Evaluation question and post competency analysis system based on grey correlation analysis

Abstract

The invention discloses an evaluation question and post competency analysis system based on grey correlation analysis, belongs to the technical field of competency analysis, and solves the problems that an existing method is independent in evaluation of each dimension, is lack of quantitative analysis of interaction and relevance among the dimensions, and cannot accurately analyze a competency and post matching relationship. The system comprises a question generation module, a test execution module, a data acquisition module and a correlation analysis module, wherein the correlation analysis module performs multi-dimensional interaction analysis on a competency information set based on a multi-dimensional analysis model; according to the invention, a correlation analysis module carries out multi-dimensional interaction analysis on a competency information set based on a multi-dimensional analysis model, the multi-dimensional analysis model calculates a correlation coefficient of test correlation data and post correlation information through gray correlation analysis, and a combination weighting method and a category joint countermeasure mechanism are combined. And the correlation degree between the capability of each dimension and the post competency is quantified, and accurate evaluation and analysis of the competency and the post are realized.

Description

Technical Field

[0001] This invention belongs to the field of competency analysis technology, specifically involving assessment items and a job competency analysis system based on grey relational analysis. Background Technology

[0002] As modern enterprise human resource management transforms towards refinement and scientification, accurately identifying job competency elements and developing highly matched assessment tools has become a core requirement for enterprise talent selection, training, and performance improvement. As a core technology of human resource management, competency analysis aims to extract the key abilities, knowledge, skills, and traits (communication skills, problem-solving skills, leadership, etc.) required for a job through a systematic approach. As the carrier of competency measurement, the degree to which the design of assessment items matches job competencies directly affects the reliability, validity, and practical application value of the assessment results.

[0003] Traditional job competency analysis primarily relies on methods such as Behavioral Event Interviews (BEI), the Delphi method, and questionnaires. These methods summarize key competency elements by inductively identifying high-frequency behavioral characteristics or drawing on expert experience. Based on this, assessment item development is often experience-driven, generating items through manual design or adaptation of established scales. Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) are then used to verify the correlation between items and competency dimensions. While these methods are widely used in practice, they still suffer from limitations such as reliance on expert experience to determine the correlation between items and competencies, and poor adaptability to small samples or incomplete information.

[0004] Chinese patent CN115511295A discloses a talent evaluation model and method based on professional competence, core competence, personality, appearance, and integrity. The method includes establishing an objective talent evaluation model based on personality, appearance, integrity, and their inherent qualities. This model includes evaluations of professional competence, core competence, personality, appearance, and integrity. The professional competence evaluation objectively assesses candidates' professional abilities, including but not limited to scoring professional learning ability, professional job skills, work experience, and educational background, thus establishing a competency model. However, existing talent evaluation models evaluate based on five dimensions: professional competence, core competence, personality, appearance, and integrity. These dimensions are evaluated independently and lack quantitative analysis of the interactions and correlations between them, making it difficult to accurately analyze the matching relationship between competency and job requirements. To address these issues, we propose an assessment question and job competency analysis system based on grey relational analysis. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by providing an assessment question and job competency analysis system based on grey relational analysis. This solves the problem that existing methods evaluate each dimension independently and lack quantitative analysis of the interaction and correlation between dimensions, thus failing to accurately analyze the relationship between competency and job matching.

[0006] This invention is implemented as follows: a system for analyzing assessment items and job competency based on grey relational analysis, the system comprising: The question generation module obtains job-related information based on distributed nodes, indexes the assessment database through the job-related information, and generates a set of job assessment questions based on clustering algorithms combined with similarity measurement algorithms. The test execution module is used to obtain the job assessment question set, randomly generate real-time assessment questions based on the job assessment question set, and simultaneously collect the test-related data of the test takers; The data acquisition module is used to obtain test-related data of test takers on real-time assessment questions, as well as basic information of test takers and job-related information. It integrates the test-related data, basic information of test takers, and job-related information into a competency information set and uploads the competency information set to the assessment database. The correlation analysis module pre-constructs a multi-dimensional analysis model based on grey relational analysis combined with category joint adversarial analysis. Iteratively trains the multi-dimensional analysis model based on the modeling sample set, performs multi-dimensional interactive analysis on the competency information set based on the multi-dimensional analysis model, and outputs the competency analysis results.

[0007] Preferably, the question generation module includes: The distributed information collection unit connects to distributed data sources and collects multi-source heterogeneous job-related information based on distributed nodes. It then uses natural language processing technology to perform structured transformation and deduplication of the job-related information. The database indexing unit constructs the index structure of the assessment database based on the job-related information after structured transformation, and traverses the question metadata in the assessment database based on the index structure. It then performs index structure-question association according to the job information range of the job-related information to construct an adaptive question tree structure. The clustering analysis unit determines the clustering association dimension of the index structure and the adaptive question tree structure based on the hierarchical clustering analysis algorithm, sets the inter-cluster similarity threshold, and combines the similarity measurement algorithm to determine the question clusters in the adaptive question tree structure that exceed the inter-cluster similarity threshold; The question generation unit is used to obtain question clusters that exceed the inter-cluster similarity threshold, perform double verification on the question clusters, generate verified question clusters, and convert the question clusters into a set of job assessment questions that interact with job-related information.

[0008] Preferably, the method for constructing an adaptive question tree structure includes: Based on the index structure, the scope of job information related to the job is determined, the scope of job information is mapped to the knowledge points to be covered, and a knowledge point set composed of the knowledge points to be covered is generated. Obtain the knowledge point set, traverse the question metadata in the assessment database based on the local search method, take the initial feasible solution as the first candidate solution, and perform iterative optimization by searching the neighborhood of the first candidate solution through the hill climbing strategy, and output the candidate solution set; Constructing a directed acyclic graph based on knowledge points and candidate solution sets using Bayesian models. ;in, This indicates the knowledge points being tested. Given a set of candidate solutions, we use knowledge point constraint functions to evaluate the quality of the candidate solution set in the directed acyclic graph (DAG), and calculate the candidate solution score for the DAG. The knowledge point constraint function is expressed as: (1) (2) in, Scoring candidate solutions Let be the prior probability of a directed acyclic graph. Represents a node The number of possible values, In the Bayesian model network, the first 1 node For the first The associated nodes of each node , For the first Individual nodes, associated nodes The first interaction 1 candidate point The total number of candidate points. For the first Individual nodes, associated nodes Interactivity level; Obtain the candidate solution scores of the directed acyclic graph, determine whether the candidate solution scores exceed the preset score filtering threshold, and retain the candidate solutions of the directed acyclic graph if the candidate solution scores exceed the preset score filtering threshold. If the candidate solution score does not exceed the preset score screening threshold, delete the candidate solution of the directed acyclic graph; Integrate the directed acyclic graph after deleting candidate solutions, determine the index structure-question association in the directed acyclic graph, and establish a hierarchical structure of the knowledge points to be examined and the candidate solution set based on the index structure-question association; Traverse the directed acyclic graph after deleting candidate solutions, and use the knowledge points and candidate solution set associated with the optimal solution of the directed acyclic graph as the root node of the adaptive problem tree structure. Then, use the root node as the starting point of the adaptive problem tree structure to associate the hierarchical structure of the knowledge points and candidate solution set to construct the adaptive problem tree structure.

[0009] Preferably, the method for determining the clustering association dimension of the index structure and adaptive question tree structure based on the hierarchical clustering analysis algorithm includes: Set up a clustering mechanism for hierarchical clustering analysis, which includes an extraction engine for clustering extraction rules. The extraction engine includes an extraction rule base, an association dimension evaluator, and an extraction executor. Based on the rule base extraction, the adaptive question tree structure is traversed, and the feature dimensions of the question metadata in the adaptive question tree are extracted. The feature dimensions of the question metadata include competency dimension, job tag, question attribute, score performance, and semantic similarity dimension. The feature dimensions of the question metadata are normalized to obtain the feature dimension set. Load the feature dimension set, use the feature dimension as the driving force and combine hierarchical clustering to derive question clusters based on different feature dimensions from the adaptive question tree structure. When deriving question clusters based on different feature dimensions, the nodes of the adaptive question tree structure are regarded as the initial clusters of the feature dimension question clusters. The inter-cluster similarity matrix between the initial clusters and the feature dimensions in the feature dimension set is calculated based on cosine similarity. The inter-cluster similarity is calculated through the inter-cluster similarity matrix. The association dimension evaluator starts with the initial cluster, and then maps the merged initial cluster to the question clusters with the highest similarity between the merged clusters and the initial clusters based on the merging strategy. A cluster similarity threshold is set based on historical experience, and its rationality is verified by combining it with the silhouette coefficient. The formula for calculating the similarity threshold is as follows: (3) in, For similarity threshold, This represents the average distance between the metadata of questions within the question cluster of the current feature dimension. This represents the average distance between question metadata within a question cluster based on other feature dimensions. The number of feature dimensions; At least one set of question clusters is obtained. The executor extracts question clusters that exceed the inter-cluster similarity threshold in the adaptive question tree structure based on Simrank++ similarity and pushes the question clusters that exceed the inter-cluster similarity threshold to the question generation unit.

[0010] Preferably, the method for double-checking the question cluster includes: Load the question cluster, perform test-retest reliability verification on the dimensional attributes of the question cluster, and determine whether the test-retest reliability of the question cluster exceeds the preset reliability threshold. The test-retest reliability is calculated based on the Pearson product-moment correlation coefficient, and the formula for calculating the test-retest reliability is as follows: (4) in, For test-retest reliability, Testers First test score, average first test score, second test score, average second test score. For the number of testers; If the test-retest reliability of the item cluster exceeds the preset reliability threshold, the factor analysis verification mechanism is triggered to perform validity testing on the item metadata of the item cluster and determine whether the average validity of the item metadata of the item cluster exceeds the preset validity threshold. If the average validity of the item metadata of the item cluster exceeds the preset validity threshold, the double validation of the item cluster is passed.

[0011] Preferably, the correlation analysis module includes: The model building unit derives a modeling sample set from the evaluation database and pre-builds a multi-dimensional analysis model based on grey relational analysis combined with category joint adversarial analysis. Iterative training of the multi-dimensional analysis model is then performed based on the modeling sample set. The interactive analysis unit is used to acquire the competency information set, take the competency information set as input, perform multi-dimensional interactive analysis on the competency information set based on the multi-dimensional analysis model, and output the competency analysis results. The visualization output unit is used to generate a competency analysis report based on the competency analysis results and to present the competency analysis report in a visual format.

[0012] Preferably, when pre-constructing the multi-dimensional analysis model based on grey relational analysis combined with category adversarial analysis, a category adversarial network is used as the initial model. The initial model also includes an input layer and an output layer. An autoencoder is embedded in the input layer. This input layer is used to process multi-source heterogeneous competency information sets, perform noise reduction and normalization on the competency information sets, autoencode the basic information of test takers in the competency information sets, establish interaction between test-related data and job-related information based on the autoencoded labels, and activate the question-competency association analysis network of the grey relational analysis module based on the autoencoded labels. A grey relational analysis module is introduced between the category adversarial network and the input layer. This grey relational analysis module incorporates a grey relational analysis algorithm. The question-competency association analysis network of the grey relational analysis module calculates the test-related data and job-related information based on the grey relational analysis algorithm. The correlation coefficient of information is weighted based on a combined weighting method, and the weighted correlation coefficients are summed to obtain the initial correlation degree between the tester and the job. The class adversarial network includes a feature extractor, a class determiner, and a feature fusion layer. The feature fusion layer introduces a multilayer perceptron network and an attention mechanism. The feature extractor extracts a correlation latent vector that integrates multi-dimensional information based on the autoencoded label, job correlation information, and the initial correlation degree. The class determiner corrects the initial correlation degree based on the correlation latent vector, generates a corrected correlation degree, and determines the true classification label of competency based on the corrected correlation degree. The feature fusion layer performs a nonlinear transformation on the corrected correlation degree based on the multilayer perceptron network, extracts the high-order interaction features of the corrected correlation degree, and fuses the high-order interaction features based on the attention mechanism to output the competency interaction degree. The competency interaction degree is used as the competency analysis result.

[0013] Preferably, the multi-dimensional analysis model training method includes: Obtain the modeling sample set, remove duplicate and missing validation samples from the modeling sample set, perform one-hot encoding on the modeling samples in the modeling sample set, and divide the encoded modeling sample set into training set and test set. Load a pre-built multi-dimensional analysis model, and preset the multi-dimensional analysis model training rounds, hyperparameters, comprehensive loss function and model optimizer; Acquire the training set, identify the job-related information in the training set, use the job-related information as the reference sequence for the initialization of grey relational analysis, use the tester's basic information as the prior information, and use the test-related data as the comparison sequence. Calculate the initial correlation matrix between the comparison sequence and the reference sequence, and use the initial correlation matrix as the input information for the class adversarial network. Obtain the initial association matrix, train the feature extractor and class determiner based on the initial association matrix, and train the feature extractor and class determiner alternately. During training, adjust the parameters of the grey relational analysis module and the class adversarial network based on the Adam optimizer. Determine whether the prediction accuracy of the competency interaction degree exceeds the preset accuracy threshold for five consecutive rounds. If it exceeds the preset accuracy threshold, stop the training of the multi-dimensional analysis model. Obtain the test set, use the test set as input, execute the multi-dimensional analysis model, output the average interaction degree of the test results, determine whether the average interaction degree exceeds the preset interaction threshold, and if the average interaction degree exceeds the preset interaction threshold, output the converged multi-dimensional analysis model.

[0014] Preferably, the method for multi-dimensional interactive analysis of competency information sets based on a multi-dimensional analysis model includes: Acquire a competency information set, self-encode the basic information of test takers in the competency information set, establish interaction between test-related data and job-related information based on the self-encoded tags, and activate the question-competency association analysis network of the grey relational analysis module based on the self-encoded tags. The topic is "Competency Association Analysis Network". Based on the grey relational analysis algorithm, the association coefficient between test data and job information is calculated. The association coefficient is weighted by the combination weighting method to obtain the association coefficient weight. The weighted association coefficients are summed to obtain the initial association degree between the tester and the job. The initial correlation degree is calculated using the following formula: (5) (6) (7) (8) in, These represent the initial correlation degree, correlation coefficient weight, and correlation coefficient, respectively. For related knowledge points of job-related information, These are the subjective and objective weights of the correlation coefficient, respectively. The weights are the subjective weights. These are the information entropy of related knowledge points and the number of related knowledge points, respectively. These are sets of related knowledge points for test-related data and interaction-related job information. These refer to the degree of interaction between testers and test-related data and job-related information, respectively. These represent the average interaction level of testers with test-related data and job-related information, respectively. Based on self-encoded tags, job-related information, and initial relevance, a latent vector of association is extracted by integrating multi-dimensional information. The initial correlation degree is corrected based on the latent correlation vector, a corrected correlation degree is generated, and the true classification label of competency is determined based on the corrected correlation degree. The correction correlation is nonlinearly transformed based on a multilayer perceptron network to extract high-order interaction features of the correction correlation. These high-order interaction features are then fused based on an attention mechanism to output the competency interaction degree, which is used as the competency analysis result.

[0015] Compared with the prior art, the embodiments of this application have the following main advantages: In this embodiment of the invention, the correlation analysis module performs multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model. The multi-dimensional analysis model calculates the correlation coefficient between test correlation data and job correlation information through grey relational analysis, and combines the combined weighting method and the category joint adversarial mechanism to quantify the correlation between each dimension of ability and job competency, thereby achieving accurate assessment and analysis of competency and job. Moreover, the multi-dimensional analysis model can be quickly adjusted according to changes in job requirements, continuously maintaining the accuracy and generalization ability of the analysis.

[0016] In this embodiment of the invention, the system is configured with a question generation module, which consists of a distributed information acquisition unit, a database indexing unit, a clustering analysis unit, and a question generation unit. The distributed information acquisition unit, database indexing unit, clustering analysis unit, and question generation unit work together to address the problem that traditional assessment question generation often relies on a single data source, making it difficult to cover multi-source heterogeneous data such as enterprise HR systems, recruitment platforms, and historical assessment databases. The distributed information acquisition unit connects to a MySQL cluster and Hadoop... HDFS, NoSQL databases, and API interfaces enable the full collection of job-related information. Simultaneously, the database indexing unit constructs a multi-dimensional index structure for the assessment database based on the structured job-related information. By traversing the association between question metadata and job information range, an adaptive question tree structure is generated. This adaptive question tree structure accurately matches job requirements with assessment questions, ensuring a high degree of relevance between the generated question set and job competency requirements. Furthermore, the clustering analysis unit groups highly similar questions together to form high-quality question clusters, further guaranteeing the objectivity and dynamic adaptability of question selection. It also provides test takers with a personalized assessment experience. The questions faced by test takers during the assessment process are carefully selected and generated based on the specific requirements of the applied or current job, making the assessment results more targeted and valuable.

[0017] In this embodiment of the invention, when constructing the adaptive question tree structure, the scope of job information is determined based on the index structure, mapping job requirements to specific knowledge points to be examined, thereby achieving accurate mapping of job information. Simultaneously, a local search hill-climbing strategy generates a high-quality candidate solution set, reducing invalid searches, ensuring question generation efficiency, and improving the relevance between candidate solutions and knowledge points. A scoring mechanism eliminates candidate solutions with incomplete knowledge point coverage or weak question relevance, improving the reliability of candidate solutions. Finally, the root node is the knowledge points and candidate solution set associated with the optimal solution in the directed acyclic graph, constructing the adaptive question tree structure. This adaptive question tree structure can dynamically adjust according to changes in job requirements, ensuring that the question tree always remains consistent with the job competency requirements. This not only improves the flexibility and adaptability of the assessment questions but also better meets the assessment needs of different positions and scenarios.

[0018] In this embodiment of the invention, when determining the clustering association dimension of the index structure and the adaptive question tree structure based on the hierarchical clustering analysis algorithm, the rule-driven mechanism and multi-dimensional feature extraction enable the clustering results to be traceable and interpretable, thereby avoiding "black box" operations during question cluster generation. The full link criterion and silhouette coefficient verification ensure that questions within a cluster are highly correlated and that differences between clusters are significant, improving the matching accuracy between questions and positions. The extraction executor determines question clusters that exceed the inter-cluster similarity threshold based on Simrank++ similarity, ensuring that the question clusters finally pushed to the question generation unit have high quality and high reliability.

[0019] In this embodiment of the invention, when performing multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model, the key features of the competency information set are learned through the input layer and compressed into a low-dimensional vector. This retains core information while removing redundancy, improving the efficiency and accuracy of subsequent analysis. The question-competency association analysis network, based on the grey relational analysis algorithm, calculates the association coefficient between test-related data and job-related information. This process, through "grey generation," processes the original data, revealing hidden nonlinear associations while balancing subjective preferences and objective laws, avoiding the bias of single weighting. Finally, a nonlinear transformation of the corrected association degree is performed based on a multilayer perceptron (MLP). High-order interactive features are fused through an attention mechanism to output the competency interaction degree. This allows the MLP to capture complex interactive relationships in the corrected association degree through multilayer nonlinear activation functions, thus avoiding the simplification problem of linear models. The attention mechanism automatically focuses on the features most important to the job, avoiding interference from irrelevant features and improving the relevance of the analysis results. The final output of the competency interaction degree, which measures comprehensive competency, not only reflects the overall matching degree between the tester and the job but also clarifies the key driving factors through the attention mechanism, making it easier for business personnel to understand. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the structure of the assessment questions and job competency analysis system based on grey relational analysis provided by the present invention.

[0021] Figure 2 The diagram illustrates the implementation process of constructing an adaptive question tree structure.

[0022] Figure 3 The diagram illustrates the implementation process of a clustering association dimension method based on hierarchical clustering analysis algorithm to determine the index structure and adaptive question tree structure.

[0023] Figure 4 The diagram illustrates the implementation process of a dual-validation method for question clusters.

[0024] Figure 5 A schematic diagram illustrating the implementation process of the multi-dimensional analysis model training method is shown.

[0025] Figure 6 This paper illustrates the implementation process of a multi-dimensional interactive analysis method for competency information sets based on a multi-dimensional analysis model. Detailed Implementation

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0027] Existing methods evaluate each dimension independently and lack quantitative analysis of the interactions and correlations between dimensions, failing to accurately analyze the relationship between competence and job matching. To address these issues, we propose a system for analyzing assessment questions and job competence based on grey relational analysis. This system comprises a question generation module 100, a test execution module 200, a data acquisition module 300, and a correlation analysis module 400. During operation, the question generation module 100 first acquires job-related information based on distributed nodes, indexes the assessment database using this information, and generates a set of job assessment questions based on a clustering algorithm combined with a similarity measurement algorithm. Then, the test execution module 200 acquires the job assessment questions. The system first generates real-time assessment questions randomly based on a set of job assessment questions, and simultaneously collects test-related data from test takers. Then, the data acquisition module 300 obtains the test-related data of test takers on the real-time assessment questions, as well as the test taker's basic information and job-related information. It integrates the test-related data, test taker's basic information, and job-related information into a competency information set, and uploads the competency information set to the assessment database. Finally, the correlation analysis module 400 pre-constructs a multi-dimensional analysis model based on grey relational analysis combined with category joint adversarial analysis. Iteratively trains the multi-dimensional analysis model based on the modeling sample set, performs multi-dimensional interactive analysis on the competency information set based on the multi-dimensional analysis model, and outputs the competency analysis results. In this embodiment of the invention, the correlation analysis module 400 performs multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model. The multi-dimensional analysis model calculates the correlation coefficient between test correlation data and job correlation information through grey relational analysis, and combines the combined weighting method and the category joint adversarial mechanism to quantify the correlation between each dimension of ability and job competency, thereby achieving accurate assessment and analysis of competency and job. Moreover, the multi-dimensional analysis model can be quickly adjusted according to changes in job requirements, continuously maintaining the accuracy and generalization ability of the analysis.

[0028] This invention provides a system for analyzing assessment questions and job competency based on grey relational analysis. Figure 1 A schematic diagram of a system for analyzing assessment items and job competency based on grey relational analysis is shown. This system specifically includes: The question generation module 100 obtains job-related information based on distributed nodes, indexes the assessment database through the job-related information, and generates a set of job assessment questions based on clustering algorithms combined with similarity measurement algorithms. In this embodiment of the invention, the question generation module 100 includes: The distributed information acquisition unit 110 connects to a distributed data source (including but not limited to MySQL clusters, Hadoop HDFS, NoSQL databases, and API interfaces), collects multi-source heterogeneous job-related information based on distributed nodes, and performs structured transformation and deduplication of the job-related information through natural language processing (NLP) technology. Database index unit 120 constructs the index structure of the assessment database based on the job association information after structure transformation, and traverses the question metadata in the assessment database based on the index structure. It performs index structure-question association according to the job information range of the job association information to construct an adaptive question tree structure. The index structure is a composite index structure of competency dimension-skill tag-question difficulty. Clustering analysis unit 130 determines the clustering association dimension of the index structure and the adaptive question tree structure based on the hierarchical clustering analysis algorithm, sets the inter-cluster similarity threshold, and combines the similarity measurement algorithm to determine the question clusters in the adaptive question tree structure that exceed the inter-cluster similarity threshold; The question generation unit 140 is used to obtain question clusters that exceed the inter-cluster similarity threshold, perform double verification on the question clusters, generate verified question clusters, and convert the question clusters into a set of job assessment questions that interact with job-related information.

[0029] In this embodiment of the invention, the system is configured with a question generation module 100, which consists of a distributed information acquisition unit 110, a database indexing unit 120, a clustering analysis unit 130, and a question generation unit 140. These units are connected via a local area network (LAN) or DTU communication to achieve data interaction. The coordinated operation of these units addresses the problem that traditional assessment question generation often relies on a single data source, making it difficult to cover multi-source heterogeneous data such as enterprise HR systems, recruitment platforms, and historical assessment databases. The distributed information acquisition unit 110 connects to a MySQL cluster (structured data) and Hadoop... HDFS (massive unstructured data), NoSQL database (semi-structured data), and API interface (real-time data) enable the full collection of job-related information. Meanwhile, the database indexing unit 120 constructs a multi-dimensional index structure for the assessment database based on the structured job-related information. By traversing the association between question metadata and job information range, an adaptive question tree structure is generated. This adaptive question tree structure accurately matches job requirements with assessment questions, ensuring that the generated question set is highly relevant to job competency requirements. The clustering analysis unit 130 groups highly similar questions together to form high-quality question clusters, further guaranteeing the objectivity and dynamic adaptability of question selection. It also provides test takers with a personalized assessment experience. The questions faced by test takers during the assessment process are carefully selected and generated based on the specific requirements of the applied or current job, making the assessment results more targeted and valuable.

[0030] The test execution module 200 is used to acquire the job assessment question set, randomly generate real-time assessment questions based on the job assessment question set, and simultaneously collect the test-related data of the test takers; The data acquisition module 300 is used to acquire test-related data of testers on real-time assessment questions, as well as basic information of testers and job-related information. It integrates the test-related data, basic information of testers, and job-related information into a competency information set and uploads the competency information set to the assessment database. The correlation analysis module 400 pre-constructs a multi-dimensional analysis model based on grey relational analysis combined with category joint adversarial analysis. Iteratively trains the multi-dimensional analysis model based on the modeling sample set, performs multi-dimensional interactive analysis on the competency information set based on the multi-dimensional analysis model, and outputs the competency analysis results.

[0031] In this embodiment of the invention, the correlation analysis module 400 includes: Model building unit 410 derives a modeling sample set from the evaluation database and pre-builds a multi-dimensional analysis model based on grey relational analysis combined with category joint adversarial analysis. Iterative training of the multi-dimensional analysis model is then performed based on the modeling sample set. The interactive analysis unit 420 is used to acquire a competency information set, take the competency information set as input, perform multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model, and output the competency analysis results. The visualization output unit 430 is used to generate a competency analysis report based on the competency analysis results and to visualize the competency analysis report.

[0032] In this embodiment, the model building unit 410, the interactive analysis unit 420, and the visualization output unit 430 can be connected via Bluetooth or 5G communication. These units cooperate collaboratively. When the interactive analysis unit 420 performs multi-dimensional interactive analysis on the competency information set based on the multi-dimensional analysis model, it introduces grey relational analysis to handle small sample and non-linear data, and enhances the model's robustness to multi-dimensional features based on a category joint adversarial mechanism. Furthermore, multilayer perceptron (MLP) and attention mechanisms can extract high-order interactive features to correct correlations, further improving the accuracy and generalization ability of the analysis.

[0033] In this embodiment, the question generation module 100, test execution module 200, data acquisition module 300, and correlation analysis module 400 are connected via Bluetooth or 5G communication. Furthermore, the assessment database is a structured data storage and management platform designed for talent assessment scenarios. The assessment database can collect, store, organize, and retrieve multi-source heterogeneous data related to the assessment. It stores collected test-related data, tester basic information, and job-related information. Test-related data includes, but is not limited to, test results (question scores, accuracy, answer time), test process behavior (number of skipped questions, number of answer modifications), and test environment information (testing device (PC / mobile), network status (latency), and other contextual data affecting test performance). Tester basic information includes, but is not limited to, tester age, name, education background, work experience, skill tags, and historical assessment records. Job-related information includes, but is not limited to, competency requirements / rules, basic job attributes, and dynamic job requirements. Additionally, the assessment database contains question metadata, including question content, assessed abilities, and related dimension tags.

[0034] In this embodiment of the invention, the correlation analysis module 400 performs multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model. The multi-dimensional analysis model calculates the correlation coefficient between test correlation data and job correlation information through grey relational analysis, and combines the combined weighting method and the category joint adversarial mechanism to quantify the correlation between each dimension of ability and job competency, thereby achieving accurate assessment and analysis of competency and job. Moreover, the multi-dimensional analysis model can be quickly adjusted according to changes in job requirements, continuously maintaining the accuracy and generalization ability of the analysis.

[0035] This invention provides a method for constructing an adaptive question tree structure. Figure 2 The diagram illustrates the implementation flow of a method for constructing an adaptive problem tree structure. The method specifically includes: S101, Based on the index structure, determine the scope of job information related to the job, map the scope of job information to the knowledge points to be covered, and generate a knowledge point set composed of the knowledge points to be covered; When determining the knowledge points to be covered, abstract job competency requirements (such as "possessing data mining skills") can be transformed into actionable knowledge points (such as "application of random forest models"), preventing questions from deviating from the core job requirements. For example, if the job information is "data analyst," it needs to cover "deep learning," "SQL," and "data analysis," while the corresponding knowledge points to be covered are "convolutional neural network models," "window functions," and "A / B test design." An index structure ensures that the knowledge point set covers all key competency dimensions of the job, avoiding the omission of important assessment points.

[0036] S102: Obtain the knowledge point set. Based on the local search method, traverse the question metadata in the evaluation database, take the initial feasible solution as the first candidate solution, and use the hill climbing strategy to search the neighborhood of the first candidate solution for iterative optimization, and output the candidate solution set. It should be noted that the local search method avoids the dependence on a single initial solution, and explores better solutions through neighborhood iteration (replacing questions and adjusting the order), thereby improving the quality of candidate solutions. The candidate solution set contains questions of different difficulties, dimensions and types, thus providing data support for building a multi-level adaptive question tree structure.

[0037] S103, Constructing a directed acyclic graph based on knowledge points and candidate solution sets in conjunction with a Bayesian model. Directed acyclic graphs (DAGs) can intuitively represent the logical dependencies between knowledge points, avoiding jumps in knowledge points when generating questions. This indicates the knowledge points being tested. Given a set of candidate solutions, a knowledge point constraint function is used to evaluate the quality of the candidate solution set in the directed acyclic graph (DAG). The candidate solution score for the DAG is calculated. In this embodiment, the knowledge point constraint function quantifies the rationality of the candidate solutions using prior probabilities and interaction metrics. The knowledge point constraint function is expressed as: (1) (2) in, Scoring candidate solutions Let be the prior probability of a directed acyclic graph. Represents a node The number of possible values, In the Bayesian model network, the first 1 node For the first The associated nodes of each node , For the first Individual nodes, associated nodes The first interaction 1 candidate point The total number of candidate points. For the first Individual nodes, associated nodes Interactivity level; S104, obtain the candidate solution scores of the directed acyclic graph, and determine whether the candidate solution scores exceed the preset score screening threshold. The score screening threshold can be 0.8-0.85. The score screening threshold can be set based on historical data or expert experience to ensure the objectivity and interpretability of the screening criteria. S105, If the score of a candidate solution exceeds the preset score screening threshold, retain the candidate solution of the directed acyclic graph. S106. If the candidate solution score does not exceed the preset score screening threshold, delete the candidate solution of the directed acyclic graph. S107 integrates the directed acyclic graph after deleting candidate solutions, determines the index structure-question association relationship in the directed acyclic graph, and establishes a hierarchical structure of the knowledge points to be examined and the candidate solution set based on the index structure-question association relationship. The root node is the knowledge point associated with the optimal solution, which can ensure that the core of the adaptive question tree structure revolves around the most core competencies of the job. When the job requirements change, the adaptive question tree structure can be quickly adjusted by updating the root node, making the adaptive question tree structure clear and ensuring the dynamic adaptability of the adaptive question tree structure. S108. Traverse the directed acyclic graph after deleting candidate solutions. Use the knowledge points and candidate solution set associated with the optimal solution of the directed acyclic graph as the root node of the adaptive problem tree structure. Then, use the root node as the starting point of the adaptive problem tree structure to associate the hierarchical structure of the knowledge points and candidate solution set to construct the adaptive problem tree structure.

[0038] In this embodiment of the invention, when constructing the adaptive question tree structure, the scope of job information is determined based on the index structure, mapping job requirements to specific knowledge points to be examined, thereby achieving accurate mapping of job information. Simultaneously, a local search hill-climbing strategy generates a high-quality candidate solution set, reducing invalid searches, ensuring question generation efficiency, and improving the relevance between candidate solutions and knowledge points. A scoring mechanism eliminates candidate solutions with incomplete knowledge point coverage or weak question relevance, improving the reliability of candidate solutions. Finally, the root node is the knowledge points and candidate solution set associated with the optimal solution in the directed acyclic graph, constructing the adaptive question tree structure. This adaptive question tree structure can dynamically adjust according to changes in job requirements, ensuring that the question tree always remains consistent with the job competency requirements. This not only improves the flexibility and adaptability of the assessment questions but also better meets the assessment needs of different positions and scenarios.

[0039] This invention provides a method for determining the clustering association dimension of the index structure and adaptive question tree structure based on a hierarchical clustering analysis algorithm. Figure 3 This diagram illustrates the implementation flow of a method for determining the clustering association dimensions of the index structure and adaptive question tree structure based on a hierarchical clustering analysis algorithm. The method specifically includes: S201, set up a clustering mechanism for hierarchical clustering analysis. The clustering mechanism includes an extraction engine for clustering extraction rules. The extraction engine includes an extraction rule base, an association dimension evaluator, and an extraction executor. The extraction rule base includes, but is not limited to, clustering target rules, feature dimension rules, extraction process rules, and evaluation rules. The extraction rule base supports dynamic adjustment of rules according to different job types (such as technical positions and management positions) or assessment scenarios (such as campus recruitment and social recruitment).

[0040] In this embodiment of the invention, the association dimension evaluator can dynamically evaluate the contribution of feature dimensions to clustering, thereby providing a basis for subsequent feature selection. The extraction executor supports flexible invocation of multiple hierarchical clustering algorithms (AGNES, DIANA) to adapt to the clustering needs of different positions. Among them, the clustering needs of different positions include, but are not limited to, fine clustering for technical positions and coarse clustering for management positions.

[0041] S202, based on the rule base extraction and traversal of the adaptive question tree structure, extracts the feature dimensions of the question metadata in the adaptive question tree. The feature dimensions of the question metadata include competency dimension, job tag, question attribute, score performance, and semantic similarity dimension. The feature dimensions of the question metadata are normalized to ensure that the clustering algorithm can fairly compare features of different dimensions and obtain a feature dimension set. S203, Load the feature dimension set, and using the feature dimensions as the driving force and combined with hierarchical clustering, derive question clusters based on different feature dimensions from the adaptive question tree structure. When deriving question clusters based on different feature dimensions, the nodes of the adaptive question tree structure are regarded as the initial clusters of the feature dimension question clusters. The inter-cluster similarity matrix between the initial clusters and the feature dimensions in the feature dimension set is calculated based on cosine similarity, and the inter-cluster similarity is calculated through the inter-cluster similarity matrix. In this embodiment of the invention, hierarchical clustering can gradually aggregate similar questions into hierarchical clusters through a merge-split mechanism, so that the hierarchical clusters conform to the difficulty of the questions and the natural distribution law. The inter-cluster similarity matrix and similarity are calculated based on cosine similarity. The cosine similarity matrix can intuitively reflect the degree of association between clusters, providing a clear quantitative basis for subsequent merging and screening.

[0042] S204, the association dimension evaluator starts with the initial cluster, and then maps the merged initial cluster to question clusters with different feature dimensions based on the cluster with the highest similarity between the merge strategy and the initial cluster. The merge strategy is based on the definition of the full link criterion in the extraction rule base. The full link criterion is to take the distance of the farthest sample between the two clusters (preferring to generate compact clusters, suitable for job-question matching scenarios). S205 sets an inter-cluster similarity threshold based on historical experience and verifies its rationality using silhouette coefficients. The silhouette coefficients quantify intra-cluster compactness and inter-cluster separation to verify whether the threshold can distinguish different clusters without compromising intra-cluster consistency, ensuring the threshold's scientific validity. The similarity threshold calculation formula is expressed as follows: (3) in, For similarity threshold, This represents the average distance between the metadata of questions within the question cluster of the current feature dimension. This represents the average distance between question metadata within a question cluster based on other feature dimensions. The number of feature dimensions; S206, obtain at least one set of question clusters, extract question clusters in the adaptive question tree structure that exceed the inter-cluster similarity threshold based on Simrank++ similarity, and push question clusters that exceed the inter-cluster similarity threshold to question generation unit 140. Simrank++ can calculate the similarity between new questions and existing clusters in real time and supports dynamic expansion of question clusters.

[0043] In this embodiment of the invention, when determining the clustering association dimension of the index structure and the adaptive question tree structure based on the hierarchical clustering analysis algorithm, the rule-driven mechanism and multi-dimensional feature extraction enable the clustering results to be traceable and interpretable, thereby avoiding "black box" operations during question cluster generation. The full link criterion and silhouette coefficient verification ensure that questions within a cluster are highly correlated and that differences between clusters are significant, improving the matching accuracy between questions and positions. The extraction executor determines question clusters that exceed the inter-cluster similarity threshold based on Simrank++ similarity, ensuring that the question clusters finally pushed to the question generation unit 140 have high quality and high reliability.

[0044] This invention provides a method for dual verification of question clusters. Figure 4 This diagram illustrates the implementation flow of a double-checking method for question clusters. The double-checking method for question clusters specifically includes: S301, Load the question cluster and perform test-retest reliability verification on the dimensional attributes of the question cluster. Test-retest reliability is an indicator that measures the consistency of test results of the question cluster at different time points and reflects the degree to which the questions are affected by random factors. S302, determine whether the test-retest reliability of the item cluster exceeds the preset reliability threshold. The test-retest reliability of the item cluster is calculated by the Pearson product-moment correlation coefficient. If the test-retest reliability is lower than the preset threshold (e.g., 0.7), it means that the item scores fluctuate greatly at different time points and cannot reliably reflect the true ability of the test takers. The test-retest reliability is calculated based on the Pearson product-moment correlation coefficient, and the formula for calculating the test-retest reliability is as follows: (4) in, For test-retest reliability, Testers First test score, average first test score, second test score, average second test score. For the number of testers; S303, If the test-retest reliability of a cluster of items exceeds a preset reliability threshold, trigger the factor analysis verification mechanism; If the test-retest reliability of a question cluster does not exceed a preset reliability threshold, a first warning message is triggered, and the system returns to step S201 to continue identifying question clusters that exceed the inter-cluster similarity threshold. If the test-retest reliability does not meet the standard, the system returns to step S201 to re-identify question clusters, forming a closed loop of verification-optimization-re-verification. This mechanism forces the question generation module 100 to continuously optimize question quality, ultimately ensuring the generation of stable question clusters.

[0045] S304. Validity is an indicator that measures whether an item cluster can accurately measure target competence and reflects the degree of matching between the items and job requirements. Validity testing can be performed using methods such as factor analysis (e.g., confirmatory factor analysis, CFA) to assess the explanatory power of the item metadata on target competence.

[0046] S305, Determine whether the average validity of the item data of the item cluster exceeds the preset validity threshold; S306. If the average validity of the item data of the item cluster exceeds the preset validity threshold, the double validation of the item cluster is passed.

[0047] If the average validity of the question data of a question cluster does not exceed the preset validity threshold, a second warning message is triggered, and the process returns to step S201 to continue identifying question clusters that exceed the inter-cluster similarity threshold.

[0048] In this embodiment of the invention, when performing dual verification on the question cluster, the dual screening mechanism of test-retest reliability verification and validity testing ensures that the question cluster is both stable and reliable, and can accurately measure the target competence. Furthermore, the closed-loop optimization characteristics and scientific evaluation characteristics of the dual verification ensure the high-quality output of the question cluster, providing key quality control support for the system's accurate assessment and dynamic adaptation to the target.

[0049] This invention provides a method for training a multi-dimensional analysis model. Figure 5 This diagram illustrates the implementation flow of a multi-dimensional analysis model training method, which specifically includes: S401, Obtain the modeling sample set, remove duplicate and missing validation samples from the modeling sample set, perform one-hot encoding on the modeling samples in the modeling sample set, and divide the encoded modeling sample set into a training set and a test set, wherein the ratio of the training set to the test set can be 7:1. S402 loads a pre-built multi-dimensional analysis model, pre-setting the training epochs, hyperparameters, comprehensive loss function, and model optimizer. The model training epochs can be 100-120, the comprehensive loss function can be the cross-entropy loss function, and the model optimizer can be the Adam optimizer. The Adam optimizer is used to adjust the parameters of the grey relational analysis module 400 and the class adversarial network, ensuring rapid convergence and stable optimization of the model during training. The Adam optimizer combines the advantages of multiple optimization algorithms, effectively improving the model's training efficiency and stability. S403, acquire the training set, identify the job-related information in the training set, use the job-related information as the reference sequence for the initialization of grey relational analysis, use the tester's basic information as the prior information, use the test-related data as the comparison sequence, calculate the initial correlation matrix between the comparison sequence and the reference sequence, and use the initial correlation matrix as the input information of the class adversarial network. S404: Obtain the initial correlation matrix. Train the feature extractor and class determiner based on the initial correlation matrix, and train the feature extractor and class determiner alternately. During training, adjust the parameters of the grey relational analysis module 400 and the class adversarial network based on the Adam optimizer. Determine whether the prediction accuracy of the competency interaction degree exceeds a preset accuracy threshold after five consecutive rounds. If it exceeds the preset accuracy threshold, stop training the multi-dimensional analysis model. In this embodiment of the invention, the feature extractor and class determiner are trained alternately, ensuring that the model achieves optimal performance in both the feature extraction and class determination stages. This alternating training method effectively avoids over-optimization of the model at a certain stage, improving the overall performance of the model. S405: Obtain the test set, use the test set as input, execute the multi-dimensional analysis model, and output the average interaction degree of the test results; S406, determine whether the average interactivity exceeds the preset interactivity threshold. The preset interactivity threshold can be 0.9-0.95. Evaluating the average interactivity through the preset interactivity threshold ensures that the competency interactivity output by the model has sufficient reliability. S407: If the average interactivity exceeds the preset interactivity threshold, output a converged multi-dimensional analysis model.

[0050] If the average interaction rate does not exceed the preset interaction threshold, return to S403 and continue iterative training of the model.

[0051] In this embodiment, when pre-constructing a multi-dimensional analysis model based on grey relational analysis combined with category adversarial analysis, a category adversarial network is used as the initial model. The initial model also includes an input layer and an output layer. A data autoencoder is embedded in the input layer. This input layer is used to process multi-source heterogeneous competency information sets, perform noise reduction and normalization on the competency information sets, autoencode the basic information of test takers in the competency information sets, establish interaction between test-related data and job-related information based on the autoencoded tags, and activate the question-competency correlation analysis network of the grey relational analysis module 400 based on the autoencoded tags. The grey relational analysis module 400 is introduced between the category adversarial network and the input layer. The grey relational analysis module 400 incorporates a grey relational analysis algorithm, and the question-competency correlation analysis network of the grey relational analysis module 400 calculates test correlations based on the grey relational analysis algorithm. The correlation coefficient between data and job-related information is weighted using a combined weighting method. The weighted correlation coefficients are then summed to obtain the initial correlation degree between the tester and the job. The class adversarial network includes a feature extractor, a class determiner, and a feature fusion layer. The feature fusion layer incorporates a multilayer perceptron network and an attention mechanism. The feature extractor extracts a correlation latent vector that integrates multi-dimensional information based on the autoencoded label, job-related information, and the initial correlation degree. The class determiner corrects the initial correlation degree based on the correlation latent vector, generates a corrected correlation degree, and determines the true competency classification label based on the corrected correlation degree. The feature fusion layer performs a nonlinear transformation on the corrected correlation degree based on the multilayer perceptron network, extracts high-order interaction features of the corrected correlation degree, and fuses the high-order interaction features based on the attention mechanism to output the competency interaction degree. The competency interaction degree is used as the competency analysis result.

[0052] This invention provides a training method for a multi-dimensional analysis model. The multi-dimensional analysis model is based on a class adversarial network and incorporates an input layer with an embedded data autoencoder. A grey relational analysis module 400 is introduced between the class adversarial network and the input layer. The grey relational analysis module 400 incorporates a grey relational analysis algorithm, thereby overcoming the problem that traditional talent assessment models rely on subjective experience or linear assumptions, making it difficult to quantify the complex relationship between test takers' abilities and job requirements. By calculating the correlation coefficient between test-related data and job-related information, the correlation strength between a single ability dimension and the job is quantified. Furthermore, the grey relational analysis algorithm introduced in the grey relational analysis module 400 can take into account the advantages of small sample adaptability, nonlinear relationship capture, and interpretability of result correlation. The class adversarial network consists of a feature extractor, a class determiner, and a feature fusion layer. Through multi-source information fusion, the feature extractor can capture the implicit correlation pattern between test takers and jobs, avoiding the one-sidedness of a single feature. The class determiner forces the generator to learn more discriminative features through adversarial training, improving the model's robustness to noise.

[0053] This invention provides a method for multi-dimensional interactive analysis of competency information sets based on a multi-dimensional analysis model. Figure 6 This diagram illustrates the implementation flow of a method for multi-dimensional interactive analysis of competency information sets based on a multi-dimensional analysis model. The method specifically includes: S501, acquire the competency information set, self-encode the basic information of the test takers in the competency information set, establish the interaction of test-related data and job-related information based on the self-encoded tags, and activate the question-competency association analysis network of the gray relational analysis module 400 based on the self-encoded tags. S502, the Item-Competency Association Analysis Network calculates the correlation coefficient between test data and job-related information based on the grey relational analysis algorithm. It assigns weights to these correlation coefficients using a combined weighting method, and then sums the weighted correlation coefficients to obtain the initial correlation degree between the test taker and the job. It's important to note that the Item-Competency Association Analysis Network is an analytical framework built on the grey relational analysis algorithm, used to calculate the correlation between assessment items and job competency requirements. By quantifying the relationship between assessment items and job competencies, it evaluates the contribution of each item to job competency, thereby generating a set of assessment items that highly matches job requirements.

[0054] The initial correlation degree is calculated using the following formula: (5) (6) (7) (8) in, These represent the initial correlation degree, correlation coefficient weight, and correlation coefficient, respectively. For related knowledge points of job-related information, These are the subjective and objective weights of the correlation coefficient, respectively. The weights are the subjective weights. These are the information entropy of related knowledge points and the number of related knowledge points, respectively. These are sets of related knowledge points for test-related data and interaction-related job information. These refer to the degree of interaction between testers and test-related data and job-related information, respectively. These represent the average interaction level of testers with test-related data and job-related information, respectively. S503 extracts the association latent vector by fusing multi-dimensional information based on self-encoded tags, job-related information, and initial association degree.

[0055] S504 corrects the initial correlation degree based on the latent correlation vector, generates a corrected correlation degree, and determines the true classification label of competency based on the corrected correlation degree. The correction of the initial correlation degree based on the latent correlation vector is achieved by a class determiner. The class determiner corrects the initial correlation degree based on the latent correlation vector, generates a corrected correlation degree, and determines the true classification label of competency. This can effectively improve the classification accuracy and reliability of the model and ensure that the output of the model has practical application value. S505 uses a multilayer perceptron network to perform a nonlinear transformation on the correction correlation, extracts the high-order interaction features of the correction correlation, and fuses the high-order interaction features based on an attention mechanism to output the competency interaction degree. The competency interaction degree is used as the competency analysis result. With the competency interaction degree as the final output, it provides a comprehensive and accurate analysis result for job competency assessment. The competency interaction degree can reflect the comprehensive matching degree between the tester and the job, and provides strong support for human resource management.

[0056] In this embodiment of the invention, when performing multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model, the key features of the competency information set are learned through the input layer and compressed into a low-dimensional vector. This retains core information while removing redundancy, improving the efficiency and accuracy of subsequent analysis. The question-competency association analysis network, based on the grey relational analysis algorithm, calculates the association coefficient between test-related data and job-related information. This process, through "grey generation," processes the original data, revealing hidden nonlinear associations while balancing subjective preferences and objective laws, avoiding the bias of single weighting. Finally, a nonlinear transformation of the corrected association degree is performed based on a multilayer perceptron (MLP). High-order interactive features are fused through an attention mechanism to output the competency interaction degree. This allows the MLP to capture complex interactive relationships in the corrected association degree through multilayer nonlinear activation functions, thus avoiding the simplification problem of linear models. The attention mechanism automatically focuses on the features most important to the job, avoiding interference from irrelevant features and improving the relevance of the analysis results. The final output of the competency interaction degree, which measures comprehensive competency, not only reflects the overall matching degree between the tester and the job but also clarifies the key driving factors through the attention mechanism, making it easier for business personnel to understand.

[0057] In summary, this invention provides an assessment question and job competency analysis system based on grey relational analysis. In this embodiment, the correlation analysis module 400 performs multi-dimensional interactive analysis on the competency information set based on a multi-dimensional analysis model. The multi-dimensional analysis model calculates the correlation coefficient between test correlation data and job correlation information through grey relational analysis, and combines the combined weighting method and the category joint adversarial mechanism to quantify the correlation degree between each dimension of ability and job competency, thereby achieving accurate assessment and analysis of competency and job. Moreover, the multi-dimensional analysis model can be quickly adjusted according to changes in job requirements, continuously maintaining the accuracy and generalization ability of the analysis.

[0058] It should be noted that, for the sake of simplicity, the foregoing embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to the present invention. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0059] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on these embodiments, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art can still combine, add, delete, or otherwise adjust the features of the various embodiments of the present invention according to the circumstances without conflict or creative effort, thereby obtaining different technical solutions that do not fundamentally depart from the concept of the present invention. These technical solutions also fall within the scope of protection of the present invention.

Claims

1. An evaluation question and post competency analysis system based on grey correlation analysis, characterized in that, The system comprises: The subject generation module obtains the post association information based on the distributed nodes, indexes the test database through the post association information, generates the post test question set based on the clustering algorithm combined with the similarity measurement algorithm; The test execution module is used for obtaining the post test question set, randomly generating real-time test questions based on the post test question set, and synchronously collecting the test association data of the testers; The data collection module is used for obtaining the test association data of the testers on the real-time test questions, obtaining the tester basic information and the post association information, integrating the test association data, the tester basic information and the post association information into the competency information set, and uploading the competency information set to the test database; The correlation analysis module pre-constructs a multi-dimensional analysis model based on the grey correlation analysis combined with the category joint confrontation, iteratively trains the multi-dimensional analysis model based on the modeling sample set, performs multi-dimensional interactive analysis on the competency information set based on the multi-dimensional analysis model, and outputs the competency analysis result.

2. The evaluation question and post competency analysis system based on grey correlation analysis according to claim 1, characterized in that: The subject generation module comprises: The distributed information collection unit is connected to the distributed data source, collects multi-source heterogeneous post association information based on the distributed nodes, and structures and de-duplicates the post association information through natural language processing technology; The database index unit constructs the index structure of the test database based on the structured post association information, traverses the question metadata in the test database based on the index structure, indexes the structure-question based on the post information range of the post association information, and constructs the adaptive question tree structure; The clustering analysis unit determines the clustering correlation dimension of the index structure and the adaptive question tree structure based on the hierarchical clustering analysis algorithm, sets the inter-cluster similarity threshold, and determines the question cluster exceeding the inter-cluster similarity threshold in the adaptive question tree structure combined with the similarity measurement algorithm; The question generation unit is used for obtaining the question cluster exceeding the inter-cluster similarity threshold, double-checking the question cluster, generating the checked question cluster, and converting the question cluster into the post test question set interacting with the post association information.

3. The evaluation question and post competency analysis system based on grey correlation analysis of claim 2, characterized in that: The method for constructing the adaptive question tree structure comprises: Based on the index structure, the post information range of the post association information is determined, the post information range is mapped to the knowledge points to be covered, and a knowledge point set composed of the knowledge points to be covered is generated; The knowledge point set is obtained, the question metadata in the test database is searched and traversed based on the local search method, the initial feasible solution is taken as the first candidate solution, the first candidate solution neighborhood is searched and iteratively optimized through the hill climbing strategy, and the candidate solution set is output; A directed acyclic graph based on the knowledge points in the knowledge point set and the candidate solution set is constructed combined with the Bayesian model, the knowledge point constraint function is used to evaluate the advantages and disadvantages of the candidate solution set in the directed acyclic graph, and the candidate solution score of the directed acyclic graph is calculated; The candidate solution score of the directed acyclic graph is obtained, it is judged whether the candidate solution score exceeds the preset score screening threshold, if the candidate solution score exceeds the preset score screening threshold, the candidate solution of the directed acyclic graph is retained; If the candidate solution score does not exceed the preset score screening threshold, the candidate solution of the directed acyclic graph is deleted. The index structure-question association relationship in the directed acyclic graph after the candidate solution is deleted is determined, and a hierarchical structure of the examination knowledge points and the candidate solution set is established based on the index structure-question association relationship; The directed acyclic graph after the candidate solution is deleted is traversed, the examination knowledge points and the candidate solution set associated with the optimal solution of the directed acyclic graph are taken as a root node of the adaptive question tree structure, and the hierarchical structure of the examination knowledge points and the candidate solution set is associated with the root node as a starting point of the adaptive question tree structure, so as to construct the adaptive question tree structure.

4. The evaluation question and post competency analysis system based on grey correlation analysis of claim 3, characterized in that: The method for determining the clustering association dimension of the index structure and the adaptive question tree structure based on the hierarchical clustering analysis algorithm comprises: A cluster clustering mechanism of hierarchical clustering analysis is set, wherein the cluster clustering mechanism comprises an extraction engine of a cluster clustering extraction rule, and the extraction engine comprises an extraction rule library, an association dimension evaluator, and an extraction executor; The adaptive question tree structure is traversed based on the extraction rule library, the feature dimension of the question metadata in the adaptive question tree is extracted, the feature dimension of the question metadata is normalized, and a feature dimension set is obtained; The feature dimension set is loaded, the feature dimension is taken as a driving factor, and a question cluster based on different feature dimensions is derived from the adaptive question tree structure by combining the hierarchical clustering method, wherein when the question cluster based on different feature dimensions is derived, the nodes of the adaptive question tree structure are regarded as initial clusters of the feature dimension question cluster, a cluster inter-similarity matrix of the initial cluster and the feature dimensions in the feature dimension set is calculated based on cosine similarity, and the cluster inter-similarity is calculated through the cluster inter-similarity matrix; The association dimension evaluator starts with the initial cluster, merges the cluster with the highest inter-similarity of the initial cluster based on a merging strategy, and maps the merged initial cluster to the question cluster of different feature dimensions; A cluster inter-similarity threshold is set through historical experience; At least one group of question clusters is obtained, the extraction executor determines the question clusters in the adaptive question tree structure that exceed the cluster inter-similarity threshold based on Simrank++ similarity, and the question clusters that exceed the cluster inter-similarity threshold are pushed to a question generation unit.

5. The evaluation question and post competency analysis system based on grey correlation analysis of claim 3, characterized in that: The method for double-checking the question cluster comprises: The question cluster is loaded, the dimension attribute of the question cluster is subjected to a test-retest reliability check, and whether the test-retest reliability of the question cluster exceeds a preset reliability threshold is determined; If the test-retest reliability of the question cluster exceeds the preset reliability threshold, a factor analysis check mechanism is triggered, the validity of the question metadata of the question cluster is verified, and whether the average validity of the question metadata of the question cluster exceeds a preset validity threshold is determined; If the average validity of the question metadata of the question cluster exceeds the preset validity threshold, the double verification of the question cluster is passed.

6. The evaluation question and post competency analysis system based on grey correlation analysis according to claim 1, characterized in that: The correlation analysis module comprises: A model construction unit that derives a modeling sample set based on an evaluation database, and pre-constructs a multi-dimensional analysis model based on gray correlation analysis combined with category joint confrontation, and iteratively trains the multi-dimensional analysis model based on the modeling sample set; An interactive analysis unit that is configured to obtain a set of competency information, take the set of competency information as input, perform multi-dimensional interactive analysis on the set of competency information based on the multi-dimensional analysis model, and output competency analysis results; A visual output unit that is configured to generate a competency analysis report according to the competency analysis results, and present the competency analysis report in a visual manner.

7. The evaluation question and post competency analysis system based on grey correlation analysis of claim 6, characterized in that: The pre-constructed multi-dimensional analysis model based on grey correlation analysis combined with category joint confrontation is an initial model, which further includes an input layer and an output layer, and a data auto-encoder is embedded in the input layer, a grey correlation analysis module is introduced between the category joint confrontation network and the input layer, the grey correlation analysis module introduces a grey correlation analysis algorithm, the category joint confrontation network includes a feature extractor, a category determinator and a feature fusion layer, and the feature fusion layer introduces a multi-layer perception network and an attention mechanism.

8. The evaluation question and post competency analysis system based on grey correlation analysis of claim 7, characterized in that: The multi-dimensional analysis model training method comprises: obtaining a modeling sample set, removing duplicate and missing verification samples from the modeling sample set, and performing one-hot encoding on the modeling samples in the modeling sample set, and dividing the encoded modeling sample set into a training set and a test set; loading the pre-constructed multi-dimensional analysis model, presetting the multi-dimensional analysis model training rounds, hyperparameters, comprehensive loss functions and model optimizers; obtaining the training set, identifying the job-related information in the training set, taking the job-related information as the reference sequence for the initialization of the grey correlation analysis, taking the tester's basic information as the prior information, and taking the test-related data as the comparison sequence, calculating the initial correlation matrix of the comparison sequence and the reference sequence, and taking the initial correlation matrix as the input information of the category joint confrontation network; obtaining the initial correlation matrix, training the feature extractor and the category determinator based on the initial correlation matrix, and alternately training the feature extractor and the category determinator, adjusting the parameters of the grey correlation analysis module and the category joint confrontation network based on the Adam optimizer during training, and determining whether the prediction accuracy of the competence interaction degree for five consecutive rounds exceeds the preset accuracy threshold, if the preset accuracy threshold is exceeded, the multi-dimensional analysis model training is stopped; obtaining the test set, taking the test set as the input, executing the multi-dimensional analysis model, outputting the average interaction degree of the test result, determining whether the average interaction degree exceeds the preset interaction threshold, if the average interaction degree exceeds the preset interaction threshold, outputting the converged multi-dimensional analysis model.

9. The evaluation question and post competency analysis system based on grey correlation analysis of claim 7, characterized in that: The multi-dimensional interaction analysis method for the competence information set based on the multi-dimensional analysis model comprises: obtaining the competence information set, auto-encoding the tester's basic information in the competence information set, establishing the interaction between the test-related data and the job-related information based on the auto-encoding label, and activating the question-competence correlation analysis network of the grey correlation analysis module based on the auto-encoding label; The question-competence correlation analysis network calculates the correlation coefficient of the test-related data and the job-related information based on the grey correlation analysis algorithm, weights the correlation coefficient based on the combination weighting method, obtains the correlation coefficient weight, and weights and sums the weighted correlation coefficient to obtain the initial correlation degree of the tester-job; extracting the correlation hidden vector of the multi-dimensional information based on the auto-encoding label, the job-related information and the initial correlation degree.

10. The evaluation question and post competency analysis system based on grey correlation analysis of claim 9, wherein: The multi-dimensional interaction analysis method for the competence information set based on the multi-dimensional analysis model further comprises: correcting the initial correlation degree based on the correlation hidden vector to generate a corrected correlation degree, and determining the real classification label of the competence based on the corrected correlation degree; The correction correlation degree is nonlinearly transformed based on a multilayer perception network, high-order interaction features of the correction correlation degree are extracted, and the high-order interaction features are fused based on an attention mechanism to output the competency interaction degree, so as to take the competency interaction degree as the competency analysis result.

Images