Post competency multi-dimensional accurate evaluation method based on machine learning

Abstract

The invention discloses a post competency multi-dimensional accurate evaluation method based on machine learning, particularly relates to the technical field of human resource management, and is used for solving the problem that an existing evaluation method cannot effectively capture non-linear interaction among competency dimensions, so that an evaluation result is not accurate. Identifying an interaction relationship between dimensions by adopting a machine learning model, when a predefined interaction relationship exists, analyzing compensation symmetry and compensation conductivity to identify a dynamic compensation mode, and calculating a compensation phase collaboration index through a time-frequency analysis method based on the dynamic compensation mode; and dynamically adjusting weight distribution of each dimension by adopting a nonlinear weighting strategy in combination with a dynamic compensation mode and a compensation phase cooperation index, and finally outputting a comprehensive competency score subjected to dynamic weight adjustment, thereby realizing accurate quantization and adaptive weight adjustment of a dynamic compensation effect between each competency dimension.

Description

Technical Field

[0001] This invention relates to the field of human resource management technology, and more specifically, to a method for multi-dimensional and accurate assessment of job competency based on machine learning. Background Technology

[0002] In the current field of human resource management, especially in the job competency assessment stage, machine learning-based data processing methods are commonly used to assist decision-making. Existing technologies typically collect multi-dimensional characteristic data of candidates, such as professional knowledge, communication and collaboration skills, and logical thinking, and then use machine learning models to independently score each dimension or calculate a comprehensive score through a predefined linear weighting method to achieve a quantitative assessment of job suitability. This approach simplifies the complex competency structure into a computable set of features, improving the efficiency of the assessment.

[0003] However, existing technologies have shortcomings: when dealing with multi-dimensional competency characteristics, they fail to effectively consider the non-linear interactions between different dimensions. Specifically, each competency dimension does not play an independent role in actual work performance. High performance in some dimensions may, to some extent, compensate for or mask deficiencies in other dimensions. This dynamic compensation effect varies in complexity depending on the specific job context and individual characteristics. Existing models based on independent scoring or linear weighting cannot capture and characterize this dynamic, non-linear relationship between dimensions, resulting in a systematic deviation between the final comprehensive evaluation results and the actual competency level in real work scenarios, reducing the accuracy and predictive value of the assessment. Summary of the Invention

[0004] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a multi-dimensional and accurate assessment method for job competency based on machine learning to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A machine learning-based method for multi-dimensional and accurate assessment of job competency includes the following steps: S1: Obtain multi-dimensional competency feature data of candidates; S2: Based on multi-dimensional competency feature data, a machine learning model is used to identify the interaction relationships between the dimensions and determine whether a predefined interaction relationship exists. S3: When a predefined interaction relationship exists, dynamic compensation patterns are identified by analyzing the compensation symmetry and compensation transmission between dimensions. S4: Based on the identified dynamic compensation pattern, the compensation phase coordination index is calculated using time-frequency analysis to quantify the coordination compensation characteristics of different dimensions in the time dimension. S5: Based on the dynamic compensation mode and the compensation phase synergy index, a nonlinear weighting strategy is adopted to dynamically adjust the weight distribution of each dimension in the calculation of the comprehensive competence score. S6: Output the overall competence score after dynamic weight adjustment.

[0006] Furthermore, obtain multi-dimensional competency characteristic data of candidates, including: Extract relevant feature data from the candidate's resume text; The standardized assessment questionnaires completed by the candidates were analyzed to obtain quantitative scoring data; Extract behavioral characteristic data from the candidate's interview process records; The feature data, quantitative scoring data, and behavioral feature data related to job requirements are standardized and integrated into a structured, multi-dimensional competency feature dataset.

[0007] Furthermore, based on multi-dimensional competency feature data, machine learning models are used to identify the interaction relationships between each dimension and determine whether predefined interaction relationships exist, including: Standardized feature data is obtained by standardizing multi-dimensional competency feature data; Construct a dimensional interaction relationship graph based on normalized feature data; A graph traversal algorithm is used to identify potential interaction paths in the dimensional interaction graph. Calculate the association strength index of potential interaction paths; The association strength index is compared with a preset association strength threshold to determine whether a predefined interaction relationship exists.

[0008] Furthermore, when predefined interaction relationships exist, dynamic compensation patterns are identified by analyzing the compensation symmetry and compensation transmission between dimensions, including: Calculate the compensation direction index between dimensions to determine compensation symmetry; Construct compensation transmission paths between dimensions to analyze compensation transmission. Calculate the indirect compensation intensity index based on the compensation transmission path; A dynamic compensation model description is generated based on the compensation direction index and the indirect compensation intensity index.

[0009] Furthermore, the compensation direction index was obtained by analyzing the asymmetric compensation relationship between dimensions in historical data using statistical hypothesis testing methods.

[0010] Furthermore, the compensation transmission path is identified through a path traversal algorithm in a directed graph structure.

[0011] Furthermore, based on the identified dynamic compensation pattern, a compensation phase coordination index is calculated using time-frequency analysis to quantify the coordination compensation characteristics of different dimensions in the time dimension, including: Extract behavioral time-series data of each competency dimension within a preset time window from the dynamic compensation model; Time-frequency transformation is performed on behavioral time-series data to obtain the frequency components of each dimension at different time scales; Calculate the phase synchronization index between frequency components in each dimension; The amplitude of coordinated fluctuations between dimensions is determined based on the phase synchronization index. The compensation phase coordination index is calculated based on the ratio between the coordinated fluctuation amplitude and the preset benchmark value.

[0012] Furthermore, calculating the phase synchronization index between frequency components of each dimension includes: extracting the instantaneous phase sequence of each frequency component; calculating the phase difference sequence between the instantaneous phase sequences; and performing statistical analysis on the phase difference sequence to obtain the phase synchronization index.

[0013] Furthermore, based on the dynamic compensation model and the compensation phase synergy index, a nonlinear weighting strategy is adopted to dynamically adjust the weight allocation of each dimension in the calculation of the comprehensive competence score, including: The type of compensation relationship between each dimension is determined based on the dynamic compensation model; The collaborative compensation contribution of each dimension is calculated based on the compensation phase collaborative index. A weighting adjustment factor is constructed by combining the type of compensation relationship and the contribution of collaborative compensation; The final weight coefficients are obtained by inputting the weight adjustment factor into the nonlinear transformation function. The original dimension weights are dynamically updated based on the final weight coefficients.

[0014] Furthermore, the output includes a comprehensive competency score adjusted for dynamic weights, comprising: Convert the dynamically weighted overall competency score into a standardized assessment data format; Standardized evaluation data is transmitted to the target user interface via a data communication interface; Render the standardized evaluation data as visual graphical elements in the target user interface; Generate interactive competency assessment reports based on visual graphic elements.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By constructing a dynamic compensation identification and quantification mechanism, the technical challenge of capturing nonlinear interactions in multidimensional competency assessment is effectively solved. Compared with the traditional static assessment model, the machine learning model is used to automatically identify the interaction relationship between dimensions, and then the compensation symmetry and transmission are analyzed to construct a dynamic compensation mode. This hierarchical processing method can accurately characterize the complex relationship of the waxing and waning between different dimensions. In particular, the compensation phase synergy index calculated by the time-frequency analysis method can quantify the synergistic fluctuation characteristics of dimensions in the time dimension, thereby breaking through the limitation of traditional methods that can only process static data, and making the assessment results closer to the dynamic changes in competency performance in actual work.

[0016] 2. By using a nonlinear weighting strategy to achieve dynamic optimization of weight allocation, the final comprehensive score can reflect both the basic level of each dimension and the synergistic compensation effect between dimensions. It can adaptively balance the contribution of each dimension according to specific job requirements and individual characteristics, thereby significantly improving the discrimination and predictive accuracy of the evaluation results. Compared with the traditional linear weighting method, it effectively avoids evaluation bias caused by ignoring the interaction between dimensions, and provides a more accurate scientific basis for talent selection decisions. Attached Figure Description

[0017] Figure 1 This is a flowchart of the machine learning-based multi-dimensional accurate assessment method for job competency according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Example: Figure 1 This invention presents a machine learning-based method for accurate multi-dimensional assessment of job competency, which includes the following steps: S1: Obtain multi-dimensional competency feature data of candidates; S2: Based on multi-dimensional competency feature data, a machine learning model is used to identify the interaction relationships between the dimensions and determine whether a predefined interaction relationship exists. S3: When a predefined interaction relationship exists, dynamic compensation patterns are identified by analyzing the compensation symmetry and compensation transmission between dimensions. S4: Based on the identified dynamic compensation pattern, the compensation phase coordination index is calculated using time-frequency analysis to quantify the coordination compensation characteristics of different dimensions in the time dimension. S5: Based on the dynamic compensation mode and the compensation phase synergy index, a nonlinear weighting strategy is adopted to dynamically adjust the weight distribution of each dimension in the calculation of the comprehensive competence score. S6: Output the overall competence score after dynamic weight adjustment.

[0020] S1: Obtain multi-dimensional competency feature data of candidates, specifically implemented as follows: When acquiring multi-dimensional competency feature data of candidates, the first step is to extract feature data related to the job requirements from the candidate's resume text. Specifically, the resume text is parsed using natural language processing (NLP) technology. A keyword library is built based on the job description, including job-related skill terms and concepts. For example, for a software engineer position, the keyword library includes programming language names such as Java and Python, development framework names such as Spring, and project management terms such as Agile development. The extraction process calculates the frequency of keywords in the resume text. The frequency value is obtained by counting the number of times each keyword appears in the resume and dividing by the total number of words in the resume text to obtain a normalized frequency value. Simultaneously, a thesaurus is used to handle semantically similar words to avoid duplicate counting; for example, "Java" and "Java programming" are considered the same keyword. Furthermore, contextual semantic analysis is used to check the contextual relevance of keywords; for example, keywords are only counted when they appear in sentences describing skills or experience, ensuring that the extracted feature data accurately reflects the candidate's skill matching. The keyword library is built based on the job description document and defined through manual review or automated extraction of common terms. The total number of words in the frequency normalization process is obtained from the character statistics of the resume text, and the normalized frequency value ranges from 0 to 1.

[0021] Next, the standardized assessment questionnaires completed by the candidates are parsed to obtain quantitative scoring data. The standardized assessment questionnaire includes multiple-choice questions and rating questions, with each question corresponding to a specific competency dimension, such as communication skills or logical thinking. The parsing process maps the candidates' options or answers to numerical scores using predefined scoring rules. For example, for a 5-point Likert scale, options from strongly disagree to strongly agree are assigned scores of 1, 2, 3, 4, and 5, respectively. For open-ended questions, a sentiment analysis algorithm is used to extract sentiment polarity scores. Sentiment polarity scores are obtained by analyzing sentiment words and context in the text; for example, positive words are assigned positive scores and negative words are assigned negative scores. The overall sentiment score is then calculated using a weighted average. The parsed scoring data undergoes consistency checks, such as calculating the Cronbach's alpha coefficient as an internal consistency indicator to ensure questionnaire reliability. Finally, quantitative scoring data for each dimension is generated and stored in tabular form. Rows represent candidate identifiers, and columns represent scores for different dimensions. The scoring rules and sentiment analysis algorithm parameters are adjusted using a pre-trained dataset, such as calibrating scoring thresholds using historical questionnaire data.

[0022] Then, behavioral feature data is extracted from the candidates' interview process records. These records include text-transcribed dialogues or audio / video recordings. Key behavioral indicators are extracted using behavioral event interviewing. For example, text analysis techniques are used to identify job competency-related behavioral descriptions in the interview records, such as teamwork or problem-solving. Specific behavioral instances are extracted using named entity recognition algorithms based on a predefined behavioral entity library, which includes common behavioral verbs and phrases. Simultaneously, time series analysis is used to calculate behavioral frequency and duration. Behavioral frequency is obtained by counting the number of times a specific behavior occurs per unit of time, such as counting the number of times teamwork occurs within a 10-minute interview segment. For audio / video recordings, speech sentiment analysis tools are used to extract intonation and facial expression features as supplementary behavioral indicators. Intonation features are obtained by analyzing the fundamental frequency and intensity of the speech signal, while facial expression features are identified through image processing techniques to recognize changes in facial expressions. All this extracted data is quantified into numerical form and stored as a behavioral feature dataset, where the behavioral entity library and feature extraction parameters are defined based on the job competency model.

[0023] Finally, the feature data, quantitative scoring data, and behavioral feature data related to job requirements were standardized and integrated into a structured multi-dimensional competency feature dataset. The standardization process employed a min-maximum normalization method, scaling the data for each feature dimension to a range of 0 to 1. This was achieved by subtracting the minimum value of each feature value from the minimum value of that dimension and then dividing by the difference between the maximum and minimum values. For example, for feature frequencies extracted from resumes, the minimum value was set to the actual minimum value of that dimension among all candidate data, and the maximum value was set to the actual maximum value of that dimension among all candidate data. For questionnaire scoring data, the minimum and maximum values ​​were set based on the scoring range, such as 1 point and 5 points. The integration process merged the three types of standardized data according to candidate identifiers, forming a structured table where each row represents all feature values ​​for a candidate, and each column represents a competency dimension. Missing values ​​were handled using mean imputation, i.e., filled with the average value of all candidates for that dimension, excluding missing values ​​during average calculation. The final multi-dimensional competency feature dataset was used for subsequent machine learning model processing, with the standardization and integration steps ensuring data consistency and comparability.

[0024] S2: Based on multi-dimensional competency feature data, a machine learning model is used to identify the interaction relationships between each dimension and determine whether a predefined interaction relationship exists. The specific implementation is as follows: When using machine learning models to identify the interaction relationships between dimensions based on multi-dimensional competency feature data, the first step is to standardize the multi-dimensional competency feature data to obtain normalized feature data. Standardization employs the Z-score method, specifically by calculating the mean and standard deviation for each feature dimension. The mean is obtained by summing the feature values ​​of all candidates in that dimension and dividing by the total number of candidates. The standard deviation is obtained by taking the square root of the average of the sum of the squares of the differences between each feature value and the mean. For example, for the communication skills dimension, assuming there are 100 candidates and their total feature values ​​are 500, the mean is 5, and the standard deviation is obtained by taking the square root of the average of the sum of the squares of the differences between each feature value and 5. During standardization, each feature value is subtracted from the mean of that dimension and then divided by the standard deviation of that dimension, resulting in normalized feature data with a mean of 0 and a standard deviation of 1. If the standard deviation of a certain dimension is 0, it indicates that all feature values ​​are the same, and the standardization process for that dimension is skipped, using the original value directly. The normalized feature data is stored in matrix form, with rows representing candidates and columns representing dimensions.

[0025] Next, a dimensional interaction graph is constructed based on the normalized feature data. The dimensional interaction graph is a graph structure where nodes represent different competency dimensions, and edges represent the interactions between dimensions. Edge construction is based on the correlation calculation between dimensions, using the Pearson correlation coefficient as the correlation indicator. The Pearson correlation coefficient is obtained by calculating the product of the covariance of two dimension feature values ​​divided by their respective standard deviations. The covariance is obtained by summing the difference between the products of the two dimension feature values ​​and their respective means, then dividing by the total number of candidates minus 1. For example, for the communication skills dimension and the teamwork dimension, their Pearson correlation coefficients are calculated. If the absolute value exceeds 0.3, an edge is added, and the edge weight is set to the correlation coefficient value. The graph structure is represented using an adjacency matrix, where the matrix elements store the edge weights, and positions without edges are set to 0. During construction, only edges with an absolute correlation coefficient greater than 0.1 are retained to filter noise. This threshold is determined based on the distribution of correlations between dimensions in historical data; for example, by analyzing data from the past 1000 candidates, the first quartile of the absolute correlation coefficient is taken as the threshold.

[0026] Then, a graph traversal algorithm is used to identify potential interaction paths in the dimensional interaction graph. The graph traversal algorithm uses a depth-first search method, recursively visiting neighboring nodes from each node and recording all simple paths; the path length is limited to no more than 3 nodes to avoid combinatorial explosion; the depth-first search is implemented by maintaining a visit stack. Starting from the starting node, it is marked as visited and pushed onto the stack, and then its unvisited neighboring nodes are traversed, repeating the process until the stack is empty; for example, starting from the communication ability node, traversing to the teamwork node, and then traversing to the problem-solving node, a path is formed; potential interaction paths are defined as all paths with a length between 2 and 3, which represent possible indirect interactions between dimensions; during the traversal, visited nodes are ignored to prevent loops, and the output is a list of paths, with each path containing a sequence of nodes.

[0027] Calculate the association strength index of potential interaction paths. The association strength index is calculated using the geometric mean of the weights of all edges in the path. The geometric mean is obtained by multiplying the weights of all edges in the path and taking the root of the number of edges in the path. If the path has only one edge, the association strength is directly equal to the weight of that edge. During the calculation, the absolute value of the edge weights is taken to ensure a positive number. The geometric mean calculation uses a logarithmic transformation to avoid numerical underflow, that is, first take the natural logarithm of the weights, sum them, divide by the number of edges, and then take the exponent to restore the value. The association strength index ranges from 0 to 1, with a larger value indicating a stronger path association.

[0028] Finally, the association strength index is compared with a preset association strength threshold to determine whether a predefined interaction relationship exists. The preset association strength threshold is set based on the distribution of path association strength of successful candidates in historical data. For example, the threshold is obtained by calculating the third and fourth quartiles of the association strength of all paths in historical data, and is assumed to be 0.5. The comparison process traverses the association strength index of all potential interaction paths. If the index of any path is greater than or equal to the preset association strength threshold, it is determined that a predefined interaction relationship exists; otherwise, it is determined that no interaction relationship exists. For example, if the association strength index of a path is 0.6 and the preset association strength threshold is 0.5, it is determined that an interaction relationship exists. The threshold setting can be adjusted according to the specific position. For example, for management positions, the threshold may be increased to 0.6, determined based on the median association strength of historical successful cases for that position. The output result is a Boolean value, used for subsequent decision-making.

[0029] S3: When predefined interaction relationships exist, dynamic compensation patterns are identified by analyzing the compensation symmetry and compensation transmission between dimensions. Specifically, this is implemented as follows: To identify dynamic compensation patterns by analyzing the symmetry and transmissibility of compensation between dimensions, the first step is to calculate the direction index of compensation between dimensions to determine the symmetry. The direction index is obtained by analyzing the asymmetric compensation relationship between dimensions in historical data using statistical hypothesis testing. Historical data includes past candidates' scores on each competency dimension and their performance ratings. Competency dimension scores are derived from multi-dimensional competency feature data, and performance ratings are derived from performance evaluation records. For each pair of dimensions, such as communication skills and problem-solving skills, the score difference is calculated across different candidate samples. The score difference is defined as the score of dimension A minus the score of dimension B; a positive difference indicates that the score of dimension A is higher than that of dimension B, and a negative difference indicates that the score of dimension B is higher than that of dimension A. The statistical hypothesis test uses a sign test method, with the null hypothesis being that there is no significant asymmetric compensation relationship between the two dimensions. The p-value is calculated using a binomial distribution by counting the number of positive and negative differences. For example, in 100 historical samples, if the communication skills dimension influences the problem-solving skills dimension... If the number of samples showing positive differences in a dimension is 70 and the number of samples showing negative differences is 30, then the p-value is calculated. The p-value is calculated using the binomial distribution formula, which is the probability that the number of positive differences is greater than or equal to 70 when the probability is 0.5. If the p-value is less than 0.05, the null hypothesis is rejected, indicating the existence of a significant asymmetric compensation relationship. The compensation direction index is defined as the number of positive differences divided by the number of negative differences. An index greater than 1 indicates that dimension A has a dominant compensation direction for dimension B, and less than 1 indicates reverse compensation. During historical data preprocessing, outliers need to be removed, for example, samples with scores outside three standard deviations are removed. The outlier threshold is calculated based on the mean and standard deviation of the scores of each dimension in the historical data. The p-value threshold for statistical hypothesis testing is set to 0.05, which is determined based on common statistical significance levels. In the calculation of the compensation direction index, if the number of negative differences is 0, the compensation direction index is set to the maximum value, such as 1000, to handle the division by zero case.

[0030] Next, we construct the compensation transmission paths between dimensions to analyze compensation transmission. The compensation transmission paths are identified using a path traversal algorithm in a directed graph structure. The directed graph is constructed based on the conditional dependencies between dimensions in historical data, where nodes represent competency dimensions and directed edges represent the direction of compensation transmission. The direction of the edges is determined by calculating conditional probabilities, defined as the probability that a high score in dimension A also results in a high score in dimension B. The high-score threshold is based on the median scores of each dimension in historical data; for example, a score greater than the median is considered a high score. The conditional probability is obtained by dividing the number of samples in historical data where both dimension A and dimension B have high scores by the number of samples where dimension A has a high score. When constructing the directed graph, only edges with conditional probabilities greater than 0.3 are retained; this threshold is based on the median of all conditional probabilities in historical data. The path traversal algorithm employs a depth-first search method, recursively visiting adjacent nodes from each node and recording all simple paths. The path length is limited to 2 to 4 nodes to avoid computational complexity explosion. Depth-first search is implemented by initializing a visit stack. Starting from the starting node, it is marked as visited and pushed onto the stack. Then, its unvisited neighbor nodes are traversed, and the process is repeated until the stack is empty. For example, starting from the communication skills node, traversing to the teamwork node, and then to the problem-solving node, a path is formed. During the traversal, visited nodes are ignored to prevent loops, and the output is a list of paths. Compensation transmission paths are defined as all paths with a length between 2 and 4, representing possible indirect compensation transmission between dimensions.

[0031] Then, the indirect compensation strength index is calculated based on the compensation transmission path. The indirect compensation strength index is calculated using the geometric mean of the conditional probabilities of all edges in the path. The geometric mean is obtained by multiplying the conditional probabilities of each edge in the path and taking the root of the number of edges. During the calculation, the conditional probabilities range from 0 to 1. The geometric mean uses logarithmic transformation to avoid numerical underflow, i.e., first taking the natural logarithm of the conditional probabilities, summing them, dividing by the number of edges, and then restoring the exponent. The indirect compensation strength index ranges from 0 to 1, with a larger value indicating a stronger compensation effect of the transmission path. For a single-edge path, the indirect compensation strength is directly equal to the conditional probability of that edge. The indirect compensation strength indices of all paths are stored in a list for subsequent analysis. In the calculation of the geometric mean, if any conditional probability is 0, the indirect compensation strength index is set to 0 to handle the zero value case.

[0032] Finally, a dynamic compensation model description is generated based on the compensation direction index and the indirect compensation intensity index. The dynamic compensation pattern description is constructed by combining compensation direction indicators and indirect compensation intensity indicators. Specifically, for each pair of dimensions with a significant compensation direction, its compensation direction indicator value and the corresponding set of transmission paths are recorded. For example, the compensation direction indicator for the communication ability dimension to the problem-solving ability dimension is 2.33, and there are two transmission paths with indirect compensation intensity indicators of 0.69 and 0.75, respectively. The dynamic compensation pattern description adopts a structured data format, including the dominant compensation dimension, the compensated dimension, the compensation direction indicator value, the list of transmission paths, and the indirect compensation intensity indicator for each path. During the generation process, only entries with a compensation direction indicator greater than 1.2 and an indirect compensation intensity indicator greater than 0.5 are retained. These thresholds are determined based on the distribution of successful cases in historical data. For example, by analyzing the compensation patterns of 100 high-performing employees, the lower quartiles of all compensation direction indicators and indirect compensation intensity indicators are calculated. The lower quartile of the compensation direction indicator is 1.2, and the lower quartile of the indirect compensation intensity indicator is 0.5. The final output dynamic compensation pattern description is used for weight adjustment decisions in subsequent steps. The dynamic compensation pattern description is stored in JSON format to ensure data parsing and scalability.

[0033] S4: Based on the identified dynamic compensation pattern, the compensation phase coordination index is calculated using time-frequency analysis to quantify the coordination compensation characteristics of different dimensions in the time dimension. The specific implementation is as follows: When calculating the compensation phase synergy index based on the identified dynamic compensation pattern using time-frequency analysis, the first step is to extract behavioral time-series data for each competency dimension within a preset time window from the dynamic compensation pattern. The dynamic compensation pattern originates from the dynamic compensation pattern description generated in the preceding steps, which includes historical behavioral records for each competency dimension. The preset time window is defined by setting a start time point and an end time point; for example, the start time point is 3 months prior to the current time point, and the end time point is the current time point. For behavioral time-series data extraction, for each competency dimension, such as communication skills or problem-solving skills, the score value of that dimension is obtained from the dynamic compensation pattern description, arranged chronologically within the preset time window. The sampling interval for the time-series data is set to a fixed period, for example, one data point per week, with a total of 12 data points corresponding to a 3-month time window. During the extraction process, if data for certain time points is missing in the dynamic compensation pattern description, linear interpolation is used to supplement them. Linear interpolation calculates the missing values ​​by averaging adjacent data points. The behavioral time-series data is stored in the form of a time-series array, with each array element containing a timestamp and a dimension score value.

[0034] Next, a time-frequency transformation is performed on the behavioral time-series data to obtain the frequency components of each dimension at different time scales. The time-frequency transformation uses the short-time Fourier transform method to convert the time-domain data into a frequency-domain representation. The short-time Fourier transform is achieved by dividing the behavioral time-series data into overlapping time windows, each with a length of 8 data points and an overlap ratio of 50%. For each time window, a Fourier transform is applied to calculate the frequency spectrum, with the frequency range set from 0.1 Hz to 1 Hz, corresponding to fluctuation periods from 1 week to 10 weeks. The frequency components are extracted as amplitude values ​​from the frequency spectrum and grouped according to the frequency scale, such as 0.1 Hz, 0.2 Hz up to 1 Hz. Each frequency component represents the fluctuation intensity of the behavioral time-series data at that frequency. In the time-frequency transformation parameters, the time window length and overlap ratio are set based on the typical fluctuation period of the behavioral data. For example, historical data analysis shows that the fluctuation period of the competency dimension is mostly concentrated between 2 and 8 weeks, so the frequency range covers this range. The transformed frequency components are stored in complex form, including amplitude and phase information, for subsequent analysis.

[0035] Then, the phase synchronization index between frequency components of each dimension is calculated. The phase synchronization index is obtained through statistical analysis after extracting the instantaneous phase sequences of each frequency component and calculating the phase difference sequence. The instantaneous phase sequence is extracted from the complex representation of the frequency component using a Hilbert transform. The Hilbert transform calculates the instantaneous phase by constructing an analytic signal, which is composed of the original signal and its Hilbert transformed signal. For example, for the communication capability dimension at a frequency component of 0.2 Hz, its instantaneous phase sequence is extracted, with the sequence length consistent with the number of time windows. The phase difference sequence is obtained by calculating the difference between the instantaneous phase sequences of two dimensions on the same frequency component, with the difference modulo... The 2π operation ensures that the result is within the range of 0 to 2π. For example, the instantaneous phase sequences of the communication ability dimension and the teamwork dimension at the 0.2 Hz frequency component are obtained by subtracting the phases at the corresponding time points to obtain the phase difference sequence. Statistical analysis uses the standard deviation of the phase difference sequence as the phase synchronization index. The standard deviation is obtained by taking the square root of the average of the sum of the squares of the differences between each value in the phase difference sequence and the average value. The smaller the phase synchronization index value, the higher the phase synchronization. The index range is between 0 and π. In the calculation, if the phase difference sequence length is 0, the phase synchronization index is set to the maximum value π to handle boundary cases.

[0036] The amplitude of coordinated fluctuation between dimensions is determined based on the phase synchronization index. The amplitude of coordinated fluctuation is calculated by dividing the phase synchronization index by 1. For example, if the phase synchronization index between the communication ability dimension and the teamwork dimension is 0.5 on the 0.2 Hz frequency component, then the amplitude of coordinated fluctuation is 1 divided by 0.5, which equals 2. A larger amplitude of coordinated fluctuation indicates stronger coordinated compensation characteristics between dimensions in the time dimension. During the calculation, if the phase synchronization index is 0, the amplitude of coordinated fluctuation is set to its maximum value, such as 1000, to avoid division by zero errors. The amplitude of coordinated fluctuation is calculated separately for each pair of dimensions and each frequency component, and finally summarized as the average of all frequency components as the overall amplitude of coordinated fluctuation. When summarizing, the weight of each frequency component is set based on its amplitude value; the larger the amplitude, the higher the weight. The weight is calculated by dividing the amplitude of each frequency component by the sum of the amplitudes of all frequency components.

[0037] Finally, the compensation phase synergy index is calculated based on the ratio between the synergistic fluctuation amplitude and the preset benchmark value. The preset benchmark value is determined by the distribution of synergistic fluctuation amplitudes of high-performing candidates in historical data, for example, by calculating the median of all synergistic fluctuation amplitudes in historical data as the preset benchmark value; the ratio is defined as the synergistic fluctuation amplitude divided by the preset benchmark value; for example, if the current synergistic fluctuation amplitude is 2 and the preset benchmark value is 1.5, then the compensation phase synergy index is 2 divided by 1.5, approximately equal to 1.33; a compensation phase synergy index value greater than 1 indicates that the synergistic compensation characteristic is higher than the historical benchmark, and less than 1 indicates that it is lower than the historical benchmark; in the calculation, if the preset benchmark value is 0, the compensation phase synergy index is set to 1 to handle abnormal situations; the preset benchmark value is updated every 6 months, and the median is recalculated based on newly accumulated historical data; the compensation phase synergy index is output in numerical form and used for weight adjustment decisions in subsequent steps.

[0038] S5: Based on the dynamic compensation model and the compensation phase synergy index, a nonlinear weighting strategy is used to dynamically adjust the weight allocation of each dimension in the calculation of the comprehensive competency score. The specific implementation is as follows: When implementing a nonlinear weighting strategy based on dynamic compensation mode and compensation phase synergy index to dynamically adjust the weight allocation of each dimension in the comprehensive competency score calculation, the type of compensation relationship between each dimension is first determined according to the dynamic compensation mode. The dynamic compensation pattern is derived from the dynamic compensation pattern description generated in the preceding steps, which includes compensation direction indicators and compensation transmission path information between each competency dimension. The compensation relationship type is classified by analyzing the characteristics of the compensation direction indicators and transmission paths. For example, the compensation direction indicator threshold is set to 1.2, which is determined based on the upper quartile of all compensation direction indicators in historical data. For each pair of dimensions, if the compensation direction indicator is greater than 1.2, the compensation relationship type is determined to be unidirectional compensation, and the dominant dimension and the compensated dimension are identified. If the compensation direction indicator is between 0.8 and 1.2, the compensation relationship type is determined to be bidirectional compensation. If the compensation direction indicator is less than 0.8, the compensation relationship type is determined to be reverse compensation. At the same time, based on the number of compensation transmission paths and the indirect compensation intensity indicator, the network compensation type is further identified. For example, when a dimension participates in more than 3 transmission paths and the average indirect compensation intensity indicator is greater than 0.6, the dimension is marked as a network compensation node. The compensation relationship type is output as a structured list, including dimension pairs, relationship type identifiers, and dominant dimension information.

[0039] Next, the collaborative compensation contribution of each dimension is calculated based on the compensation phase synergy index. The compensation phase synergy index comes from the compensation phase synergy index value of each dimension calculated in the previous steps; the collaborative compensation contribution represents the relative importance of each dimension in the overall collaborative compensation, and is calculated through normalization. Specifically, firstly, the compensation phase synergy indices of all competency dimensions are collected, and their sum is calculated; then, for each dimension, the collaborative compensation contribution is defined as the compensation phase synergy index of that dimension divided by the sum of the compensation phase synergy indices of all dimensions; during the calculation, if the sum of the compensation phase synergy indices is 0, the collaborative compensation contribution of all dimensions is set to an equal value, for example, 1 for each dimension divided by the total number of dimensions; the collaborative compensation contribution is stored in vector form for subsequent steps.

[0040] Then, a weight adjustment factor is constructed by combining the compensation relationship type and the collaborative compensation contribution. The weight adjustment factor is calculated by mapping the compensation relationship type to the basic adjustment value and combining it with the collaborative compensation contribution. The basic adjustment value is set based on the compensation relationship type. For example, for one-way compensation, the basic adjustment value of the dominant dimension is set to 1.5, and the basic adjustment value of the compensated dimension is set to 0.5; for two-way compensation, the basic adjustment values ​​of both dimensions are set to 1.0; for reverse compensation, the basic adjustment value of the dominant dimension is set to 0.5, and the basic adjustment value of the compensated dimension is set to 1.5; for network compensation, the basic adjustment value is set to 1.2. The weight adjustment factor is calculated by multiplying the basic adjustment value by the collaborative compensation contribution. The setting of the basic adjustment value is based on the statistical analysis of the impact of different compensation relationship types on performance in historical data, such as determining the adjustment range through regression analysis. The weight adjustment factor output is a value for each dimension, ranging from 0 to 1.

[0041] The weight adjustment factor is input into a nonlinear transformation function to obtain the final weight coefficients. The nonlinear transformation function is an sigmoid function; its purpose is to smoothly map the weight adjustment factor to the interval between 0 and 1, avoiding the influence of extreme values. In the calculation, if the weight adjustment factor is 0, the final weight coefficient is set to 0.5 to handle boundary cases. The final weight coefficients are stored as a vector for weight updates.

[0042] Finally, the original dimension weights are dynamically updated based on the final weight coefficients. The original dimension weights are derived from initial settings or historical averages; for example, the original weight for communication skills is 0.3, for teamwork is 0.25, and for problem-solving is 0.2. Dynamic updates use exponential smoothing. The updated weights are equal to the smoothing parameter multiplied by the original weights plus 1, minus the smoothing parameter multiplied by the final weight coefficient. The smoothing parameter is set to 0.7, a value determined based on the need to balance stability and sensitivity in time series analysis. After the update, the weights of all dimensions are normalized to ensure the sum of the weights is 1. Normalization is achieved by dividing each dimension weight by the sum of all dimension weights. For example, if the sum of the dimension weights after the update is 1.2, then each dimension weight is divided by 1.2 to obtain the normalized weight. The dynamically updated weights are used to calculate the comprehensive competency score, completing the weight allocation adjustment.

[0043] S6: Output the overall competence score after dynamic weight adjustment. The specific implementation is as follows: When outputting the overall competence score after dynamic weight adjustment, the overall competence score after dynamic weight adjustment is first converted into a standardized evaluation data format. The dynamically weighted overall competence score is derived from the overall score and its dimensions generated in the preceding steps. The standardized assessment data format is defined using JSON and includes a fixed field structure. For example, fields include candidate identifier, assessment timestamp, overall competence score value, a list of scores for each dimension, and weight version information. During the conversion process, the dynamically weighted overall competence score is mapped to these fields. For instance, the candidate identifier uses a string type, the assessment timestamp uses ISO 8601 format, the overall competence score value uses a floating-point number type, and the list of scores for each dimension uses an array structure, with each array element containing a dimension name string and a floating-point score value. The weight version information records the version identifier of the weight adjustment, used to trace the weight calculation process. During conversion, if there are missing values ​​in the dynamically weighted overall competence score, default values ​​are used to fill them, such as filling with 0 when a score is missing. The field definitions of the standardized assessment data format are based on common industry assessment data exchange standards and were determined after analyzing the data formats of multiple assessment systems. The converted data is stored as a JSON file or string to ensure machine readability and cross-platform compatibility.

[0044] Next, standardized evaluation data is transmitted to the target user interface via a data communication interface. The data communication interface is implemented using a RESTful API interface within the HTTP protocol. The transmission process is completed using a client-server architecture, where the client is a web browser or mobile application hosting the target user interface, and the server is a backend system storing the standardized evaluation data. The RESTful API interface uses the POST method, with the request URL set to a fixed resource endpoint. The request body contains the standardized evaluation data, encoded in JSON format. During transmission, a timeout of 30 seconds is set, determined based on network latency testing. If transmission fails, a retry mechanism is implemented, with a maximum of 3 retries. The retry interval uses an exponential backoff strategy, for example, 1 second for the first retry, 2 seconds for the second, and 4 seconds for the third. The target user interface is defined as a web application interface or mobile application interface, and its URL or application identifier is pre-configured in the system. After transmission is complete, the server returns a status code to confirm receipt; for example, status code 200 indicates success, status code 400 indicates a client error, and status code 500 indicates a server error. Error handling includes logging and user notifications, such as sending an alert email to the administrator in case of transmission failure.

[0045] Then, the standardized evaluation data is rendered as a visual graphical element in the target user interface. The rendering process uses JavaScript libraries such as D3.js or Chart.js; visual graphic elements include bar charts, radar charts, and trend charts; for example, bar charts display scores for each dimension, radar charts display relative performance across multiple dimensions, and trend charts display historical score changes; during rendering, standardized evaluation data is first retrieved from the target user interface's data storage and parsed in JSON format; then, chart instances are initialized according to the graphic type, for example, the X-axis of a bar chart is set to the dimension name, the Y-axis to the score value, and the number of axes in a radar chart equals the number of dimensions; visual attributes of graphic elements are set through configuration parameters, for example, bar charts use blue tones, and radar charts use semi-transparent orange fill color; during rendering, data binding is accomplished by mapping score values ​​from the standardized evaluation data to graphic coordinates, for example, the height of each bar in a bar chart is calculated by multiplying the score value by a scaling factor, which is dynamically adjusted based on the Y-axis range; graphic elements support responsive layout, automatically adjusting size when the user interface size changes; after rendering, graphic elements are embedded in a specified container of the target user interface, such as an HTML div element; the type and style parameters of visual graphic elements are managed through configuration files, allowing adjustments based on user preferences.

[0046] Finally, an interactive competency assessment report is generated based on visual graphic elements. The interactive competency assessment report is implemented by combining multiple visual graphic elements and adding interactive functions. The report structure includes a summary section, a detailed analysis section, and a recommendation section. The summary section displays the overall competency score and key dimension scores, the detailed analysis section embeds bar charts, radar charts, and trend charts, and the recommendation section generates text descriptions based on score analysis. Interactive functions include click highlighting, data filtering, and report export. For example, clicking on a dimension point in the radar chart highlights the corresponding data for that dimension in all graphs and displays detailed score information. Data filtering allows users to select a time range or specific candidates for comparison, with filtering conditions entered via drop-down menus or sliders. The report export function supports PDF and Excel formats and is implemented using libraries such as jsPDF or SheetJS. During the generation process, the report content is dynamically generated; for example, the summary section uses a template engine to populate HTML paragraphs by parsing the overall score and dimension scores from the standardized assessment data. Interactive functions are implemented through event listeners; for example, click event listeners are added to graphic elements to update the report content when triggered. The interactive competency assessment report adopts a responsive design to ensure proper display on different devices. After the report is generated, it is stored on a server or locally and accessed through the target user interface.

[0047] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0048] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0049] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0050] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0051] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0052] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0053] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-dimensional and accurate assessment method for job competency based on machine learning, characterized in that, Includes the following steps: S1: Obtain multi-dimensional competency feature data of candidates; S2: Based on multi-dimensional competency feature data, a machine learning model is used to identify the interaction relationships between the dimensions and determine whether a predefined interaction relationship exists. S3: When a predefined interaction relationship exists, dynamic compensation patterns are identified by analyzing the compensation symmetry and compensation transmission between dimensions. S4: Based on the identified dynamic compensation pattern, the compensation phase coordination index is calculated using time-frequency analysis to quantify the coordination compensation characteristics of different dimensions in the time dimension. S5: Based on the dynamic compensation mode and the compensation phase synergy index, a nonlinear weighting strategy is adopted to dynamically adjust the weight distribution of each dimension in the calculation of the comprehensive competence score. S6: Output the overall competence score after dynamic weight adjustment.

2. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 1, characterized in that, Obtain multi-dimensional competency characteristic data of candidates, including: Extract relevant feature data from the candidate's resume text; The standardized assessment questionnaires completed by the candidates were analyzed to obtain quantitative scoring data; Extract behavioral characteristic data from the candidate's interview process records; The feature data, quantitative scoring data, and behavioral feature data related to job requirements are standardized and integrated into a structured, multi-dimensional competency feature dataset.

3. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 1, characterized in that, Based on multi-dimensional competency feature data, a machine learning model is used to identify the interaction relationships between each dimension and determine whether predefined interaction relationships exist, including: Standardized feature data is obtained by standardizing multi-dimensional competency feature data; Construct a dimensional interaction relationship graph based on normalized feature data; A graph traversal algorithm is used to identify potential interaction paths in the dimensional interaction graph. Calculate the association strength index of potential interaction paths; The association strength index is compared with a preset association strength threshold to determine whether a predefined interaction relationship exists.

4. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 1, characterized in that, When predefined interaction relationships exist, dynamic compensation patterns are identified by analyzing the compensation symmetry and compensation transmission between dimensions, including: Calculate the compensation direction index between dimensions to determine compensation symmetry; Construct compensation transmission paths between dimensions to analyze compensation transmission. Calculate the indirect compensation intensity index based on the compensation transmission path; A dynamic compensation model description is generated based on the compensation direction index and the indirect compensation intensity index.

5. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 4, characterized in that, The compensation direction index is obtained by analyzing the asymmetric compensation relationship between dimensions in historical data using statistical hypothesis testing methods.

6. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 4, characterized in that, The compensation transmission path is identified using a path traversal algorithm in a directed graph structure.

7. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 1, characterized in that, Based on the identified dynamic compensation pattern, a compensation phase coordination index is calculated using time-frequency analysis to quantify the coordination compensation characteristics of different dimensions in the time dimension, including: Extract behavioral time-series data of each competency dimension within a preset time window from the dynamic compensation model; Time-frequency transformation is performed on behavioral time-series data to obtain the frequency components of each dimension at different time scales; Calculate the phase synchronization index between frequency components in each dimension; The amplitude of coordinated fluctuations between dimensions is determined based on the phase synchronization index. The compensation phase coordination index is calculated based on the ratio between the coordinated fluctuation amplitude and the preset benchmark value.

8. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 7, characterized in that, The calculation of phase synchronization index between frequency components of each dimension includes: extracting the instantaneous phase sequence of each frequency component; calculating the phase difference sequence between the instantaneous phase sequences; and performing statistical analysis on the phase difference sequence to obtain the phase synchronization index.

9. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 1, characterized in that, Based on a dynamic compensation model and a compensation phase synergy index, a nonlinear weighting strategy is used to dynamically adjust the weight allocation of each dimension in the calculation of the comprehensive competency score, including: The type of compensation relationship between each dimension is determined based on the dynamic compensation model; The collaborative compensation contribution of each dimension is calculated based on the compensation phase collaborative index. A weighting adjustment factor is constructed by combining the type of compensation relationship and the contribution of collaborative compensation; The final weight coefficients are obtained by inputting the weight adjustment factor into the nonlinear transformation function. The original dimension weights are dynamically updated based on the final weight coefficients.

10. The method for multi-dimensional and accurate assessment of job competency based on machine learning according to claim 1, characterized in that, Output the overall competence score after dynamic weight adjustment, including: Convert the dynamically weighted overall competency score into a standardized assessment data format; Standardized evaluation data is transmitted to the target user interface via a data communication interface; Render the standardized evaluation data as visual graphical elements in the target user interface; Generate interactive competency assessment reports based on visual graphic elements.

Images