculated, and graph data in which the user node and the job posting node are commonly connected can be provided to a user terminal along with the career path and the path suitability score.
[0031] The above instructions identify a basis node shared by the user node and the announcement node on the static ontology when calculating the path suitability score, and the basis node includes at least one of the task, industry, or necessary knowledge that the user node and the announcement node have in common, and can provide the reason why the career path was proposed by visualizing it through the path suitability score and the basis node.
[0032] The above instructions search for one or more intermediate nodes connecting the user node and the announcement node when the user node and the announcement node are not directly connected, and the career path can sequentially connect the user node, the intermediate node, and the announcement node.
[0033] The above instructions, when calculating the path fit score, sum the weights of multiple edges existing on the career path, and may assign a higher weight if the edge is an essential prerequisite.
[0034] The above instructions can map the announcement node and the user node by analyzing the edge based on a domain context including a higher-level concept by analyzing the hierarchical structure defined in the static ontology.
[0035] The above instructions can be controlled to identify a deficiency in capabilities between the user node and the announcement node, identify conflicting nodes between the user node and the announcement node, and calculate and provide a learning path or intermediate job path to supplement the deficiency in capabilities based on node and edge relationships on the static ontology.
[0036] The above instructions can control the extraction of organizational culture preference information and compensation condition information from the user-side data fields, the extraction of organizational culture characteristic information and compensation condition information from the company-side data fields, mapping them to corresponding nodes on the static ontology, and calculating organizational culture fit and compensation condition fit to reflect in the matching result.
[0037] The above instructions can control the generation and provision of a candidate report including at least one of the career path, the path fit score, the job fit, the lack of competency, the organizational culture fit, the compensation condition fit, and the items requiring verification.
[0038] The above instructions can be controlled to identify items requiring verification during the interview process based on the above company-side data fields and the above user-side data fields, and to generate and provide interview verification questions by referencing together the fields extracted from public recruitment announcement data and the fields extracted from the above company insider interviews.
[0039] The above instructions store the corporate data fields extracted from the corporate insider interview-based unstructured conversation data by connecting them to corresponding original utterance segments, assign different weights to the corporate data fields according to the respondent role type of the corporate insider interview, wherein the role type includes at least one of an HR manager, an incumbent, a team leader, and a hiring decision-maker, and control can classify mutually inconsistent items as items requiring verification by comparing fields extracted from public recruitment announcement data with fields extracted from the corporate insider interview.
[0040] The above static ontology is managed through a version control system, and the instructions can store version information of the static ontology and control the application of an updated version of the static ontology to the mapping of the user node and the announcement node.
[0041] A customized self-introduction letter generation system according to an embodiment of the present invention includes a memory for storing instructions and a processor, wherein, when the instructions are executed by the processor, the system receives structured career data of a user and collects external documents related to said applicant company using an applicant company input from a user terminal as a key value, said external documents include at least one of recruitment announcement data, job descriptions, and unstructured conversational data based on interviews with corporate insiders, text-mines corporate analysis information and job requirement information from said external documents to generate and store a corporate customized report, embeds first experience data classified into the user's situation, task, behavior, and result stored in a database, calculates a similarity score with said job requirement information, extracts and stores said first experience data as second experience data if the similarity score is greater than or equal to a preset threshold, classifies the self-introduction letter questions of said applicant company into one of a plurality of predefined question types, analyzes paragraphs among successful self-introduction letters stored in the database that match said question type to learn a standardized writing logic, and uses said writing logic as a framework. By inserting the aforementioned second experience data and performing prompt engineering that combines it with the aforementioned company-customized report as the basis for creation, control can be exercised to generate a user-customized self-introduction draft.
[0042] When classifying the above-mentioned question types, the above-mentioned instructions can analyze the intent of the self-introduction questions of the applicant company and match them to one of the pre-established categories including at least one of the motivation for application type, job competency type, and post-employment aspiration type.
[0043] The above instructions, when learning the above standardized writing logic, analyze the sentences of the above successful self-introduction letter based on situation, task, action, result, and reflection, and can apply this as a template when generating the above user-customized self-introduction letter draft.
[0044] The above instructions can control the dynamic assignment of weights representing the character count or emphasis weight for each item of the situation, task, action, result, and reflection according to the above question type, and the generation of the user-customized self-introduction draft based on the assigned weights.
[0045] The above instructions can control the generation of the user-customized self-introduction draft by applying different generation logic or weights according to the type of the output to be generated, the company information, the job requirements information, the question type, the user's personal characteristics, and the field configuration of the structured career data.
[0046] The above instructions can control the analysis or evaluation of at least one of item suitability, job suitability, corporate suitability, experience specificity, performance clarity, logical consistency, factual consistency with original experience, stylistic naturalness, and interview risk for the generated user-customized self-introduction draft, and detect and correct expressions judged to be AI-generated style.
[0047] The above instructions can interactively receive a user's request for modification, correction, analysis, or evaluation of the user-customized personal statement draft, convert the request into at least one of stylistic conditions, emphasis conditions, character count conditions, risk avoidance conditions, or evaluation criteria conditions, rewrite the relevant paragraph or the entire draft based on the converted conditions, and control saving versions before and after modification.
[0048] The above instructions can control the updating of the user's structured career data by reflecting additional information collected from the user's response to the request for modification, correction, analysis, or evaluation.
[0049] The above instructions group the user-customized self-introduction draft, the company-customized report, and the second experience data by the support company, and upon a user's inquiry request, provide the user-customized self-introduction draft, the company-customized report, and the second experience data in a time series.
[0050] The above instructions can generate and present a question requesting additional information from the user, and upon receiving the user's response to the question, can rewrite the paragraph by transforming the response to fit the context.
[0051] The above instructions can crawl the company's semi-annual or annual reports and job descriptions from recruitment sites when collecting the above external documents, and generate the above company-customized report.
[0052] The above instructions can be stored as a structured data set by mapping the business content of the above semi-annual report with the qualification requirements or preferred items of the above job description.
[0053] In addition to generating and correcting the user-customized self-introduction draft, the above instructions can control the generation or correction of at least one of an interview answer, resume, career description, application input, performance evaluation, and performance review based on the structured career data and the company-customized report.
[0054] The above instructions allow for the generation of the user-customized self-introduction draft, the automatic generation of a subheading at the top summarizing the key contents of the second experience data, and the adjustment of the length to meet the character limit set for the self-introduction questions of the applicant company.
[0055] The above instructions can humanize the draft of the self-introduction into natural sentences by applying conjunctions and endings frequently used in the above-mentioned successful self-introduction. Effects of the invention
[0056] The present invention can automatically structure unstructured experiences into sophisticated STARR datasets while minimizing the user's cognitive load through an interactive agent that identifies missing data in the user's experience text and provides customized follow-up questions.
[0057] In addition, the present invention can analyze the competency gap with the target job by mapping structured experience data to a domain-specific static ontology, and can visualize and recommend a step-by-step career transition path including intermediate jobs.
[0058] This enables users to move beyond a short-term perspective and secure a foundation to proactively design a realistic and systematic mid-to-long-term career roadmap based on their current competency levels.
[0059] Furthermore, the present invention can automatically generate and edit customized self-introduction letters that match the company's requirements and the user's core competencies through the generation of customized reports based on corporate disclosure data, the selection of optimal experience data through vector embedding, and prompt engineering reflecting weights for each type of support question.
[0060] Furthermore, the present invention extracts unstructured career information existing in a user's tacit knowledge into specific data fields through a user experience-based input structure, and can construct user-specific structured career data including multiple experiences, episodes, competencies, achievements, preferences, and goals; enhances the explainability and verifiability of the data by storing the original utterance and the extracted data fields in a linked manner; and enables repeated utilization for generating subsequent AI results and B2B recruitment matching by mapping the relationships between the extracted data fields.
[0061] Furthermore, the present invention structures unstructured corporate data into nodes and edges on an ontology map to construct data sets by company, job, team, and required competency, and by connecting the user-side data map and the corporate-side data map, it is possible to calculate job fit, organizational culture fit, compensation condition fit, lack of competency, and items requiring verification that are more precise and explainable than simple keyword matching. Brief explanation of the drawing
[0062] FIGS. 1 and 2 are flowcharts of an experience organization system for customized job search utilizing questions according to an embodiment of the present invention. Figure 3 is a flowchart for generating follow-up questions of a performance-oriented scenario in an experience organization system for customized job search using questions according to an embodiment of the present invention. Figure 4 is a flowchart for generating tail questions of a complex scenario in an experience organization system for customized job search using questions according to an embodiment of the present invention. FIG. 5 is a flowchart for generating tail questions of a process-oriented scenario in an experience organization system for customized job search using questions according to an embodiment of the present invention. FIG. 6 is a flowchart of the process of building structured career data in an experience organization system for customized job search using questions according to an embodiment of the present invention. FIGS. 6, FIGS. 7, and FIGS. 8 are flowcharts of an ontology-based career path recommendation system according to an embodiment of the present invention. FIG. 9 is a flowchart of an ontology-based career path recommendation system according to one embodiment of the present invention. FIG. 10 is a diagram showing an example of a graph data provision step of an ontology-based career path recommendation system according to one embodiment of the present invention. FIG. 11 is a flowchart for generating and saving a company-customized report in a customized self-introduction letter generation system according to one embodiment of the present invention. FIG. 12 is a flowchart for extracting and storing second experience data of a customized self-introduction letter generation system according to one embodiment of the present invention. FIG. 13 is a flowchart for completing a draft of a self-introduction letter in a customized self-introduction letter generation system according to one embodiment of the present invention. FIG. 14 is a diagram showing the self-introduction questions of a supporting company in a customized self-introduction generation system according to one embodiment of the present invention. FIG. 15 is a flowchart of the editing of a draft self-introduction letter in a customized self-introduction letter generation system according to one embodiment of the present invention. Specific details for implementing the invention
[0063] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the attached drawings. Prior to this, terms and words used in this specification and claims should not be interpreted as being limited to their ordinary or dictionary meanings, and should be interpreted in a meaning and concept consistent with the technical spirit of the present invention, based on the principle that the inventor can appropriately define the concept of the terms to best describe his invention.
[0064] Therefore, the embodiments described in this specification and the configurations illustrated in the drawings are merely the most preferred embodiments of the present invention and do not represent all of the technical ideas of the present invention; thus, it should be understood that various equivalents and modifications that can replace them may exist at the time of filing this application.
[0065] FIGS. 1 and 2 are flowcharts of an experience organization system for customized job search utilizing questions according to an embodiment of the present invention.
[0066] Referring to FIGS. 1 and 2, an experience organization system for customized job exploration using questions may include a memory and a processor for storing instructions.
[0067] When the instructions are executed by the processor, the system can receive initial experience data from the user terminal.
[0068] The system may first output situational questions (S10) specialized for job applications so that the user can easily recall experiences without feeling uncertain. These questions may be in the form of inquiries regarding core competencies frequently asked during the hiring process. For example, questions such as "Do you have any project experience you were most passionate about?", "Recall a situation where you collaborated to achieve a common goal," or "How did you cope when faced with an unexpected problem?" may be presented at the top of the interface.
[0069] The user may input initial experience data as an answer to these questions. The system may receive initial experience data from the user terminal (S20). Here, initial experience data may be unstructured text data written by the user as they come to mind regarding their job experience. For example, it may be fragmentary sentences or a list of keywords such as "I increased sales during my last internship" or "I resolved discord among team members." The system may collect this raw data through a text input interface or a voice recognition interface to utilize it as basic data for analysis.
[0070] Instructions can request the user terminal to re-enter the experience if the processor is unable to grasp the context of the initial experience data. That is, the instructions can control the processor to determine that the inability to grasp the context of the initial experience data constitutes a state of insufficient information and to output a guide message requesting the user terminal to re-enter the experience.
[0071] The system can perform validation on initial experience data received from a user terminal. If the text entered by the user is excessively short, consists of a sequence of short-answer keywords lacking sentence components, or is a combination of meaningless strings, the system may determine through a natural language processing engine that it is impossible to understand the context of the data.
[0072] For example, if the input data consists only of abstract words without descriptions of specific situations or actions, such as "working hard" or "sales," the system can detect that the confidence score of the scenario classification model is below a preset threshold and determine this as an insufficient information state.
[0073] When a state of insufficient information is detected, the system can be controlled to halt subsequent processes, such as scenario classification and follow-up question generation, and to output a guide message requesting the user to re-enter the experience. In this case, the guide message may be in the form of a template that encourages the user to write specific details, rather than a simple warning window requesting re-entry.
[0074] For example, a correct writing example can be presented in a pop-up or at the bottom of the input field, along with a guidance message such as, "Please write at least one sentence including the specific situation of the experience and the role you performed." This re-entry request process prevents data with low analytical value from entering the system, thereby guaranteeing the quality and accuracy of the overall experience summary service.
[0075] Subsequently, the system can classify the initial experience data into one of a plurality of scenarios, including performance-oriented, complex, and process-oriented types (S30). That is, the system can classify the experience into one of a plurality of scenarios (S30) by analyzing the context and included key keywords of the received initial experience data using natural language processing (NLP) technology.
[0076] In the present invention, scenarios may include performance-oriented, process-oriented, and mixed types depending on the nature of the experience.
[0077] Specifically, the performance-oriented type may focus on clear numerical results or business impact, such as increased sales, cost reduction, or improved efficiency. The process-oriented type, on the other hand, may emphasize qualitative behavioral processes, such as logical thinking for goal achievement, collaboration and communication among team members, and methods for resolving unexpected problems.
[0078] The composite type is a form in which the characteristics of the two aforementioned types are combined and synthesized, and it may be a case where a causal relationship is revealed in which specific problem-solving actions lead to clear quantitative results. For example, an experience that includes both the process of establishing and executing a specific marketing strategy and the concrete conversion rate figures achieved as a result can be classified as the composite type.
[0079] The system can classify initial experience data as performance-oriented scenarios if it includes quantitative indicators such as numbers, achievement rates, revenue, and efficiency, or if it emphasizes business impact. Conversely, if the content emphasizes qualitative capabilities or the logical flow of actions—such as communication, collaboration, conflict resolution, and planning intent—it can be classified as process-oriented scenarios. This classification process goes beyond simply dividing categories and can serve as a criterion for determining which data to prioritize securing in subsequent stages.
[0080] The system may load a predefined reference data structure from memory (S40) according to a determined scenario. The reference data structure may be a set of data fields that must be included for the experience to be composed of a complete episode. This structure may be designed based on a framework (e.g., STARR method) that includes at least one of Situation data, Task data, Action data, Result data, and Reflection data.
[0081] The system can load different standard data structures for each scenario (S40). For example, if classified as a performance-oriented scenario, the system can call a data structure configured so that numerical evidence or before-and-after comparison data is necessarily included in the result data field.
[0082] On the other hand, in the case of a process-oriented scenario, a weighted data structure can be invoked so that specific problem-solving methods or one's role are described in detail in the task data or behavioral data fields. By establishing a data collection framework optimized for the characteristics of each experience, this can serve as a foundation for enabling customized experience organization rather than uniform questions.
[0083] The system can identify missing data among the initial experience data that is not mapped to a reference data structure (S50). That is, the system can load a predefined reference data structure according to a classified scenario and identify missing data (S50) by analyzing whether the received initial experience data satisfies the requirements of the structure.
[0084] After identifying missing data (S50), the system can determine whether missing data exists. Subsequently, if it is determined that missing data exists, the system can generate and provide a follow-up question.
[0085] Here, the standard data structure is based on the STARR framework, which includes Situation, Task, Action, Result, and Reflection, but may be a dynamic template in which the mandatory status and weight of each item are set differently according to the previously classified scenarios (performance-oriented, process-oriented, and complex types).
[0086] The system can perform the process of breaking down initial experience data into sentence units through natural language processing (NLP)-based named entity recognition (NER) and semantic analysis, and mapping them to each slot of a reference data structure. At this time, slots that are not mapped and are empty, or items that are mapped but whose data specificity is below a preset threshold, can be identified as missing data.
[0087] Specifically, in the case of a performance-oriented scenario, the system can set the Result item within the STARR framework as the top priority for verification. Even if the initial experience data contains a sentence stating "sales increased" and is mapped to the Result item, if specific figures or comparison criteria are not extracted, the system may not recognize this as complete data and may identify the quantitative performance indicator as missing data.
[0088] On the other hand, in the case of process-oriented scenarios, the system can focus on verifying the logical connectivity between Action and Task items. If a user mentions the problem situation but lacks a description of the specific strategies they formulated and executed to resolve it, the system can identify the specific actions taken for problem solving as missing data.
[0089] Furthermore, in the case of complex scenarios, the system can verify the causal relationship between Action and Result. Even if action and result data exist separately, if the causal link stating "I performed Action A, and Result B occurred" is not clearly identifiable within the text, the system can consider the explanation of this causal relationship itself as missing data. This process of identifying missing data can serve as a reference point for precisely targeting customized follow-up questions to be generated in subsequent steps.
[0090] The system can generate customized follow-up questions by applying a prompt engineering algorithm that reflects the content of the scenario and initial experience data, and provide the follow-up questions to the user terminal (S60).
[0091] That is, the system can generate and provide customized follow-up questions (S60) by applying a prompt engineering algorithm that reflects the content of classified scenarios and initial experience data to supplement identified missing data. The system can generate and provide customized follow-up questions (S60) by applying a prompt engineering algorithm that reflects the content of classified scenarios and initial experience data to supplement identified missing data based on domain research.
[0092] Here, a prompt engineering algorithm may be a command combination technology that controls a Large Language Model (LLM) to act as an agent leading a conversation based on domain expertise, performing the persona of an expert interviewer that embodies the user experience. Rather than simply listing fixed questions, the system can understand the context entered by the user and respond to it to naturally generate conversational questions in real time that ask for missing information.
[0093] The system can generate structured prompts that include roles, context, goals, and constraints to generate follow-up questions. For example, the system can internally combine prompts in the form of "You are a recruitment expert with 10 years of experience (role). The user entered 'increased sales' but there is no specific figure (context). Create a question asking for specific performance figures without offending the user (goal). However, the question must be written concisely in a single sentence (constraint)." and input them into a language model.
[0094] The system can dynamically change the target parameters of the prompt according to the previously classified scenarios. If the initial experience data is a performance-oriented scenario, the system can include an instruction in the prompt to "obtain quantitative figures for the result data."
[0095] Accordingly, the follow-up questions generated can take the form of eliciting specific figures, such as, "It is impressive that you increased sales. Do you happen to remember specifically what percentage it grew compared to the previous year, or what the exact amount achieved was?"
[0096] On the other hand, in the case of a process-oriented scenario, the system can include an instruction in the prompt to "secure specific methodologies for behavioral data."
[0097] For example, in response to the input "Resolved a conflict within the team," the system can generate follow-up questions asking for details of the action, such as, "What specific mediation methods did you employ during the conflict? Was there a special conversational strategy used to coordinate the opinions of both parties?" These generated follow-up questions are provided through the chat interface on the user's device, allowing the user to elaborate on their experience in a comfortable environment, as if conversing with a real mentor.
[0098] Furthermore, in the case of complex scenarios, the system can include a directive in the prompt to "strengthen the logical causal relationship between behavioral data and result data." Even if the user mentions both the behavior and the result, if the link between the two elements is weak, the system can generate a question to verify this.
[0099] For example, in response to the input "We achieved a 20% increase in sales by introducing a new marketing tool (Action, Action in the STARR structure) (Result, Result in the STARR structure)," the system can generate in-depth questions that go beyond a simple listing of facts to ask about the key factors of the performance, such as "Which feature of the tool you introduced played a decisive role in the increase in sales? Can you explain the causal relationship?"
[0100] The generated follow-up questions are provided through the chat interface of the user's terminal, allowing users to elaborate on their experiences in a comfortable environment, as if conversing with a real mentor.
[0101] The instructions can control the generation of a follow-up question that includes at least one of the specific time, place, or organization information where the user's experience occurred, if the missing data is identified as situation data.
[0102] Contextual data is an essential element for explaining the background of an experience, and it can play a role in ensuring the objectivity of the experience by defining the environment in which the performance or behavior took place.
[0103] The system can detect whether spatiotemporal information is missing from the initial data entered by the user through text mining.
[0104] For example, if a user only enters "conducted and launched an app development project" without specifying the exact timeframe or affiliation, the system may determine that it is unclear whether the experience was a "personal project," a "university capstone design," or a "corporate internship," and identify the context data as missing.
[0105] Accordingly, the system can generate customized follow-up questions to secure specific situational information. The system can generate questions that simultaneously inquire about the timeframe and affiliation, such as, "When was this project conducted? Was it part of a school course, or was it a club activity or an external competition?" Alternatively, it can generate questions that indirectly gauge the scale of the collaboration experienced by the user by asking about the size of the organization, such as, "What was the size of the department you worked in?"
[0106] This process of acquiring contextual data can serve as important supporting evidence to prove the recency of experience or emphasize organizational fit when writing cover letters in the future. The system analyzes user responses to extract keywords such as 'second half of 2023,' 'marketing team intern,' and 'startup,' and can automatically update these into structured contextual data fields.
[0107] The instructions can control the generation of follow-up questions asking about the Key Performance Indicator (KPI) of the task or problems encountered during the execution of the task when the missing data is identified as task data.
[0108] Task data is information that defines what the user had to solve in a given situation, and it can include how challenging the task was or what specific goals it had, beyond simply listing the given tasks.
[0109] If the system determines that the specificity of the goal is lacking in the user's initial input data, it can identify this as missing data in the task data. For example, if a user inputs, "I have written a new marketing plan," the system recognizes the action of "writing a plan," but since it cannot determine whether the goal of the plan is "attracting new customers," "raising brand awareness," or "preventing existing customer churn," it may determine that the task data is insufficient.
[0110] Accordingly, the system can generate follow-up questions asking about the task's key performance indicators. For example, through questions such as, "What was the most important goal of the proposal? Were specific key performance indicators (KPIs, e.g., click-through rate, conversion rate, etc.) set to be achieved?", the system can encourage users to define goals using numerical values or clear metrics.
[0111] In addition, the system can generate questions regarding difficulties or problems encountered during the task execution process. The value of experience can be evaluated more highly in the process of overcoming obstacles than in the performance of ordinary tasks.
[0112] If a user simply enters "I completed the team project," the system can generate a question such as, "Were there any unexpected obstacles or constraints (e.g., lack of time, budget cuts, technical limitations) during the project?" This prompts the user to recall the urgent situation or difficult problems encountered at the time, which can serve as a background that enhances the value of the action data described later.
[0113] The reference data structure may further include action data, and instructions can control the generation of follow-up questions asking about the tool used or strategy applied by the user if missing data is identified as action data.
[0114] Behavioral data defines the specific efforts and methodologies directly performed by the user to solve a task, and can serve as the most critical indicator for evaluating a user's practical capabilities in a self-introduction letter.
[0115] The system can identify cases where a result (STARR, Result) or situation (STARR, Status) exists in the user's input data, but the explanation of how it was resolved is missing or abstract as missing data in the behavioral data.
[0116] For example, if the input is "analyzed data and found a problem," the system may determine that while it recognized the act of "analysis," it omitted information regarding which tool was used or what criteria were used to analyze the data.
[0117] Accordingly, the system can generate follow-up questions asking for specific tools based on the user's job field or entered keywords.
[0118] For example, you can encourage users to specify their hard skills through questions such as, "Did you use Excel for data analysis, or did you utilize statistical tools like Python or R? Please tell me if there were any specific functions or libraries you used."
[0119] In addition, the system can generate follow-up questions asking about the strategy applied by the user. This can be particularly effective in business situations or conflict scenarios where there is no predetermined answer.
[0120] If a user inputs "I appeased customer complaints," the system can generate questions asking about specific behavioral strategies, such as, "Did you have any special know-how or persuasion strategies of your own when dealing with dissatisfied customers? Did you focus on presenting alternatives rather than offering unconditional apologies?" These questions can act as a trigger, encouraging users to express their soft skills or problem-solving capabilities in concrete terms, aspects they may have previously overlooked.
[0121] The instructions can control the generation of a question asking for quantitative data representing changes before and after the user's action as the tail question when the missing data is identified as result data.
[0122] Result data is the final output of tasks and actions performed by a user, and it can serve as an objective indicator proving how that experience actually contributed to an organization or project. In particular, since changes expressed in numbers are more persuasive than simple descriptions in the hiring process, securing quantitative data can be considered a key step in organizing experience.
[0123] The system can identify cases where positive adjectives (e.g., "better," "improved," "increased") exist in the initial data entered by the user, but specific figures supporting them are absent, as missing values in the resulting data.
[0124] For example, if the input is "Sales increased after changing the display method while working part-time at the store," the system recognizes the result as "sales increase," but since there is no basis to determine whether the increase is insignificant or groundbreaking, it may classify it as something that needs improvement.
[0125] Accordingly, the system can generate follow-up questions asking for quantitative data that can compare changes before and after performing an action. For example, through questions such as, "What were the average daily sales before and after changing the display method, respectively? Or do you remember approximately what percentage (%) the sales growth rate was?", the system can induce users to convert performance results they vaguely remember into specific numbers.
[0126] In addition, the system can generate questions regarding quantitative changes in various dimensions, such as time, efficiency, and personnel, as well as financial performance. If a user inputs, "I increased work processing speed," the system can generate a question asking about time efficiency, such as, "Specifically, how many minutes was a task that previously took one hour reduced?" The data obtained through these questions can be transformed into impactful performance statements when writing a cover letter in the future, such as, "Reduced deadline tasks that previously took 60 minutes to 30 minutes, a 50% reduction."
[0127] The reference data structure may further include reflection data, and instructions can control the generation of follow-up questions asking about lessons learned or improvements through the result data if missing data is identified as reflection data.
[0128] Reflective data defines a user's inner realization or professional insight derived from the results of experience, and beyond a simple record of facts, it can serve as an important metric for evaluating a user's potential for growth.
[0129] The system can identify cases where specific outcomes or behaviors are stated in the user's input data, but there is a lack of description regarding what meaning the experience left for the user, as missing data in the reflective data.
[0130] For example, even if a result such as "won the grand prize in a competition" is entered, if there is no mention of "what was learned" or "how the importance of teamwork was felt" during the process, the system may judge this as an incomplete state where the experience has not been concluded.
[0131] Accordingly, the system can analyze the result data and generate follow-up questions asking for corresponding lessons.
[0132] For example, users who have entered successful results can be encouraged to describe their analysis of the success factors and their commitment to continuous improvement through questions such as, "What was the biggest lesson learned from this success? If you were to apply this successful approach to your next project, what aspects would you like to improve?"
[0133] In addition, the system can generate questions asking about shortcomings or areas for improvement.
[0134] If a user inputs, "I completed the project, but I had difficulty managing the schedule," the system can generate a question such as, "If you could go back to the start of the project, how would you reduce the risk of schedule management?" Such questions can help users transform failures or regrets from mere disappointment into concrete improvement plans. Consequently, the resulting reflective data can be utilized in the concluding paragraph of a cover letter as a powerful message demonstrating the user's professional expertise and mature attitude.
[0135] If it is determined that there are multiple identified missing data, the instructions can be controlled to sequentially generate follow-up questions according to a pre-set priority algorithm to minimize the user's cognitive load and induce a natural conversation flow.
[0136] Based on the analysis of initial experience data, two or more items among Situation, Task, Action, Result, and Reflection may be identified as missing data simultaneously.
[0137] In this case, providing questions for all missing items at once can put psychological pressure on the user, potentially lowering the quality of the answers or causing them to leave the service. Therefore, the system can determine the priority of queries among multiple missing data and provide follow-up questions step-by-step, starting with the item with the highest priority.
[0138] The system can determine the priority of missing data by applying a sequential query method based on the logical flow of experience development. Based on the narrative structure of the STARR framework, the system can assign basic priorities in the order of situation, task, action, result, and reflection.
[0139] For example, if situation, action, and result data are all missing from the input by the user, the system may first generate and provide context data follow-up questions to understand the background of the experience. That is, when the user answers the above context data follow-up questions, the system can then perform a circular question-and-answer session that sequentially asks about actions and results.
[0140] In addition, the system can apply a dynamic priority allocation method based on scenario types. The system can assign the highest priority to the most critical data according to the previously classified scenarios.
[0141] If the experience is classified as performance-oriented, the system can set securing result data as the top priority and first generate follow-up questions asking for numerical performance.
[0142] On the other hand, if the experience is classified as process-oriented, the system may determine the concretization of behavioral data as the most urgent task and prioritize placing follow-up questions asking about the problem-solving method.
[0143] Furthermore, instructions can control the merging of two highly correlated missing data points among multiple missing data points to generate a single complex tail question.
[0144] For example, if both situation data and task data are missing, the system can efficiently reduce the number of question-and-answer cycles by combining them into a single sentence that simultaneously fills in the affiliated organization and assigned key objectives. The system can provide high-priority follow-up questions, receive user responses, and immediately after updating the corresponding missing items, reconstruct an interactive prompt that naturally continues the context of the next follow-up question based on the remaining low-priority missing data.
[0145] As a result, the system can provide a user experience that feels like putting together a puzzle while conversing with an expert.
[0146] Before receiving the above career-related initial input data, the instructions can identify the user's input purpose and control the loading of the standard data structure according to the input purpose.
[0147] Here, the input purpose refers to the user's ultimate goal of organizing their experiences through this system; therefore, even if the experience data is identical, the data fields to be emphasized and their structure may differ depending on the purpose for which it is organized.
[0148] The system may provide an interface on the user terminal that allows the user to select an input purpose before the user accesses the service platform and begins organizing experiences. The interface may visually present input purpose categories including at least one of writing a self-introduction letter, preparing for an interview, writing a resume, writing a career description, job search, preparing for a job change, career counseling, job switching, salary negotiation, performance evaluation, performance review, recruitment matching, and career path recommendation.
[0149] Users can select the category that best suits their current situation and usage goals, and the system can set criteria for subsequent processes using the selected category as a key value.
[0150] The system can load different reference data structures from memory depending on the selected input purpose. For example, if the input purpose is selected as writing a self-introduction letter, the system can load a reference data structure designed to focus on collecting experience information based on the STARR framework.
[0151] On the other hand, if the input purpose is selected as recruitment matching or career path recommendation, the system can load an extended standard data structure to collect not only experience information but also the user's competency information, goal information, preference information, compensation information, organizational culture information, and matching judgment information.
[0152] Additionally, if the input purpose is selected as salary negotiation, the system can load a reference data structure with weighted quantitative performance data and compensation information fields.
[0153] If the input purpose is selected as personnel appraisal or performance review, the system can load a reference data structure with weighted result and reflection data within a specific period. If the input purpose is selected as preparation for job transition or changing jobs, the system can load a reference data structure that enhances the user's transferable competency and goal information fields.
[0154] This branching of standard data structures based on input purposes can serve as a foundation for improving the accuracy and utility of subsequent services by ensuring that data is structured in an optimized form according to the purpose of use, even if the same user inputs the same experience.
[0155] The above missing data may include at least one of a missing state in which information corresponding to a specific field of the reference data structure is absent, a deficient state in which information exists in the field but the reliability is below a preset threshold, an uncertain state in which the possibility of interpreting the extracted information is multiple or the basis for inference is weak, and a conflict state in which information extracted from multiple utterances of the user is mutually contradictory.
[0156] Here, missing data goes beyond the conventional concept of simple missing values and can be a concept that encompasses all states deemed to require supplementation as a result of evaluating data quality and reliability from various angles.
[0157] Specifically, a missing state may refer to a case where information corresponding to a specific field of the reference data structure does not exist at all in the user's initial input data.
[0158] For example, if a user enters only "successfully completed the project," the system may classify the context data field as missing because the time, place, or affiliation information to be mapped to the context data field is absent.
[0159] An insufficient state may refer to a case where information exists in the field, but the reliability of that information is below a preset threshold. The system can calculate a specificity score of the information extracted from the user's utterance through a natural language processing model and classify it as an insufficient state if the score is below the threshold (e.g., 0.7).
[0160] For example, even if information such as "sales increased slightly" is mapped to a result data field, if a low reliability score is calculated due to a lack of quantitative figures, the system can identify this as an insufficient state.
[0161] An uncertain state may refer to a case where information extracted from a user's utterance has multiple possible interpretations or where the basis for inference is weak.
[0162] For example, if a user inputs "played an important role in the team," the system can classify the behavioral data field as uncertain because the expression "important role" can be interpreted in multiple ways—whether it refers to a leader role, a key operational role, or a decision-maker role.
[0163] A conflict state may refer to a case where information extracted from multiple utterances of a user is mutually contradictory.
[0164] For example, if a user initially enters that they "proactively led a team project," but later answers to a follow-up question that they "performed auxiliary tasks according to the team leader's instructions," the system may classify this as a conflict state by determining that conflicting information exists in the behavior data field.
[0165] The system can be controlled to generate customized tail questions corresponding to the identified data status. In other words, even for the same data field, different types of tail questions can be generated depending on the status of the missing data.
[0166] If identified as missing, the system can generate a direct query to obtain the basic information of the corresponding field from scratch.
[0167] For example, you can create questions that ask about the time, place, and affiliation simultaneously, such as "When, where, and to which organization did you conduct this experience?"
[0168] If identified as insufficient, the system can generate questions that supplement specificity while acknowledging the previously entered information.
[0169] For example, you can create a question requesting supplementation while citing an existing answer, such as, "You mentioned that you contributed to the increase in sales; could you tell me specifically in numbers how much the sales changed?"
[0170] If identified as an uncertain state, the system may generate a multiple-choice or clarification question to clarify which of the multiple possible interpretations applies.
[0171] For example, you can guide the user to clearly define their role by presenting multiple options, such as, "You mentioned that you played an important role in the team, but specifically what was that role? For example, was it a decision-maker who determined the team's direction, a person in charge of core operations, or another type of role?"
[0172] If a conflict is identified, the system can present conflicting information to the user and generate a verification question to confirm which one is correct.
[0173] For example, you can generate questions that directly point out contradictions, such as, "You previously stated that you proactively led the project, but later answered that you performed auxiliary tasks under the team leader's direction. Could you tell me which of the two answers is closer to your role in the project, or whether your role changed depending on the time period?"
[0174] The process of classifying such missing data by status and generating customized tail questions can go beyond simply filling empty slots and have the effect of sophisticatedly enhancing the qualitative completeness of the data.
[0175] When there are multiple identified missing data or multiple data fields requiring supplementation, the instructions can calculate the priority of the follow-up questions by considering at least one of the narrative development order of the reference data structure, whether it is a core field according to the scenario type, the degree of relevance to subsequent functions, and the user's fatigue or the possibility of disengaging from the conversation, and control the provision of the follow-up questions sequentially according to the calculated priority.
[0176] When multiple missing data exist simultaneously, providing questions for all missing items randomly or simultaneously can increase the user's cognitive load and degrade the quality of the answers. Therefore, the system can dynamically calculate the priority of follow-up questions by comprehensively considering multiple criteria.
[0177] The system can utilize the narrative development order of the reference data structure as a criterion for calculating priorities. In the case of the STARR framework, since it has a structure in which Situation, Task, Action, Result, and Reflection unfold according to a temporal and logical flow, the system can assign basic priorities based on this order.
[0178] For example, if both situation data and result data are missing, the system may provide follow-up questions regarding the situation data before follow-up questions regarding the result data to first secure the background of the experience.
[0179] The system can utilize the presence of core fields based on scenario type as a criterion for calculating priorities. Since the most important data fields differ depending on the previously classified scenarios, the system can assign priority weights to the core fields of the corresponding scenarios.
[0180] For example, if both behavioral and outcome data are missing in an experience classified as a performance-oriented scenario, the system may prioritize providing follow-up questions regarding quantitative indicators of outcome data—which is the core of performance-oriented scenarios—over follow-up questions regarding behavioral data.
[0181] The system can utilize the relevance to subsequent functions as a criterion for calculating priorities. Depending on the previously identified user input purpose, specific data fields may play a more significant role when utilizing subsequent services.
[0182] For example, if the input purpose is writing a self-introduction letter, the system can calculate a relatively high priority for behavioral and reflective data, which are key elements of the letter. On the other hand, if the input purpose is recruitment matching, the system can calculate a high priority for fields related to competency and compensation information used in the matching algorithm.
[0183] The system can utilize user fatigue or the likelihood of disengaging from the conversation as criteria for calculating priorities. The system can quantify user fatigue by monitoring trends in response length, response time, and specificity scores in real time. If fatigue exceeds a preset threshold, the system can prevent user disengagement by prioritizing short-answer or multiple-choice questions that place less burden on the user. Conversely, at a point where user responses are active and highly specific, the system can prioritize more in-depth questions that require thought.
[0184] The system can calculate a priority score for each missing data item by comprehensively considering the aforementioned multiple criteria and provide follow-up questions sequentially, starting with the items with the highest scores. Whenever a user completes an answer to a follow-up question, the system can recalculate the priorities of the remaining missing data items to dynamically adjust the flow of the conversation.
[0185] Instructions can control storing the above-mentioned answer data as original conversation data and storing the career data field extracted from the above-mentioned original conversation data by connecting it to the corresponding original utterance section.
[0186] Conventional data processing methods had limitations in that they extracted only key information from user utterances and stored it in a structured form, while discarding or failing to manage the original utterances themselves. However, preserving original conversational data may be required to verify information loss or interpretation errors occurring during the structuring process and to reuse the context of the user's utterance in subsequent services.
[0187] The system can store all response data entered by the user through a chat interface or voice recognition interface in a database in chronological order in the form of original conversation data. The original conversation data can be stored in a structured log form that includes the time of the user's utterance, the content of the utterance, and the system's question information corresponding to that utterance.
[0188] The system can extract career data fields from original conversation data through natural language processing. At this time, the system can store each extracted career data field by linking it to the original utterance segment that served as the basis for the information.
[0189] Specifically, the system can store utterance segment identifiers (e.g., utterance index, start character position, end character position) that identify specific sentences or phrases of the original utterance as metadata for the career data field.
[0190] For example, if a user utters, "I worked as a marketing team intern in the second half of 2023 and planned a new member acquisition campaign, achieving 30% growth compared to the previous month," the system can preserve this original utterance as is, store the value "second half of 2023, marketing team intern" in the context data field, and link it together with the source utterance section.
[0191] Likewise, the behavior data field can store a value "planning a new member acquisition campaign" connected to its underlying utterance, and the result data field can store a value "30% growth compared to the previous month" connected to its underlying utterance.
[0192] As such, storing the connection with the original utterance segment can serve as a foundation for verifying the accuracy of career data fields extracted in subsequent services, generating natural sentences that reflect the context of the original utterance when creating a self-introduction letter, or enabling users to verify and modify the source of their experience data.
[0193] Instructions can control the assignment of at least one reliability type among explicit utterance-based, inference-based, uncertainty, and conflict to the career data field.
[0194] Conventional data processing methods were limited to expressing the reliability of extracted data as a single numerical score, failing to provide qualitative information regarding the basis on which the data was derived. However, when utilizing data in subsequent services, qualitative classification information that allows for a more intuitive understanding of the data's source and derivation method may be required, rather than simple numerical scores.
[0195] In the process of extracting career data fields from original conversation data through natural language processing, the system can assign different reliability types depending on the derivation method of each field.
[0196] Explicit utterance-based reliability types can be assigned to career data fields extracted based on information directly specified by the user in the original conversation.
[0197] For example, if time information is extracted by a user clearly uttering "started in September 2023," the system can assign an explicit utterance-based confidence type to the corresponding situation data field.
[0198] Inference-based reliability types can be assigned to career data fields derived by the system through inference from the context of the utterance or surrounding information, even if the user has not explicitly specified them.
[0199] For example, if a user says "a project conducted in the first semester of the fourth year of college," the system may refer to the user's academic records to convert that time into a specific year and month and store it, and may assign an inference-based confidence type to this inferred value.
[0200] An uncertain confidence type may be assigned when information extracted from a user's utterance has multiple possible interpretations or when the confidence score of the extraction model is below a preset threshold.
[0201] For example, if a user's utterance is ambiguous and it is unclear whether the extracted value corresponds to behavioral data or result data, the system can assign an uncertain confidence type to the corresponding field.
[0202] A conflict reliability type may be assigned when information extracted from multiple user utterances is mutually contradictory or inconsistent with external reference data. The system may assign a conflict reliability type to the corresponding field when a user provides different answers regarding the same fact, or when the content of a user's utterance contradicts other experience data already stored in the database.
[0203] Assigning such reliability types can enable differentiated processing when utilizing data in subsequent services.
[0204] For example, when generating a self-introduction letter, the system prioritizes the use of explicit utterance-based data to ensure factual consistency, while utilizing inference-based data but requesting reconfirmation from the user if necessary. By controlling uncertain or conflicting data to require explicit user verification before use, it is possible to prevent incorrect information from being reflected in subsequent services.
[0205] FIG. 3 is a flowchart for the generation of follow-up questions for a performance-oriented scenario in an experience organization system for customized job search using questions according to an embodiment of the present invention. FIG. 4 is a flowchart for the generation of follow-up questions for a complex scenario in an experience organization system for customized job search using questions according to an embodiment of the present invention. FIG. 5 is a flowchart for the generation of follow-up questions for a process-oriented scenario in an experience organization system for customized job search using questions according to an embodiment of the present invention. FIG. 6 is a flowchart for the process of constructing structured career data in an experience organization system for customized job search using questions according to an embodiment of the present invention.
[0206] Referring to FIG. 3, the instructions can check (S41) whether numerical indicators exist in the result data when the initial experience data is classified as performance-oriented. Subsequently, if numerical indicators are absent in the result data, the system can be controlled to generate and provide a follow-up question asking for numerical indicators (S61). If the numerical indicators exist, the reference data can be converted into structured career data and stored (S80).
[0207] That is, the instructions can determine whether numerical indicators are absent in the result data when the initial experience data is classified as performance-oriented, and if they are absent, generate and provide a follow-up question asking for the numerical indicators (S61).
[0208] In this context, since performance-oriented scenarios deal with experiences that have a clear business impact, such as increased revenue, reduced costs, and improved efficiency, numbers that can prove this can be the most important factor in determining the credibility of the experience.
[0209] The system can detect patterns in user input data through natural language processing where descriptive terms indicating performance (e.g., "improved," "reduced," "achieved") are present, but quantitative figures indicating the degree (e.g., "10%," "5 million won," "3 days") are not included.
[0210] For example, if a user enters "attracted many new members through a marketing campaign," the system may determine that the expression "many" is subjective and identify it as lacking numerical indicators.
[0211] Accordingly, the system can generate follow-up questions that encourage users to recall specific figures. The system can generate questions that ask for both absolute numbers and relative ratios, such as, "How many specifically did the number of members increase? Or do you remember the approximate percentage (%) of the growth compared to the previous month?" Through this, users can concretize vague performance into objective data, such as "attracting 1,000 new members" or "150% growth compared to the previous month."
[0212] Furthermore, the system can generate questions inquiring about numerical indicators across various dimensions, such as time, personnel, and frequency, in addition to monetary figures. If the input states, "Meeting time was shortened," the system can request specific data to prove time efficiency, such as, "How many minutes did a meeting that previously took one hour take on average?" This process can perform the key function of transforming a user's subjective claims into objective facts that recruiters can trust.
[0213] Referring to FIG. 4, instructions can identify a causal relationship (S42) in which behavioral data results in result data when the initial experience data is classified as a complex type. Subsequently, if the above causal relationship does not exist, the system can control the generation and provision of a connectivity-enhancing question asking about the causal relationship as a follow-up question (S62). If the above causal relationship exists, the reference data can be converted into structured career data and stored (S80).
[0214] That is, the instructions can control the generation and provision of a connection-enhancing question as a follow-up question (S62) that asks about the causal relationship in which the behavior data leads to the result data, when the initial experience data is classified as a complex type and there is a lack of causal relationship between the behavior data and the result data.
[0215] Complex scenarios deal with a high-quality experience that includes both the specific problem-solving process and the resulting quantitative outcomes; however, if the logical link between the two elements is weak, it may amount to nothing more than a mere listing of facts.
[0216] The system can analyze whether there is a lack of causal descriptions connecting behaviors and results, even if behaviors and results exist separately in the user's input data.
[0217] For example, if a user inputs "introduced a new marketing tool (action)" and "sales increased by 20% (result)," but there is no explanation of 'why or how' the introduction of the tool led to the increase in sales, the system may determine that there is a lack of causal relationship. Since recruiters generally want to verify whether the strategy intended by the applicant actually worked, rather than simply the fact that the results were good, it may be required to strengthen the aforementioned link.
[0218] Accordingly, the system can generate connectivity-enhancing questions that identify the black box between behavior and results. The system can generate questions asking about key factors of performance, such as, "Which specific feature of the marketing tool you introduced contributed decisively to the increase in sales? Can you explain the correlation?"
[0219] In addition, the system can generate questions that induce the user's analytical thinking, such as, "What do you think is the reason that this strategy was the most effective among various attempts?" By reinforcing the answers to these questions, the self-introduction can erase the impression that "success was achieved by luck" and instead possess a persuasive logical structure demonstrating that "success was inevitably produced through thorough strategy and execution."
[0220] The instructions can identify a causal relationship in which behavioral data leads to result data when the initial experience data is classified as a complex type (S42).
[0221] Referring to FIG. 5, instructions can check the problem-solving content of the behavior data (S43) when the initial experience data is classified as process-oriented. If the problem-solving content is missing in the behavior data, instructions can control the generation and provision of a problem-solving question as a follow-up question (S63).
[0222] That is, in a process-oriented scenario, the problem-solving content of the behavioral data is checked, and if problem-solving content exists, the reference data can be converted into structured career data and stored (S80). However, if problem-solving content does not exist, follow-up questions can be generated and provided (S63).
[0223] In this context, a process-oriented scenario can refer to experiences that emphasize the logical flow of overcoming situations and the specificity of actions, such as conflict resolution, the exercise of leadership, and adaptation to new environments, rather than quantitative figures. What recruiters in the hiring market focus on most in these experiences may not be "what was done," but rather "what thoughts and intentions were behind the actions."
[0224] The system can analyze whether the problem situation is clearly described in the user's input data, but whether specific actions or solution logic to resolve it are missing.
[0225] For example, if a user inputs, "I completed the project by mediating differences of opinion among team members," the system may determine that while the act of "mediation" exists, there is a lack of specific logic regarding the solution—such as what alternatives were proposed after listening to both sides or what persuasive arguments were used. This can be viewed as insufficient to demonstrate active problem-solving capabilities that go beyond the role of a simple mediator.
[0226] Accordingly, the system can generate problem-solving questions that ask about the specific solutions the user implemented at the time. The system can generate questions that delve into the details of actions, such as, "What specific compromise did you propose to narrow the differences in opinion? Or were there any specific supporting data or communication methods used to persuade both parties?"
[0227] In addition, the system can generate questions regarding crisis management skills in response to input such as "coordinated with sudden schedule changes," asking, "How did you readjust work priorities to meet the schedule? Or what negotiation process did you go through to secure additional resources?" Through these questions, users describe specific "strategies" and "executions" rather than abstract "efforts," which can serve as decisive evidence in their cover letters to demonstrate their professional expertise and problem-solving abilities.
[0228] Referring again to FIG. 1, the system can receive answer data input from a user terminal regarding a follow-up question and update the missing data (S70). That is, the system can identify the missing data, determine whether the missing data exists, and if the missing data exists, generate and provide a follow-up question (S60) to receive new answer data from the user terminal (S70). The system can receive the new data, update the data in the missing data section, and determine again whether the missing data exists (S50).
[0229] The system receives answer data input from the user terminal regarding the provided follow-up question (S70), analyzes it using natural language processing, and can update the corresponding missing data item of the previously loaded reference data structure (S70).
[0230] In this case, the response data may be text or voice data entered by the user through a chat interface, etc., and the system can perform a slot filling process to extract key entities from this response and fill the empty slots.
[0231] Immediately after the missing data is updated with the answer data, the system can determine again (S50) whether missing data still exists within the reference data structure. If the user's answer is not sufficiently specific, deviates from the intent of the question, or if one missing item is filled but other related required items remain empty, the system can determine that missing data remains.
[0232] For example, if a user only answers "it has increased quite a lot" to the question "Please tell me the performance figures," the system may consider the missing data state not to have been resolved because the 'quantitative figures' have still not been secured.
[0233] If missing data exists, the generation and delivery of tail questions can be repeated. By continuously updating the missing data through tail questions, the presence of missing data can be determined. Ultimately, the system can repeat the tail question generation and delivery steps until the missing data is completely resolved.
[0234] The system can change the intensity or method of questioning by reflecting the context of the previous answer. In the previous example, the system can encourage users to input data without giving up by lowering the hurdle for an answer—such as asking, "If it is difficult to verify specific figures, could you at least provide a rough estimate or a before-and-after comparison ratio?"—or by asking again with specific examples. This circular question-and-answer process can continue until all essential items of the reference data structure possess a level of specificity exceeding a threshold.
[0235] With the update of missing data, the absence of missing data can be considered as a case where the standard data structure is satisfied. Depending on the performance-oriented, complex, and process-oriented scenarios, the weight of situation data (Situation, S in STARR), task data (T in STARR), action data (A in STARR), result data (R in STARR), and reflection data (R in STARR) may differ.
[0236] If all criteria data structures with different weights for each of these scenarios are satisfied, the system can convert the updated criteria data into structured career data through follow-up questions and store it (S80).
[0237] In other words, if the system determines that the standard data structure is satisfied, it can terminate the iterative process, convert the updated standard data into structured career data, and store it in the database.
[0238] Here, experience data refers to complete data objects structured and refined according to the STARR framework, rather than initial unstructured text. The stored experience data can be utilized as core source data for various job support services in subsequent stages, such as generating users' self-introductions, extracting anticipated interview questions, or creating career analysis reports.
[0239] Referring to FIG. 6, when the standard data structure is satisfied, the system may identify multiple experience candidates or episodes from one initial input data or conversation session prior to storage (S80) (S71), extract career data fields for each experience candidate or episode (S72), extract common competencies, recurring patterns, or strengths among multiple experiences of the same user (S73), map the relationships between the extracted career data fields (S74), and then convert and store them as user-specific structured career data (S80).
[0240] Meanwhile, the above instructions can identify multiple experience candidates or multiple episodes (S71) from a single initial input data or a single conversation session. That is, the system does not necessarily store a user's input as a single experience, but can analyze through natural language processing whether the input contains multiple distinct experiences or events. For example, if a user inputs, "During my internship, I performed data analysis, improved the customer service process, and managed the team project schedule," the system can separate this into multiple experience candidates, such as 'data analysis experience,' 'customer service process improvement experience,' and 'team schedule management experience,' instead of storing it as a single experience.
[0241] The system can individually extract career data fields corresponding to the reference data structure for each of the aforementioned multiple experience candidates or episodes (S72). Additionally, the system can extract common competencies, recurring patterns, strengths, preferences, or career directions that appear repeatedly among multiple experiences of the same user (S73) and store them as a structured career data set for each user. For example, if 'data-driven problem-solving competency' is commonly extracted from the three experience candidates, the system can structure this as the user's representative strength field so that it is utilized preferentially in subsequent functions.
[0242] Furthermore, the above instructions can map the relationships between multiple extracted career data fields (S74). The relationships may include at least one of the following: a relationship between specific experience and specific competency, a relationship between specific behavior and specific performance, a relationship between specific tools used and specific job competency, a relationship between specific problem-solving methods and specific work styles, a relationship between specific preference conditions and specific organizational culture fit, or a basis relationship between specific experience data and subsequent generated results. For example, the system may interconnect behavior data such as 'improvement of customer response process' and performance data such as 'reduction of response time by 30%', and store the 'CRM tool' data used in the corresponding behavior by linking it with 'customer management job competency'.
[0243] Structured career data with mapped relationships between data fields in this manner enables tracking of which experience and competency data a specific outcome is derived from in subsequent functions such as resume generation, interview response generation, career path recommendation, or recruitment matching, and can function as foundational data that allows the same data to be repeatedly utilized in multiple subsequent functions.
[0244] FIGS. 7 and FIGS. 8 are flowcharts of an ontology-based career path recommendation system according to an embodiment of the present invention.
[0245] Referring to FIGS. 7 and 8, the system can build and store a static ontology (S90), map job posting data and user experience data to an ontology node (S100), calculate a career path and path suitability score (S110), and provide graph data to a user terminal (S120).
[0246] Instructions can be used to construct a static ontology that defines at least one of the standard job, task, industry, required knowledge, company, team, required competency, technology, equipment, tool, organizational culture, compensation conditions, evaluation criteria, and verification items within a specific industry domain as an ontology node (S91), and defines the association, hierarchy, utilization, and premise relationships between the ontology nodes as ontology edges (S92), and store it in a database (S90).
[0247] In this embodiment, for clarity of explanation, the semiconductor industry, characterized by a complex value chain and technical connectivity, is used as an example, but the concept of the present invention is not limited thereto.
[0248] The system can define fixed standard node types by reflecting the technical characteristics of the semiconductor industry.
[0249] Specifically, a standard job node can be a concept that unifies job titles, which are called differently by each company, into industry standards.
[0250] For example, job titles that are used interchangeably as 'Process Design', 'Process Architect', and 'PA' can be defined as a single standard node called 'Process Design Engineer' to eliminate ambiguity in job names. Industry and process nodes can define detailed domains such as the eight major processes of semiconductor manufacturing, and nodes such as 'Lithography', 'Etching', 'Deposition', and 'Packaging' can be created.
[0251] In addition, task nodes can define specific actions (Operations) performed in the corresponding job. For example, 'recipe optimization', 'yield analysis', 'chamber maintenance', etc. may be included.
[0252] The required knowledge and competency nodes can encompass tools, equipment, materials, and knowledge essential for job performance.
[0253] Due to the nature of the semiconductor industry, equipment such as 'ASML Scanner' or 'Lam Research Etcher', materials such as 'ArF Immersion', software tools such as 'Spotfire' or 'Python', and theoretical knowledge such as 'Plasma Physics' can be subdivided and managed as separate nodes.
[0254] The nodes defined above can be interconnected through edges with direction and meaning to form a knowledge graph.
[0255] The hierarchical relationships between the aforementioned ontology nodes can define a classification system of superordinate and subordinate concepts. For example, the 'etching process' node can be connected as a subordinate concept (SUB_OF) of the 'front-end process' node to form a hierarchical structure of the process.
[0256] The utilization relationships between the above-mentioned ontology nodes can define that specific equipment or tools are used in a specific process. For example, the 'EUV lithography equipment' node can be connected to the 'photo process' node via an edge marked 'used_in'.
[0257] The associations between the aforementioned ontology nodes can define connections between jobs and processes, or between companies and jobs. For example, a 'Process Engineer' node is closely connected to a 'Unit Process' node via 'related_to', and a specific 'Semiconductor Manufacturer' node can be connected to the corresponding job via 'has'.
[0258] The premise relationships between the aforementioned ontology nodes can define the knowledge or experience essential for performing a specific job. The static ontology constructed in this way functions as a domain map that structures the knowledge system of the industry domain, and can be periodically updated through a version control system and stored in a database. This can serve as a reference dictionary for interpreting and normalizing the dynamic job postings described later.
[0259] The system can normalize job posting data and user experience data into standard terms defined in a static ontology to create job posting nodes and user nodes, and map them to ontology nodes (S100). At this time, the instructions can map job posting nodes and user nodes by analyzing the hierarchical structure defined in the static ontology during the mapping and analyzing edges based on a domain context that includes higher-level concepts.
[0260] The system can perform natural language processing by receiving job descriptions (JDs) composed of unstructured text and user resume data. The system can extract key keywords related to job duties, tasks, and technology stacks from the input text and undergo a normalization process that compares them with standard nodes of a previously constructed static ontology.
[0261] For example, if a job posting states "responsible for etching among the 8 major semiconductor processes," the system can convert this into a standard process node called 'Etching'. Additionally, text such as "experience in operating Lam equipment preferred" can be normalized into a specific equipment node called 'Lam Research' and connected to the job posting node.
[0262] The system does not simply determine whether text matches, but can grasp the domain context by utilizing the hierarchy and association of the ontology.
[0263] For example, the semiconductor industry can have a structure in which upper-level processes, lower-level detailed processes, and equipment and components dependent on specific processes are intricately intertwined. Therefore, it may be necessary to determine whether they can be connected within the domain context, even if the superficial keywords differ.
[0264] Specifically, the system can expand the mapping scope by analyzing the inclusion relationship (SUB_OF) between super-concepts and sub-concepts in the ontology. For example, even if a user describes "Front-end Process engineering experience" in their resume without specific process names, the system can recognize that "photo," "etching," and "deposition" processes exist as sub-concepts of "Front-end Process" in the ontology.
[0265] Accordingly, the system can potentially connect the corresponding user node with lower process nodes, or conversely, link a user node with the keyword 'front-end process' to an announcement node requesting an 'etching process' through a higher-level concept matching.
[0266] Furthermore, the system can perform contextual reasoning based on the used-in relationship between equipment and processes. If a user inputs "ASML Scanner operable," the system can identify, through ontology knowledge, that the ASML Scanner is essential equipment used in the "photo process," even if the word "photo process" is not present in the text. Accordingly, the system can automatically map the user node to the "photo process" node and connect it to an announcement node seeking photo process engineers.
[0267] This domain context-based matching can resolve semantic inconsistencies arising from different job titles (e.g., Planner vs. Event Planner) or discrepancies in wording between job postings and resumes. Consequently, the system can enhance recommendation accuracy by going beyond simple keyword matching and performing precise node mapping based on industry structure and technological correlations.
[0268] Instructions can search for an ontology edge connected from a user node to an announcement node on a static ontology to calculate a career path that can be moved between jobs, and calculate a path fit score (S110) based on the connection strength or distance of the ontology edge.
[0269] The system does not simply find job postings with matching job titles, but can explore a logical path from the user's current location (Start Node) to the target job posting (Target Node) on the ontology graph.
[0270] At this point, the system can calculate a path suitability score. For example, a higher score can be assigned as the edge distance (Hop Count) between the user node and the job posting node is shorter, or as the weighted sum of the connected edges is higher. This quantifies the difficulty or suitability of a user changing jobs or moving to that position, and can serve as a criterion for providing quantitative indicators to the user, such as "This job is 95% suitable."
[0271] When calculating the path fit score, the instructions sum the weights of multiple edges existing on the career path, but may assign a higher weight if the edge is a prerequisite.
[0272] Here, the path fit score can be a metric that quantifies the technical feasibility and probability of success when the user moves to the target job. Not all connections (edges) on the ontology have equal importance, and 'essential conditions' required for performing a specific job can be distinguished from 'preferential conditions' or 'loose associations' that are merely desirable.
[0273] The system can differentially set default weights according to each edge type during the static ontology construction phase. For example, a low weight (e.g., 1.0) can be assigned to simple 'Company-has-Job' relationships or broad 'Related-to' relationships, while a high weight (e.g., 5.0) can be assigned to 'Prerequisite' relationships or 'Used-in' relationships that have a decisive impact on job performance.
[0274] To take the semiconductor industry as a specific example, consider a scenario where an 'Equipment Engineer' user wishes to transition their career to a 'Process Engineer' role. While both roles are connected to the 'Semiconductor Manufacturing' industry node, this connection alone may not be sufficient to justify the transition. In this case, the system can identify that the 'Process Engineer' node is connected to the 'Plasma Physics' knowledge node as a prerequisite.
[0275] If the user's experience data includes projects or coursework related to 'plasma physics' and is connected to the corresponding knowledge node, the system can significantly increase the path fit score by summing the high weights assigned to the edges on this path.
[0276] On the other hand, if only low-weighted edges are connected, such as the fact that they simply worked at the same company or proficiency in using general office software (Excel), the path fit score may be calculated low.
[0277] Consequently, the system can provide highly credible reasons for recommendations regarding paths that yield high scores, such as, "Since you possess knowledge of plasma physics, an essential competency for a process engineer, there is a greater than 90% chance of a job transition." This weighted algorithm can serve as a key mechanism to prevent misrecommendations caused by simple keyword matching and to propose practical career paths based on domain expertise.
[0278] In addition, when calculating the path fit score, the instructions can control identifying the basis node shared by the user node and the announcement node on the static ontology, and providing the reason why the career path was proposed through the path fit score and the basis node.
[0279] Here, the basis node may include at least one of the tasks, industries, or necessary knowledge that the user node and the announcement node have in common.
[0280] Specifically, in the semiconductor industry, we can assume a situation where the user is currently an 'etching process engineer' and the recommended job posting is for a 'process integration (PI) engineer'. Although the two job nodes may not be directly connected on the ontology, the system can identify the common basis node connecting the two nodes.
[0281] For example, grounds can be based on the fact that both jobs require knowledge of 'plasma physics,' have in common that they operate 'Lam Research Etcher (Equipment),' or share the task of 'yield improvement.'
[0282] The system can visualize these supporting nodes and provide them to the user terminal. Rather than simply notifying the result as "Job A is recommended," it can explain the reason for the recommendation, such as, "Since you possess plasma knowledge and experience with etching equipment, we recommend a process integration role where you can utilize these skills."
[0283] The system can help users intuitively understand the validity of a career path by highlighting supporting nodes, such as 'plasma' and 'etching equipment,' located between the user and job posting nodes through a graph-based UI. This recommendation method, equipped with 'path, score, and supporting evidence,' goes beyond simply providing information to offer users insights into how their competencies can be transferred to other roles.
[0284] FIG. 9 is a flowchart of an ontology-based career path recommendation system according to an embodiment of the present invention. FIG. 10 is a diagram showing an example of a graph data provision step (S120) of an ontology-based career path recommendation system according to an embodiment of the present invention.
[0285] Referring to FIGS. 9 and 10, instructions can control providing graph data in which the user node and the announcement node are commonly connected, along with the career path and path suitability score, to the user terminal (S120).
[0286] Unlike existing methods that simply list recommendation results in the form of a text list, this can enhance the persuasiveness of recommendations by visually showing how users and job postings are connected within an ontology. On the user terminal's display, a graph is output with the user node as the starting point and the job posting node as the endpoint, and common technology, equipment, and process nodes existing between the two nodes can be visualized as being connected by edges.
[0287] Additionally, instructions can propose a step-by-step career path by exploring one or more intermediate nodes connecting the user node and the job posting node when the user node and the job posting node are not directly connected or the connection strength is low—that is, when immediate movement is difficult due to a large gap in job competencies even if connected. The system can control the career path to sequentially connect the user node, intermediate nodes, and job posting nodes.
[0288] To illustrate with the semiconductor industry as an example, we can assume a situation where a user currently performing the 'Equipment Maintenance' role, responsible for repairing and maintaining semiconductor equipment, wishes to transition to the 'Process Integration (PI)' role, which involves designing the flow of all eight major processes and managing yield.
[0289] In the ontology, 'facility maintenance' nodes are strongly connected to hardware kinematics or electronic circuit knowledge, whereas 'process integration' nodes can be connected to semiconductor device physics or data statistical analysis.
[0290] Therefore, the system determines that the direct connection strength between the two roles is below a threshold, identifying that the probability of acceptance is low if applying immediately. In this case, the system can identify 'Unit Process Engineer' as the optimal intermediate node to serve as a bridge to fill the gap between the two roles.
[0291] The system can generate and provide the logical basis for the selection of these intermediate nodes as specific scenarios as follows.
[0292] Through your current equipment maintenance role, you possess an exceptional understanding of the chamber structure and hardware mechanisms of etching equipment. This capability serves as a strong differentiator for performing the role of a unit process engineer (an intermediate node in a static ontology) handling process recipes, as you are better able than anyone to grasp the impact of equipment status on process results.
[0293] Therefore, the system proposes that the primary goal is to first transition to a unit process engineer role and acquire experience in optimizing process conditions within a specific process (e.g., etching) and the ability to analyze defects.
[0294] Subsequently, based on your accumulated in-depth knowledge of unit processes, I recommend a roadmap to grow into a Process Integration (PI) Engineer (the ultimate goal), who must coordinate the interaction between upstream and downstream processes. Since the PI role cannot improve overall yield without an understanding of unit processes, your bottom-up career path will become a rare competitive advantage.
[0295] Along with the above scenario, the system can provide a sequential path leading from the Equipment (Start), Unit Process (Bridge), and Integration (Goal) roles as visualized graph data. In this case, on each edge (arrow) of the graph, the capabilities that can be carried to move to that stage and the capabilities that must be newly acquired can be distinguished and displayed in the form of tags.
[0296] For example, on the first edge (from equipment to unit processes), tags such as #EquipmentTroubleshootingCapability and #HardwareUnderstanding are activated to emphasize ease of career transition, while on the second edge (from unit processes to integration), tags such as #YieldAnalysis, #DevicePhysics, and #DataStatistics are displayed to guide users on which competencies they should focus on developing in the future.
[0297] This intermediate node exploration function can perform a practical consulting function that prevents users from applying too aggressively while risking immediate rejection, and helps them strategically reach their target job by establishing a mid-to-long-term career roadmap spanning two to three years.
[0298] The system according to the present invention can receive user-side structured career data and company-side unstructured data, and the company-side unstructured data may include not only job posting data but also unstructured conversational data based on company insider interviews.
[0299] Conventional recruitment matching systems have limitations in that they cannot obtain qualitative and in-depth information, such as the actual work environment, team atmosphere, specific tasks involved in performing the job, or the ideal candidate profile that recruiters truly desire, as they rely solely on the text of publicly available job postings as input data.
[0300] This information asymmetry reduced the accuracy of matching between users and companies and could lead to problems such as early resignation after joining or job mismatches.
[0301] Accordingly, the system can utilize unstructured conversational data collected through interviews with corporate HR managers, current employees in the relevant roles, team leaders, or hiring decision-makers as additional input data. Unstructured conversational data based on insider interviews may contain rich qualitative information, including the work environment, team composition, decision-making structure, preferred candidate profiles, and evaluation criteria, which are not specified in job postings.
[0302] The system can store unstructured conversational data based on corporate insider interviews in a database in the form of original utterance data. This original utterance data can be stored in a structured log format that includes the respondent's utterance time, content, identification information, role type, and question information corresponding to the utterance. This serves as a foundation for tracing the source of extracted corporate data fields and reconfirming the context of the original utterance during subsequent verification processes.
[0303] The system can extract corporate data fields from original utterance data through natural language processing. The extracted corporate data fields may include at least one of job requirement competencies, task information, necessary knowledge, tools used, organizational culture characteristic information, compensation condition information, evaluation criteria, and preferred talent profile.
[0304] The system can generate user nodes and job posting nodes by extracting user-side data fields from user-side structured career data and normalizing both the user-side and corporate-side data fields using standard terms defined in a static ontology. In this process, by mapping the data fields from both sides to their respective ontology nodes, the system can perform precise matching based on domain context rather than mere text similarity.
[0305] The system can calculate at least one of job fit, job transition potential, lack of competencies, organizational culture fit, compensation condition fit, and items requiring verification or grounds for recommendation, along with career path and path fit scores, based on edge exploration results between user nodes and job posting nodes on a static ontology.
[0306] Here, job fit can be a metric that quantifies how well a user's current competencies match the competencies required by the job posting. Job transition potential can be a metric that quantifies the difficulty and probability of success when a user transitions from their current role to the role in the job posting. Deficient competencies can be the set of items among the required competencies of the job posting that the user does not possess.
[0307] Organizational culture fit may be an indicator that quantifies the degree of alignment between a user's preference for organizational culture and the characteristics of the company's organizational culture. Compensation condition fit may be an indicator that quantifies the degree of alignment between a user's preference for compensation conditions and the company's compensation conditions. Items requiring verification may be those where there is a discrepancy between information in public job postings and insider interviews, or items that are difficult to judge clearly from user data, requiring additional confirmation during subsequent procedures such as interviews.
[0308] Instructions can be controlled to identify missing competencies between user nodes and job posting nodes, and to calculate and provide learning paths or intermediate job paths to supplement missing competencies based on node and edge relationships on a static ontology.
[0309] Lack of competencies should not be limited to a mere list of competencies that the user does not possess; they can provide tangible value to the user when provided with specific guidance on how to supplement them.
[0310] After mapping user nodes and announcement nodes onto a static ontology, the system can identify required knowledge, task, or tool nodes connected to announcement nodes that are not connected to user nodes as lacking capabilities.
[0311] For example, if the announcement node is connected to the 'Plasma Physics', 'ASML Scanner Operation Experience', and 'Yield Analysis' nodes, while the user node is connected only to the 'ASML Scanner Operation Experience' node, the system can identify 'Plasma Physics' and 'Yield Analysis' as lacking capabilities.
[0312] The system can generate a supplementary path for each identified deficiency competency. If the deficiency competency corresponds to a knowledge node, the system can explore learning resource nodes, course nodes, or certification nodes connected to the knowledge node on a static ontology through presupposition or utilization relationships and present them as learning paths.
[0313] For example, if 'plasma physics' is identified as a lacking competency, the system can provide online lectures, university courses, or related book information linked to the corresponding knowledge node as a learning path.
[0314] If the lack of competency corresponds to job performance experience, the system can explore intermediate job nodes on the static ontology where such experience can be naturally accumulated and present them as intermediate job paths.
[0315] For example, if the job posting node is 'Process Integration Engineer' and the user node is 'Facility Maintenance Engineer,' experience in 'Yield Analysis' may be identified as a lacking competency. In this case, the system can suggest 'Unit Process Engineer' as an intermediate job node to provide a step-by-step path to move to Process Integration Engineer after accumulating experience in yield analysis within that role.
[0316] The generation and provision of such pathways to supplement insufficient competencies can perform a practical function, helping users design their growth direction from a mid-to-long-term career development perspective, rather than simply settling for matching results at the current point in time.
[0317] Meanwhile, the above instructions can identify conflicting nodes that are opposed between the user node and the announcement node. Here, a conflicting node may refer to a node in which a node generated from a user-side data field and a node generated from a corporate-side data field are placed in a conflicting relationship with each other on the static ontology.
[0318] For example, if a user node is connected to a node that 'prefers autonomous and independent work styles,' while a job posting node extracted from corporate insider interviews is connected to a node that 'prefers work styles that strictly adhere to established processes,' the system can identify these two nodes as conflicting nodes in terms of organizational culture.
[0319] Alternatively, if a user-side compensation condition node conflicts with an enterprise-side compensation condition node by exceeding a pre-set allowable range, it can also be identified as a conflict node.
[0320] The system may classify the identified conflict nodes as items requiring verification, or provide them by including them as candidate risk factors in candidate reports or matching results.
[0321] In addition, the system can support the identification of risks related to organizational culture fit or compensation conditions that are not revealed during the documentation stage by generating interview verification questions based on the aforementioned conflict nodes.
[0322] Through this, the system can enhance the reliability and verifiability of matching results by presenting not only the basis nodes commonly connected to the user node and the announcement node, but also the points where the data from both sides conflict, in an explainable form.
[0323] Instructions can control the extraction of organizational culture preference information and compensation condition information from user-side data fields, the extraction of organizational culture characteristic information and compensation condition information from enterprise-side data fields, mapping them respectively to corresponding nodes on a static ontology, and calculating organizational culture fit and compensation condition fit to reflect in the matching results.
[0324] Conventional recruitment matching has limitations as it focuses primarily on the alignment of job competencies, leading to early attrition after joining due to mismatches in organizational culture or compensation conditions, even when job fit is high. Therefore, matching that comprehensively considers non-job-related factors may be required.
[0325] The system can extract information on users' organizational culture preferences from user-side structured career data. This information may include, for example, preference for horizontal or vertical organizational structures, degree of autonomy, collaboration methods, speed of decision-making, and emphasis on work-life balance. Additionally, the system can extract information on users' compensation conditions, such as desired salary range, preference for performance-based bonus proportions, preference for stock options or RSUs, and priority of employee benefits.
[0326] The system can extract information on corporate organizational culture characteristics from unstructured corporate data, particularly unstructured conversational data based on insider interviews. Insider interviews can be the most effective source for understanding the nuances of actual organizational culture that are not specified in job postings. For example, if a current employee states, "Our team holds open discussion meetings every week," the system can extract "horizontal discussion culture" as a characteristic of that company's organizational culture. Additionally, the system can extract information regarding corporate compensation conditions, such as salary tables, performance-based bonus systems, and employee benefits.
[0327] The system can map the extracted information on organizational culture and compensation conditions from both sides to corresponding nodes defined on a static ontology. The static ontology may pre-define standard nodes such as 'horizontal organizational structure', 'flexible working hours', and 'performance-based compensation', and the system can map the extracted information by normalizing it to these standard nodes.
[0328] The system can calculate organizational culture fit and compensation condition fit based on mapped nodes. Fit can be calculated by quantifying the degree of agreement between the set of organizational culture nodes preferred by the user and the set of organizational culture nodes possessed by the company; it can be calculated not only through a simple intersection ratio but also using a weighted average method that reflects the user's weight (importance) for each item.
[0329] The calculated organizational culture fit and compensation condition fit scores can be reflected in the matching results along with the previously calculated job fit and path fit scores. The system can calculate a comprehensive fit score by integrating these and provide it to the user terminal, or it can visualize each fit score as an individual indicator to enable the user to identify their strengths and concerns from various angles.
[0330] Instructions can control the generation and provision of a candidate report that includes at least one of a career path, path fit score, job fit, lack of competencies, organizational culture fit, compensation condition fit, and items requiring verification.
[0331] Candidate reports can serve as valuable informational assets for both users and companies. From the user's perspective, they can be utilized as self-assessment data to objectively identify their strengths, weaknesses, and areas for future improvement. From the company's perspective, they can serve as comprehensive evaluation data to review a candidate's capabilities from various angles during the hiring decision-making process.
[0332] The system can automatically generate candidate reports for each user. Candidate reports may include the following components.
[0333] The candidate report may include the user's current competency profile. This may be a list of ontology nodes connected to the user node, namely a set of job experience, knowledge, tool usage experience, and task performance experience possessed by the user.
[0334] The candidate report may include the results of a matching analysis with the recommended job posting. The matching analysis results may present job fit scores, path fit scores, organizational culture fit scores, and compensation condition fit scores in the form of quantitative figures and visualized graphs.
[0335] The candidate report may include career path visualizations. This can be data visualizing a direct path from a user node to a job posting node, or a step-by-step path via intermediate nodes, in the form of a graph.
[0336] The candidate report may include deficient competencies and supplementation guides. This may be a document presenting a list of additional competencies the user needs to acquire to become a suitable candidate for the job posting, along with learning paths or intermediate job paths to supplement them.
[0337] The candidate report may include items requiring verification. This could be a list of items where information differs between the public job posting and insider interviews, or items that are difficult to clearly determine from user information and require further confirmation during the interview process.
[0338] The system can provide the generated candidate report to user or corporate terminals in the form of PDF, HTML, or a dashboard. Additionally, by displaying the original utterance segment and basis node for each component of the candidate report, the system can transparently provide the source and basis for the report's content.
[0339] Instructions can be controlled to identify items requiring verification during the interview process based on corporate-side data fields and user-side data fields, and to generate and provide interview verification questions by referencing fields extracted from public recruitment data and fields extracted from corporate insider interviews together.
[0340] The interview can be a crucial step in verifying information obtained during the document screening stage and gaining a deeper understanding of the candidate. However, if interviewers use the same questions for all candidates, it becomes difficult to verify candidates based on their specific characteristics, and the effectiveness of the interview may be diminished.
[0341] The system can automatically generate interview verification questions by comprehensively analyzing user-side and corporate-side data fields. In doing so, by referencing both fields extracted from public recruitment data and those extracted from insider interviews, the system can generate questions that simultaneously reflect official requirements and requirements from the actual work environment.
[0342] For example, if a public job posting specifies 'experience using Python,' and an utterance is extracted from an insider interview stating that "in reality, data extraction skills using SQL are more important than Python," the system can generate interview questions to verify the user's SQL experience.
[0343] Specifically, questions such as, "It has been confirmed that you have experience using Python. This role frequently utilizes SQL during the data extraction phase; do you have experience handling complex joins or window functions using SQL?" can be automatically generated.
[0344] In addition, the system can generate verification questions to confirm the user's willingness to supplement or learning plans for items identified as areas where the user has insufficient competencies.
[0345] For example, if 'plasma physics' is identified as a lacking competency, the system can generate a question such as, "Understanding plasma processes is essential for this role. Do you have a study plan or experience with self-study in this field?"
[0346] The generated interview verification questions can be provided to the interviewer's terminal and utilized as a tool to simultaneously ensure consistency and depth in the interview, and can also be provided to the user's terminal in advance to serve as a guide for the user to prepare for the interview.
[0347] Instructions store corporate data fields extracted from unstructured conversational data based on corporate insider interviews by linking them to corresponding original utterance segments, assign different weights to corporate data fields according to the respondent role type of the corporate insider interview, wherein the role type includes at least one of HR manager, current employee, team leader, and hiring decision maker, and control can classify mutually inconsistent items as items requiring verification by comparing fields extracted from public job posting data with fields extracted from corporate insider interviews.
[0348] When the system extracts corporate data fields from corporate insider interviews, it can store not only the extracted values but also the original utterance segments that served as the basis for those values in conjunction. Specifically, the system can store utterance segment identifiers identifying specific sentences or phrases of the original utterance, the time of utterance, respondent identification information, and the respondent role type as metadata for the corporate data fields.
[0349] Storing the connection with these original utterance segments can serve as a foundation for tracing the source of data extracted in subsequent stages, utilizing the context of the original utterance when generating interview verification questions, or comparing and verifying evidence when conflicts arise between data.
[0350] The system can assign different weights to corporate data fields extracted from insider interviews based on the respondent's role type. This is because even the same information can have different reliability and influence depending on who speaks it.
[0351] Specifically, data fields extracted from the utterances of HR managers may be assigned high weights for items related to HR administration, such as recruitment procedures, formal requirements, and compensation policies. Data fields extracted from the utterances of current employees may be assigned high weights for items related to practical work, such as actual job duties, tools used, and the daily work environment.
[0352] Data fields extracted from team leaders' utterances can be assigned high weights to items related to team operations, such as team composition, collaboration methods, evaluation criteria, and team culture. Data fields extracted from hiring decision-makers' utterances can be assigned high weights to strategic items, such as talent profiles, hiring priorities, and long-term vision.
[0353] The system can identify discrepancies by comparing fields extracted from public recruitment data with those extracted from insider interviews. Discrepancies in information between recruitment announcements and the actual work environment can occur frequently.
[0354] For example, if a public job posting specifies "small team of 5 or fewer people," but a utterance is extracted from an interview with an employee stating "actually, I work in a team of 10 people," the system can classify the team size item as an item requiring verification.
[0355] In addition, if the public job posting specifies that "working from home is possible," but the utterance "working four days a week is the principle" is extracted during the team leader interview, the system may classify the work type item as an item requiring verification.
[0356] Items requiring verification classified in this way can be shared with users in advance so that they can be guided to check directly during the interview, and can also be utilized to enhance the effectiveness of the interview by being prioritized when generating interview verification questions.
[0357] The instructions can control that the static ontology is managed through a version control system, stores version information of the static ontology, and applies the updated version of the static ontology to the mapping of user nodes and announcement nodes.
[0358] The knowledge system of an industrial domain can constantly change over time. New technologies emerge, existing job functions are subdivided or integrated, and new tools or equipment may be introduced. Therefore, a method of constructing a static ontology at a single point in time and utilizing it as is may have limitations in that it fails to reflect the actual structure of the domain as time passes.
[0359] Accordingly, the system can systematically manage static ontologies through a version control system. The system assigns new version information whenever a static ontology is updated and can preserve previous versions of the static ontology in a separate memory area. The version information may include a version number, the time of the update, a summary of the update details, and information regarding the person responsible for the update.
[0360] By applying the most up-to-date version of the static ontology when mapping new user nodes or announcement nodes, the system can perform matching that reflects the knowledge system of the industry domain, which changes in real time. At the same time, by preserving previous versions of the static ontology, the system can guarantee the reproducibility and consistency of matching results performed in the past.
[0361] For example, if a new process technology called the 'GAA (Gate-All-Around) transistor' emerges in the semiconductor industry, the system can add the technology as a new node, connect it hierarchically with the existing 'transistor structure' node, and save it as a new version of the static ontology. Subsequently, new user nodes or announcement nodes are mapped based on this updated ontology, enabling accurate matching related to GAA technology.
[0362] Furthermore, if the application of an updated static ontology may alter the matching results of existing mapped user or announcement nodes, the system can update the results by remapping those nodes. These versioning and remapping functions can serve as a core mechanism enabling the static ontology to flexibly accommodate industrial changes as a living knowledge system.
[0363] FIG. 11 is a flowchart for generating and saving a company-customized report in a customized self-introduction letter generation system according to one embodiment of the present invention.
[0364] Referring to FIG. 11, the instructions can be controlled to collect external documents related to a support company (S140) using a support company input (S130) from a user terminal as a key value when executed by a processor, and to generate and save a company-specific report (S150) by text mining company analysis information and job requirement information from the external documents.
[0365] That is, the user can select a company and a job function (S130) and input information into the user terminal. The system can collect external documents of the supported company through the information input by the user into the user terminal (S140). After that, the system can generate and save a company-specific report (S150).
[0366] Here, the company-specific report is data that summarizes the current situation of the company the user intends to apply to and the core competencies required for the relevant job, and can be used as a basis for writing during the self-introduction letter creation process.
[0367] When the system collects external documents (S140), it can crawl (S141) the company's public disclosure materials, such as semi-annual reports or annual reports, and job descriptions (JDs) posted on recruitment sites.
[0368] Specifically, the system can secure objective and macroscopic information, such as a company's financial status, major projects, and R&D status, by accessing the Financial Supervisory Service's Data Acquisition and Relations System (DART) or a company's IR page to collect the most recently issued business reports.
[0369] At the same time, the system can crawl major recruitment platforms such as JobKorea, Saramin, and LinkedIn, or corporate recruitment websites, to collect specific job descriptions and qualification requirements for the relevant job postings. The instructions can generate and save a company-specific report (S150) using the crawled information.
[0370] The instructions can extract meaningful information by performing natural language processing (NLP)-based text mining on the collected text data (S142). The system can extract key keywords that the company is currently focusing on (e.g., transition to eco-friendly materials, adoption of AI, expansion into global markets, etc.) by focusing on analyzing the chapters 'Business Overview' and 'Directors' Management Diagnosis and Analysis Opinions' from the vast amount of text within the business report.
[0371] Furthermore, instructions can be stored as a structured data set by mapping the business content of the semi-annual report with the qualifications or preferred requirements of the job description. This can be a process of identifying the link between the company's macroscopic management goals and the microscopic competencies required of the job performer.
[0372] For example, if the system extracts the business objective "expansion of market share for next-generation automotive semiconductors" from a business report and the qualification requirement "MCU design experience preferred" from a job description, the system can map these two to generate a structured data pair stating, "The business objective is expansion of automotive semiconductors, and the required capability is MCU design."
[0373] Such a structured data set can be used as foundational data to automatically generate logical motivations for application when writing a cover letter, such as "I will utilize my MCU design experience (job competency) to contribute to your company's goal of expanding the automotive semiconductor market (business content)."
[0374] Furthermore, when collecting the aforementioned external documents (S140), the instructions may further perform a process (S143) of researching social and economic issues related to the aforementioned supporting company through a web search engine or news aggregator. This allows for the supplementation of real-time market trends or external risks of the company that are difficult to grasp solely from static disclosure materials (business reports, etc.) that are updated quarterly.
[0375] For example, the system can crawl web documents such as 'latest news,' 'press releases,' and 'industry trend reports' using the name of the applicant company as a keyword. In this process, the system can focus on extracting keywords related to social issues, such as the company's ESG (Environmental, Social, and Governance) management activities, labor-management relations, or community contributions. Additionally, regarding economic issues, it can analyze the impact of macroeconomic factors on the company, such as the effects of exchange rate fluctuations on imports and exports, changes in profitability due to rising raw material prices, or the entry of competitors into the market.
[0376] The socio-economic issue data collected in this way can be classified and stored as opportunity or threat factors in customized corporate reports. This can then be utilized as a logical basis for writing timely and insightful motivations when generating a cover letter, such as, "I was deeply impressed by your company's efforts to secure a stable supply chain even amidst the rapidly changing exchange rate market (economic issue)."
[0377] FIG. 12 is a flowchart for extracting and storing second experience data of a customized self-introduction letter generation system according to one embodiment of the present invention.
[0378] Referring to FIG. 12, when the instructions are executed by the processor, they can embed (S160) first experience data classified into user situation, task, action, result, and reflection stored in the database, and calculate a similarity score (S170) with the previously analyzed job requirement information.
[0379] Here, the first experience data refers to the archive of all experiences the user has created over time, and the system can find materials perfectly suited for the current job posting from this vast data pool.
[0380] The system can utilize natural language processing models (e.g., SBERT, BERT, etc.) to perform an embedding process that converts primary experience data in text and job requirement information into a high-dimensional vector space. At this time, it is possible to quantify contextual meaning rather than simply checking for word match.
[0381] For example, if the job requirement information is "data-driven decision-making ability" and the user experience data contains the content "analyzed sales trends using Excel functions," the two texts can have semantically similar vector values, even though their superficial words are different.
[0382] The system can calculate a similarity score (S170) by calculating the cosine similarity between two vectors, and then determine (S180) whether this score is greater than or equal to a preset threshold (e.g., 0.75). If the similarity score is greater than or equal to the preset threshold, the corresponding first experience data can be extracted and stored as second experience data (S190). If the similarity score is less than the preset threshold, the first experience data can be embedded again (S160), and the similarity score between the embedded first experience data and the job requirement information can be calculated (S170), and the step of determining whether the similarity score is greater than or equal to the preset threshold (S180) can be repeated.
[0383] Here, secondary experience data may refer to key material from the entire experience that is determined to be the most suitable for the target company and job.
[0384] For example, if an applicant applies for a 'marketing position,' the system can calculate a higher similarity score for "experience increasing sales by rearranging display locations" rather than "simple cashiering tasks" within the 'convenience store part-time job experience (1st experience data),' and extract this as 2nd experience data. This screening process can be an essential step to ensure that the self-introduction letter does not list irrelevant experiences but instead precisely targets the job competencies that the recruiter is interested in. The extracted 2nd experience data can be used as the core framework for answers to each question during the self-introduction letter generation stage.
[0385] FIG. 13 is a flowchart for completing a draft of a self-introduction letter in a customized self-introduction letter generation system according to one embodiment of the present invention. FIG. 14 is a diagram showing the self-introduction letter questions of a supporting company in a customized self-introduction letter generation system according to one embodiment of the present invention.
[0386] Referring to FIGS. 13 and 14, the instructions can classify the self-introduction questions of the applicant company into one of a plurality of predefined question types (S200), and then classify the questions of the successful self-introductions stored in the database (S200). Afterwards, the instructions can control the learning of a standardized writing logic (S220) by analyzing the paragraphs among the successful self-introductions that match the question types (S210).
[0387] At this time, when classifying the question types (S200), the instructions can analyze the intent of the self-introduction questions of the applicant company using natural language processing and match them to one of the pre-set categories including at least one of the application motivation type, job competency type, challenge experience type, collaboration type, and post-employment aspiration type.
[0388] The motivation type may be a type that asks for clear reasons why the user chose the industry and company, and why they want to perform the specific job.
[0389] The job competency type may be a type that requires the user to demonstrate the knowledge, skills, and attitudes possessed to perform the relevant job through specific experiences.
[0390] The Challenge Experience type may be a type that evaluates the user's passion and problem-solving skills through the experience of setting high goals for oneself, overcoming obstacles, and achieving them.
[0391] The collaborative type may be a type that evaluates communication skills and teamwork through experiences of resolving conflicts arising in team projects or organizational life, or achieving common goals by cooperating with others.
[0392] The post-employment aspiration type may be a question asking for specific plans on how to grow and contribute to the company in the short and long term after joining.
[0393] The question types presented herein are categorized from questions frequently required in the recruitment market, but the present invention is not limited thereto and may additionally include various types such as 'strengths and weaknesses of personality', 'values', and 'views on social issues'.
[0394] The system can extract cases with similar patterns from the database of successful self-introduction letters according to classified question types, and learn standardized writing logic (S220) by analyzing their sentence structures.
[0395] At this time, the instructions analyze the sentences of the successful personal statement and learn based on the STARR framework, which includes Situation, Task, Action, Result, and Reflection, and can apply this as a template when generating a user-customized personal statement draft (S250).
[0396] For example, if classified as a challenge experience type question, the system can template the logic of a successful personal statement in which the sentences unfold in the order of "Past difficult situation (S) - Task faced (T) - Specific actions taken to overcome it (A) - Resulting outcome (R) - Reflections (R)".
[0397] Subsequently, instructions can be controlled to generate a user-customized self-introduction draft (S250) by performing prompt engineering that uses the above-mentioned learned writing logic as a framework, inserts the previously extracted second experience data as content (S230), and combines the company-customized report (company analysis information) as the basis for writing (S240).
[0398] Specifically, the system can input the following structured prompts to a large-scale language model (LLM): "You are a professional employment consultant. (1) Write according to the STARR structure of a successful 'challenge experience-type' self-introduction. (2) Use the user's 'secondary experience data (e.g., experience of completing a marathon)' as material. (3) At this time, use 'your company's ideal tenacious talent' found in the company report as a connecting link." The draft generated through this process can be a result that possesses both logical completeness and corporate suitability, rather than simply generating sentences.
[0399] Since recruiters must review numerous documents, eye-catching subheadings such as "Perseverance learned from completing a marathon" can be an important factor in improving readability. Therefore, instructions can generate a user-customized self-introduction draft (S250), then automatically generate a subheading (S251) by summarizing the key content of the second experience data and place it at the top.
[0400] In addition, the system can recognize the character limit (e.g., 500 characters, 1000 characters) set in the self-introduction questions of the applicant company and automatically adjust the length of the generated draft (S252). If the content is long, it can be summarized focusing on the key points, and if it is short, additional explanations can be added to optimize it to within ±10% of the character limit.
[0401] Finally, the instructions can humanize the draft of the self-introduction into natural sentences by applying conjunctions and endings frequently used in successful self-introductions (S253). Text written by artificial intelligence can often feel stiff or like a translation. Accordingly, the system can complete the draft of the self-introduction in a style that appears as if it was written with careful consideration by a human, by placing appropriate conjunctions such as "therefore," "consequently," and "however," and processing natural endings such as "~did" and "~think" (S260).
[0402] Instructions can classify the personal statement questions of the target company into motivation, job competency, challenge experience, collaboration, and future aspiration types, and then dynamically assign weights indicating the character count or emphasis weight for the detailed items of the STARR framework (situation, task, action, result, reflection) to match the intent of each question type.
[0403] According to the analysis of successful personal statements, since the factors recruiters focus on evaluating differ depending on the nature of the questions, the system can control the application of differential weighting for each item within the total text based on type.
[0404] Specifically, when the question type is classified as 'motivation for application,' the system may assign high weight to Reflection and Task items to emphasize that the user's values align with the company's vision. As an example of the process, the system may structure the logic with a flow of "Reason for becoming interested in the industry (S) - Recognition of key market challenges facing the company (T) - Efforts made to join the company (A) - Results of preparation (R) - Reasons why my values align with the company's vision and justification for joining (Reflection)," and input a prompt into a large-scale language model that allocates more than 40% of the total amount to Reflection.
[0405] If the item type is classified as 'Job Competency Type,' the system may assign the highest weight to Action and Result items to demonstrate the user's practical applicability and expertise.
[0406] As an example of a process, the system can configure a flow of "Background of past projects (S) - Technical or practical tasks to be solved (T) - Specific problem-solving processes and tool utilization using job knowledge (A) - Quantitative performance indicators (R) - Confidence in job performance (Reflection)".
[0407] At this time, the system can allocate more than 60% of the total volume to actions (A) and results (R), thereby generating a draft that highlights specific job skills and quantifiable performance rather than simple appreciation.
[0408] If the item type is classified as 'Challenge Experience Type,' the system may assign high weight to Task and Action items to highlight the process of overcoming the difficulty and limitations of the goal.
[0409] As an example of a process, the system can configure a flow of "Existing limitations (S) - Difficulty of self-set challenging goals (T) - Differentiated efforts attempted despite the risk of failure (A) - Final outcome (R) - Insights gained through the challenge (Reflection)." The system can be controlled to deepen the text generation regarding the reason for setting a task (T) that others avoid and the persistent actions (A) taken to overcome it.
[0410] If the item type is classified as 'collaboration type,' the system can assign weights to Situation and Action items to evaluate communication skills and teamwork in conflict situations.
[0411] As an example of a process, the system can structure a flow of "conflict of opinion or crisis situation arising during a team project (S) - Common goal the team must achieve (T) - Specific communication methods used to mediate conflict and divide roles (A) - Creation of synergy and achievement of goals through teamwork (R) - Realization of the value of collaboration within the organization (Reflection)." The system can focus the prompt on the description of actions (A) that persuaded others and elicited cooperation, rather than on individual unilateral achievements.
[0412] If the question type is classified as 'post-employment aspirations,' the system can substitute past experiences with future plans and assign high weight to Task and Action items. The prompt can be adjusted to present a roadmap for post-employment rather than a past-experience-based STARR.
[0413] As an example of the process, the system can generate a draft that emphasizes a specific master plan (A) for the job by configuring the logic in the flow of “Current situation and market environment of the company analyzed through a customized report (S) - Detailed job goals to contribute to after joining (T) - Specific implementation plans and competency development methods for each year, such as the 1st, 3rd, and 5th years (A) - Expected effects on the growth of the company (R) - Long-term commitment (Reflection)”.
[0414] By applying STARR weight profiles, which are set differently for each question type, as parameters for prompt engineering, the system can control large-scale language models to go beyond simply writing contextually appropriate text and generate structured text that accurately matches the hiring manager's evaluation criteria and intentions.
[0415] FIG. 15 is a flowchart of the editing of a draft self-introduction letter in a customized self-introduction letter generation system according to one embodiment of the present invention.
[0416] Referring to FIG. 15, the instructions can generate and present a question requesting additional information from the user after the user-customized self-introduction draft is completed (S260) (S270), and determine whether the user's response has been received (S280).
[0417] When the user's response to a question is received, the system can control the rewriting of the corresponding paragraph (S290) by converting the response to fit the context. This interaction may be intended to supplement the generated draft when it has an overall logical structure but lacks specific details known only to the user, such as specific figures or proper nouns. After the paragraph rewriting (S290), the system can save the revised version of the rewritten self-introduction draft (S300).
[0418] The system can analyze each paragraph of a draft generated through a natural language processing model to identify sentences lacking specificity or arguments with weak evidence.
[0419] For example, if a draft includes the sentence "I significantly increased work efficiency at my previous job," the system may determine that the expression "significantly" is ambiguous. Accordingly, the system can generate customized questions for the user, such as "You mentioned that you increased work efficiency; specifically, what tools did you use? Or could you tell me in numbers how much (%) your work time was reduced?" and present them through a chat interface or a pop-up window.
[0420] Even if a user enters a colloquial or short answer to this question, such as "I used an Excel macro, and the time was reduced from one hour to ten minutes," the system can convert it to fit the context. The system does not simply paste the input text, but can reprocess it by taking into account the formal writing style of a self-introduction letter and the flow of the preceding and succeeding sentences.
[0421] The system can rewrite the paragraph as follows (S290) by reflecting the user's response: "I established an automated process by introducing Excel macros to repetitive tasks that previously took 1 hour. As a result, I was able to reduce the task processing time to 10 minutes and achieve an improvement in work efficiency of approximately 83%." The rewritten paragraph in this way incorporates the user's unique experiences in detail, which can increase the credibility and persuasiveness of the self-introduction. If the system does not receive the user's response within a preset time, it can maintain the content of the previously completed draft of the self-introduction and save it as the final version (S310).
[0422] Instructions can group user-customized cover letter drafts, company-customized reports, and second experience data by supported company and store them in a database, and control the provision of user-customized cover letter drafts, company-customized reports, and second experience data in a time series upon a user's query request.
[0423] Since the job preparation process is not a one-off event but a long-term race involving applying to dozens of companies and experiencing repeated rejections and acceptances, the ability to systematically manage application history may be required.
[0424] The system can archive three key data sets (draft of cover letter, company analysis report, and key experience materials used) generated whenever a user applies to a specific company, grouped into a single package, at the 'Applied Company Folder' level.
[0425] Specifically, by storing customized reports for each company, the system allows users to review which specific issues or talent profiles they focused on when formulating their strategies. Additionally, by storing secondary experience data, it enables users to identify which specific experiences from their extensive backgrounds were effective for the relevant job role.
[0426] When the user passes the document screening and prepares for an interview, the system can retrieve and provide this stored data in a time series. If the user has a history of applying to the company in the past, the system can provide a timeline that allows the user to grasp changes in application strategy at a glance, such as, "When applying in the first half of 2023, I emphasized experience A with the keyword 'challenge' but was rejected, and this time in the second half of 2024, I wrote by emphasizing experience B with the keyword 'collaboration'."
[0427] These company-specific data grouping and time-series provision features can serve as a powerful interview guide, helping users thoroughly understand the logic and evidence they wrote and provide consistent answers to anticipated interviewer questions (e.g., "What preparations did you make to apply to our company?", "Please elaborate on the experience you wrote about in your cover letter").
[0428] The system according to the present invention can receive structured career data of a user. Here, the user's structured career data may refer to experience data structured into a sophisticated data set through the experience organization system for customized job search described above.
[0429] The system can minimize the omission or distortion of information that may occur during the self-introduction generation process by utilizing structured data, which is already organized into situations, tasks, actions, results, and reflections according to the STARR framework, as input, rather than receiving unstructured text directly from the user terminal.
[0430] In addition, the system eliminates the inconvenience of users having to repeatedly input the same experiences and provides an efficient work environment where experience data, once organized, can be reused by multiple support companies.
[0431] When collecting external documents, the system can collect not only job posting data and job descriptions but also unstructured conversational data based on corporate insider interviews.
[0432] Unstructured conversational data based on corporate insider interviews is qualitative data collected through interviews with HR managers, current employees, team leaders, or hiring decision-makers, and may include actual work environments, team culture, preferred candidate profiles, evaluation criteria, etc., which are not specified in publicly available job postings.
[0433] The system can extract corporate analysis information and job requirements information from unstructured conversational data based on corporate insider interviews through natural language processing, and generate customized reports for the company by integrating this with information extracted from job postings and job descriptions.
[0434] This multifaceted information integration goes beyond simple keyword matching and can serve as a foundation for comprehensively understanding the type of talent a company actually seeks. It can also be utilized to express more authentic and differentiated motivations for application and job suitability when creating a cover letter.
[0435] The instructions can control the dynamic assignment of weights representing the character count or emphasis weight for each item—situation, task, action, result, and reflection—depending on the question type, and generate a user-customized personal statement draft based on the assigned weights.
[0436] Conventional methods of writing personal statements had limitations in that they applied the same STARR weight to all questions, resulting in answers with a uniform structure regardless of the question's intent. However, in the actual recruitment market, the factors that recruiters focus on evaluating vary depending on the nature of the questions, so a differentiated weight distribution for each question may be required.
[0437] The system can store a predefined weight profile in memory for each of the previously classified item types. The weight profile may be a set of numerical values representing the STARR items to be emphasized in the corresponding item type, the proportion each item should occupy in the total volume, and the level of emphasis.
[0438] Specifically, if the question type is classified as motivational, the system may assign the highest weight to the context and reflection items. This is because motivational questions focus on evaluating the alignment between the user's values and the corporate vision.
[0439] For example, the system can apply a weighted profile that allocates more than 40% of the total volume to reflection items and about 30% to situation items.
[0440] If the item type is classified as job competency, the system may assign the highest weight to behavior and result items. This is because job competency items focus on evaluating the user's practical performance abilities and quantitative results.
[0441] For example, the system can apply a weighted profile by allocating more than 35% to behavior items and more than 30% to result items to emphasize specific job skills and quantifiable performance.
[0442] If the item type is classified as a post-employment aspiration type, the system assigns high weight to task and behavior items, but can adjust the weight profile so that past experiences are replaced with future plans.
[0443] For example, the system can apply weighted profiles by mapping short-term and medium-to-long-term action plans after hiring to action items, and mapping the analysis of the company's current business environment to situation items.
[0444] The system can convert applied weight profiles into prompt engineering parameters and input them into a large-scale language model. In this process, the prompts may include the volume ratio assigned to each STARR item, the emphasis level, and instructions regarding the key content to be covered in that item. Through this, the system can go beyond simply generating contextually appropriate text and produce structured text that precisely matches the intent of the item.
[0445] In addition, the system does not statically fix the weight profile but can dynamically adjust it according to the characteristics of the user's experience data or the requirements of the supporting company.
[0446] For example, if the user's second experience data contains a wealth of quantitative performance indicators, the system can further increase the weight of the result item to maximize those strengths. Conversely, if the user's experience has a strong qualitative growth narrative, the system can increase the weight of the reflection item to reveal the user's differentiated strengths.
[0447] The instructions can analyze or evaluate at least one of item suitability, job suitability, corporate suitability, experience specificity, performance clarity, logical consistency, factual consistency with original experience, stylistic naturalness, and interview risk for the generated user-customized self-introduction draft, and control the detection and correction of expressions judged to be AI-generated style.
[0448] When generating personal statements using generative AI, there is a risk that, although they may appear natural on the surface, they may deviate from the user's actual experiences or be immediately detected by recruiters due to the standardized writing style characteristic of AI, potentially leading to a negative evaluation. Therefore, multi-faceted quality evaluation and correction of the generated draft may be required.
[0449] The system can generate the following multiple evaluation indicators immediately after a user-customized personal statement draft is created.
[0450] Question suitability can be a quantitative indicator of how well the generated draft aligns with the requirements of the applicant's self-introduction questions. The system analyzes the core keywords and intent of the questions and evaluates whether those keywords and intents are faithfully addressed in the draft.
[0451] Job fit can be a quantitative indicator of how well the experience and competencies described in the draft align with the requirements of the applied position. The system can calculate a score by determining the semantic similarity between the job requirements information included in the customized report and the content of the draft.
[0452] Corporate tailoring can be an indicator evaluating whether the draft contains content specialized for the target company, rather than being a generic cover letter. The system can assess the frequency and depth of phrases within the draft that directly mention or link to the company's business activities, ideal candidate profile, or job requirements.
[0453] Experiential specificity can be an indicator evaluating whether the experience described in the draft does not remain an abstract description but includes specific facts and context. The system can analyze the density of specific information within the draft, such as time, place, affiliation, tools, and methodology.
[0454] Performance clarity may be an indicator that evaluates whether the results described in the draft include quantitative figures or clear comparison criteria. The system can assess the presence and reliability of quantitative data, such as numbers, ratios, time reductions, and personnel changes, within the result descriptions in the draft.
[0455] Logical consistency can be an indicator evaluating whether the overall structure of the draft unfolds logically without contradictions. The system can analyze whether causal relationships between situations, tasks, actions, results, and reflections are clearly connected.
[0456] Factual consistency with original experience can be an indicator evaluating whether the content of the draft factually matches the original experience stored in the user's structured career data. The system can verify whether the draft contains information arbitrarily generated or exaggerated by artificial intelligence by comparing factual information extracted from the draft with the corresponding fields in the original experience data. This can be a key function for preventing damage to credibility caused by factual distortion during the recruitment process.
[0457] Stylistic naturalness can be an indicator that evaluates whether the writing style of a draft reads as naturally as if it were written by a human. The system can calculate a score by detecting patterns that frequently appear in AI-generated text, such as standardized expressions, overly formal writing styles, or the use of unnatural conjunctions.
[0458] Interview risk may be an indicator identifying parts of the draft that could be difficult to answer if verified through additional questions during the interview stage. The system can automatically identify ambiguous descriptions, exaggerated expressions, or claims that are difficult to verify within the draft and provide advance warnings to the user.
[0459] The system can perform automatic correction on specific parts if an item is identified that shows a score below a certain threshold among the aforementioned evaluation indicators. In particular, if the stylistic naturalness indicator is calculated to be low, the system can perform a correction operation to detect expressions judged to be AI-generated styles and replace them with more natural expressions.
[0460] Expressions identified as AI-generated writing styles may include, for example, standardized phrases such as "it was a very impressive experience," "when viewed from various perspectives," or "inferring from these points," patterns of using an excessive number of adverbs within a single sentence, or patterns of repeating the same sentence structure.
[0461] The system can automatically identify these expressions by utilizing a pre-trained artificial intelligence style detection model and generate alternative expressions by referencing natural expression patterns extracted from a database of successful personal statements.
[0462] Instructions can interactively receive a user's request for modification, correction, analysis, or evaluation of a user-customized personal statement draft, convert said request into at least one of stylistic conditions, emphasis conditions, character count conditions, risk avoidance conditions, or evaluation criteria conditions, rewrite the relevant paragraph or the entire draft based on the converted conditions, and control saving versions before and after modification.
[0463] Writing a personal statement is not a one-off output, but rather a process of iterative refinement tailored to the user's intentions and market demands. Therefore, if the generated draft does not precisely meet the user's expectations, an interactive interface may be required that allows the user to intuitively and efficiently request revisions.
[0464] After a user-customized draft of a self-introduction is generated, the system can provide a means to input requests for modification, correction, analysis, and evaluation via a chat interface on the user's terminal. Users can input their intentions using free natural language expression, and the system can analyze the user's input through natural language processing and convert it into specific conditions.
[0465] Stylistic conditions may be applied when a user wishes to change the writing style of a draft. For example, if a user inputs "write it a little softer" or "change it to a professional tone," the system can convert this into a stylistic condition to reconstruct the vocabulary selection, ending processing, and sentence structure of the corresponding paragraph.
[0466] Emphasis conditions can be applied when a user wants to highlight or weaken specific content.
[0467] For example, if a user inputs "emphasize the leadership part more" or "highlight the collaboration experience more," the system can convert this into an emphasis condition and expand the volume and details of the description related to the corresponding keyword.
[0468] The character count condition may be a condition applied when the user wants to adjust the length of the draft.
[0469] For example, if a user inputs "reduce by 100 characters" or "match to 1,000 characters," the system converts this into a character count condition and can precisely adjust the length while maintaining the core content.
[0470] Risk avoidance conditions may be applied when a user wishes to remove parts of a draft that carry high interview risk or are difficult to verify. For example, if a user inputs, "Remove expressions that could cause difficulties during an interview" or "Replace difficult-to-verify content with different wording," the system can convert this into a risk avoidance condition and replace the relevant expressions with safer and more verifiable ones.
[0471] Evaluation criteria conditions may be conditions applied when a user wishes to restructure a draft to meet specific evaluation criteria.
[0472] For example, if a user inputs "increase performance clarity" or "rewrite to increase the corporate customization score," the system can convert this into evaluation criteria conditions and rewrite the draft to prioritize meeting those evaluation metrics.
[0473] The system can rewrite the relevant paragraph or the entire draft based on the converted conditions. During rewriting, the system can maintain overall consistency by precisely modifying only the parts that require change while preserving the overall context of the original draft. Additionally, the system can provide an interface that allows for a visual comparison of the differences before and after the rewrite, enabling users to intuitively verify the changes.
[0474] The system can store all versions, both before and after modification, in the database. Version information can be stored in a structured form that includes the modification time, the modification request details, applied conditions, and the text before and after the modification. This allows users to immediately roll back to a previous version and provides a working environment where they can compare multiple versions to select the optimal presentation.
[0475] In addition, the system can respond to user requests for analysis or evaluation through the same interactive interface.
[0476] For example, if a user inputs "Analyze the strengths and weaknesses of this draft" or "Tell me expected interview questions," the system can organize and provide the strengths and weaknesses based on the draft's evaluation metrics, or analyze the content of the draft to automatically generate and provide expected interview questions.
[0477] Instructions can control updates to reflect additional information collected from the user's response to a request for modification, correction, analysis, or evaluation in the user's structured career data.
[0478] The process of writing and revising a personal statement can serve as an opportunity for users to re-examine their experiences from various angles and recall previously unorganized details. The additional information collected during this process should not be merely used for the specific personal statement and then discarded; rather, it may have value as structured career data that can be reused for various subsequent services.
[0479] The system analyzes all response data entered by the user during the self-introduction editing process to identify new information not included in existing structured career data.
[0480] For example, if the system presents an additional question such as "Please let us know if there are any specific tools used in the project," and the user responds, "I used Excel macros and the Python pandas library," the system can extract tool information such as 'Excel macros' and 'Python pandas' from this response and automatically add it to the behavior data field of the corresponding experience data.
[0481] In addition, if a user enters correction information during the correction process, such as "Actually, this project was conducted for 6 months, not 3 months," the system can update the context data field of the corresponding experience data with the corrected value.
[0482] Furthermore, if an experience newly mentioned by the user during the editing process is distinct and independent from existing experience data, the system can add it to the database as a new experience data item. This can have the effect of enabling users to naturally expand their experience archives through the writing of their self-introduction letters.
[0483] For each newly extracted or updated piece of information, the system may store metadata specifying its source. This metadata may include the time at which the information was extracted, identifying information from the extracted self-introduction, and the question or request that triggered the information.
[0484] This allows users to track the context in which their structured career data was updated and maintains control to review or revert those updates if necessary.
[0485] The function of reflecting structured career data with additional information collected during this modification process can serve as a core mechanism that enables the personal statement generation system to operate as a career data infrastructure that continuously accumulates and refines users' experience assets, going beyond being a tool for providing a single output.
[0486] As described above, although the present invention has been explained by limited embodiments and drawings, the present invention is not limited thereto, and it is obvious that various modifications and variations are possible within the scope of the technical spirit of the present invention and the equivalent scope of the claims described below by those skilled in the art to which the present invention belongs.
Claims
Claim 1 In a customized personal statement generation system, a memory for storing instructions; The system includes a processor, and when the instructions are executed by the processor, the system receives the user's structured career data and, using the applicant company input from the user terminal as a key value, collects external documents related to the applicant company and unstructured data based on company-side interviews, wherein the external documents include at least one of recruitment announcement data and job descriptions, generates and stores a company-customized report by text mining company analysis information and job requirement information from the external documents and the unstructured data, embeds first experience data classified by the user's situation, task, behavior, and result stored in a database, calculates a similarity score with the job requirement information, extracts and stores the first experience data as second experience data if the similarity score is greater than or equal to a preset threshold, classifies the applicant company's self-introduction questions into one of a plurality of predefined question types, learns a standardized writing logic by analyzing paragraphs matching the question type among reference documents including successful self-introductions stored in the database, uses the writing logic as a framework, inserts the second experience data, and the company A system characterized by controlling the generation of a user-customized self-introduction draft by performing prompt engineering that combines customized reports as a basis for writing. Claim 2 A system according to claim 1, wherein the instructions, when classifying the question types, analyze the intent of the self-introduction questions of the applicant company and match them to one of a pre-set category including at least one of the application motivation type, job competency type, and post-employment aspiration type. Claim 3 A system according to claim 1, characterized in that the instructions analyze the sentences of the reference document when learning the standardized writing logic, learn based on situations, tasks, actions, results, and reflections, and apply this as a template when generating the user-customized self-introduction draft. Claim 4 A system characterized in that, in claim 3, the instructions dynamically assign weights representing the number of characters or emphasis weights for each item of the situation, task, action, result, and reflection according to the item type, and control the generation of the user-customized self-introduction draft based on the assigned weights. Claim 5 A system according to claim 1, wherein the instructions analyze or evaluate at least one of item suitability, job suitability, corporate suitability, experience specificity, performance clarity, logical consistency, factual consistency with original experience, stylistic naturalness, and interview risk for the generated user-customized self-introduction draft, and control to detect and correct expressions judged to be AI-generated styles. Claim 6 A system according to claim 1, wherein the instructions interactively receive a user's request for modification, correction, analysis, or evaluation of the user-customized self-introduction draft, convert the request into at least one of a stylistic condition, an emphasis condition, a character count condition, a risk avoidance condition, or an evaluation criterion condition, rewrite the corresponding paragraph or the entire draft based on the converted condition, and control the saving of versions before and after modification. Claim 7 A system according to claim 6, characterized in that the instructions control updating the user's structured career data by reflecting additional information collected from the user's response to the request for modification, correction, analysis, or evaluation. Claim 8 A system according to claim 1, wherein the instructions group the user-customized self-introduction draft, the company-customized report, and the second experience data by the supporting company, and, upon a user's inquiry request, provide the user-customized self-introduction draft, the company-customized report, and the second experience data in a time series.
Citation Information
Patent Citations
Personalized self-introduction coaching service method and system using big data and artificial intelligence
KR1020210023344A
System for providing artificial intelligence based coverletter analysis service
KR1020230119416A
Artificial intelligence-based system that automatically creates recruitment channels, application forms, self-introductions, and announcements to recruit talent with organizational fit and job suitability suited to customer companies
KR1020250120076A
Metaverse Based Self Introduction Writing System Using Artificial Intelligence
KR102714266B1
Method for providing self introduction document using large language models and system using the same
KR102732264B1
