Method and system for competence assessment based on multi-agent evidence directional matching

Abstract

The present application belongs to the technical field of human resource intelligence, and particularly relates to a competence evaluation method and system based on multi-agent evidence directional matching, comprising: using a preset analysis algorithm to analyze candidate information into a set of evidence segments; recalling matching rules corresponding to each evidence segment from a preset rule base; based on a plurality of preset checking agents and corresponding matching rules, performing multidirectional checking, cross verifying a plurality of checking results, and obtaining a competence evaluation result; and based on a preset experience memory enhancement algorithm and the competence evaluation result, performing updating. The present application combines candidate information analysis, rule recalling, multi-checking agent cooperation, cross verification and experience memory enhancement to realize intelligent evaluation of candidate competence, and through directional matching of evidence segments and ability dimensions and a scoring mode taking a hierarchical evaluation basis as a reference context, avoids word waste caused by enumerative evaluation, and reduces evaluation cost.

Description

Technical Field

[0001] This invention belongs to the field of intelligent human resources technology in electronic digital data processing, specifically relating to a competency assessment method and system based on multi-agent evidence-oriented matching. Background Technology

[0002] With the development of digital recruitment and talent management in enterprises, competency assessment based on candidate resume information has become an important technical means for talent screening, job matching, and talent inventory. Traditional competency assessment methods usually rely on manual review of resumes and recruitment personnel's experience to make a comprehensive judgment on candidates' educational background, work experience, project experience, and skill levels. However, with the rapid increase in the number of resumes and the increasing complexity of job competency requirements, manual assessment methods suffer from problems such as low assessment efficiency, high subjectivity, and inconsistent evaluation standards, making it difficult to meet the needs of large-scale talent screening and precise talent evaluation.

[0003] Existing AI-based resume evaluation methods typically employ natural language processing (NLP) to parse resume text and use keyword matching, text similarity calculation, or machine learning models to predict candidates' abilities. For example, by extracting job titles, skill keywords, and years of work experience from resumes and matching them with preset job requirements, a candidate evaluation result is generated. However, due to the unstructured nature of resume text, the same ability information may be described in different ways, making traditional keyword matching methods prone to missing relevant information. Furthermore, single semantic matching models struggle to comprehensively consider rule reliability, historical experience, and job-specific requirements, resulting in insufficient accuracy and adaptability of the evaluation results.

[0004] Furthermore, with the application of large language models in competency assessment scenarios, existing methods typically combine the complete resume text with each competency dimension one by one before inputting it into the large language model for enumerated assessment. This means that the entire resume text is input for each competency dimension for reasoning. In this approach, the resume text is repeatedly input in assessment requests for each dimension, with a large amount of content irrelevant to the current dimension occupying the model's context window. This results in high lexical consumption, high assessment costs, and slow response times. Moreover, redundant information can easily interfere with the model's focus on key evidence, affecting the accuracy of the scores. Simultaneously, existing assessment methods usually only output the scoring results, lacking structured reflection of the scoring reasons. The interpretability of the assessment process is insufficient, and the historical experience of assessment experts is difficult to continuously accumulate and reuse, failing to form a self-evolving capability that improves accuracy with use. Therefore, how to improve the accuracy, interpretability, and efficiency of competency assessment, reduce the lexical consumption of large language models, and achieve continuous accumulation and reuse of assessment experience has become an urgent technical problem to be solved. Summary of the Invention

[0005] (a) Purpose of the invention The purpose of this invention is to provide a competency assessment method and system based on multi-agent evidence-oriented matching. By combining candidate data parsing, rule recall, multi-verification agent collaboration, cross-validation, and experience memory enhancement, it achieves intelligent assessment of candidate competency. Through multi-dimensional evidence extraction and dynamic rule optimization, it improves the accuracy, consistency, and reliability of the assessment, while reducing the cost of manual assessment and enabling continuous optimization and intelligent iteration of the assessment system. Furthermore, this invention avoids the waste of lexical units caused by enumeration-based assessment and reduces assessment costs by using targeted matching of evidence fragments and competency dimensions and a scoring method that uses hierarchical evaluation criteria as reference context. Simultaneously, it generates comprehensive evaluation criteria for each competency dimension and updates the rule base using an experience memory enhancement algorithm, achieving self-evolution of the assessment system.

[0006] (II) Technical Solution To address the above problems, this invention provides a competency assessment method based on multi-agent evidence-oriented matching, comprising: Obtain candidate information, which includes at least one of job application information and publicly available information; Using a preset parsing algorithm, the candidate information is parsed into a set of evidence fragments; Based on the set of evidence fragments, the matching rules corresponding to each of the evidence fragments are recalled from the preset rule base; Based on multiple preset verification agents and corresponding matching rules, the evidence fragments in the evidence fragment set are verified from multiple perspectives to obtain multiple verification results. According to the preset cross-validation mechanism, multiple verification results are cross-validated to obtain the competency assessment result; Based on the preset experience memory enhancement algorithm and the competency assessment results, the preset rule base is updated to obtain the updated rule base, and the competency assessment of the next candidate's information is performed based on the updated rule base.

[0007] In another aspect of the present invention, preferably, the step of parsing candidate data into a set of evidence fragments using a preset parsing algorithm includes: The candidate information is parsed to extract the text content and retain the document structure information to obtain the original text. Based on semantic topic types, the original text is semantically segmented to obtain multiple valid evidence fragments; Entity recognition is performed on each of the valid evidence fragments to obtain the corresponding entity information; Each valid piece of evidence is classified according to intent to obtain the corresponding intent category and intent confidence level. The set of evidence fragments is constructed based on the entity information, intent category, and intent confidence level corresponding to each valid evidence fragment.

[0008] In another aspect of the present invention, preferably, the step of recalling matching rules corresponding to each of the evidence fragments from a preset rule base based on the set of evidence fragments includes: The evidence fragments are matched with a preset rule base using a pre-defined artificial intelligence model to obtain a set of candidate rules; The candidate rule set is subjected to quality filtering to remove candidate rules with a confidence level lower than a preset threshold, thereby obtaining the corresponding matching rule.

[0009] In another aspect of the present invention, preferably, the verification agent includes: a hard condition verifier; Based on the hard condition validator and the corresponding matching rules, the evidence fragment is validated, including: Determine the hard criteria rules in the matching rules corresponding to the target position; The evidence fragments are matched with the hard condition rules to obtain hard condition matching results. The hard condition rules include negative score rules, marker rules, bonus score rules, and green channel rules. Based on the rule weights and hard condition matching results corresponding to each of the hard condition rules, calculate the comprehensive hard condition score corresponding to the evidence fragment; The comprehensive score of the hard conditions is compared with the preset hard condition threshold to obtain the hard condition verification result.

[0010] In another aspect of the present invention, preferably, the verification agent includes: a basic competency assessment verifier; Based on the aforementioned basic competency assessment verifier and corresponding matching rules, the evidence fragments are verified, including: Perform capability dimension matching on each of the aforementioned evidence fragments to obtain the capability dimension corresponding to each of the aforementioned evidence fragments; Based on the capability dimension corresponding to each of the evidence fragments, the evidence fragments, the corresponding matching rules, and the preset artificial intelligence model, the capability dimension analysis results corresponding to each of the evidence fragments are obtained; Based on the analysis results of the aforementioned capability dimensions, calculate the corresponding dimension score for each capability dimension; Based on the dimension scores corresponding to each capability dimension and the preset competency weights corresponding to each capability dimension, a basic competency score is calculated, which is the basic competency verification result.

[0011] In another aspect of the present invention, preferably, the step of performing capability dimension matching on each of the evidence fragments to obtain the capability dimension corresponding to each of the evidence fragments includes: Based on the job competency model corresponding to the target position, determine the set of candidate competency dimensions; Calculate the correlation between each of the evidence fragments and each of the candidate capability dimensions in the candidate capability dimension set to obtain the corresponding correlation score; Evidence fragments whose relevance scores meet preset matching conditions are identified as evidence fragment-capability dimension matching pairs with candidate capability dimensions, thereby obtaining the capability dimension corresponding to each evidence fragment.

[0012] In another aspect of the present invention, preferably, obtaining the capability dimension analysis result corresponding to each of the evidence fragments based on the capability dimension corresponding to each of the evidence fragments, the evidence fragments, the corresponding matching rules, and the preset artificial intelligence model includes: Obtain the grading criteria corresponding to each of the aforementioned capability dimensions, wherein the grading criteria represent the evaluation standards for different capability levels under the corresponding capability dimension; For each piece of evidence, the grading criteria corresponding to the capability dimension that matches the evidence piece are used as the reference context, and the evaluation prompt information is constructed by combining the evidence piece and the corresponding matching rules. The evaluation prompts are input into the preset artificial intelligence model to obtain the single evidence score of the evidence fragment on the corresponding ability dimension and the corresponding scoring basis. For multiple single evidence scores under the same capability dimension, the dimension score corresponding to the capability dimension is obtained by combining the evidence relevance and / or evidence confidence corresponding to each single evidence score. By summarizing multiple scoring criteria under the same capability dimension, a comprehensive evaluation criterion corresponding to the capability dimension is obtained. The comprehensive evaluation criterion is used to characterize the evidence supporting the candidate's score under the capability dimension and the reasons for the evaluation.

[0013] In another aspect of the present invention, preferably, the verification agent includes: a risk detection verifier; Based on the risk detection verifier and the corresponding matching rules, the evidence fragment is verified, including: The risk detection dimensions are determined, and the evidence fragments are verified based on the corresponding matching rules to obtain the risk scores for each risk detection dimension. The risk detection dimensions include the risk of gaps in resume, the risk of frequent job hopping, the risk of conflicting resumes, and the risk of questionable academic qualifications. The risk scores of each risk detection dimension are weighted and summed to obtain a comprehensive risk score. Based on the comparison between the comprehensive risk score and the preset risk level threshold, the corresponding risk level is determined, and the risk level is the risk verification result.

[0014] In another aspect of the present invention, preferably, the step of cross-validating multiple validation results according to a preset cross-validation mechanism to obtain a competency assessment result includes: Based on the results of the hard condition verification, the basic competency verification, and the risk verification, calculate the consistency index among the verification results. The consistency index is compared with a preset conflict threshold to obtain a conflict determination result; Based on the conflict determination results, each verification result, consistency index, and corresponding preset comprehensive weight, a comprehensive evaluation result is calculated to obtain the competency evaluation result.

[0015] In another aspect of the present invention, preferably, comparing the consistency index with a preset conflict threshold to obtain a conflict determination result includes: When the consistency index is greater than or equal to the preset conflict threshold, the conflict determination result is that there is no verification conflict; When the consistency index is less than the preset conflict threshold, the conflict determination result indicates that a verification conflict exists, and the conflict resolution mechanism is activated, including: When the result of the hard condition verification is that the job hard conditions are not met, the competency assessment result is subject to a level limit so that the competency assessment result does not exceed a preset level threshold. Based on the degree of influence of multiple verification results on the consistency index, the target verification result that causes the verification conflict is determined; Obtain the evidence fragments, matching rules, and verification process information corresponding to the target verification result, construct review information, and input the review information into a preset artificial intelligence model for special review to obtain the review result; Based on the review results, the target verification results are corrected, and the consistency index is recalculated. When the recalculated consistency index is greater than or equal to the preset conflict threshold, the competency assessment result is generated based on the corrected multiple verification results. When the recalculated consistency index is less than the preset conflict threshold, a manual review identifier is generated, and a manual review is performed.

[0016] In another aspect of the present invention, preferably, the step of updating the preset rule base based on the preset experience memory enhancement algorithm and the competency assessment result to obtain an updated rule base, and then performing competency assessment on the next candidate's data based on the updated rule base, includes: Based on the competency assessment results, feedback information is generated; The corresponding rule update type is determined based on the feedback information, and the update type includes correct match, mismatch, and incorrect match. The rule information is updated according to the update type to obtain the updated rule information, which includes rule confidence, rule application conditions, rule rating information and rule status. Based on the updated rule information, an updated rule base is generated; The updated rule base is then updated, and a competency assessment is performed on the next candidate's information based on the updated rule base.

[0017] In another aspect, preferably, the invention further includes: Based on the job requirements corresponding to the target position, the job requirement matching degree of the candidate is determined. The job requirement matching degree is used to characterize the degree of satisfaction between the candidate's educational background, major, years of work experience, professional qualifications and skill requirements and the job requirements. Based on preset evaluation weights, the job requirement matching degree and the competency assessment result are weighted and summed to obtain a comprehensive competency assessment result.

[0018] In another aspect, preferably, a competency assessment system based on multi-agent evidence-oriented matching includes: Acquisition module: Acquires candidate information, which includes at least one of job application information and publicly available information; Parsing module: Using a preset parsing algorithm, the candidate information is parsed into a set of evidence fragments; Recall module: Based on the set of evidence fragments, recall the matching rules corresponding to each of the evidence fragments from a preset rule base; Verification module: Based on multiple preset verification agents and corresponding matching rules, it performs multi-faceted verification on the evidence fragments in the evidence fragment set and obtains multiple verification results; Cross-validation module: Based on a preset cross-validation mechanism, cross-validate multiple validation results to obtain competency assessment results; Enhancement module: Based on the preset experience memory enhancement algorithm and the competency assessment results, the preset rule base is updated to obtain the updated rule base, and the competency assessment of the next candidate's information is performed based on the updated rule base.

[0019] (III) Beneficial Effects The above-described technical solution of the present invention has the following beneficial technical effects: This invention transforms unstructured candidate data into a set of evidence fragments containing information such as time, position, projects, skills, and achievements through a parsing algorithm, improving the completeness and accuracy of resume information extraction. By utilizing artificial intelligence models and rule confidence, it enhances the accuracy of evaluation rule matching and reduces information omissions caused by differences in textual expression. Simultaneously, multiple verification agents analyze candidate information from multiple dimensions, including hard requirements, competence, and resume risks, and a cross-validation mechanism is used to determine consistency and resolve conflicts, reducing biases from single-validation assessments and improving the reliability of evaluation results. Furthermore, an experience-memory enhancement algorithm dynamically optimizes the rule base based on feedback from historical evaluation results, enabling continuous accumulation and iterative updates of evaluation rules. This improves subsequent evaluation effectiveness without retraining the model, increasing talent evaluation efficiency, reducing manual screening costs, and enhancing the interpretability of the evaluation process. Furthermore, by atomizing candidate data into evidence fragments and performing targeted matching of these fragments with capability dimensions, scoring is conducted using only the corresponding grading criteria for each capability dimension as the reference context. This avoids enumerating all capability dimensions by combining complete candidate data, significantly reducing the lexical consumption and computational overhead of the AI ​​model and improving evaluation efficiency. Simultaneously, by summarizing the scoring criteria of each evidence fragment to generate a comprehensive evaluation criterion for the capability dimensions, the evaluation results become interpretable. Moreover, by updating the rule base through an experience memory enhancement algorithm, the evaluation system possesses self-evolution capabilities, and the evaluation effect continuously improves with the number of uses. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of one embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0022] Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0023] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0024] Example 1 A competency assessment method based on multi-agent evidence-oriented matching. Figure 1 An overall flowchart of one embodiment of the present invention is shown, as follows: Figure 1 As shown, it includes: Candidate information is obtained, including at least one of job application information and publicly available information. Job application information includes at least one of the following: resumes, cover letters, personal statements, portfolios, application registration information, educational background, work experience, project experience, professional skills, professional qualifications, training experience, awards, and other personal information related to the target position. Publicly available information includes at least one of the following: personal resume information, project achievement information, professional achievement information, professional experience information, publicly published information, and other publicly available information reflecting the candidate's professional abilities and job competence, generated by the candidate on authorized public platforms, recruitment platforms, career information platforms, academic platforms, corporate information platforms, and / or other publicly accessible information carriers. When obtaining the candidate information, source identification, time identification, and data type identification are applied to candidate information from different sources. The obtained candidate information is then deduplicated, formatted uniformly, and screened for validity to form a candidate information set for subsequent competency assessment.

[0025] Using a pre-defined parsing algorithm, candidate information is parsed into a set of evidence fragments. The candidate information is unstructured resume text, supporting various file formats, including Word, PDF, and plain text. For Word resumes, the document parsing interface can be used to read the document's internal structure information, extracting text content, paragraph hierarchy, and table field information. For PDF resumes, a text extraction algorithm is used to obtain the text stream, and paragraph structure is restored using text position information to avoid information errors caused by PDF formatting. For plain text resumes, paragraphs are divided based on line breaks to obtain initial text data. This embodiment includes: parsing the candidate information to extract text content and retain document structure information to obtain the original text; further preprocessing of the original text can be performed, including removing invalid characters, repeated spaces, header / footer information, irrelevant explanatory information, and other noise, and standardizing the text format to improve the accuracy of subsequent semantic analysis.

[0026] Based on semantic topic types, the original text is semantically segmented to obtain multiple valid evidence fragments. Specifically, the original text is divided according to professional event information, time information, job information, and semantic relationships in the resume, so that each evidence fragment corresponds to an independent professional experience, project experience, or ability performance event. For example, continuous text containing the employment period, job title, project description, and achievement indicators is divided into a professional event evidence fragment. Through semantic segmentation, subsequent rule matching can analyze specific ability performances, avoiding information interference caused by the entire resume text.

[0027] Entity recognition is performed on each valid evidence fragment to obtain corresponding entity information. A pre-defined entity recognition model is used to extract key entities from the valid evidence fragments. Entity information includes, but is not limited to, time entities, organizational entities, position entities, project entities, skill entities, educational background entities, and performance indicator entities. Specifically, the time entity is used to determine the candidate's tenure and project duration; the organizational entity is used to determine the employing unit or educational institution; the position entity is used to determine the job level; project and skill entities are used to characterize the candidate's professional abilities; and performance indicator entities are used to characterize the candidate's actual work performance. The entity recognition model can employ a deep learning-based named entity recognition model, such as a sequence labeling model based on pre-trained language models like BERT, RoBERTa, and ERNIE, or a BiLSTM-CRF model or an entity extraction model based on artificial intelligence models.

[0028] Each valid evidence fragment is classified according to its intent to obtain the corresponding intent category and intent confidence score. Specifically, a pre-defined text classification model can be used to perform semantic understanding on the evidence fragments. Based on the information type expressed by the valid evidence fragments, they can be divided into categories such as job description, project experience, work achievements, educational background, skill certificates, self-evaluation, or other categories, and the confidence score of the corresponding category can be output. Here, the intent category indicates the competency evaluation direction corresponding to the evidence fragment, and the intent confidence score indicates the reliability of the current classification result. The text classification model can be a multi-label text classification model based on a pre-trained language model, such as a classification model built based on pre-trained models like BERT, RoBERTa, and ERNIE, or a semantic classification model based on an artificial intelligence model.

[0029] Based on the entity information, intent category, and intent confidence level corresponding to each valid evidence fragment, the evidence fragment set is constructed. This evidence fragment set serves as input data for subsequent rule retrieval, multi-verification agent collaborative analysis, and comprehensive competency assessment.

[0030] Based on the set of evidence fragments, matching rules corresponding to each evidence fragment are retrieved from a preset rule base; specifically, in this embodiment, this includes: The evidence fragment is matched with a pre-defined rule base using a pre-set artificial intelligence model to obtain a set of candidate rules. The evidence fragment is then input into the artificial intelligence model for semantic feature extraction to obtain the corresponding semantic features. Based on these semantic features, the correlation between the evidence semantic features and the semantic features of each evaluation rule in the rule base is calculated. Multiple evaluation rules related to the evidence fragment are then selected from the rule base to form the set of candidate rules. The artificial intelligence model can also determine the relevance of the candidate rules based on the candidate's experience type, ability type, professional field, job responsibilities, time frame, and / or achievement description contained in the evidence fragment, thereby determining the degree of matching between each candidate rule and the evidence fragment. The artificial intelligence model may include large language models, natural language processing models, multimodal artificial intelligence models, domain-specific artificial intelligence models, etc.

[0031] The candidate rule set is quality filtered to remove candidate rules with a confidence level below a preset threshold, thus obtaining the corresponding matching rules. Rule confidence level characterizes the reliability of the association between a candidate rule and its corresponding evidence fragment. It can be determined based on at least one of the following: semantic similarity between the evidence fragment and the candidate rule, the degree of matching of key entities, consistency of evaluation dimensions, the degree to which the rule's applicability conditions are met, and the historical usage effect of the rule. When the rule confidence level of a candidate rule reaches or exceeds the preset threshold, the candidate rule is determined as the matching rule corresponding to the evidence fragment; when the rule confidence level is below the preset threshold, the corresponding candidate rule is removed to avoid evaluation rules with low relevance or insufficient applicability from participating in subsequent verification processes. The preset threshold is used to judge the reliability of the matching between the candidate rule and its corresponding evidence fragment, and is determined based on historical candidate data, historical rule matching results, and corresponding manual annotation results. Specifically, evidence fragments are extracted from historical candidate data, and the preset artificial intelligence model is used to obtain candidate rules and rule confidence levels corresponding to each historical evidence fragment. The candidate rules are compared with manually labeled valid matching rules, and the rule matching accuracy and / or effective recall rate are statistically analyzed under different rule confidence level intervals. A preset threshold is determined based on preset matching accuracy and / or effective recall rate requirements, ensuring that candidate rules with rule confidence levels reaching the preset threshold have a matching reliability that meets preset requirements. In actual operation, the preset threshold is adjusted based on the verification results obtained after subsequent verification of the matching rules by the verification agent. When the effective matching ratio corresponding to low-confidence rules is lower than the preset ratio, the preset threshold is increased; when a large number of valid rules are filtered out due to the preset threshold, resulting in a rule recall rate lower than the preset recall rate, the preset threshold is decreased. Here, the preset threshold can be 0.3 to ensure a minimum quality level for recalled rules.

[0032] Based on multiple preset verification agents and corresponding matching rules, the evidence fragments in the evidence fragment set are verified from multiple perspectives to obtain multiple verification results; wherein, the multiple preset verification agents include: a hard condition verifier; based on the hard condition verifier and the corresponding matching rules, the evidence fragments are verified, including: The hard condition rules in the matching rules corresponding to the target position are determined. Hard condition rules describe the job requirements that must be met or are given priority for the position. These may include educational requirements, professional requirements, years of work experience requirements, qualification certificate requirements, skill requirements, language proficiency requirements, age requirements, industry experience requirements, and special ability requirements. Each hard condition rule is pre-configured with a corresponding rule type and rule weight. The rule weight indicates the importance of the hard condition in the target position. Different positions may have different rule weights; for example, professional skills have a higher weight than language proficiency in R&D positions, while communication and language skills have relatively higher weights in marketing positions.

[0033] The evidence fragments are matched against the hard condition rules to obtain hard condition matching results. These hard condition rules include negative score rules, flag rules, bonus score rules, and green channel rules. Each hard condition rule is matched and judged item by item based on the entity information, intent category, and corresponding text content in the evidence fragments. For example, when the hard condition rule is a bachelor's degree or above, the candidate's educational qualification is assessed based on the educational qualification entity; when the hard condition rule is more than five years of R&D experience, the candidate's cumulative tenure is calculated based on the time, position, and institution entities; when the hard condition rule is familiarity with Java development, the candidate's skills are assessed based on the skill entity. Based on the matching results of each hard condition rule, corresponding hard condition matching results are generated. Negative scoring rules indicate job requirements that the candidate does not meet, such as insufficient education, unsuitable major, or age exceeding the job limit; a score of -1 is given. Marked scoring rules indicate conditions that need to be recorded but do not directly affect the score, such as overseas study experience, a background from a top university, or special industry experience, for subsequent comprehensive analysis; a score of 0 is given. Bonus scoring rules indicate conditions that meet the preferred job criteria, such as relevant professional qualification certificates, required language skills, or high professional achievements; a score of +1 is given. Green channel scoring rules indicate special advantages required for the position, such as industry-rare skill certifications, national or international awards, or experience leading major projects; these are given higher weight in subsequent evaluations and a score of +2 is given.

[0034] Based on the rule weights and matching results of each of the aforementioned hard condition rules, a comprehensive hard condition score is calculated for the evidence fragment. The score value is determined according to the rule type of each hard condition rule, and a weighted calculation is performed using the rule weights of each rule to obtain a comprehensive hard condition score that reflects the degree of matching between the candidate and the job's hard conditions. The rule weights can be pre-configured according to the job type or dynamically adjusted according to the company's recruitment strategy to adapt to the talent selection needs of different positions.

[0035] The comprehensive score of the hard requirements is compared with a preset hard requirement threshold to obtain the hard requirement verification result. When the comprehensive score of the hard requirements is greater than or equal to the preset hard requirement threshold, the candidate is determined to meet the job's hard requirements, and a hard requirement pass flag is generated; when the comprehensive score of the hard requirements is lower than the preset hard requirement threshold, the candidate is determined not to meet the job's hard requirements, and a hard requirement fail flag is generated. In some embodiments, for rapid screening scenarios, the subsequent evaluation process can be terminated directly based on the hard requirement verification result; for comprehensive evaluation scenarios such as talent inventory and talent reserves, even if the hard requirements are not passed, subsequent competency assessments and risk detection can continue, and the corresponding hard requirement risks can be marked in the final evaluation report to facilitate comprehensive decision-making by the evaluators.

[0036] Furthermore, in this embodiment, the preset multiple verification agents include: a basic competency assessment verifier; Based on the aforementioned basic competency assessment verifier and corresponding matching rules, the evidence fragments are verified, including: The capability dimensions of each piece of evidence are matched to obtain the capability dimension corresponding to each piece of evidence. To reduce the lexical consumption and computational overhead of the artificial intelligence model, this embodiment adopts a scoring method of targeted matching between evidence fragments and capability dimensions, avoiding enumeration-based evaluation. Specifically, it includes: Based on the job competency model corresponding to the target position, determine the set of candidate competency dimensions; The correlation between each of the evidence fragments and each of the candidate capability dimensions in the candidate capability dimension set is calculated to obtain the corresponding correlation score; for example, the correlation between the evidence fragments and each capability dimension is calculated based on the semantic similarity between the semantic features of the evidence fragments and the dimensional descriptions of each capability dimension, or based on the association between the entity information and intent category of the evidence fragments and each capability dimension. Evidence fragments whose relevance scores meet preset matching conditions are identified as evidence fragment-ability dimension matching pairs with candidate capability dimensions, thus obtaining the capability dimension corresponding to each evidence fragment. Only evidence fragment-ability dimension pairs whose relevance scores meet the preset matching conditions are input into the preset artificial intelligence model for scoring. For capability dimensions whose relevance scores do not meet the preset matching conditions, the artificial intelligence model is not invoked again in conjunction with the evidence fragment. Therefore, it is unnecessary to enumerate and evaluate each capability dimension in the candidate capability dimension set with complete candidate information. The original resume does not need to be repeatedly input in the evaluation request for each dimension, significantly reducing the lexical consumption and computational overhead of the artificial intelligence model. Simultaneously, it reduces the interference of irrelevant information on scoring reasoning and improves the accuracy of scoring.

[0037] Based on the capability dimension corresponding to each of the aforementioned evidence fragments, the evidence fragments themselves, the corresponding matching rules, and the preset artificial intelligence model, the capability dimension analysis results corresponding to each of the aforementioned evidence fragments are obtained, including: Obtain the grading criteria corresponding to each of the aforementioned ability dimensions. The grading criteria represent the evaluation standards for different ability levels under the corresponding ability dimension. For example, for each ability dimension, predefine the ability performance description and typical behavioral characteristics corresponding to each level from L1 to L5. For each piece of evidence, the grading criteria corresponding to the capability dimension that matches the evidence piece are used as the reference context, and the evaluation prompt information is constructed by combining the evidence piece and the corresponding matching rules. The evaluation prompts are input into the preset artificial intelligence model to obtain the single-evidence score and corresponding scoring basis for the evidence fragment on the corresponding ability dimension. Since the evaluation prompts only include the grading criteria for the ability dimension matching the evidence fragment, rather than the evaluation standards for all ability dimensions, the length of the input context is further compressed, reducing word consumption. The artificial intelligence model determines the level corresponding to the ability performance reflected in the evidence fragment based on the grading criteria, and outputs the single-evidence score and corresponding scoring basis for the evidence fragment on the corresponding ability dimension. The scoring basis explains which specific content in the evidence fragment and which standards in the grading criteria the artificial intelligence model used to arrive at the score.

[0038] For multiple single evidence scores under the same capability dimension, the dimension score corresponding to the capability dimension is obtained by combining the evidence relevance and / or evidence confidence corresponding to each single evidence score; for example, multiple single evidence scores are weighted and fused according to the rule confidence corresponding to each single evidence score and the degree of matching between the evidence fragment and the matching rule.

[0039] Multiple scoring criteria under the same competency dimension are summarized to obtain a comprehensive evaluation criterion corresponding to that competency dimension. This comprehensive evaluation criterion is used to characterize the evidence supporting the candidate's score in that competency dimension and the reasons for the evaluation. Simultaneously, multiple scoring criteria under the same competency dimension are summarized to obtain a comprehensive evaluation criterion for that corresponding competency dimension. This comprehensive evaluation criterion is used to explain the reasons why the candidate obtained the score in that corresponding competency dimension. For example, an artificial intelligence model can deduplicate, merge, and summarize multiple scoring criteria to form a holistic explanation of the score for that dimension. This allows evaluators to understand the specific evidence and reasons why the candidate was rated at the corresponding level, improving the interpretability of the evaluation results.

[0040] Based on the analysis results of the aforementioned ability dimensions, the dimension score corresponding to each ability dimension is calculated. Based on the ability level, rule confidence, and the degree of matching between the evidence fragment and the matching rule, multiple analysis results under the same ability dimension are comprehensively calculated, which may be multiplied or added, to obtain the dimension score of the corresponding ability dimension. The ability level is used to reflect the candidate's performance level in the corresponding ability dimension, the rule confidence is used to reflect the reliability of the matching rule, and the degree of matching is used to reflect the degree of conformity between the evidence fragment and the rule. By combining the above factors, the dimension score can more objectively reflect the candidate's actual ability level.

[0041] Based on the dimensional scores and preset competency weights for each competency dimension, a comprehensive basic competency score is calculated. This comprehensive basic competency score serves as the basic competency verification result. Specifically, according to the competency model corresponding to the target position, corresponding competency weights are pre-configured for each competency dimension. Different weight configuration methods can be used for different positions. For example, for R&D positions, professional technical skills, innovation capabilities, and problem analysis abilities have higher weights; for management positions, leadership skills, communication and coordination skills, and project management skills have higher weights. The dimensional scores of each competency dimension are weighted and integrated according to their corresponding competency weights to obtain the candidate's comprehensive basic competency score. This comprehensive basic competency score is output as the basic competency verification result, providing an evaluation basis for subsequent cross-validation mechanisms. The competency model corresponding to the target position is a pre-established structured competency evaluation model for the target position, including at least one candidate competency dimension, the competency level corresponding to each candidate competency dimension, the grading evaluation criteria for each competency level, and the competency weight corresponding to each candidate competency dimension.

[0042] Furthermore, in this embodiment, the preset multiple verification intelligent agents include: a risk detection verifier; Based on the risk detection verifier and the corresponding matching rules, the evidence fragment is verified, including: Risk detection dimensions are determined, and the evidence fragments are verified based on corresponding matching rules to obtain risk scores for each risk detection dimension. These dimensions include resume gap risk, frequent job-hopping risk, resume conflict risk, and questionable educational background risk. Risk analysis is performed on the candidate's resume information based on time information, institution information, position information, educational background information, and corresponding matching rules in the evidence fragments. Each risk detection dimension corresponds to different risk judgment rules. For resume gap risk, the time interval between two work experiences is calculated based on the start and end times of adjacent work experiences. When the time interval between two adjacent work experiences exceeds a preset time threshold, a resume gap risk is determined, and the corresponding risk score is determined based on the length of the gap; the longer the time interval, the higher the corresponding risk score. When there is evidence of educational experience, entrepreneurial experience, freelance experience, or other evidence that can reasonably explain the reason for the gap, the risk score can be corrected according to the corresponding matching rules. For frequent job-hopping risk, the average tenure of each candidate's work experience is calculated, and the number of historical jobs is counted. When the average tenure is below a preset threshold, a risk of frequent job hopping is identified. Simultaneously, the risk score is adjusted based on the number of past tenures; if a candidate's past tenure exceeds a preset threshold, the corresponding risk score is increased to reflect personnel stability risk. For resume conflict risk, a candidate's resume timeline is established based on the time, institution, and position information in each piece of evidence, and consistency checks are performed on the temporal and positional relationships between different evidence pieces. When there are multiple full-time positions within the same time period, significantly abnormal job promotion order, or time discrepancies between work experience and project experience, a resume conflict risk is identified, and a corresponding risk score is generated. For educational background questionable risk, consistency analysis is performed on the candidate's school name, educational level, major, graduation date, etc., based on educational information and corresponding matching rules, and preset educational information verification rules are used to determine if there are any anomalies in the educational information. When educational information is missing, inconsistent in description, contradicts publicly available educational information, or has obvious logical conflicts, a risk of educational background questionable is identified, and a corresponding risk score is generated.

[0043] A weighted sum of the risk scores from each risk detection dimension is used to obtain a comprehensive risk score. Specifically, based on the company's pre-defined risk preferences, corresponding risk weights are assigned to each risk detection dimension, with different dimensions potentially having different levels of importance. For example, for high-risk positions such as finance and security, the weights corresponding to resume conflict risk and questionable educational background risk can be appropriately increased; for innovative positions such as those in the internet industry, the weights corresponding to frequent job-hopping risk can be appropriately decreased. The comprehensive risk score is calculated based on the risk scores and risk weights corresponding to each risk detection dimension, ensuring that the comprehensive risk score fully reflects the candidate's overall resume risk level.

[0044] Based on the comparison between the comprehensive risk score and the preset risk level threshold, the corresponding risk level is determined, which is the risk verification result. Specifically, the comprehensive risk score is divided into three levels: low risk (e.g., 0-0.3), medium risk (e.g., 0.3-0.6), and high risk (0.6-1.0). When the comprehensive risk score is lower than the first risk level threshold, it is judged as low risk; when the comprehensive risk score is between the first and second risk level thresholds, it is judged as medium risk, and a risk warning message is generated; when the comprehensive risk score is greater than or equal to the second risk level threshold, it is judged as high risk, and a high risk warning message is generated, suggesting further background investigation or manual review.

[0045] According to a pre-defined cross-validation mechanism, multiple verification results are cross-validated to obtain competency assessment results. This introduces the consensus mechanism of distributed systems into the field of talent assessment. Credibility is quantified using a consistency index. Multiple validators are treated as independent assessment nodes, and consistency analysis is used to determine the degree of coordination between the outputs of each assessment node, thereby reducing the possibility of misjudgment from a single verification result and improving the reliability and stability of the final assessment result. Specifically, this includes: Based on the results of the hard condition verification, competency verification, and risk verification, a consistency index is calculated among the verification results. Each verification result is then normalized to map the scores output by different verifiers to a unified numerical range, thereby eliminating dimensional differences between different scoring systems. Then, the normalized hard condition verification results, competency verification results, and risk verification results are used to construct a comprehensive evaluation vector, and the consistency index is calculated based on the dispersion among the verification results. The consistency index characterizes the degree of consistency between the evaluation results of multiple verifiers. When the output results of each verifier are relatively close, the consistency index is high, indicating that the evaluations of the candidates by different verifiers tend to be consistent. When the output results of each verifier differ significantly, the consistency index is low, indicating that there is a significant evaluation conflict between different verifiers, requiring further conflict analysis. The following formula can be used to calculate the consistency index: Wherein, CIX represents the consistency index, and V represents the evaluation vector composed of the results of hard condition verification, basic competence verification, and risk verification. The standard deviation of the evaluation vector is used to characterize the degree of dispersion among the verification results; This represents the mean of the evaluation vector. The minimum value is set to prevent the denominator from being zero.

[0046] The consistency index is compared with a preset conflict threshold to obtain a conflict determination result. Based on the conflict determination result, each verification result, consistency index, and corresponding preset comprehensive weight are used to calculate a comprehensive evaluation result, thus obtaining the competency evaluation result. The preset comprehensive weights are then used to perform a weighted summation of each verification result and consistency index to obtain the competency evaluation result.

[0047] Furthermore, in this embodiment, the consistency index is compared with a preset conflict threshold to obtain a conflict determination result, including: When the consistency index is greater than or equal to the preset conflict threshold, the conflict determination result is that there is no verification conflict. The conflict threshold is used to determine whether there is a significant conflict between the output results of multiple verifiers, and its value can be preset according to different job types, corporate risk preferences, or historical assessment data. For example, the conflict threshold can be set to 0.75. When the consistency index is greater than or equal to the conflict threshold, it indicates that the assessment results output by the hard condition verifier, competency assessment verifier, and risk detection verifier have a high degree of consistency, and there is no significant conflict between the verifiers. The conflict determination result is that there is no verification conflict, and a comprehensive assessment can be directly performed based on the verification results.

[0048] When the consistency index is less than the preset conflict threshold, the conflict determination result indicates that a verification conflict exists, and the conflict resolution mechanism is activated, including: When the hard requirement verification result indicates that the job requirements are not met, the competency assessment result is subject to a level restriction to ensure that it does not exceed a preset level threshold. The hard requirement verification result serves as the fundamental constraint for job qualifications and has the highest priority. Even if a candidate has a high overall competency score, if they do not meet the job's educational requirements, professional requirements, qualification certificate requirements, or other mandatory job requirements, the final assessment level will be restricted. For example, the final assessment level can be restricted to no higher than Level L3 to ensure that the basic job requirements are met first, preventing strong soft skills from masking deficiencies in basic qualifications.

[0049] Based on the impact of multiple verification results on the consistency index, the target verification result causing the verification conflict is determined. The contributions of the hard condition verification results, competency verification results, and risk verification results to the consistency index are analyzed separately. The degree of deviation of each verification result from the overall evaluation mean is calculated, and the verification result with the largest deviation is determined as the target verification result. When multiple verification results simultaneously exhibit significant deviations, all multiple verification results can be used as target verification results in subsequent review processes.

[0050] The system acquires the evidence fragments, matching rules, and verification process information corresponding to the target verification result, constructs review information, and inputs this review information into a preset artificial intelligence model for specialized review to obtain the review result. Specifically, the review information includes, but is not limited to, the original text content of the target evidence fragment, entity recognition results, intent classification results, recall matching rules, rule matching relationships, scoring results corresponding to each verifier, verification basis, and reasoning process. The preset artificial intelligence model, in conjunction with the above review information, re-performs semantic understanding and logical reasoning on the target verification result, analyzes whether the target evidence fragment meets the applicable conditions of the corresponding rules, determines whether the verification result has rule matching errors, semantic understanding biases, or insufficient evidence, and outputs the specialized review result and corresponding review reasons.

[0051] Based on the review results, the target verification results are corrected, and the consistency index is recalculated. When the special review results are consistent with the original verification results, the original verification results are kept unchanged. When the special review results are inconsistent with the original verification results, the corresponding score, capability level, or risk level is corrected based on the special review results, and the consistency index is recalculated based on the corrected multiple verification results to assess whether the verifiers have reached a consistent state again.

[0052] When the recalculated consistency index is greater than or equal to the preset conflict threshold, the competency assessment result is generated based on the corrected multiple verification results; this indicates that the verification conflict has been effectively resolved. At this time, the competency assessment result is generated based on the corrected multiple verification results, and the final assessment report is output.

[0053] When the recalculated consistency index is less than the preset conflict threshold, a manual review flag is generated for manual review. This indicates that the automatic conflict resolution mechanism still cannot eliminate the evaluation discrepancies among multiple verifiers. At this point, a manual review flag is generated, and the corresponding candidate information, target evidence fragments, evaluation results of each verifier, special review results, and conflict reasons are submitted to the expert review queue for manual review. Through the above conflict resolution mechanism, while fully leveraging the advantages of collaborative evaluation by multiple verifiers, multi-level verification methods such as priority constraints, conflict localization, special review by artificial intelligence models, and manual review are introduced. This achieves automatic identification, automatic correction, and manual fallback handling of verification conflicts, not only improving the accuracy and consistency of the final competency assessment results but also enhancing the interpretability and reliability of the assessment process.

[0054] Based on a preset experience memory enhancement algorithm and the competency assessment results, the preset rule base is updated to obtain an updated rule base. Based on the updated rule base, a competency assessment is performed on the next candidate, including: Based on the competency assessment results, feedback information is generated. This feedback information reflects the accuracy of the current assessment results and may originate from confirmation of the assessment results by recruiters, business managers, or domain experts. It may also come from feedback on subsequent hiring decisions, probationary period performance, performance evaluations, and other practical applications of the candidate. The feedback information includes at least rule matching status, competency rating results, rule applicability, and corresponding correction suggestions, indicating the effectiveness of the current rules in the actual assessment process.

[0055] The corresponding rule update type is determined based on the feedback information. The update type includes correct match, mismatch, and incorrect match. When the feedback information indicates that the current rule can accurately match the evidence fragment and the evaluation result is correct, the update type is determined to be correct match. When the rule can match the target evidence fragment, but there is a certain deviation in the capability level, scope of application, or evaluation result, the update type is determined to be mismatch. When there is an incorrect match between the rule and the evidence fragment, resulting in an obviously incorrect evaluation result, the update type is determined to be incorrect match.

[0056] The rule information is updated according to the update type to obtain updated rule information, which includes rule confidence, rule applicability conditions, rule rating information, and rule status. When an expert confirms that a rule matches the corresponding evidence fragment correctly, the rule is determined to have high applicability and reliability in the current scenario, and a positive reinforcement learning strategy is executed. Specifically, the rule confidence is increased, the historical usage count of the rule is increased, and the corresponding usage frequency is recalculated. The rule confidence gradually approaches the preset maximum value in a decreasing increment, so that the rule gradually increases its credibility with multiple correct feedbacks, without excessive confidence increase due to a single feedback, thus ensuring the smoothness and stability of the rule update process. At the same time, the updated usage count is used to calculate the rule usage frequency in the subsequent multi-factor recall process, increasing the recall priority of the rule in similar evidence fragments.

[0057] When experts believe that the current rule matches the corresponding evidence fragment, but there are deviations in the corresponding competency level, applicable conditions, or rating results, it is determined to be a revised rating feedback, and a rule differentiation strategy is implemented. Specifically, while retaining the original rule, a new rule is generated based on the expert's revision results, and an initial rule confidence level is assigned to the new rule, while the rule confidence level of the original rule is appropriately reduced. The new rule inherits the basic attributes of the original rule and is improved by combining the expert's revised applicable conditions, competency level, or rule description, making the new rule more applicable to the current specific scenario, while the original rule continues to be retained for other application scenarios with a wider range of applications. Through rule differentiation, not only can existing experience and knowledge be preserved, but the fine-grained rules in the rule base can also be continuously enriched, improving the assessment accuracy for different positions and different evidence scenarios.

[0058] When an expert explicitly marks a rule's matching result as incorrect, it is determined that the current rule is being misapplied, and a negative learning strategy is implemented. Specifically, the rule's confidence is reduced according to a preset negative learning rate, gradually lowering its priority in subsequent rule recall processes. When the rule's confidence drops below a preset obsolescence threshold, the rule's status is updated to pending review or obsolescence, suspending its participation in subsequent rule recalls until further expert review and confirmation. Rules confirmed as invalid can be permanently marked as obsolescence rules; rules confirmed to still have some applicability can be revised based on review comments and then reactivated.

[0059] Furthermore, to ensure the stability and controllability of the rule update process, this embodiment sets different learning rate parameters for positive reinforcement, rule differentiation, and negative learning. Specifically, the positive reinforcement learning rate controls the rate of increase in rule confidence, the rule differentiation coefficient determines the initial confidence of new rules, the inhibition coefficient controls the rate of decrease in the confidence of original rules, and the negative learning rate controls the degree of punishment for erroneous rules. Different learning rate parameters can be pre-configured based on historical evaluation data, rule update frequency, and actual enterprise application needs, and satisfy preset parameter constraints. This allows correct rules to gradually accumulate confidence, biased rules to continuously evolve and refine, and erroneous rules to be promptly removed from the evaluation process, thereby achieving continuous optimization and dynamic evolution of the rule base. This enables the continuous accumulation of experiential knowledge and adaptive rule updates.

[0060] Based on the updated rule information, an updated rule base is generated; a new version is automatically generated each time a rule is updated. Version management can include: each time a rule is changed, confidence is updated, rule is differentiated, or rule is deprecated, the version number is automatically incremented, such as v2.3→v2.4; old version rules are retained for traceable auditing of historical evaluation results and are not directly physically deleted; the rule base supports multiple versions in parallel and canary release, and new rules can be gradually implemented proportionally. The updated rule base is then updated, and competency assessment is performed on the next candidate's data based on the updated rule base. This allows for continuous improvement in rule recall accuracy, verification result reliability, and competency assessment precision without retraining, thus forming a competency assessment system with self-learning and self-optimization capabilities.

[0061] Furthermore, this embodiment also includes: determining the candidate's job requirement matching degree based on the job requirements corresponding to the target position; and performing a weighted summation of the job requirement matching degree and the competency assessment result to obtain a comprehensive competency assessment result. The job requirement matching degree is used to characterize the candidate's educational background, professional requirements, years of work experience, professional qualifications, skill requirements, and / or other preset job conditions and their fulfillment of the job requirements corresponding to the target position; these are different evaluation dimensions from the competency assessment result. After obtaining the job requirement matching degree and the competency assessment result, the job requirement matching degree and the competency assessment result are weighted and summed according to preset evaluation weights to obtain a comprehensive competency assessment result. The competency assessment result characterizes the candidate's comprehensive competency level determined based on their experience, skills, project experience, professional abilities, and other competency evidence; the job requirement matching degree characterizes the degree of fit between the candidate's comprehensive abilities and the specific job requirements of the target position. By weighted fusion of the two evaluation dimensions, the comprehensive competency assessment result simultaneously reflects the candidate's actual ability level and their fit with the target position.

[0062] Furthermore, in this embodiment, the method is executed collaboratively by multiple intelligent agents, which at least include: a parsing agent, used to parse the candidate information into a set of evidence fragments; a recall agent, used to perform rule matching between the evidence fragments and the rule base; multiple verification agents, corresponding to a hard condition verifier, a competency assessment verifier, and a risk detection verifier, used to perform corresponding multi-faceted verifications; a review agent, used to perform a special review for conflict resolution when verification conflicts exist; and a rule learning agent, used to update the rule base based on feedback information from the evaluators. Each of the intelligent agents executes its corresponding task based on a preset artificial intelligence model, and transmits evidence fragments, matching rules, verification results, and review results through a preset intelligent agent collaboration protocol, forming a multi-agent collaborative evaluation pipeline; as the rule base is continuously updated during the evaluation process, the evaluation capabilities of each intelligent agent are improved, enabling the entire evaluation system to form a self-evolving mechanism that becomes more accurate with use.

[0063] Example 2 A competency assessment system based on multi-agent evidence-oriented matching includes: Acquisition module: Acquires candidate information, which includes at least one of job application information and publicly available information; Parsing module: Using a preset parsing algorithm, the candidate information is parsed into a set of evidence fragments; Recall module: Based on the set of evidence fragments, recall the matching rules corresponding to each of the evidence fragments from a preset rule base; Verification module: Based on multiple preset verification agents and corresponding matching rules, it performs multi-faceted verification on the evidence fragments in the evidence fragment set and obtains multiple verification results; Cross-validation module: Based on a preset cross-validation mechanism, cross-validate multiple validation results to obtain competency assessment results; Enhancement module: Based on the preset experience memory enhancement algorithm and the competency assessment results, the preset rule base is updated to obtain the updated rule base, and the competency assessment of the next candidate's information is performed based on the updated rule base.

[0064] It should be understood that the specific embodiments described above are merely illustrative or explanatory of the principles of the invention and do not constitute a limitation thereof. Therefore, any modifications, equivalent substitutions, improvements, etc., made without departing from the spirit and scope of the invention should be included within the protection scope of the invention. Furthermore, the appended claims are intended to cover all variations and modifications falling within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

[0065] The present invention has been described above with reference to embodiments thereof. However, these embodiments are merely illustrative and not intended to limit the scope of the invention. The scope of the invention is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of the invention, and all such substitutions and modifications should fall within the scope of the invention.

[0066] Although embodiments of the present invention have been described in detail, it should be understood that various changes, substitutions, and modifications can be made to the embodiments of the present invention without departing from the spirit and scope of the invention.

[0067] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A competency assessment method based on multi-agent evidence-oriented matching, characterized in that, include: Obtain candidate information, which includes at least one of job application information and publicly available information; Using a preset parsing algorithm, the candidate information is parsed into a set of evidence fragments; Based on the set of evidence fragments, the matching rules corresponding to each of the evidence fragments are recalled from the preset rule base; Based on multiple preset verification agents and corresponding matching rules, the evidence fragments in the evidence fragment set are verified from multiple perspectives to obtain multiple verification results. According to the preset cross-validation mechanism, multiple verification results are cross-validated to obtain the competency assessment result; Based on the preset experience memory enhancement algorithm and the competency assessment results, the preset rule base is updated to obtain the updated rule base, and the competency assessment of the next candidate's information is performed based on the updated rule base.

2. The competency assessment method based on multi-agent evidence-oriented matching according to claim 1, characterized in that, The process of parsing candidate information into a set of evidence fragments using a preset parsing algorithm includes: The candidate information is parsed to extract the text content and retain the document structure information to obtain the original text. Based on semantic topic types, the original text is semantically segmented to obtain multiple valid evidence fragments; Entity recognition is performed on each of the valid evidence fragments to obtain the corresponding entity information; Each valid piece of evidence is classified according to intent to obtain the corresponding intent category and intent confidence level. The set of evidence fragments is constructed based on the entity information, intent category, and intent confidence level corresponding to each valid evidence fragment.

3. The competency assessment method based on multi-agent evidence-oriented matching according to claim 2, characterized in that, The step of recalling matching rules corresponding to each of the evidence fragments from a preset rule base based on the set of evidence fragments includes: The evidence fragments are matched with a preset rule base using a pre-defined artificial intelligence model to obtain a set of candidate rules; The candidate rule set is subjected to quality filtering to remove candidate rules with a confidence level lower than a preset threshold, thereby obtaining the corresponding matching rule.

4. The competency assessment method based on multi-agent evidence-oriented matching according to claim 3, characterized in that, The verification agent includes: a hard condition verifier; Based on the hard condition validator and the corresponding matching rules, the evidence fragment is validated, including: Determine the hard criteria rules in the matching rules corresponding to the target position; The evidence fragments are matched with the hard condition rules to obtain hard condition matching results. The hard condition rules include negative score rules, marker rules, bonus score rules, and green channel rules. Based on the rule weights and hard condition matching results corresponding to each of the hard condition rules, calculate the comprehensive hard condition score corresponding to the evidence fragment; The comprehensive score of the hard conditions is compared with the preset hard condition threshold to obtain the hard condition verification result.

5. The competency assessment method based on multi-agent evidence-oriented matching according to claim 4, characterized in that, The verification agent includes: a basic competency assessment verification device; Based on the aforementioned basic competency assessment verifier and corresponding matching rules, the evidence fragments are verified, including: Perform capability dimension matching on each of the aforementioned evidence fragments to obtain the capability dimension corresponding to each of the aforementioned evidence fragments; Based on the capability dimension corresponding to each of the evidence fragments, the evidence fragments, the corresponding matching rules, and the preset artificial intelligence model, the capability dimension analysis results corresponding to each of the evidence fragments are obtained; Based on the analysis results of the aforementioned capability dimensions, calculate the corresponding dimension score for each capability dimension; Based on the dimension scores corresponding to each capability dimension and the preset competency weights corresponding to each capability dimension, a basic competency score is calculated, which is the basic competency verification result.

6. The competency assessment method based on multi-agent evidence-oriented matching according to claim 5, characterized in that, The step of matching the capability dimensions of each of the evidence fragments to obtain the capability dimension corresponding to each of the evidence fragments includes: Based on the job competency model corresponding to the target position, determine the set of candidate competency dimensions; Calculate the correlation between each of the evidence fragments and each of the candidate capability dimensions in the candidate capability dimension set to obtain the corresponding correlation score; Evidence fragments whose relevance scores meet preset matching conditions are identified as evidence fragment-capability dimension matching pairs with candidate capability dimensions, thereby obtaining the capability dimension corresponding to each evidence fragment.

7. The competency assessment method based on multi-agent evidence-oriented matching according to claim 5, characterized in that, The step of obtaining the capability dimension analysis results corresponding to each of the evidence fragments based on the capability dimension corresponding to each of the evidence fragments, the evidence fragments, the corresponding matching rules, and the preset artificial intelligence model includes: Obtain the grading criteria corresponding to each of the aforementioned capability dimensions, wherein the grading criteria represent the evaluation standards for different capability levels under the corresponding capability dimension; For each piece of evidence, the grading criteria corresponding to the capability dimension that matches the evidence piece are used as the reference context, and the evaluation prompt information is constructed by combining the evidence piece and the corresponding matching rules. The evaluation prompts are input into the preset artificial intelligence model to obtain the single evidence score of the evidence fragment on the corresponding ability dimension and the corresponding scoring basis. For multiple single evidence scores under the same capability dimension, the dimension score corresponding to the capability dimension is obtained by combining the evidence relevance and / or evidence confidence corresponding to each single evidence score. By summarizing multiple scoring criteria under the same capability dimension, a comprehensive evaluation criterion corresponding to the capability dimension is obtained. The comprehensive evaluation criterion is used to characterize the evidence supporting the candidate's score under the capability dimension and the reasons for the evaluation.

8. The competency assessment method based on multi-agent evidence-oriented matching according to claim 5, characterized in that, The verification agent includes: a risk detection verifier; Based on the risk detection verifier and the corresponding matching rules, the evidence fragment is verified, including: The risk detection dimensions are determined, and the evidence fragments are verified based on the corresponding matching rules to obtain the risk scores for each risk detection dimension. The risk detection dimensions include the risk of gaps in resume, the risk of frequent job hopping, the risk of conflicting resumes, and the risk of questionable academic qualifications. The risk scores of each risk detection dimension are weighted and summed to obtain a comprehensive risk score. Based on the comparison between the comprehensive risk score and the preset risk level threshold, the corresponding risk level is determined, and the risk level is the risk verification result.

9. The competency assessment method based on multi-agent evidence-oriented matching according to claim 8, characterized in that, The step of cross-validating multiple validation results according to a preset cross-validation mechanism to obtain a competency assessment result includes: Based on the results of the hard condition verification, the basic competency verification, and the risk verification, calculate the consistency index among the verification results. The consistency index is compared with a preset conflict threshold to obtain a conflict determination result; Based on the conflict determination results, each verification result, consistency index, and corresponding preset comprehensive weight, a comprehensive evaluation result is calculated to obtain the competency evaluation result.

10. The competency assessment method based on multi-agent evidence-oriented matching according to claim 9, characterized in that, The step of comparing the consistency index with a preset conflict threshold to obtain a conflict determination result includes: When the consistency index is greater than or equal to the preset conflict threshold, the conflict determination result is that there is no verification conflict; When the consistency index is less than the preset conflict threshold, the conflict determination result indicates that a verification conflict exists, and the conflict resolution mechanism is activated, including: When the result of the hard condition verification is that the job hard conditions are not met, the competency assessment result is subject to a level limit so that the competency assessment result does not exceed a preset level threshold. Based on the degree of influence of multiple verification results on the consistency index, the target verification result that causes the verification conflict is determined; Obtain the evidence fragments, matching rules, and verification process information corresponding to the target verification result, construct review information, and input the review information into a preset artificial intelligence model for special review to obtain the review result; Based on the review results, the target verification results are corrected, and the consistency index is recalculated. When the recalculated consistency index is greater than or equal to the preset conflict threshold, the competency assessment result is generated based on the corrected multiple verification results. When the recalculated consistency index is less than the preset conflict threshold, a manual review identifier is generated, and a manual review is performed.

11. The competency assessment method based on multi-agent evidence-oriented matching according to claim 10, characterized in that, The preset rule base is updated based on the preset experience memory enhancement algorithm and the competency assessment results to obtain an updated rule base. Competency assessment is then performed on the next candidate's data based on the updated rule base, including: Based on the competency assessment results, feedback information is generated; The corresponding rule update type is determined based on the feedback information, and the update type includes correct match, mismatch, and incorrect match. The rule information is updated according to the update type to obtain the updated rule information, which includes rule confidence, rule application conditions, rule rating information and rule status. Based on the updated rule information, an updated rule base is generated; The updated rule base is then updated, and a competency assessment is performed on the next candidate's information based on the updated rule base.

12. The competency assessment method based on multi-agent evidence-oriented matching according to claim 1, characterized in that, Also includes: Based on the job requirements corresponding to the target position, the job requirement matching degree of the candidate is determined. The job requirement matching degree is used to characterize the degree of satisfaction between the candidate's educational background, major, years of work experience, professional qualifications and skill requirements and the job requirements. Based on preset evaluation weights, the job requirement matching degree and the competency assessment result are weighted and summed to obtain a comprehensive competency assessment result.

13. A competency assessment system based on multi-agent evidence-oriented matching, characterized in that, include: Acquisition module: Acquires candidate information, which includes at least one of job application information and publicly available information; Parsing module: Using a preset parsing algorithm, the candidate information is parsed into a set of evidence fragments; Recall module: Based on the set of evidence fragments, recall the matching rules corresponding to each of the evidence fragments from a preset rule base; Verification module: Based on multiple preset verification agents and corresponding matching rules, it performs multi-faceted verification on the evidence fragments in the evidence fragment set and obtains multiple verification results; Cross-validation module: Based on a preset cross-validation mechanism, cross-validate multiple validation results to obtain competency assessment results; Enhancement module: Based on the preset experience memory enhancement algorithm and the competency assessment results, the preset rule base is updated to obtain the updated rule base, and the competency assessment of the next candidate's information is performed based on the updated rule base.

Images