Human resource matching method and system based on machine learning

Abstract

The invention discloses a human resource matching method and system based on machine learning, and relates to the technical field of human resource matching, and the method comprises the steps: obtaining employee attribute data and post demand data; generating a preliminary matching set and pheromone concentration distribution by adopting an optimization matching algorithm; performing quantitative verification and precision evaluation on each matching pair in the preliminary matching set by adopting a matching evaluation model to generate a precise matching scheme; establishing a co-evolution mechanism, and dynamically adjusting the calculation weight of each evaluation coefficient in the matching evaluation model according to the pheromone concentration; based on a preset team construction requirement, selecting a candidate team from the accurate matching scheme, and performing efficiency prediction on the candidate team by adopting a team collaborative simulation model to generate a team matching scheme; through fusion of bionic optimization and a mechanical model, efficient screening and accurate evaluation from massive employee data are realized, and the intelligent level and decision quality of human resource configuration are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of human resource matching technology, and specifically to a human resource matching method and system based on machine learning. Background Technology

[0002] The current field of human resource matching mainly relies on two major technical approaches. One is a data-driven approach based on statistics, such as machine learning algorithms like the weight of evidence method, logistic regression, or random forest. These methods mine and analyze a large amount of historical application data and successful job cases to build predictive models to achieve automated screening and recommendation. The other is a rule-based expert system or knowledge graph approach, which relies on manually defined competency models, job profiles, and matching rules to achieve person-job matching through logical reasoning. However, existing technologies have certain limitations. Although the first type of method has an efficiency advantage in handling explicit features and clear historical patterns, its models heavily rely on high-quality, large-scale, and clearly labeled training data, making it difficult to cope with cold-start scenarios such as dynamic changes in recruitment needs, vague job descriptions, or scarce positions. The interpretability of the models is usually weak, and they cannot effectively integrate complex evaluation dimensions such as deep soft skills and cultural fit. The second type of method can embed domain knowledge, but its rule maintenance costs are high, its flexibility is insufficient, and it is difficult to adapt to the unique cultures and management styles of different organizations. Moreover, the system performance is highly dependent on the completeness and accuracy of expert experience, and there are obvious bottlenecks in large-scale application and generalization to complex scenarios. At the same time, both types of existing technologies treat matching as a static, one-time decision, lacking a continuous tracking and feedback learning mechanism for matching results, and failing to conduct forward-looking simulation and evaluation of personnel combinations from the perspective of overall team effectiveness. This means that the optimal individual matching may not be able to converge into team synergy. Therefore, developing a machine learning-based human resource matching method is of great significance. Summary of the Invention

[0003] The purpose of this invention is to provide a human resource matching method and system based on machine learning to address the shortcomings in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a human resource matching method based on machine learning, comprising: Obtain employee attribute data and job requirement data; An optimized matching algorithm is used to perform global exploration and self-organizing matching of employee attribute data and job requirement data to generate a preliminary matching set and pheromone concentration distribution. A matching evaluation model is used to quantitatively verify and evaluate the accuracy of each matching pair in the preliminary matching set, and to generate an accurate matching scheme. Establish a collaborative evolution mechanism to identify high-success matching patterns based on pheromone concentration distribution and dynamically adjust the calculation weights of each evaluation coefficient in the matching evaluation model; Based on preset team building requirements, candidate teams are selected from the precise matching schemes, and a team collaboration simulation model is used to predict the effectiveness of the candidate teams and generate team matching schemes.

[0005] In a preferred embodiment, the steps of obtaining employee attribute data and job requirement data are as follows: Employee attribute data includes skill sets, work experience values, and collaboration preference characteristics; Job requirements data includes a set of skill requirements, experience requirements, and team culture characteristics; The feature extraction algorithm is used to structure employee attribute data and job requirement data to generate employee feature tags and job feature tags; Employee feature vectors and job requirement vectors are generated based on employee feature tags and job feature tags.

[0006] In a preferred embodiment, the step of using an optimized matching algorithm to perform global exploration and self-organizing matching of employee attribute data and job requirement data to generate a preliminary matching set and pheromone concentration distribution is as follows: Construct a bipartite graph model with employee nodes and job nodes as elements, where the edges in the bipartite graph model represent the potential matching relationship between employees and jobs; Initialize the pheromone concentration of each edge in the bipartite graph model, and calculate the heuristic factor of each edge based on the basic similarity between the employee feature vector and the job requirement vector; Multiple rounds of iterative exploration were conducted on the bipartite graph model; Based on the current pheromone concentration and heuristic factor of each edge, multiple different matching sequences are generated through a probability selection function. Each matching sequence consists of a set of non-conflicting edges, representing a matching scheme. Calculate the weighted sum of similarity between the employee feature vectors and the job requirement vectors of all matching pairs in the matching sequence to generate a comprehensive quality assessment. Based on the comprehensive quality assessment results, the pheromone concentration of each edge constituting the matching sequence is updated, with edges corresponding to high-quality matching sequences receiving a larger pheromone concentration increment; The iteration termination condition is preset. Based on the final pheromone concentration of each edge in the bipartite graph model, the edges with the highest pheromone concentration are selected for each job node. The employee nodes corresponding to the edges are added to the candidate list of the job. The candidate lists of all jobs are integrated to generate a preliminary matching set, and the pheromone concentration values ​​of each edge are output as the pheromone concentration distribution.

[0007] In a preferred embodiment, the step of using a matching evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the preliminary matching set, and generating an accurate matching scheme, is as follows: For each matching pair in the initial matching set, input the matching evaluation model; Each skill item and its proficiency value in the employee's skill set is mapped to the corresponding tooth height and tooth width geometric parameters on the drive gear; Map each skill requirement item and its level value in the set of job skill requirements to the corresponding tooth height and tooth width geometric parameters on the driven gear; Calculate the matching degree of geometric parameters of all corresponding teeth on the driving gear and the driven gear, and generate the skill tooth profile matching degree; Based on the logical dependencies between skill items, a series gear transmission chain is constructed, and the employee's work experience value is quantified as the input torque of the transmission chain to simulate and calculate the mechanical transmission efficiency of the transmission chain. Calculate the similarity between the employee collaboration preference feature vector and the job team culture feature vector, and map this similarity to a lubrication coefficient that reduces transmission loss; Based on preset weights, the skill tooth profile matching degree, mechanical transmission efficiency and lubrication coefficient are weighted and fused to calculate and generate the accurate matching degree score of the matching pair. Aggregate all evaluated matching pairs and their exact match scores to generate an exact match solution.

[0008] In a preferred embodiment, the steps of establishing a co-evolutionary mechanism, identifying high-success matching patterns based on pheromone concentration distribution, and dynamically adjusting the calculation weights of each evaluation coefficient in the matching evaluation model are as follows: From the pheromone concentration distribution, edges with pheromone concentration values ​​higher than a preset threshold are selected, and matching pairs corresponding to these edges are obtained to form a set of high-success matching patterns. Analyze each matching pair in the high-success matching pattern set and extract three evaluation coefficients generated by the matching evaluation model: skill tooth profile matching degree, mechanical transmission efficiency, and lubrication coefficient. Calculate the statistical contribution of each evaluation coefficient for all matching pairs in the high-success matching pattern set, wherein the statistical contribution of the evaluation coefficient is determined based on the correlation between the numerical distribution of the coefficient in the set and the exact match score. Based on the statistical contribution of each evaluation coefficient, the preset weights of skill tooth profile matching degree, mechanical transmission efficiency and lubrication coefficient in the matching evaluation model are adjusted in the same direction, so that the evaluation coefficient with high statistical contribution will receive higher calculation weight after the weight adjustment.

[0009] In a preferred embodiment, the step of selecting candidate teams from the precise matching scheme based on preset team building requirements, and using a team collaboration simulation model to predict the effectiveness of the candidate teams and generate a team matching scheme is as follows: Pre-defined team building requirements include job roles, number of personnel, and collaborative relationship network; Based on the team building requirements, at least one employee with the highest precision match score for each job role is selected from the precise matching scheme to form a candidate team; Input the candidate teams into the team collaboration simulation model and output the team matching scheme.

[0010] In a preferred embodiment, the step of inputting candidate teams into the team collaboration simulation model and outputting team matching solutions is as follows: Map each member of the candidate team to a member's individual gear in the team gear set; Map members' work experience values ​​to the tooth width of their individual gears; The number of skills in a member's skill set is mapped to the number of teeth on the member's personal gear. The collaborative preference characteristics of members are mapped to the friction coefficient of the individual gear teeth of each member; Based on the collaborative relationship network defined in the team building requirements, the meshing connection relationship between the individual gears of each member in the team gear set is determined, and the team transmission topology is constructed. The total amount of team tasks is quantified as input torque and loaded onto the team transmission topology; The predicted collaborative efficiency value of the candidate team is calculated based on the ratio of the sum of the torques of the output gears at the end of the transmission topology to the input torque. By integrating team transmission topology and predicted collaboration efficiency values, a team matching scheme is generated.

[0011] This invention also provides a machine learning-based human resource matching system, comprising: Data acquisition module: Acquires employee attribute data and job requirement data; Job matching module: Connected to the data acquisition module, it uses an optimized matching algorithm to perform global exploration and self-organized matching of employee attribute data and job requirement data, generating a preliminary matching set and pheromone concentration distribution; Matching and evaluation module: Connected to the job matching module, it uses a matching and evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the initial matching set, and generates an accurate matching plan; Parameter update module: Connects with the matching evaluation module to establish a co-evolution mechanism, identifies high-success matching patterns based on pheromone concentration distribution, and dynamically adjusts the calculation weights of each evaluation coefficient in the matching evaluation model; Team matching module: Connected to the parameter update module, it selects candidate teams from the precise matching schemes based on preset team building requirements, and uses a team collaboration simulation model to predict the effectiveness of the candidate teams and generate team matching schemes.

[0012] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs a human resource intelligent matching system with continuous learning capabilities by integrating biomimetic optimization algorithms and mechanical structure models. First, it utilizes biomimetic optimization algorithms to simulate swarm intelligence, conducting efficient global exploration across a vast array of employee and job combinations to quickly identify potential matching areas. Then, it introduces a sophisticated gear meshing model, mapping multi-dimensional features such as skills, experience, and collaboration into quantifiable mechanical parameters for in-depth evaluation, achieving refined scoring of matched pairs. Simultaneously, a collaborative evolution mechanism is designed, enabling the model to automatically adjust the weight parameters of the evaluation criteria based on feedback from historical successful matching cases. This allows the system to adapt to specific organizational cultures and stage-specific needs. This complete closed loop, from broad-area exploration to precise evaluation, and from practical feedback to model optimization, significantly improves the accuracy, interpretability, and long-term adaptability of the matching system, providing companies with stable and reliable intelligent talent allocation decision support.

[0013] 2. This invention extends individual matching results to team collaboration efficiency simulation, achieving a leap from static job matching to dynamic team building. Based on high-precision individual matching, it innovatively simulates the entire team as a gear transmission system. According to the team structure, it maps each member to a gear with specific parameters, defines the collaborative relationship as the meshing between gears, and quantifies the task load as the system's input torque. By calculating efficiency losses during the transmission process, the system can predict the overall team's collaborative efficiency value and accurately identify potential collaboration bottlenecks and risk interfaces. This allows decision-makers not only to understand who is suitable for a particular position but also to anticipate the overall work capacity and potential internal friction points of this group after forming a team. This mechanism provides unprecedented quantitative assessment and predictive planning tools for complex scenarios such as project team building and cross-departmental collaborative group establishment, greatly reducing the team's trial-and-error costs. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a flowchart of the method of the present invention.

[0016] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 As shown in this embodiment, a human resource matching method based on machine learning includes: S1. Obtain employee attribute data and job requirement data; S2. An optimized matching algorithm is used to perform global exploration and self-organized matching of employee attribute data and job requirement data to generate a preliminary matching set and pheromone concentration distribution. S3. Use a matching evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the preliminary matching set, and generate an accurate matching scheme. S4. Establish a collaborative evolution mechanism to identify high-success matching patterns based on pheromone concentration distribution and dynamically adjust the calculation weights of each evaluation coefficient in the matching evaluation model. S5. Based on the preset team building requirements, select candidate teams from the precise matching schemes, and use the team collaboration simulation model to predict the effectiveness of the candidate teams and generate team matching schemes. As described in steps S1-S5 above, the existing human resource matching field mainly relies on two major technical approaches. One is a data-driven approach based on statistics, such as machine learning algorithms like the weight of evidence method, logistic regression, or random forest. These methods mine and analyze a large amount of historical job application data and successful job cases to build predictive models to achieve automated screening and recommendation. The other is a rule-based expert system or knowledge graph approach, which relies on manually defined competency models, job profiles, and matching rules to achieve person-job matching through logical reasoning. However, existing technologies have certain limitations. Although the first type of method has an efficiency advantage in handling explicit features and clear historical patterns, its models heavily rely on high-quality, large-scale, and clearly labeled training data, making it difficult to cope with cold-start scenarios such as dynamic changes in recruitment needs, vague job descriptions, or scarce positions. The interpretability of the models is usually weak, and they cannot effectively integrate complex evaluation dimensions such as deep soft skills and cultural fit. The second type of method can embed domain knowledge, but its rule maintenance costs are high, its flexibility is insufficient, and it is difficult to adapt to the unique cultures and management styles of different organizations. Moreover, the system performance is highly dependent on the completeness and accuracy of expert experience, and there are obvious bottlenecks in large-scale application and generalization to complex scenarios. At the same time, both types of existing technologies treat matching as a static, one-time decision, lacking a continuous tracking and feedback learning mechanism for matching results, and failing to conduct forward-looking simulation and evaluation of personnel combinations from the perspective of overall team effectiveness. This means that the optimal individual matching may not be able to converge into team synergy. This invention constructs a human resource intelligent matching system with continuous learning capabilities by integrating biomimetic optimization algorithms and mechanical structure models. First, it utilizes biomimetic optimization algorithms to simulate swarm intelligence, conducting efficient global exploration across a massive number of employee and job combinations to quickly identify potential matching areas. Then, it introduces a sophisticated gear meshing model, mapping multi-dimensional features such as skills, experience, and collaboration into quantifiable mechanical parameters for in-depth evaluation, achieving refined scoring of matching pairs. Simultaneously, a collaborative evolution mechanism is designed, enabling the model to automatically adjust the weight parameters of the evaluation criteria based on feedback from historical successful matching cases. This allows the system to adapt to specific organizational cultures and stage-specific needs. This complete closed loop, from broad-area exploration to precise evaluation, and from practical feedback to model optimization, significantly improves the accuracy, interpretability, and long-term adaptability of the matching system, providing companies with stable and reliable intelligent talent allocation decision support. By extending individual matching results to team collaboration efficiency simulation, a leap from static job matching to dynamic team building has been achieved. Building upon high-precision individual matching, the entire team is innovatively simulated as a gear transmission system. Based on the team structure, each member is mapped to a gear with specific parameters, the collaborative relationship is defined as the meshing between gears, and the task load is quantified as the system's input torque. By calculating efficiency losses during the transmission process, the system can predict the overall team's collaborative efficiency and accurately identify potential collaboration bottlenecks and risk interfaces. This allows decision-makers not only to understand who is suitable for a particular position but also to anticipate the overall work capacity and potential internal friction points of this group as a team. This mechanism provides unprecedented quantitative assessment and predictive planning tools for complex scenarios such as project team building and cross-departmental collaborative group establishment, significantly reducing the team's trial-and-error costs.

[0019] In one embodiment, step S1 of obtaining employee attribute data and job requirement data includes: S11. Employee attribute data includes skill sets, work experience values, and collaboration preference characteristics; S12. Job requirements data includes a set of skill requirements, experience requirements, and team culture characteristics. S13. Use feature extraction algorithms to structure employee attribute data and job requirement data to generate employee feature tags and job feature tags. S14. Generate employee feature vectors and job requirement vectors based on employee feature tags and job feature tags; As described in steps S11-S14 above, in the data acquisition and processing stage, the system first collects raw employee attribute data and job requirement data from the human resources database and recruitment management platform. For employee attribute data, the skill set is extracted by parsing the text of the project experience and skill certificate sections of the employee's resume, using named entity recognition technology to extract specific skill items such as programming languages ​​or professional software names, and quantifying them into numerical proficiency values ​​using proficiency descriptive words such as proficiency, familiarity, and understanding appearing in the text context. The work experience value is calculated based on the start and end dates in the employee's work history, and weighted and summed according to the complexity and scale of the projects they participated in. Collaboration preference characteristics are comprehensively evaluated by analyzing the employee's historical communication patterns on the internal collaboration platform and the results of the team project role questionnaire survey, and transformed into a multi-dimensional feature vector. For job requirement data, the skill set is further analyzed by parsing the text of the project experience and skill certificate sections of the employee's resume, using named entity recognition technology to extract specific skill items such as programming languages ​​or professional software names, and quantifying them into numerical proficiency values ​​using proficiency descriptive words such as proficiency, familiarity, and understanding appearing in the text context. The work experience value is calculated by calculating the cumulative years of work experience based on the start and end dates in the employee's work history, and weighted summation is performed based on the complexity and scale of the projects they participated in. The collaboration preference characteristics are comprehensively evaluated by analyzing the employee's historical communication patterns on the internal collaboration platform and the results of the role questionnaire survey on team projects, and transformed into a multi-dimensional feature vector. Skill requirements and levels are obtained from job description text through keyword extraction and semantic analysis. Experience requirements are obtained by parsing the years of experience explicitly required in the job qualifications section. Team culture characteristics are characterized by text sentiment analysis and theme modeling based on the publicly stated values ​​of the department to which the position belongs and records of team activities. Subsequently, feature extraction algorithms, mainly based on pre-trained models in the field of natural language processing, are used to semantically encode and classify the above unstructured text descriptions, transforming the free text information of employees and positions into standardized skill tags, experience level tags, and culture dimension tags. Finally, based on these tags, word embedding technology is used to map each discrete tag into a dense vector in a high-dimensional space. All tag vectors of the same individual or position are weighted and concatenated or mean pooled to generate fixed-length employee feature vectors and job requirement vectors, providing structured numerical input for subsequent matching algorithms.

[0020] In one embodiment, step S2, which uses an optimized matching algorithm to perform global exploration and self-organizing matching of employee attribute data and job requirement data to generate a preliminary matching set and pheromone concentration distribution, includes: S21. Construct a bipartite graph model with employee nodes and job nodes as elements, where the edges in the bipartite graph model represent the potential matching relationship between employees and jobs; S22. Initialize the pheromone concentration of each edge in the bipartite graph model, and calculate the heuristic factor of each edge based on the basic similarity between the employee feature vector and the job requirement vector. S23. Conduct multiple rounds of iterative exploration on the bipartite graph model; S24. Based on the current pheromone concentration and heuristic factor of each edge, generate multiple different matching sequences through a probability selection function. Each matching sequence consists of a set of non-conflicting edges, representing a matching scheme. S25. Calculate the weighted sum of similarity between the employee feature vectors and job requirement vectors of all matching pairs in the matching sequence to generate a comprehensive quality assessment. S26. Based on the comprehensive quality assessment results, update the pheromone concentration of each edge constituting the matching sequence, where the edges corresponding to high-quality matching sequences receive a larger pheromone concentration increment. S27. Preset the iteration termination condition. Based on the final pheromone concentration of each edge in the bipartite graph model, select the edges with the highest pheromone concentration for each job node, add the employee node corresponding to the edge to the candidate list of the job, integrate the candidate lists of all jobs to generate a preliminary matching set, and output the pheromone concentration distribution of each edge. As described in steps S21-S27 above, the system first constructs a bipartite graph model based on the employee feature vector and job requirement vector sets, where each employee and each job is treated as an independent node. Fully connected edges are established between all employee nodes and all job nodes to represent any potential matching relationship. During initialization, each edge is assigned the same and relatively small initial pheromone concentration value, and a heuristic factor for each edge is calculated. This factor is obtained by calculating the cosine similarity between the employee feature vector and the job requirement vector connected by the edge in the skill dimension. Subsequently, the system enters a multi-round iterative exploration phase. In each iteration, multiple agents are simulated to simultaneously explore paths. Each agent starts by randomly selecting an unmatched employee node, dynamically calculates the selection probability according to the pheromone concentration and heuristic factor of each edge in the current round using a roulette wheel selection method, and selects an edge to connect to the corresponding job node based on this probability. The employee and job are then marked as matched. This process is repeated until all employee nodes or one of the job nodes is fully matched, thus generating a complete matching sequence composed of multiple non-conflicting edges. The comprehensive quality assessment of this sequence is then performed using… The system achieves this by calculating the overall similarity between the feature vectors of each matched employee and job in the sequence. This overall similarity is a weighted sum of skill matching, experience fit, and the matching between employee collaboration preferences and job team culture characteristics. After evaluation, the pheromone concentration on each edge in the sequence is increased proportionally to the overall quality evaluation score. At the end of this iteration, the system applies a global evaporation operation to the pheromone concentration of all edges in the bipartite graph model, multiplying each pheromone concentration value by a evaporation coefficient less than one to simulate the natural evaporation of pheromones. The above iterative process continues until the preset maximum number of iterations is reached or the change in pheromone concentration distribution tends to stabilize. Finally, based on the converged pheromone concentration distribution, the system selects the edges with the highest pheromone concentration connected to each job node, usually selecting the top three edges. The employee nodes corresponding to these edges are included in the candidate list of the job. After summing the candidate lists of all jobs, a preliminary matching set is formed. At the same time, the final pheromone concentration value of each edge in the bipartite graph model is output as the pheromone concentration distribution.

[0021] In one embodiment, step S3, which uses a matching evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the preliminary matching set to generate an accurate matching scheme, includes: S31. For each matching pair in the preliminary matching set, input the matching evaluation model; S32. Map each skill item and its proficiency value in the employee's skill set to the corresponding tooth height and tooth width geometric parameters on the drive gear; S33. Map each skill requirement item and its level value in the set of job skill requirements to the corresponding tooth height and tooth width geometric parameters on the driven gear; S34. Calculate the matching degree of geometric parameters of all corresponding teeth on the driving gear and the driven gear, and generate the tooth profile matching degree. S35. Based on the logical dependencies between skill items, construct a series gear transmission chain, quantify the employee's work experience value as the input torque of the transmission chain, and simulate and calculate the mechanical transmission efficiency of the transmission chain. S36. Calculate the similarity between the employee collaboration preference feature vector and the job team culture feature vector, and map the similarity to a lubrication coefficient that reduces transmission loss. S37. Based on the preset weights, the skill tooth profile matching degree, mechanical transmission efficiency and lubrication coefficient are weighted and fused to calculate and generate the accurate matching degree score of the matching pair. S38. Aggregate all evaluated matching pairs and their exact match scores to generate an exact match solution; As described in steps S31-S38 above, the matching evaluation model processes each input employee-job matching pair independently. First, it performs skill tooth profile mapping, where the proficiency value of each skill item in the employee's skill set is directly mapped to a specific tooth height value on the corresponding tooth of the drive gear through a preset linear transformation function. The breadth or type of the skill item is mapped to a tooth width value. The process involves the system mapping the employee's skill data to the gear's geometric parameters using a predefined and definite numerical conversion standard. First, the system sets two basic mapping rules: one is to define a corresponding tooth height value range for skill proficiency, for example, mapping the lowest proficiency to a basic tooth height value and the highest proficiency to a maximum tooth height value, with all proficiency values ​​between the two defined as a range. The proficiency values ​​are linearly scaled proportionally within this range to ensure a stable and continuous proportional relationship between proficiency and tooth width. Secondly, a mapping table is created between skill types and tooth width values, where different skill types are pre-assigned fixed tooth width values ​​to represent the varying breadth of their roles in collaboration. In practice, for each skill item in an employee's skill set, the system performs two mapping operations simultaneously: firstly, it calculates a precise tooth height value based on the specific proficiency value according to the aforementioned linear scaling rule; secondly, it directly obtains a definite tooth width value by querying the mapping table based on the skill type. This process is repeated until all skill items are converted into corresponding tooth height and tooth width parameter pairs. These parameters collectively define the complete tooth profile of the driving gear representing the employee's competency profile. Each skill requirement item and its level value in the job skill requirement set are mapped to the tooth height and tooth width of the corresponding teeth on the driven gear according to the same rules. Then, the skill tooth profile matching degree is calculated by comparing the geometric parameters of the tooth height and tooth width of all corresponding teeth on the driving and driven gears item by item, calculating the matching ratio of each parameter, and taking the geometric mean of all ratios as the final matching overlap. Next, a series gear transmission chain model is constructed based on the pre-defined logical dependency relationship diagram between skill items. The employee's work experience value is quantified into the initial input torque of the transmission chain through a nonlinear function, and the torque transmission process in the transmission chain is simulated, considering the factors affecting each gear pair. The efficiency loss caused by tooth profile matching is used to calculate the final mechanical transmission efficiency. The steps are as follows: First, the system quantifies the employee's work experience value into the initial input torque of the transmission system according to a preset nonlinear transformation rule. This rule ensures that the torque increases rapidly when the experience value is in a low range, and gradually slows down as the experience value increases. For example, a logarithmic function or an exponential decay function is used for mapping, so that the torque corresponding to five years of experience may be significantly higher than that of one year of experience, but the torque corresponding to fifteen years of experience is not simply equivalent to the sum of three five-year experiences. After quantization, the system connects the gears representing each skill in sequence according to the preset logical dependencies between skill items to construct a transmission chain model. The starting end of the transmission chain receives the aforementioned initial input torque.In the simulated torque transmission process, the system calculates sequentially starting from the first pair of gears in the transmission chain. For each pair of meshing gears in the chain, the system obtains the pre-calculated tooth profile matching degree value and uses this value directly as the instantaneous efficiency coefficient of that stage of transmission. The torque is multiplied by the efficiency coefficient of each stage as it flows through each pair of gears, resulting in attenuation. The output torque of the previous stage becomes the input torque of the next stage. This process continues until the torque is transmitted to the end of the transmission chain. Finally, the mechanical transmission efficiency is calculated by dividing the output torque obtained at the end of the transmission chain by the initial input torque. This ratio directly reflects the effective output torque that the employee's experience can retain after experiencing all skill matching degree losses. The system calculates the ratio; simultaneously, it calculates the cosine similarity between the employee collaboration preference feature vector and the job team culture feature vector in multidimensional space, and directly maps this similarity value to a lubrication coefficient between 0 and 1 through a linear scaling function; finally, the model performs weighted summation on the calculated skill tooth profile matching degree, mechanical transmission efficiency, and lubrication coefficient according to a weight configuration file generated by a co-evolution mechanism, generating a precise matching degree score between 0 and 100; the system repeats the above complete process for all matching pairs in the initial matching set, and aggregates all scoring matching pairs according to job position and employee number, generating a structured precise matching scheme data table as output.

[0022] In one embodiment, step S4, which establishes a co-evolutionary mechanism to identify high-success matching patterns based on pheromone concentration distribution and dynamically adjusts the calculation weights of each evaluation coefficient in the matching evaluation model, includes: S41. From the pheromone concentration distribution, filter out the edges whose pheromone concentration values ​​are higher than the preset threshold, obtain the matching pairs corresponding to these edges, and form a set of high-success matching patterns. S42. Analyze each matching pair in the high-success matching pattern set and extract the three evaluation coefficients generated by the matching evaluation model: skill tooth profile matching degree, mechanical transmission efficiency, and lubrication coefficient. S43. Calculate the statistical contribution of each evaluation coefficient for all matching pairs in the high-success matching pattern set, wherein the statistical contribution of the evaluation coefficient is determined based on the correlation between the numerical distribution of the coefficient in the set and the exact match score. S44. Based on the statistical contribution of each evaluation coefficient, the preset weights of skill tooth profile matching degree, mechanical transmission efficiency and lubrication coefficient in the matching evaluation model are adjusted in the same direction, so that the evaluation coefficient with high statistical contribution will get higher calculation weight after the weight adjustment. As described in steps S41-S44 above, the co-evolution mechanism first automatically filters out all edges with pheromone concentration values ​​higher than a preset absolute threshold or relative quantile threshold from the global pheromone concentration distribution. The employee job matching pairs corresponding to these edges are then determined as historical successful patterns and constitute a set of high-success matching patterns. Subsequently, the system traverses each matching pair in this set and accurately extracts three specific values ​​generated by the pair in the most recent evaluation from the stored matching evaluation result records: skill tooth profile matching degree, mechanical transmission efficiency, and lubrication coefficient. Next, the system uses the Pearson correlation coefficient calculation method to analyze the data of all matching pairs in this set, calculating the correlation coefficients between the skill tooth profile matching degree numerical sequence and the corresponding precise matching degree score sequence, the correlation coefficients between the mechanical transmission efficiency numerical sequence and the score sequence, and the correlation coefficients between the lubrication coefficient numerical sequence and the score sequence. These three calculated correlation coefficients... The absolute value is directly defined as the initial statistical contribution of each evaluation coefficient. The system then normalizes these three initial statistical contributions to ensure their sum is 1.0, ultimately obtaining the standardized contribution weights of each coefficient. Finally, based on a preset learning rate parameter, the system synchronously adjusts the calculated weights of the currently stored skill tooth profile matching degree, mechanical transmission efficiency, and lubrication coefficient in the matching evaluation model according to their proportional relationship with their respective standardized contribution weights. Specifically, the new weight equals the old weight plus the learning rate multiplied by the difference between the standardized contribution and the old weight, thus ensuring that evaluation coefficients with higher contributions will have a larger weight in the final score in the next evaluation cycle. This completes a full cycle of co-evolutionary learning and parameter update. The preset learning rate parameter is a fixed small value set by the administrator during system initialization, typically selected between 0.01 and 0.1, for example, 0.05. The determination of this parameter is based on a balance between system stability and adaptation speed: a higher learning rate allows the model weights to adjust rapidly based on feedback, suitable for scenarios where matching criteria need to quickly adapt to changes in business conditions, but may lead to fluctuations in evaluation results; a lower learning rate allows the weights to evolve slowly and smoothly, suitable for scenarios that prioritize stability and avoid overfitting to historical noise, but may result in a slow learning speed. This parameter is a fixed value and does not change automatically during the iteration process, ensuring that the weight adjustment process is controllable and predictable. That is, if the quality of historical matching data is high and the business environment changes rapidly, a slightly higher learning rate, such as 0.1, can be selected; if there is a lot of data noise or the business environment is stable, a lower learning rate, such as 0.01, should be selected. In the initial stage of system deployment, an empirical value, such as 0.05, is usually used as the default value, and can then be fine-tuned based on the stability and speed of model convergence during actual operation.

[0023] In one embodiment, step S5, which involves selecting candidate teams from a precise matching scheme based on preset team building requirements and using a team collaboration simulation model to predict the effectiveness of the candidate teams and generate a team matching scheme, includes: S51. Pre-defined team building requirements include job roles, number of personnel, and collaborative relationship network; S52. Based on the team building requirements, select at least one employee with the highest precision match score for each job role from the precise matching scheme to form a candidate team; S53. Input the candidate teams into the team collaboration simulation model and output the team matching scheme; As described in steps S51-S53 above, the system first receives preset team building requirements. The job roles define the required functional categories for the team, such as software development engineers and product managers. The number of personnel specifies the exact number of people required for each role, such as two engineers and one manager. The collaboration relationship network is defined in the form of a directed graph, specifying the reporting relationships and task delivery processes between roles. Based on these requirements, the system iterates through the data tables in the precise matching scheme. For each job role listed in the building requirements, it retrieves all employees labeled as that role and their precise matching scores, sorts them from highest to lowest score, and automatically selects the specified number of employees with the highest ranking for each role, thus forming a structured candidate team list. Subsequently, this candidate team is input into the team collaboration simulation model. The model first generates a corresponding personal gear model for each member of the team. The member's work experience value is mapped to the gear's tooth width parameter through a linear function; the total number of skills in the member's skill set is mapped to the gear's tooth count parameter; and the member's collaboration preference feature vector is mapped to the friction coefficient parameter that affects the gear's meshing surface friction behavior. Then the model... Based on the defined collaborative relationship network in the team building requirements, the model connects the individual gear models of each member according to their corresponding role relationships to construct a complete team gear transmission topology. In this topology, for each pair of gears with a direct connection, the model quantifies their collaborative compatibility by calculating the cosine similarity of their collaborative preference feature vectors. At the start of the simulation, the model converts an estimated total team task into a standardized input torque value and loads it onto the driving gear mapped by the team leader role. The system then simulates the step-by-step transmission process of this torque in the transmission topology. Each time the torque flows through a pair of meshing gears, its value is attenuated according to the collaborative compatibility of that gear pair; the lower the collaborative compatibility, the greater the torque loss. After the torque is transmitted to all the output gears at the end of the transmission chain, the model summarizes the remaining torque values ​​of all the end gears as the total output. By calculating the ratio of the total output torque to the initial input torque, the predicted collaborative efficiency value of the candidate team is finally obtained. By integrating the collaborative compatibility of each gear pair in the above transmission topology and the predicted collaborative efficiency value, the system generates a team matching scheme that includes quantitative performance prediction and structural analysis.

[0024] In one embodiment, step S53, which involves inputting candidate teams into a team collaboration simulation model and outputting a team matching solution, includes: S531. Map each member of the candidate team to a member's personal gear in the team gear set; S532. Map the members' work experience value to the tooth width of the members' personal gear; S533. Map the number of skills in a member's skill set to the number of teeth on the member's personal gear. S534. Map the members' collaborative preference characteristics to the friction coefficient of the individual gear teeth of each member; S535. Based on the collaborative relationship network defined in the team building requirements, determine the meshing connection relationship between the individual gears of each member in the team gear set, and construct the team transmission topology. S536. Quantify the total team task as input torque and load it onto the team transmission topology; S537. Calculate the predicted collaborative efficiency value of the candidate team based on the ratio of the sum of the torques of the output gears at the end of the transmission topology to the input torque. S538: Integrate team transmission topology and predicted collaboration efficiency values ​​to generate team matching schemes; As described in steps S531-S538 above, in the team collaboration simulation model, the system first generates a corresponding personal gear for each member of the candidate team. The member's work experience value is mapped to the gear's tooth width parameter through a preset linear transformation rule; for example, 0 to 20 years of experience linearly corresponds to a tooth width of 5 mm to 15 mm. The total number of skills in the member's skill set is directly mapped to the number of teeth on the gear, with one skill corresponding to one tooth. Simultaneously, the member's collaboration preference feature vector is processed into a scalar value between 0 and 1 through a normalization function, and this scalar value is directly defined as the friction coefficient of the gear's tooth surface. After parameterizing all members' personal gears, the system, based on the collaboration relationship network explicitly defined in the team building requirements (usually defined as an adjacency matrix or edge list), meshes and connects each member's personal gear according to the connection relationship of their corresponding role in the matrix, thereby constructing a complete team transmission topology. Subsequently, the system completes the team... The total task volume of the team, such as the number of person-days in the project, is quantified into a specific input torque value through a predefined conversion coefficient, and this torque is loaded onto the drive gear marked as the starting point in the transmission topology. During the simulation, the torque is transmitted sequentially along the path defined by the transmission topology. For each pair of meshing gears, the system calculates the cooperation compatibility of the gear pair in real time. This compatibility is obtained by comparing the cosine similarity of the cooperation preference feature vectors of the members represented by the two gears. The transmission efficiency of the torque is equal to the compatibility value, so the output torque is equal to the input torque multiplied by the compatibility. The torque is finally transmitted to the end output gear after passing through all meshing paths. The system summarizes the torque values ​​of all end gears to obtain the total output torque. Finally, the predicted cooperation efficiency value of the candidate team is obtained by calculating the ratio of the total output torque to the initial input torque. The system integrates the complete transmission topology structure, which includes all individual gear parameters, gear meshing relationships, cooperation compatibility, and predicted cooperation efficiency values, into a structured data report, thus generating the final team matching scheme.

[0025] Example 2, please refer to Figure 2 As shown in this embodiment, a human resource matching system based on machine learning includes: Data acquisition module: Acquires employee attribute data and job requirement data; Job matching module: Connected to the data acquisition module, it uses an optimized matching algorithm to perform global exploration and self-organized matching of employee attribute data and job requirement data, generating a preliminary matching set and pheromone concentration distribution; Matching and evaluation module: Connected to the job matching module, it uses a matching and evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the initial matching set, and generates an accurate matching plan; Parameter update module: Connects with the matching evaluation module to establish a co-evolution mechanism, identifies high-success matching patterns based on pheromone concentration distribution, and dynamically adjusts the calculation weights of each evaluation coefficient in the matching evaluation model; Team matching module: Connected to the parameter update module, it selects candidate teams from the precise matching schemes based on preset team building requirements, and uses a team collaboration simulation model to predict the effectiveness of the candidate teams and generate team matching schemes.

[0026] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A human resource matching method based on machine learning, characterized in that, Obtain employee attribute data and job requirement data; An optimized matching algorithm is used to perform global exploration and self-organizing matching of employee attribute data and job requirement data to generate a preliminary matching set and pheromone concentration distribution. A matching evaluation model is used to quantitatively verify and evaluate the accuracy of each matching pair in the preliminary matching set, and to generate an accurate matching scheme. Establish a collaborative evolution mechanism to identify high-success matching patterns based on pheromone concentration distribution and dynamically adjust the calculation weights of each evaluation coefficient in the matching evaluation model; Based on preset team building requirements, candidate teams are selected from the precise matching schemes, and a team collaboration simulation model is used to predict the effectiveness of the candidate teams and generate team matching schemes.

2. The human resource matching method based on machine learning according to claim 1, characterized in that, The steps for obtaining employee attribute data and job requirement data are as follows: Employee attribute data includes skill sets, work experience values, and collaboration preference characteristics; Job requirements data includes a set of skill requirements, experience requirements, and team culture characteristics; The feature extraction algorithm is used to structure employee attribute data and job requirement data to generate employee feature tags and job feature tags; Employee feature vectors and job requirement vectors are generated based on employee feature tags and job feature tags.

3. The human resource matching method based on machine learning according to claim 1, characterized in that, The steps involved in using an optimized matching algorithm to perform global exploration and self-organizing matching of employee attribute data and job requirement data, and to generate a preliminary matching set and pheromone concentration distribution, are as follows: Construct a bipartite graph model with employee nodes and job nodes as elements, where the edges in the bipartite graph model represent the potential matching relationship between employees and jobs; Initialize the pheromone concentration of each edge in the bipartite graph model, and calculate the heuristic factor of each edge based on the basic similarity between the employee feature vector and the job requirement vector; Multiple rounds of iterative exploration were conducted on the bipartite graph model; Based on the current pheromone concentration and heuristic factor of each edge, multiple different matching sequences are generated through a probability selection function. Each matching sequence consists of a set of non-conflicting edges, representing a matching scheme. Calculate the weighted sum of similarity between the employee feature vectors and the job requirement vectors of all matching pairs in the matching sequence to generate a comprehensive quality assessment. Based on the comprehensive quality assessment results, the pheromone concentration of each edge constituting the matching sequence is updated, with edges corresponding to high-quality matching sequences receiving a larger pheromone concentration increment; The iteration termination condition is preset. Based on the final pheromone concentration of each edge in the bipartite graph model, the edges with the highest pheromone concentration are selected for each job node. The employee nodes corresponding to the edges are added to the candidate list of the job. The candidate lists of all jobs are integrated to generate a preliminary matching set, and the pheromone concentration values ​​of each edge are output as the pheromone concentration distribution.

4. The human resource matching method based on machine learning according to claim 1, characterized in that, The steps for using a matching evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the preliminary matching set, and to generate an accurate matching scheme, are as follows: For each matching pair in the initial matching set, input the matching evaluation model; Each skill item and its proficiency value in the employee's skill set is mapped to the corresponding tooth height and tooth width geometric parameters on the drive gear; Map each skill requirement item and its level value in the set of job skill requirements to the corresponding tooth height and tooth width geometric parameters on the driven gear; Calculate the matching degree of geometric parameters of all corresponding teeth on the driving gear and the driven gear, and generate the skill tooth profile matching degree; Based on the logical dependencies between skill items, a series gear transmission chain is constructed, and the employee's work experience value is quantified as the input torque of the transmission chain to simulate and calculate the mechanical transmission efficiency of the transmission chain. Calculate the similarity between the employee collaboration preference feature vector and the job team culture feature vector, and map this similarity to a lubrication coefficient that reduces transmission loss; Based on preset weights, the skill tooth profile matching degree, mechanical transmission efficiency and lubrication coefficient are weighted and fused to calculate and generate the accurate matching degree score of the matching pair. Aggregate all evaluated matching pairs and their exact match scores to generate an exact match solution.

5. The human resource matching method based on machine learning according to claim 1, characterized in that, The steps for establishing a co-evolutionary mechanism, identifying high-success matching patterns based on pheromone concentration distribution, and dynamically adjusting the calculation weights of each evaluation coefficient in the matching evaluation model are as follows: From the pheromone concentration distribution, edges with pheromone concentration values ​​higher than a preset threshold are selected, and matching pairs corresponding to these edges are obtained to form a set of high-success matching patterns. Analyze each matching pair in the high-success matching pattern set and extract three evaluation coefficients generated by the matching evaluation model: skill tooth profile matching degree, mechanical transmission efficiency, and lubrication coefficient. Calculate the statistical contribution of each evaluation coefficient for all matching pairs in the high-success matching pattern set, wherein the statistical contribution of the evaluation coefficient is determined based on the correlation between the numerical distribution of the coefficient in the set and the exact match score. Based on the statistical contribution of each evaluation coefficient, the preset weights of skill tooth profile matching degree, mechanical transmission efficiency and lubrication coefficient in the matching evaluation model are adjusted in the same direction, so that the evaluation coefficient with high statistical contribution will receive higher calculation weight after the weight adjustment.

6. The human resource matching method based on machine learning according to claim 1, characterized in that, The steps for selecting candidate teams from precise matching schemes based on preset team building requirements, and generating team matching schemes by using a team collaboration simulation model to predict the effectiveness of candidate teams are as follows: Pre-defined team building requirements include job roles, number of personnel, and collaborative relationship network; Based on the team building requirements, at least one employee with the highest precision match score for each job role is selected from the precise matching scheme to form a candidate team; Input the candidate teams into the team collaboration simulation model and output the team matching scheme.

7. The human resource matching method based on machine learning according to claim 1, characterized in that, The steps for inputting candidate teams into the team collaboration simulation model and outputting team matching solutions are as follows: Map each member of the candidate team to a member's individual gear in the team gear set; Map members' work experience values ​​to the tooth width of their individual gears; The number of skills in a member's skill set is mapped to the number of teeth on the member's personal gear. The collaborative preference characteristics of members are mapped to the friction coefficient of the individual gear teeth of each member; Based on the collaborative relationship network defined in the team building requirements, the meshing connection relationship between the individual gears of each member in the team gear set is determined, and the team transmission topology is constructed. The total amount of team tasks is quantified as input torque and loaded onto the team transmission topology; The predicted collaborative efficiency value of the candidate team is calculated based on the ratio of the sum of the torques of the output gears at the end of the transmission topology to the input torque. By integrating team transmission topology and predicted collaboration efficiency values, a team matching scheme is generated.

8. A machine learning-based human resource matching system, used to implement the machine learning-based human resource matching method according to any one of claims 1-7, characterized in that, Data acquisition module: Acquires employee attribute data and job requirement data; Job matching module: Connected to the data acquisition module, it uses an optimized matching algorithm to perform global exploration and self-organized matching of employee attribute data and job requirement data, generating a preliminary matching set and pheromone concentration distribution; Matching and evaluation module: Connected to the job matching module, it uses a matching and evaluation model to quantitatively verify and evaluate the accuracy of each matching pair in the initial matching set, and generates an accurate matching plan; Parameter update module: Connects with the matching evaluation module to establish a co-evolution mechanism, identifies high-success matching patterns based on pheromone concentration distribution, and dynamically adjusts the calculation weights of each evaluation coefficient in the matching evaluation model; Team matching module: Connected to the parameter update module, it selects candidate teams from the precise matching schemes based on preset team building requirements, and uses a team collaboration simulation model to predict the effectiveness of the candidate teams and generate team matching schemes.

Images