Intelligent man-post matching model management method and system based on data analysis

Abstract

The invention relates to the technical field of person and post matching, in particular to a person and post intelligent matching model management method and system based on data analysis, and the method comprises the steps: collecting historical recruitment data, and constructing a job seeker feature vector and a post feature vector; extracting job seeker feature vectors and post feature vectors corresponding to the job application results to form a historical job application feedback data set, and calculating a non-prejudice feature matching degree to construct a prejudice feedback data set; extracting non-prejudice feedback data based on the prejudice feedback data set and the historical job-hunting feedback data set, and performing reconstruction optimization on the historical job-hunting feedback data set; constructing a non-prejudice man-and-post matching data set, and training the man-and-post intelligent matching model by using the non-prejudice man-and-post matching data set; and applying the trained person and post intelligent matching model to post recommendation, and collecting subsequent job application result data to optimize the model. According to the method, the prejudice factors in the historical recruitment data can be effectively eliminated, and the fairness of the matching result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of person-job matching, and specifically to a person-job intelligent matching model management method and system based on data analysis. Background Art

[0002] In today's digital employment market, intelligent job-person matching models play a key role in job recommendations. As companies pursue more efficient talent screening and job seekers increasingly expect accurate job recommendations, the application of these models is becoming increasingly widespread.

[0003] However, existing intelligent job-person matching models have significant flaws. The collection and utilization of historical recruitment application data often confounds numerous factors unrelated to applicants' professional skills, such as age, gender, and place of origin. Past recruitment practices have shown numerous cases where applicants, even those possessing the core professional skills required for a position, often face failure in the job search process due to non-hard skill factors such as age not matching the company's implicit preferences, gender not meeting the stereotype of a specific position, or place of origin triggering regional bias.

[0004] When this biased data is incorporated into the training set of intelligent job matching models, the models will unconsciously perpetuate these biases when subsequently applied to job recommendations. For example, they may over-select candidates within a specific age range, gender, or place of origin, while overlooking other candidates who equally or even better meet the job requirements. This can cause matching results to deviate significantly from the principles of fairness and impartiality, causing many job seekers to miss out on suitable development opportunities and making it difficult for them to receive fair recommendations based on their professional competence, reducing their efficiency in finding suitable positions in the job market.

[0005] Therefore, a human-job intelligent matching model management method and system based on data analysis is proposed. Summary of the Invention

[0006] The purpose of the present invention is to provide a method and system for managing a person-job intelligent matching model based on data analysis. The method collects historical recruitment data to construct a job applicant feature vector and a job feature vector; extracts the job applicant feature vector and the job feature vector corresponding to the job search results to form a historical job search feedback dataset, calculates the unbiased feature matching degree to construct a biased feedback dataset; extracts unbiased feedback data based on the biased feedback dataset and the historical job search feedback dataset, and reconstructs and optimizes the historical job search feedback dataset; constructs an unbiased person-job matching dataset, and uses the unbiased person-job matching dataset to train a person-job intelligent matching model; applies the trained person-job intelligent matching model to job recommendations, and collects subsequent job search result data to optimize the model. The present invention can effectively eliminate bias factors in historical recruitment data and improve the fairness of matching results.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for managing a person-job intelligent matching model based on data analysis, comprising:

[0009] Collect historical recruitment data; the historical recruitment data includes historical resume data, historical job data and historical job search result data;

[0010] Constructing a feature mapping dictionary, and constructing a job applicant feature vector and a position feature vector based on the feature mapping dictionary and the historical recruitment data; the job applicant feature vector and the position feature vector include a hard requirement feature sub-vector, a soft preference feature sub-vector, and a potential bias feature sub-vector;

[0011] Extracting the corresponding job seeker feature vector and the job feature vector according to the historical job search result data to form a historical job search feedback data set;

[0012] Calculating an unbiased feature matching degree based on the job applicant feature vector and the job feature vector, and constructing a biased feedback dataset based on the unbiased feature matching degree and the historical job search feedback dataset;

[0013] extracting non-biased feedback data based on the biased feedback dataset and the historical job application feedback dataset, and reconstructing and optimizing the historical job application feedback dataset according to the non-biased feedback data to obtain a reconstructed and optimized feedback dataset;

[0014] Constructing an unbiased person-job matching dataset based on the reconstructed optimization feedback dataset, and using the unbiased person-job matching dataset to train a person-job intelligent matching model;

[0015] The trained intelligent job-person matching model is applied to job recommendations, and subsequent job search result data is collected, and the intelligent job-person matching model is optimized based on the job search result data.

[0016] Preferably, the steps for calculating the unbiased feature matching degree are:

[0017] Calculate the hard requirement matching degree based on the hard requirement feature subvectors of the job applicant and the position:

[0018]

[0019] Among them, M hard Indicates the matching degree of the hard requirements; Indicates the weight of the kth hard requirement feature; sim() represents the similarity calculation function; and Respectively represent the vector components corresponding to the k-th hard requirement feature of the job applicant and the position; n represents the number of hard requirement features;

[0020] Calculate the soft preference matching degree based on the soft preference feature subvectors of the job seeker and the position:

[0021]

[0022] Among them, M soft Indicates the soft preference matching degree; represents the weight of the lth soft preference feature; V l soft and V l ' soft represent the vector components corresponding to the lth soft preference feature of the job seeker and the position respectively; m represents the number of the soft preference features;

[0023] The non-biased feature matching degree is calculated based on the hard requirement matching degree and the soft preference matching degree:

[0024] M nonbias =α·M hard +(1-α)·M soft ;

[0025] Among them, M nonbias represents the unbiased feature matching degree; α represents the balance coefficient.

[0026] Preferably, constructing the biased feedback dataset based on the non-biased feature matching degree and the historical job search feedback dataset includes:

[0027] Set the unbiased matching threshold and define the biased feedback dataset:

[0028] D bias ={d h |M nonbias (d h )>θ and Y h =0};

[0029] Among them, D bias represents the bias feedback dataset; d h represents the hth historical job search feedback data; M nonbias (d h ) represents the unbiased feature matching degree corresponding to the historical job application feedback data in the hth item; θ represents the unbiased matching degree threshold; Y h Indicates the hiring result of the historical job application feedback data described in Article h, Y h =0 means not hired, Y h =1 means hired.

[0030] Preferably, extracting the unbiased feedback data based on the biased feedback data set and the historical job application feedback data set, and reconstructing and optimizing the historical job application feedback data set according to the unbiased feedback data includes:

[0031] deleting the biased feedback dataset from the historical job application feedback dataset to obtain a reconstructed job application feedback dataset;

[0032] Extracting successful job search feedback data from the reconstructed job search feedback dataset and calculating a person-job matching similarity matrix; each element in the person-job matching similarity matrix represents the person-job matching similarity between each piece of successful job search feedback data and each piece of bias feedback data;

[0033] Selecting the successful job search feedback data whose person-job matching similarity is greater than the person-job matching threshold as the unbiased feedback data;

[0034] For each piece of the unbiased feedback data, counting the number of times the person-job matching similarity is greater than the person-job matching threshold, calculating a bias weight, and sampling the unbiased feedback data according to the bias weight to obtain an optimized unbiased feedback data set;

[0035] The reconstructed job application feedback dataset and the optimized unbiased feedback dataset are combined to obtain the reconstructed optimized feedback dataset.

[0036] Preferably, a dual-tower model is used to construct an intelligent job matching model, and the loss function is set as:

[0037] L=L match +β1·L bias +β2·L reg ;

[0038]

[0039] L reg =||W||2;

[0040] Wherein, L represents the loss function; L match represents the matching loss; L bias Indicates bias control loss; L reg represents the regularization loss; β1 and β2 represent the loss weights; N represents the number of samples; y i Indicates the matching label value of sample i; m i Indicates the matching degree of sample i output by the model; log() represents the logarithmic function; represents the unbiased feature matching degree of sample i; W represents the weight set of the model; || ||2 represents the Euclidean norm.

[0041] Preferably, setting the matching label value of the sample includes: the job seeker in the sample did not apply for the position in the sample, the corresponding matching label value is λ1; the job seeker in the sample applied for the position in the sample but did not enter the interview, the corresponding matching label value is λ2; the job seeker in the sample applied for the position in the sample and entered the interview but failed the interview, the corresponding matching label value is λ3; the job seeker in the sample applied for the position in the sample and was hired, the corresponding matching label value is λ4.

[0042] A data analysis-based intelligent job matching model management system, including:

[0043] A historical data collection module collects historical recruitment data; the historical recruitment data includes historical resume data, historical job data and historical job search result data;

[0044] A person-job feature construction module constructs a feature mapping dictionary, and constructs a job applicant feature vector and a job feature vector based on the feature mapping dictionary and the historical recruitment data; the job applicant feature vector and the job feature vector include a hard requirement feature sub-vector, a soft preference feature sub-vector, and a potential bias feature sub-vector;

[0045] A feedback data extraction module extracts the corresponding job seeker feature vector and the job feature vector according to the historical job search result data to form a historical job search feedback data set;

[0046] a biased data extraction module, which calculates a non-biased feature matching degree based on the job applicant feature vector and the job feature vector, and constructs a biased feedback dataset based on the non-biased feature matching degree and the historical job application feedback dataset;

[0047] a data reconstruction and optimization module, which extracts unbiased feedback data based on the biased feedback data set and the historical job application feedback data set, and reconstructs and optimizes the historical job application feedback data set according to the unbiased feedback data to obtain a reconstructed and optimized feedback data set;

[0048] The model training optimization module constructs an unbiased person-job matching dataset based on the reconstructed optimization feedback dataset, and uses the unbiased person-job matching dataset to train the person-job intelligent matching model; applies the trained person-job intelligent matching model to job recommendations, collects subsequent job search result data, and optimizes the person-job intelligent matching model based on the job search result data.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. We constructed a candidate and position feature vector that includes a sub-vector of hard requirement features, a sub-vector of soft preference features, and a sub-vector of potential bias features. This allows for comprehensive and accurate quantification of both candidate and position information. By eliminating potential bias features and calculating the unbiased feature matching degree based solely on the sub-vectors of hard requirement features and soft preference features, we effectively avoid the interference of non-professional factors in matching results, allowing the matching process to focus more on the actual requirements of the position and the matching of the candidate's abilities and qualities. This improves the accuracy and fairness of job matching, provides a more scientific data foundation for model training, and fundamentally reduces the impact of historical data bias on matching results.

[0051] 2. A biased feedback dataset is constructed by setting a threshold based on the degree of unbiased feature matching, accurately identifying job application feedback data that may be biased. By removing the biased feedback dataset and calculating the person-job matching similarity matrix, high-similarity successful cases are screened to form unbiased feedback data. This data is then sampled and optimized based on bias weights, and finally combined with the reconstructed job application feedback dataset to produce a reconstructed and optimized feedback dataset. Mining and utilizing historical data effectively corrects biases in the original data, improving the quality and representativeness of the historical job application feedback dataset. This ensures that the training data better reflects the true person-job matching relationship, enhances the scientific nature of model training, reduces the risk of unfair matching caused by historical data bias, and promotes the performance of the intelligent person-job matching model.

[0052] 3. Set up a loss function that includes matching loss, bias control loss, and regularization loss, and refine the sample matching label values to cover various situations, including job seekers not applying, not being interviewed, failing the interview, and being hired. The design of the loss function enables the model to balance matching accuracy and fairness during training, effectively handling complex person-job matching relationships and preventing overfitting. Refined sample label settings provide more accurate training feedback, allowing the model to learn the matching degree of different states, thereby making more detailed predictions, helping to continuously optimize the model, improve the model's reliability and adaptability in practical applications, and better address bias issues in person-job matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a flow chart of a method for managing a person-job intelligent matching model based on data analysis according to the present invention;

[0054] Figure 2 A schematic diagram of the process of reconstructing and optimizing the historical job search feedback dataset of the present invention;

[0055] Figure 3 This is a structural diagram of a human-job intelligent matching model management system based on data analysis of the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] See also Figures 1 to 3 The present invention provides a method and system for managing a person-job intelligent matching model based on data analysis. The technical solution is as follows:

[0058] Example 1:

[0059] This embodiment addresses the problem of bias in job recommendations for fresh graduates on a certain recruitment platform. In order to provide fairer and more suitable job recommendations to each job seeker and help talents find the most suitable job opportunities, a person-job intelligent matching model management method based on data analysis is applied. Figure 1 Shown, including:

[0060] Collect historical recruitment data; the historical recruitment data includes historical resume data, historical job data and historical job search result data;

[0061] Constructing a feature mapping dictionary, and constructing a job applicant feature vector and a position feature vector based on the feature mapping dictionary and the historical recruitment data; the job applicant feature vector and the position feature vector include a hard requirement feature sub-vector, a soft preference feature sub-vector, and a potential bias feature sub-vector;

[0062] Extracting the corresponding job seeker feature vector and the job feature vector according to the historical job search result data to form a historical job search feedback data set;

[0063] Calculating an unbiased feature matching degree based on the job applicant feature vector and the job feature vector, and constructing a biased feedback dataset based on the unbiased feature matching degree and the historical job search feedback dataset;

[0064] extracting non-biased feedback data based on the biased feedback dataset and the historical job application feedback dataset, and reconstructing and optimizing the historical job application feedback dataset according to the non-biased feedback data to obtain a reconstructed and optimized feedback dataset;

[0065] Constructing an unbiased person-job matching dataset based on the reconstructed optimization feedback dataset, and using the unbiased person-job matching dataset to train a person-job intelligent matching model;

[0066] The trained intelligent job-person matching model is applied to job recommendations, and subsequent job search result data is collected, and the intelligent job-person matching model is optimized based on the job search result data.

[0067] Table 1 shows a partial example of a feature mapping dictionary, constructed based on generally accepted industry knowledge. Table 2 shows an example of historical job search feedback data, including the candidate's feature vector, the position feature vector for the position the candidate applied for, and the corresponding job search results. The table also displays the specific information of the corresponding job search results.

[0068] Table 1 Example of feature mapping dictionary

[0069] Feature Type Feature Name Hard requirement features Education Hard requirement features major Hard requirement features Professional skills Soft preference features Project Experience Soft preference features Awards Potentially biased characteristics age Potentially biased characteristics Place of origin

[0070] Table 2 Example of historical job search feedback data

[0071] Person-job matching ID Job Applicant ID Position ID Whether to interview Interview results Final Result M001 R001 J001 yes pass Hire M002 R002 J001 yes pass Hire M003 R003 J001 no none Not hired M004 R004 J002 yes pass Hire M005 R005 J003 yes Failed Not hired

[0072] Furthermore, the calculation steps of the unbiased feature matching degree are:

[0073] Calculate the hard requirement matching degree based on the hard requirement feature subvectors of the job applicant and the position:

[0074]

[0075] Among them, M hard Indicates the matching degree of the hard requirements; Indicates the weight of the kth hard requirement feature; sim() represents the similarity calculation function; and Respectively represent the vector components corresponding to the k-th hard requirement feature of the job applicant and the position; n represents the number of hard requirement features;

[0076] Calculate the soft preference matching degree based on the soft preference feature subvectors of the job seeker and the position:

[0077]

[0078] Among them, M soft Indicates the soft preference matching degree; represents the weight of the lth soft preference feature; V l soft and V l ' soft represent the vector components corresponding to the lth soft preference feature of the job seeker and the position respectively; m represents the number of the soft preference features;

[0079] The non-biased feature matching degree is calculated based on the hard requirement matching degree and the soft preference matching degree:

[0080] M nonbias =α·M hard +(1-α)·M soft;

[0081] Among them, M nonbias represents the unbiased feature matching degree; α represents the balance coefficient.

[0082] Furthermore, the similarity calculation function adopts different calculation methods according to the feature type. The text description feature uses word embedding technology to convert the text description into a vector, and cosine similarity is used as the similarity calculation function; the categorical feature uses one-hot encoding, and the similarity is determined by judging whether the encoded vectors are equal. When they are equal, the similarity value is 1, and when they are not equal, the similarity value is 0; the numerical feature uses numerical normalization, and the similarity is calculated based on the normalized numerical difference.

[0083] The unbiased feature matching degree is calculated based on the hard requirement feature sub-vector and the soft preference feature sub-vector, which clearly distinguishes the core matching factors and possible bias factors between positions and job seekers, effectively avoiding excessive interference of non-professional skill factors on the matching results, improving the initial accuracy and fairness of person-job matching, and providing a more reliable data basis for subsequent screening and training.

[0084] Furthermore, constructing the biased feedback dataset based on the non-biased feature matching degree and the historical job search feedback dataset includes:

[0085] Set the unbiased matching threshold and define the biased feedback dataset:

[0086] D bias ={d h |M nonbias (d h )>θ and Y h =0};

[0087] Among them, D bias represents the bias feedback dataset; d h represents the hth historical job search feedback data; M nonbias (d h ) represents the unbiased feature matching degree corresponding to the historical job application feedback data in the hth item; θ represents the unbiased matching degree threshold; Y h Indicates the hiring result of the historical job application feedback data described in Article h, Y h =0 means not hired, Y h =1 means hired.

[0088] Each piece of historical job search feedback data includes a job seeker feature vector, a job feature vector of one of the positions the job seeker applied for, and the corresponding job search result.

[0089] By setting a threshold based on the unbiased feature matching degree to construct a biased feedback dataset, we can quickly and accurately identify the parts of historical job application feedback data that may be affected by bias, so that subsequent data processing and model optimization can specifically solve the bias problems in historical data, reduce the negative impact on model training, improve the model's ability to learn fair and just matching, and enhance the overall performance of the model.

[0090] Table 3 shows an example of the calculation results for the unbiased feature match, where the balance coefficient is 0.6. The unbiased match threshold is 0.8. As shown in Tables 2 and 3, the unbiased feature match for the historical job application feedback data with the job ID M003 exceeds the unbiased match threshold, and the job application result is a rejection. Therefore, historical job application feedback data M003 is considered biased feedback data.

[0091] Table 3 Examples of unbiased feature matching

[0092] Person-job matching ID Hard matching requirements Soft preference matching Unbiased feature matching M001 0.95 0.85 0.91 M002 0.90 0.82 0.87 M003 0.85 0.75 0.81 M004 0.98 0.88 0.94 M005 0.82 0.70 0.77

[0093] Furthermore, if Figure 2 As shown, extracting the unbiased feedback data based on the biased feedback data set and the historical job application feedback data set, and reconstructing and optimizing the historical job application feedback data set according to the unbiased feedback data includes:

[0094] deleting the biased feedback dataset from the historical job application feedback dataset to obtain a reconstructed job application feedback dataset;

[0095] Extracting successful job search feedback data from the reconstructed job search feedback dataset and calculating a person-job matching similarity matrix; each element in the person-job matching similarity matrix represents the person-job matching similarity between each piece of successful job search feedback data and each piece of bias feedback data;

[0096] Selecting the successful job search feedback data whose person-job matching similarity is greater than the person-job matching threshold as the unbiased feedback data;

[0097] For each piece of the unbiased feedback data, counting the number of times the person-job matching similarity is greater than the person-job matching threshold, calculating a bias weight, and sampling the unbiased feedback data according to the bias weight to obtain an optimized unbiased feedback data set;

[0098] The reconstructed job application feedback dataset and the optimized unbiased feedback dataset are combined to obtain the reconstructed optimized feedback dataset.

[0099] Furthermore, the person-job matching similarity is calculated using the cosine similarity method; the calculation of the bias weight includes: counting the number of times the person-job matching similarity of each unbiased feedback data is greater than the person-job matching threshold, dividing the number by the number of biased feedback data to obtain the initial weight, and normalizing the initial weights of all unbiased feedback data to obtain the bias weight.

[0100] By reconstructing and optimizing the historical job search feedback dataset, fully mining and utilizing the effective information in the historical data, and correcting the deviation of the original data, the reconstructed and optimized dataset can more truly reflect the objective laws of person-job matching, improve data quality, enhance the effectiveness of model training, and reduce the possibility of incorrect matching due to historical data bias.

[0101] Furthermore, the dual-tower model is used to construct an intelligent job matching model, and the loss function is set as:

[0102] L=L match +β1·L bias +β2·L reg ;

[0103]

[0104] L reg =||W||2;

[0105] Wherein, L represents the loss function; L match represents the matching loss; L bias Indicates bias control loss; L reg represents the regularization loss; β1 and β2 represent the loss weights; N represents the number of samples; y i Indicates the matching label value of sample i; m i Indicates the matching degree of sample i output by the model; log() represents the logarithmic function; represents the unbiased feature matching degree of sample i; W represents the weight set of the model; || ||2 represents the Euclidean norm.

[0106] A twin-tower model is adopted and a comprehensive loss function including matching loss, bias control loss and regularization loss is set, so that the model can fully take into account the accuracy, fairness and prevention of overfitting problems of matching during the training process. The twin-tower model structure helps to better process the characteristic information of job seekers and positions. The loss function can improve the adaptability and decision-making ability of the model, ensuring that the matching results output by the model both meet the actual job requirements and reflect the principles of fairness and justice.

[0107] Furthermore, setting the matching label value of the sample includes: if the job seeker in the sample did not apply for the position in the sample, the corresponding matching label value is λ1; if the job seeker in the sample applied for the position in the sample but did not enter the interview, the corresponding matching label value is λ2; if the job seeker in the sample applied for the position in the sample and entered the interview but did not pass the interview, the corresponding matching label value is λ3; if the job seeker in the sample applied for the position in the sample and was hired, the corresponding matching label value is λ4. In this embodiment, λ1 = 0, λ2 = 0.3, λ3 = 0.7, and λ4 = 1.

[0108] The sample matching degree label values are finely set to cover various states of job seekers in the job search process, providing the model with richer and more accurate training feedback information, enabling the model to learn the characteristics and rules of different job search stages more deeply, further improving the model's understanding and judgment of the person-job matching relationship, enhancing the model's prediction accuracy and reliability, optimizing the model's training effect, and improving the model's performance in practical applications.

[0109] The present invention constructs a method for managing a person-job intelligent matching model based on data analysis. First, by collecting and analyzing historical recruitment data, a feature mapping dictionary including hard requirements, soft preferences and potential bias features is constructed to generate feature vectors of job seekers and positions, laying a data foundation for subsequent matching; then, based on the feature vector, the non-biased feature matching degree is calculated, and the biased feedback data set is screened out by setting a threshold, and reconstructed and optimized in combination with the historical feedback data, samples are extracted from successful cases with high similarity, and a reconstructed and optimized feedback data set is generated by weight calculation, which improves the representativeness and fairness of the data while reducing the influence of historical bias data; finally, the reconstructed and optimized feedback data set is used to construct a non-biased person-job matching data set to train a person-job intelligent matching model. The present invention can effectively eliminate bias factors in historical recruitment data, improve the fairness of job recommendation results, and significantly improve the matching accuracy and applicability of the model.

[0110] Example 2:

[0111] The present invention also provides a data analysis-based intelligent matching model management system for people and jobs. Figure 3 Shown, including:

[0112] A historical data collection module collects historical recruitment data; the historical recruitment data includes historical resume data, historical job data and historical job search result data;

[0113] A person-job feature construction module constructs a feature mapping dictionary, and constructs a job applicant feature vector and a job feature vector based on the feature mapping dictionary and the historical recruitment data; the job applicant feature vector and the job feature vector include a hard requirement feature sub-vector, a soft preference feature sub-vector, and a potential bias feature sub-vector;

[0114] A feedback data extraction module extracts the corresponding job seeker feature vector and the job feature vector according to the historical job search result data to form a historical job search feedback data set;

[0115] a biased data extraction module, which calculates a non-biased feature matching degree based on the job applicant feature vector and the job feature vector, and constructs a biased feedback dataset based on the non-biased feature matching degree and the historical job application feedback dataset;

[0116] a data reconstruction and optimization module, which extracts unbiased feedback data based on the biased feedback data set and the historical job application feedback data set, and reconstructs and optimizes the historical job application feedback data set according to the unbiased feedback data to obtain a reconstructed and optimized feedback data set;

[0117] The model training optimization module constructs an unbiased person-job matching dataset based on the reconstructed optimization feedback dataset, and uses the unbiased person-job matching dataset to train the person-job intelligent matching model; applies the trained person-job intelligent matching model to job recommendations, collects subsequent job search result data, and optimizes the person-job intelligent matching model based on the job search result data.

[0118] Furthermore, the calculation steps of the unbiased feature matching degree are:

[0119] Calculate the hard requirement matching degree based on the hard requirement feature subvectors of the job applicant and the position:

[0120]

[0121] Among them, M hard Indicates the matching degree of the hard requirements; Indicates the weight of the kth hard requirement feature; sim() represents the similarity calculation function; and Respectively represent the vector components corresponding to the k-th hard requirement feature of the job applicant and the position; n represents the number of hard requirement features;

[0122] Calculate the soft preference matching degree based on the soft preference feature subvectors of the job seeker and the position:

[0123]

[0124] Among them, M softIndicates the soft preference matching degree; represents the weight of the lth soft preference feature; V l soft and V l ' soft represent the vector components corresponding to the lth soft preference feature of the job seeker and the position respectively; m represents the number of the soft preference features;

[0125] The non-biased feature matching degree is calculated based on the hard requirement matching degree and the soft preference matching degree:

[0126] M nonbias =α·M hard +(1-α)·M soft ;

[0127] Among them, M nonbias represents the unbiased feature matching degree; α represents the balance coefficient.

[0128] Furthermore, constructing the biased feedback dataset based on the non-biased feature matching degree and the historical job search feedback dataset includes:

[0129] Set the unbiased matching threshold and define the biased feedback dataset:

[0130] D bias ={d h |M nonbias (d h )>θ and Y h =0};

[0131] Among them, D bias represents the bias feedback dataset; d h represents the hth historical job search feedback data; M nonbias (d h ) represents the unbiased feature matching degree corresponding to the historical job application feedback data in the hth item; θ represents the unbiased matching degree threshold; Y h Indicates the hiring result of the historical job application feedback data described in Article h, Y h =0 means not hired, Y h =1 means hired.

[0132] Furthermore, extracting the unbiased feedback data based on the biased feedback dataset and the historical job application feedback dataset, and reconstructing and optimizing the historical job application feedback dataset based on the unbiased feedback data includes:

[0133] deleting the biased feedback dataset from the historical job application feedback dataset to obtain a reconstructed job application feedback dataset;

[0134] Extracting successful job search feedback data from the reconstructed job search feedback dataset and calculating a person-job matching similarity matrix; each element in the person-job matching similarity matrix represents the person-job matching similarity between each piece of successful job search feedback data and each piece of bias feedback data;

[0135] Selecting the successful job search feedback data whose person-job matching similarity is greater than the person-job matching threshold as the unbiased feedback data;

[0136] For each piece of the unbiased feedback data, counting the number of times the person-job matching similarity is greater than the person-job matching threshold, calculating a bias weight, and sampling the unbiased feedback data according to the bias weight to obtain an optimized unbiased feedback data set;

[0137] The reconstructed job application feedback dataset and the optimized unbiased feedback dataset are combined to obtain the reconstructed optimized feedback dataset.

[0138] Furthermore, the dual-tower model is used to construct an intelligent job matching model, and the loss function is set as:

[0139] L=L match +β1·L bias +β2·L reg ;

[0140]

[0141] L reg =||W||2;

[0142] Wherein, L represents the loss function; L match represents the matching loss; L bias Indicates bias control loss; L reg represents the regularization loss; β1 and β2 represent the loss weights; N represents the number of samples; y i Indicates the matching label value of sample i; m i Indicates the matching degree of sample i output by the model; log() represents the logarithmic function; represents the unbiased feature matching degree of sample i; W represents the weight set of the model; || ||2 represents the Euclidean norm.

[0143] Furthermore, setting the matching label value of the sample includes: the job seeker in the sample did not apply for the position in the sample, the corresponding matching label value is λ1; the job seeker in the sample applied for the position in the sample but did not enter the interview, the corresponding matching label value is λ2; the job seeker in the sample applied for the position in the sample and entered the interview but failed the interview, the corresponding matching label value is λ3; the job seeker in the sample applied for the position in the sample and was hired, the corresponding matching label value is λ4.

[0144] Table 4 compares the job matching results before and after reconstructing and optimizing the historical job search feedback dataset using the proposed method, using multiple metrics. The data shows that eliminating biased data not only increases job seeker engagement but also significantly improves the ultimate job search success rate.

[0145] Table 4 Matching effect analysis

[0146] Performance indicators Before optimization After optimization Improvement Resume delivery rate 15％ 28％ +13％ Interview invitation rate 20％ 35％ +15％ Job search success rate 18％ 32％ +14％

[0147] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for managing a person-job intelligent matching model based on data analysis, characterized in that: include: Collect historical recruitment data; the historical recruitment data includes historical resume data, historical job data and historical job search result data; Constructing a feature mapping dictionary, and constructing a job applicant feature vector and a position feature vector based on the feature mapping dictionary and the historical recruitment data; the job applicant feature vector and the position feature vector include a hard requirement feature sub-vector, a soft preference feature sub-vector, and a potential bias feature sub-vector; Extracting the corresponding job seeker feature vector and the job feature vector according to the historical job search result data to form a historical job search feedback data set; Calculating an unbiased feature matching degree based on the job applicant feature vector and the job feature vector, and constructing a biased feedback dataset based on the unbiased feature matching degree and the historical job search feedback dataset; extracting non-biased feedback data based on the biased feedback dataset and the historical job application feedback dataset, and reconstructing and optimizing the historical job application feedback dataset according to the non-biased feedback data to obtain a reconstructed and optimized feedback dataset; Constructing an unbiased person-job matching dataset based on the reconstructed optimization feedback dataset, and using the unbiased person-job matching dataset to train a person-job intelligent matching model; The trained intelligent job-person matching model is applied to job recommendations, and subsequent job search result data is collected, and the intelligent job-person matching model is optimized based on the job search result data.

2. The method for managing a person-job intelligent matching model based on data analysis according to claim 1, characterized in that: The calculation steps of the unbiased feature matching degree are as follows: Calculate the hard requirement matching degree based on the hard requirement feature subvectors of the job applicant and the position: Among them, M hard Indicates the matching degree of the hard requirements; Indicates the weight of the kth hard requirement feature; sim() represents the similarity calculation function; and Respectively represent the vector components corresponding to the k-th hard requirement feature of the job applicant and the position; n represents the number of hard requirement features; Calculate the soft preference matching degree based on the soft preference feature subvectors of the job seeker and the position: Among them, M soft Indicates the soft preference matching degree; represents the weight of the lth soft preference feature; V l soft and V l ' soft represent the vector components corresponding to the lth soft preference feature of the job seeker and the position respectively; m represents the number of the soft preference features; The non-biased feature matching degree is calculated based on the hard requirement matching degree and the soft preference matching degree: M nonbias =α·M hard +(1-a)·M soft ; Among them, M nonbias represents the unbiased feature matching degree; α represents the balance coefficient.

3. The method for managing a person-job intelligent matching model based on data analysis according to claim 1, characterized in that: Constructing the biased feedback dataset based on the non-biased feature matching degree and the historical job search feedback dataset includes: Set the unbiased matching threshold and define the biased feedback dataset: D bias = {d h | M nonbias (d h ) > θ and Y h = 0}; Among them, D bias represents the bias feedback dataset; d h represents the hth historical job search feedback data; M nonbias (d h ) represents the unbiased feature matching degree corresponding to the historical job application feedback data in the hth item; θ represents the unbiased matching degree threshold; Y h Indicates the hiring result of the historical job application feedback data described in Article h, Y h =0 means not hired, Y h =1 means hired.

4. The method for managing a person-job intelligent matching model based on data analysis according to claim 1, characterized in that: Extracting the unbiased feedback data based on the biased feedback dataset and the historical job application feedback dataset, and reconstructing and optimizing the historical job application feedback dataset based on the unbiased feedback data includes: deleting the biased feedback dataset from the historical job application feedback dataset to obtain a reconstructed job application feedback dataset; Extracting successful job search feedback data from the reconstructed job search feedback dataset and calculating a person-job matching similarity matrix; each element in the person-job matching similarity matrix represents the person-job matching similarity between each piece of successful job search feedback data and each piece of bias feedback data; Selecting the successful job search feedback data whose person-job matching similarity is greater than the person-job matching threshold as the unbiased feedback data; For each piece of the unbiased feedback data, counting the number of times the person-job matching similarity is greater than the person-job matching threshold, calculating a bias weight, and sampling the unbiased feedback data according to the bias weight to obtain an optimized unbiased feedback data set; The reconstructed job application feedback dataset and the optimized unbiased feedback dataset are combined to obtain the reconstructed optimized feedback dataset.

5. The method for managing a person-job intelligent matching model based on data analysis according to claim 1, characterized in that: The dual-tower model is used to build an intelligent job matching model, and the loss function is set as: L=L match +β1·L bias +β2·L reg ; L reg =||W||2; Wherein, L represents the loss function; L match represents the matching loss; L bias Indicates bias control loss; L reg represents the regularization loss; β1 and β2 represent the loss weights; N represents the number of samples; y i Indicates the matching label value of sample i; m i Indicates the matching degree of sample i output by the model; log() represents the logarithmic function; represents the unbiased feature matching degree of sample i; W represents the weight set of the model; || ||2 represents the Euclidean norm.

6. The method for managing a person-job intelligent matching model based on data analysis according to claim 5 is characterized in that: Setting the matching label value of the sample includes: the job seeker in the sample did not apply for the position in the sample, the corresponding matching label value is λ1; the job seeker in the sample applied for the position in the sample but did not enter the interview, the corresponding matching label value is λ2; the job seeker in the sample applied for the position in the sample and entered the interview but failed the interview, the corresponding matching label value is λ3; the job seeker in the sample applied for the position in the sample and was hired, the corresponding matching label value is λ4.

7. A data analysis-based intelligent matching model management system for people and jobs, characterized by: include: A historical data collection module collects historical recruitment data; the historical recruitment data includes historical resume data, historical job data and historical job search result data; A person-job feature construction module constructs a feature mapping dictionary, and constructs a job applicant feature vector and a job feature vector based on the feature mapping dictionary and the historical recruitment data; the job applicant feature vector and the job feature vector include a hard requirement feature sub-vector, a soft preference feature sub-vector, and a potential bias feature sub-vector; A feedback data extraction module extracts the corresponding job seeker feature vector and the job feature vector according to the historical job search result data to form a historical job search feedback data set; a biased data extraction module, which calculates a non-biased feature matching degree based on the job applicant feature vector and the job feature vector, and constructs a biased feedback dataset based on the non-biased feature matching degree and the historical job application feedback dataset; a data reconstruction and optimization module, which extracts unbiased feedback data based on the biased feedback data set and the historical job application feedback data set, and reconstructs and optimizes the historical job application feedback data set according to the unbiased feedback data to obtain a reconstructed and optimized feedback data set; The model training optimization module constructs an unbiased person-job matching dataset based on the reconstructed optimization feedback dataset, and uses the unbiased person-job matching dataset to train the person-job intelligent matching model; applies the trained person-job intelligent matching model to job recommendations, collects subsequent job search result data, and optimizes the person-job intelligent matching model based on the job search result data.

Images