Method and system for accurate matching and intelligent recommendation of people and sentry

Abstract

The invention discloses an intelligent recommendation method and system for accurate matching of personnel and sentry, and relates to the technical field of intelligent data recommendation, and the method comprises the steps: obtaining local data of an enterprise, building a local model, defining a local feature matrix, carrying out the mapping, merging into a global feature, synchronously optimizing a local gradient, and carrying out the dynamic weighted aggregation into a global gradient, updating and outputting the global features based on the global gradient; on the basis of the updated global features, extracting an embedded vector and an adjacency relation through a domain knowledge base to establish an initial heterogeneous graph, aggregating node information in the initial heterogeneous graph through a multilayer neural network to output an aggregated heterogeneous graph, obtaining job seeker data, extracting job seeker features, and calculating a worker-post matching comprehensive score through a dual-path matching score; and selecting a post according to the person-post matching comprehensive score, displaying the post to the job seeker, and generating a matching record according to the characteristics of the job seeker, and storing the matching record. According to the method, the defects in the aspects of data privacy protection, algorithm fairness and semantic modeling are effectively overcome.

Description

Technical Field

[0001] This invention relates to the field of data-driven intelligent recommendation technology, and in particular to a method and system for intelligent recommendation of precise matching of people and jobs. Background Technology

[0002] With the rapid development of information technology and the arrival of the big data era, the methods of enterprise recruitment and job seeker matching are gradually evolving towards digitalization and intelligence. Traditional human resource management systems mainly rely on keyword matching, rule filtering, or simple statistical analysis to achieve preliminary matching between people and positions. However, these methods are difficult to fully explore the deep semantic relationships and dynamic change patterns behind the data, resulting in widespread low matching efficiency and inaccurate matching. In recent years, with the rise of artificial intelligence, natural language processing, and data mining technologies, job matching methods based on machine learning and deep learning have gradually become research hotspots. These methods provide a more intelligent solution for recruitment recommendations by embedding job seeker characteristics and job descriptions, combined with the mining capabilities of multi-layer neural networks. At the same time, the introduction of knowledge graph technology has made it possible to model domain knowledge in job matching, and significantly improved the interpretability of recommendation systems by explicitly associating different entities.

[0003] However, existing research still has certain limitations. Existing technical models have significant shortcomings in terms of fairness and personalized recommendations, and lack in-depth utilization of domain rules and business knowledge. They also have deficiencies in terms of algorithmic fairness and the scientific nature of matching models. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method and system for intelligent recommendation of precise matching of people and jobs, which solves the problems that existing technical models have significant shortcomings in terms of fairness and personalized recommendation, lack in-depth utilization of domain rules and business knowledge, and still have deficiencies in terms of algorithmic fairness and the scientific nature of matching models.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for intelligent recommendation of precise matching between people and jobs, comprising, Acquire local enterprise data and build a local model. Define the local feature matrix and map it to merge it into global features. Simultaneously optimize the local gradient and dynamically weight and aggregate it into a global gradient. Update the global features based on the global gradient and output them. Based on the updated global features, an initial heterogeneous graph is established by extracting embedding vectors and adjacency relationships through a domain knowledge base. A multi-layer neural network is used to aggregate node information in the initial heterogeneous graph to output an aggregated heterogeneous graph. Job seeker data is obtained and job seeker features are extracted. A comprehensive score for job matching is calculated through dual-path matching scores. Job positions are selected and displayed to job seekers based on their overall job-person matching score, and matching records are generated and stored based on the job seeker's characteristics.

[0007] As a preferred embodiment of the intelligent recommendation method for precise matching of personnel and positions described in this invention, the step of acquiring local enterprise data, establishing a local model, defining a local feature matrix, mapping it, and then merging it into global features refers to acquiring local data for each enterprise. Preprocessing is performed, and a local model is defined using a deep learning network to extract basic job features. Deep learning is used to train the local model, and a local loss function is constructed to optimize the local model parameters. ; The local gradient is obtained by differentiating the local model parameters from the local loss function. ; The local model parameters are iteratively updated using gradient descent based on local gradients. After each update, the loss function is calculated. The iteration stops when the loss function converges, and the local model parameters and local gradients after the last iteration are output. The local model is updated based on the iterated local model parameters, and basic job features are extracted. A validation set is randomly selected from the training data. The local model is then validated using this validation set, and the validation loss is calculated. ; Calculate enterprise weights by verifying losses. ; A local characteristic matrix for each company is formed based on the basic characteristics of job positions. ; Perform a unified mapping of the local feature matrices of all enterprises; Obtain the local feature matrix after mapping all enterprises A global feature matrix is ​​generated by fusing enterprise weights. .

[0008] As a preferred embodiment of the intelligent recommendation method for precise matching of people and jobs described in this invention, the method involves: optimizing local gradients and dynamically weighting and aggregating them into global gradients; updating global features based on the global gradients and outputting the regularization correction factor calculated by differentiating the validation loss. ; Local gradients are modified by regularization factors. Make corrections; Noise is generated using a Laplace distribution. ; Adding Laplacian noise to the corrected local gradient generates a privacy-enhanced local gradient. ; Obtain the privacy-preserving local gradients of all enterprises The global gradient is calculated using weighted aggregation. ; Through global gradient For the global characteristic matrix Optimize to obtain the optimized global feature matrix .

[0009] As a preferred embodiment of the intelligent recommendation method for precise matching of people and jobs described in this invention, the method involves: establishing an initial heterogeneous graph based on updated global features by extracting embedding vectors and adjacency relationships from a domain knowledge base; and aggregating node information in the initial heterogeneous graph using a multi-layer neural network to output an aggregated heterogeneous graph, which is obtained from the optimized global feature matrix using a domain knowledge base. Extract job information, using each job as an initial heterogeneous graph node, and then extract the global feature matrix corresponding to that job. The row combination in the middle is used as the job feature as the initial node embedding vector. Calculate the initial node embedding vector for each pair of initial heterogeneous graph nodes. The cosine similarity is used to identify initial heterogeneous graph node pairs with a cosine similarity greater than the similarity threshold as adjacent nodes. Connecting edges are formed between adjacent nodes, and an initial heterogeneous graph is constructed based on the initial heterogeneous graph nodes and connecting edges. The node embedding vectors of the initial heterogeneous graph are dynamically aggregated using a multi-layer neural network; Repeat the aggregation operation until the network reaches a specified depth K, then stop and output the final aggregated heterogeneous graph.

[0010] As a preferred embodiment of the intelligent recommendation method for precise matching of people and jobs described in this invention, the following steps are included: acquiring job seeker data and extracting job seeker features, and calculating a comprehensive person-job matching score through dual-path matching scores. This involves acquiring job seeker data, extracting job seeker features from it, and calculating the similarity between the embedding vector of each node and the job seeker features based on an aggregated heterogeneous graph as a similarity score. ; Based on the enterprise's local data definition symbol logic rule library ; Define rule matching function If the job seeker meets the job requirements If the value is 1, then the value is 1; otherwise, the value is 0. By rule matching function Calculate rule-level scores ; Similarity score and rule-level scoring Weighted calculation of the overall score for person-job matching .

[0011] As a preferred embodiment of the intelligent recommendation method for precise matching of people and jobs described in this invention, the step of selecting jobs to display to job seekers based on the comprehensive score of the matching of people and jobs refers to sorting the jobs from high to low according to the comprehensive score of the matching of people and jobs, obtaining a job recommendation list for job seekers, and obtaining the company and job information of the job in the job recommendation list and displaying them accordingly for job seekers to view.

[0012] As a preferred embodiment of the intelligent recommendation method for precise matching of people and jobs described in this invention, the step of generating and storing matching records based on job seeker characteristics refers to generating job matching records based on job seeker characteristics and a job recommendation list, generating job seeker profiles based on job seeker characteristics, and storing the job matching records in the job seeker profiles.

[0013] Secondly, this invention provides a system for precise matching and intelligent recommendation of people and jobs, comprising, The feature extraction module is used to acquire local enterprise data, build a local model, define the local feature matrix, map and merge it into global features, simultaneously optimize local gradients and dynamically weight and aggregate them into global gradients, update global features based on global gradients and output them. The job matching module is used to establish an initial heterogeneous graph based on updated global features, extracting embedding vectors and adjacency relationships from the domain knowledge base, aggregating node information in the initial heterogeneous graph through a multi-layer neural network, outputting an aggregated heterogeneous graph, acquiring job seeker data and extracting job seeker features, and calculating a comprehensive job matching score through dual-path matching scores. The display record module is used to select positions to display to job seekers based on the comprehensive score of person-job matching, and to generate and store matching records based on job seeker characteristics.

[0014] Thirdly, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the intelligent recommendation method for precise matching of people and jobs as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent recommendation method for precise matching of people and jobs as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: This invention acquires local enterprise data to establish a local model, achieves global fusion of enterprise data through feature mapping, optimizes local gradients and weighted aggregates them into global gradients; further, it utilizes global features to construct an initial heterogeneous graph, aggregates node information through a multi-layer graph neural network, and combines a dual-path scoring calculation strategy to generate accurate matching recommendation results between job seekers and positions, effectively solving the shortcomings of existing technologies in data privacy protection, algorithmic fairness and semantic modeling, and improving the accuracy of job matching. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the intelligent recommendation method for precise matching of people and jobs in Example 1.

[0019] Figure 2 This is a structural diagram of the intelligent recommendation system for precise matching of people and jobs in Example 1. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a method for intelligent recommendation of precise matching between people and jobs, including the following steps: S1. Obtain local enterprise data and build a local model. Define the local feature matrix and map it to merge it into global features. Simultaneously optimize the local gradient and dynamically weight and aggregate it into a global gradient. Update the global features based on the global gradient and output them. Specifically, acquiring local enterprise data and building a local model, defining local feature matrices, mapping them, and then merging them into global features refers to acquiring local data for each enterprise. Preprocessing is performed, and a local model is defined using a deep learning network to extract basic job features. Convolutional neural networks, SVM classifiers, etc., can be used. The input to the local model is local enterprise data, and the output is basic job features, such as job category, required skill tags, salary range, years of experience, etc. The local model is trained with deep learning, and a local loss function is constructed to optimize the local model parameters. : in For local loss functions, For the predicted job basic features of the local data of the j-th enterprise, The actual job-related basic characteristics of the local data of the j-th enterprise are obtained through training data, where n is the number of training data. By using deep learning techniques such as convolutional neural networks (CNN) and support vector machines (SVM) classifiers, the local model can mine the feature distribution of enterprise job requirements and fully capture information features such as job categories, skill requirements, salary ranges and years of experience. This targeted feature extraction avoids the problem of data silos and also improves the data expression ability of local job features. The local gradient is obtained by differentiating the local model parameters from the local loss function. : in This is a partial derivative operation; The local model parameters are iteratively updated using gradient descent based on local gradients. After each update, the loss function is calculated. The iteration stops when the loss function converges, and the local model parameters and local gradients after the last iteration are output. The local model is updated based on the iterated local model parameters, and basic job features are extracted. A validation set is randomly selected from the training data. The local model is then validated using this validation set, and the validation loss is calculated. : in To determine the number of data points in the validation set, k is the data index of the validation set. The dynamic calculation of validation loss allows for a comprehensive evaluation of the performance of different enterprise models and the allocation of weights based on performance. Compared with the uniform weight strategy in the prior art, the dynamic weight allocation mechanism in this invention avoids the negative impact of low-quality data on the global feature matrix. Calculate enterprise weights by verifying losses. : in This is a smoothing factor to prevent the validation loss from being zero. A local characteristic matrix for each company is formed based on the basic characteristics of job positions. : Where n is the number of job positions in the enterprise's local data, and d is the basic feature dimension, including job category, job requirement skill tags, salary range, years of experience, etc. Skill embedding processing generates semantic embeddings through Word2Vec or BERT models, which can capture contextual information in skill labels. For example, "Python" and "database" can establish an approximate relationship in the embedding space through pre-trained models, thereby enhancing the semantic accuracy of skill feature representation. To standardize the dimensions and representations of different organizational characteristics, a unified mapping is performed on the local feature matrices of all enterprises: in This is the mapped local feature matrix. Here is the mapping function for the i-th enterprise, used for normalization and unifying the local matrix representation; Where the mapping function Includes the following operations: One-hot encoding of categorical features: This transforms discrete categorical variables into high-dimensional one-hot vectors, facilitating fusion with other features, for example: Job type = (Python developer, Java engineer, data analyst) — One-hot encoding: ; Embedding of skill features: Embedding vectors are generated using pre-trained models of skill text (such as Word2Vec and BERT), for example: "Python, Spring, Database" - Word2Vec Embedding: ; Numerical feature normalization: Normalizing continuous variables such as years of experience or salary range to... ; Obtain the local feature matrix after mapping all enterprises A global feature matrix is ​​generated by fusing enterprise weights. : Where N is the number of enterprises.

[0024] The global feature matrix is ​​generated by dynamically fusing local feature matrices and weights. The weights are dynamically adjusted based on the validation loss to ensure the scientific contribution of each enterprise's data to the global features. This mechanism can significantly improve the accuracy of the global feature matrix, making it more comprehensive and fair.

[0025] Furthermore, local gradients are optimized and dynamically weighted to form global gradients. Global features are updated based on global gradients, and the regularization correction factor is calculated by differentiating the validation loss. : Where m is the number of data in the validation set; The regularization correction process can automatically strengthen the contribution of enterprises with large validation losses to the global gradient, thereby reflecting the characteristics of minority samples in the data of such enterprises more meticulously. The regularization correction factor is calculated by the validation loss and combined with the gradient correction coefficient to adjust the influence of the local gradient of each enterprise, ensuring that the global model learns the key information of each enterprise's data. Compared with the traditional method of using fixed weights or simple averaging, the introduction of dynamic weights effectively avoids the risk of the global model deviating due to the poor quality of individual abnormal enterprise data. Local gradients are modified by regularization factors. Make corrections: in The corrected local gradient, This is the gradient correction coefficient, used to control the correction magnitude, typically set to 0.1~0.5; Noise is generated using a Laplace distribution. : in Here, x is a privacy budget parameter, a random variable in a Laplace distribution used to represent a value sampled from that distribution; Adding Laplacian noise to the corrected local gradient generates a privacy-enhanced local gradient. : By using the Laplace distribution to generate noise and superimposing it into the corrected gradient, the gradient data uploaded to the central server does not have the ability to directly trace the local data. This differential privacy protection method effectively preserves the optimization direction of the gradient while ensuring privacy and prevents optimization failure due to noise interference. Through the superposition of random noise, the privacy gradient uploaded each time has randomness in value, which further increases the security of the system when facing external cracking. Obtain the privacy-preserving local gradients of all enterprises The global gradient is calculated using weighted aggregation. : Through global gradient For the global characteristic matrix Optimize to obtain the optimized global feature matrix : in The learning rate; The above global feature matrix and global gradient All of them are formed through local aggregation, therefore they have the same dimensions and represent all enterprises, through global gradients. For the global characteristic matrix The update is actually performed using global gradients. The elements in each dimension of the global feature matrix represent the elements in each dimension. The elements in the corresponding dimension are updated element by element to achieve the overall global feature matrix. Update.

[0026] Element-wise optimization allows each feature dimension of the global feature matrix to be finely adjusted according to the corresponding global gradient, avoiding the drawback that overall updates may ignore local characteristics. By dynamically weighting and fusing data from multiple enterprises, combined with the global gradient optimization mechanism, the problem of inconsistent data representation caused by different job requirements among enterprises is solved. The global feature matrix not only reflects the common needs of enterprises, but also takes into account the personalized characteristics of each enterprise.

[0027] S2. Based on the updated global features, an initial heterogeneous graph is established by extracting embedding vectors and adjacency relationships through the domain knowledge base. The node information in the initial heterogeneous graph is aggregated and the aggregated heterogeneous graph is output through a multi-layer neural network. Job seeker data is obtained and job seeker features are extracted. The comprehensive score of job matching is calculated through dual-path matching scores. Specifically, based on the updated global features, an initial heterogeneous graph is established by extracting embedding vectors and adjacency relationships through a domain knowledge base. A multi-layer neural network then aggregates node information in the initial heterogeneous graph to output an aggregated heterogeneous graph, which is derived from the optimized global feature matrix using the domain knowledge base. Extract job information, using each job as an initial heterogeneous graph node, and then extract the global feature matrix corresponding to that job. The row combination in the middle is used as the job feature as the initial node embedding vector. Calculate the initial node embedding vector for each pair of initial heterogeneous graph nodes. The cosine similarity is used to identify initial heterogeneous graph node pairs with a cosine similarity greater than the similarity threshold as adjacent nodes. Connecting edges are formed between adjacent nodes, and an initial heterogeneous graph is constructed based on the initial heterogeneous graph nodes and connecting edges. Dynamically aggregate the node embedding vectors of the initial heterogeneous graph using a multi-layer neural network: in Let i be the set of adjacent nodes. Let j be the node embedding vector of the neighbor node j at the k-th layer. Let be the trainable weight matrix of the k-th layer. and Let be the degree of node i and node j, and represent the number of neighboring nodes of node i. For activation function, Let i be the node embedding vector of node i in the (k+1)th layer; Repeat the aggregation operation until the network reaches a specified depth K, then stop and output the final aggregated heterogeneous graph.

[0028] By extracting job information from the global feature matrix into high-dimensional embedding vectors, this invention can represent the complex basic features of jobs. Furthermore, by combining a domain knowledge base, it enriches the semantic information of the vectors. Compared with traditional rule-based or keyword-based feature processing methods, this high-dimensional representation method can more fully capture the implicit associations of job skill requirements, thus laying a solid foundation for complex semantic matching. The cosine similarity of node embedding vectors is used to define whether nodes establish adjacency relationships, and a similarity threshold is used to control edge generation. This dynamic adjacency design avoids the shortcomings of traditional graph construction that relies on fixed rules, and can dynamically define edges based on job information. A multi-layer neural network is used to dynamically aggregate node embedding vectors, enabling each job node to simultaneously absorb important feature information from itself and neighboring nodes, thereby learning a more comprehensive job representation. Each layer dynamically adjusts node features through weight matrices and adjacency relationships, enabling nodes to progressively build implicit relationship networks between jobs. Normalization reduces the interference of connection number differences on node feature aggregation, making gradient updates more stable. Meanwhile, the nonlinear transformation of the activation function enhances the model's ability to capture complex feature relationships. The multi-layer network can gradually aggregate features of remote neighboring nodes, providing deeply related information for the target node. This design is particularly suitable for modeling specific needs across skills and domains in job matching, solving the problem of fragmented and isolated job features in existing technologies. The aggregation result outputs the final node embedding vector through multi-layer aggregation operations. This embedding vector can comprehensively integrate local job features and their relationship information with other jobs, thereby guiding the execution of downstream job recommendation tasks and optimizing the accuracy of job matching for job seekers.

[0029] Furthermore, job seeker data is acquired and job seeker features are extracted. A comprehensive person-job matching score is calculated using dual-path matching. This involves acquiring job seeker data and extracting job seeker features, including desired positions, skills, desired salaries, and work experience. The similarity score is then calculated based on the similarity between the embedding vector of each node and the job seeker features using an aggregated heterogeneous graph. ; By calculating the similarity between job seeker features and job embedding vectors, the degree of matching between job seeker features and job requirements can be quantified. This approach is more flexible than traditional keyword-based methods and can capture the potential relationship between job requirements and job seeker skills through a high-dimensional semantic space. Based on the enterprise's local data definition symbol logic rule library This is used to verify whether job applicants meet the job requirements. The total number of rules is defined as Q, for example: Rule 1: "5 years of Python programming experience is equivalent to a Java junior developer certification"; Rule 2: "Mastering basic machine learning skills is equivalent to having mathematical modeling ability"; Define rule matching function If the job seeker meets the job requirements If the value is 1, then the value is 1; otherwise, the value is 0. By rule matching function Calculate rule-level scores : in For the qth rule The weight, For the qth rule The rule matching function value; By using a symbolic logic rule base to verify whether job seekers meet job requirements, constraints from domain knowledge or industry standards can be introduced. For example, the rule "5 years of Python programming experience is equivalent to a Java junior developer certification" reflects the substitutability or equivalence between job skills, improving the practicality of the matching results. The symbolic logic rule base allows for dynamic adjustment of rule content and weights based on job characteristics. For technical positions, rules may focus on skill requirements, while for management positions, rules may emphasize experience years and salary matching. This flexibility significantly improves the system's applicability. Similarity score and rule-level scoring Weighted calculation of the overall score for person-job matching : in and Let be the weighting coefficient for the score, and , .

[0030] The comprehensive score calculation combines similarity score and rule-level score to improve matching accuracy from two dimensions: data-driven matching and domain logic matching. This dual-path scoring mechanism makes up for the shortcomings of a single method, enabling the recommendation algorithm to handle multilingual and complex features while meeting the strict requirements of job business rules.

[0031] S3. Select job positions based on the comprehensive score of person-job matching and display them to job seekers, and generate and store matching records based on job seeker characteristics; Specifically, selecting and displaying job postings based on the overall person-job matching score means sorting job postings from highest to lowest based on the overall person-job matching score, obtaining a job recommendation list for job seekers, and obtaining the corresponding company and job information (job requirements, salary, etc.) for job seekers to view in the job recommendation list.

[0032] Furthermore, generating and storing matching records based on job seeker characteristics refers to generating job matching records based on job seeker characteristics and the job recommendation list, generating job seeker profiles based on job seeker characteristics, and storing the job matching records in the job seeker profiles.

[0033] This embodiment also provides a system for intelligent recommendation of precise matching of people and jobs, including: The feature extraction module is used to acquire local enterprise data, build a local model, define the local feature matrix, map and merge it into global features, simultaneously optimize local gradients and dynamically weight and aggregate them into global gradients, update global features based on global gradients and output them. The job matching module is used to establish an initial heterogeneous graph based on updated global features, extracting embedding vectors and adjacency relationships from the domain knowledge base, aggregating node information in the initial heterogeneous graph through a multi-layer neural network, outputting an aggregated heterogeneous graph, acquiring job seeker data and extracting job seeker features, and calculating a comprehensive job matching score through dual-path matching scores. The display record module is used to select positions to display to job seekers based on the comprehensive score of person-job matching, and to generate and store matching records based on job seeker characteristics.

[0034] This embodiment also provides a computer device applicable to the method of intelligent recommendation for precise matching of people and jobs, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method of intelligent recommendation for precise matching of people and jobs as proposed in the above embodiment. The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0035] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for intelligent recommendation of precise matching of people and jobs as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0036] In summary, this invention acquires local enterprise data to build a local model, achieves global fusion of enterprise data through feature mapping, optimizes local gradients and weighted aggregates them into global gradients; further, it utilizes global features to construct an initial heterogeneous graph, aggregates node information through a multi-layer graph neural network, and combines a dual-path scoring calculation strategy to generate accurate matching recommendation results between job seekers and positions. This effectively solves the shortcomings of existing technologies in terms of data privacy protection, algorithmic fairness, and semantic modeling, and improves the accuracy of job matching.

[0037] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent recommendation of precise matching between people and jobs, characterized in that: include, Acquire local enterprise data and build a local model. Define the local feature matrix and map it to merge it into global features. Simultaneously optimize the local gradient and dynamically weight and aggregate it into a global gradient. Update the global features based on the global gradient and output them. Based on the updated global features, an initial heterogeneous graph is established by extracting embedding vectors and adjacency relationships through a domain knowledge base. A multi-layer neural network is used to aggregate node information in the initial heterogeneous graph to output an aggregated heterogeneous graph. Job seeker data is obtained and job seeker features are extracted. A comprehensive score for job matching is calculated through dual-path matching scores. Job positions are selected and displayed to job seekers based on their overall job-person matching score, and matching records are generated and stored based on the job seeker's characteristics.

2. The method for precise matching and intelligent recommendation of people and jobs as described in claim 1, characterized in that: The process of acquiring local enterprise data, establishing a local model, defining a local feature matrix, mapping it, and then merging it into global features refers to acquiring local data from each enterprise. Preprocessing is performed, and a local model is defined using a deep learning network to extract basic job features. Deep learning is used to train the local model, and a local loss function is constructed to optimize the local model parameters. ; The local gradient is obtained by differentiating the local model parameters from the local loss function. ; The local model parameters are iteratively updated using gradient descent based on local gradients. After each update, the loss function is calculated. The iteration stops when the loss function converges, and the local model parameters and local gradients after the last iteration are output. The local model is updated based on the iterated local model parameters, and basic job features are extracted. A validation set is randomly selected from the training data. The local model is then validated using this validation set, and the validation loss is calculated. ; Calculate enterprise weights by verifying losses. ; A local characteristic matrix for each company is formed based on the basic characteristics of job positions. ; Perform a unified mapping of the local feature matrices of all enterprises; Obtain the local feature matrix after mapping all enterprises A global feature matrix is ​​generated by fusing enterprise weights. .

3. The method for precise matching and intelligent recommendation of people and jobs as described in claim 2, characterized in that: The local gradient is optimized and dynamically weighted to form a global gradient. Global features are updated based on the global gradient, and a regularization correction factor is calculated by taking the derivative of the validation loss. ; Local gradients are modified by regularization factors. Make corrections; Noise is generated using a Laplace distribution. ; Adding Laplacian noise to the corrected local gradient generates a privacy-enhanced local gradient. ; Obtain the privacy-preserving local gradients of all enterprises The global gradient is calculated using weighted aggregation. ; Through global gradient For the global characteristic matrix Optimize to obtain the optimized global feature matrix .

4. The method for precise matching and intelligent recommendation of people and jobs as described in claim 3, characterized in that: Based on the updated global features, an initial heterogeneous graph is established by extracting embedding vectors and adjacency relationships from a domain knowledge base. A multi-layer neural network then aggregates node information in the initial heterogeneous graph to output an aggregated heterogeneous graph, which is derived from the optimized global feature matrix using the domain knowledge base. Extract job information, using each job as an initial heterogeneous graph node, and then extract the global feature matrix corresponding to that job. The row combination in the middle is used as the job feature as the initial node embedding vector. Calculate the initial node embedding vector for each pair of initial heterogeneous graph nodes. The cosine similarity is used to identify initial heterogeneous graph node pairs with a cosine similarity greater than the similarity threshold as adjacent nodes. Connecting edges are formed between adjacent nodes, and an initial heterogeneous graph is constructed based on the initial heterogeneous graph nodes and connecting edges. The node embedding vectors of the initial heterogeneous graph are dynamically aggregated using a multi-layer neural network; Repeat the aggregation operation until the network reaches a specified depth K, then stop and output the final aggregated heterogeneous graph.

5. The method for precise matching and intelligent recommendation of people and jobs as described in claim 4, characterized in that: The process of acquiring job seeker data and extracting job seeker features, and calculating a comprehensive person-job matching score through dual-path matching, refers to acquiring job seeker data, extracting job seeker features from it, and calculating the similarity between the embedding vector of each node and the job seeker features based on the aggregated heterogeneous graph as a similarity score. ; Based on the enterprise's local data definition symbol logic rule library ; Define rule matching function If the job seeker meets the job requirements If the value is 1, then the value is 1; otherwise, the value is 0. By rule matching function Calculate rule-level scores ; Similarity score and rule-level scoring Weighted calculation of the overall score for person-job matching .

6. The method for precise matching and intelligent recommendation of people and jobs as described in claim 5, characterized in that: The step of selecting and displaying job postings to job seekers based on the comprehensive score of job matching refers to sorting job postings from highest to lowest according to the comprehensive score of job matching, obtaining a job recommendation list for job seekers, and then obtaining the company and job information of the corresponding job postings and displaying them in the job recommendation list for job seekers to view.

7. The method for precise matching and intelligent recommendation of people and jobs as described in claim 6, characterized in that: The process of generating and storing matching records based on job seeker characteristics refers to generating job matching records based on job seeker characteristics and a job recommendation list, generating job seeker profiles based on job seeker characteristics, and storing the job matching records in the job seeker profiles.

8. A system for precise matching and intelligent recommendation of personnel and positions, based on the method for precise matching and intelligent recommendation of personnel and positions as described in any one of claims 1 to 7, characterized in that: include, The feature extraction module is used to acquire local enterprise data, build a local model, define the local feature matrix, map and merge it into global features, simultaneously optimize local gradients and dynamically weight and aggregate them into global gradients, update global features based on global gradients and output them. The job matching module is used to establish an initial heterogeneous graph based on updated global features, extracting embedding vectors and adjacency relationships from the domain knowledge base, aggregating node information in the initial heterogeneous graph through a multi-layer neural network, outputting an aggregated heterogeneous graph, acquiring job seeker data and extracting job seeker features, and calculating a comprehensive job matching score through dual-path matching scores. The display record module is used to select positions to display to job seekers based on the comprehensive score of person-job matching, and to generate and store matching records based on job seeker characteristics.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent recommendation method for precise matching of people and jobs as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent recommendation method for precise matching of people and jobs as described in any one of claims 1 to 7.

Images