Personnel portrait construction method and system based on multi-dimensional talent evaluation model

Abstract

The invention provides a personnel portrait construction method and system based on a multi-dimensional talent evaluation model, and belongs to the technical field of human resource management and data processing, and the method comprises the steps: collecting the original data of talents, constructing a tree-level comprehensive quality index system, and outputting the index score vectors of the talents; performing dimension reduction processing on the original data to extract key features, calculating objective weights of the key features, and outputting weight vectors; inputting the key features and the weight vectors into a random forest machine learning model to calculate a preliminary evaluation score; on the basis of the key features, an analytic hierarchy process is adopted to calculate expert correction scores; integrating the two scores to calculate a comprehensive evaluation score, and constructing a personal portrait based on the index score vector, the comprehensive evaluation score and the key features; and updating the personal portrait, monitoring the index score vector or the comprehensive evaluation score, and generating growth early warning information for the talents meeting the early warning conditions. According to the method, multi-dimensional data are integrated, the weight is objectively calculated, and the portrait is dynamically updated.

Description

Technical Field

[0001] This application belongs to the field of human resource management and data processing technology, specifically involving a method and system for constructing personnel profiles based on a multi-dimensional talent evaluation model. Background Technology

[0002] In corporate human resource management, talent evaluation and personnel profiling are core components of talent selection, promotion, training, and team configuration. Traditional talent evaluation methods often rely on the subjective judgment of interviewers or single performance data, resulting in narrow evaluation dimensions, strong subjectivity, and a lack of data support, making it difficult to comprehensively reflect the overall capabilities and potential of talent. Furthermore, existing personnel profiling is largely based on static basic information, failing to update talent's ability changes, behavioral dynamics, and performance fluctuations in real time. This leads to a disconnect between the profiling and the actual situation of the talent, hindering the provision of accurate support for corporate human resource decisions.

[0003] With the development of big data and machine learning technologies, some methods have begun to try to introduce multi-dimensional data for talent evaluation, but the following shortcomings still exist: First, the data integration is low, and structured data (such as performance scores) and unstructured data (such as work behavior logs) are not effectively integrated; second, the determination of feature weights depends on expert experience, which is not objective enough; and third, the personnel profile lacks a dynamic update mechanism, which cannot adapt to changes in talent growth and job requirements.

[0004] Therefore, there is an urgent need for a talent evaluation and personnel profile construction method that can achieve deep integration of multi-dimensional data, objective weight calculation, and dynamic profile updates to solve the above problems. Summary of the Invention

[0005] In a first aspect, embodiments of this application provide a method for constructing personnel profiles based on a multi-dimensional talent evaluation model, comprising the following steps: S100. Collect basic attribute data, ability data, performance data, behavioral data, and social data of talents as raw data, construct a tree-like hierarchical comprehensive quality index system based on the raw data, and output a structured index score vector for each talent. S200. Based on the structured index score vector, the principal component analysis algorithm is used to reduce the dimensionality of the original data and extract key features. The objective weights of each key feature are calculated by the entropy weight method, and the weight vector composed of each objective weight is output. S300. The weight vector composed of key features and objective weights is used as input features and fed into a pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent. S400. Using key characteristics as evaluation indicators, the analytic hierarchy process (AHP) is used to calculate the expert-corrected score for talent. S500. Calculate the talent's comprehensive evaluation score based on the talent's initial evaluation score and the talent's expert-corrected score, and construct a personal profile based on the structured indicator score vector, comprehensive evaluation score, and key characteristics; S600 updates individual profiles through timed, event-triggered, or evaluation-triggered methods, and monitors structured indicator score vectors or comprehensive evaluation scores through preset early warning rules, generating growth early warning information for talents that meet the early warning conditions.

[0006] Furthermore, the specific steps of step S100 are as follows: S101. Collect basic attribute data from the enterprise human resources system; collect capability data from the enterprise training system and third-party assessment platforms; collect performance data from the enterprise performance management system; collect behavioral data from the logs of the enterprise office automation system and project management tools; collect social data from the application programming interface of the enterprise instant messaging tool and the database of the enterprise evaluation system; S102. Construct a tree-like hierarchical structure for the comprehensive quality index system, wherein the tree-like hierarchical structure includes leaf nodes and non-leaf nodes, and the leaf nodes are associated with specific data items or basic calculation models; S103. Assign weights to each node in the tree-like hierarchical structure; S104. Based on the collected raw data, calculate the initial values ​​of the leaf nodes; S105. Based on the weights assigned to each node, the values ​​of non-leaf nodes are weighted and aggregated from bottom to top until the root node is reached; S106. Output a structured index score vector consisting of the root node value and the node values ​​of each level.

[0007] Furthermore, the specific steps of step S200 are as follows: S201. After preprocessing the collected raw data, a standardized data matrix is ​​constructed; S202. Standardized data matrix Calculate the covariance matrix The calculation formula is:

[0008] in, For standardized data matrix The mean matrix formed by the column mean vectors, Let m be the number of talents and n be the number of indicators; S203. On the covariance matrix Perform eigenvalue decomposition and calculate eigenvalues. and the corresponding feature vectors And sort them in descending order of their eigenvalues; S204. Select the top k eigenvalues ​​as key features based on the preset cumulative variance contribution rate threshold, and generate a key feature matrix. ; in, It is the projection matrix composed of the first k eigenvectors; S205. Based on the key feature matrix The objective weights of each key feature are calculated using the entropy weight method; S206. Output the weight vector composed of each objective weight.

[0009] Furthermore, the specific steps of step S205 are as follows: S2051. Calculate the entropy value of the j-th key feature. :

[0010] in, , Let be the eigenvalue of the i-th talent in the j-th key feature matrix, and n be the total number of talents; S2052. Calculate the difference coefficient of the j-th key feature. : ; S2053. Calculate the weight of the j-th key feature. :

[0011] Where k is the total number of key features.

[0012] Furthermore, the specific steps of step S300 are as follows: S301. Merge the key feature matrix with the weight vector to form a weighted input feature set; S302. Input the input feature set into the pre-trained random forest regression model, and use multiple decision trees inside the random forest model to vote or average the input features; S303. Summarize the outputs of all decision trees to obtain the preliminary evaluation score for each talent. .

[0013] Furthermore, the specific steps of step S302 are as follows: S3021. Construct a historical talent dataset containing a historical key feature matrix, a historical weight vector, and corresponding manually labeled real evaluation scores, and divide the historical talent dataset into a training set and a validation set; S3022. Initialize a random forest model containing T decision trees and set model parameters including the maximum depth of decision trees and the minimum number of samples required for node splits; S3023. Train the random forest model using the training set: Impurity is determined by the Gini index or information entropy. Select key features and thresholds that can minimize the impurity of node data for node splitting; The node splitting process is executed recursively to construct each decision tree; The recursive process stops splitting when any of the following conditions are met, and the current node is marked as a leaf node: The current node contains fewer samples than the minimum number of samples required for the node to split. The impurity of the current node is lower than the preset purity threshold; The decision tree reaches its maximum depth as set. S3024. Use the validation set to evaluate the prediction accuracy of the trained random forest model. Once the prediction accuracy meets the preset requirements, save the structure and parameters of the random forest model.

[0014] Furthermore, the specific steps of step S400 are as follows: S401. Using the extracted key features as evaluation indicators for the analytic hierarchy process (AHP), obtain a set number of domain experts to conduct pairwise importance comparisons of each evaluation indicator, and construct a judgment matrix. ; S402. Calculate the judgment matrix Consistency index and consistency ratio :

[0015]

[0016] in, To determine the largest eigenvalue of a matrix, To determine the order of a matrix; The average random consistency index; S403. When the consistency ratio When the judgment matrix passes the consistency test, the eigenvectors of the judgment matrix are calculated and normalized to obtain the subjective weights of each key feature. ; S404. Using Subjective Weighting Key characteristics of each talent Perform a weighted summation to calculate the expert-corrected score. : .

[0017] Furthermore, the specific steps of step S500 are as follows: S501. Based on the preliminary evaluation score Score correction by experts The comprehensive evaluation score of talents is calculated using a weighted fusion formula. :

[0018] in, These are the scoring weights for the random forest regression model; S502. Map the comprehensive evaluation score to a preset score range, with each score range corresponding to a level, and determine the evaluation level label based on the score range in which the comprehensive evaluation score falls; S503. Based on the node values ​​related to basic attribute data, capability data and historical performance in the structured indicator score vector, generate a static profile including basic information tags, capability tags and historical performance tags; S504. Based on the node values ​​in the structured indicator score vector that are related to real-time behavior, social interaction, or behavior and social interaction within a set time period, and the comprehensive evaluation score. This generates a dynamic profile that includes real-time behavior tags, capability change trend tags, and performance dynamic tags. S505. Display capability tags or key features in radar chart format, and display evaluation level tags in word cloud format.

[0019] Furthermore, the specific steps of step S600 are as follows: S601. Perform dynamic image updates: Steps S100 to S500 are re-executed according to a fixed cycle, and the static and dynamic profiles are updated using the latest collected raw data and comprehensive evaluation scores. When an event is detected that a talent has completed important training, obtained certification, or changed positions, the job information in the ability tags and basic information tags in the static profile is updated in real time. When the change in the comprehensive evaluation score compared to the previous record exceeds a preset threshold, the dynamic performance tag in the dynamic profile and the evaluation level tag generated based on the comprehensive evaluation score are updated. S602. Pre-build early warning types and the early warning conditions and rule thresholds for each type; S603. Based on real-time or periodic structured indicator score vectors or comprehensive evaluation scores, determine whether the preset warning conditions are met; S604. Generate early warning records for talents that meet the early warning conditions, and generate early warning reports containing cause analysis and adjustment suggestions according to the early warning type.

[0020] Secondly, embodiments of this application also provide a personnel profile construction system based on a multi-dimensional talent evaluation model, including: The data collection and indicator system construction module is used to collect basic attribute data, ability data, performance data, behavioral data and social data of talents as raw data, and construct a tree-like hierarchical comprehensive quality indicator system based on the raw data, and output the structured indicator score vector of each talent. The feature extraction and weighting module is used to extract key features from the original data by using the principal component analysis algorithm based on the structured index score vector, and to calculate the objective weight of each key feature by using the entropy weight method, and output the weight vector composed of each objective weight. The machine learning evaluation module is used to input the weight vector composed of key features and objective weights as input features into a pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent. The expert correction evaluation module is used to calculate the expert correction score of talents using key characteristics as evaluation indicators and the analytic hierarchy process. The comprehensive evaluation and profile building module is used to calculate the comprehensive evaluation score of talents based on their initial evaluation scores and expert-corrected scores, and to build a personal profile based on the structured indicator score vector, comprehensive evaluation score and key features. The dynamic update and early warning module is used to update personal profiles through timed, event-triggered, or evaluation-triggered methods, and monitor the structured indicator score vector or comprehensive evaluation score through preset early warning rules, generating growth early warning information for talents who meet the early warning conditions.

[0021] As can be seen from the above technical solutions, this application has the following advantages: The personnel profile construction method and system based on a multi-dimensional talent evaluation model provided in this application comprehensively reflects the overall capabilities and potential of talents by integrating multi-dimensional data on basic attributes, abilities, performance, behavior, and social interactions, avoiding the limitations of traditional reliance on single data. It calculates the objective weights of each key feature using the entropy weight method, reducing interference from subjective human factors and making talent evaluation more reasonable and accurate. By combining timed, event-based, and evaluation trigger mechanisms, it achieves dynamic updates to personnel profiles, ensuring that the profiles reflect the latest status of talents in real time and avoiding disconnection from reality. It provides data-driven support for talent selection, promotion, training, and team configuration, achieving efficient and rational enterprise human resource management. Attached Figure Description

[0022] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating the personnel profile construction method based on a multi-dimensional talent evaluation model of the present invention.

[0024] Figure 2 This is a schematic diagram of the personnel profile construction system based on a multi-dimensional talent evaluation model according to the present invention. Detailed Implementation

[0025] The various embodiments of this disclosure will be described more fully in the detailed steps of the personnel profile construction method based on the multi-dimensional talent evaluation model described below. This disclosure may have various embodiments, and adjustments and changes may be made therein. However, it should be understood that there is no intention to limit the various embodiments of this disclosure to the specific embodiments disclosed herein, but rather this disclosure should be understood to cover all adjustments, equivalents, and / or alternatives falling within the spirit and scope of the various embodiments of this disclosure.

[0026] This embodiment provides a method for constructing personnel profiles based on a multi-dimensional talent evaluation model. It integrates multi-dimensional data, constructs a dynamic evaluation system, realizes accurate talent profiles and real-time early warning, and improves the rationality and timeliness of human resource management.

[0027] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1 The diagram shows a flowchart of a personnel profile construction method based on a multi-dimensional talent evaluation model in a specific embodiment. The method includes the following steps: S100. Collect basic attribute data, ability data, performance data, behavioral data, and social data of talents as raw data, construct a tree-like hierarchical comprehensive quality index system based on the raw data, and output a structured index score vector for each talent. It should be noted that data was collected from multiple dimensions, including basic attributes, abilities, performance, behavior, and social interactions, covering all aspects of talent and providing a data foundation for subsequent talent evaluation and profile construction. The sources of the data were clearly defined, ensuring the accuracy and reliability of the data and avoiding biases in talent evaluation caused by data quality issues. By constructing a tree-like hierarchical comprehensive quality index system, structured index score vectors were output, providing standardized data input for feature extraction and evaluation model construction. S200. Based on the structured index score vector, the principal component analysis algorithm is used to reduce the dimensionality of the original data and extract key features. The objective weights of each key feature are calculated by the entropy weight method, and the weight vector composed of each objective weight is output. It should be noted that using principal component analysis (PCA) for dimensionality reduction can effectively extract key features, reduce data dimensionality, and improve the training efficiency and performance of the model. Furthermore, calculating the objective weights of each key feature using the entropy weight method reduces the interference of subjective human factors, making talent evaluation more reasonable and accurate. S300. The weight vector composed of key features and objective weights is used as input features and input into the pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent. It should be noted that the pre-trained random forest machine learning model can quickly calculate the initial evaluation score of talents, providing a basis for expert correction and comprehensive evaluation; the generalization ability and anti-overfitting ability of the random forest model can effectively handle high-dimensional data and complex features, improving the accuracy and reliability of talent evaluation. S400. Using key characteristics as evaluation indicators, the analytic hierarchy process (AHP) is used to calculate the expert-corrected score for talent. It should be noted that by constructing a judgment matrix through pairwise importance comparisons of key features by domain experts and conducting consistency checks, the experience and knowledge of experts can be integrated into talent evaluation, improving the rationality and reliability of the evaluation. The combination of expert-corrected scores and preliminary evaluation scores achieves the integration of subjective and objective evaluation, making up for the shortcomings of a single evaluation method. S500. Calculate the talent's comprehensive evaluation score based on the talent's initial evaluation score and the talent's expert-corrected score, and construct a personal profile based on the structured indicator score vector, comprehensive evaluation score, and key characteristics; It should be noted that the comprehensive evaluation score of talents is calculated by weighted fusion formula, which takes into account the results of machine learning evaluation and expert correction evaluation, making the evaluation more comprehensive and accurate. Based on the comprehensive evaluation score and key features, static and dynamic profiles are constructed, which can comprehensively reflect the current status and dynamic changes of talents, and provide an intuitive visualization tool for enterprise human resource management. S600. Update personal profiles through timed, event-triggered, or evaluation-triggered methods, and monitor structured indicator score vectors or comprehensive evaluation scores through preset early warning rules, generating growth early warning information for talents that meet the early warning conditions; It should be noted that the dynamic updating of the profile is achieved through timed, event-based, and evaluation-triggered mechanisms, ensuring that the profile can reflect the latest status of the talent in real time, avoiding the profile from becoming out of touch with the actual situation, and maintaining the timeliness and accuracy of the profile. By monitoring the structured indicator score vector or comprehensive evaluation score through preset early warning rules, early warning information is generated for talents that meet the early warning conditions, providing timely early warning and intervention support for the enterprise's talent management.

[0029] This embodiment collects multi-dimensional data on basic attributes, capabilities, performance, behavior, and social interactions. It employs principal component analysis and entropy weighting for feature extraction and weight calculation, and combines random forest model and analytic hierarchy process to achieve accurate evaluation. Dynamic updates and early warning mechanisms ensure the timeliness of the profile, providing enterprises with reasonable decision support.

[0030] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the specific implementation process in this embodiment, another method for constructing personnel profiles based on a multi-dimensional talent evaluation model is provided, taking the student training and management of a flight school as an application scenario. This method includes the following steps: S100. Collect basic attribute data, ability data, performance data, behavioral data, and social data of talents as raw data. Based on the raw data, construct a tree-structured hierarchical comprehensive quality index system and output a structured index score vector for each talent. The specific steps of step S100 are as follows: S101. Collect basic attribute data from the enterprise's human resources system, including age, education, major, years of work experience, and job category; collect capability data from the enterprise's training system and third-party assessment platforms, including professional skill certificates, training results, and skill assessment scores; collect performance data from the enterprise's performance management system, including annual / quarterly performance ratings, project completion rates, and KPI achievement values; collect behavioral data from the enterprise's office automation system and project management tool logs, including work duration distribution, task collaboration frequency, and meeting participation; and collect social data from the enterprise's instant messaging tool's application programming interface and the enterprise's evaluation system database, including team communication frequency, cross-departmental collaboration frequency, and peer evaluation scores. For example, five types of raw data from 3,000 students were collected from various business systems of the school, and a hierarchical comprehensive quality index system was constructed, as shown in Table 1 below: Table 1

[0031] During the data collection process, structured data is automatically synchronized through the system API interface, while unstructured data (such as flight training logs and student peer review texts) is captured and stored in batches to ensure data integrity and timeliness. S102. Construct a tree-like hierarchical structure for the comprehensive quality index system, wherein the tree-like hierarchical structure includes leaf nodes and non-leaf nodes, and the leaf nodes are associated with specific data items or basic calculation models; S103. Assign weights to each node in the tree-like hierarchical structure; S104. Calculate the initial values ​​of the leaf nodes based on the collected raw data; S105. Based on the weights assigned to each node, the values ​​of non-leaf nodes are weighted and aggregated from bottom to top until the root node is reached; S106. Output a structured index score vector consisting of the root node value and the node values ​​of each level; For example, a three-level tree-like hierarchical structure is constructed, with the root node being "Student's Comprehensive Quality," second-level nodes including five categories: "Basic Qualities," "Professional Abilities," "Practical Performance," "Collaboration Ability," and "Growth Potential," and third-level leaf nodes associated with specific data items or calculation models; for example: Root node: Student's overall quality Second-level node: Professional capabilities Leaf nodes: Flight simulation operation score, flight theory score, and emergency response capability assessment score. Weights are assigned to each node, and the Analytic Hierarchy Process (AHP) is used to assist in decision-making. The total weight of the root node is 1, and the weights of the second-level nodes are as follows: basic literacy 0.2, professional ability 0.35, practical performance 0.2, teamwork ability 0.15, and growth potential 0.1. The weights of the leaf nodes are allocated according to the importance of their respective second-level nodes. For example, under "professional ability", the weights of the leaf nodes are: flight simulation operation score 0.4, flight theory score 0.35, and emergency response ability assessment score 0.25. The initial value of the leaf node is calculated based on the collected raw data. For example, a student's original "flight simulation operation score" is 89 points, which is standardized to 0.89 (out of 1 point). The non-leaf node values ​​are weighted and aggregated from bottom to top. For example, the value of the "Professional Ability" node is 89 × 0.4 + 95 × 0.35 + 82 × 0.25 = 89.15 points, which is 0.8915 on a 1-point scale. The root node "Student Comprehensive Quality" value is the sum of the values ​​of each secondary node × their corresponding weights. Output a structured index score vector, for example [0.92 (basic literacy), 0.8915 (professional ability), 0.87 (practical performance), 0.91 (collaboration ability), 0.85 (growth potential), 0.886 (root node)]; S200. Based on the structured index score vector, principal component analysis is used to reduce the dimensionality of the original data and extract key features. The objective weights of each key feature are then calculated using the entropy weight method, and a weight vector composed of these objective weights is output. The specific steps of step S200 are as follows: S201. After preprocessing the collected raw data, a standardized data matrix is ​​constructed; Specifically, the preprocessing steps are as follows: S2011. For missing values ​​in continuous data in the collected raw data, the mean imputation method is used; for missing values ​​in discrete data in the collected raw data, the mode imputation method is used; and outliers are identified and removed from the imputed raw data based on the 3σ principle to complete the data cleaning. S2012. Perform min-max standardization on the cleaned continuous data to eliminate the influence of dimensions. The standardization formula is:

[0032] in, Let be the original value of the i-th talent on the j-th continuous index. and Let $\frac{j}{j}$ be the minimum and maximum values ​​of the j-th continuous indicator among all talents, respectively. Let be the standardized value of the i-th talent on the j-th continuous index; S2013. Perform word segmentation on text data in unstructured behavioral data and social data, and convert the segmented text data into numerical structured feature vectors through the bag-of-words model to complete the structured transformation; S2014. Integrate the processed data and construct a standardized data matrix; For example, data cleaning: missing values ​​in continuous data are filled with the mean (e.g., if a student's "emergency response ability assessment score" is missing, it is filled with the mean of 78 points for the same indicator among 50 students in the same major); missing values ​​in discrete data are filled with the mode (e.g., if "student category" is missing, it is filled with "full-time on-campus student"); outliers are removed based on the 3σ principle, such as the 8 data points where the "course completion rate" exceeds the range of 60%-100%. Standardization processing: Min-max standardization is performed on continuous data. For example, a student's original score for "flight theory training" is 95 points. The minimum score for this indicator among students in the same batch is 70 points and the maximum score is 98 points. The standardized value is (95-70) / (98-70)≈0.8929. Structured transformation: Unstructured data (such as student peer review text "This student has solid flying skills, strong teamwork awareness, and quick emergency response") is segmented using jieba and converted into structured feature vectors [1,1,1,0,...] using the bag-of-words model; Construct a standardized data matrix with 3000 rows (number of students) × 56 columns (number of indicators); S202. Standardized data matrix Calculate the covariance matrix The calculation formula is:

[0033] in, For standardized data matrix The mean matrix formed by the column mean vectors, Let m be the number of talents and n be the number of indicators; S203. On the covariance matrix Perform eigenvalue decomposition and calculate eigenvalues. and the corresponding feature vectors And sort them in descending order of their eigenvalues; S204. Select the top k eigenvalues ​​as key features based on the preset cumulative variance contribution rate threshold, and generate a key feature matrix. ; in, It is the projection matrix composed of the first k eigenvectors; For example, principal component analysis (PCA) is used to reduce the dimensionality of the standardized data matrix: the covariance matrix is ​​calculated, and the covariance matrix C is calculated for the 3000×56 standardized data matrix X, using the following formula:

[0034] in, The mean matrix is ​​given, and m=3000 represents the number of students. The covariance matrix is ​​decomposed using eigenvalues, and the top 15 eigenvalues ​​with a cumulative variance contribution rate ≥88% are selected as key features, generating a 15×56 key feature matrix Y (the projection matrix U is composed of the top 15 eigenvectors). ); S205. Based on the key feature matrix The objective weights of each key feature are calculated using the entropy weight method; the specific steps of step S205 are as follows: S2051. Calculate the entropy value of the j-th key feature. :

[0035] in, , Let be the feature value of the i-th talent in the j-th key feature matrix, and n be the total number of talents; S2052. Calculate the difference coefficient of the j-th key feature. : ; S2053. Calculate the weight of the j-th key feature. :

[0036] Where k is the total number of key features; S206. Output the weight vector composed of each objective weight; For example, the weights are calculated using the entropy weighting method based on the key feature matrix: the entropy value of the j-th key feature is calculated. Taking the fourth key feature (corresponding to the principal component related to "flight simulation operation") as an example, =0.78; Calculate the coefficient of difference. =0.22; Calculate weights Assuming the sum of the difference coefficients of the 15 key features is 2.13, the weight of the 4th key feature... =0.22 / 2.13≈0.1033; The output weight vector W = [0.072, 0.085, 0.091, 0.1033, 0.068, 0.079, 0.088, 0.065, 0.073, 0.082, 0.069, 0.076, 0.084, 0.071, 0.073]; S300. The weight vector composed of key features and objective weights is used as input features and fed into a pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent; the specific steps of step S300 are as follows: S301. Merge the key feature matrix with the weight vector to form a weighted input feature set; For example, the key feature matrix Y (3000×15) is fused with the weight vector W (1×15), and the weighted input feature set Z=Y×W is obtained through matrix multiplication. T Z is a vector of 3000×1; S302. Input the input feature set into the pre-trained random forest regression model, and use multiple decision trees within the random forest model to vote or average the input features; the specific steps of step S302 are as follows: S3021. Construct a historical talent dataset containing a historical key feature matrix, a historical weight vector, and corresponding manually labeled real evaluation scores, and divide the historical talent dataset into a training set and a validation set; S3022. Initialize a random forest model containing T decision trees and set model parameters including the maximum depth of decision trees and the minimum number of samples required for node splits; S3023. Train the random forest model using the training set: Impurity is determined by the Gini index or information entropy. Select key features and thresholds that can minimize the impurity of node data for node splitting; The node splitting process is executed recursively to construct each decision tree; The recursive process stops splitting when any of the following conditions are met, and the current node is marked as a leaf node: The current node contains fewer samples than the minimum number of samples required for the node to split. The impurity of the current node is lower than the preset purity threshold; The decision tree reaches its maximum depth as set. S3024. Use the validation set to evaluate the prediction accuracy of the trained random forest model. Once the prediction accuracy meets the preset requirements, save the structure and parameters of the random forest model. S303. Summarize the outputs of all decision trees to obtain the preliminary evaluation score for each talent. ; For example, a historical dataset is constructed: the historical key feature matrix, weight vector and manually labeled real evaluation scores (1-10 points) of 1,500 students of the school in the past 4 years are collected and divided into a training set (1,050 records) and a validation set (450 records) in a 7:3 ratio.

[0037] Model training: Initialize a random forest regression model containing 120 decision trees, set the maximum depth of the decision trees to 10, and the minimum number of samples for node splitting to 6; use information entropy to measure impurity, recursively split until the stopping condition is met, and complete the model training.

[0038] Model validation: The validation set yielded a mean absolute error (MAE) of 0.29 and a coefficient of determination (R²) of 0.89, meeting the preset accuracy requirements. The model parameters were then saved.

[0039] Preliminary evaluation score calculation: The input feature set Z is input into the model, and the average of the prediction results from 120 decision trees is taken. This is the preliminary evaluation score for a flight technology student. =9.1 points; S400. Using key characteristics as evaluation indicators, the analytic hierarchy process (AHP) is employed to calculate the expert-corrected score for the talent. The specific steps of S400 are as follows: S401. Using the extracted key features as evaluation indicators for the analytic hierarchy process (AHP), a set number of domain experts are obtained to conduct pairwise importance comparisons of each evaluation indicator, and a judgment matrix is ​​constructed. ; S402. Calculate the judgment matrix Consistency index and consistency ratio :

[0040]

[0041] in, To determine the largest eigenvalue of a matrix, To determine the order of a matrix; The average random consistency index; S403. When the consistency ratio When the judgment matrix passes the consistency test, the eigenvectors of the judgment matrix are calculated and normalized to obtain the subjective weights of each key feature. ; For example, eight experts in the field (including three flight teaching professors, two academic management experts, and three senior flight instructors) were invited to construct a judgment matrix using 1-9 scaling with 15 key characteristics as evaluation indicators, and the maximum eigenvalue was calculated. =15.32, Consistency Index The average random consistency index RI = 1.59, and the consistency ratio CR = 0.0229 / 1.59 ≈ 0.0144 < 0.1, thus passing the consistency test; S404. Using Subjective Weighting Key characteristics of each talent Perform a weighted summation to calculate the expert-corrected score. : ; For example, the subjective weight vector is obtained by normalizing the eigenvectors of the judgment matrix. The expert-corrected score is calculated by weighted summation:

[0042] The expert-corrected scores of the above trainees =8.8 points; S500. Calculate the talent's comprehensive evaluation score based on the initial evaluation score and the expert-corrected score, and construct a personal profile based on the structured indicator score vector, comprehensive evaluation score, and key characteristics; the specific steps of step S500 are as follows: S501. Based on the preliminary evaluation score Score correction by experts The comprehensive evaluation score of talents is calculated using a weighted fusion formula. :

[0043] in, These are the scoring weights for the random forest regression model; S502. Map the comprehensive evaluation score to a preset score range, with each score range corresponding to a level, and determine the evaluation level label based on the score range in which the comprehensive evaluation score falls; For example, using a weighted fusion formula, the scoring weight α of the random forest model is 0.75 (determined through cross-validation): Substituting the data, we get: S = 0.75 × 9.1 + 0.25 × 8.8 = 9.025 points, which is mapped to the "Excellent" grade range (8.5-10 points). The evaluation grade label is determined to be "Excellent". S503. Based on the node values ​​related to basic attribute data, capability data and historical performance in the structured indicator score vector, generate a static profile including basic information tags, capability tags and historical performance tags; S504. Based on the node values ​​in the structured indicator score vector that are related to real-time behavior, social interaction, or behavior and social interaction within a set time period, and the comprehensive evaluation score. This generates a dynamic profile that includes real-time behavior tags, capability change trend tags, and performance dynamic tags. S505. Display capability tags or key features in the form of radar charts, and display evaluation level tags in the form of word clouds; For example, a static profile is generated based on a structured index score vector, which includes basic information tags {20 years old, undergraduate student, majoring in flight technology, 2-year enrollment period, full-time student}, ability tags {Civil Aviation English Level 4, advanced flight simulation operation (89 points), excellent flight theory (95 points)}, and historical performance tags {excellent academic performance in the first semester of the 2024-2025 academic year, 100% course completion rate}. Dynamic profile: Based on real-time data, generate real-time behavior tags {18 group collaborations in the past 30 days, 16 monthly flight practice hours, core learning period 8:00-11:00}, ability change trend tags {7 points improved in flight simulation operation score in the past 6 months, 5 points improved in emergency response ability}, and performance dynamic tags {1 level improved in academic grade in the last semester compared to the previous academic year, comprehensive evaluation score 9.025 (excellent)}; Visualization: The capability tags are displayed in the form of a radar chart, clearly presenting the capability scores of each dimension; the evaluation level tags and core capability tags are displayed in the form of a word cloud, highlighting core strengths; S600. Update individual profiles through timed, event-triggered, or evaluation-triggered methods, and monitor structured indicator score vectors or comprehensive evaluation scores through preset early warning rules, generating growth early warning information for talents who meet the early warning conditions; the specific steps of step S600 are as follows: S601. Perform dynamic image updates: Steps S100 to S500 are re-executed according to a fixed cycle, and the static and dynamic profiles are updated using the latest collected raw data and comprehensive evaluation scores. When an event is detected that a talent has completed important training, obtained certification, or changed positions, the job information in the ability tags and basic information tags in the static profile is updated in real time. When the change in the comprehensive evaluation score compared to the previous record exceeds a preset threshold, the dynamic performance tag in the dynamic profile and the evaluation level tag generated based on the comprehensive evaluation score are updated. For example, scheduled updates: Set the first week of each semester as a fixed update cycle, re-execute steps S100-S500, collect the latest data from the previous semester, and update the historical performance tags in the static profile and all tags in the dynamic profile; Event-triggered updates: When it is detected that a trainee has completed "Advanced Flight Technology Training" (important training), obtained "Civil Aviation English Level 6 Certificate" (certification acquisition), or changed majors (job change), the professional information in the ability tag and basic information tag will be updated in real time; Evaluation trigger update: The preset threshold for changes in the comprehensive evaluation score is ±8%. If the student's comprehensive evaluation score for the next semester is 8.3, which is 8.03% lower than the previous semester's score of 9.025, the performance dynamic label and evaluation level label in the dynamic profile will be updated to {2024-2025 academic year, second semester, academic level is good, comprehensive evaluation score is 8.3 (good)}. S602. Pre-build early warning types and the early warning conditions and rule thresholds for each type; S603. Based on real-time or periodic structured indicator score vectors or comprehensive evaluation scores, determine whether the preset warning conditions are met; S604. Generate early warning records for talents that meet the early warning conditions, and generate early warning reports containing cause analysis and adjustment suggestions according to the early warning type; The warning report generates differentiated content based on the viewer's identity: a report containing personal adjustment plans is generated for trainees, and a report containing teaching guidance strategies is generated for instructors; For example, the preset warning types include three categories: academic warning, ability improvement warning, and practical participation warning, with clearly defined warning conditions and thresholds; for example, the academic warning condition is that the core course KPI achievement value is <70%, and the ability improvement warning condition is that the key ability score has decreased for two consecutive cycles with a cumulative decrease of >10%; Real-time monitoring of structured indicator score vectors and comprehensive evaluation scores shows that a student in a certain major has achieved 65% of the core course KPI, which meets the academic warning criteria. Generate alert records and generate differentiated alert reports based on the viewer's identity: Student Perspective Report: Includes self-attribution analysis (e.g., insufficient study time for core courses) and personalized adjustment plans (e.g., developing a daily 2-hour dedicated tutoring plan and scheduling tutor Q&A sessions). Instructor's Perspective Report: Includes analysis of causes (such as students' lack of in-depth understanding of navigation theory) and teaching guidance strategies (such as recommending specialized tutoring materials, arranging one-on-one Q&A sessions, and adjusting teaching cases to suit students' cognition).

[0044] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0045] like Figure 2 As shown, the following are embodiments of the personnel profile construction system based on a multi-dimensional talent evaluation model provided in this disclosure. This system and the personnel profile construction method based on a multi-dimensional talent evaluation model in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the personnel profile construction system based on a multi-dimensional talent evaluation model, please refer to the embodiments of the personnel profile construction method based on a multi-dimensional talent evaluation model described above.

[0046] The system includes: The data collection and indicator system construction module is used to collect basic attribute data, ability data, performance data, behavioral data and social data of talents as raw data, and construct a tree-like hierarchical comprehensive quality indicator system based on the raw data, and output the structured indicator score vector of each talent. The feature extraction and weighting module is used to extract key features from the original data by using the principal component analysis algorithm based on the structured index score vector, and to calculate the objective weight of each key feature by using the entropy weight method, and output the weight vector composed of each objective weight. The machine learning evaluation module is used to input the weight vector composed of key features and objective weights as input features into a pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent. The expert correction evaluation module is used to calculate the expert correction score of talents using key characteristics as evaluation indicators and the analytic hierarchy process. The comprehensive evaluation and profile building module is used to calculate the comprehensive evaluation score of talents based on their initial evaluation scores and expert-corrected scores, and to build a personal profile based on the structured indicator score vector, comprehensive evaluation score and key features. The dynamic update and early warning module is used to update personal profiles through timed, event-triggered, or evaluation-triggered methods, and monitor the structured indicator score vector or comprehensive evaluation score through preset early warning rules, generating growth early warning information for talents who meet the early warning conditions.

[0047] This embodiment achieves accurate talent profiling and real-time early warning through the interactive collaboration of modules including data collection and indicator system construction, feature extraction and weighting, machine learning evaluation, expert correction evaluation, comprehensive evaluation and profile construction, and dynamic updating and early warning, thereby improving the rationality and timeliness of human resource management.

[0048] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing personnel profiles based on a multi-dimensional talent evaluation model, characterized in that, Includes the following steps: S100. Collect basic attribute data, ability data, performance data, behavioral data, and social data of talents as raw data, construct a tree-like hierarchical comprehensive quality index system based on the raw data, and output a structured index score vector for each talent. S200. Based on the structured index score vector, the principal component analysis algorithm is used to reduce the dimensionality of the original data and extract key features. The objective weights of each key feature are calculated by the entropy weight method, and the weight vector composed of each objective weight is output. S300. The weight vector composed of key features and objective weights is used as input features and fed into a pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent. S400. Using key characteristics as evaluation indicators, the analytic hierarchy process (AHP) is used to calculate the expert-corrected score for talent. S500. Calculate the talent's comprehensive evaluation score based on the talent's initial evaluation score and the talent's expert-corrected score, and construct a personal profile based on the structured indicator score vector, comprehensive evaluation score, and key characteristics; S600 updates individual profiles through timed, event-triggered, or evaluation-triggered methods, and monitors structured indicator score vectors or comprehensive evaluation scores through preset early warning rules, generating growth early warning information for talents that meet the early warning conditions.

2. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 1, characterized in that, The specific steps of step S100 are as follows: S101. Collect basic attribute data from the enterprise's human resources system; collect capability data from the enterprise's training system and third-party assessment platforms; collect performance data from the enterprise's performance management system; Collect behavioral data from logs of enterprise office automation systems and project management tools; Collect social data from the application programming interfaces (APIs) of enterprise instant messaging tools and the databases of enterprise evaluation systems; S102. Construct a tree-like hierarchical structure for the comprehensive quality index system, wherein the tree-like hierarchical structure includes leaf nodes and non-leaf nodes, and the leaf nodes are associated with specific data items or basic calculation models; S103. Assign weights to each node in the tree-like hierarchical structure; S104. Based on the collected raw data, calculate the initial values ​​of the leaf nodes; S105. Based on the weights assigned to each node, the values ​​of non-leaf nodes are weighted and aggregated from bottom to top until the root node is reached; S106. Output a structured index score vector consisting of the root node value and the node values ​​of each level.

3. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 1, characterized in that, The specific steps of step S200 are as follows: S201. After preprocessing the collected raw data, a standardized data matrix is ​​constructed; S202. Standardized data matrix Calculate the covariance matrix The calculation formula is: in, For standardized data matrix The mean matrix formed by the column mean vectors, Let m be the number of talents and n be the number of indicators; S203. On the covariance matrix Perform eigenvalue decomposition and calculate eigenvalues. and the corresponding feature vectors And sort them in descending order of their eigenvalues; S204. Select the top k eigenvalues ​​as key features based on the preset cumulative variance contribution rate threshold, and generate a key feature matrix. ; in, It is the projection matrix composed of the first k eigenvectors; S205. Based on the key feature matrix The objective weights of each key feature are calculated using the entropy weight method; S206. Output the weight vector composed of each objective weight.

4. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 3, characterized in that, The specific steps of step S205 are as follows: S2051. Calculate the entropy value of the j-th key feature. : in, , Let be the eigenvalue of the i-th talent in the j-th key feature matrix, and n be the total number of talents; S2052. Calculate the difference coefficient of the j-th key feature. : ； S2053. Calculate the weight of the j-th key feature. : Where k is the total number of key features.

5. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 1, characterized in that, The specific steps of step S300 are as follows: S301. Merge the key feature matrix with the weight vector to form a weighted input feature set; S302. Input the input feature set into the pre-trained random forest regression model, and use multiple decision trees inside the random forest model to vote or average the input features; S303. Summarize the outputs of all decision trees to obtain the preliminary evaluation score for each talent. .

6. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 5, characterized in that, The specific steps of step S302 are as follows: S3021. Construct a historical talent dataset containing a historical key feature matrix, a historical weight vector, and corresponding manually labeled real evaluation scores, and divide the historical talent dataset into a training set and a validation set; S3022. Initialize a random forest model containing T decision trees and set model parameters including the maximum depth of decision trees and the minimum number of samples required for node splits; S3023. Train the random forest model using the training set: Impurity is determined by the Gini index or information entropy. Select key features and thresholds that can minimize the impurity of node data for node splitting; The node splitting process is executed recursively to construct each decision tree; The recursive process stops splitting when any of the following conditions are met, and the current node is marked as a leaf node: The current node contains fewer samples than the minimum number of samples required for the node to split. The impurity of the current node is lower than the preset purity threshold; The decision tree reaches its maximum depth as set. S3024. Use the validation set to evaluate the prediction accuracy of the trained random forest model. Once the prediction accuracy meets the preset requirements, save the structure and parameters of the random forest model.

7. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 1, characterized in that, The specific steps of step S400 are as follows: S401. Using the extracted key features as evaluation indicators for the analytic hierarchy process (AHP), obtain a set number of domain experts to conduct pairwise importance comparisons of each evaluation indicator, and construct a judgment matrix. ; S402. Calculate the judgment matrix Consistency index and consistency ratio : in, To determine the largest eigenvalue of a matrix, To determine the order of a matrix; The average random consistency index; S403. When the consistency ratio When the judgment matrix passes the consistency test, the eigenvectors of the judgment matrix are calculated and normalized to obtain the subjective weights of each key feature. ; S404. Using Subjective Weighting Key characteristics of each talent Perform a weighted summation to calculate the expert-corrected score. : 。 8. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 1, characterized in that, The specific steps of step S500 are as follows: S501. Based on the preliminary evaluation score Score correction by experts The comprehensive evaluation score of talents is calculated using a weighted fusion formula. : in, These are the scoring weights for the random forest regression model; S502. Map the comprehensive evaluation score to a preset score range, with each score range corresponding to a level, and determine the evaluation level label based on the score range in which the comprehensive evaluation score falls; S503. Based on the node values ​​related to basic attribute data, capability data and historical performance in the structured indicator score vector, generate a static profile including basic information tags, capability tags and historical performance tags; S504. Based on the node values ​​in the structured indicator score vector that are related to real-time behavior, social interaction, or behavior and social interaction within a set time period, and the comprehensive evaluation score. This generates a dynamic profile that includes real-time behavior tags, capability change trend tags, and performance dynamic tags. S505. Display capability tags or key features in radar chart format, and display evaluation level tags in word cloud format.

9. The method for constructing personnel profiles based on a multi-dimensional talent evaluation model according to claim 8, characterized in that, The specific steps of step S600 are as follows: S601. Perform dynamic image updates: Steps S100 to S500 are re-executed according to a fixed cycle, and the static and dynamic profiles are updated using the latest collected raw data and comprehensive evaluation scores. When an event is detected that a talent has completed important training, obtained certification, or changed positions, the job information in the ability tags and basic information tags of the static profile is updated in real time. When the change in the comprehensive evaluation score compared to the previous record exceeds a preset threshold, the dynamic performance tag in the dynamic profile and the evaluation level tag generated based on the comprehensive evaluation score are updated. S602. Pre-build early warning types and the early warning conditions and rule thresholds for each type; S603. Based on real-time or periodic structured indicator score vectors or comprehensive evaluation scores, determine whether the preset warning conditions are met; S604. Generate early warning records for talents that meet the early warning conditions, and generate early warning reports containing cause analysis and adjustment suggestions according to the early warning type.

10. A personnel profile construction system based on a multi-dimensional talent evaluation model, characterized in that, include: The data collection and indicator system construction module is used to collect basic attribute data, ability data, performance data, behavioral data and social data of talents as raw data, and construct a tree-like hierarchical comprehensive quality indicator system based on the raw data, and output the structured indicator score vector of each talent. The feature extraction and weighting module is used to extract key features from the original data by using the principal component analysis algorithm based on the structured index score vector, and to calculate the objective weight of each key feature by using the entropy weight method, and output the weight vector composed of each objective weight. The machine learning evaluation module is used to input the weight vector composed of key features and objective weights as input features into a pre-trained random forest machine learning model to calculate the preliminary evaluation score of the talent. The expert correction evaluation module is used to calculate the expert correction score of talents using key characteristics as evaluation indicators and the analytic hierarchy process. The comprehensive evaluation and profile building module is used to calculate the comprehensive evaluation score of talents based on their initial evaluation scores and expert-corrected scores, and to build a personal profile based on the structured indicator score vector, comprehensive evaluation score and key features. The dynamic update and early warning module is used to update personal profiles through timed, event-triggered, or evaluation-triggered methods, and monitor the structured indicator score vector or comprehensive evaluation score through preset early warning rules, generating growth early warning information for talents who meet the early warning conditions.

Images