Human resource risk management and control method and system based on big data

Abstract

The invention discloses a human resource risk management and control method and system based on big data, and relates to the technical field of human resource management, and the method comprises the following components: S1, a data collection and integration step, S2, an intelligent risk early warning step, S3, a visual risk monitoring step, and S4, a risk response step. According to the invention, human resource related data is comprehensively collected from a plurality of data sources, deep integration and cleaning are carried out, and a recurrent neural network risk prediction model based on a multilayer attention mechanism and deep reinforcement learning is utilized, so that the system can intelligently identify and early warn potential human resource risks, and the human resource risk prediction efficiency is improved. According to the method, the accuracy of risk identification is improved, the timeliness of risk management and control is ensured through real-time data updating and a dynamic early warning mechanism, and enterprises can quickly respond to risk signals and take effective measures for intervention, so that the possibility and the influence degree of risk occurrence are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of human resource management, and in particular to a human resource risk management method and system based on big data. Background Art

[0002] As enterprises expand in size and their business scope, human resource management faces more and more challenges. Human resource risks, such as employee turnover, poor performance, and recruitment errors, not only affect the daily operations of enterprises, but may also cause significant losses to their long-term development. In recent years, the rapid development of big data technology has provided new ideas and means for human resource management. Through big data technology, enterprises can collect and analyze human resource-related data from multiple data sources, revealing the laws and trends behind the data, thereby achieving accurate identification and early warning of human resource risks.

[0003] Traditional technologies have shortcomings. Traditional systems are often unable to fully collect and integrate all relevant data sources, resulting in missing and inaccurate data, which in turn affects the accuracy of risk identification. Secondly, traditional systems lack real-time and dynamic capabilities in risk warning and are unable to respond to risk signals in a timely manner, causing companies to miss the best time to intervene.

[0004] To sum up, traditional technology intelligent risk warning systems have many limitations when processing massive and complex human resource data, and cannot meet the modern enterprise's needs for accurate, real-time and intelligent risk management. Therefore, it is particularly important to develop a human resource risk management method and system based on big data. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the existing technology and provide a human resource risk management method and system based on big data. It can comprehensively collect and analyze human resource related data, and use a recurrent neural network risk prediction model based on multi-layer attention mechanism and deep reinforcement learning to achieve accurate identification and early warning of human resource risks.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a human resource risk management method based on big data, the specific steps of the method are: S1. Data collection and integration steps Collect human resources related data from multiple data sources. For internal enterprise systems, use customized interface programs to extract basic employee information, attendance records, and performance evaluation results from the human resources management system according to specific rules, obtain salary data and budget information from the financial system, and collect work tasks and communication record information from the office automation system. For external channels, use web crawlers combined with API interfaces to capture employee social dynamics from social media platforms, cooperate with data suppliers to obtain talent supply and demand information from the talent market database, and use web crawlers to extract industry trend data from industry report websites. The collected data includes structured, semi-structured, and unstructured data. When cleaning data, target outlier data , using an outlier detection algorithm based on fuzzy logic and probability statistics to construct a fuzzy membership function To evaluate the possibility of abnormality, the formula is: ,in is the data feature vector, for inconsistent data , based on the data consistency rule base , combined with semantic analysis and data constraint judgment processing, data conversion is based on knowledge graph The data mapping method is as follows: ,in is the cleaned data. It is the target data model. Data fusion adopts the Bayesian network model based on quantum particle swarm optimization. The quantum particle swarm optimization algorithm optimizes the Bayesian network structure and parameters. Each data source in the Bayesian network is a node, and the edge represents the dependency relationship. By calculating the probability distribution of the data source data , combined with the weights determined by quantum particle swarm optimization To achieve fusion, the formula is: ,in It is Data after conversion from one data source to another; S2. Intelligent risk warning steps Construct a recurrent neural network risk prediction model based on a multi-layer attention mechanism combined with deep reinforcement learning. The input layer takes the preprocessed data Convert to Vector , the attention mechanism layer highlights important information through the weight calculation method, the formula is: ,in , are the query and key vectors, is the vector dimension, is the semantic feature vector, is the context information vector. The deep reinforcement learning module learns the optimal strategy by interacting with the environment and defines the reward function ,in , , is the weight coefficient, is the risk prediction accuracy, It is the timeliness of early warning. It is other factors, the recurrent neural network layer captures the long-term dependency and dynamic characteristics of the data, and the output layer performs risk prediction; The training uses the cross entropy loss function Combined with adversarial training to optimize parameters, the cross entropy loss function takes into account prediction uncertainty and risk severity , the formula is: ,in is the true value, is the predicted value; Determine input features for different risk types and build sentiment analysis models and topic models Extract feature vectors and use preset warning indicators and threshold , automatically identify risk signals, and early warning indicators are based on risk factor weights and eigenvalues Calculation, the formula is: , the threshold is determined based on the historical data quantile and the enterprise’s risk appetite; S3. Visual risk monitoring steps The visualization technology based on WebGL and virtual reality (VR) is used to present human resource risks, and the dynamic dashboard obtains risk data in real time. , generate risk index values ​​through the indicator calculation model, and the comprehensive risk index The formula is calculated by nonlinear weighted summation of multiple risk indicators: ,in is the weight, The risk map is the value of each risk indicator, showing the changing trend through a variety of charts, using VR to achieve 3D visualization and interactive operation. The risk map is based on geographic information or organizational structure, using a method based on geographic information system (GIS) and organizational network analysis to convert risk data into Spatial mapping is performed, and the level of risk is indicated by visual elements such as colors and icons. VR is used to achieve immersive browsing. The platform provides interactive functions, and users can view detailed risk information and analysis reports through operations. The reports include risk descriptions, cause analysis, and historical data comparisons, and are presented through natural language generation technology; S4. Risk Response Steps Based on the preset risk response strategy library , automatically generates personalized response plans, and adopts a hybrid intelligent algorithm based on case reasoning and rule reasoning for different risk types and severity. Case reasoning is based on the case library. Retrieve similar cases, and calculate similarity based on semantic similarity and structural similarity The hybrid algorithm is: ,in is the weight coefficient. Based on similar cases and rule reasoning, the system adjusts and optimizes the response strategy. According to the organizational structure and personnel authority information, the system uses an intelligent allocation algorithm based on roles and task priorities to assign response tasks to relevant responsible persons. This algorithm takes into account the role of the responsible person. ,ability , Workload Factor, the formula is: ,in It is a response task, and the implementation of response measures is monitored using a task tracking algorithm based on blockchain technology. The algorithm uses the characteristics of blockchain to record the start time of the task , Estimated completion time , Actual completion time , execution status information, triggering task reminders and status updates through smart contract technology.

[0007] Furthermore, in the data collection step, the data collection strategy is further optimized in view of the diversity and complexity of the data in the internal system and external channels of the enterprise. For the internal system of the enterprise, a data cache and pre-fetch mechanism is introduced to establish a local cache for the data with low update frequency but frequent query in the human resource management system. , regularly update, use pre-fetching technology for data with high update frequency, collect data from external channels, and use a distributed web crawler architecture on social media platforms to introduce data filtering and screening mechanisms, and evaluate weights based on data credibility. Determine the data to be collected, establish a data subscription and push mechanism for the talent market database and industry report website, and use a method based on a multi-dimensional trust model to evaluate data credibility. The formula is: ,in The authority of the data source. is the quality of historical data, It is user evaluation.

[0008] Furthermore, in the intelligent risk warning step, the input features of the risk prediction model are further mined, the internal cultural atmosphere and value factors of the enterprise are introduced, and relevant features are extracted through sentiment analysis and topic modeling of internal text data. Sentiment analysis adopts a multimodal sentiment analysis model based on deep learning. , the formula is: ,in is text data, It is voice data. It is image data, and topic modeling uses an algorithm based on variational autoencoder , the formula is: ,in It is text data. For different risk type feature combinations, a feature selection method based on genetic algorithm is used. The fitness function formula is: ,in , is the weight coefficient, is the accuracy, is the recall rate.

[0009] Furthermore, in the intelligent risk warning step, the structure and training process of the deep learning model are further optimized. In terms of the model structure, a feature enhancement module based on a capsule network is introduced between the attention mechanism layers. The formula is: ,in It is the feature vector processed by the attention mechanism. The recurrent neural network layer adopts the LSTM and GRU hybrid structure. The training process adopts the training strategy based on curriculum learning according to the model performance. and data complexity Adjust the difficulty of training data, the formula is: ,in It's a difficulty adjustment.

[0010] Furthermore, in the visual risk monitoring step, the personalized display interface design for different user roles is further deepened. For senior managers of enterprises, strategic decision-making support functions are added. Through in-depth mining and analysis of risk data, combined with the strategic goals of the enterprise, and market environment , a strategic decision model is constructed based on scenario analysis and decision tree, and the formula is: ,in It is risk data. For human resources managers, we add talent development planning functions and adopt a talent development planning model based on machine learning. , the formula is: ,in It is employee data. It is the career goal of employees. For department managers, it provides team collaboration risk analysis function and adopts team collaboration risk assessment model based on social network analysis. , the formula is: ,in It is the communication data among team members. It is the task allocation data.

[0011] Furthermore, in the risk response step, the risk response strategy library Dynamically update and optimize, and regularly collect data on the results of countermeasure execution , combined with the actual risk treatment effect evaluation standards , using a reinforcement learning-based strategy update algorithm, by calculating the reward value of strategy execution , adjust the response strategies in the strategy library according to the reward value, and the reward value calculation takes into account the degree of risk reduction , cost investment , Improved employee satisfaction Factor, the formula is: ,in , , It is the weight coefficient, which improves the effectiveness and adaptability of the response plan by continuously iterating and updating the strategy library.

[0012] Furthermore, in the entire risk management process, a data security and privacy protection mechanism is established, and encryption algorithms are used for the collected and stored data. Encryption is performed to ensure the security of data during transmission and storage. The encryption algorithm uses a method based on quantum key distribution and homomorphic encryption, which can not only ensure the security of the key, but also realize direct calculation of encrypted data. For data involving employee privacy, access control strategies are used during data use. Limit data access rights. Access control policies are set based on roles and data sensitivity. Only users with corresponding permissions can access data of specific sensitivity. At the same time, detailed records of data access and use are kept to form audit logs. , in order to conduct data security audit and traceability.

[0013] Furthermore, during the operation of the system, an adaptive feedback adjustment mechanism is introduced to regularly collect system operation index data. , including risk prediction accuracy , Timeliness of early warning 、Effects of implementation of countermeasures According to these index data, an adaptive adjustment algorithm based on fuzzy control is used to adjust the index data and the preset target value. Compare and calculate deviation and the rate of change of deviation , according to the fuzzy rule base Determine the adjustment parameters and make corresponding adjustments to each module of the system. The fuzzy rule base is established based on a large amount of historical data and expert experience. Through continuous learning and optimization, the system can automatically adjust according to actual operating conditions to improve the overall performance of risk management.

[0014] On the other hand, a human resource risk management and control system based on big data is characterized in that the system includes a data collection and integration module, an intelligent risk warning module, a visual risk monitoring module and a risk response module: The data collection and integration module collects human resources related data from multiple data sources. For the internal system of the enterprise, it uses a customized interface program to extract basic employee information, attendance records, and performance evaluation results from the human resources management system according to specific rules, obtains salary data and budget information from the financial system, and collects work tasks and communication record information from the office automation system. For external channels, it uses a web crawler combined with an API interface to capture employee social dynamics from social media platforms, cooperates with data suppliers to obtain talent supply and demand information from the talent market database, and uses a web crawler to extract industry trend data from industry report websites. The collected data includes structured, semi-structured, and unstructured data. When cleaning data, outlier data is targeted , using an outlier detection algorithm based on fuzzy logic and probability statistics to construct a fuzzy membership function To evaluate the possibility of abnormality, the formula is: ,in is the data feature vector, for inconsistent data , based on the data consistency rule base , combined with semantic analysis and data constraint judgment processing, data conversion is based on knowledge graph The data mapping method is as follows: ,in is the cleaned data. It is the target data model. Data fusion adopts the Bayesian network model based on quantum particle swarm optimization. The quantum particle swarm optimization algorithm optimizes the Bayesian network structure and parameters. Each data source in the Bayesian network is a node, and the edge represents the dependency relationship. By calculating the probability distribution of the data source data , combined with the weights determined by quantum particle swarm optimization To achieve fusion, the formula is: ,in It is Data after conversion from one data source to another; The intelligent risk warning module: constructs a recurrent neural network risk prediction model based on a multi-layer attention mechanism combined with deep reinforcement learning. The input layer takes the preprocessed data Convert to Vector , the attention mechanism layer highlights important information through the weight calculation method, the formula is: ,in , are the query and key vectors, is the vector dimension, is the semantic feature vector, is the context information vector. The deep reinforcement learning module learns the optimal strategy by interacting with the environment and defines the reward function ,in , , is the weight coefficient, is the risk prediction accuracy, It is the timeliness of early warning. It is other factors, the recurrent neural network layer captures the long-term dependency and dynamic characteristics of the data, and the output layer performs risk prediction; The training uses the cross entropy loss function Combined with adversarial training to optimize parameters, the cross entropy loss function takes into account prediction uncertainty and risk severity , the formula is: ,in is the true value, is the predicted value; Determine input features for different risk types and build sentiment analysis models and topic models Extract feature vectors and use preset warning indicators and threshold , automatically identify risk signals, and early warning indicators are based on risk factor weights and eigenvalues Calculation, the formula is: , the threshold is determined based on the historical data quantile and the enterprise’s risk appetite; The visual risk monitoring module: uses visualization technology based on the fusion of WebGL and virtual reality (VR) to present human resource risks, and the dynamic dashboard obtains risk data in real time , generate risk index values ​​through the indicator calculation model, and the comprehensive risk index The formula is calculated by nonlinear weighted summation of multiple risk indicators: ,in is the weight, The risk map is the value of each risk indicator, showing the changing trend through a variety of charts, using VR to achieve 3D visualization and interactive operation. The risk map is based on geographic information or organizational structure, using a method based on geographic information system (GIS) and organizational network analysis to convert risk data into Spatial mapping is performed, and the level of risk is indicated by visual elements such as colors and icons. VR is used to achieve immersive browsing. The platform provides interactive functions, and users can view detailed risk information and analysis reports through operations. The reports include risk descriptions, cause analysis, and historical data comparisons, and are presented through natural language generation technology; The risk response module: according to the preset risk response strategy library , automatically generates personalized response plans, and adopts a hybrid intelligent algorithm based on case reasoning and rule reasoning for different risk types and severity. Case reasoning is based on the case library. Retrieve similar cases, and calculate similarity based on semantic similarity and structural similarity The hybrid algorithm is: ,in is the weight coefficient. Based on similar cases and rule reasoning, the system adjusts and optimizes the response strategy. According to the organizational structure and personnel authority information, the system uses an intelligent allocation algorithm based on roles and task priorities to assign response tasks to relevant responsible persons. This algorithm takes into account the role of the responsible person. ,ability , Workload Factor, the formula is: ,in It is a response task, and the implementation of response measures is monitored using a task tracking algorithm based on blockchain technology. The algorithm uses the characteristics of blockchain to record the start time of the task , Estimated completion time , Actual completion time , execution status information, triggering task reminders and status updates through smart contract technology.

[0015] Compared with the existing technology, this human resource risk management method and system based on big data has the following beneficial effects: 1. The present invention comprehensively collects human resource related data from multiple data sources, and conducts in-depth integration and cleaning. By using a recurrent neural network risk prediction model based on a multi-layer attention mechanism and deep reinforcement learning, the system can intelligently identify and warn of potential human resource risks. This method not only improves the accuracy of risk identification, but also ensures the timeliness of risk management through real-time data updates and dynamic warning mechanisms. Enterprises can respond quickly to risk signals and take effective measures to intervene, thereby reducing the possibility and impact of risks.

[0016] 2. Through the visualization technology based on the fusion of WebGL and virtual reality (VR), the present invention enables enterprises to intuitively understand the distribution and changing trends of human resource risks. At the same time, the system can also automatically generate personalized response plans based on the preset risk response strategy library to provide scientific decision-making support for enterprise managers. This not only enhances the pertinence and effectiveness of decision-making, but also improves the overall efficiency of human resource management through automation and intelligent means. Enterprise managers can focus more on strategic planning and business development without having to pay too much attention to tedious daily risk management and control work.

[0017] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is a process operation diagram of a human resource risk management method based on big data; Figure 2 This is a process operation diagram of a human resource risk management and control system based on big data. DETAILED DESCRIPTION

[0020] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0021] Embodiment 1 This embodiment describes a large Internet company with tens of thousands of employees, whose business covers multiple fields and whose personnel turnover is frequent.

[0022] Extract employee basic information, attendance and performance evaluation results from the enterprise human resources management system, obtain salary and budget data from the financial system, collect work tasks and communication records from the office automation system, use web crawlers and APIs to capture employee social dynamics from social media, obtain talent supply and demand information from the talent market database, extract industry trend data from industry report websites, and analyze outlier data. , using fuzzy logic and probability statistics algorithms, by constructing fuzzy membership functions Evaluate the possibility of abnormality, the formula is , process inconsistent data based on the data consistency rule base, and use the knowledge graph To convert data, the formula is , the Bayesian network model optimized by quantum particle swarm is used to fuse data, and the formula is .

[0023] Construct a recurrent neural network model that combines multi-layer attention mechanism with deep reinforcement learning. The input layer converts the preprocessed data into vectors, and the attention mechanism layer uses the formula Highlight important information and define reward function , the cross entropy loss function is combined with adversarial training to optimize the parameters, the formula is , determine the input features for employee turnover risk, construct sentiment analysis and topic model to extract feature vectors, and and thresholds to automatically identify risk signals.

[0024] Using WebGL and VR fusion technology, the dynamic dashboard obtains risk data in real time to generate risk indicator values. The risk comprehensive index is calculated through the formula Calculation, display changing trends through charts, risk maps present risks based on geographic information or organizational structure, and use VR to achieve immersive browsing and interaction.

[0025] According to the risk response strategy library, a hybrid algorithm of case reasoning and rule reasoning is used to generate response plans. The case similarity calculation adopts the formula , the system allocates tasks according to the organizational structure and personnel authority using an intelligent allocation algorithm based on roles and task priorities. The formula is , using blockchain technology to monitor task execution.

[0026] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modifications to the technical contents disclosed above without departing from the scope of the technical solution of the present invention. Any simple modification, different changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A human resource risk management method based on big data, characterized in that: The specific steps of this method are: S1. Data collection and integration steps Collect human resources related data from multiple data sources. For internal enterprise systems, use customized interface programs to extract basic employee information, attendance records, and performance evaluation results from the human resources management system according to specific rules, obtain salary data and budget information from the financial system, and collect work tasks and communication record information from the office automation system. For external channels, use web crawlers combined with API interfaces to capture employee social dynamics from social media platforms, cooperate with data suppliers to obtain talent supply and demand information from the talent market database, and use web crawlers to extract industry trend data from industry report websites. The collected data includes structured, semi-structured, and unstructured data. When cleaning data, target outlier data , using an outlier detection algorithm based on fuzzy logic and probability statistics to construct a fuzzy membership function To evaluate the possibility of abnormality, the formula is: ,in is the data feature vector, for inconsistent data , based on the data consistency rule base , combined with semantic analysis and data constraint judgment processing, data conversion is based on knowledge graph The data mapping method is as follows: ,in is the cleaned data. It is the target data model. Data fusion adopts the Bayesian network model based on quantum particle swarm optimization. The quantum particle swarm optimization algorithm optimizes the Bayesian network structure and parameters. Each data source in the Bayesian network is a node, and the edge represents the dependency relationship. By calculating the probability distribution of the data source data , combined with the weights determined by quantum particle swarm optimization To achieve fusion, the formula is: ,in It is Data after conversion from one data source to another; S2. Intelligent risk warning steps Construct a recurrent neural network risk prediction model based on a multi-layer attention mechanism combined with deep reinforcement learning. The input layer takes the preprocessed data Convert to Vector , the attention mechanism layer highlights important information through the weight calculation method, the formula is: ,in , are the query and key vectors, is the vector dimension, is the semantic feature vector, is the context information vector. The deep reinforcement learning module learns the optimal strategy by interacting with the environment and defines the reward function ,in , , is the weight coefficient, is the risk prediction accuracy, It is the timeliness of early warning. It is other factors, the recurrent neural network layer captures the long-term dependency and dynamic characteristics of the data, and the output layer performs risk prediction; The training uses the cross entropy loss function Combined with adversarial training to optimize parameters, the cross entropy loss function takes into account prediction uncertainty and risk severity , the formula is: ,in is the true value, is the predicted value; Determine input features for different risk types and build sentiment analysis models and topic models Extract feature vectors and use preset warning indicators and threshold , automatically identify risk signals, and early warning indicators are based on risk factor weights and eigenvalues Calculation, the formula is: , the threshold is determined based on the historical data quantile and the enterprise’s risk appetite; S3. Visual risk monitoring steps The visualization technology based on WebGL and virtual reality (VR) is used to present human resource risks, and the dynamic dashboard obtains risk data in real time. , generate risk index values ​​through the indicator calculation model, and the comprehensive risk index The formula is calculated by nonlinear weighted summation of multiple risk indicators: ,in is the weight, The risk map is the value of each risk indicator, showing the changing trend through a variety of charts, using VR to achieve 3D visualization and interactive operation. The risk map is based on geographic information or organizational structure, using a method based on geographic information system (GIS) and organizational network analysis to convert risk data into Spatial mapping is performed, and the level of risk is indicated by visual elements such as colors and icons. VR is used to achieve immersive browsing. The platform provides interactive functions, and users can view detailed risk information and analysis reports through operations. The reports include risk descriptions, cause analysis, and historical data comparisons, and are presented through natural language generation technology; S4. Risk Response Steps Based on the preset risk response strategy library , automatically generates personalized response plans, and adopts a hybrid intelligent algorithm based on case reasoning and rule reasoning for different risk types and severity. Case reasoning is based on the case library. Retrieve similar cases, and calculate similarity based on semantic similarity and structural similarity The hybrid algorithm is: ,in is the weight coefficient. Based on similar cases and rule reasoning, the system adjusts and optimizes the response strategy. According to the organizational structure and personnel authority information, the system uses an intelligent allocation algorithm based on roles and task priorities to assign response tasks to relevant responsible persons. This algorithm takes into account the role of the responsible person. ,ability , Workload Factor, the formula is: ,in It is a response task, and the execution of response measures is monitored using a task tracking algorithm based on blockchain technology. The algorithm uses the characteristics of blockchain to record the start time of the task , Estimated completion time , Actual completion time , execution status information, triggering task reminders and status updates through smart contract technology.

2. According to the method of human resource risk management based on big data according to claim 1, it is characterized in that: In the data collection step, the data collection strategy is further optimized in view of the diversity and complexity of the data in the internal system and external channels of the enterprise. For the internal system of the enterprise, a data cache and pre-fetch mechanism is introduced to establish a local cache for the data with low update frequency but frequent query in the human resource management system. , regularly update, use pre-fetching technology for data with high update frequency, collect data from external channels, and use a distributed web crawler architecture on social media platforms to introduce data filtering and screening mechanisms, and evaluate weights based on data credibility. Determine the data to be collected, establish a data subscription and push mechanism for the talent market database and industry report website, and use a method based on a multi-dimensional trust model to evaluate data credibility. The formula is: ,in The authority of the data source. is the quality of historical data, It is user evaluation.

3. According to the method of human resource risk management based on big data in claim 1, it is characterized in that: In the intelligent risk warning step, the input features of the risk prediction model are further mined, the internal cultural atmosphere and value factors of the enterprise are introduced, and relevant features are extracted through sentiment analysis and topic modeling of internal text data. Sentiment analysis adopts a multimodal sentiment analysis model based on deep learning. , the formula is: ,in is text data, It is voice data. It is image data, and topic modeling uses an algorithm based on variational autoencoder , the formula is: ,in It is text data. For different risk type feature combinations, a feature selection method based on genetic algorithm is used. The fitness function formula is: ,in , is the weight coefficient, is the accuracy, is the recall rate.

4. According to the method of human resource risk management based on big data in claim 1, it is characterized in that: In the intelligent risk warning step, the structure and training process of the deep learning model are further optimized. In terms of the model structure, a feature enhancement module based on a capsule network is introduced between the attention mechanism layers. The formula is: ,in It is the feature vector processed by the attention mechanism. The recurrent neural network layer adopts a hybrid structure of LSTM and GRU. The training process adopts a training strategy based on curriculum learning according to the model performance. and data complexity Adjust the difficulty of training data, the formula is: ,in It's a difficulty adjustment.

5. According to the method of human resource risk management based on big data in claim 1, it is characterized in that: In the visual risk monitoring step, the personalized display interface design for different user roles is further deepened. For senior managers of enterprises, strategic decision-making support functions are added. Through in-depth mining and analysis of risk data, combined with the strategic goals of the enterprise, and market environment , a strategic decision model is constructed based on scenario analysis and decision tree, and the formula is: ,in It is risk data. For human resources managers, we add talent development planning functions and adopt a talent development planning model based on machine learning. , the formula is: ,in It is employee data. It is the career goal of employees. For department managers, it provides team collaboration risk analysis function and adopts team collaboration risk assessment model based on social network analysis. , the formula is: ,in It is the communication data among team members. It is the task allocation data.

6. The method for human resource risk management based on big data according to claim 1 is characterized in that: In the risk response step, the risk response strategy library Dynamically update and optimize, and regularly collect data on the results of countermeasure execution , combined with the actual risk treatment effect evaluation standards , using a reinforcement learning-based strategy update algorithm, by calculating the reward value of strategy execution , adjust the response strategies in the strategy library according to the reward value, and the reward value calculation takes into account the degree of risk reduction , cost investment , Improved employee satisfaction Factor, the formula is: ,in , , is the weight coefficient.

7. The method for human resource risk management based on big data according to claim 1 is characterized in that: In the entire risk management process, a data security and privacy protection mechanism is established, and encryption algorithms are used for the collected and stored data. Encryption is performed using a combination of quantum key distribution and homomorphic encryption. For data involving employee privacy, access control policies are used during data use. Limit data access rights. Access control policies are set based on roles and data sensitivity. Only users with corresponding permissions can access data of specific sensitivity. At the same time, detailed records of data access and use are kept to form audit logs. .

8. The method for human resource risk management based on big data according to claim 1 is characterized in that: During the operation of the system, an adaptive feedback adjustment mechanism is introduced to regularly collect system operation index data. , including risk prediction accuracy , Timeliness of early warning 、Effects of implementation of countermeasures According to these index data, an adaptive adjustment algorithm based on fuzzy control is used to adjust the index data and the preset target value. Compare and calculate deviation and the rate of change of deviation , according to the fuzzy rule base Determine the tuning parameters.

9. A human resource risk management and control system based on big data, characterized in that: The system includes data collection and integration module, intelligent risk warning module, visual risk monitoring module and risk response module: The data collection and integration module collects human resources related data from multiple data sources. For the internal system of the enterprise, it uses a customized interface program to extract basic employee information, attendance records, and performance evaluation results from the human resources management system according to specific rules, obtains salary data and budget information from the financial system, and collects work tasks and communication record information from the office automation system. For external channels, it uses a web crawler combined with an API interface to capture employee social dynamics from social media platforms, cooperates with data suppliers to obtain talent supply and demand information from the talent market database, and uses a web crawler to extract industry trend data from industry report websites. The collected data includes structured, semi-structured, and unstructured data. When cleaning data, outlier data is targeted , using an outlier detection algorithm based on fuzzy logic and probability statistics to construct a fuzzy membership function To evaluate the possibility of abnormality, the formula is: ,in is the data feature vector, for inconsistent data , based on the data consistency rule base , combined with semantic analysis and data constraint judgment processing, data conversion is based on knowledge graph The data mapping method is as follows: ,in is the cleaned data. It is the target data model. Data fusion adopts the Bayesian network model based on quantum particle swarm optimization. The quantum particle swarm optimization algorithm optimizes the Bayesian network structure and parameters. Each data source in the Bayesian network is a node, and the edge represents the dependency relationship. By calculating the probability distribution of the data source data , combined with the weights determined by quantum particle swarm optimization To achieve fusion, the formula is: ,in It is Data after conversion from one data source to another; The intelligent risk warning module: constructs a recurrent neural network risk prediction model based on a multi-layer attention mechanism combined with deep reinforcement learning. The input layer takes the preprocessed data Convert to Vector , the attention mechanism layer highlights important information through the weight calculation method, the formula is: ,in , are the query and key vectors, is the vector dimension, is the semantic feature vector, is the context information vector. The deep reinforcement learning module learns the optimal strategy by interacting with the environment and defines the reward function ,in , , is the weight coefficient, is the risk prediction accuracy, It is the timeliness of early warning. It is other factors, the recurrent neural network layer captures the long-term dependency and dynamic characteristics of the data, and the output layer performs risk prediction; The training uses the cross entropy loss function Combined with adversarial training to optimize parameters, the cross entropy loss function takes into account prediction uncertainty and risk severity , the formula is: ,in is the true value, is the predicted value; Determine input features for different risk types and build sentiment analysis models and topic models Extract feature vectors and use preset warning indicators and threshold , automatically identify risk signals, and early warning indicators are based on risk factor weights and eigenvalues Calculation, the formula is: , the threshold is determined based on the historical data quantile and the enterprise’s risk appetite; The visual risk monitoring module: uses visualization technology based on the fusion of WebGL and virtual reality (VR) to present human resource risks, and the dynamic dashboard obtains risk data in real time , generate risk index values ​​through the indicator calculation model, and the comprehensive risk index The formula is calculated by nonlinear weighted summation of multiple risk indicators: ,in is the weight, The risk map is the value of each risk indicator, showing the changing trend through a variety of charts, using VR to achieve 3D visualization and interactive operation. The risk map is based on geographic information or organizational structure, using a method based on geographic information system (GIS) and organizational network analysis to convert risk data into Spatial mapping is performed, and the level of risk is indicated by color and icon visual elements. VR is used to achieve immersive browsing. The platform provides interactive functions, and users can view detailed risk information and analysis reports through operations. The reports include risk descriptions, cause analysis, and historical data comparisons, and are presented through natural language generation technology; The risk response module: according to the preset risk response strategy library , automatically generates personalized response plans, and adopts a hybrid intelligent algorithm based on case reasoning and rule reasoning for different risk types and severity. Case reasoning is based on the case library. Retrieve similar cases, and calculate similarity based on semantic similarity and structural similarity The hybrid algorithm is: ,in is the weight coefficient. Based on similar cases and rule reasoning, the system adjusts and optimizes the response strategy. According to the organizational structure and personnel authority information, the system uses an intelligent allocation algorithm based on roles and task priorities to assign response tasks to relevant responsible persons. This algorithm takes into account the role of the responsible person. ,ability , Workload Factor, the formula is: ,in It is a response task, and the execution of response measures is monitored using a task tracking algorithm based on blockchain technology. The algorithm uses the characteristics of blockchain to record the start time of the task , Estimated completion time , Actual completion time , execution status information, triggering task reminders and status updates through smart contract technology.

Images