Artificial Intelligence-Based Employee Position Matching and Deployment Method and System

Abstract

The present invention provides a method and system for employee position matching and allocation based on artificial intelligence, which relates to the field of artificial intelligence technology and includes: generating an optimized employee ability feature vector based on historical employee work data, extracting the employee ability development trend, combining team collaboration data, obtaining an employee comprehensive ability vector by using a cross-modal contrastive learning network, calculating a development potential index, constructing a position skill knowledge graph based on position description data and historical recruitment data, generating a position skill feature matrix and a team structure feature vector through a dynamic heterogeneous graph neural network and an adaptive graph attention network, combining the employee comprehensive ability vector, calculating an employee-position matching degree score matrix, inputting the matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model, constructing a reward function, generating and evaluating a deployment plan through hierarchical task decomposition and Monte Carlo tree search, and finally outputting a deployment execution plan.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to an employee position matching and allocation method and system based on artificial intelligence. Background Art

[0002] With the rapid development of artificial intelligence technology, enterprises' demand for efficient and accurate employee position matching and allocation is increasing day by day. Traditional employee position matching and allocation methods mainly rely on manual experience and subjective judgment, with low efficiency and prone to matching errors, and it is difficult to meet the rapidly changing needs of enterprises;

[0003] Currently, the existing technologies mainly perform text matching using employees' resume information and job descriptions, or matching based on employees' skill tags and job requirements. However, there are still problems such as only considering employees' static skills and experience, being difficult to capture employees' ability development trends and potential, ignoring the impact of teamwork factors on employees' performance, and being unable to make real-time adjustments according to the development and changes of enterprises and the improvement of employees' abilities;

[0004] Therefore, there is an urgent need for a solution to solve the problems existing in the prior art. Summary of the Invention

[0005] Embodiments of the present invention provide an employee position matching and allocation method and system based on artificial intelligence, which can at least solve some of the problems existing in the prior art.

[0006] In the first aspect of the embodiments of the present invention, an employee position matching and allocation method based on artificial intelligence is provided, including:

[0007] Collecting employees' historical work data, where the employees' historical work data includes work performance data, skill assessment data, and teamwork data, performing multi-scale feature extraction on the employees' historical work data to obtain an initial employee ability feature vector and adding it to a pre-set diffusion probability model, generating an optimized employee ability feature vector in combination with a Markov chain sampling process, adding the work performance data and the skill assessment data to a pre-set spatio-temporal hybrid network model, extracting the employees' ability development trend through convolution operations, adding the employees' ability development trend, the optimized employee ability feature vector, and the teamwork data to a cross-modal contrast learning network, solving for an employee comprehensive ability vector based on a contrast loss function and a dynamic temperature parameter and adding it to a feature enhancement network, and obtaining a development potential index in combination with a multi-scale feature pyramid and an adaptive feature aggregation module;

[0008] Collect job description data and historical recruitment data and add them to a pre - constructed generative pre - trained language model for semantic understanding and knowledge extraction, construct a job skill knowledge graph, add the job skill knowledge graph to a dynamic heterogeneous graph neural network, and the dynamic heterogeneous graph neural network generates a job skill feature matrix through time - varying topology learning, heterogeneous sub - graph sampling, and graph structure distillation, and combines an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector. Add the job skill feature matrix, the comprehensive employee ability vector, and the team structure feature vector to a neural architecture search network, and obtain an employee - job matching score matrix through parameter learning, hyper - network optimization, and knowledge transfer;

[0009] Input the employee - job matching score matrix and the development potential index into a distributed multi - agent reinforcement learning model, construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the job matching score, the development potential index, and team collaboration data, perform iterative training based on an exploration strategy of intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, and analyze the employee performance data and team collaboration data after deployment in combination with a causal inference network, and adjust the parameters of the multi - agent reinforcement learning model in combination with a meta - policy optimization algorithm.

[0010] In an alternative embodiment,

[0011] Collect employees' historical work data, where the employees' historical work data includes work performance data, skill assessment data, and team collaboration data. Perform multi - scale feature extraction on the employees' historical work data to obtain an initial employee ability feature vector and add it to a pre - set diffusion probability model, and generate an optimized employee ability feature vector in combination with a Markov chain sampling process. Add the work performance data and the skill assessment data to a pre - set spatio - temporal hybrid network model, and extract the employee ability development trend through convolution operations. Add the employee ability development trend, the optimized employee ability feature vector, and the team collaboration data to a cross - modal contrast learning network, and solve for a comprehensive employee ability vector based on a contrast loss function and a dynamic temperature parameter and add it to a feature enhancement network. Combine a multi - scale feature pyramid and an adaptive feature aggregation module to obtain a development potential index, including:

[0012] Collect the historical work data of employees, where the historical work data of employees includes work performance data, skill assessment data, and teamwork data. Divide the historical work data of employees into a training set, a validation set, and a test set. Build a convolutional neural network model. Input the data in the training set into the convolutional neural network model, calculate the output feature vector through forward propagation, calculate the difference between the output feature vector and the actual ability level of employees based on the loss function, update the network parameters through backpropagation, use the data in the validation set to evaluate the generalization performance of the model and adjust the hyperparameters, input the historical work data of employees into the trained convolutional neural network model, and output the initial employee ability feature vector;

[0013] Build the graph structure of the diffusion probability model, use the initial employee ability feature vector as the node feature, build the edges between nodes based on the organizational structure and project cooperation information, calculate the similarity between different node feature vectors, define the transition probability matrix based on the similarity, randomly select a node as the starting node, perform random walk sampling according to the transition probability matrix, record the accessed node sequence to form a sampling path, perform average aggregation on the node feature vectors in the sampling path, and repeat the random walk sampling multiple times to obtain the distributed employee ability feature vectors after diffusion;

[0014] Use the distributed employee ability feature vectors after diffusion as the initial distribution of Markov chain sampling, calculate the conditional probability distribution between feature vectors, define the Markov chain transition matrix based on the conditional probability distribution, sample the initial state from the initial distribution, perform state transition iteratively according to the Markov chain transition matrix to form a Markov chain, repeat the Markov chain sampling multiple times to obtain a state sequence, perform average aggregation on the feature vectors in the state sequence, and output the optimized employee ability feature vector;

[0015] Build a cross-modal contrastive learning network containing three feature encoders. The three feature encoders are respectively used to process the optimized employee ability feature vector, the employee ability development trend vector, and the teamwork data. Map the three-modal data to a common feature space through the three feature encoders, introduce a dynamic temperature parameter, calculate the contrast loss between similar sample pairs and dissimilar sample pairs based on the dynamic temperature parameter, optimize the feature encoders and the dynamic temperature parameter to align different-modal features in the common space, and splice the three feature vectors to obtain the employee comprehensive ability vector;

[0016] Construct a feature enhancement network that includes a multi-scale feature pyramid and an adaptive feature aggregation module. Input the comprehensive employee ability vector into the bottom layer of the feature pyramid, generate feature maps of different scales through convolution and pooling operations, calculate the feature aggregation weights through the attention mechanism, use the adaptive feature aggregation module to perform weighted fusion on the feature maps of different scales, map the fused multi-scale features to a predetermined range through a fully connected layer, and output the employee development potential index.

[0017] In an alternative embodiment,

[0018] Calculate the contrastive loss between similar sample pairs and dissimilar sample pairs based on the dynamic temperature parameter as shown in the following formula

[0019] ;

[0020] where L represents the contrastive loss, f(x i ) represents the feature vector of sample x i , f(x j ) represents the feature vector of sample x j , r represents the dynamic temperature parameter, sim() represents the similarity metric function, f(x k ) represents the feature vector of sample x k , and k represents the sample index.

[0021] In an alternative embodiment,

[0022] Collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction. Construct a job skill knowledge graph, add the job skill knowledge graph to a dynamic heterogeneous graph neural network. The dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation, combines an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector, and adds the job skill feature matrix, the comprehensive employee ability vector, and the team structure feature vector to a neural architecture search network. Through parameter learning, hypernetwork optimization, and knowledge transfer, obtain an employee-job matching score matrix including:

[0023] Collect job description data in the human resources system and historical recruitment data in the recruitment system, perform data cleaning operations to remove redundant information and outliers in the job description data and the historical recruitment data, perform data standardization processing to construct a normalized job description corpus and a historical recruitment corpus, input the job description corpus and the historical recruitment corpus into a pre-constructed generative pre-trained language model, call the named entity recognition module to identify skill entities and knowledge entities in the job description corpus and the historical recruitment corpus, call the relation extraction module to extract the association relationships between the skill entities and the knowledge entities and the job, and establish a job skill knowledge graph with the job as the central node and the skill entities and the knowledge entities as the attribute nodes;

[0024] Construct a dynamic heterogeneous graph neural network model, input the job skill knowledge graph into the dynamic heterogeneous graph neural network model, calculate the connection weights between nodes based on the time-varying topology learning strategy, dynamically update the graph structure according to the connection weights, construct a heterogeneous subgraph sampler, input the node type information and edge type information of the job skill knowledge graph, sample and generate multiple local subgraphs based on node semantic similarity, perform graph structure distillation operations on the local subgraphs, map the job node features and skill node features to a low-dimensional feature space, and fuse the job node features and the skill node features to generate a job skill feature matrix;

[0025] Construct an adaptive graph attention network model, input employee rank information, department information, and reporting relationship information, perform hierarchical sampling operations to generate junior team subgraphs, intermediate team subgraphs, and senior team subgraphs; calculate the attention weights of the nodes in the junior team subgraphs, the intermediate team subgraphs, and the senior team subgraphs, and fuse the feature information of team members at different levels based on the attention weights to output a team structure feature vector;

[0026] Construct a neural architecture search network model, input the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector into the neural architecture search network model, call the parameter sharing module to share and optimize the network parameters of different candidate architectures, call the hypernetwork optimization module to search for the optimal matching model structure, construct a knowledge transfer module, input the historical job matching data and team collaboration data into the knowledge transfer module, extract cross-job and cross-team transfer features, and input the transfer features into the matching model for model training to output an employee-job matching degree score matrix.

[0027] In an alternative embodiment,

[0028] Construct a heterogeneous subgraph sampler that inputs the node type information and edge type information of the job skill knowledge graph, samples and generates multiple local subgraphs based on node semantic similarity, performs graph structure distillation operations on the local subgraphs, maps the job node features and skill node features to a low-dimensional feature space, and fuses the job node features and the skill node features to generate a job skill feature matrix, including:

[0029] Construct a heterogeneous subgraph sampler, where the heterogeneous subgraph sampler includes a node semantic similarity calculation module. The node semantic similarity calculation module includes a multi-head self-attention layer and a feed-forward neural network. Input the node type information and edge type information of the job skill knowledge graph into the heterogeneous subgraph sampler, calculate the dot product similarity between node feature vectors through the multi-head self-attention layer, and calculate the semantic similarity matrix between nodes based on the dot product similarity and the node type information;

[0030] Construct an adaptive sampling module. Input the semantic similarity matrix between nodes and the node degree distribution information into the adaptive sampling module. Increase the sampling weight for highly central nodes and decrease the sampling weight for marginal nodes based on the semantic similarity matrix between nodes, and perform sampling operations according to the preset job node ratio and skill node ratio to generate multiple local subgraphs;

[0031] Construct a graph structure distillation module. The graph structure distillation module includes a dual-path encoder. The dual-path encoder includes a feature encoding path and a structure encoding path. The feature encoding path includes a graph convolutional network, and the structure encoding path includes a graph isomorphism network. Input the local subgraph into the dual-path encoder, extract node local features through the graph convolutional network, extract global topological features through the graph isomorphism network, and use a gating mechanism to fuse the node local features and the global topological features to obtain node feature representations;

[0032] Construct a contrastive learning module. Input the node feature representations into the contrastive learning module. Construct positive sample pairs from the node feature representations in the original feature space and the node feature representations in the low-dimensional embedding space, and construct shuffled node feature representations as negative samples. Perform feature mapping operations based on the positive sample pairs and the negative samples, and map the job node features and the skill node features to the low-dimensional feature space respectively;

[0033] Construct a feature fusion module, where the feature fusion module includes a node-level fusion layer and a graph-level fusion layer. The node-level fusion layer includes a cross-attention mechanism, and the graph-level fusion layer includes a graph pooling module. Input the mapped job node features and skill node features into the feature fusion module. Calculate the feature importance weights between the job node features and the skill node features through the cross-attention mechanism, and aggregate the node features through the graph pooling module to obtain graph-level features. Adopt a residual connection mechanism to connect feature representations at different levels, and fuse the graph-level features and the feature representations through an adaptive feature aggregation module to generate a job-skill feature matrix.

[0034] In an alternative embodiment,

[0035] Input the employee-job matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model. Construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the job matching degree score, the development potential index, and team collaboration data, perform iterative training based on an exploration strategy with intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, and analyze the employee performance data and team collaboration data after deployment in combination with a causal inference network. Parameter adjustment of the multi-agent reinforcement learning model in combination with a meta-policy optimization algorithm includes:

[0036] Input the employee-job matching degree score matrix and the employee development potential index into the distributed multi-agent reinforcement learning model. Use a neural network to extract features from the employee-job matching degree score matrix to obtain an employee-job matching feature vector, and use a deep learning network to normalize the employee development potential index to obtain a standardized potential index.

[0037] Generate a personnel distribution state vector in the form of a high-dimensional sparse vector. The dimension of the personnel distribution state vector is the product of the number of employees and the number of jobs. Set the vector element corresponding to the employee's current job to 1, and the remaining elements to 0. Establish a team structure state matrix, and the team structure state matrix uses an adjacency matrix form to describe the reporting relationship between team members. Combine the personnel distribution state vector and the team structure state matrix to form a state space.

[0038] Generate an employee deployment plan matrix in the form of a two-dimensional matrix. The number of rows of the employee deployment plan matrix is equal to the total number of employees, and the number of columns is equal to the total number of positions. Set the matrix elements where employees are assigned to corresponding positions to 1, and set the unassigned matrix elements to 0. Use the employee deployment plan matrix as the action space;

[0039] Calculate the matching degree score by calculating the similarity between the employee position matching feature vector and the preset position requirement vector. Multiply the standardized potential index by the preset position development coefficient to obtain the potential score. Calculate the collaboration density among team members based on the team structure status matrix to obtain the collaboration score. Construct a reward function for the matching degree score, the potential score, and the collaboration score and solve to obtain the reward value;

[0040] Decompose the overall deployment task into local team deployment subtasks and cross-departmental deployment subtasks. For the local team deployment subtasks, use the deep reinforcement learning algorithm to train the agent to perform personnel deployment within the team. For the cross-departmental deployment subtasks, use the hierarchical reinforcement learning algorithm to train the agent to complete cross-departmental personnel deployment. Combine the training results of the local team deployment subtasks and the training results of the cross-departmental deployment subtasks to generate an initial deployment plan;

[0041] Use the value network prediction module to predict the long-term benefits of the initial deployment plan. Calculate the improvement value of different deployment decisions on future team performance based on the graph neural network. Use the path evaluation module to evaluate the deployment path. Calculate the cumulative benefits of different deployment plans through the Monte Carlo tree search algorithm. Select the plan with the highest evaluation value as the optimal deployment plan;

[0042] Collect the employee performance data and team collaboration data after implementing the optimal deployment plan. Input the employee performance data and the team collaboration data into the causal inference network. Analyze the causal relationship between deployment decisions and performance and collaboration through the structural equation model to generate a causal relationship diagram. Extract the key causal path to obtain the causal analysis result. According to the causal analysis result, use the gradient descent algorithm to optimize the network parameters in the distributed multi-agent reinforcement learning model, update the feature representation method of the state space, adjust the constraint conditions of the action space, and modify the calculation weight of the reward value to achieve the adaptive optimization of the model parameters.

[0043] In an alternative embodiment,

[0044] Calculating the improvement value of different deployment decisions on future team performance based on the graph neural network, using the path evaluation module to evaluate the deployment path, calculating the cumulative benefits of different deployment plans through the Monte Carlo tree search algorithm, and selecting the plan with the highest evaluation value as the optimal deployment plan includes:

[0045] Input the employee node feature vector and the position node feature vector into a pre-set graph neural network model. The employee node feature vector includes the skill features and performance data of the employee, and the position node feature vector includes the position requirement features and task description features. Based on the employee node feature vector and the position node feature vector, construct the collaborative relationship edge features and the position matching relationship edge features;

[0046] Use a message passing neural network to update the node hidden state, perform a feature aggregation operation on the employee node feature vector and the position node feature vector to generate a node representation vector, and perform an edge feature aggregation operation on the collaborative relationship edge features and the position matching relationship edge features to generate an edge representation vector;

[0047] Construct a training dataset based on the pre-acquired historical deployment data. The historical deployment data includes historical deployment plans and team performance indicators. Input the training dataset into the graph neural network model, and use the backpropagation algorithm to optimize the network parameters of the graph neural network model;

[0048] Input the initial deployment plan into the graph neural network model, calculate the team performance improvement value of each deployment decision based on the node representation vector and the edge representation vector, and generate a decision evaluation vector;

[0049] Construct a deployment decision tree, use the initial deployment plan as the root node, generate child nodes based on the employee deployment constraints, each child node represents a deployment state, and calculate the performance evaluation score of each child node based on the decision evaluation vector;

[0050] Use the upper confidence bound algorithm to select child nodes. Calculate the upper confidence bound value of each child node according to the average performance evaluation score of the child node, the number of visits to the parent node, the number of visits to the child node, and a preset exploration coefficient. Select the child node with the largest upper confidence bound value for expansion. The upper confidence bound value increases the exploration term on the basis of the average performance evaluation score of the child node. The exploration term increases with the increase of the number of visits to the parent node and decreases with the increase of the number of visits to the child node;

[0051] Perform an expansion operation on the selected child node, generate a new deployment decision based on the employee deployment constraints, and add the new deployment decision as a new child node to the deployment decision tree;

[0052] Use the graph neural network model to evaluate the performance of the newly added child node, calculate the team performance improvement value of the newly added child node, and backpropagate the team performance improvement value along the decision path to the root node to update the number of visits and the average performance evaluation score of each node on the path;

[0053] Repeat the operations of node selection, node expansion, performance evaluation, and result feedback until the preset search depth is reached, and select the decision path with the highest average performance evaluation score in the deployment decision tree as the optimal deployment plan.

[0054] In the second aspect of the embodiments of the present invention, there is provided an employee position matching and deployment system based on artificial intelligence, including:

[0055] A first unit, configured to collect historical work data of employees, where the historical work data of employees includes work performance data, skill evaluation data, and teamwork data, perform multi-scale feature extraction on the historical work data of employees to obtain an initial employee ability feature vector and add it to a pre-set diffusion probability model, generate an optimized employee ability feature vector in combination with a Markov chain sampling process, add the work performance data and the skill evaluation data to a pre-set spatio-temporal hybrid network model, extract the employee ability development trend through a convolution operation, add the employee ability development trend, the optimized employee ability feature vector, and the teamwork data to a cross-modal contrastive learning network, solve for an employee comprehensive ability vector based on a contrastive loss function and a dynamic temperature parameter and add it to a feature enhancement network, and obtain a development potential index in combination with a multi-scale feature pyramid and an adaptive feature aggregation module;

[0056] A second unit, configured to collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction, construct a job skill knowledge graph, add the job skill knowledge graph to a dynamic heterogeneous graph neural network, and the dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation, perform hierarchical sampling and dynamic aggregation in combination with an adaptive graph attention network to obtain a team structure feature vector, add the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector to a neural architecture search network, and obtain an employee-job matching degree score matrix through parameter learning, hypernetwork optimization, and knowledge transfer;

[0057] The third unit is used to input the employee-position matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model, construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the position matching degree score, the development potential index and the team collaboration data, perform iterative training based on the exploration strategy of intrinsic motivation, optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan, evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, analyze the employee performance data and the team collaboration data after deployment in combination with a causal inference network, and adjust the parameters of the multi-agent reinforcement learning model in combination with a meta-policy optimization algorithm.

[0058] In the third aspect of the embodiments of the present invention,

[0059] a kind of electronic device is provided, including:

[0060] a processor;

[0061] a memory for storing instructions executable by the processor;

[0062] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0063] In the fourth aspect of the embodiments of the present invention,

[0064] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0065] In the present invention, through technologies such as multi-scale feature extraction, spatio-temporal hybrid network model and cross-modal contrast learning, the employee capabilities are evaluated more comprehensively, and in combination with the generative pre-trained language model and the dynamic heterogeneous graph neural network, the position requirements are understood more deeply, so as to improve the matching accuracy. By using the historical work data of employees, the development trend of employee capabilities is extracted through convolution operations, and in combination with the multi-scale feature pyramid and the adaptive feature aggregation module, the development potential index is generated to more accurately predict the future development potential of employees. The distributed multi-agent reinforcement learning model is adopted, iterative training is carried out in combination with the position matching degree, the development potential and the team collaboration data, and evaluation and adjustment are carried out through the Monte Carlo tree search and the causal inference network to generate a better deployment plan, thereby improving the overall efficiency of the team. Description of the Drawings

[0066] Figure 1 It is a schematic flowchart of the method for employee position matching and deployment based on artificial intelligence according to the embodiments of the present invention;

[0067] Figure 2 This is a schematic structural diagram of an artificial intelligence-based employee position matching and allocation system according to an embodiment of the present invention. Specific implementation manners

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] The technical solutions of the present invention will be described in detail below with specific embodiments. The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0070] Figure 1 This is a flowchart of an artificial intelligence-based employee position matching and allocation method according to an embodiment of the present invention. As Figure 1 shown, the method includes:

[0071] S1. Collect historical employee work data, where the historical employee work data includes work performance data, skill assessment data, and teamwork data. Perform multi-scale feature extraction on the historical employee work data to obtain an initial employee ability feature vector and add it to a pre-set diffusion probability model. Combine the Markov chain sampling process to generate an optimized employee ability feature vector. Add the work performance data and the skill assessment data to a pre-set spatio-temporal hybrid network model, and extract the employee ability development trend through convolution operations. Add the employee ability development trend, the optimized employee ability feature vector, and the teamwork data to a cross-modal contrast learning network, and solve for an employee comprehensive ability vector based on a contrast loss function and a dynamic temperature parameter and add it to a feature enhancement network. Combine a multi-scale feature pyramid and an adaptive feature aggregation module to obtain a development potential index;

[0072] The diffusion probability model is a model based on stochastic processes, used to describe the process of information or substance propagation or diffusion in time and space. The Markov chain sampling process is a stochastic sampling method based on Markov chains, used to generate samples from complex probability distributions. The spatio-temporal hybrid network model is a deep learning model that combines spatial information and temporal information, commonly used to process spatio-temporal sequence data, such as video analysis and traffic flow prediction. The cross-modal contrastive learning network is a deep learning method for learning the associations between different modal data. By comparing the representations of different modal data, it learns a cross-modal shared feature space and is widely applied to multi-modal learning tasks such as vision-language and image-audio. The contrastive loss function is a loss function used to train the cross-modal contrastive learning network. The multi-scale feature pyramid is a network structure for extracting multi-scale features, usually used in tasks such as image segmentation and object detection.

[0073] In an alternative embodiment,

[0074] Collect historical work data of employees, where the historical work data of employees includes work performance data, skill assessment data, and teamwork data. Perform multi-scale feature extraction on the historical work data of employees to obtain an initial employee ability feature vector and add it to a pre-set diffusion probability model. Combine the Markov chain sampling process to generate an optimized employee ability feature vector. Add the work performance data and the skill assessment data to a pre-set spatio-temporal hybrid network model, and extract the employee ability development trend through convolution operations. Add the employee ability development trend, the optimized employee ability feature vector, and the teamwork data to the cross-modal contrastive learning network. Solve based on the contrastive loss function and the dynamic temperature parameter to obtain an employee comprehensive ability vector and add it to the feature enhancement network. Combine the multi-scale feature pyramid and the adaptive feature aggregation module to obtain the development potential index, including:

[0075] Collect historical work data of employees, where the historical work data of employees includes work performance data, skill assessment data, and teamwork data. Divide the historical work data of employees into a training set, a validation set, and a test set. Construct a convolutional neural network model. Input the data in the training set into the convolutional neural network model, calculate the output feature vector through forward propagation, calculate the difference between the output feature vector and the actual ability level of employees based on the loss function, update the network parameters through backpropagation. Use the data in the validation set to evaluate the generalization performance of the model and adjust the hyperparameters. Input the historical work data of employees into the trained convolutional neural network model to output an initial employee ability feature vector;

[0076] Construct the graph structure of the diffusion probability model, use the initial employee ability feature vector as the node feature, construct the edges between nodes based on the organizational structure and project cooperation information, calculate the similarity between different node feature vectors, define the transition probability matrix based on the similarity, randomly select a node as the starting node, perform random walk sampling according to the transition probability matrix, record the accessed node sequence to form a sampling path, aggregate the average values of the node feature vectors in the sampling path, repeat the random walk sampling multiple times to obtain the distribution of the diffused employee ability feature vectors;

[0077] Use the distribution of the diffused employee ability feature vectors as the initial distribution for Markov chain sampling, calculate the conditional probability distribution between feature vectors, define the Markov chain transition matrix based on the conditional probability distribution, sample the initial state from the initial distribution, perform state transitions iteratively according to the Markov chain transition matrix to form a Markov chain, repeat the Markov chain sampling multiple times to obtain a state sequence, aggregate the average values of the feature vectors in the state sequence, and output the optimized employee ability feature vector;

[0078] Construct a cross-modal contrastive learning network containing three feature encoders. The three feature encoders are respectively used to process the optimized employee ability feature vector, the employee ability development trend vector, and the team collaboration data. Map the three-modal data to a common feature space through the three feature encoders, introduce a dynamic temperature parameter, calculate the contrast loss between similar sample pairs and dissimilar sample pairs based on the dynamic temperature parameter, optimize the feature encoders and the dynamic temperature parameter to align different-modal features in the common space, and splice the three feature vectors to obtain the employee comprehensive ability vector;

[0079] Construct a feature enhancement network containing a multi-scale feature pyramid and an adaptive feature aggregation module. Input the employee comprehensive ability vector into the bottom layer of the feature pyramid, generate feature maps of different scales through convolution and pooling operations, calculate the feature aggregation weights through the attention mechanism, use the adaptive feature aggregation module to perform weighted fusion on the feature maps of different scales, map the fused multi-scale features to a predetermined range through a fully connected layer, and output the employee development potential index.

[0080] The transition probability matrix is a core concept in the Markov chain, which is used to describe the probability of transferring from one state to another. Each element of the transition probability matrix represents the probability of the current state transferring to the next state.

[0081] Collect historical work data of employees, including work performance data, such as the number of completed projects, project quality scores, customer satisfaction, etc.; skill assessment data, such as professional skill level ratings, general skill test results, etc.; and teamwork data, such as team contribution, communication and collaboration evaluations, etc. For example, collect the work performance data of an employee in the past three years, including the number of projects completed each year, project quality scores, customer satisfaction scores, etc., the skill assessment data includes the results of professional skill level ratings, and the teamwork data includes the evaluations of the employee's contribution and communication and collaboration by team members. Divide the collected data into a training set, a validation set, and a test set, with proportions of 70%, 15%, and 15% respectively.

[0082] Build a convolutional neural network model. The input is the historical work data of employees, and the output is the initial employee ability feature vector. Input the training set data into the convolutional neural network model, calculate the output feature vector through forward propagation, and calculate the difference between the output feature vector and the actual ability level of the employee based on the loss function. For example, use the mean squared error as the loss function. Update the network parameters through the backpropagation algorithm to minimize the loss function. Use the validation set data to evaluate the generalization performance of the model and adjust the hyperparameters, such as the learning rate, convolutional kernel size, etc. Input all the historical work data of employees into the trained convolutional neural network model to obtain the initial employee ability feature vector for each employee.

[0083] Build a diffusion probability model. Use the initial employee ability feature vector as the node feature, and build the edges between nodes based on the organizational structure and project cooperation information. For example, if two employees are in the same department or have collaborated on a project, establish an edge between them. Calculate the similarity between different node feature vectors, such as using cosine similarity. Define the transition probability matrix based on the similarity, and each element in the matrix represents the probability of transitioning from one node to another. Randomly select a node as the starting node, perform random walk sampling according to the transition probability matrix, and record the accessed node sequence to form a sampling path. Aggregate the node feature vectors in the sampling path by taking the average. Repeat the random walk sampling multiple times to obtain the distribution of the diffused employee ability feature vectors.

[0084] Perform Markov chain sampling. Use the distribution of the diffused employee ability feature vectors as the initial distribution for Markov chain sampling. Calculate the conditional probability distribution between the feature vectors, and define the Markov chain transition matrix based on the conditional probability distribution. Sample the initial state from the initial distribution, and perform state transitions iteratively according to the Markov chain transition matrix to form a Markov chain. Repeat the Markov chain sampling multiple times to obtain a state sequence. Aggregate the feature vectors in the state sequence by taking the average and output the optimized employee ability feature vector.

[0085] Build a cross-modal contrastive learning network, which includes three feature encoders, respectively used to process the optimized employee ability feature vector, employee ability development trend vector, and team collaboration data. The employee ability development trend vector is obtained by inputting the work performance data and skill assessment data into the spatio-temporal hybrid network model and extracting through convolutional operations. Map the three-modal data to the common feature space through their respective feature encoders. Introduce a dynamic temperature parameter, calculate the contrastive loss between similar sample pairs and dissimilar sample pairs based on the dynamic temperature parameter, and optimize the feature encoder and the dynamic temperature parameter to align different-modal features in the common space. Concatenate the three feature vectors to obtain the employee comprehensive ability vector.

[0086] Build a feature enhancement network, which includes a multi-scale feature pyramid and an adaptive feature aggregation module. Input the employee comprehensive ability vector into the bottom layer of the feature pyramid, and generate feature maps of different scales through convolutional and pooling operations. Calculate the feature aggregation weights through the attention mechanism, and use the adaptive feature aggregation module to perform weighted fusion on the feature maps of different scales. Map the fused multi-scale features to a predetermined range through a fully connected layer, and output the employee development potential index.

[0087] In this embodiment, by comprehensively considering the employee's historical work data, ability development trend, and team collaboration information, it is possible to evaluate the employee's development potential more comprehensively and accurately. Through technologies such as convolutional neural networks, diffusion probability models, and Markov chain sampling, the interference of human factors is reduced, the objectivity of the evaluation is improved, an automated evaluation process is realized, the evaluation efficiency is greatly improved, and the evaluation cost is reduced.

[0088] In an alternative embodiment,

[0089] The contrastive loss between similar sample pairs and dissimilar sample pairs is calculated based on the dynamic temperature parameter as shown in the following formula

[0090] ;

[0091] where L represents the contrastive loss, f(x i ) represents the feature vector of sample x i , f(x j ) represents the feature vector of sample x j , r represents the dynamic temperature parameter, sim() represents the similarity metric function, f(x k ) represents the feature vector of sample x k , and k represents the sample index.

[0092] where the sample x i and the sample x jDenote positive samples. Both \(i\) and \(j\) represent positive sample indices and have different values. The minimum values of both \(i\) and \(j\) are 1, and the maximum values are both the size of the current batch. Sample \(x\) k Denote negative samples. \(k\) represents the negative sample index, so the value of \(k\) is different from those of \(i\) and \(j\). The minimum value of \(k\) is 1, and the maximum value is the size of the current batch.

[0093] In this embodiment, by dynamically adjusting the temperature parameter, the contributions of similar sample pairs and dissimilar sample pairs to the loss function can be better balanced, thereby improving the accuracy of model learning. The dynamic temperature parameter can make the model more robust to noise and outliers. By using the dynamic temperature parameter, the convergence speed of the model can be accelerated, thereby reducing the training time.

[0094] S2. Collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction to construct a job skill knowledge graph. Add the job skill knowledge graph to a dynamic heterogeneous graph neural network. The dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation, and combines an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector. Add the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector to a neural architecture search network, and obtain an employee-job matching score matrix through parameter learning, hypernetwork optimization, and knowledge transfer;

[0095] The generative pre-trained language model is a natural language processing model based on unsupervised learning. By pre-training the model with a large-scale corpus, it learns the latent structure and grammar knowledge of the language. The dynamic heterogeneous graph neural network is a neural network for modeling heterogeneous graph data (including different types of nodes and edges), which can dynamically learn the spatio-temporal change characteristics of each node and edge in the graph. The time-varying topology learning is a technique for learning the topological structure changes in time series data, especially for those network structures with time changes, such as social networks, communication networks, etc. The graph structure distillation is a model compression method that extracts and learns a more concise and efficient graph structure representation from a complex graph neural network through distillation technology. The neural architecture search network is a technique for automatically designing neural network architectures. Through the search space and search strategy, the neural architecture search network can automatically explore the best network structure without manual intervention.

[0096] In an alternative embodiment,[[]]

[0097] Collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction, construct a job skill knowledge graph, add the job skill knowledge graph to a dynamic heterogeneous graph neural network, and the dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation, combines an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector, and adds the job skill feature matrix, the comprehensive employee ability vector, and the team structure feature vector to a neural architecture search network, and obtains an employee-job matching score matrix through parameter learning, hypernetwork optimization, and knowledge transfer, including:

[0098] Collect job description data in the human resources system and historical recruitment data in the recruitment system, perform data cleaning operations to remove redundant information and outliers in the job description data and the historical recruitment data, perform data standardization processing to construct a normalized job description corpus and a historical recruitment corpus, input the job description corpus and the historical recruitment corpus into a pre-constructed generative pre-trained language model, call the named entity recognition module to identify skill entities and knowledge entities in the job description corpus and the historical recruitment corpus, call the relation extraction module to extract the association relationships between the skill entities and the knowledge entities and the job, and establish a job skill knowledge graph with the job as the central node and the skill entities and the knowledge entities as attribute nodes;

[0099] Construct a dynamic heterogeneous graph neural network model, input the job skill knowledge graph into the dynamic heterogeneous graph neural network model, calculate the connection weights between nodes based on the time-varying topology learning strategy, dynamically update the graph structure according to the connection weights, construct a heterogeneous subgraph sampler, input the node type information and edge type information of the job skill knowledge graph, sample and generate multiple local subgraphs based on node semantic similarity, perform graph structure distillation operations on the local subgraphs, map the job node features and skill node features to a low-dimensional feature space, and fuse the job node features and the skill node features to generate a job skill feature matrix;

[0100] Construct an adaptive graph attention network model, input employee rank information, department information, and reporting relationship information, perform hierarchical sampling operations to generate junior team subgraphs, intermediate team subgraphs, and senior team subgraphs; calculate the attention weights of the nodes in the junior team subgraphs, the intermediate team subgraphs, and the senior team subgraphs, and fuse the feature information of team members at different levels based on the attention weights to output a team structure feature vector;

[0101] Construct a neural architecture search network model. Input the job skill feature matrix, the comprehensive employee ability vector, and the team structure feature vector into the neural architecture search network model. Invoke the parameter sharing module to share and optimize the network parameters of different candidate architectures. Invoke the hypernetwork optimization module to search for the optimal matching model structure. Construct a knowledge transfer module. Input the historical job matching data and team collaboration data into the knowledge transfer module, extract the transfer features across jobs and teams, and input the transfer features into the matching model for model training, and output the employee-job matching degree score matrix.

[0102] The named entity recognition module is a natural language processing technology used to identify entities with specific meanings (such as person names, place names, dates, etc.) from text. The local subgraph refers to a subset extracted from the overall graph, usually used to represent certain specific nodes and their direct adjacency relationships in the graph.

[0103] Collect job description data from the human resources system and historical recruitment data from the recruitment system. For example, job description data includes information such as job names, responsibility descriptions, and job requirements, and historical recruitment data includes information such as applicant resumes, interview evaluations, and employment results. Clean the collected data to remove redundant information and outliers. For example, delete duplicate job descriptions and applicant resumes lacking key information. Then, perform standardization processing on the data. For example, unify the formats of different dates and standardize the salary ranges in different units. Finally, construct a standardized job description corpus and a historical recruitment corpus.

[0104] Input the processed job description corpus and historical recruitment corpus into a pre-constructed generative pre-trained language model. For example, use the BERT model as the pre-trained language model. Use the named entity recognition module in the model to identify skill entities and knowledge entities in the corpus. For example, identify the "Java programming" skill entity from "Proficient in Java programming". At the same time, use the relation extraction module to extract the association relationships between skill entities and knowledge entities and jobs. For example, extract the association relationship between the "Java programming" skill and the "Software Engineer" job. Finally, establish a job skill knowledge graph with the job as the central node and skill entities and knowledge entities as attribute nodes. For example, the "Software Engineer" job is connected to skill nodes such as "Java programming" and "Software development".

[0105] Build a dynamic heterogeneous graph neural network model and input the constructed job-skill knowledge graph into this model. Calculate the connection weights between nodes based on the time-varying topology learning strategy. For example, dynamically adjust the connection weights between skill nodes and job nodes according to the co-occurrence frequency and importance of skills. Dynamically update the graph structure according to the calculated connection weights. Build a heterogeneous subgraph sampler and input the node type information and edge type information of the job-skill knowledge graph. For example, input job nodes, skill nodes, and their connection relationships. Sample and generate multiple local subgraphs based on node semantic similarity. For example, combine job nodes with similar skill requirements and their related skill nodes into a subgraph. Perform graph structure distillation operations on the generated local subgraphs to map job node features and skill node features to a low-dimensional feature space. For example, use a graph convolutional network to extract node features and perform dimensionality reduction. Fuse job node features and skill node features to generate a job-skill feature matrix.

[0106] Build an adaptive graph attention network model. Input employee rank information, department information, and reporting relationship information. For example, input information such as the job level of employees, their affiliated departments, and their direct superiors. Perform hierarchical sampling operations to generate junior team subgraphs, intermediate team subgraphs, and senior team subgraphs. For example, form junior team subgraphs with grass-roots employees in the same department, form intermediate team subgraphs with department managers and grass-roots employees, and form senior team subgraphs with higher-level managers and subordinate department managers. Calculate the attention weights of nodes in the junior team subgraphs, intermediate team subgraphs, and senior team subgraphs. For example, calculate their attention weights according to the contribution and influence of employees in the team. Based on the calculated attention weights, fuse the feature information of team members at different levels and output a team structure feature vector.

[0107] Build a neural architecture search network model. Input the job-skill feature matrix, the comprehensive ability vector of employees, and the team structure feature vector into this model. Call the parameter sharing module to share and optimize the network parameters of different candidate architectures. Call the hypernetwork optimization module to search for the optimal matching model structure. Build a knowledge transfer module and input historical job matching data and team collaboration data into the knowledge transfer module. For example, input past successful employee-job matching cases and efficient team collaboration cases. Extract transfer features across jobs and teams. For example, extract the general skill requirements between different jobs and the collaboration patterns between different teams. Input the extracted transfer features into the matching model for model training. Finally, output an employee-job matching degree score matrix.

[0108] In this embodiment, by constructing a job skill knowledge graph and a dynamic heterogeneous graph neural network, the job skill requirements and employee capabilities can be understood more accurately, thereby improving the matching accuracy. Through the adaptive graph attention network model, the interaction relationships and team structure characteristics among team members can be better captured, thereby optimizing the team structure and improving team efficiency. Through the neural architecture search network model, the optimal matching model structure can be automatically searched, and combined with knowledge transfer learning, intelligent employee-job matching can be achieved, reducing labor costs.

[0109] In an alternative embodiment,

[0110] Construct a heterogeneous subgraph sampler, input the node type information and edge type information of the job skill knowledge graph, sample and generate multiple local subgraphs based on node semantic similarity, perform graph structure distillation operations on the local subgraphs, map the job node features and skill node features to a low-dimensional feature space, and fuse the job node features and the skill node features to generate a job skill feature matrix, including:

[0111] Construct a heterogeneous subgraph sampler, the heterogeneous subgraph sampler includes a node semantic similarity calculation module, the node semantic similarity calculation module includes a multi-head self-attention layer and a feed-forward neural network, input the node type information and edge type information of the job skill knowledge graph into the heterogeneous subgraph sampler, calculate the dot product similarity between node feature vectors through the multi-head self-attention layer, and calculate the inter-node semantic similarity matrix based on the dot product similarity and the node type information;

[0112] Construct an adaptive sampling module, input the inter-node semantic similarity matrix and node degree distribution information into the adaptive sampling module, increase the sampling weight for high-centrality nodes and decrease the sampling weight for marginal nodes based on the inter-node semantic similarity matrix, and perform sampling operations according to the preset job node ratio and skill node ratio to generate multiple local subgraphs;

[0113] Construct a graph structure distillation module, the graph structure distillation module includes a dual-path encoder, the dual-path encoder includes a feature encoding path and a structure encoding path, where the feature encoding path includes a graph convolutional network, and the structure encoding path includes a graph isomorphism network. Input the local subgraph into the dual-path encoder, extract node local features through the graph convolutional network, extract global topological features through the graph isomorphism network, and fuse the node local features and the global topological features using a gating mechanism to obtain node feature representations;

[0114] Construct a contrastive learning module. Input the node feature representation into the contrastive learning module. Construct positive sample pairs from the node feature representation in the original feature space and the node feature representation in the low-dimensional embedding space, and construct shuffled node feature representations as negative samples. Perform a feature mapping operation based on the positive sample pairs and the negative samples, and map the job node features and the skill node features to the low-dimensional feature space respectively.

[0115] Construct a feature fusion module. The feature fusion module includes a node-level fusion layer and a graph-level fusion layer. The node-level fusion layer includes a cross-attention mechanism, and the graph-level fusion layer includes a graph pooling module. Input the mapped job node features and skill node features into the feature fusion module. Calculate the feature importance weights between the job node features and the skill node features through the cross-attention mechanism, and aggregate the node features through the graph pooling module to obtain graph-level features. Use a residual connection mechanism to connect feature representations at different levels, and fuse the graph-level features and the feature representations through an adaptive feature aggregation module to generate a job-skill feature matrix.

[0116] The dot product similarity is a method for calculating the similarity between two vectors. By calculating the dot product of two vectors, the degree of similarity in the same direction can be measured. The dual-path encoder is a deep learning model architecture that processes different features or information of the input data through two parallel paths (usually two different neural networks or modules). The positive sample refers to a sample that matches the target category or query condition in a machine learning or information retrieval task, and the negative sample refers to a sample that does not match the target category or query condition in a machine learning or information retrieval task.

[0117] Construct a heterogeneous subgraph sampler, which contains a node semantic similarity calculation module composed of a multi-head self-attention layer and a feed-forward neural network. Input the node type information (e.g., job nodes, skill nodes) and edge type information (e.g., has, needs) of the job-skill knowledge graph into the sampler. The multi-head self-attention layer calculates the dot product similarity between node feature vectors, and combines the node type information to calculate the semantic similarity matrix between nodes. For example, if two nodes are both skill nodes and their feature vector similarities are high, then their semantic similarities will also be high.

[0118] Build an adaptive sampling module. Input the semantic similarity matrix between nodes and the node degree distribution information into this module. According to the semantic similarity matrix between nodes, increase the sampling weight for highly central nodes (e.g., job nodes with many skills or skill nodes required by many jobs), and decrease the sampling weight for peripheral nodes. Perform sampling operations to generate multiple local subgraphs according to the preset ratios of job nodes and skill nodes (e.g., job nodes:skill nodes = 1:3). For example, around a specific job node, sample skill nodes with high semantic similarity to it and other job nodes related to these skill nodes to form a local subgraph.

[0119] Build a graph structure distillation module, which includes a dual-path encoder composed of a feature encoding path and a structure encoding path. The feature encoding path uses a graph convolutional network to extract node local features. The structure encoding path uses a graph isomorphism network to extract global topological features. Input the local subgraph into the dual-path encoder to extract node local features and global topological features respectively. Adopt a gating mechanism to fuse these two types of features to obtain node feature representations. For example, the gating mechanism can dynamically adjust their weights according to the importance of the local features and global topological features of the nodes.

[0120] Build a contrastive learning module. Input the node feature representations into this module. Construct positive sample pairs from the node feature representations in the original feature space and the node feature representations in the low-dimensional embedding space. Construct shuffled node feature representations as negative samples. Perform feature mapping operations based on the positive sample pairs and negative samples to map job node features and skill node features to the low-dimensional feature space respectively. For example, a fully connected layer can be used to map high-dimensional features to the low-dimensional space.

[0121] Build a feature fusion module, which includes a node-level fusion layer and a graph-level fusion layer. The node-level fusion layer uses a cross-attention mechanism to calculate the feature importance weights between job node features and skill node features. The graph-level fusion layer uses a graph pooling module to aggregate node features to obtain graph-level features. Adopt a residual connection mechanism to connect feature representations at different levels. Fuse the graph-level features and feature representations through an adaptive feature aggregation module to generate a job-skill feature matrix. For example, the adaptive feature aggregation module can dynamically adjust their weights according to the importance of features at different levels. Suppose a local subgraph contains one job node and three skill nodes. After feature extraction and fusion, a vector containing job features and skill features can be obtained, and this vector is the job-skill feature representation of this local subgraph. Combine the job-skill feature representations of all local subgraphs to obtain the final job-skill feature matrix.

[0122] In this embodiment, through heterogeneous subgraph sampling, rich local structural information in the job skill knowledge graph can be captured. Through graph structure distillation, complex graph structure information can be compressed into a low-dimensional feature space, thereby improving the model's expressive ability. The contrastive learning module can learn more discriminative node feature representations, thereby enhancing the model's generalization ability. Mapping high-dimensional features to a low-dimensional space can reduce the model's computational complexity and improve the model's training efficiency.

[0123] S3. Input the employee-job matching degree score matrix and the development potential index into the distributed multi-agent reinforcement learning model, construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the job matching degree score, the development potential index, and the team collaboration data, perform iterative training based on the exploration strategy of intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, and analyze the employee performance data and team collaboration data after deployment in combination with the causal inference network. Adjust the parameters of the multi-agent reinforcement learning model in combination with the meta-policy optimization algorithm.

[0124] The distributed multi-agent reinforcement learning model is a reinforcement learning method in a multi-agent system, where multiple agents cooperate or compete in a shared environment. Each agent maximizes its individual or global reward by learning interaction strategies with other agents and the environment. The exploration strategy of intrinsic motivation is a reinforcement learning strategy that promotes exploration by providing intrinsic motivation to the agent. The Monte Carlo tree search model is a search algorithm for decision-making problems, commonly used in fields such as chess, games, and robots. It evaluates different nodes in the decision tree by simulating random game processes and selects the optimal action. The path evaluation module is a module for evaluating the performance and effect of an agent or system on a given path. The causal inference network is a network based on causal relationship modeling, aiming to infer the behavior or influence of a system by observing and analyzing the causal relationships between variables.

[0125] In an alternative embodiment,

[0126] Input the employee-position matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model, construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the position matching degree score, the development potential index and the team collaboration data, perform iterative training based on the exploration strategy of intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, and analyze the employee performance data and team collaboration data after deployment in combination with a causal inference network. Adjust the parameters of the multi-agent reinforcement learning model in combination with a meta-policy optimization algorithm, including:

[0127] Input the employee-position matching degree score matrix and the employee development potential index into a distributed multi-agent reinforcement learning model. Use a neural network to extract features from the employee-position matching degree score matrix to obtain an employee position matching feature vector, and use a deep learning network to normalize the employee development potential index to obtain a standardized potential index;

[0128] Generate a personnel distribution state vector in the form of a high-dimensional sparse vector, where the dimension of the personnel distribution state vector is the product of the number of employees and the number of positions. Set the vector element corresponding to the position where the employee is currently located to 1, and the remaining elements to 0. Establish a team structure state matrix, and the team structure state matrix uses an adjacency matrix form to describe the reporting relationship between team members. Combine the personnel distribution state vector and the team structure state matrix to form a state space;

[0129] Generate an employee deployment plan matrix in the form of a two-dimensional matrix, where the number of rows of the employee deployment plan matrix is equal to the total number of employees, and the number of columns is equal to the total number of positions. Set the matrix element where the employee is assigned to the corresponding position to 1, and the unassigned matrix element to 0. Use the employee deployment plan matrix as the action space;

[0130] Calculate the similarity between the employee position matching feature vector and a preset position requirement vector to obtain a matching degree score, multiply the standardized potential index by a preset position development coefficient to obtain a potential score, calculate the collaboration density between team members based on the team structure state matrix to obtain a collaboration score, construct a reward function for the matching degree score, the potential score and the collaboration score, and solve to obtain a reward value;

[0131] Decompose the overall deployment task into local team deployment subtasks and cross - departmental deployment subtasks. For the local team deployment subtasks, use a deep reinforcement learning algorithm to train an agent to perform personnel deployment within the team. For the cross - departmental deployment subtasks, use a hierarchical reinforcement learning algorithm to train an agent to complete cross - departmental personnel deployment. Combine the training results of the local team deployment subtasks and the cross - departmental deployment subtasks to generate an initial deployment plan;

[0132] Use a value network prediction module to predict the long - term benefits of the initial deployment plan, calculate the improvement value of different deployment decisions on future team performance based on a graph neural network, use a path evaluation module to evaluate the deployment path, calculate the cumulative benefits of different deployment plans through the Monte Carlo tree search algorithm, and select the plan with the highest evaluation value as the optimal deployment plan;

[0133] Collect the employee performance data and team collaboration data after implementing the optimal deployment plan, input the employee performance data and the team collaboration data into a causal inference network, analyze the causal relationship between deployment decisions and performance, collaboration through a structural equation model to generate a causal relationship graph, extract the key causal path to obtain the causal analysis result, and according to the causal analysis result, use the gradient descent algorithm to optimize the network parameters in the distributed multi - agent reinforcement learning model, update the feature representation method of the state space, adjust the constraint conditions of the action space, and modify the calculation weight of the reward value to achieve the adaptive optimization of the model parameters.

[0134] The high - dimensional sparse vector form is a vector representation method in which the representation dimension of the data is high, but most of its elements are zero (sparse).

[0135] Collect the data of employees' job matching degree scores and development potential index. For example, obtain the matching degree scores of each employee for each job through an evaluation system to form a matrix. For instance, the matching degree score of employee A for job 1 is 0.8, and for job 2 is 0.6, etc. At the same time, obtain the development potential index of each employee. For example, the potential index of employee A is 90 points, and that of employee B is 85 points, etc. Input these data into the distributed multi - agent reinforcement learning model.

[0136] Pre - process the input data. Use a neural network to extract the features of the employee job matching degree score matrix to obtain the job matching feature vector of each employee. For example, the feature vector of employee A may be [0.75, 0.2, 0.05], indicating that they are more suitable for technical positions. At the same time, use a deep learning network to normalize the employee development potential index to obtain the standardized potential index. For example, normalize 90 points and 85 points to 0.9 and 0.85 respectively.

[0137] Construct the state space. Represent the personnel distribution state as a high-dimensional sparse vector, where the dimension of the vector is the number of employees multiplied by the number of positions. If an employee is in a certain position, the corresponding vector element value is 1; otherwise, it is 0. For example, if there are 3 employees and 3 positions, and employee A is in position 1, the vector is [1, 0, 0, 0, 0, 0, 0, 0, 0]. At the same time, construct the team structure state matrix, and use the adjacency matrix to describe the reporting relationship between team members. For example, if employee A reports to employee B, the corresponding element values of A and B in the matrix are 1. Combine the personnel distribution state vector and the team structure state matrix to form the state space.

[0138] Construct the action space. Use a two-dimensional matrix to represent the employee deployment plan. The number of rows of the matrix is equal to the total number of employees, and the number of columns is equal to the total number of positions. If an employee is assigned to a certain position, the corresponding matrix element value is 1; otherwise, it is 0.

[0139] Construct the reward function. Calculate the similarity between the employee's position matching feature vector and the preset position requirement vector to obtain the matching score. Multiply the standardized potential index by the preset position development coefficient to obtain the potential score. Calculate the collaboration density between team members based on the team structure state matrix to obtain the collaboration score. Perform weighted summation on the matching score, potential score, and collaboration score to obtain the reward value.

[0140] Conduct model training. Decompose the overall deployment task into local team deployment subtasks and cross-departmental deployment subtasks. For local team deployment subtasks, use the deep reinforcement learning algorithm to train the agent to perform personnel deployment within the team. For cross-departmental deployment subtasks, use the hierarchical reinforcement learning algorithm to train the agent to complete cross-departmental personnel deployment. Combine the training results of local team deployment subtasks and cross-departmental deployment subtasks to generate an initial deployment plan.

[0141] Evaluate the initial deployment plan. Use the value network prediction module to predict the long-term benefits of the initial deployment plan. For example, predict the improvement value of the team performance in the next three months. Calculate the impact of different deployment decisions on the future team performance based on the graph neural network. Use the path evaluation module to evaluate the deployment path and calculate the cumulative benefits of different deployment plans. Select the plan with the highest evaluation value as the optimal deployment plan through the Monte Carlo tree search algorithm.

[0142] Execute the optimal deployment plan and optimize the model. Collect the employee performance data and team collaboration data after the implementation of the plan. Input these data into the causal inference network, analyze the causal relationship between the deployment decision and performance / collaboration, and generate a causal relationship diagram. According to the results of the causal analysis, use the gradient descent algorithm to optimize the network parameters in the distributed multi-agent reinforcement learning model, update the feature representation method of the state space, adjust the constraint conditions of the action space, and modify the calculation weights of the reward values to achieve the adaptive optimization of the model parameters.

[0143] In this embodiment, the deployment plan is automatically generated by an intelligent algorithm, reducing manual intervention, which helps to improve the personnel deployment efficiency. The deployment is based on multi-dimensional factors such as job matching degree, development potential, and team collaboration, which helps to achieve the best allocation of personnel, thereby improving the overall performance of the team. Considering the development potential of employees helps to provide more suitable positions and development opportunities for employees and promote the personal growth of employees.

[0144] In an alternative embodiment,

[0145] Calculating the improvement value of different deployment decisions on the future team performance based on the graph neural network, evaluating the deployment path using a path evaluation module, calculating the cumulative benefits of different deployment plans through the Monte Carlo tree search algorithm, and selecting the plan with the highest evaluation value as the optimal deployment plan includes:

[0146] Input the employee node feature vector and the job node feature vector into a pre-set graph neural network model. The employee node feature vector includes the skill features and performance data of the employee, and the job node feature vector includes the job requirement features and task description features. Based on the employee node feature vector and the job node feature vector, construct the collaborative relationship edge feature and the job matching relationship edge feature;

[0147] Use the message passing neural network to update the node hidden state, perform a feature aggregation operation on the employee node feature vector and the job node feature vector to generate a node representation vector, and perform an edge feature aggregation operation on the collaborative relationship edge feature and the job matching relationship edge feature to generate an edge representation vector;

[0148] Construct a training data set based on the pre-obtained historical deployment data. The historical deployment data includes historical deployment plans and team performance indicators. Input the training data set into the graph neural network model, and use the backpropagation algorithm to optimize the network parameters of the graph neural network model;

[0149] Input the initial deployment plan into the graph neural network model, calculate the team performance improvement value of each deployment decision based on the node representation vector and the edge representation vector, and generate a decision evaluation vector;

[0150] Construct a deployment decision tree, take the initial deployment plan as the root node, generate child nodes based on the employee deployment constraints, each child node represents a deployment state, and calculate the performance evaluation score of each child node based on the decision evaluation vector;

[0151] Use the Upper Confidence Bound (UCB) algorithm to select child nodes. Calculate the UCB value of each child node according to the average performance evaluation score of the child node, the number of visits to the parent node, the number of visits to the child node, and a preset exploration coefficient. Select the child node with the maximum UCB value for expansion. The UCB value adds an exploration term to the average performance evaluation score of the child node. The exploration term increases with the increase in the number of visits to the parent node and decreases with the increase in the number of visits to the child node;

[0152] Perform an expansion operation on the selected child node, generate a new deployment decision based on the employee deployment constraints, and add the new deployment decision as a new child node to the deployment decision tree;

[0153] Use the graph neural network model to evaluate the performance of the newly added child node, calculate the team performance improvement value of the newly added child node, and backpropagate the team performance improvement value along the decision path to the root node to update the number of visits and the average performance evaluation score of each node on the path;

[0154] Repeat the operations of node selection, node expansion, performance evaluation, and result backpropagation until the preset search depth is reached. Select the decision path with the highest average performance evaluation score in the deployment decision tree as the optimal deployment plan.

[0155] The message passing neural network is a graph neural network model for processing graph-structured data. The Upper Confidence Bound algorithm is a strategy for balancing exploration and exploitation, commonly used in reinforcement learning, optimal decision-making, and multi-armed bandit problems. The decision path refers to the sequence of a series of steps or actions in the decision-making process, usually used to describe the specific process experienced by an agent or algorithm when making decisions.

[0156] Collect employee information, including skills, past performance, etc., to form an employee feature vector. At the same time, collect job information, such as job responsibilities, required skills, etc., to form a job feature vector. In addition, it is necessary to collect collaborative relationship data between team members and matching data between employees and positions to construct collaborative relationship edge features and job matching relationship edge features. For example, Xiao Zhang is good at programming and communication, and has excellent performance. This information constitutes his employee feature vector. The programming engineer position requires proficiency in Java and Python and strong communication skills. This information constitutes the feature vector of the position. Xiao Zhang and Xiao Li often cooperate and have a high degree of tacit cooperation, which constitutes the collaborative relationship edge feature between them. Xiao Zhang's skills are highly matched with the requirements of the programming engineer position, which constitutes the job matching relationship edge feature between them.

[0157] The collected employee feature vectors, job feature vectors, collaboration relationship edge features, and job matching relationship edge features are input into the graph neural network. The graph neural network aggregates these features to generate node representation vectors and edge representation vectors. For example, through the message passing mechanism, Xiao Zhang's node representation vector will incorporate the information of his neighbor nodes (colleagues with whom he collaborates) and the information of the edges connected to him (matching relationships with positions).

[0158] The graph neural network model is trained using historical deployment data, including past deployment plans and corresponding team performance indicators. The model parameters are continuously adjusted through the back-propagation algorithm so that the model can accurately predict the impact of different deployment plans on team performance. For example, after Xiao Zhang was transferred to the programming engineer position in the past, the team performance increased by 10%. This data is input into the model for training, so that the model can learn the positive impact of this deployment relationship on performance.

[0159] After the model training is completed, it is used to evaluate new deployment plans. The initial deployment plan is input into the trained graph neural network model. The model will calculate the improvement value of each deployment decision on team performance based on the node representation vector and edge representation vector, and generate a decision evaluation vector. For example, if Xiao Wang is transferred from the testing position to the development position, the model predicts that the team performance will improve by 5%, and this value is recorded in the decision evaluation vector.

[0160] The initial deployment plan is used as the root node of the decision tree. Different child nodes are generated according to the employee deployment constraints, such as budget constraints, staffing, etc. Each child node represents a possible deployment state. The performance evaluation score of each child node is calculated using the decision evaluation vector. For example, the root node represents the current team staffing, one child node represents transferring Xiao Wang to the development position, and another child node represents transferring Xiao Li to the product manager position.

[0161] The Monte Carlo tree search algorithm is adopted, and the Upper Confidence Bound (UCB) algorithm is used to select child nodes for expansion. This algorithm comprehensively considers the average performance evaluation score of child nodes, the visit counts of parent and child nodes, and a preset exploration coefficient, and selects the child node with the maximum UCB value for expansion. For example, although the current average performance evaluation score of a child node is not high, but its visit count is small and it still has great potential, the UCB algorithm will preferentially select such nodes for exploration.

[0162] Expand the selected child nodes to generate new deployment decisions, and add these decisions as new child nodes to the decision tree. For example, expand the child node that selects to transfer Xiao Wang to the development position to generate new child nodes, such as transferring Xiao Li to the product manager position or keeping Xiao Li's position unchanged.

[0163] Use the graph neural network model to evaluate the performance of the newly added child nodes, calculate the value of the team performance improvement, and backpropagate the results to the root node to update the visit counts and average performance evaluation scores of each node on the path.

[0164] Repeat the operations of node selection, node expansion, performance evaluation, and result backpropagation until the preset search depth is reached. Finally, select the decision path with the highest average performance evaluation score in the decision tree as the optimal deployment plan.

[0165] In this embodiment, by using the graph neural network model, the impact of different deployment plans on team performance can be predicted more accurately, avoiding the limitations of traditional methods that rely on subjective judgment. The Monte Carlo tree search algorithm can search for the optimal deployment plan globally, avoiding local optimal solutions, so as to find the plan that maximally improves team performance. By taking employee characteristics, position characteristics, and the relationships between them as inputs, the model can more comprehensively consider the impact of personnel relationships and position matching on team performance, and thus formulate a more reasonable deployment plan.

[0166] Figure 2 For the structural schematic diagram of the employee position matching and deployment system based on artificial intelligence according to the embodiment of the present invention, as Figure 2 shown, the system includes:

[0167] The first unit is used to collect historical work data of employees. Among them, the historical work data of employees includes work performance data, skill assessment data, and teamwork data. Multi-scale feature extraction is performed on the historical work data of employees to obtain an initial employee ability feature vector, which is added to a pre-set diffusion probability model. Combining with the Markov chain sampling process to generate an optimized employee ability feature vector. Add the work performance data and the skill assessment data to a pre-set spatio-temporal hybrid network model, and extract the development trend of employee ability through convolution operation. Add the development trend of employee ability, the optimized employee ability feature vector, and the teamwork data to a cross-modal contrastive learning network, and solve to obtain an employee comprehensive ability vector based on a contrastive loss function and a dynamic temperature parameter, which is added to a feature enhancement network. Combining a multi-scale feature pyramid and an adaptive feature aggregation module to obtain a development potential index;

[0168] The second unit is used to collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction, construct a job skill knowledge graph, and add the job skill knowledge graph to a dynamic heterogeneous graph neural network. The dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation. Combining with an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector. Add the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector to a neural architecture search network, and obtain an employee-job matching degree score matrix through parameter learning, hypernetwork optimization, and knowledge transfer;

[0169] The third unit is used to input the employee-job matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model, construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the job matching degree score, the development potential index, and the teamwork data, perform iterative training based on an exploration strategy of intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, and analyze the employee performance data and teamwork data after deployment in combination with a causal inference network. Combine a meta-policy optimization algorithm to adjust the parameters of the multi-agent reinforcement learning model.

[0170] In the third aspect of the embodiments of the present invention,

[0171] Provide an electronic device, including:

[0172] A processor;

[0173] A memory for storing processor-executable instructions;

[0174] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0175] In the fourth aspect of the embodiments of the present invention,

[0176] A computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0177] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for performing various aspects of the present invention are loaded.

[0178] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based method for employee position matching and deployment, characterized in that, Including: Collect historical work data of employees, where the historical work data of employees includes work performance data, skill assessment data, and teamwork data. Perform multi-scale feature extraction on the historical work data of employees to obtain an initial employee ability feature vector and add it to a pre-set diffusion probability model. Combine the Markov chain sampling process to generate an optimized employee ability feature vector. Add the work performance data and the skill assessment data to a pre-set spatio-temporal hybrid network model, and extract the development trend of employee ability through convolution operations. Add the development trend of employee ability, the optimized employee ability feature vector, and the teamwork data to a cross-modal contrastive learning network, and solve for the comprehensive employee ability vector based on the contrast loss function and dynamic temperature parameter and add it to a feature enhancement network. Combine a multi-scale feature pyramid and an adaptive feature aggregation module to obtain a development potential index, including: Collect historical work data of employees, where the historical work data of employees includes work performance data, skill assessment data, and teamwork data. Divide the historical work data of employees into a training set, a validation set, and a test set. Build a convolutional neural network model, input the data in the training set into the convolutional neural network model, calculate the output feature vector through forward propagation, calculate the difference between the output feature vector and the actual ability level of employees based on the loss function, update the network parameters through backpropagation, use the data in the validation set to evaluate the generalization performance of the model and adjust the hyperparameters, and input the historical work data of employees into the trained convolutional neural network model to output an initial employee ability feature vector; Build the graph structure of the diffusion probability model, use the initial employee ability feature vector as the node feature, build the edges between nodes based on the organizational structure and project cooperation information, calculate the similarity between different node feature vectors, define the transition probability matrix based on the similarity, randomly select a node as the starting node, perform random walk sampling according to the transition probability matrix, record the visited node sequence to form a sampling path, perform average aggregation on the node feature vectors in the sampling path, and repeat the random walk sampling multiple times to obtain the distribution of the diffused employee ability feature vectors; Use the distribution of the diffused employee ability feature vectors as the initial distribution of Markov chain sampling, calculate the conditional probability distribution between feature vectors, define the Markov chain transition matrix based on the conditional probability distribution, sample the initial state from the initial distribution, perform state transitions iteratively according to the Markov chain transition matrix to form a Markov chain, repeat the Markov chain sampling multiple times to obtain a state sequence, perform average aggregation on the feature vectors in the state sequence, and output the optimized employee ability feature vector; Construct a cross-modal contrastive learning network containing three feature encoders, which are respectively used to process the optimized employee ability feature vectors, employee ability development trend vectors, and team collaboration data. Map the three-modal data to a common feature space through the three feature encoders, introduce a dynamic temperature parameter, calculate the contrastive loss between similar sample pairs and dissimilar sample pairs based on the dynamic temperature parameter, optimize the feature encoders and the dynamic temperature parameter to align different-modal features in the common space, and splice the three feature vectors to obtain an employee comprehensive ability vector; Construct a feature enhancement network containing a multi-scale feature pyramid and an adaptive feature aggregation module. Input the employee comprehensive ability vector into the bottom layer of the feature pyramid, generate feature maps of different scales through convolution and pooling operations, calculate the feature aggregation weights through the attention mechanism, use the adaptive feature aggregation module to perform weighted fusion on the feature maps of different scales, map the fused multi-scale features to a predetermined range through a fully connected layer, and output the employee development potential index; The contrastive loss between similar sample pairs and dissimilar sample pairs is calculated based on the dynamic temperature parameter as shown in the following formula: Among them, L represents the contrastive loss, f(x i ) represents the feature vector of the sample x i , f(x j ) represents the feature vector of the sample x j , r represents the dynamic temperature parameter, sim() represents the similarity metric function, f(x k ) represents the feature vector of the sample x k , and k represents the sample index; Collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction. Construct a job skill knowledge graph, add the job skill knowledge graph to a dynamic heterogeneous graph neural network. The dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation, combines an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector, and adds the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector to a neural architecture search network to obtain an employee-job matching score matrix through parameter learning, hypernetwork optimization, and knowledge transfer; Input the employee-job matching score matrix and the development potential index into a distributed multi-agent reinforcement learning model. Construct a state space containing the current personnel distribution state and team structure state, construct an action space based on the deployment plan space, construct a reward function based on the job matching score, development potential index, and team collaboration data, perform iterative training based on the exploration strategy of intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, analyze the employee performance data and team collaboration data after deployment in combination with a causal inference network, and adjust the parameters of the multi-agent reinforcement learning model in combination with a meta-policy optimization algorithm.

2. The method according to claim 1, wherein Collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction, construct a job skill knowledge graph, add the job skill knowledge graph to a dynamic heterogeneous graph neural network, and the dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation. Combine with an adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector. Add the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector to a neural architecture search network, and obtain an employee-job matching score matrix through parameter learning, hypernetwork optimization, and knowledge transfer, including: Collect job description data in the human resources system and historical recruitment data in the recruitment system, perform data cleaning operations to remove redundant information and outliers in the job description data and the historical recruitment data, perform data standardization processing to construct a normalized job description corpus and a historical recruitment corpus, input the job description corpus and the historical recruitment corpus into a pre-constructed generative pre-trained language model, call the named entity recognition module to identify skill entities and knowledge entities in the job description corpus and the historical recruitment corpus, call the relationship extraction module to extract the association relationships between the skill entities and the knowledge entities and the job, and establish a job skill knowledge graph with the job as the central node and the skill entities and the knowledge entities as attribute nodes; Construct a dynamic heterogeneous graph neural network model, input the job skill knowledge graph into the dynamic heterogeneous graph neural network model, calculate the connection weights between nodes based on the time-varying topology learning strategy, dynamically update the graph structure according to the connection weights, construct a heterogeneous subgraph sampler, input the node type information and edge type information of the job skill knowledge graph, sample and generate multiple local subgraphs based on node semantic similarity, perform graph structure distillation operations on the local subgraphs, map the job node features and skill node features to a low-dimensional feature space, and fuse the job node features and the skill node features to generate a job skill feature matrix; Construct an adaptive graph attention network model, input employee rank information, department information, and reporting relationship information, perform hierarchical sampling operations to generate junior team subgraphs, intermediate team subgraphs, and senior team subgraphs; calculate the attention weights of the nodes in the junior team subgraphs, the intermediate team subgraphs, and the senior team subgraphs, fuse the feature information of team members at different levels based on the attention weights, and output a team structure feature vector; Construct a neural architecture search network model. Input the job skill feature matrix, the comprehensive employee ability vector, and the team structure feature vector into the neural architecture search network model. Invoke the parameter sharing module to share and optimize the network parameters of different candidate architectures. Invoke the hypernetwork optimization module to search for the optimal matching model structure. Construct a knowledge transfer module. Input the historical job matching data and team collaboration data into the knowledge transfer module, extract the transfer features across jobs and teams, and input the transfer features into the matching model for model training to output the employee-job matching degree score matrix.

3. The method according to claim 2, wherein Construct a heterogeneous subgraph sampler. Input the node type information and edge type information of the job skill knowledge graph. Sample and generate multiple local subgraphs based on node semantic similarity. Perform graph structure distillation operations on the local subgraphs, map the job node features and skill node features to a low-dimensional feature space, and fuse the job node features and the skill node features to generate a job skill feature matrix, including: Construct a heterogeneous subgraph sampler. The heterogeneous subgraph sampler includes a node semantic similarity calculation module. The node semantic similarity calculation module includes a multi-head self-attention layer and a feed-forward neural network. Input the node type information and edge type information of the job skill knowledge graph into the heterogeneous subgraph sampler. Calculate the dot product similarity between node feature vectors through the multi-head self-attention layer, and calculate the node-to-node semantic similarity matrix based on the dot product similarity and the node type information. Construct an adaptive sampling module. Input the node-to-node semantic similarity matrix and the node degree distribution information into the adaptive sampling module. Increase the sampling weight for highly central nodes and decrease the sampling weight for marginal nodes based on the node-to-node semantic similarity matrix. Perform sampling operations according to the preset job node ratio and skill node ratio to generate multiple local subgraphs. Construct a graph structure distillation module. The graph structure distillation module includes a dual-path encoder. The dual-path encoder includes a feature encoding path and a structure encoding path. The feature encoding path includes a graph convolutional network, and the structure encoding path includes a graph isomorphism network. Input the local subgraph into the dual-path encoder. Extract node local features through the graph convolutional network, extract global topological features through the graph isomorphism network, and use a gating mechanism to fuse the node local features and the global topological features to obtain node feature representations. Construct a contrastive learning module. Input the node feature representations into the contrastive learning module. Construct positive sample pairs from the node feature representations in the original feature space and the node feature representations in the low-dimensional embedding space, and construct shuffled node feature representations as negative samples. Perform feature mapping operations based on the positive sample pairs and the negative samples, and map the job node features and the skill node features to the low-dimensional feature space respectively. Construct a feature fusion module, where the feature fusion module includes a node-level fusion layer and a graph-level fusion layer. The node-level fusion layer includes a cross-attention mechanism, and the graph-level fusion layer includes a graph pooling module. Input the mapped job node features and skill node features into the feature fusion module. Calculate the feature importance weights between the job node features and the skill node features through the cross-attention mechanism, and aggregate the node features through the graph pooling module to obtain graph-level features. Use a residual connection mechanism to connect feature representations at different levels, and fuse the graph-level features and the feature representations through an adaptive feature aggregation module to generate a job-skill feature matrix.

4. The method according to claim 1, wherein Input the employee-job matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model. Construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the job matching degree score, the development potential index, and team collaboration data, perform iterative training based on an exploration strategy of intrinsic motivation, and optimize the deployment strategy through a hierarchical task decomposition mechanism to generate an initial deployment plan. Evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, and analyze the employee performance data and team collaboration data after deployment in combination with a causal inference network. Adjust the parameters of the multi-agent reinforcement learning model in combination with a meta-policy optimization algorithm, including: Input the employee-job matching degree score matrix and the employee development potential index into a distributed multi-agent reinforcement learning model. Use a neural network to extract features from the employee-job matching degree score matrix to obtain an employee-job matching feature vector, and use a deep learning network to normalize the employee development potential index to obtain a standardized potential index. Generate a personnel distribution state vector in the form of a high-dimensional sparse vector, where the dimension of the personnel distribution state vector is the product of the number of employees and the number of jobs. Set the vector element corresponding to the job where the employee is currently located to 1, and the remaining elements to 0. Establish a team structure state matrix, and the team structure state matrix uses an adjacency matrix to describe the reporting relationship between team members. Combine the personnel distribution state vector and the team structure state matrix to form a state space. Generate an employee deployment plan matrix in the form of a two-dimensional matrix, where the number of rows of the employee deployment plan matrix is equal to the total number of employees, and the number of columns is equal to the total number of jobs. Set the matrix element where the employee is assigned to the corresponding job to 1, and the unassigned matrix elements to 0. Use the employee deployment plan matrix as the action space. Calculate the similarity between the employee-job matching feature vector and a preset job requirement vector to obtain a matching degree score, multiply the standardized potential index by a preset job development coefficient to obtain a potential score, calculate the collaboration density between team members based on the team structure state matrix to obtain a collaboration score, construct a reward function for the matching degree score, the potential score, and the collaboration score, and solve to obtain a reward value. Decompose the overall deployment task into local team deployment subtasks and cross-departmental deployment subtasks. For the local team deployment subtasks, use a deep reinforcement learning algorithm to train an agent to perform personnel deployment within the team. For the cross-departmental deployment subtasks, use a hierarchical reinforcement learning algorithm to train an agent to complete cross-departmental personnel deployment. Combine the training results of the local team deployment subtasks and the training results of the cross-departmental deployment subtasks to generate an initial deployment plan; Use a value network prediction module to predict the long-term benefits of the initial deployment plan. Calculate the improvement value of different deployment decisions on future team performance based on a graph neural network. Use a path evaluation module to evaluate the deployment path. Calculate the cumulative benefits of different deployment plans through the Monte Carlo tree search algorithm, and select the plan with the highest evaluation value as the optimal deployment plan; Collect the employee performance data and team collaboration data after executing the optimal deployment plan. Input the employee performance data and the team collaboration data into a causal inference network. Analyze the causal relationship between deployment decisions and performance and collaboration through a structural equation model to generate a causal relationship diagram. Extract the key causal paths to obtain the causal analysis results. According to the causal analysis results, use the gradient descent algorithm to optimize the network parameters in the distributed multi-agent reinforcement learning model, update the feature representation method of the state space, adjust the constraint conditions of the action space, and modify the calculation weights of the reward values to achieve adaptive optimization of the model parameters.

5. The method according to claim 4, wherein Calculating the improvement value of different deployment decisions on future team performance based on a graph neural network, using a path evaluation module to evaluate the deployment path, and calculating the cumulative benefits of different deployment plans through the Monte Carlo tree search algorithm, and selecting the plan with the highest evaluation value as the optimal deployment plan includes: Input the employee node feature vector and the position node feature vector into a pre-set graph neural network model. The employee node feature vector includes the skill features and performance data of the employee, and the position node feature vector includes the position requirement features and task description features. Based on the employee node feature vector and the position node feature vector, construct the collaboration relationship edge features and the position matching relationship edge features; Use a message passing neural network to update the node hidden state, perform a feature aggregation operation on the employee node feature vector and the position node feature vector to generate a node representation vector, and perform an edge feature aggregation operation on the collaboration relationship edge features and the position matching relationship edge features to generate an edge representation vector; Construct a training data set based on the pre-acquired historical deployment data. The historical deployment data includes historical deployment plans and team performance indicators. Input the training data set into the graph neural network model, and use the backpropagation algorithm to optimize the network parameters of the graph neural network model; Input the initial deployment plan into the graph neural network model, and calculate the team performance improvement value of each deployment decision based on the node representation vector and the edge representation vector to generate a decision evaluation vector; Construct a deployment decision tree, use the initial deployment plan as the root node, generate child nodes based on employee deployment constraints, each child node represents a deployment status, and calculate the performance evaluation score of each child node based on the decision evaluation vector; Use the Upper Confidence Bound (UCB) algorithm to select child nodes. Calculate the UCB value of each child node according to the average performance evaluation score of the child node, the number of visits to the parent node, the number of visits to the child node, and a preset exploration coefficient. Select the child node with the maximum UCB value for expansion. The UCB value adds an exploration term to the average performance evaluation score of the child node. The exploration term increases with the increase in the number of visits to the parent node and decreases with the increase in the number of visits to the child node; Perform an expansion operation on the selected child node, generate a new deployment decision based on the employee deployment constraints, and add the new deployment decision as a new child node to the deployment decision tree; Use the graph neural network model to evaluate the performance of the newly added child node, calculate the team performance improvement value of the newly added child node, and backpropagate the team performance improvement value along the decision path to the root node to update the number of visits and the average performance evaluation score of each node on the path; Repeat the operations of node selection, node expansion, performance evaluation, and result backpropagation until the preset search depth is reached. Select the decision path with the highest average performance evaluation score in the deployment decision tree as the optimal deployment plan.

6. An employee position matching and deployment system based on artificial intelligence, which is used to implement the method described in any one of the foregoing claims 1-5, and is characterized in that It includes: The first unit is used to collect historical work data of employees. The historical work data of employees includes work performance data, skill evaluation data, and team collaboration data. Perform multi-scale feature extraction on the historical work data of employees to obtain an initial employee ability feature vector and add it to a pre-set diffusion probability model. Combine the Markov chain sampling process to generate an optimized employee ability feature vector. Add the work performance data and the skill evaluation data to a pre-set spatio-temporal hybrid network model, and extract the employee ability development trend through convolution operations. Add the employee ability development trend, the optimized employee ability feature vector, and the team collaboration data to a cross-modal contrastive learning network, and solve for the employee comprehensive ability vector based on the contrast loss function and the dynamic temperature parameter and add it to the feature enhancement network. Combine the multi-scale feature pyramid and the adaptive feature aggregation module to obtain the development potential index; The second unit is used to collect job description data and historical recruitment data and add them to a pre-constructed generative pre-trained language model for semantic understanding and knowledge extraction, construct a job skill knowledge graph, and add the job skill knowledge graph to a dynamic heterogeneous graph neural network. The dynamic heterogeneous graph neural network generates a job skill feature matrix through time-varying topology learning, heterogeneous subgraph sampling, and graph structure distillation. Combine the adaptive graph attention network for hierarchical sampling and dynamic aggregation to obtain a team structure feature vector. Add the job skill feature matrix, the employee comprehensive ability vector, and the team structure feature vector to a neural architecture search network, and obtain the employee-job matching degree score matrix through parameter learning, hypernetwork optimization, and knowledge transfer; The third unit is used to input the employee-position matching degree score matrix and the development potential index into a distributed multi-agent reinforcement learning model, construct a state space including the current personnel distribution state and the team structure state, construct an action space based on the deployment plan space, construct a reward function based on the position matching degree score, the development potential index and the team collaboration data, perform iterative training based on the exploration strategy of intrinsic motivation, optimize the deployment strategy through a hierarchical task decomposition mechanism, generate an initial deployment plan, evaluate the initial deployment plan through the value network prediction module and the path evaluation module in the Monte Carlo tree search model, output a deployment execution plan according to the evaluation result, analyze the employee performance data and the team collaboration data after deployment in combination with a causal inference network, and adjust the parameters of the multi-agent reinforcement learning model in combination with a meta-policy optimization algorithm.

7. An electronic device, characterized in that, It includes: a processor; a memory for storing instructions executable by the processor; wherein, the processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, the method according to any one of claims 1 to 5 is implemented.

Images