Intelligent talent management method and system for chain enterprises

Abstract

The invention relates to the technical field of data processing, and discloses an intelligent talent management method and system for chain enterprises, and the method comprises the steps: carrying out the multi-dimensional data collection and preprocessing of an internal system and an external data source of an enterprise, and obtaining a multi-dimensional employee data set; performing talent network construction, and fusing personal attributes and network attributes to obtain a multi-level talent distribution network; extracting multi-dimensional evaluation indexes according to the multi-dimensional employee data set and carrying out weight distribution to obtain an evaluation index system; performing feature index calculation on the multi-dimensional employee data set according to the evaluation index system to obtain a feature evaluation index set; and performing graph convolutional neural network and multi-scale aggregation analysis on the feature evaluation index set and the multi-level talent distribution network to obtain a plurality of talent recognition results, and matching an intelligent management strategy of each talent recognition result, so that intelligent talent management of chain enterprises is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an intelligent talent management method and system for a chain enterprise. Background Art

[0002] Chain enterprises are facing increasingly fierce market competition and huge challenges in talent management. Traditional talent management methods can no longer meet the needs of chain enterprises for rapid expansion and efficient operation. Enterprises are in urgent need of a more intelligent and precise talent management method. However, there are still many problems in talent management in current chain enterprises, such as scattered data, information islands, a single evaluation system, and inaccurate talent identification.

[0003] These problems make it difficult for chain enterprises to fully and accurately understand the capabilities and potential of their employees, and to formulate effective talent development strategies. In addition, traditional talent management methods often ignore the relationship network and teamwork factors among employees, and fail to give full play to the overall value of talents. In this context, it has become an urgent need for chain enterprises to use advanced technologies such as artificial intelligence and big data analysis to build an intelligent talent management method. Summary of the invention

[0004] The present invention provides a talent intelligent management method and system for a chain enterprise, and the present invention realizes the talent intelligent management for the chain enterprise.

[0005] In a first aspect, the present invention provides a talent intelligent management method for a chain enterprise, the talent intelligent management method for a chain enterprise comprising: Collect and preprocess multi-dimensional data from the enterprise's internal systems and external data sources to obtain a multi-dimensional employee data set; Constructing a talent network for the multi-dimensional employee data set, integrating individual attributes and network attributes, and obtaining a multi-level talent distribution network; Extracting multidimensional evaluation indicators based on the multidimensional employee data set, and assigning weights to the multidimensional evaluation indicators through a hierarchical analysis method to obtain an evaluation indicator system; Calculating characteristic indicators of the multi-dimensional employee data set according to the evaluation indicator system to obtain a characteristic evaluation indicator set; A graph convolutional neural network and multi-scale aggregation analysis are performed on the feature evaluation index set and the multi-level talent distribution network to obtain multiple talent identification results, and an intelligent management strategy is matched to each talent identification result.

[0006] In a second aspect, the present invention provides a chain enterprise's intelligent talent management system, the chain enterprise's intelligent talent management system comprising: The acquisition module is used to collect and preprocess multi-dimensional data from the enterprise's internal system and external data sources to obtain a multi-dimensional employee data set; A construction module is used to construct a talent network for the multi-dimensional employee data set, integrating individual attributes and network attributes to obtain a multi-level talent distribution network; An allocation module, used for extracting multi-dimensional evaluation indicators according to the multi-dimensional employee data set, and performing weight allocation on the multi-dimensional evaluation indicators through a hierarchical analysis method to obtain an evaluation indicator system; A calculation module, used to calculate characteristic indicators of the multi-dimensional employee data set according to the evaluation indicator system to obtain a characteristic evaluation indicator set; The identification module is used to perform graph convolutional neural network and multi-scale aggregation analysis on the feature evaluation index set and the multi-level talent distribution network to obtain multiple talent identification results and match the intelligent management strategy for each talent identification result.

[0007] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned intelligent talent management method for chain enterprises.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned intelligent talent management method for chain enterprises.

[0009] In the technical solution provided by the present invention, by integrating and preprocessing multiple data sources inside and outside the enterprise, employee information can be obtained comprehensively and accurately, overcoming the problems of data dispersion and information islands in traditional methods, integrating personal attributes and network attributes, and constructing a multi-level talent distribution network, which can not only reflect the personal characteristics of employees, but also show the relationship and team structure between employees. The evaluation indicators are scientifically weighted by the hierarchical analysis method, and a comprehensive and objective evaluation system is established, avoiding the shortcomings of single evaluation criteria and strong subjectivity in traditional methods. Quantitative and qualitative indicators are standardized and numerically processed, and key features are extracted by dimensionality reduction and cluster analysis, which can more accurately characterize the ability and potential of employees. Using deep learning technology, feature learning is performed while considering the network structure, which can better capture the complex relationship between employees and team dynamics, and improve the accuracy of talent identification. Through the aggregation of features at the three levels of team, department and company, multi-scale talent identification is realized, and the performance and potential of employees at different organizational levels can be comprehensively evaluated. According to the identification results, the corresponding management strategy is automatically matched, and the personalization and intelligence of talent management are realized, which helps to improve management efficiency and effectiveness. This method has good scalability and can adapt to chain enterprises of different sizes and types, and can be adjusted and optimized according to the specific needs of the enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 A schematic diagram of the steps of the intelligent talent management method for a chain enterprise according to an embodiment of the present invention; Figure 2 It is a structural diagram of the intelligent talent management system of a chain enterprise in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The embodiment of the present invention provides a method and system for intelligent management of talents of a chain enterprise. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 An embodiment of the chain enterprise's talent intelligent management method in the embodiment of the present invention includes: Step S1: collect and preprocess multi-dimensional data from the enterprise internal system and external data sources to obtain a multi-dimensional employee data set; It is understandable that the execution subject of the present invention may be a chain enterprise's intelligent talent management system, or a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as an execution subject as an example.

[0014] Specifically, data is extracted from the enterprise ERP, CRM and OA systems to obtain employees' basic information, work history and performance records. Based on the basic information of employees, log analysis is performed on the enterprise instant messaging system and email system to obtain employees' communication network data. At the same time, text mining is performed on employees' work history to extract each employee's skill tag set. Based on the employee skill tag set, keyword matching is performed on the industry talent market database to obtain the supply and demand of external talents. Data cleaning and standardization are performed on the basic information, work history, performance records and employee communication network data extracted from different systems to obtain structured employee data. The data cleaning process includes steps such as removing duplicate data, correcting erroneous data, and filling missing data to ensure data consistency and accuracy. Standardization processing includes converting the data into a unified format and unit to make it conform to a unified standard, which is convenient for subsequent data analysis and processing. Desensitization processing is performed on structured employee data. The desensitization process includes removing or blurring sensitive information involving personal privacy, such as name, contact information, ID number, etc., to ensure that employees' personal privacy information will not be leaked during data analysis and use. Conduct data quality assessment on desensitized employee data, including data integrity, accuracy, consistency, and timeliness. Through data quality assessment, identify quality problems in the data and generate a data quality report. According to the data quality report, interpolate and process outliers for low-quality data. Interpolation refers to the reasonable filling of missing data, and outlier processing refers to the identification and processing of outliers in the data to ensure the quality and reliability of the data. Through the processing steps, the corrected employee data is obtained. The corrected employee data is integrated with the external talent supply and demand situation, including associating, matching, and integrating data from different sources to form a unified, multi-dimensional employee data set.

[0015] Step S2: construct a talent network for the multi-dimensional employee data set, integrate personal attributes and network attributes, and obtain a multi-level talent distribution network; Specifically, feature extraction is performed on the multi-dimensional employee data set to obtain each employee's personal attribute vector and initial network relationship matrix. Based on the initial network relationship matrix, a weighted graph is constructed for the working relationship between employees to generate an initial talent network graph. In this process, the relationship between each employee in the network is represented in a weighted manner, reflecting the intensity of their interaction at work. The initial talent network graph is encoded by the graph neural network algorithm to obtain a low-dimensional representation of the employee node. The low-dimensional representation can better capture the position and role of employees in the network and simplify the complexity of the network. Based on the low-dimensional representation, the employee's personal attribute vector is feature fused to obtain the fused employee feature vector. The multi-head attention mechanism is applied to calculate the fused employee feature vector to obtain the similarity and correlation strength between employees and update the relationship matrix. Based on the updated relationship matrix, the initial talent network graph is updated with edge weights and structural optimization to generate an optimized talent network graph, ensuring that the network graph not only reflects the actual working relationship between employees, but also integrates their personal characteristics and the strength of their mutual correlation. Cluster analysis is performed on the optimized talent network graph to identify employee community structures at different levels. The community structure shows the distribution of employees in the network and the characteristics of group behavior. According to the employee community structure, each community is analyzed for centrality and influence, key nodes and core members in the community are identified, and community feature descriptions and key node sets are generated. Through description and collection, communities at different levels are hierarchically aggregated to build a multi-level network structure including individual, team and organizational levels. Dynamic evolution analysis is performed on the multi-level network structure to observe and predict the changing trend of the network structure over time, and a multi-level talent distribution network is obtained.

[0016] Step S3, extracting multidimensional evaluation indicators based on the multidimensional employee data set, and assigning weights to the multidimensional evaluation indicators through the hierarchical analysis method to obtain an evaluation indicator system; Specifically, data analysis was performed on the multi-dimensional employee data set to extract an initial set of indicators including performance, ability and quality, potential development and team contribution. In order to ensure that the initial set of indicators is highly relevant to the company's strategic goals and job requirements, text analysis was performed on the company's strategic goals and job requirements to extract a set of keywords related to the company's strategy. Semantic similarity was calculated between the keyword set and the initial set of indicators to build a preliminary evaluation system. Based on the preliminary evaluation system, principal component analysis was performed on the historical talent evaluation data to calculate the explained variance ratio of each evaluation indicator. By threshold screening of the explained variance ratio, multi-dimensional evaluation indicators were determined to reflect the comprehensive quality and development potential of employees. Based on the multi-dimensional evaluation indicators, a judgment matrix of the hierarchical analysis method was constructed to determine the relative importance of each indicator. By calculating the eigenvalues ​​of the judgment matrix, especially the maximum eigenvalue and its corresponding eigenvector, the relative weights of the indicators were obtained, reflecting the importance of each evaluation indicator in the overall evaluation system. A consistency test was performed on the relative importance matrix, the consistency ratio of the hierarchical analysis method was calculated, and the rationality of the judgment matrix was ensured to ensure the logic and reliability of the weight distribution. On this basis, a sensitivity analysis of the indicator weights is conducted, and the influence of the weight changes on the evaluation results is examined by adjusting the element values ​​in the judgment matrix to obtain the stability evaluation results. According to the stability evaluation results, the weights of the multi-dimensional evaluation indicators are appropriately adjusted and normalized to ensure that the weight of each indicator is between 0 and 1, and the sum is 1, so as to obtain the final evaluation indicator system.

[0017] Step S4: Calculate characteristic indicators of the multi-dimensional employee data set according to the evaluation indicator system to obtain a characteristic evaluation indicator set; Specifically, the data type analysis is performed on each indicator in the evaluation indicator system, and the indicators are divided into a quantitative indicator set and a qualitative indicator set. The quantitative indicator set includes numerical data, such as performance scores and years of work experience, while the qualitative indicator set includes text data, such as employee evaluations and feedback. According to the quantitative indicator set, the numerical data in the multi-dimensional employee data set is normalized, and the data of different dimensions are converted to the same dimension through a standardization method to obtain a standardized quantitative indicator value. At the same time, the text data in the qualitative indicator set is processed, and a word segmentation operation is performed to divide the text into several meaningful words, and then part-of-speech tagging is performed to determine the part of speech of each word. Finally, entity recognition is performed to identify key entities in the text, such as names, places, and organization names. After this series of processing, structured text features are obtained. According to the structured text features, the text is converted into a vector representation, and each word is converted into a numerical vector through natural language processing technology to obtain a numerical representation of the qualitative indicator. The standardized quantitative indicator value and the numerical representation of the qualitative indicator are feature fused to generate an initial feature vector set. In order to reduce the data dimension and computational complexity, the initial feature vector set is subjected to dimensionality reduction processing. Commonly used methods include principal component analysis and linear discriminant analysis. After dimensionality reduction processing, the reduced feature matrix is ​​obtained, and the high-dimensional data is visualized and represented as a two-dimensional scatter plot for easy observation and analysis. According to the two-dimensional scatter plot representation, data point density clustering analysis is performed, and similar data points are grouped by density clustering algorithms (such as DBSCAN) to obtain the preliminary grouping results of employees. In order to identify the key indicators that best represent the characteristics of employees, the feature importance scores are calculated for the preliminary grouping results and the initial feature vector set. By evaluating the contribution of each feature in the classification task, the key feature subset is identified. According to the key feature subset, the initial feature vector set is reconstructed and normalized, and the key features are recombined and standardized to ensure that the eigenvalues ​​are in the same dimension, and the final feature evaluation indicator set is obtained.

[0018] Step S5: Perform graph convolutional neural network and multi-scale aggregation analysis on the feature evaluation index set and the multi-level talent distribution network to obtain multiple talent identification results, and match the intelligent management strategy for each talent identification result.

[0019] Specifically, the feature evaluation index set is standardized to obtain a node feature matrix to ensure that feature data of different dimensions are compared on the same scale. The initial adjacency matrix is ​​constructed according to the multi-level talent distribution network, and the degree matrix is ​​calculated for the initial adjacency matrix. The normalized adjacency matrix is ​​obtained by normalizing the adjacency matrix to ensure that the connection relationship between different nodes in the network diagram is reasonably represented. The node feature matrix and the normalized adjacency matrix are input into the first layer of graph convolution layer, and the feature calculation is performed through matrix multiplication operation and ReLU activation function to obtain the first hidden layer representation. The ReLU activation function can introduce nonlinear characteristics, so that the model can learn more complex features. In order to prevent the model from overfitting, the first hidden layer representation is dropout, and 50% of the neurons are randomly set to zero to obtain a sparse hidden layer representation, which helps to improve the generalization ability of the model. The sparse hidden layer representation is input into the second layer of graph convolution layer, and the feature calculation is performed again through matrix multiplication operation and softmax activation function to obtain the output layer representation. The softmax activation function can convert the output into a probability distribution for classification tasks. Perform talent category labeling operations on each node to preliminarily determine the classification label of each employee. In order to analyze the talent distribution at a higher level, perform average pooling operations on node features at the team level based on the preliminary talent category labels to obtain the team-level talent distribution vector. Perform department-level maximum pooling operations on the team-level talent distribution vector to obtain the department-level talent distribution vector, and input the department-level talent distribution vector into the fully connected layer. Perform feature calculations through the weight matrix and sigmoid activation function to obtain the company-level talent distribution vector. The sigmoid activation function can compress the output value between 0 and 1, making it suitable as a probability value. Perform connection operations on the team-level talent distribution vector, the department-level talent distribution vector, and the company-level talent distribution vector to integrate talent information at different levels and obtain multiple talent identification results. Divide multiple talent identification results into K talent types, and match the corresponding intelligent management strategy for each talent type through the preset strategy mapping table. According to the characteristics and needs of different talents, formulate corresponding management measures and development plans to achieve refined and intelligent management of talents.

[0020] Generate a random mask matrix for the first hidden layer representation to get the dropout mask, then perform element-wise multiplication of the first hidden layer representation and the dropout mask, randomly set 50% of the neurons to zero, and generate a sparse hidden layer representation. Perform matrix multiplication on the sparse hidden layer representation and the normalized adjacency matrix to get the aggregated node representation. This operation aims to fuse the information of adjacent nodes so that the representation of each node contains not only its own features but also the features of its neighboring nodes. Perform matrix multiplication on the aggregated node representation and the weight matrix of the second graph convolution layer to generate the original output. Linear transformation is used to further explore the complex relationship between nodes. Batch normalization is performed on the original output. By calculating the mean and variance, each batch of data is scaled and translated to get the normalized output. Batch normalization can speed up the training process and help stabilize the performance of the model. The normalized output is input into the LeakyReLU activation function for nonlinear transformation to get the activated feature representation. Unlike the traditional ReLU activation function, LeakyReLU will also have a small output value when the input is less than zero to avoid the death of neurons. Perform global maximum pooling on the activated feature representation, extract the maximum value of each feature channel, and obtain the pooled feature vector. Global maximum pooling can effectively retain the significant features of each feature channel, reduce the data dimension and reduce the risk of overfitting. Input the pooled feature vector into the fully connected layer, perform linear transformation through the weight matrix, and generate the target feature vector. The function of the fully connected layer is to map the high-dimensional pooled feature vector to the target feature space. The probability distribution of each category of the target feature vector is calculated by the softmax function to obtain the probability output of each category. The softmax function can convert the target feature vector into a probability distribution, so that the output result can be interpreted as the probability value of each category. According to the probability output, the argmax function is used to select the category index with the highest probability, determine the final classification result, and obtain the output layer representation.

[0021] In the embodiment of the present invention, by integrating and preprocessing multiple data sources inside and outside the enterprise, employee information can be obtained comprehensively and accurately, overcoming the problems of data dispersion and information islands in traditional methods, integrating personal attributes and network attributes, and constructing a multi-level talent distribution network, which can not only reflect the personal characteristics of employees, but also show the relationship and team structure between employees. The evaluation indicators are scientifically weighted by the hierarchical analysis method, and a comprehensive and objective evaluation system is established, avoiding the shortcomings of single evaluation criteria and strong subjectivity in traditional methods. Quantitative and qualitative indicators are standardized and numerically processed, and key features are extracted by dimensionality reduction and cluster analysis, which can more accurately characterize the ability and potential of employees. Using deep learning technology, feature learning is performed while considering the network structure, which can better capture the complex relationship between employees and team dynamics, and improve the accuracy of talent identification. Through the aggregation of features at the three levels of team, department and company, multi-scale talent identification is realized, and the performance and potential of employees at different organizational levels can be comprehensively evaluated. According to the identification results, the corresponding management strategy is automatically matched, and the personalization and intelligence of talent management are realized, which helps to improve management efficiency and effectiveness. This method has good scalability and can adapt to chain enterprises of different sizes and types, and can be adjusted and optimized according to the specific needs of the enterprises.

[0022] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Extract data from the enterprise ERP, CRM and OA systems to obtain basic employee information, work history and performance records; Based on the basic information of employees, we analyze the logs of the enterprise instant messaging system and email system to obtain the employee communication network data, and conduct text mining on the work history to obtain the employee skill tag set; Based on the employee skill tag set, keyword matching is performed on the industry talent market database to obtain the external talent supply and demand situation; Clean and standardize employee basic information, work history, performance records, and employee communication network data to obtain structured employee data; Based on the structured employee data, employees’ personal privacy information is desensitized to obtain desensitized employee data, and the desensitized employee data is evaluated for data quality to obtain a data quality report; According to the data quality report, low-quality data is interpolated and outliers are processed to obtain corrected employee data, and the corrected employee data is fused with external talent supply and demand information to obtain a multi-dimensional employee data set.

[0023] Specifically, through data interfaces or customized data extraction scripts, basic information of employees is obtained from the enterprise's ERP system, including basic information such as name, gender, date of birth, contact information, and position; employee work history information is extracted from the CRM system to record detailed career paths such as employees' various positions in the company, project participation, and promotion records; employee performance records are obtained from the OA system, including monthly, quarterly, and annual performance evaluation results and related comments and scores. These data constitute the initial data set of employee basic information, work history, and performance records. Based on the basic information of employees, log analysis is performed on the enterprise's instant messaging system and email system. Chat records in the instant messaging system and email logs in the email system are extracted to reflect the communication network and interaction frequency of employees at work. Through natural language processing technology, these logs are segmented, part-of-speech tagged, and semantically analyzed to identify the communication patterns and cooperative relationships between employees and obtain employee communication network data. At the same time, text mining is performed on employees' work resumes. Using technologies such as keyword extraction and topic modeling, the skill tag set of employees is extracted from the job description. For example, an employee has used a certain programming language in multiple projects, managed a team of so many people, etc. These skill tags can fully reflect the professional ability and expertise of employees. According to the skill tag set, keyword matching is performed on the industry talent market database. By matching the skill tags of employees within the enterprise with the job requirements and recruitment information of the external talent market, the current market demand for these skills, that is, the external talent supply and demand situation, is obtained. For example, if a certain skill is in high demand and low supply in the external market, then the value and shortage of the skill are very high. In order to ensure the quality and consistency of the data, data cleaning and standardization are performed on employee basic information, work resumes, performance records, and employee communication network data. Data cleaning includes steps such as removing duplicate data, filling missing values, and correcting erroneous data. For example, for missing values ​​in performance records, the mean interpolation method can be used to fill missing values. Standardization processing includes converting data into a unified format and unit so that data from different sources can be compared on the same dimension. According to structured employee data, in order to protect the personal privacy of employees, these data are desensitized. Desensitization includes anonymizing or blurring sensitive information such as name, ID number, and contact information, such as replacing the name with a code name, and representing some digits of the phone number with asterisks, so that the data can be used without revealing personal privacy. Perform data quality assessment on desensitized employee data, including data integrity, accuracy, and consistency. Generate a data quality report by calculating the missing rate, error rate, and duplication rate of each data field to identify and improve data quality issues. Based on the data quality report, interpolate and process outliers for low-quality data. For example, for some outliers, use the standard deviation method to identify and process abnormal data. Fusion the corrected employee data with external talent supply and demand.Integrate internal employee data with external market data to form a comprehensive, multi-dimensional employee data set. For example, through relational database or data warehouse technology, match employee skill tags with external market demand tags to generate a comprehensive data set that includes employee basic information, work history, performance records, communication networks, and external market demand.

[0024] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Feature extraction is performed on the multi-dimensional employee data set to obtain the employee's personal attribute vector and the initial network relationship matrix. Based on the initial network relationship matrix, a weighted graph of the working relationships between employees is constructed to obtain the initial talent network graph. The initial talent network graph is encoded using a graph neural network algorithm to obtain a low-dimensional representation of employee nodes. According to the low-dimensional representation of employee nodes, the employee's personal attribute vector is feature fused to obtain the fused employee feature vector. The fused employee feature vector is analyzed by multi-head attention mechanism to calculate the similarity and correlation strength between employees to obtain the updated relationship matrix. According to the updated relationship matrix, the edge weights of the initial talent network graph are updated and the structure is optimized to obtain the optimized talent network graph. The optimized talent network graph is then clustered and analyzed to obtain the employee community structures at different levels. According to the employee community structure, each community is analyzed for centrality and influence, and community feature descriptions and key node sets are obtained. The community feature descriptions and key node sets are hierarchically aggregated to construct a multi-level network structure including individual, team and organizational levels. Through dynamic evolution analysis of multi-level network structure, a multi-level talent distribution network is obtained.

[0025] Specifically, feature extraction is performed on the multi-dimensional employee data set, and the basic information and work history of employees are converted into numerical personal attribute vectors. For example, the employee's years of work, performance score, number of projects participated in, etc. can all constitute part of the personal attribute vector. Based on the employee's interaction records in the instant messaging system and email system, an initial network relationship matrix is ​​constructed. Each element in the matrix Represents the frequency or intensity of interaction between employee i and employee j, forming an initial representation of the working relationship between employees. Based on the initial network relationship matrix, a weighted graph of the working relationship between employees is constructed to obtain an initial talent network graph. In the network graph, nodes represent employees, and the weight of edges represents the strength of the working relationship between employees. In this way, the working relationship and interaction of employees within the enterprise are intuitively represented. The node encoding calculation of the initial talent network graph is performed through the graph neural network algorithm. The feature matrix of the node And the normalized adjacency matrix Input graph convolutional network and calculate by formula: ; in represents the node representation of layer I, represents the weight matrix of layer I, represents the activation function, such as ReLU. In the first layer, , through several layers of graph convolution operations, a low-dimensional representation of employee nodes is obtained According to the low-dimensional representation , perform feature fusion on the employee’s personal attribute vector to form a fused employee feature vector. Analyze the fused feature vector through the multi-head attention mechanism to calculate the similarity and correlation strength between employees. The multi-head attention mechanism can capture richer feature information through multiple sets of linear transformations and attention calculations. The formula is as follows: ; Where Q, K, and V represent query, key, and value matrices, respectively. The dimension of the key. The calculated similarity and association strength can generate an updated relationship matrix. According to the updated relationship matrix, the edge weights of the initial talent network graph are updated and the structure is optimized to obtain the optimized talent network graph. Through the optimization process, the actual working relationship and interaction between employees can be more accurately reflected. Cluster analysis is performed on the optimized talent network graph, such as using a density-based clustering algorithm (DBSCAN) to identify the employee community structure at different levels in the network. The community structure shows the distribution of employees within the enterprise and the group relationship. According to the employee community structure, centrality analysis and influence evaluation are performed on each community. Centrality analysis can identify key nodes in the community by calculating indicators such as degree centrality, closeness centrality, and betweenness centrality of nodes. Influence evaluation can evaluate the overall influence of nodes on the community by analyzing the connection of nodes inside and outside the community. Based on these analysis results, community feature descriptions and key node sets are generated. The community feature descriptions and key node sets are hierarchically aggregated to construct a multi-level network structure including individual, team, and organizational levels. For example, individual communities are aggregated into teams, teams are aggregated into departments, and finally the overall organizational structure of the enterprise is formed. Multi-level network structures can fully display the distribution and relationships of talents within the enterprise. Dynamic evolution analysis of multi-level network structures is performed to predict future trends by observing and simulating changes in network structures over time. For example, through time series analysis and dynamic network models, the impact of employee mobility, team changes, and organizational reorganization is studied to obtain a multi-level talent distribution network.

[0026] In a specific embodiment, the process of executing step S3 may specifically include the following steps: Analyze the multi-dimensional employee data set to extract the initial indicator set including performance, ability and quality, potential development and team contribution. Based on the initial indicator set, conduct text analysis on the company's strategic goals and job requirements to obtain a set of keywords related to the company's strategy. Calculate the semantic similarity between the keyword set and the initial indicator set to obtain a preliminary evaluation system; According to the preliminary evaluation system, the historical talent evaluation data was subjected to principal component analysis to obtain the explained variance ratio of each indicator, and the explained variance ratio was subjected to threshold screening to obtain multi-dimensional evaluation indicators; According to the multi-dimensional evaluation indicators, the judgment matrix of the hierarchical analysis method is constructed to obtain the relative importance matrix between the indicators, and the eigenvalue calculation of the relative importance matrix is ​​performed to obtain the maximum eigenvalue and the corresponding eigenvector; According to the maximum eigenvalue and eigenvector, the consistency ratio of the hierarchical analysis method is calculated to judge the rationality of the matrix, obtain the indicator weight, conduct sensitivity analysis on the indicator weight, and obtain the stability evaluation result by adjusting the element value in the judgment matrix; According to the stability assessment results, the indicator weights of the multi-dimensional evaluation indicators are adjusted and normalized to obtain the evaluation indicator system.

[0027] Specifically, relevant data is obtained from various systems of the enterprise (such as ERP, CRM and OA systems). These data include employee performance scores, project participation, skill assessments and teamwork records. Through data cleaning and integration, these data are converted into a quantifiable initial indicator set. For example, performance can be measured by employee performance scores, ability and quality can be reflected by skill assessment results, potential development can be reflected by promotion records and training results, and team contribution can be measured by project participation and colleague evaluation. According to the initial indicator set, text analysis is performed on the enterprise's strategic goals and job requirements. The enterprise's strategic documents and job descriptions are collected, and these texts are segmented and keyword extracted through natural language processing technology to obtain a set of keywords related to the enterprise's strategy. Keywords can reflect the current focus and future development direction of the enterprise. For example, if the enterprise's strategic goals include "technological innovation" and "market expansion", then the corresponding keyword set may include "innovation ability", "R&D results", "marketing" and "customer relationship". The semantic similarity of the keyword set and the initial indicator set is calculated, and the preliminary evaluation system is constructed by calculating the similarity between each keyword and each indicator. Use a word vector model (such as Word2Vec or GloVe) to convert keywords and indicators into vector representations, and then use the cosine similarity calculation formula: ; Where A and B represent the vectors of keywords and indicators respectively. Calculate the similarity value between them to determine which indicators are most relevant to the company's strategic goals. According to the preliminary evaluation system, perform principal component analysis on historical talent evaluation data. Principal component analysis is a dimensionality reduction technique that converts raw data into several principal components through linear transformation. These principal components can explain most of the variance of the raw data. By solving the eigenvalues ​​and eigenvectors of the covariance matrix, the explained variance ratio of each principal component is obtained. The formula is as follows: ; in represents the proportion of variance explained by the ith principal component, Represents the variance of the i-th principal component, and n is the total number of principal components. A threshold screening is performed on the explained variance ratio, and the principal components that can explain most of the variance are selected to obtain multidimensional evaluation indicators. According to the multidimensional evaluation indicators, a judgment matrix of the hierarchical analysis method is constructed to determine the relative importance of each indicator. The judgment matrix is ​​a pairwise comparison matrix. The elements in the matrix represent the relative importance between two indicators, which is determined by expert scoring or data analysis. The eigenvalue calculation is performed on the relative importance matrix to solve the maximum eigenvalue and the corresponding eigenvector. The normalized result of the eigenvector is the weight of each indicator. In order to verify the rationality of the judgment matrix, the consistency ratio is calculated. The calculation formula of the consistency ratio is: ; in, is the consistency indicator, is the maximum eigenvalue of the judgment matrix, is the order of the matrix, is a random consistency index. Generally, if CR is less than 0.1, the judgment matrix is ​​considered to have satisfactory consistency. According to the maximum eigenvalue and eigenvector, the consistency ratio of the hierarchical analysis method is calculated, the rationality of the judgment matrix is ​​judged, and the indicator weight is obtained. A sensitivity analysis is performed on the indicator weights. By adjusting the element values ​​in the judgment matrix and observing the changes in the weights, the stability evaluation results are obtained. In this way, the stability and reliability of the indicator weights are ensured under different circumstances. According to the stability evaluation results, the indicator weights of the multidimensional evaluation indicators are adjusted and normalized to obtain the final evaluation indicator system. The normalization process is to ensure that the sum of all indicator weights is 1, which is convenient for subsequent comprehensive evaluation and comparison.

[0028] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Perform data type analysis on each indicator in the evaluation indicator system to obtain a quantitative indicator set and a qualitative indicator set; According to the quantitative indicator set, the numerical data in the multi-dimensional employee data set is normalized to obtain standardized quantitative indicator values, and the text data in the qualitative indicator set is segmented, POS tagged, and entity recognized to obtain structured text features; According to the structured text features, the text is converted into a vector representation to obtain the numerical representation of the qualitative indicators, and the standardized quantitative indicator values ​​and the numerical representation of the qualitative indicators are subjected to feature fusion to obtain an initial feature vector set; Perform dimensionality reduction processing on the initial feature vector set to obtain a feature matrix after dimensionality reduction, and perform high-dimensional data visualization mapping on the feature matrix after dimensionality reduction to obtain a two-dimensional scatter plot representation; Perform data point density cluster analysis based on the two-dimensional scatter plot representation to obtain preliminary grouping results of employees, and calculate feature importance scores for the preliminary grouping results and the initial feature vector set to obtain key feature subsets; The initial feature vector set is reconstructed and normalized according to the key feature subset to obtain a set of feature evaluation indicators.

[0029] Specifically, the data is classified and organized. Each indicator in the evaluation indicator system may involve different data types, such as numerical data such as employee performance scores and years of work, and text data such as employee evaluations and feedback. Through data type analysis, these indicators are divided into quantitative indicator sets and qualitative indicator sets. The quantitative indicator set includes all numerical data, such as performance scores, project participation times, years of work, etc.; the qualitative indicator set includes all text data, such as employee evaluations, skill descriptions, and work feedback. According to the quantitative indicator set, the numerical data in the multi-dimensional employee data set is normalized, and data of different dimensions are converted to the same dimension for comparison and analysis, so as to obtain standardized quantitative indicator values. At the same time, for the text data in the qualitative indicator set, natural language processing operations such as word segmentation, part-of-speech tagging, and entity recognition are performed. Word segmentation is to divide the text into separate words, part-of-speech tagging is to determine the part of speech of each word, and entity recognition is to identify key entities in the text, such as names of people, places, and organizations. These processing steps can help the system convert unstructured text data into structured text features. According to the structured text features, the text is converted into a vector representation. Common methods include the Bag of Words model, TF-IDF (Term Frequency-Inverse Document Frequency) and word vector models (Word2Vec, GloVe). For example, using the TF-IDF method, the importance of each word in the document is calculated and represented as a vector. Suppose there is a text containing n words, and the vector calculated by TF-IDF is represented as TF-IDF= ,in Represents the TF-IDF value of the i-th word. Thus, the qualitative indicators are converted into numerical representations. The standardized quantitative indicator values ​​and the numerical representations of the qualitative indicators are subjected to feature fusion to obtain an initial feature vector set. Data from different sources are integrated to form a comprehensive feature representation. Quantitative and qualitative features are fused into a unified vector representation through splicing or weighting methods. In order to reduce the data dimension and reduce the computational complexity, the initial feature vector set is subjected to dimensionality reduction processing. Common dimensionality reduction methods include principal component analysis (PCA), linear discriminant analysis (LDA) and t-SNE. Taking PCA as an example, by solving the eigenvalues ​​and eigenvectors, the high-dimensional data is projected into a low-dimensional space to obtain the feature matrix after dimensionality reduction. For example, suppose the original feature matrix is , the dimension reduction formula of PCA is ,in is the eigenvector matrix of PCA, is the feature matrix after dimensionality reduction. The high-dimensional data is visualized and mapped on the feature matrix after dimensionality reduction to obtain a two-dimensional scatter plot representation. t-SNE is a commonly used high-dimensional data visualization method. By maintaining the local structure of high-dimensional data and mapping it to a low-dimensional space, an intuitive two-dimensional scatter plot is obtained. Scatter plots can help observe the distribution and clustering characteristics of data points. According to the two-dimensional scatter plot representation, data point density clustering analysis is performed. Common density clustering algorithms include DBSCAN, which identifies different clusters by analyzing the density of data points. The clustering results of DBSCAN can not only identify high-density areas, but also detect noise points. Through cluster analysis, the preliminary grouping results of employees are obtained. In order to identify key features, the feature importance scores are calculated for the preliminary grouping results and the initial feature vector set. Common methods include decision trees, random forests, etc., which identify the features that have the greatest impact on the grouping results by calculating the importance score of each feature in the classification task. The initial feature vector set is reconstructed and normalized according to the key feature subset to obtain a set of feature evaluation indicators. The purpose of reconstruction is to retain key features, remove redundant features, and improve the efficiency and accuracy of the model. Normalization is to ensure that all features are in the same dimension, making subsequent analysis more reliable. Through the above steps, a high-quality feature evaluation index set is obtained, which fully reflects the various aspects of the employee's ability and contribution.

[0030] In a specific embodiment, the process of executing step S5 may specifically include the following steps: Standardize the feature evaluation index set to obtain the node feature matrix; According to the multi-level talent distribution network, an initial adjacency matrix is ​​constructed, and the degree matrix of the initial adjacency matrix is ​​calculated to obtain a normalized adjacency matrix; The node feature matrix and the normalized adjacency matrix are input into the first graph convolution layer, and the features are calculated through matrix multiplication and ReLU activation function to obtain the first hidden layer representation; Perform a dropout operation on the first hidden layer representation, randomly set 50% of the neurons to zero, and obtain a sparse hidden layer representation. The sparse hidden layer representation is input into the second graph convolution layer, and feature calculation is performed through matrix multiplication and softmax activation function to obtain the output layer representation. Perform talent category labeling operations on the output layer representation to obtain preliminary talent category labels for each node, and perform average pooling operations on node features at the team level based on the preliminary talent category labels to obtain team-level talent distribution vectors; Perform department-level maximum pooling on the team-level talent distribution vector to obtain the department-level talent distribution vector, and input the department-level talent distribution vector into the fully connected layer. Perform feature calculation through the weight matrix and sigmoid activation function to obtain the company-level talent distribution vector. Connecting the team-level talent distribution vector, the department-level talent distribution vector, and the company-level talent distribution vector to obtain multiple talent identification results; The multiple talent identification results are divided into K talent types, and a corresponding intelligent management strategy is matched for each talent type according to a preset strategy mapping table.

[0031] Specifically, each indicator is normalized so that data with different features can be compared on the same scale to obtain the node feature matrix , where each row represents a standardized feature vector of an employee and each column represents a feature. Construct the initial adjacency matrix based on the multi-level talent distribution network The adjacency matrix is ​​a square matrix, where Represents the strength of the relationship between employee i and employee j. For example, if two employees work in the same team and have more cooperation, the strength of their relationship will be higher. Based on the adjacency matrix, calculate the degree matrix , which is a diagonal matrix where each diagonal element Represents the degree of node i, that is, the number of connections of node i. Calculate the normalized adjacency matrix The formula is: ; Through normalization, we can ensure that the node relationships in the network are properly considered in feature calculation. and the normalized adjacency matrix Input the first layer of graph convolution layer, perform feature calculation through matrix multiplication and ReLU activation function, and get the first hidden layer representation The calculation formula of the graph convolution layer is: ; in, is the weight matrix of the first layer, and ReLU is the activation function. This formula combines the features of a node with the features of its neighboring nodes, and obtains a new representation of each node through linear transformation and nonlinear activation function. Perform a dropout operation and randomly set 50% of the neurons to zero to obtain a sparse hidden layer representation. The dropout operation can prevent the model from overfitting and improve the generalization ability of the model. The sparse hidden layer representation is input into the second layer of graph convolution layer, and the feature calculation is performed through matrix multiplication and softmax activation function to obtain the output layer representation. The calculation formula of the second graph convolution layer is: ; in, is the weight matrix of the second layer, and softmax is the activation function used to convert the output into a probability distribution. Perform talent category labeling operations to obtain preliminary talent category labels for each node. These labels can be determined by the maximum probability principle, that is, the category with the highest probability is selected as the category label of the node. According to the preliminary talent category labels, perform average pooling operations on the node features at the team level to obtain a team-level talent distribution vector. The average pooling operation forms an overall feature representation of the team by taking the average of the node features within the same team. Perform a department-level maximum pooling operation on the team-level talent distribution vector to obtain a department-level talent distribution vector. The maximum pooling operation selects the maximum value on each feature dimension to highlight the most significant features. Input the department-level talent distribution vector into the fully connected layer, perform feature calculations through the weight matrix and sigmoid activation function, and obtain the company-level talent distribution vector. The calculation formula of the fully connected layer is: ; in, is the weight matrix, is the input vector, is the bias vector, is a sigmoid activation function, which is used to compress the output between 0 and 1. The team-level talent distribution vector, department-level talent distribution vector, and company-level talent distribution vector are connected to form a comprehensive feature representation to obtain multiple talent identification results. Through these results, employees are divided into different talent types. Assuming that there are K preset talent types, each talent identification result is mapped to the corresponding intelligent management strategy through a strategy mapping table. The strategy mapping table matches different talent types to corresponding management measures based on the management needs and talent characteristics of the enterprise. For example, for technical innovation talents, the enterprise may provide more training and R&D opportunities; for management talents, the enterprise may provide more leadership development and team management training.

[0032] In a specific embodiment, the execution step performs a dropout operation on the first hidden layer representation, randomly sets 50% of the neurons to zero, obtains a sparse hidden layer representation, and inputs the sparse hidden layer representation into the second graph convolution layer, performs feature calculation through matrix multiplication and softmax activation function, and obtains the output layer representation. The process can specifically include the following steps: Generate a random mask matrix for the first hidden layer representation to obtain a dropout mask, and perform element-wise multiplication of the first hidden layer representation and the dropout mask to obtain a sparse hidden layer representation; Perform matrix multiplication on the sparse hidden layer representation and the normalized adjacency matrix to obtain the aggregated node representation, and perform matrix multiplication on the aggregated node representation and the weight matrix of the second graph convolutional layer to obtain the original output; Perform batch normalization on the original output, calculate the mean and variance, and perform scaling and translation to obtain the normalized output. The normalized output is input into the LeakyReLU activation function for nonlinear transformation to obtain the activated feature representation. Perform a global maximum pooling operation on the activated feature representation, extract the maximum value of each feature channel, obtain the pooled feature vector, and input the pooled feature vector into the fully connected layer for linear transformation to obtain the target feature vector; The softmax function is used to calculate the probability distribution of each category of the target feature vector to obtain the probability output. Based on the probability output, the argmax function is used to select the category index with the largest probability to obtain the output layer representation.

[0033] Specifically, for the first hidden layer, Perform random mask matrix generation. Assume The dimension is ,in is the number of nodes, is the dimension of the feature. Generate a Random mask matrices of the same shape , where each element is zero with probability 0.5 and one with probability 0.5. This process can be expressed by the following formula: ; Will and Perform element-wise multiplication to obtain a sparse hidden layer representation : ; Among them, © represents element-wise multiplication. Through the dropout operation, 50% of the neurons are randomly set to zero to prevent overfitting and improve the generalization ability of the model. With the normalized adjacency matrix Perform matrix multiplication to obtain the aggregated node representation: ; By weighted averaging the features of each node with those of its neighboring nodes, feature propagation and aggregation are achieved. The aggregated node representation is combined with the weight matrix of the second graph convolution layer. Perform matrix multiplication to get the original output: ; Perform batch normalization on the raw output to speed up the training process, stabilize the model training, and improve the model performance. The batch normalization steps include calculating the mean and variance, and then scaling and translating. The specific formula is as follows: ; in, is the mean, is the variance, is a very small number used to prevent division by zero. To zoom and pan: ; in, and are trainable scaling and translation parameters. Through batch normalization, we get the normalized output . Input the normalized output into the LeakyReLU activation function for nonlinear transformation. The formula of LeakyReLU is as follows: ; in, is a very small constant, usually 0.01. Through the LeakyReLU activation function, the activated feature representation is obtained . The activated feature representation Perform a global maximum pooling operation. Global maximum pooling compresses the feature vector to a smaller size by extracting the maximum value of each feature channel while retaining the most important information. Assume The dimension is , after global maximum pooling, we get the pooling feature vector , whose dimensions are , the formula is as follows: ; The pooled feature vector Input the fully connected layer for linear transformation to obtain the target feature vector : ; in, is the weight matrix of the fully connected layer, is the bias vector. The target feature vector is transformed into Calculate the probability distribution of each category and get the probability output : ; in, is the probability of the ith category, is the i-th element of the target feature vector. Output according to probability , use the argmax function to select the category index with the highest probability and get the output layer representation.

[0034] The above describes the chain enterprise's talent intelligent management method in the embodiment of the present invention. The following describes the chain enterprise's talent intelligent management system in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an embodiment of the chain enterprise's intelligent talent management system includes: The acquisition module is used to collect and preprocess multi-dimensional data from the enterprise's internal system and external data sources to obtain a multi-dimensional employee data set; The construction module is used to construct a talent network for multi-dimensional employee data sets, integrating individual attributes and network attributes to obtain a multi-level talent distribution network; An allocation module is used to extract multi-dimensional evaluation indicators based on the multi-dimensional employee data set, and to allocate weights to the multi-dimensional evaluation indicators through the hierarchical analysis method to obtain an evaluation indicator system; A calculation module is used to calculate characteristic indicators of a multi-dimensional employee data set according to an evaluation indicator system to obtain a characteristic evaluation indicator set; The identification module is used to perform graph convolutional neural network and multi-scale aggregation analysis on the feature evaluation indicator set and the multi-level talent distribution network to obtain multiple talent identification results and match the intelligent management strategy for each talent identification result.

[0035] Through the collaboration of the above components, and through the integration and preprocessing of multiple data sources inside and outside the enterprise, employee information can be obtained comprehensively and accurately, overcoming the problems of data dispersion and information islands in traditional methods, integrating personal attributes and network attributes, and building a multi-level talent distribution network, which can not only reflect the personal characteristics of employees, but also show the relationship and team structure between employees. Through the hierarchical analysis method, the evaluation indicators are scientifically weighted, and a comprehensive and objective evaluation system is established, avoiding the shortcomings of single evaluation criteria and strong subjectivity in traditional methods. Standardization and numerical processing of quantitative and qualitative indicators, and extraction of key features through dimensionality reduction and cluster analysis can more accurately characterize the ability and potential of employees. Using deep learning technology, feature learning while considering the network structure can better capture the complex relationship between employees and team dynamics, and improve the accuracy of talent identification. Through the aggregation of features at the three levels of team, department and company, multi-scale talent identification is achieved, which can comprehensively evaluate the performance and potential of employees at different organizational levels. Automatically matching the corresponding management strategy according to the identification results realizes the personalization and intelligence of talent management, which helps to improve management efficiency and effectiveness. This method has good scalability and can adapt to chain enterprises of different sizes and types, and can be adjusted and optimized according to the specific needs of the enterprises.

[0036] The present invention also provides a computer device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the chain enterprise's intelligent talent management method in the above-mentioned embodiments.

[0037] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the chain enterprise's intelligent talent management method.

[0038] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0039] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0040] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A talent intelligent management method for a chain enterprise, characterized in that: The chain enterprise's intelligent talent management method includes: Collect and preprocess multi-dimensional data from the enterprise's internal systems and external data sources to obtain a multi-dimensional employee data set; Constructing a talent network for the multi-dimensional employee data set, integrating individual attributes and network attributes, and obtaining a multi-level talent distribution network; Extracting multidimensional evaluation indicators based on the multidimensional employee data set, and assigning weights to the multidimensional evaluation indicators through a hierarchical analysis method to obtain an evaluation indicator system; Calculating characteristic indicators of the multi-dimensional employee data set according to the evaluation indicator system to obtain a characteristic evaluation indicator set; A graph convolutional neural network and multi-scale aggregation analysis are performed on the feature evaluation index set and the multi-level talent distribution network to obtain multiple talent identification results, and an intelligent management strategy is matched to each talent identification result.

2. The chain enterprise's intelligent talent management method according to claim 1, characterized in that: The multi-dimensional data collection and preprocessing of the enterprise internal system and external data sources are performed to obtain a multi-dimensional employee data set, including: Extract data from the enterprise ERP, CRM and OA systems to obtain basic employee information, work history and performance records; Based on the basic information of the employees, log analysis is performed on the enterprise instant messaging system and email system to obtain employee communication network data, and text mining is performed on the work resume to obtain a set of employee skill tags; According to the employee skill tag set, keyword matching is performed on the industry talent market database to obtain external talent supply and demand conditions; Performing data cleaning and standardization on the employee's basic information, work history, performance record, and employee communication network data to obtain structured employee data; Desensitizing the employee's personal privacy information according to the structured employee data to obtain desensitized employee data, and performing data quality assessment on the desensitized employee data to obtain a data quality report; According to the data quality report, low-quality data is interpolated and outliers are processed to obtain corrected employee data, and the corrected employee data is fused with external talent supply and demand conditions to obtain a multi-dimensional employee data set.

3. The chain enterprise's intelligent talent management method according to claim 1, characterized in that: The talent network is constructed for the multi-dimensional employee data set, and personal attributes and network attributes are integrated to obtain a multi-level talent distribution network, including: Extracting features from the multi-dimensional employee data set to obtain employee personal attribute vectors and an initial network relationship matrix, and constructing a weighted graph of working relationships between employees based on the initial network relationship matrix to obtain an initial talent network graph; Performing node encoding calculation on the initial talent network graph through a graph neural network algorithm to obtain a low-dimensional representation of employee nodes; According to the low-dimensional representation of the employee node, the employee personal attribute vector is feature fused to obtain a fused employee feature vector, and the fused employee feature vector is subjected to multi-head attention mechanism analysis to calculate the similarity and correlation strength between employees to obtain an updated relationship matrix; According to the updated relationship matrix, the edge weights of the initial talent network graph are updated and the structure is optimized to obtain an optimized talent network graph, and cluster analysis is performed on the optimized talent network graph to obtain employee community structures at different levels; According to the employee community structure, each community is analyzed for centrality and influence, and a community feature description and a key node set are obtained. The community feature description and the key node set are hierarchically aggregated to construct a multi-level network structure including individual, team and organizational levels. The multi-level network structure is dynamically analyzed to obtain a multi-level talent distribution network.

4. The method for intelligent talent management of a chain enterprise according to claim 1, characterized in that: The multi-dimensional evaluation index is extracted according to the multi-dimensional employee data set, and the multi-dimensional evaluation index is weighted by the analytic hierarchy process to obtain an evaluation index system, including: Performing data analysis on the multi-dimensional employee data set to extract an initial indicator set including performance, ability and quality, potential development, and team contribution, and performing text analysis on the enterprise strategic goals and job requirements based on the initial indicator set to obtain a set of keywords related to the enterprise strategy; Calculating semantic similarity between the keyword set and the initial indicator set to obtain a preliminary evaluation system; According to the preliminary evaluation system, the historical talent evaluation data is subjected to principal component analysis to obtain the explained variance ratio of each indicator, and the explained variance ratio is subjected to threshold screening to obtain multi-dimensional evaluation indicators; According to the multi-dimensional evaluation indicators, a judgment matrix of the hierarchical analysis method is constructed to obtain a relative importance matrix between the indicators, and eigenvalue calculation is performed on the relative importance matrix to obtain the maximum eigenvalue and the corresponding eigenvector; According to the maximum eigenvalue and eigenvector, the consistency ratio of the hierarchical analysis method is calculated, the rationality of the judgment matrix is ​​judged, the indicator weight is obtained, the sensitivity analysis is performed on the indicator weight, and the stability evaluation result is obtained by adjusting the element value in the judgment matrix; According to the stability evaluation result, the indicator weights of the multi-dimensional evaluation indicators are adjusted and normalized to obtain an evaluation indicator system.

5. The chain enterprise's intelligent talent management method according to claim 4 is characterized in that: The characteristic index calculation of the multi-dimensional employee data set according to the evaluation index system to obtain a characteristic evaluation index set includes: Performing data type analysis on each indicator in the evaluation indicator system to obtain a quantitative indicator set and a qualitative indicator set; According to the quantitative indicator set, the numerical data in the multi-dimensional employee data set is normalized to obtain standardized quantitative indicator values, and the text data in the qualitative indicator set is segmented, POS tagged and entity recognized to obtain structured text features; According to the structured text features, the text is converted into a vector representation to obtain a numerical representation of the qualitative indicator, and the standardized quantitative indicator value and the numerical representation of the qualitative indicator are subjected to feature fusion to obtain an initial feature vector set; Performing dimensionality reduction processing on the initial feature vector set to obtain a feature matrix after dimensionality reduction, and performing high-dimensional data visualization mapping on the feature matrix after dimensionality reduction to obtain a two-dimensional scatter plot representation; Performing data point density cluster analysis based on the two-dimensional scatter plot representation to obtain a preliminary grouping result of the employees, and calculating feature importance scores for the preliminary grouping result and the initial feature vector set to obtain a key feature subset; The initial feature vector set is reconstructed and normalized according to the key feature subset to obtain a feature evaluation index set.

6. The method for intelligent talent management of chain enterprises according to claim 1, characterized in that: The graph convolutional neural network and multi-scale aggregation analysis are performed on the feature evaluation index set and the multi-level talent distribution network to obtain multiple talent identification results, and an intelligent management strategy is matched for each talent identification result, including: Standardizing the feature evaluation index set to obtain a node feature matrix; Constructing an initial adjacency matrix according to the multi-level talent distribution network, and performing degree matrix calculation on the initial adjacency matrix to obtain a normalized adjacency matrix; Inputting the node feature matrix and the normalized adjacency matrix into the first graph convolution layer, performing feature calculation through matrix multiplication operation and ReLU activation function, and obtaining the first hidden layer representation; Perform a dropout operation on the first hidden layer representation, randomly set 50% of the neurons to zero, obtain a sparse hidden layer representation, and input the sparse hidden layer representation into the second graph convolution layer, perform feature calculation through matrix multiplication and softmax activation function, and obtain an output layer representation; Performing a talent category label operation on the output layer representation to obtain a preliminary talent category label for each node, and performing an average pooling operation on node features at the team level based on the preliminary talent category label to obtain a team-level talent distribution vector; Performing a department-level maximum pooling operation on the team-level talent distribution vector to obtain a department-level talent distribution vector, and inputting the department-level talent distribution vector into a fully connected layer, performing feature calculation through a weight matrix and a sigmoid activation function to obtain a company-level talent distribution vector; Performing a connection operation on the team-level talent distribution vector, the department-level talent distribution vector, and the company-level talent distribution vector to obtain multiple talent identification results; The multiple talent identification results are divided into K talent types, and a corresponding intelligent management strategy is matched for each talent type according to a preset strategy mapping table.

7. The method for intelligent talent management of chain enterprises according to claim 6, characterized in that: The step of performing a dropout operation on the first hidden layer representation, randomly setting 50% of the neurons to zero, obtaining a sparse hidden layer representation, and inputting the sparse hidden layer representation into the second graph convolution layer, performing feature calculation through matrix multiplication and softmax activation function, and obtaining an output layer representation includes: Generate a random mask matrix for the first hidden layer representation to obtain a dropout mask, and perform element-wise multiplication operation on the first hidden layer representation and the dropout mask to obtain a sparse hidden layer representation; Performing matrix multiplication on the sparse hidden layer representation and the normalized adjacency matrix to obtain an aggregated node representation, and performing matrix multiplication on the aggregated node representation and the weight matrix of the second graph convolutional layer to obtain an original output; Performing batch normalization on the original output, calculating the mean and variance, scaling and translating, obtaining a normalized output, and inputting the normalized output into a LeakyReLU activation function for nonlinear transformation to obtain an activated feature representation; Performing a global maximum pooling operation on the activated feature representation, extracting the maximum value of each feature channel to obtain a pooled feature vector, and inputting the pooled feature vector into a fully connected layer for linear transformation to obtain a target feature vector; The probability distribution of each category of the target feature vector is calculated using the softmax function to obtain a probability output, and based on the probability output, the argmax function is used to select the category index with the largest probability to obtain an output layer representation.

8. An intelligent talent management system for chain enterprises, characterized in that: Used to execute the talent intelligent management method of a chain enterprise as described in any one of claims 1 to 7, the talent intelligent management system of the chain enterprise comprises: The acquisition module is used to collect and preprocess multi-dimensional data from the enterprise's internal system and external data sources to obtain a multi-dimensional employee data set; A construction module is used to construct a talent network for the multi-dimensional employee data set, integrating personal attributes and network attributes to obtain a multi-level talent distribution network; An allocation module, used for extracting multi-dimensional evaluation indicators according to the multi-dimensional employee data set, and performing weight allocation on the multi-dimensional evaluation indicators through a hierarchical analysis method to obtain an evaluation indicator system; A calculation module, used to calculate characteristic indicators of the multi-dimensional employee data set according to the evaluation indicator system to obtain a characteristic evaluation indicator set; The identification module is used to perform graph convolutional neural network and multi-scale aggregation analysis on the feature evaluation index set and the multi-level talent distribution network to obtain multiple talent identification results and match the intelligent management strategy for each talent identification result.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, the method for intelligent talent management of a chain enterprise as claimed in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the talent intelligent management method for a chain enterprise according to any one of claims 1 to 7.

Images