Intelligent man-post matching method based on deep learning

Abstract

The invention discloses an intelligent person and post matching method based on deep learning, and the method comprises the following steps: S1, collecting candidate person information and post requirement information, and carrying out the preprocessing and feature extraction; s2, successfully matched cases are extracted from historical recruitment data to train an imitation learning model, an imitation agent is generated, and an expert behavior mode is learned through the imitation agent; s3, initializing eagle individuals, and evaluating a matching effect of each eagle individual based on a fitness function; s4, in a global search stage, realizing cooperative work of the individual eagles by using an eagle group cluster strategy, and optimizing a global matching effect based on a matching evaluation result and an information sharing mechanism; s5, further optimizing the matching strategy of the eagle individuals by adopting a local search mode; and S6, when the change of the recruitment market is detected, triggering a dynamic adaptation mechanism, and finally generating an optimal man-post matching result. According to the method, deep learning, imitation learning and eagle group clustering strategies are fully utilized, and intelligent man-post matching optimization is realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recruitment technology, and in particular to an intelligent method for person-job matching based on deep learning. Background Art

[0002] In the field of human resource management and recruitment, person-job matching has always been a key link for enterprises to optimize talent management, improve work efficiency and reduce recruitment costs. Traditional person-job matching methods mainly rely on manual resume screening, interview evaluation and rule-based screening systems. These methods can meet basic recruitment needs to a certain extent, but with the intensification of market competition, the increase in talent mobility and the increasing complexity of job requirements, the limitations of traditional methods are gradually emerging.

[0003] In the traditional job matching method, companies usually conduct preliminary screening of candidates through hard conditions such as keyword matching, years of experience screening, and educational requirements. Although this method can quickly screen out candidates who do not meet the basic requirements, its biggest problem is that it over-relies on hard skill matching and ignores soft skills, cultural adaptability, psychological motivation and other factors that are crucial to long-term employment relationships. In addition, traditional matching methods often lack dynamic adjustment capabilities. Once the company's recruitment needs change, such as adjustments to job responsibilities and changes in team culture, the existing matching system cannot quickly adapt to these changes, resulting in inefficient recruitment processes and low matching quality.

[0004] In recent years, machine learning and data-driven methods have been gradually applied to job matching, mainly relying on big data analysis, statistical modeling, and simple classification algorithms. For example, some recruitment platforms and human resource management systems have begun to use algorithms based on logistic regression, random forests, and support vector machines (SVM) to match candidates with positions. Compared with traditional manual screening methods, these methods can improve the level of automation to a certain extent and provide more objective matching suggestions based on data. However, such methods still have several defects. First, many machine learning models mainly rely on structured data, such as work experience, education, and skill lists, when dealing with job matching tasks, and it is difficult to effectively process unstructured data such as candidate resumes, interview records, and social media information. In addition, many existing methods adopt static matching strategies and cannot make adaptive adjustments as the market changes, resulting in the inability to update matching rules in a timely manner when the recruitment market fluctuates.

[0005] With the development of deep learning technology, especially the progress of natural language processing (NLP) and neural networks, more and more studies have tried to apply deep learning to job matching. For example, some methods use convolutional neural networks (CNN) or long short-term memory networks (LSTM) to process text data, extract candidate resume features, job description features, and perform similarity matching. These methods have improved the matching accuracy to a certain extent, but there are still many problems. On the one hand, these methods only rely on static text matching between candidates and positions, and do not take into account the dynamics of the recruitment market, that is, changes in job requirements, trends in candidate skill development, etc. In addition, although deep learning models can extract features, traditional supervised learning methods still require a large amount of labeled historical recruitment data for training, and high-quality recruitment data is difficult to obtain, and existing models cannot fully utilize the matching experience of experts.

[0006] Therefore, how to provide an intelligent method for person-job matching based on deep learning is an issue that technical personnel in this field urgently need to solve. Summary of the invention

[0007] One purpose of the present invention is to propose an intelligent method for people-job matching based on deep learning. The present invention makes full use of deep learning, imitation learning and eagle swarm clustering strategy to achieve intelligent people-job matching optimization. The multi-dimensional characteristics of candidates and positions are extracted through deep neural networks, and the matching accuracy is improved by combining imitation agent learning expert matching strategy. The introduction of eagle swarm clustering strategy improves global search and local optimization capabilities to avoid matching falling into local optimality. A dynamic adaptation mechanism is adopted to monitor changes in the recruitment market in real time, and adjust the matching strategy to adapt to new needs. Compared with traditional methods, the present invention has the advantages of high matching accuracy, strong optimization ability, fast adaptation to market changes and high intelligence level.

[0008] According to an embodiment of the present invention, the intelligent method for matching people with jobs based on deep learning includes the following steps:

[0009] S1. Collect candidate information and job requirement information, perform preprocessing and feature extraction, and generate candidate feature vectors and job feature vectors;

[0010] S2. Extract successful matching cases from historical recruitment data to train the imitation learning model, generate imitation agents, and learn expert behavior patterns through the imitation agents to obtain preliminary person-job matching strategies;

[0011] S3. Initialize the falcon individuals according to the candidate feature vector, the job feature vector and the preliminary person-job matching strategy. Each falcon individual represents the possibility of a person-job matching. The matching effect of each falcon individual is evaluated based on the fitness function to obtain the matching evaluation result.

[0012] S4. In the global search phase, the falcon cluster strategy is used to achieve the collaborative work of individual falcons. Based on the matching evaluation results and information sharing mechanism, the falcons are explored in the entire search space, and the optimal matching information is transmitted to other falcon individuals to optimize the global matching effect.

[0013] S5. Based on the feedback provided by the imitating agent, a local search method is used to further optimize the matching strategy of the falcon individuals, including skill matching, cultural adaptation and psychological motivation;

[0014] S6. Monitor changes in the recruitment market. When changes in the recruitment market are detected, the dynamic adaptation mechanism is triggered. The imitation agent updates the matching strategy based on the latest market data, and adjusts the search direction and strategy of the falcon individual according to the changes in the recruitment market to generate the optimal person-job matching results.

[0015] Optionally, the S2 specifically includes:

[0016] S21. Filter successfully matched case samples from historical recruitment data, annotate the candidate feature vector and job feature vector in each case, and form a training data set;

[0017] S22, performing data cleaning and missing value processing on the training data set, and using a unified standardization method to map all features to the same numerical range to obtain an input data set;

[0018] S23, constructing a deep neural network of an imitation learning model, wherein the input layer is used to receive candidate feature vectors and job feature vectors, the hidden layer uses multi-layer nonlinear transformation to extract expert decision patterns, and the output layer generates matching score predictions;

[0019] S24. Based on the multi-dimensional characteristics between candidates and positions, a comprehensive matching scoring function is defined. The comprehensive matching scoring function is composed of skill similarity, cultural similarity, psychological motivation similarity and interaction terms:

[0020] S m =α·sim(V c ,V p )+β·sim(C c ,C p )+γ·sim(M c ,M p )+

[0021] δ·(sim(V c ,C c )×sim(V p ,C p ));

[0022] Among them, S m represents the comprehensive matching score, Vc represents the skill characteristics of the candidate, C c Indicates the cultural characteristics of the candidate, M c Represents the psychological motivation characteristics of the candidate, V p Indicates the skill characteristics required for the position, C p Indicates the cultural characteristics required by the position, M p represents the psychological motivation characteristics of the job requirements, sim represents the similarity function, α, β, γ and δ represent weight coefficients, and satisfy α+β+γ+δ=1;

[0023] S25. Use the back propagation algorithm to train the imitation learning model and define the loss function L to measure the difference between the predicted match score and the historical successful match score in the form of a combination of weighted mean square error and regularization:

[0024]

[0025] Where N represents the number of training samples, w i represents the weighting factor of the i-th training sample, represents the i-th matching score predicted by the imitation learning model, represents the successful matching score of the expert in the i-th training sample, λ represents the regularization coefficient, and θ k represents the training parameters of the k-th layer network, and K represents the total number of layers;

[0026] S26. Encapsulate the trained imitation learning model into an imitation agent, and obtain a preliminary person-job matching strategy by inputting new candidate feature vectors and job feature vectors.

[0027] Optionally, the S3 specifically includes:

[0028] S31, based on the candidate feature vector C and the job feature vector P, combined with the preliminary person-job matching strategy, initialize the falcon individuals, each falcon individual represents the possibility of a person-job matching, and construct the falcon individual set H = {H 1 ,H 2 ,…,H M}, where H i represents the i-th falcon individual, M represents the total number of falcon individuals;

[0029] S32. Initialize the matching parameters of each falcon individual using Gaussian distribution:

[0030] X i =X 0 +σ·N(0,I);

[0031] Among them, X i represents the matching parameter vector of the i-th falcon individual, X 0represents the matching parameter generated by the preliminary job matching strategy, σ represents the disturbance coefficient, and N(0,I) represents the normal distribution with zero mean and unit variance;

[0032] S33. For each falcon individual, a matching evaluation model based on the fitness function is constructed. The fitness function is composed of skill fitness, cultural fitness, and psychological motivation fitness:

[0033] F(H i )=λ 1 f skill (C,P,H i )+λ 2 f culture (C,P,H i )+λ 3 f motive (C,P,H i );

[0034] Among them, F(H i ) represents the comprehensive fitness of the i-th falcon individual, λ 1 , 2 and λ 3 represents the weight factor, satisfying λ 1 +λ 2 +λ 3 =1;

[0035]

[0036] Among them, f skill (C,P,H i ) represents the skill fitness of the i-th falcon individual, v i,j represents the candidate's score on the jth skill, p i,j represents the demand score of the position on the jth skill, and m represents the total number of skills;

[0037]

[0038] Among them, f culture (C,P,H i ) represents the cultural fitness of the i-th falcon individual, c i,p represents the candidate's preference value on the pth cultural factor, p i,p represents the adaptation requirement of the position on the pth cultural factor, m 1 represents the total number of cultural factors;

[0039]

[0040] Among them, f motive (C,P,H i ) represents the psychological motivation fitness of the i-th falcon individual, mi,q represents the candidate's inclination on the qth psychological motivation factor, p i,q represents the adaptation requirement of the position on the qth psychological motivation factor, m 2 Indicates the total number of psychological motivational factors;

[0041] S34, after calculating the comprehensive fitness value of each falcon individual, the falcon individuals are screened probabilistically, and the entropy value of the fitness distribution of the falcon individuals is calculated using the matching entropy regularization method to evaluate the quality of the overall matching solution;

[0042] S35. Based on the comprehensive fitness evaluation results of the screened falcon individuals, a matching evaluation result is obtained.

[0043] Optionally, the S4 specifically includes:

[0044] S41, based on the falcon individual set and comprehensive fitness F(H i ) constructs the global search space Ω:

[0045] Ω={X|X min ≤X≤X max};

[0046] Where X represents the matching parameter vector of the falcon individual, X min represents the minimum boundary of the search space, X max represents the maximum boundary of the search space;

[0047] S42. In each round of iteration, the individual with the highest comprehensive fitness is selected as the global guiding individual, and the optimal matching information is propagated through the dynamic information sharing mechanism. The matching parameter update process of the falcon individual is:

[0048]

[0049] in, represents the matching parameter vector of the i-th falcon individual in the t+1th round, represents the matching parameter vector of the i-th falcon individual in the tth round, w i represents the weight of the i-th falcon individual, represents the matching parameter vector of the current optimal individual, N(0,I) represents the normal distribution with zero mean and unit variance, μ 1 represents the global adjustment factor, μ 2 represents the random exploration factor;

[0050] S43. In the global search process, a Levy flight mechanism is introduced to enhance the coverage of the search space, so that the falcon individuals can perform long-distance searches. The Levy flight mechanism generates random step lengths using Levy distribution, and the falcon individuals randomly select long-distance jump searches to avoid falling into local optimality.

[0051] S44, passing the matching results after the global search to the local optimization stage through the information sharing mechanism.

[0052] Optionally, the S5 specifically includes:

[0053] S51, selecting falcon individuals whose fitness is higher than a set threshold from the falcon individual set obtained in the global search phase, and constructing a matching optimization candidate set in combination with the historical matching data of the imitation agent;

[0054] S52. In the skill matching optimization stage, based on the successful matching cases provided by the imitation agent, analyze the differences between the candidate's skill characteristics and the job requirements, and optimize the skill weight allocation;

[0055] S53. In the cultural adaptation optimization stage, based on the feedback information on cultural matching from the imitation agent, the cultural adaptability of the falcon individual is partially corrected. By adjusting the influencing factors of the cultural adaptation vector, the matching plan is made more in line with the organizational culture, values ​​and team collaboration methods of the company where the position is located, thereby improving the fit between the candidate and the corporate culture.

[0056] S54. In the psychological motivation matching optimization stage, combined with the matching success cases analyzed by the imitation agent, the psychological motivation adaptability of the falcon individual is dynamically adjusted. By adjusting the psychological motivation weight in the matching strategy, the candidate's career development goals, work style preferences, and incentive mechanism needs are more consistent with the job requirements;

[0057] S55. During the local search process, the optimized matching results are evaluated in real time according to the feedback information of the imitation agent, and the optimized matching results are stored in the historical matching database.

[0058] Optionally, the S6 specifically includes:

[0059] S61. Continuously monitor recruitment market data, collect information including job requirements, candidate skill changes, industry trends and corporate culture changes, and build a market dynamics data set;

[0060] S62. Perform layered processing on the market dynamics data set, extract market change indicators, and generate a market change vector R d =(r 1 ,r 2 ,…,r Y ), where r i represents the quantitative value of the i-th market change indicator, and Y represents the total number of market change indicators;

[0061] S63: When the market change threshold is detected to exceed the preset value T m After that, the dynamic adaptation mechanism is triggered, and the imitation agent reads the market change vector Rd , and update the matching strategy parameter set:

[0062] S new =S old +λ m ·R d ;

[0063] Among them, S new represents the updated matching strategy parameters, S old represents the original matching strategy parameters, λ m represents the market adaptation factor;

[0064] S64, according to the updated matching strategy parameter S new , adjust the search direction of the falcon individual set. During the adjustment process, according to the comprehensive fitness F(H i ) to re-evaluate, so that the falcon individuals move closer to the optimal matching solution, and finally form an optimized falcon individual set;

[0065] S65. Based on the optimized falcon individual set, the match with the highest comprehensive fitness is selected as the optimal person-job matching result.

[0066] The beneficial effects of the present invention are:

[0067] First, the present invention is optimized at the feature extraction level, not only focusing on the hard skill matching between candidates and positions, but also introducing soft factors such as cultural adaptability and psychological motivation. Multi-dimensional features are extracted through deep neural networks (such as LSTM, CNN, etc.), making the matching process more comprehensive, thereby effectively avoiding the limitations of traditional rule-based or keyword matching methods. At the same time, the present invention uses imitation learning to construct an imitation agent, so that the system can learn the matching decision-making model of experts, thereby reducing dependence on large amounts of labeled data, and generating high-quality matching strategies in the early matching stage, providing a better search starting point for subsequent optimization.

[0068] Secondly, the present invention effectively improves the global search capability by introducing the hawk group clustering strategy. The falcon individuals adopt a group collaboration mechanism during the search process, so that the matching process not only relies on individual exploration, but also can improve the overall search efficiency through information sharing to avoid falling into the local optimum. In addition, in the local optimization stage, combined with the feedback information of the imitation agent, the falcon individuals can dynamically adjust the matching strategy according to the historical matching data, so that the optimization not only stays at the static matching level, but can be deeply optimized according to multi-dimensional characteristics such as skills, cultural adaptability and psychological motivation, thereby improving the accuracy and long-term stability of job matching.

[0069] In addition, the present invention has a strong ability to adapt to the market. Traditional job matching systems often use static rules for matching, which cannot cope with the rapid changes in the recruitment market. The present invention, through a dynamic adaptation mechanism, can monitor changes in market demand in real time, such as adjustments to job requirements, evolution of candidate skill trends, changes in industry recruitment standards, etc., and automatically adjust matching strategies to optimize the search direction and strategy of individual falcons. This mechanism ensures that the system can continuously optimize matching solutions as the market environment changes, thereby improving the effectiveness of recruitment matching and avoiding the problem of matching failure due to market changes.

[0070] Finally, the present invention not only improves the intelligent level of job matching, but also optimizes the long-term adaptability of the matching system. By combining global search, local optimization and market dynamic adaptation, the present invention establishes a more accurate matching mechanism between candidates and jobs, thereby improving recruitment efficiency, reducing recruitment costs, optimizing the human resource management process of enterprises, and ultimately promoting the further development of intelligent recruitment technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0072] Figure 1 This is the overall flow chart of the deep learning-based intelligent method for person-job matching proposed in the present invention. DETAILED DESCRIPTION

[0073] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0074] refer to Figure 1 , the intelligent method of job matching based on deep learning includes the following steps:

[0075] S1. Collect candidate information and job requirement information, perform preprocessing and feature extraction, and generate candidate feature vectors and job feature vectors;

[0076] S2. Extract successful matching cases from historical recruitment data to train the imitation learning model, generate imitation agents, and learn expert behavior patterns through the imitation agents to obtain preliminary person-job matching strategies;

[0077] S3. Initialize the falcon individuals according to the candidate feature vector, the job feature vector and the preliminary person-job matching strategy. Each falcon individual represents the possibility of a person-job matching. The matching effect of each falcon individual is evaluated based on the fitness function to obtain the matching evaluation result.

[0078] S4. In the global search phase, the falcon cluster strategy is used to achieve the collaborative work of individual falcons. Based on the matching evaluation results and information sharing mechanism, the falcons are explored in the entire search space, and the optimal matching information is transmitted to other falcon individuals to optimize the global matching effect.

[0079] S5. Based on the feedback provided by the imitating agent, a local search method is used to further optimize the matching strategy of the falcon individuals, including skill matching, cultural adaptation and psychological motivation;

[0080] S6. Monitor changes in the recruitment market. When changes in the recruitment market are detected, the dynamic adaptation mechanism is triggered. The imitation agent updates the matching strategy based on the latest market data, and adjusts the search direction and strategy of the falcon individual according to the changes in the recruitment market to generate the optimal person-job matching results.

[0081] In this implementation, S2 specifically includes:

[0082] S21. Filter successfully matched case samples from historical recruitment data, annotate the candidate feature vector and job feature vector in each case, and form a training data set;

[0083] S22, performing data cleaning and missing value processing on the training data set, and using a unified standardization method to map all features to the same numerical range to obtain an input data set;

[0084] S23, constructing a deep neural network of an imitation learning model, wherein the input layer is used to receive candidate feature vectors and job feature vectors, the hidden layer uses multi-layer nonlinear transformation to extract expert decision patterns, and the output layer generates matching score predictions;

[0085] S24. Based on the multi-dimensional characteristics between candidates and positions, a comprehensive matching scoring function is defined. The comprehensive matching scoring function is composed of skill similarity, cultural similarity, psychological motivation similarity and interaction terms:

[0086] S m =α·sim(V c ,V p )+β·sim(C c ,C p )+γ·sim(M c ,M p )+

[0087] δ·(sim(V c ,C c )×sim(V p ,C p ));

[0088] Among them, S m represents the comprehensive matching score, Vc represents the skill characteristics of the candidate, C c Indicates the cultural characteristics of the candidate, M c Represents the psychological motivation characteristics of the candidate, V p Indicates the skill characteristics required for the position, C p Indicates the cultural characteristics of the job requirements, M p represents the psychological motivation characteristics of the job requirements, sim represents the similarity function, α, β, γ and δ represent weight coefficients, and satisfy α+β+γ+δ=1;

[0089] S25. Use the back propagation algorithm to train the imitation learning model and define the loss function L to measure the difference between the predicted match score and the historical successful match score in the form of a combination of weighted mean square error and regularization:

[0090]

[0091] Where N represents the number of training samples, w i represents the weighting factor of the i-th training sample, represents the i-th matching score predicted by the imitation learning model, represents the successful matching score of the expert in the i-th training sample, λ represents the regularization coefficient, and θ k represents the training parameters of the k-th layer network, and K represents the total number of layers;

[0092] S26. Encapsulate the trained imitation learning model into an imitation agent, and obtain a preliminary person-job matching strategy by inputting new candidate feature vectors and job feature vectors.

[0093] In this implementation, S3 specifically includes:

[0094] S31, based on the candidate feature vector C and the job feature vector P, combined with the preliminary person-job matching strategy, initialize the falcon individuals, each falcon individual represents the possibility of a person-job matching, and construct the falcon individual set H = {H 1 ,H 2 ,…,H M}, where H i represents the i-th falcon individual, M represents the total number of falcon individuals;

[0095] S32. Initialize the matching parameters of each falcon individual using Gaussian distribution:

[0096] X i =X 0 +σ·N(0,I);

[0097] Among them, X i represents the matching parameter vector of the i-th falcon individual, X 0represents the matching parameter generated by the preliminary job matching strategy, σ represents the disturbance coefficient, and N(0,I) represents the normal distribution with zero mean and unit variance;

[0098] S33. For each falcon individual, a matching evaluation model based on the fitness function is constructed. The fitness function is composed of skill fitness, cultural fitness, and psychological motivation fitness:

[0099] F(H i )=λ 1 f skill (C,P,H i )+λ 2 f culture (C,P,H i )+λ 3 f motive (C,P,H i );

[0100] Among them, F(H i ) represents the comprehensive fitness of the i-th falcon individual, λ 1 , 2 and λ 3 represents the weight factor, satisfying λ 1 +λ 2 +λ 3 =1;

[0101]

[0102] Among them, f skill (C,P,H i ) represents the skill fitness of the i-th falcon individual, v i,j represents the candidate's score on the jth skill, p i,j represents the demand score of the position on the jth skill, and m represents the total number of skills;

[0103]

[0104] Among them, f culture (C,P,H i ) represents the cultural fitness of the i-th falcon individual, c i,p represents the candidate's preference value on the pth cultural factor, p i,p represents the adaptation requirement of the position on the pth cultural factor, m 1 represents the total number of cultural factors;

[0105]

[0106] Among them, f motive (C,P,H i ) represents the psychological motivation fitness of the i-th falcon individual, mi,q represents the candidate's inclination on the qth psychological motivation factor, p i,q represents the adaptation requirement of the position on the qth psychological motivation factor, m 2 Indicates the total number of psychological motivational factors;

[0107] S34, after calculating the comprehensive fitness value of each falcon individual, the falcon individuals are screened probabilistically, and the entropy value of the fitness distribution of the falcon individuals is calculated using the matching entropy regularization method to evaluate the quality of the overall matching solution;

[0108] S35. Based on the comprehensive fitness evaluation results of the screened falcon individuals, a matching evaluation result is obtained.

[0109] In this implementation manner, the S4 specifically includes:

[0110] S41, based on the falcon individual set and comprehensive fitness F(H i ) constructs the global search space Ω:

[0111] Ω={X|X min ≤X≤X max};

[0112] Where X represents the matching parameter vector of the falcon individual, X min represents the minimum boundary of the search space, X max represents the maximum boundary of the search space;

[0113] S42. In each round of iteration, the individual with the highest comprehensive fitness is selected as the global guiding individual, and the optimal matching information is propagated through the dynamic information sharing mechanism. The matching parameter update process of the falcon individual is:

[0114]

[0115] in, represents the matching parameter vector of the i-th falcon individual in the t+1th round, represents the matching parameter vector of the i-th falcon individual in the tth round, w i represents the weight of the i-th falcon individual, represents the matching parameter vector of the current optimal individual, N(0,I) represents the normal distribution with zero mean and unit variance, μ 1 represents the global adjustment factor, μ 2 represents the random exploration factor;

[0116] S43. In the global search process, a Levy flight mechanism is introduced to enhance the coverage of the search space, so that the falcon individuals can perform long-distance searches. The Levy flight mechanism generates random step lengths using Levy distribution, and the falcon individuals randomly select long-distance jump searches to avoid falling into local optimality.

[0117] S44, passing the matching results after the global search to the local optimization stage through the information sharing mechanism.

[0118] In this implementation manner, S5 specifically includes:

[0119] S51, selecting falcon individuals whose fitness is higher than a set threshold from the falcon individual set obtained in the global search phase, and constructing a matching optimization candidate set in combination with the historical matching data of the imitation agent;

[0120] S52. In the skill matching optimization stage, based on the successful matching cases provided by the imitation agent, analyze the differences between the candidate's skill characteristics and the job requirements, and optimize the skill weight allocation;

[0121] S53. In the cultural adaptation optimization stage, based on the feedback information on cultural matching from the imitation agent, the cultural adaptability of the falcon individual is partially corrected. By adjusting the influencing factors of the cultural adaptation vector, the matching plan is made more in line with the organizational culture, values ​​and team collaboration methods of the company where the position is located, thereby improving the fit between the candidate and the corporate culture.

[0122] S54. In the psychological motivation matching optimization stage, combined with the matching success cases analyzed by the imitation agent, the psychological motivation adaptability of the falcon individual is dynamically adjusted. By adjusting the psychological motivation weight in the matching strategy, the candidate's career development goals, work style preferences, and incentive mechanism needs are more consistent with the job requirements;

[0123] S55. During the local search process, the optimized matching results are evaluated in real time according to the feedback information of the imitation agent, and the optimized matching results are stored in the historical matching database.

[0124] In this implementation manner, S6 specifically includes:

[0125] S61. Continuously monitor recruitment market data, collect information including job requirements, candidate skill changes, industry trends and corporate culture changes, and build a market dynamics data set;

[0126] S62. Perform layered processing on the market dynamics data set, extract market change indicators, and generate a market change vector R d =(r 1 ,r 2 ,…,r Y ), where r i represents the quantitative value of the i-th market change indicator, and Y represents the total number of market change indicators;

[0127] S63: When the market change threshold is detected to exceed the preset value T mAfter that, the dynamic adaptation mechanism is triggered, and the imitation agent reads the market change vector R d , and update the matching strategy parameter set:

[0128] S new =S old +λ m ·R d ;

[0129] Among them, S new represents the updated matching strategy parameters, S old represents the original matching strategy parameters, λ m represents the market adaptation factor;

[0130] S64, according to the updated matching strategy parameter S new , adjust the search direction of the falcon individual set. During the adjustment process, according to the comprehensive fitness F(H i ) to re-evaluate, so that the falcon individuals move closer to the optimal matching solution, and finally form an optimized falcon individual set;

[0131] S65. Based on the optimized falcon individual set, the match with the highest comprehensive fitness is selected as the optimal person-job matching result.

[0132] Embodiment 1:

[0133] In order to verify the feasibility of the present invention in implementation, the present invention is applied to the intelligent recruitment system of a large Internet company, which needs to recruit more than 5,000 employees each year, involving multiple positions such as software development, data analysis, product operation, and marketing. The traditional recruitment method mainly relies on manual screening of resumes and comprehensive evaluation combined with the experience of the interviewer. This method is not only time-consuming, but also due to the influence of human subjective factors, the accuracy of candidate matching is low, resulting in some job candidates having difficulty adapting after joining the job, and even the proportion of resignation during the probation period is as high as 30%. In addition, the traditional matching method is mainly based on hard conditions such as academic qualifications and work experience for screening, ignoring the degree of matching between candidates and positions in terms of soft skills, cultural adaptability, psychological motivation, etc., resulting in some candidates having difficulty adapting to corporate culture or team collaboration methods when their technical capabilities are in line, further exacerbating the recruitment failure rate.

[0134] The present invention was piloted in the company's recruitment system to optimize the job matching process for three core positions: software development, data analysis, and product operation. First, the system uses deep neural networks (such as LSTM and CNN) to analyze more than 100,000 historical recruitment data, extracting multi-dimensional features of candidates and positions, including skills, cultural adaptability, psychological motivation, teamwork tendency, etc., and trains the imitation agent through the imitation learning model to learn the matching strategy of recruitment experts and build a more accurate matching mechanism.

[0135] In the actual recruitment process, the system first collects the candidate's resume information, online assessment results, social media career information, etc., and uses neural networks to extract features to generate the candidate's feature vector. At the same time, job requirement information is also parsed into feature vectors. Based on the eagle swarm clustering strategy, the system initializes multiple falcon individuals, each of which represents a possible matching combination, and evaluates the matching effect through the fitness function. In the global search stage, the falcon individuals work together to continuously optimize the matching strategy and use the information sharing mechanism to improve the search efficiency. Subsequently, in the local optimization stage, combined with the feedback of the imitation agent, the matching strategy is further fine-tuned to ensure that the final matched candidates and positions are optimally matched in terms of skills, culture, psychological motivation, etc.

[0136] The dynamic adaptation mechanism of the present invention plays an important role when the recruitment market changes. During the three months of the pilot, the company's data analysis positions were adjusted from traditional SQL and data visualization to more big data processing and AI modeling due to changes in market demand. Under the traditional recruitment model, this adjustment often requires the recruitment team to manually modify the screening conditions and slowly adjust the matching rules in subsequent recruitments, resulting in slow recruitment response and delayed talent acquisition. The present invention continuously monitors market changes and automatically adjusts matching strategies, so that the system can quickly adapt to changes in job requirements and improve the accuracy and timeliness of talent recommendations.

[0137] After three months of pilot application, the present invention has significantly improved recruitment efficiency and matching accuracy compared to traditional recruitment methods. In traditional recruitment methods, HR needs an average of 4 days to complete a round of resume screening, while the present invention, based on an intelligent matching system, shortens the screening time to 1.2 days. At the same time, the candidate-position matching score (out of 10 points) increased from 6.8 to 8.9, the pass rate of the probation period after joining the company increased from 70% to 91%, the recruitment cost was reduced by 30%, and the HR workload was reduced by 60%.

[0138] Table 1 Comparison of the effects of traditional recruitment methods and the intelligent job matching method of the present invention

[0139]

[0140] First, in terms of resume screening time, traditional recruitment methods usually require recruiters to manually screen resumes, and each round of screening takes an average of 4 days. The present invention introduces an intelligent matching system that combines deep learning and imitation learning, which can quickly analyze the multi-dimensional characteristics between candidates and positions, realize intelligent screening, and greatly shorten the screening time to 1.2 days, improving the overall efficiency by 70%. This means that companies can complete preliminary screening in a shorter time and improve the response speed of the recruitment process. This optimization effect is particularly important in the recruitment scenarios of large companies or high-mobility industries.

[0141] Secondly, in terms of candidate matching score, the present invention extracts multi-dimensional features of candidates, including hard skills, soft skills, cultural adaptability and psychological motivation, through deep neural network, and combines the Eagle Group cluster strategy for global optimization, so that the final recommended person-job matching score is increased from 6.8 points to 8.9 points, an increase of 30.8%. This optimization shows that the intelligent matching strategy of the present invention can more comprehensively consider the adaptability of candidates and positions than traditional methods, avoiding the problem of poor matching caused by relying solely on hard condition screening.

[0142] In terms of interview success rate, the optimization effect of the present invention is particularly obvious. In the traditional recruitment method, due to the extensive resume screening, some candidates are found to be unqualified for the position only after entering the interview stage, resulting in an interview success rate of only 45%. However, the present invention uses the imitation learning model and the eagle swarm optimization strategy to make the recommended candidates of higher quality and increase the interview success rate to 72%, an increase of 60%. This means that the recruitment team can reduce the number of inefficient interviews and focus on the in-depth evaluation of highly matched candidates, thereby further improving recruitment efficiency.

[0143] In terms of the probation pass rate, the probation pass rate of traditional recruitment methods is 70%, which means that 30% of new employees are still eliminated or leave during the probation period, which brings additional recruitment costs and job vacancy pressure to the company. After the optimization of the present invention, the probation pass rate is increased to 91%, an increase of 30%, indicating that the intelligent matching system can more accurately predict the long-term matching degree between candidates and positions, reduce the recruitment failure rate, and improve the employee retention rate, thereby reducing the risk of talent loss in the company.

[0144] In terms of the stability rate within six months after joining the company, the matching strategy of the present invention makes the candidates more adaptable to the positions, and the proportion of employees still working six months after joining the company increased from 65% to 87%, an increase of 33.8%. This data further proves the advantages of the present invention in long-term matching, avoiding the high turnover rate problem caused by poor matching in traditional methods.

[0145] In terms of recruitment costs, the present invention significantly reduces the recruitment expenses of enterprises. Under the traditional recruitment model, due to the long screening and interview process, enterprises need to invest a lot of human resources, resulting in each round of recruitment costs as high as 1 million yuan. The present invention makes the recruitment process more efficient through automated matching and optimization strategies, and ultimately reduces the recruitment costs to 700,000 yuan, a 30% reduction in costs. In enterprises with large-scale recruitment needs, this optimization effect can bring significant economic benefits.

[0146] Finally, in terms of the average workload of HR, in traditional recruitment methods, the HR team needs to invest a lot of time in resume screening, interview arrangement and other links, making the overall workload reach 100%. The intelligent matching system of the present invention can automatically complete resume screening, preliminary matching evaluation, and provide high-precision candidate recommendations, ultimately reducing the workload of the HR team by 60%, allowing the HR team to focus more on more strategic work such as interviews, talent management and employee development, and further improving the company's human resources management level.

[0147] In summary, the present invention has demonstrated excellent optimization effects in multiple key links of the recruitment process, which not only greatly shortens the recruitment cycle and improves the matching accuracy, but also effectively reduces the recruitment cost and talent loss rate. Through the intelligent person-job matching system, the present invention improves the human resource management efficiency of the enterprise while ensuring the quality of recruitment, and has high practical value and broad market application prospects.

[0148] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent method for matching people with jobs based on deep learning, characterized by: The steps include: S1. Collect candidate information and job requirement information, perform preprocessing and feature extraction, and generate candidate feature vectors and job feature vectors; S2. Extract successful matching cases from historical recruitment data to train the imitation learning model, generate imitation agents, and learn expert behavior patterns through the imitation agents to obtain preliminary person-job matching strategies; S3. Initialize the falcon individuals according to the candidate feature vector, the job feature vector and the preliminary person-job matching strategy. Each falcon individual represents the possibility of a person-job matching. The matching effect of each falcon individual is evaluated based on the fitness function to obtain the matching evaluation result. S4. In the global search phase, the falcon cluster strategy is used to achieve the collaborative work of individual falcons. Based on the matching evaluation results and information sharing mechanism, the falcons are explored in the entire search space, and the optimal matching information is transmitted to other falcon individuals to optimize the global matching effect. S5. Based on the feedback provided by the imitating agent, a local search method is used to further optimize the matching strategy of the falcon individuals, including skill matching, cultural adaptation and psychological motivation; S6. Monitor changes in the recruitment market. When changes in the recruitment market are detected, the dynamic adaptation mechanism is triggered. The imitation agent updates the matching strategy based on the latest market data, and adjusts the search direction and strategy of the falcon individual according to the changes in the recruitment market to generate the optimal person-job matching results.

2. The deep learning-based intelligent method for matching people with jobs according to claim 1 is characterized in that: The S2 specifically includes: S21. Filter successfully matched case samples from historical recruitment data, annotate the candidate feature vector and job feature vector in each case, and form a training data set; S22, performing data cleaning and missing value processing on the training data set, and using a unified standardization method to map all features to the same numerical range to obtain an input data set; S23, constructing a deep neural network of an imitation learning model, wherein the input layer is used to receive candidate feature vectors and job feature vectors, the hidden layer uses multi-layer nonlinear transformation to extract expert decision patterns, and the output layer generates matching score predictions; S24. Based on the multi-dimensional characteristics between candidates and positions, a comprehensive matching scoring function is defined. The comprehensive matching scoring function is composed of skill similarity, cultural similarity, psychological motivation similarity and interaction terms: S m =α·sim(V c ,V p )+β·sim(C c ,W p )+γ·sim(M c ,M p )+ δ·(sim(V c , C c )×sim(V p , C p )); where S m represents the comprehensive matching score, V c represents the skill characteristics of the candidate, C c Indicates the cultural characteristics of the candidate, M c Represents the psychological motivation characteristics of the candidate, V p Indicates the skill characteristics required for the position, C p Indicates the cultural characteristics of the job requirements, M p represents the psychological motivation characteristics of the job requirements, sim represents the similarity function, α, β, γ and δ represent weight coefficients, and satisfy α+β+γ+δ=1; S25. Use the back propagation algorithm to train the imitation learning model and define the loss function L to measure the difference between the predicted match score and the historical successful match score in the form of a combination of weighted mean square error and regularization: Where N represents the number of training samples, w i represents the weighting factor of the i-th training sample, represents the i-th matching score predicted by the imitation learning model, represents the successful matching score of the expert in the i-th training sample, λ represents the regularization coefficient, and θ k represents the training parameters of the k-th layer network, and K represents the total number of layers; S26. Encapsulate the trained imitation learning model into an imitation agent, and obtain a preliminary person-job matching strategy by inputting new candidate feature vectors and job feature vectors.

3. The deep learning-based intelligent method for matching people with jobs according to claim 1 is characterized in that: The S3 specifically includes: S31. Based on the candidate feature vector C and the job feature vector P, the falcon individuals are initialized in combination with the preliminary person-job matching strategy. Each falcon individual represents the possibility of a person-job matching. The falcon individual set H = {H1, H2, ..., H M }, where H i represents the i-th falcon individual, M represents the total number of falcon individuals; S32. Initialize the matching parameters of each falcon individual using Gaussian distribution: X i =X0+σ·N(0,I); Among them, X i represents the matching parameter vector of the i-th falcon individual, X0 represents the matching parameter generated by the preliminary man-job matching strategy, σ represents the perturbation coefficient, and N(0,I) represents the normal distribution with zero mean and unit variance; S33. For each falcon individual, a matching evaluation model based on the fitness function is constructed. The fitness function is composed of skill fitness, cultural fitness, and psychological motivation fitness: F(H i )=λ1f skill (C,P,H i )+λ2f culture (C,P,H i )+λ3f motive (C,P,H i ); Among them, F(H i ) represents the comprehensive fitness of the i-th falcon individual, λ1, λ2 and λ3 represent weight factors, satisfying λ1+λ2+λ3=1; Among them, f skill (C,P,H i ) represents the skill fitness of the i-th falcon individual, v i,j represents the candidate's score on the jth skill, p i,j represents the demand score of the position on the jth skill, and m represents the total number of skills; Among them, f culture (C,P,H i ) represents the cultural fitness of the i-th falcon individual, c i,p represents the candidate's preference value on the pth cultural factor, p i,p represents the adaptation requirement of the position on the pth cultural factor, and m1 represents the total number of cultural factors; Among them, f motive (C,P,H i ) represents the psychological motivation fitness of the i-th falcon individual, m i,q represents the candidate's inclination on the qth psychological motivation factor, p i,q represents the adaptation requirement of the position on the qth psychological motivation factor, and m2 represents the total number of psychological motivation factors; S34, after calculating the comprehensive fitness value of each falcon individual, the falcon individuals are screened probabilistically, and the entropy value of the fitness distribution of the falcon individuals is calculated using the matching entropy regularization method to evaluate the quality of the overall matching solution; S35. Based on the comprehensive fitness evaluation results of the screened falcon individuals, a matching evaluation result is obtained.

4. The deep learning-based intelligent method for matching people with jobs according to claim 1 is characterized in that: The S4 specifically includes: S41, based on the falcon individual set and comprehensive fitness F(H i ) constructs the global search space Ω: Ω={X∣X min ≤X≤X max }; Where X represents the matching parameter vector of the falcon individual, X min represents the minimum boundary of the search space, X max represents the maximum boundary of the search space; S42. In each round of iteration, the individual with the highest comprehensive fitness is selected as the global guiding individual, and the optimal matching information is propagated through the dynamic information sharing mechanism. The matching parameter update process of the falcon individual is: in, represents the matching parameter vector of the i-th falcon individual in the t+1th round, represents the matching parameter vector of the i-th falcon individual in the tth round, w i represents the weight of the i-th falcon individual, represents the matching parameter vector of the current optimal individual, N(0,I) represents the normal distribution with zero mean and unit variance, μ1 represents the global adjustment factor, and μ2 represents the random exploration factor; S43. In the global search process, a Levy flight mechanism is introduced to enhance the coverage of the search space, so that the falcon individuals can perform long-distance searches. The Levy flight mechanism generates random step lengths using Levy distribution, and the falcon individuals randomly select long-distance jump searches to avoid falling into local optimality. S44, passing the matching results after the global search to the local optimization stage through the information sharing mechanism.

5. The deep learning-based intelligent method for matching people with jobs according to claim 1 is characterized in that: The S5 specifically includes: S51, selecting falcon individuals whose fitness is higher than a set threshold from the falcon individual set obtained in the global search phase, and constructing a matching optimization candidate set in combination with the historical matching data of the imitation agent; S52. In the skill matching optimization stage, based on the successful matching cases provided by the imitation agent, analyze the differences between the candidate's skill characteristics and the job requirements, and optimize the skill weight allocation; S53. In the cultural adaptation optimization stage, based on the feedback information on cultural matching from the imitation agent, the cultural adaptability of the falcon individual is partially corrected. By adjusting the influencing factors of the cultural adaptation vector, the matching plan is made more in line with the organizational culture, values ​​and team collaboration methods of the company where the position is located, thereby improving the fit between the candidate and the corporate culture. S54. In the psychological motivation matching optimization stage, combined with the matching success cases analyzed by the imitation agent, the psychological motivation adaptability of the falcon individual is dynamically adjusted. By adjusting the psychological motivation weight in the matching strategy, the candidate's career development goals, work style preferences, and incentive mechanism needs are more consistent with the job requirements; S55. During the local search process, the optimized matching results are evaluated in real time according to the feedback information of the imitation agent, and the optimized matching results are stored in the historical matching database.

6. The deep learning-based intelligent method for matching people with jobs according to claim 1 is characterized in that: The S6 specifically includes: S61. Continuously monitor recruitment market data, collect information including job requirements, candidate skill changes, industry trends and corporate culture changes, and build a market dynamics data set; S62. Perform layered processing on the market dynamics data set, extract market change indicators, and generate a market change vector R d =(r1,r2,…,r Y ), where r i represents the quantitative value of the i-th market change indicator, and Y represents the total number of market change indicators; S63: When the market change threshold is detected to exceed the preset value T m After that, the dynamic adaptation mechanism is triggered, and the imitation agent reads the market change vector R d , and update the matching strategy parameter set: S new =S old +λ m ·R d ; Among them, S new represents the updated matching strategy parameters, S old represents the original matching strategy parameters, λ m represents the market adaptation factor; S64, according to the updated matching strategy parameter S new , adjust the search direction of the falcon individual set. During the adjustment process, according to the comprehensive fitness F(H i ) to re-evaluate, so that the falcon individuals move closer to the optimal matching solution, and finally form an optimized falcon individual set; S65. Based on the optimized falcon individual set, the match with the highest comprehensive fitness is selected as the optimal person-job matching result.

Images