Human resource data management method based on big data

Abstract

The invention provides a human resource data management method based on big data, and belongs to the technical field of data management, and the method comprises the steps: carrying out the initialization of a Gaussian mixture model based on human resource data and the number of scene types, inputting the human resource data into the initialized Gaussian mixture model, and obtaining a plurality of clusters; the scene type comprises a recruitment scene, a training scene and a performance evaluation scene; for each data point in the human resource data, calculating a clustering soft membership degree of the data point based on the initialized Gaussian mixture model; performing iterative updating on all clusters based on the clustering soft membership of all data points and an expectation maximization iterative algorithm until an iterative convergence condition is met; determining a recruitment data set, a training data set and a performance evaluation data set based on the target cluster obtained by the last update; and performing data management on the recruitment data set, the training data set and the performance evaluation data set. The management efficiency of the human resource data can be improved.

Description

Technical Field

[0001] The present application belongs to the field of data management technology, and more specifically, relates to a human resources data management method based on big data. Background Art

[0002] In the digital age, corporate human resource management generates massive amounts of data, and how to effectively manage and utilize this data has become a key issue.

[0003] Currently, existing human resources data management technologies have limited data integration capabilities. Human resources data within an enterprise comes from a wide range of sources, spanning multiple systems with inconsistent data formats and standards. This makes it difficult for existing technologies to efficiently integrate this multi-source, heterogeneous data. Furthermore, when applying existing technologies to massive amounts of data, it often struggles to quickly obtain the required data and analytical results. All of these factors hinder enterprises from obtaining comprehensive and accurate information support during their human resources decision-making processes, reducing management efficiency and the scientific nature of their decisions. Therefore, a more efficient human resources data management approach based on big data is urgently needed. Summary of the Invention

[0004] The purpose of this application is to provide a human resources data management method based on big data to improve the efficiency of human resources data management.

[0005] A first aspect of an embodiment of the present application provides a human resources data management method based on big data, comprising: Initializing a Gaussian mixture model based on the human resources data and the number of scenario types, and inputting the human resources data into the initialized Gaussian mixture model to obtain multiple clusters; the scenario types include recruitment scenarios, training scenarios, and performance evaluation scenarios, and each scenario type corresponds to a cluster; For each data point in the human resources data, a cluster soft membership of the data point is calculated based on the initialized Gaussian mixture model; the cluster soft membership refers to the probability that the data point belongs to each cluster; all clusters are iteratively updated based on the cluster soft membership of all data points and the expectation-maximization iterative algorithm until the rate of change of cluster parameters between two adjacent iterations meets the iterative convergence condition; A recruitment dataset, a training dataset, and a performance evaluation dataset are determined based on the target clustering obtained from the last update; and data management is performed on the recruitment dataset, the training dataset, and the performance evaluation dataset respectively.

[0006] A second aspect of the embodiments of the present application provides a human resources data management device based on big data, comprising: A model initialization module is used to initialize the Gaussian mixture model based on the human resources data and the number of scenario types, and input the human resources data into the initialized Gaussian mixture model to obtain multiple clusters; the scenario types include recruitment scenarios, training scenarios, and performance evaluation scenarios, and each scenario type corresponds to a cluster; a cluster updating module for calculating, for each data point in the human resources data, a cluster soft membership of the data point based on the initialized Gaussian mixture model; the cluster soft membership refers to the probability that the data point belongs to each cluster; and iteratively updating all clusters based on the cluster soft membership of all data points and an expectation-maximization iterative algorithm until the rate of change of cluster parameters between two adjacent iterations satisfies an iterative convergence condition; The classification management module is used to determine the recruitment data set, training data set and performance evaluation data set based on the target clustering obtained in the last update; and perform data management on the recruitment data set, training data set and performance evaluation data set respectively.

[0007] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned human resources data management method based on big data when executing the computer program.

[0008] According to a fourth aspect of an embodiment of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned human resources data management method based on big data are implemented.

[0009] The beneficial effects of the human resources data management method based on big data provided by the embodiment of the present application are: on the one hand, the embodiment of the present application uses the Gaussian Mixture Model (GMM) to initialize clustering according to human resources data and different scenario types, and uses the Expectation-Maximization Algorithm (EM) for iterative updates, which can accurately divide the data into three types of data sets: recruitment, training, and performance evaluation, thereby improving the accuracy and scientificity of data classification. On the other hand, the embodiment of the present application manages the classified data sets separately, and can formulate exclusive management strategies for different scenario requirements, thereby avoiding the management difficulties caused by data mixing. This not only effectively improves the efficiency of human resources data management, but also provides more accurate and targeted data support for the company's human resources decision-making, helping the company to optimize the human resources management process. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0011] Figure 1 A flowchart of a human resources data management method based on big data provided in one embodiment of the present application; Figure 2 A structural block diagram of a human resources data management device based on big data provided in one embodiment of the present application; Figure 3 A schematic block diagram of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0012] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0013] In order to make the purpose, technical solutions and advantages of this application clearer, specific embodiments will be described below with reference to the accompanying drawings.

[0014] Please refer to Figure 1 , Figure 1 This is a flow chart of a human resources data management method based on big data provided in one embodiment of the present application. The method can be executed by an electronic device. Specifically, the method can include S101 to S103.

[0015] S101: Initialize a Gaussian mixture model based on human resources data and the number of scenario types, and input the human resources data into the initialized Gaussian mixture model to obtain multiple clusters; the scenario types include recruitment scenarios, training scenarios, and performance evaluation scenarios, and each scenario type corresponds to a cluster.

[0016] In this embodiment, human resources data may include multi-dimensional structured / unstructured data such as unclassified employee basic information, recruitment data, training records, performance data, etc. Scenario type refers to the classification of business application scenarios, such as recruitment scenarios, training scenarios, and performance evaluation scenarios. Each scenario corresponds to specific data characteristics and business goals. The Gaussian mixture model is a probabilistic generation model for clustering. It assumes that the data is a mixture of multiple Gaussian distributions, and each cluster corresponds to a Gaussian component, which in this embodiment corresponds to a scenario type. The parameters of the Gaussian mixture model may include the number of Gaussian components (i.e., the number of clusters), the mean vector of each component (indicating the center of the scenario data), the covariance matrix (indicating the discreteness of the data distribution), and the prior probability (indicating the proportion of scenario data).

[0017] This embodiment uses a Gaussian mixture model to soft-cluster human resources data, probabilistically assigning each data point to a Gaussian component corresponding to a different scenario, thereby achieving scenario-based data segmentation. For example, recruitment scenario data exhibits a high probability distribution in the dimensions of "job keywords" and "resume matching", while performance data exhibits a high probability distribution in the dimensions of "KPI completion rate" and "rating level".

[0018] For example, this embodiment preprocesses human resources data, such as removing missing values ​​and outliers and unifying time formats. This embodiment also performs word segmentation and keyword extraction on unstructured data such as resume text. It also performs one-hot encoding on categorical data such as academic qualifications, generating feature vectors such as [length of service, academic qualification code, TF-IDF value of job keywords].

[0019] This example initializes the model by setting the number of Gaussian components to 3, corresponding to three types of scenarios. This example initializes parameters based on prior knowledge: the mean vector is set based on typical scenario data; the covariance matrix is ​​initialized to the identity matrix; and the prior probability is set based on the proportion of scenarios in historical data.

[0020] This example iteratively optimizes each Gaussian component, calculating the probability that each data point belongs to each scenario cluster. For example, the probability that a data point belongs to the recruitment scenario is 0.6, the probability that it belongs to the training scenario is 0.4, and the probability that it belongs to the performance scenario is 0.2. Based on this probability, this example re-estimates the mean, covariance, and prior probability of each Gaussian component to maximize the data log-likelihood. This iterative optimization step is repeated until the parameters converge or the preset number of iterations is reached.

[0021] This embodiment can assign each data point to the scenario cluster with the highest probability and generate a scenario label for data management and analysis of subsequent business scenarios.

[0022] S102: For each data point in the human resources data, the cluster soft membership of the data point is calculated based on the initialized Gaussian mixture model; the cluster soft membership refers to the probability that the data point belongs to each cluster; based on the cluster soft membership of all data points and the expectation-maximization iterative algorithm, all clusters are iteratively updated until the rate of change of cluster parameters between clusters obtained in two adjacent iterations meets the iterative convergence condition.

[0023] In this embodiment, the cluster soft membership of the data point is calculated based on the initialized Gaussian mixture model, specifically including: Based on the initialized Gaussian mixture model, the cluster soft membership of the data point is calculated using the first formula; The first formula is: ; in, ; ; in, For data points The probability of belonging to cluster k, that is, the data point The clustering soft membership of is the prior probability of cluster k, is the probability density function of the Gaussian distribution, is the i-th data point in the human resources data, each data point is a vector with multiple features, is the mean vector of cluster k, is the diagonal covariance matrix of cluster k, j is the index of the cluster, K is the total number of clusters, and d is the number of data points The characteristic dimension of is the diagonal covariance matrix The determinant of is an exponential function, is a vector The transpose of is the diagonal covariance matrix The inverse matrix of .

[0024] In this embodiment, cluster soft membership refers to the probability that a data point belongs to cluster k, with values ​​ranging from [0, 1], and the sum of all cluster probabilities equals 1. Cluster soft membership measures the degree of association between a data point and each scenario. For example, a particular employee's data may have a 60% probability of belonging to the "training scenario," a 30% probability of belonging to the "performance scenario," and a 30% probability of belonging to the "recruitment scenario." The first formula measures the contribution of a data point to cluster k by multiplying the prior probability by the Gaussian likelihood N(·). The denominator is then normalized to yield the probability that the data point belongs to cluster k. The expectation-maximization iterative algorithm includes the following steps: Step E: Calculating the soft membership of all data points using the current model parameters to quantify the probability of association between the data and the scenario; Step M: Updating the model parameters based on the soft membership of all data points to maximize the data log-likelihood; Convergence condition: The iteration terminates when the parameter update amplitude is less than a threshold or the rate of change in the log-likelihood stabilizes.

[0025] For example, this embodiment cleans, extracts features, and normalizes human resources data to generate a d-dimensional feature matrix. This embodiment sets initial model parameters based on the number of scenario types, K, = 3. This embodiment first calculates the cluster soft membership of all data points (i.e., the d-dimensional feature matrix), then updates the model parameters, and repeats until convergence.

[0026] This embodiment assigns each data point to a cluster whose soft membership meets the threshold based on a preset threshold, generating a scenario label. For example, if the threshold is set to 50%, and the soft memberships of a data point in the recruitment scenario, training scenario, and performance evaluation scenario are 80%, 60%, and 45%, respectively, the data point is labeled with the dual label recruitment and training, and is classified into the data sets corresponding to these two scenario types.

[0027] This embodiment avoids the defect of rigid clustering that can only perform single-type division through flexible data partitioning, and is applicable to cross-scenario data of this embodiment, such as employee development data involving both training and performance. Through iterative optimization, this embodiment's model can automatically learn the feature distribution of different scenarios, such as adjusting keyword weights as positions change in recruitment scenarios. The soft membership of this embodiment can be used for scenario association analysis, such as identifying reserve talent groups that are highly correlated with both recruitment and training scenarios, providing a multi-dimensional basis for human resource decision-making. The diagonal covariance matrix setting of this embodiment can reduce computational complexity and is applicable to high-dimensional human resource data.

[0028] S103: Determine a recruitment dataset, a training dataset, and a performance evaluation dataset based on the target cluster obtained in the last update; and perform data management on the recruitment dataset, the training dataset, and the performance evaluation dataset respectively.

[0029] In this embodiment, target clustering refers to the clustering results obtained after iterative optimization of human resources data using a Gaussian mixture model. Its core is to probabilistically or deterministically partition data by business scenario, forming dataset groups that closely match actual business needs. Each target cluster corresponds to a specific business scenario. Parameters for target clustering may include cluster centers (mean vectors), scenario labels, and data point soft membership thresholds. Recruitment datasets may include data related to talent selection, such as job requirements, resume information, interview evaluations, and certifications. Training datasets may include data related to employee development, such as training course records, assessment results, and study time. Performance evaluation datasets, used for employee assessment, may include data such as KPI completion rates and attendance records. Data management refers to performing operations such as storage, cleaning, analysis, and security management based on the characteristics of datasets in different scenarios to improve data availability.

[0030] This embodiment categorizes and stores human resources data by scenario tags based on the soft or hard membership results of target clustering, customizing management data based on scenario requirements. For example, recruitment data prioritizes real-time performance and candidate matching efficiency, training data focuses on process documentation and effectiveness evaluation, and performance data emphasizes objectivity and periodic analysis. This embodiment allows data points to belong to multiple datasets simultaneously, making it suitable for complex business scenarios.

[0031] Exemplarily, determining the recruitment dataset, training dataset, and performance evaluation dataset based on the target cluster obtained by the last update may include: Based on the cluster soft membership output by the Gaussian mixture model, data attribution rules are formulated. For example, if the cluster soft membership of a data point in the target cluster for a certain scenario type is greater than or equal to 0.6, the data point is directly assigned to the scenario dataset corresponding to that cluster. If the cluster soft membership of a data point in the target cluster for all scenario types is less than 0.3, it is considered noise data and filtered or re-clustered. After the various datasets are divided according to the rules, this embodiment can manage the recruitment dataset, training dataset, and performance evaluation dataset separately.

[0032] For example, data management for a recruitment dataset, a training dataset, and a performance evaluation dataset may include: (1) Perform batch data partitioning through the data processing engine: Read the human resources data with cluster probability labels output by the Gaussian mixture model. Traverse each data according to the preset attribution rules: For data with a single strong attribution feature, directly write it to the database table of the corresponding scenario, such as the "recruitment_data" table in MySQL; for cross-scenario data, generate a JSON file containing multi-scenario probabilities, store it in the "cross-scenario data" directory, and store the data in the corresponding multiple scenario database tables; for noisy data, write it to the "abnormal data" temporary table and annotate the original features for subsequent analysis.

[0033] (2) A tiered storage strategy is used that combines hot data with cold data. Hot data refers to data from the past year, while cold data refers to data older than one year. This embodiment can store hot data in a high-performance database to support real-time query and analysis, such as real-time candidate matching in recruitment scenarios. This embodiment can archive cold data to a distributed file system and regularly clean up invalid records, such as closed job recruitment data, to reduce storage costs.

[0034] (3) This embodiment can perform real-time updates, quality control, and effect analysis on the recruitment dataset. For example, new resumes are automatically synchronized with the recruitment platform at dawn every day. The matching degree is calculated with the recruitment scenario cluster center through the keyword matching algorithm to generate a subset of highly matched candidates. Sensitive information in the resume is irreversibly encrypted, and read-only permissions are set for the recruitment team. Duplicate resumes are automatically deduplicated. The resume conversion rate of each recruitment channel is calculated weekly, and job descriptions are optimized based on changes in cluster centers.

[0035] This embodiment also allows for historical tracing of training data, effectiveness evaluation, and intelligent recommendation. For example, a training profile is created for each employee, linking course records, assessment results, and skill tags along a timeline, supporting batch queries by cluster tags. Training datasets are compared with performance datasets quarterly, and training effectiveness is verified through correlation analysis, triggering optimization processes for low-correlation courses. Based on the gap between an employee's current competency cluster and their target cluster, training courses that complement their skills are automatically recommended.

[0036] This embodiment also allows for periodic processing, attribution analysis, and long-term archiving of training data. For example, on the 10th of each month, attendance and KPI completion data for employees in each department are automatically collected and compared with performance scenario clusters to generate a performance warning list. This warning list is then linked to training datasets to analyze whether poor performance is due to skill deficiencies, and personalized improvement plans are generated. At the end of each year, historical performance data is backed up to cold storage, supporting multi-dimensional career development trend modeling.

[0037] From the above, it can be concluded that, on the one hand, this embodiment uses a Gaussian mixture model to initialize clustering based on human resources data and different scenario types, and uses the expectation maximization algorithm for iterative updates, which can accurately divide the data into three types of data sets: recruitment, training, and performance evaluation, thereby improving the accuracy and scientificity of data classification. On the other hand, this embodiment manages the classified data sets separately, and can formulate exclusive management strategies for different scenario requirements, avoiding the management difficulties caused by data mixing. This not only effectively improves the efficiency of human resources data management, but also provides more accurate and targeted data support for the company's human resources decision-making, helping companies optimize their human resources management processes.

[0038] This embodiment uses a Gaussian mixture model to implement scenario-based soft clustering and dynamic management of human resources data, which can avoid single classification, support multi-label attribution of cross-scenario data, and improve the adaptability of complex businesses. This embodiment is based on iteratively optimized clustering centers, which automatically captures changes in scenario characteristics and can enhance the pertinence of data analysis. This embodiment shortens data query response time through hierarchical storage and automated processing of data, and scenario-based management and control strengthen privacy protection and decision-making accuracy. The soft membership introduced in this embodiment supports scenario association analysis and provides multi-dimensional data support for talent training, performance improvement, etc.

[0039] In one embodiment of the present application, initializing the Gaussian mixture model based on human resources data and the number of scene types includes: Determine the initial mean vector based on human resource data; Determine the number of clusters of the Gaussian mixture model according to the number of scene types; Initialize the Gaussian mixture model based on the initial mean vector and the number of clusters.

[0040] In this embodiment, determining the initial mean vector based on human resources data specifically includes: Randomly select K data points from the human resources data and use the eigenvalues ​​of the K data points as the initial mean vector, or: The data features of each scenario type are determined based on prior knowledge of human resources data, and the initial mean vector is determined based on the data features of each scenario type.

[0041] In this embodiment, the initial mean vector refers to the center position of each Gaussian component in the Gaussian mixture model, representing the center of the data feature distribution of the corresponding scenario type. Each mean vector is a d-dimensional vector, where d is the number of data features. For example, the mean vector of the recruitment scenario may include feature values ​​such as high job matching, medium educational requirements, and low performance scores. The number of clusters refers to the number of Gaussian components in the Gaussian mixture model, which directly corresponds to the number of scenario types. For example, K=3 corresponds to the three scenarios of recruitment, training, and performance evaluation. The number of clusters determines how many business scenario categories the model divides the data into. Prior knowledge refers to the scene feature patterns summarized based on business experience or historical data. For example, recruitment data usually contains job keywords, candidate education and other features. Prior knowledge can be used to manually set a more reasonable initial mean vector to accelerate model convergence.

[0042] This embodiment can randomly select K data points from human resources data as the initial mean vector. Random sampling is used to make the initial points cover the data space as much as possible to avoid local optimality. For example, 3 candidate resumes are randomly selected from the recruitment data as the initial mean, representing the three categories of "high match", "medium match" and "low match". This embodiment can define scenario feature templates based on prior knowledge. For example, the recruitment scenario feature template may include a resume matching degree greater than or equal to 0.8, length of service greater than or equal to 3 years, and a current salary less than 10,000. The training scenario feature template may include a training participation rate greater than 90%, a performance score greater than 3.5, etc. This embodiment can convert each scenario feature template into a numerical vector and use business experience to accelerate convergence. It is particularly suitable for scenarios with complex data distribution or small sample size.

[0043] In this embodiment, the Gaussian mixture model is initialized based on the initial mean vector and the number of clusters, specifically including: Calculate the variance of each feature in the initial mean vector and initialize an initial diagonal covariance matrix for all clusters based on the variance of each feature; Predict the amount of human resources data corresponding to each scenario type based on the data source of human resources data; Determine the initial prior probability of all clusters based on the amount of data corresponding to all scenario types; the initial prior probability is the proportion of the data in each cluster in the human resources data; The Gaussian mixture model is initialized based on the initial mean vector, the number of clusters, the initial diagonal covariance matrix of all clusters, and the initial prior probabilities of all clusters.

[0044] In this embodiment, the diagonal covariance matrix is ​​used to describe the degree of dispersion of the data in each cluster in each feature dimension. The diagonal elements are the variances of each feature, and the off-diagonal elements are 0. Each element on the diagonal corresponds to the variance of a feature, such as age variance and performance score variance. The prior probability refers to the proportion of each cluster (scenario type) in the overall data, that is, the initial probability that a data point belongs to a certain cluster. For example, the recruitment scenario accounts for 40%, the training scenario accounts for 30%, and the performance scenario accounts for 30%, with corresponding prior probabilities of 0.4, 0.3, and 0.3, respectively. Predicting the amount of data corresponding to each scenario type based on the data source of human resources data refers to estimating the proportion of data volume of each scenario type based on the business system or process that generates the data. For example, the data generated by the recruitment system accounts for approximately 40% of the total data.

[0045] This embodiment provides a better initial state for GMM by reasonably estimating data distribution parameters such as covariance matrix and prior probability, accelerates convergence and avoids local optimum.

[0046] For example, this embodiment first calculates the diagonal covariance matrix: for each initial mean vector, the variance of each feature of the corresponding data subset is calculated, such as the matching variance = 0.16 and the educational background variance = 0.09. A diagonal matrix is ​​then constructed based on the extracted feature variances. For example, the recruitment scenario covariance matrix Σ1 = diag([0.16, 0.09, ..., 0.02]) is constructed, where each diagonal element corresponds to the variance of a feature.

[0047] This embodiment can calculate the data volume proportion of recruitment scenarios, training scenarios, and performance scenarios based on the number of historical data records generated by each business system in the past 12 months, such as 100,000 records generated by the recruitment system, 50,000 records generated by the training system, and 50,000 records generated by the performance system. Based on this historical experience data, the proportion of scenario data volume in the human resources data to be processed can be predicted as a prior probability.

[0048] This embodiment integrates parameters such as the mean vector, diagonal covariance matrix, and prior probability, and substitutes these parameters into the GMM as initialization model parameters to prepare for the EM algorithm iteration.

[0049] This embodiment introduces a reasonable initial covariance matrix and prior probabilities to reduce the number of EM algorithm iterations. The prior probabilities in this embodiment reflect the true distribution of business transactions, preventing clustering results from being biased towards scenarios with large data volumes. The diagonal covariance matrix in this embodiment assumes feature independence, which aligns with the characteristics of human resources data, simplifies the model, and maintains business interpretability.

[0050] In one embodiment of the present application, all clusters are iteratively updated based on the cluster soft membership of all data points and the expectation-maximization iterative algorithm until the cluster parameter change rate between two adjacent iterations meets the iterative convergence condition, including: Based on the cluster soft membership of all data points and the expectation maximization iterative algorithm, the initial mean vector, initial diagonal covariance matrix and initial prior probability of all clusters are updated respectively to obtain the intermediate mean vector, intermediate diagonal covariance matrix and intermediate prior probability respectively; Calculate the Euclidean distance between the initial mean vector and the intermediate mean vector, calculate the average change of the matrix elements of the initial diagonal covariance matrix and the intermediate diagonal covariance matrix, and calculate the rate of change of the initial prior probability and the intermediate prior probability; The Euclidean distance, the average change of matrix elements and the change rate are weighted summed to obtain the clustering parameter change rate; When the clustering parameter change rate is less than or equal to a first threshold, determining that the iterative convergence condition is met; When the clustering parameter change rate is greater than the first threshold, the intermediate mean vector, intermediate diagonal covariance matrix and intermediate prior probability of all clusters are updated respectively, and the iterative convergence conditions of the updated intermediate mean vector, intermediate diagonal covariance matrix and intermediate prior probability are judged.

[0051] In this embodiment, the Euclidean distance is used to measure the spatial distance between the initial mean and the intermediate mean vectors, and the Euclidean distance is used to evaluate the change in the mean vector before and after iteration. The smaller the distance, the more stable the cluster center. The average change in matrix elements refers to the average fluctuation range of each feature variance in the diagonal covariance matrix, which is used to reflect the stability of the data distribution within the cluster, such as judging whether the variance of the KPI completion rate in the performance scenario converges. The rate of change of the prior probability refers to the difference in the proportion of cluster data before and after iteration, which is used to measure the dynamic adjustment range of the scenario data volume. The clustering parameter change rate is an overall convergence indicator obtained by weighted summation of the changes in the comprehensive mean vector, covariance matrix, and prior probability. When the clustering parameter change rate is lower than the preset threshold, the model parameters are considered stable and the iteration is terminated.

[0052] This example uses soft membership to calculate the contribution of each data point to the clustering parameters. Data points with higher soft memberships have a greater impact on the mean vector update. For example, if an employee's performance data has a membership of 0.8 in the "low-performance cluster," their eigenvalues ​​will be given a higher weight when updating the cluster's mean.

[0053] Considering the limitations of single metrics: considering only the mean distance ignores changes in covariance, and diverging feature variances can lead to ambiguous clustering. Considering only the prior probability fails to reflect structural changes in the data distribution. Therefore, this embodiment combines changes in the mean, covariance, and prior probability to more comprehensively determine whether the model is stable.

[0054] Exemplarily, the EM algorithm iteration includes: Step E: Use the current scenario parameters to calculate the probability that each piece of data belongs to each business scenario of recruitment, training, and performance.

[0055] Step M: Update the typical feature values ​​of the scene, the data fluctuation range, and the proportion of scene data based on the probability.

[0056] Weight setting: When the calculation focuses on the stability of the scene center, this embodiment can increase the weight of the typical eigenvalue change; when data fluctuations need to be controlled, the weight of the fluctuation range change can be increased.

[0057] Threshold adjustment: Initially set a loose threshold (such as 10%) for rapid iteration, and then tighten it to 0.1% for fine-tuning to avoid invalid cycles.

[0058] Exception handling: This embodiment can detect parameter oscillation or iteration stagnation, triggering manual review or forcibly ending the iteration.

[0059] This embodiment combines the mean vector, covariance matrix, and prior probability changes to avoid misjudgments based on a single indicator, improving model stability and reliability. This embodiment utilizes soft membership weighted parameter updates to adapt to data distribution characteristics, accelerating convergence. This embodiment allows for flexible adjustment of weights and thresholds to adapt to dynamic changes in human resources data and accurately identify abnormal data.

[0060] In one embodiment of the present application, data management is performed separately for the recruitment dataset, the training dataset, and the performance evaluation dataset, including: For any of the recruitment dataset, training dataset, and performance evaluation dataset: The data set is divided into data types to obtain a text data subset, an image data subset, and a voice data subset; Construct semantic vector representation based on text data subset, perform feature extraction on image data subset and gesture and speech data subset respectively to obtain image features and speech features; The semantic vector representation, image features and speech features are input into the multimodal data fusion model to obtain data analysis results, and data management is performed based on the data analysis results.

[0061] In this embodiment, data type division refers to splitting a single data set into three subsets: text, image, and voice according to the data form. For example, a recruitment data set includes resume text, candidate photos, and interview recordings. A training data set includes course description text, operation video screenshots, and lecturer audio. A performance evaluation data set includes comment text, 360-degree evaluation charts, and job report recordings. Semantic vector representation refers to converting text data into computer-understandable numerical vectors to capture semantic information. To construct a semantic vector representation, word embedding methods such as Word2Vec can be used to map words to a low-dimensional vector space, or a pre-trained language model such as BERT can be used to generate context-related sentence vectors.

[0062] This embodiment can extract visual image features such as facial expressions and document structure through a convolutional neural network. This embodiment can also extract acoustic features such as voice color, speaking speed, and pitch through Mel-frequency cepstral coefficients or Wav2Vec. The multimodal data fusion model can be used to integrate text, image, and speech features and output a unified analysis result. The fusion strategy of the multimodal data fusion model can be early fusion: inputting the neural network after splicing vectors at the feature layer, or late fusion: modeling each modality separately and then integrating the decision.

[0063] This example uses multimodal fusion to mine complementary information between data sets, improving the accuracy of human resources analysis. For any of the recruitment, training, and performance evaluation datasets, this example first performs data preprocessing. For example, text cleaning involves removing HTML tags and standardizing punctuation in resumes. Image normalization involves resizing photos to 224×224 pixels and normalizing pixel values. Speech framing involves cutting recordings into 25ms frames and extracting MFCC features.

[0064] For text data, this example uses BERT to generate 768-dimensional semantic vectors. For image data, this example uses ResNet50 to extract 2048-dimensional visual features. For speech data, this example uses Wav2Vec to generate 512-dimensional acoustic vectors. This example concatenates the text (768), image (2048), and speech (512) vectors into a 3328-dimensional vector, assigns weights to different modalities, and finally outputs the analysis results through a fully connected layer.

[0065] Exemplarily, the model construction of the multimodal data fusion model may include: Input layer: receives three types of feature vectors.

[0066] Fusion layer: Calculate weights through the attention mechanism.

[0067] Output layer: regression prediction (such as interview score) or classification (such as job matching type).

[0068] This embodiment fuses multimodal data—text, images, and voice—to mine complementary information and improve analysis accuracy. This embodiment can detect cross-modal inconsistencies and pinpoint data anomalies or candidate potential. The automated multimodal analysis of this embodiment can shorten manual processing time, improve efficiency, and save costs.

[0069] Corresponding to the human resources data management method based on big data in the above embodiment, Figure 2 This is a structural block diagram of a human resources data management device based on big data provided by an embodiment of the present application. For ease of explanation, only the parts related to the embodiment of the present application are shown. Figure 2The human resources data management device 20 based on big data includes: a model initialization module 21, a clustering update module 22 and a classification management module 23.

[0070] The model initialization module 21 is used to initialize the Gaussian mixture model based on the human resources data and the number of scenario types, and input the human resources data into the initialized Gaussian mixture model to obtain multiple clusters; the scenario types include recruitment scenarios, training scenarios, and performance evaluation scenarios, and each scenario type corresponds to a cluster; The cluster updating module 22 is configured to calculate the cluster soft membership of each data point in the human resources data based on the initialized Gaussian mixture model; the cluster soft membership refers to the probability that the data point belongs to each cluster; and iteratively update all clusters based on the cluster soft membership of all data points and the expectation-maximization algorithm until the rate of change of cluster parameters between two consecutive iterations meets the iterative convergence condition. The classification management module 23 is used to determine the recruitment data set, the training data set and the performance evaluation data set based on the target cluster obtained in the last update; and perform data management on the recruitment data set, the training data set and the performance evaluation data set respectively.

[0071] In one embodiment of the present application, the model initialization module 21 is specifically configured to determine an initial mean vector based on human resources data; Determine the number of clusters of the Gaussian mixture model according to the number of scene types; Initialize the Gaussian mixture model based on the initial mean vector and the number of clusters.

[0072] In one embodiment of the present application, the model initialization module 21 is further configured to randomly select K data points from the human resources data and use the eigenvalues ​​of the K data points as the initial mean vector, or: The data features of each scenario type are determined based on prior knowledge of human resources data, and the initial mean vector is determined based on the data features of each scenario type.

[0073] In one embodiment of the present application, the model initialization module 21 is further configured to calculate the variance of each feature in the initial mean vector, and initialize an initial diagonal covariance matrix for all clusters based on the variance of each feature; Predict the amount of human resources data corresponding to each scenario type based on the data source of human resources data; Determine the initial prior probability of all clusters based on the amount of data corresponding to all scenario types; the initial prior probability is the proportion of the data in each cluster in the human resources data; The Gaussian mixture model is initialized based on the initial mean vector, the number of clusters, the initial diagonal covariance matrix of all clusters, and the initial prior probabilities of all clusters.

[0074] In one embodiment of the present application, the cluster updating module 22 is specifically configured to calculate the cluster soft membership of the data point using the first formula based on the initialized Gaussian mixture model; The first formula is: ; in, ; ; in, For data points The probability of belonging to cluster k, that is, the data point The clustering soft membership of is the prior probability of cluster k, is the probability density function of the Gaussian distribution, is the i-th data point in the human resources data, each data point is a vector with multiple features, is the mean vector of cluster k, is the diagonal covariance matrix of cluster k, j is the index of the cluster, K is the total number of clusters, and d is the number of data points The characteristic dimension of is the diagonal covariance matrix The determinant of is an exponential function, is a vector The transpose of is the diagonal covariance matrix The inverse matrix of .

[0075] In one embodiment of the present application, the cluster updating module 22 is further configured to update the initial mean vectors, initial diagonal covariance matrices, and initial prior probabilities of all clusters based on the cluster soft membership of all data points and the expectation-maximization iterative algorithm, to obtain intermediate mean vectors, intermediate diagonal covariance matrices, and intermediate prior probabilities, respectively; Calculate the Euclidean distance between the initial mean vector and the intermediate mean vector, calculate the average change of the matrix elements of the initial diagonal covariance matrix and the intermediate diagonal covariance matrix, and calculate the rate of change of the initial prior probability and the intermediate prior probability; The Euclidean distance, the average change of matrix elements and the change rate are weighted summed to obtain the clustering parameter change rate; When the clustering parameter change rate is less than or equal to a first threshold, determining that the iterative convergence condition is met; When the clustering parameter change rate is greater than the first threshold, the intermediate mean vector, intermediate diagonal covariance matrix and intermediate prior probability of all clusters are updated respectively, and the iterative convergence conditions of the updated intermediate mean vector, intermediate diagonal covariance matrix and intermediate prior probability are judged.

[0076] In one embodiment of the present application, the classification management module 23 is specifically configured to: for any one of the recruitment dataset, the training dataset, and the performance evaluation dataset: The data set is divided into data types to obtain a text data subset, an image data subset, and a voice data subset; Construct semantic vector representation based on text data subset, perform feature extraction on image data subset and gesture and speech data subset respectively to obtain image features and speech features; The semantic vector representation, image features and speech features are input into the multimodal data fusion model to obtain data analysis results, and data management is performed based on the data analysis results.

[0077] See also Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided in one embodiment of the present application. Figure 3 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of the modules in the above-mentioned device embodiments, such as Figure 2 The functions of the model initialization module 21, the clustering update module 22 and the classification management module 23 are shown.

[0078] It should be understood that in the embodiment of the present application, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0079] The input device 302 may include a touchpad, a fingerprint collection sensor (for collecting user fingerprint information and fingerprint direction information), a microphone, etc. The output device 303 may include a display (LCD, etc.), a speaker, etc.

[0080] The memory 304 may include a read-only memory and a random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about model parameters.

[0081] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiment of the present application can execute the implementation method described in the embodiment of the human resources data management method based on big data provided in the embodiment of the present application, and can also execute the implementation method of the electronic device 300 described in the embodiment of the present application, which will not be repeated here.

[0082] In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, all or part of the process of the method in the above embodiment is implemented. The computer program can also be used to instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of each of the above method embodiments are implemented. The computer program includes computer program code, which can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.

[0083] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, the computer-readable storage medium can include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or is about to be output.

[0084] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0085] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0086] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or can be an electrical, mechanical or other form of connection.

[0087] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0088] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0089] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A human resources data management method based on big data, characterized in that: include: Initializing a Gaussian mixture model based on the human resources data and the number of scenario types, and inputting the human resources data into the initialized Gaussian mixture model to obtain multiple clusters; the scenario types include recruitment scenarios, training scenarios, and performance evaluation scenarios, and each scenario type corresponds to a cluster; For each data point in the human resources data, calculating the cluster soft membership of the data point based on the initialized Gaussian mixture model; The cluster soft membership refers to the probability that the data point belongs to each cluster; All clusters are iteratively updated based on the cluster soft membership of all data points and the expectation maximization iterative algorithm until the cluster parameter change rate between two adjacent iterations meets the iterative convergence condition; A recruitment dataset, a training dataset, and a performance evaluation dataset are determined based on the target clustering obtained from the last update; and data management is performed on the recruitment dataset, the training dataset, and the performance evaluation dataset respectively.

2. The human resources data management method based on big data according to claim 1, characterized in that: Initializing the Gaussian mixture model based on the human resources data and the number of scenario types includes: Determine the initial mean vector based on human resource data; Determine the number of clusters of the Gaussian mixture model according to the number of scene types; A Gaussian mixture model is initialized based on the initial mean vector and the number of clusters.

3. The human resources data management method based on big data according to claim 2, characterized in that: The determining of the initial mean vector based on the human resources data includes: Randomly select K data points from the human resources data and use the eigenvalues ​​of the K data points as the initial mean vector, or: The data features of each scenario type are determined based on prior knowledge of human resources data, and the initial mean vector is determined based on the data features of each scenario type.

4. The human resources data management method based on big data according to claim 2, characterized in that: Initializing the Gaussian mixture model based on the initial mean vector and the number of clusters includes: Calculating the variance of each feature in the initial mean vector, and initializing an initial diagonal covariance matrix for all clusters based on the variance of each feature; Predicting the amount of human resources data corresponding to each scenario type based on the data source of the human resources data; Determining initial prior probabilities of all clusters based on the amount of data corresponding to all scenario types; the initial prior probabilities are the proportion of data in each cluster in the human resources data; A Gaussian mixture model is initialized based on the initial mean vector, the number of clusters, the initial diagonal covariance matrix of all clusters, and the initial prior probabilities of all clusters.

5. The human resources data management method based on big data according to claim 1, characterized in that: The calculating the cluster soft membership of the data point based on the initialized Gaussian mixture model includes: Based on the initialized Gaussian mixture model, the cluster soft membership of the data point is calculated using the first formula; The first formula is: ； in, ; ； in, For data points The probability of belonging to cluster k, that is, the data point The clustering soft membership of is the prior probability of cluster k, is the probability density function of the Gaussian distribution, is the i-th data point in the human resources data, each data point is a vector with multiple features, is the mean vector of cluster k, is the diagonal covariance matrix of cluster k, j is the index of the cluster, K is the total number of clusters, and d is the number of data points The characteristic dimension of is the diagonal covariance matrix The determinant of is an exponential function, is a vector The transpose of is the diagonal covariance matrix The inverse matrix of .

6. The human resources data management method based on big data according to claim 1, characterized in that: The clustering soft membership based on all data points and the expectation maximization iterative algorithm iteratively updates all clusters until the clustering parameter change rate between two adjacent iterations meets the iterative convergence condition, including: Based on the cluster soft membership of all data points and the expectation maximization iterative algorithm, the initial mean vector, initial diagonal covariance matrix and initial prior probability of all clusters are updated respectively to obtain the intermediate mean vector, intermediate diagonal covariance matrix and intermediate prior probability respectively; Calculate the Euclidean distance between the initial mean vector and the intermediate mean vector, calculate the average change of the matrix elements of the initial diagonal covariance matrix and the intermediate diagonal covariance matrix, and calculate the rate of change of the initial prior probability and the intermediate prior probability; Performing a weighted summation on the Euclidean distance, the average change of the matrix elements, and the change rate to obtain a clustering parameter change rate; When the clustering parameter change rate is less than or equal to a first threshold, determining that an iterative convergence condition is satisfied; When the clustering parameter change rate is greater than a first threshold, the intermediate mean vectors, intermediate diagonal covariance matrices and intermediate prior probabilities of all clusters are updated respectively, and the iterative convergence conditions of the updated intermediate mean vectors, intermediate diagonal covariance matrices and intermediate prior probabilities are judged.

7. The human resources data management method based on big data according to claim 1, characterized in that: The data management of the recruitment dataset, training dataset, and performance evaluation dataset respectively includes: For any one of the recruitment dataset, the training dataset, and the performance evaluation dataset: The data set is divided into data types to obtain a text data subset, an image data subset, and a voice data subset; Constructing a semantic vector representation based on the text data subset, and performing feature extraction on the image data subset and the gesture and speech data subset to obtain image features and speech features; The semantic vector representation, the image features, and the speech features are input into a multimodal data fusion model to obtain a data analysis result, and data management is performed based on the data analysis result.

8. A human resources data management device based on big data, characterized in that: include: A model initialization module is used to initialize the Gaussian mixture model based on the human resources data and the number of scenario types, and input the human resources data into the initialized Gaussian mixture model to obtain multiple clusters; the scenario types include recruitment scenarios, training scenarios, and performance evaluation scenarios, and each scenario type corresponds to a cluster; A clustering update module, configured to calculate, for each data point in the human resources data, a cluster soft membership of the data point based on the initialized Gaussian mixture model; The cluster soft membership refers to the probability that the data point belongs to each cluster; All clusters are iteratively updated based on the cluster soft membership of all data points and the expectation maximization iterative algorithm until the cluster parameter change rate between two adjacent iterations meets the iterative convergence condition; The classification management module is used to determine the recruitment data set, training data set and performance evaluation data set based on the target clustering obtained in the last update; and perform data management on the recruitment data set, training data set and performance evaluation data set respectively.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images