Person-post matching method, device, equipment, medium and product

Abstract

The application discloses a kind of person-post matching method, device, equipment, medium and product, it is related to information processing technical field, method includes: by receiving recruitment post data and resume data;By pre-constructed person-post matching model, the recruitment post data and resume data are carried out data matching, obtain matching result, the person-post matching model is based on recruitment vector model, information extraction model, weight evaluation model and matching degree scoring model and is constructed to obtain.Thereby, by the person-post matching model that is constructed by recruitment vector model, information extraction model, weight evaluation model and matching degree scoring model, the recruitment post data and resume data are carried out data matching, obtain matching result, it is not aligned in prior art that post requirement and resume advantage are carried out, leading to enterprise recruitment quality is not high and the problem of brain drain, improve the accuracy of person-post matching.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, in particular to a person-post matching method, device, equipment, medium and product. BACKGROUND

[0002] In the post recruitment of an enterprise, resume information of multiple job seekers will be received. Finding the most matched job seeker from the resume information needs to go through information extraction and person-post matching operations. Information extraction is an important link in the person-post matching task, and the quality of key data extraction directly affects the matching result. Person-post matching is because the requirements of different posts and the personal advantages of resume information are different.

[0003] However, the information extraction method designed by the existing person-post matching scheme does not fully integrate domain knowledge, the extraction performance is limited, and the potential of the matching algorithm cannot be fully utilized. Secondly, the extraction dimensions designed in the existing person-post matching scheme are less, which will lead to the inability to evaluate the matching degree of the post and the resume from multiple aspects. In the matching part, different posts have different focuses. Although the corresponding weight can be evaluated by designing the post requirement focus dimension, this scheme is only suitable for the scenario of matching resumes according to person-post matching, and does not consider the advantages embodied in the resume when matching the post based on talents.

[0004] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a person-post matching method, device, equipment, medium and product, which aims to solve the technical problem that the existing technology does not align the post requirements with the resume advantages, resulting in low recruitment quality and talent loss of enterprises.

[0006] To achieve the above purpose, the present application provides a person-post matching method, which comprises:

[0007] receiving recruitment post data and resume data;

[0008] performing data matching on the recruitment post data and the resume data through a pre-constructed person-post matching model to obtain a matching result, wherein the person-post matching model is constructed based on a recruitment vector model, an information extraction model, a weight evaluation model and a matching degree scoring model.

[0009] In an embodiment, before the step of performing data matching on the recruitment post data and the resume data through the pre-constructed person-post matching model to obtain a matching result, the method further comprises:

[0010] obtaining historical post information and historical resume information;

[0011] performing clustering analysis on the historical post information to obtain a human-post matching dimension;

[0012] performing information extraction on the historical post information and historical resume information based on the human-post matching dimension through a first deep learning and natural language processing model to obtain a first extraction result;

[0013] constructing a vector model optimization dataset according to the first extraction result, and training an initial vector model based on the vector model optimization dataset to obtain a recruitment vector model;

[0014] performing knowledge distillation on the first deep learning and natural language processing model according to the first extraction result to obtain an information extraction model;

[0015] performing dimension weight evaluation on the human-post matching dimension and historical resume information through a second deep learning and natural language processing model to obtain an evaluation result;

[0016] performing knowledge distillation on the second deep learning and natural language processing model according to the evaluation result to obtain a weight evaluation model;

[0017] inputting the first extraction result and the evaluation result into a third deep learning and natural language processing model to perform matching degree scoring to obtain a first scoring result;

[0018] performing knowledge distillation on the third deep learning and natural language processing model according to the first scoring result to obtain a matching degree scoring model;

[0019] constructing a human-post matching model according to the recruitment vector model, the information extraction model, the matching degree scoring model, and the weight evaluation model.

[0020] In an embodiment, the step of performing knowledge distillation on the first deep learning and natural language processing model according to the first extraction result to obtain an information extraction model comprises:

[0021] performing dimension classification on the first extraction result through the first deep learning and natural language processing model based on the human-post matching dimension to obtain a first classification result;

[0022] performing dimension classification on the first extraction result through a classification model based on the human-post matching dimension to obtain a second classification result;

[0023] performing cross-validation on the first classification result and the second classification result to obtain a dimension classification dataset;

[0024] According to the first extraction result, knowledge distillation is performed on the first deep learning and natural language processing model to obtain an initial information extraction model;

[0025] According to the dimension classification dataset, a subnetwork of the initial information extraction model is activated through a gating network, and a nonlinear transformation is performed on the initial information extraction model through an activation function to obtain an information extraction model.

[0026] In an embodiment, the step of performing knowledge distillation on the third deep learning and natural language processing model according to the first scoring result to obtain a matching degree scoring model comprises:

[0027] Knowledge distillation is performed on the third deep learning and natural language processing model to obtain an initial matching degree scoring model;

[0028] The first extraction result and the evaluation result are input into the initial matching degree scoring model to obtain a second scoring result;

[0029] Difference calculation is performed based on the first scoring result and the second scoring result to obtain a cross-entropy loss;

[0030] Difference calculation is performed based on the probability distribution of the first scoring result and the probability distribution of the second scoring result to obtain a divergence loss;

[0031] The initial matching degree scoring model is optimized based on the cross-entropy loss and the divergence loss to obtain a matching degree scoring model.

[0032] In an embodiment, the step of performing data matching on the recruitment post data and the resume data through a pre-constructed person-post matching model to obtain a matching result comprises:

[0033] Information extraction is performed on the recruitment post data and the resume data through the information extraction model to obtain a second extraction result;

[0034] Weight evaluation is performed on the recruitment post data and the resume data through the weight evaluation model to obtain post dimension weights and resume dimension weights;

[0035] The second extraction result is vectorized through the recruitment vector model to obtain a vector representation of the second extraction result;

[0036] Based on the post dimension weights and the resume dimension weights, similarity sorting is performed on the second extraction result to obtain a plurality of candidate information;

[0037] The plurality of candidate information is matched through the matching degree scoring model to obtain a matching result.

[0038] In an embodiment, after the step of obtaining a matching result by matching the recruitment post data and the resume data through the pre-constructed human-post matching model, the method further comprises:

[0039] receiving a post recruitment result;

[0040] querying the resume data according to the post recruitment result to obtain employee information;

[0041] sending an initial employee evaluation form to an employee management end according to the employee information, and filling the initial employee evaluation form by the employee management end to obtain a final employee evaluation form;

[0042] receiving the final employee evaluation form, and optimizing the human-post matching model through the final employee evaluation form to obtain a model optimization result.

[0043] In addition, to achieve the above-mentioned purpose, the present application also provides a human-post matching device, which comprises:

[0044] a receiving module, configured to receive recruitment post data and resume data;

[0045] a matching module, configured to match the recruitment post data and the resume data through a pre-constructed human-post matching model to obtain a matching result, wherein the human-post matching model is constructed based on a recruitment vector model, an information extraction model, a weight evaluation model, and a matching degree scoring model.

[0046] In addition, to achieve the above-mentioned purpose, the present application also provides a human-post matching device, which comprises: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is configured to implement the steps of the human-post matching method as described above.

[0047] In addition, to achieve the above-mentioned purpose, the present application also provides a storage medium, which is a computer readable storage medium, and a computer program is stored on the storage medium, wherein the computer program is executed by a processor to implement the steps of the human-post matching method as described above.

[0048] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which comprises a computer program, wherein the computer program is executed by a processor to implement the steps of the human-post matching method as described above.

[0049] The one or more technical solutions provided by the present application have at least the following technical effects:

[0050] The person-post matching method, device, equipment, medium and product provided by the embodiment of the application receive recruitment post data and resume data, perform data matching on the recruitment post data and the resume data through a pre-constructed person-post matching model, obtain a matching result, and the person-post matching model is constructed based on a recruitment vector model, an information extraction model, a weight evaluation model and a matching degree scoring model. Thus, the recruitment post data and the resume data are matched through the person-post matching model constructed based on the recruitment vector model, the information extraction model, the weight evaluation model and the matching degree scoring model, and a matching result is obtained, thereby solving the problem that the post requirements and the resume advantages are not aligned in the prior art, resulting in low recruitment quality of enterprises and talent loss, and improving the accuracy of person-post matching. BRIEF DESCRIPTION OF DRAWINGS

[0051] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate preferred embodiments of the application and, together with the description, serve to explain the principles of the application.

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the accompanying drawings required to be used in the embodiments or the prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.

[0053] Figure 1 The flowchart provided for the person-post matching method embodiment one of the application;

[0054] Figure 2 The flowchart provided for the person-post matching method of the application involves constructing a person-post matching model;

[0055] Figure 3 The flowchart provided for the person-post matching method embodiment two of the application;

[0056] Figure 4 The flowchart provided for the person-post matching method of the application involves a person-post matching model for matching;

[0057] Figure 5 The brief flowchart provided for the person-post matching method of the embodiment two of the application;

[0058] Figure 6 The module structure diagram of the person-post matching device of the embodiment of the application;

[0059] Figure 7 The device structure diagram of the hardware running environment involved in the person-post matching method in the embodiment of the application.

[0060] The object, the functional characteristics and the advantages of the present application will be further explained in conjunction with the embodiments and with reference to the drawings. DETAILED DESCRIPTION

[0061] It should be understood that the specific embodiments described herein merely exemplify the technical solutions of the present application, and are not intended to limit the present application.

[0062] In order to better understand the technical solutions of the present application, the following will be described in detail in conjunction with the drawings of the specification and the specific embodiments.

[0063] The main solution of the embodiment of the application is: obtaining historical post information and historical resume information; performing cluster analysis on the historical post information to obtain a human-post matching dimension; performing information extraction on the historical post information and the historical resume information based on the human-post matching dimension through a first deep learning and natural language processing model to obtain a first extraction result; constructing a vector model optimization data set according to the first extraction result, training an initial vector model based on the vector model optimization data set to obtain a recruitment vector model; performing knowledge distillation on the first deep learning and natural language processing model according to the first extraction result to obtain an information extraction model; performing dimension weight evaluation on the human-post matching dimension and the historical resume information through a second deep learning and natural language processing model to obtain an evaluation result; performing knowledge distillation on the second deep learning and natural language processing model according to the evaluation result to obtain a weight evaluation model; inputting the first extraction result and the evaluation result into a third deep learning and natural language processing model to perform matching degree scoring to obtain a first scoring result; performing knowledge distillation on the third deep learning and natural language processing model according to the first scoring result to obtain a matching degree scoring model; constructing a human-post matching model according to the recruitment vector model, the information extraction model, the matching degree scoring model and the weight evaluation model. Performing dimension classification on the first extraction result through the first deep learning and natural language processing model based on the human-post matching dimension to obtain a first classification result; performing dimension classification on the first extraction result through a classification model based on the human-post matching dimension to obtain a second classification result; performing cross-validation on the first classification result and the second classification result to obtain a dimension classification data set; performing knowledge distillation on the first deep learning and natural language processing model according to the first extraction result to obtain an initial information extraction model; activating a subnetwork of the initial information extraction model through a gating network according to the dimension classification data set, and performing nonlinear transformation on the initial information extraction model through an activation function to obtain an information extraction model. Performing knowledge distillation on the third deep learning and natural language processing model to obtain an initial matching degree scoring model; inputting the first extraction result and the evaluation result into the initial matching degree scoring model to obtain a second scoring result; performing difference calculation based on the first scoring result and the second scoring result to obtain a cross-entropy loss; performing difference calculation based on a probability distribution of the first scoring result and a probability distribution of the second scoring result to obtain a divergence loss; performing model optimization on the initial matching degree scoring model based on the cross-entropy loss and the divergence loss to obtain a matching degree scoring model.The information extraction model is used for information extraction on the recruitment post data and the resume data, to obtain a second extraction result; the weight evaluation model is used for weight evaluation on the recruitment post data and the resume data, to obtain post dimension weight and resume dimension weight; the recruitment vector model is used for vectorization on the second extraction result, to obtain a vector representation of the second extraction result; the post dimension weight and the resume dimension weight are used for similarity sorting on the second extraction result, to obtain a plurality of candidate information; the matching degree scoring model is used for matching on the plurality of candidate information, to obtain a matching result. A post recruitment result is received; the resume data is queried based on the post recruitment result to obtain recruitment personnel information; an initial employee evaluation table is sent to an employee management end based on the recruitment personnel information, the initial employee evaluation table is filled in by the employee management end, and a final employee evaluation table is obtained; the final employee evaluation table is received, and the human-post matching model is optimized based on the final employee evaluation table, to obtain a model optimization result. Thus, the problem that the post requirements are not aligned with the resume advantages in the prior art, resulting in low enterprise recruitment quality and talent loss, is solved, the post matching is realized, and the accuracy of human-post matching is improved. Based on the present application, the model parameter quantity is multiplied in reality, the matching time and accuracy are affected under limited computing power, and the content generated by the model prediction is too long, which also increases the time consumption of human-post matching, thereby the problem of low accuracy is solved. A human-post matching method is designed, the effectiveness of the human-post matching method is verified when the post is matched, and finally the accuracy of human-post matching by the present application is obviously improved.

[0064] In the present embodiment, for the convenience of description, the following is described with the human-post matching device as the execution subject.

[0065] In the prior art, the information extraction in human-post matching does not fully integrate field knowledge, the extraction performance is limited, and the potential of the matching algorithm cannot be fully developed. Secondly, the extraction dimensions designed in the existing human-post matching scheme are less, and the extraction dimensions of the post and the resume are not aligned. The unaligned dimensions will cause more information interference when extracting features in the vector space, and the less extraction dimensions will not be able to evaluate the matching degree of the post and the resume from multiple perspectives. In addition, different posts have different focuses in human-post matching, so most schemes design to evaluate the corresponding weight according to the post demand focus. However, this scheme is only applicable to the scenario of matching resumes based on human-post matching, and does not consider the advantages embodied in the resume when matching talents to posts in the human-post matching task. In addition, the matching sorting part is different from the HR evaluation method in the real scene, which leads to the need to improve the explainability of the quantitative matching result, thereby reducing the accuracy of human-post matching in the prior art.

[0066] The application provides a solution, a post-person matching model is constructed by a recruitment vector model, an information extraction model, a weight evaluation model and a matching degree scoring model, and the received recruitment post data and resume data are matched by using the post-person matching model, the accuracy and intelligent degree of post-person matching are improved, and more high-quality services are provided for users.

[0067] From the above embodiments, the post-person matching model constructed by the recruitment vector model, the information extraction model, the weight evaluation model and the matching degree scoring model is used to match the recruitment post data and the resume data, and the matching result is obtained, the problem that the post requirements are not aligned with the resume advantages in the prior art, resulting in low recruitment quality of enterprises and talent loss, is solved, and the accuracy of post-person matching is improved.

[0068] It should be noted that the execution subject of the embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device, a post-person matching device, etc. capable of realizing the above functions. The post-person matching device is taken as an example to describe the embodiment and the following embodiments.

[0069] Based on this, the embodiment of the application provides a post-person matching method, referring to Figure 1 , Figure 1 The flowchart of the first embodiment of the post-person matching method of the present application is shown.

[0070] In this embodiment, the post-person matching method includes steps S01-S02:

[0071] Step S01, receiving recruitment post data and resume data;

[0072] Before the embodiment is described, it should be clear that in the post recruitment of enterprises, resume information of multiple job seekers will be received, and the most matched job seeker needs to be found from the resume information through information extraction and post-person matching operations. Information extraction is an important link in the post-person matching task, and the quality of key data extraction directly affects the matching result, and post-person matching is because different posts have different emphases, so it is necessary to evaluate the corresponding weight according to the post demand emphasis dimension.

[0073] However, the information extraction method designed by the existing post-person matching scheme does not fully integrate domain knowledge, and the extraction performance is limited, and the main problems are as follows: ①the information extraction link lacks domain knowledge integration, resulting in incomplete key data recognition; ②the evaluation dimension design is single, and it is difficult to cover the multi-dimensional characteristics of post requirements; ③the matching algorithm only focuses on the weight of post requirements in one direction, ignoring the core advantages of talents (such as special skills, project results, etc.) reflected in the resume, resulting in mismatch of high-potential candidates and causing hidden loss of talents.

[0074] Therefore, in this embodiment, the recruitment position data and the resume data are preferentially received. The recruitment position data includes the basic attributes of the position (such as the position name, the job level system, the salary range, and the work location), the ability requirements (such as the technical stack weight, the framework requirement, the certificate hard condition, and the project experience time limit), and the implicit data (such as the technical depth index, the team coordination feature, and the industry specific data). These recruitment position data are for the employer to formulate, and the employer can modify the recruitment position data according to the specific needs. For example, for the position of software development, the basic attributes can be:

[0075] (1) Position name: software engineer, front-end development engineer, back-end development engineer, full-stack engineer, etc.

[0076] (2) Job level system: such as junior, intermediate, senior, or more subdivided job level system such as intern, junior engineer, senior engineer, technical manager, technical director, etc.

[0077] (3) Salary range: for example, annual salary range, monthly salary range, or other salary structure (such as fixed salary + bonus, stock options, etc.).

[0078] (4) Work location: including company headquarters, branch, or remote work (Hybrid / Remote). It can also be subdivided to specific cities or regions.

[0079] (5) Company nature: such as start-up company, large internet company, foreign company, traditional industry, etc.

[0080] The ability requirements include:

[0081] (1) Technical stack weight: programming languages (such as Java, Python, JavaScript, etc.) required to be mastered by the applicant, tools (such as Docker, Kubernetes, etc.), databases (such as MySQL, PostgreSQL, etc.), and related platforms (such as cloud platform, DevOps tools, etc.).

[0082] (2) Framework requirement: for front-end development positions, React, Vue.js, or Angular may be required, and for back-end development positions, Spring Boot, Django, etc. may be required.

[0083] (3) Certificate hard condition: requires specific technical certification or certificate, such as AWS certification, Google Cloud certification, PMP, etc.

[0084] (4) Project experience time limit: requires 3 years or more of project experience, or specific technical stack related experience time limit.

[0085] (5) Education background: Bachelor's degree or above is required, or whether there is a specific field of professional requirements (such as computer science, software engineering, etc.);

[0086] (6) Language requirements: fluent English is required, especially for foreign companies or multinational companies.

[0087] And the implicit data includes:

[0088] (1) Technical depth index: the candidate is required to have certain architecture design ability, algorithm and data structure depth understanding, or large-scale distributed system design experience;

[0089] (2) Team collaboration features: the post requires good team cooperation ability, cross-department communication ability, project management ability, etc.;

[0090] (3) Industry-specific data: requires specific industry knowledge, such as specific development experience in the financial industry, medical industry;

[0091] (4) Soft skills requirements: communication skills, problem-solving skills, stress management skills, etc.

[0092] (5) Work mode / culture: emphasizes agile development, DevOps culture, or requires support for rapid iteration, innovative thinking, etc.

[0093] In the process of talent recruitment, in addition to the recruitment post data (i.e. the requirements of the enterprise), there is also resume data (i.e. the relevant information of the candidate), including the candidate's personal basic information (such as name, contact information, job-seeking intention and personal profile, etc.), education background (such as school name, degree, major and graduation time, etc.), work experience (such as company name and position, work time, job responsibilities and achievements, and use of technology or tools, etc.), skills and certificates (such as technical skills, soft skills and certificates, etc.), but when the candidate submits the resume, there may be a case of mass mailing, that is, sending the resume information to multiple recruitment platforms, companies or positions, which will cause the resume that does not meet the requirements to be sent to the company's recruitment department, resulting in waste of resources for the company to screen during talent recruitment. In addition, some candidates' resume content is simple or cannot grasp the key points of the company's requirements during filling, both of which may lead to the fact that the advantages in the resume are not considered during the matching of the post talent, resulting in the problem of talent loss.

[0094] Step S02, data matching of the recruitment post data and the resume data is performed through a pre-constructed post-personnel matching model to obtain a matching result, and the post-personnel matching model is constructed based on a recruitment vector model, an information extraction model, a weight evaluation model, and a matching degree scoring model.

[0095] After receiving the recruitment post data and the resume data, it is defaulted that the enterprise's talent matching task is received, at this time, the recruitment post data and the talent matching data are matched by the talent matching model constructed by the recruitment vector model, the information extraction model, the weight evaluation model and the matching degree scoring model to obtain the most suitable candidate resume (i.e. the matching structure) for the above recruitment post data. In order to solve the problem of talent loss, the recruitment vector model in the embodiment converts the description of the recruitment post and the content in the resume into a vector form, and the text information of each post and resume is converted into a numerical vector, which is convenient for calculating the similarity between them. The information extraction model automatically extracts key information such as job title, skill requirement, work experience, education background, etc. from the post recruitment data and the resume data. The weight evaluation model assigns different weights to different parts of the recruitment post and the resume, for example, the "work experience" part of the post may have a greater weight than the "interests and hobbies" part, and the "professional skills" part in the resume has a higher weight. The weight of different information can be adjusted according to the important needs of the position, and the matching degree scoring model calculates the matching degree score of the resume and the post according to the similarity between the post and the resume and the weight of each part of the information, and outputs a matching degree score.

[0096] In the embodiment, the talent matching model constructed by the recruitment vector model, the information extraction model, the weight evaluation model and the matching degree scoring model is used to match the recruitment post data and the resume data to obtain the most suitable candidate resume, solving the problem that the post requirements and resume advantages are not aligned in the prior art, resulting in low recruitment quality and talent loss.

[0097] Specifically, before using the talent matching model to match the recruitment post data and the resume data, the construction of the talent matching model needs to be completed in advance, so before the step of matching the recruitment post data and the resume data by the pre-constructed talent matching model to obtain the matching result in the above step S02, the method further comprises:

[0098] Step S0201, obtaining historical post information and historical resume information;

[0099] Step S0202, performing cluster analysis on the historical post information to obtain a talent matching dimension;

[0100] Step S0203, based on the talent matching dimension, performing information extraction on the historical post information and the historical resume information by a first deep learning and natural language processing model to obtain a first extraction result;

[0101] Step S0204, constructing a vector model optimization dataset according to the first extraction result, and training an initial vector model based on the vector model optimization dataset to obtain a recruitment vector model;

[0102] Step S0205, knowledge distillation is performed on the first deep learning and natural language processing model according to the first extraction result to obtain an information extraction model;

[0103] Step S0206, dimension weight evaluation is performed on the person-post matching dimension and historical resume information by a second deep learning and natural language processing model to obtain an evaluation result;

[0104] Step S0207, knowledge distillation is performed on the second deep learning and natural language processing model according to the evaluation result to obtain a weight evaluation model;

[0105] Step S0208, the first extraction result and the evaluation result are input into a third deep learning and natural language processing model for matching degree scoring to obtain a first scoring result;

[0106] Step S0209, knowledge distillation is performed on the third deep learning and natural language processing model according to the first scoring result to obtain a matching degree scoring model;

[0107] Step S02010, a person-post matching model is constructed according to the recruitment vector model, the information extraction model, the matching degree scoring model and the weight evaluation model.

[0108] The model construction scheme in this embodiment is as shown in Figure 2 The deep learning and natural language processing model used is DeepSeek R1, and other deep learning and natural language processing models such as GPT series (GPT-4o), Gemini, Kimi and ERNIE Bot can also be used. After receiving the historical post information and the historical resume information, clustering analysis is performed on the requirements of a large number of posts, and the matching dimensions can be divided into six parts (i.e., person-post matching dimensions), which are basic information, education experience, industry background, project experience, professional skills and general ability.

[0109] The vector matching and matching degree evaluation of the subsequent link need to be based on the extracted information, and the quality of the extraction will directly affect the effect of the retrieval, therefore, the information extraction not only needs to identify the related content, but also needs to complete the analysis, classification, summary and the like of the content, for this, the embodiment constructs an information extraction MoE model through a series of processes, that is, the first deep learning and natural language processing model extracts information based on the historical post information and the historical resume information, after obtaining the first extraction result, the first deep learning and natural language processing model is knowledge distilled to obtain the information extraction model.

[0110] The embodiment also constructs a recruitment field vector model fine-tuning data set based on the first extraction result, and trains the vector model based on the data set, so that the structure of the vector space is more in line with the semantic relationship in the field, enhances the understanding of professional vocabulary, and based on the fine-tuned vector model, the distinguishability of different posts and technical fields (such as front-end and back-end development) can be improved in the semantic space, and the pertinence of the post matching can be improved.

[0111] In addition, the extracted information will be used for the matching degree evaluation and the Top n retrieval tasks, in the matching degree evaluation task, the use of the extracted information can effectively shorten the context of the model input, make the key information densely distributed, and reduce the influence of the interference information on the matching result, and for the same reason, the extracted text is vectorized, and the matching calculation of each dimension is based on the vector, which can effectively improve the recall rate of this link, therefore, the first extraction result is also used to construct a vector model optimization data set in the embodiment, and then the initial vector model is trained through the vector model optimization data set to obtain a recruitment vector model, which improves the ability of the vector model in vectorizing specific information.

[0112] In addition, the model with large-scale parameters improves the comprehensive ability of the model while increasing the demand for computing resources, and the tasks in the recruitment field are relatively clear, therefore, the embodiment distills the ability of DeepSeek R1 in the recruitment field to a small model and performs targeted strengthening, which can achieve better results in limited resources, especially in the matching degree evaluation task, the use of the distilled small model can realize fast scoring, reordering and recommendation in limited computing resources, therefore, the embodiment selects DeepSeek R1 as the teacher model, respectively completes the dimension weight evaluation and the matching degree scoring in the subsequent, forms the knowledge distillation data set corresponding to each task, and finally performs knowledge distillation through the knowledge distillation data set to obtain the matching degree scoring model and the weight evaluation model.

[0113] It can be seen that in the present embodiment, deep learning and natural language processing models are used for information extraction, dimension weight evaluation and matching degree scoring in the recruitment field. Subsequently, in order to enhance the accuracy of the model and reduce the required computing power of the model, knowledge distillation is performed on the deep learning and natural language processing models that perform information extraction, dimension weight evaluation and matching degree scoring respectively, obtaining an information extraction model, a matching degree scoring model and a weight evaluation model, and finally completing the construction of the person-post matching model. This makes the model more accurate in extracting information, evaluating weights and scoring matching degrees when dealing with recruitment post data and resume data, and solves the problem of talent loss caused by not considering the advantages in the resume in the person-post matching task.

[0114] More specifically, after the step S02 of obtaining the matching result by matching the recruitment post data and the resume data through the pre-constructed person-post matching model, the method further comprises:

[0115] Step S03, receiving a post recruitment result;

[0116] Step S04, querying the resume data according to the post recruitment result to obtain employee information;

[0117] Step S05, according to the employee information, sending an initial employee evaluation form to an employee management end, filling the initial employee evaluation form by the employee management end to obtain a final employee evaluation form;

[0118] Step S06, receiving the final employee evaluation form, and optimizing the person-post matching model through the final employee evaluation form to obtain a model optimization result.

[0119] The system receives a post recruitment result from a job seeking platform or a recruitment system, which usually includes information such as the resume of the job seeker, the recruitment post, the recruitment time, etc. Then the received recruitment result is stored in the database, and each recruitment record is identified with its corresponding post information.

[0120] Then, according to the post recruitment result, the resume data is queried to obtain employee information. According to the recruitment result, the corresponding resume data is queried to obtain the detailed information of the employee (such as name, contact information, education, work experience, skill specialty, etc.), which provides data support for subsequent analysis.

[0121] According to the employee information, the corresponding initial employee evaluation form is generated, i.e. the personal information and work completion of the employee, and the initial employee evaluation form is sent to the employee management end. In the present embodiment, the employee management end at least includes an HR system and the superior leader corresponding to the employee.

[0122] In the employee evaluation form contains some basic questions, such as work experience, skill level, job fit, etc., for HR personnel and superiors to conduct preliminary evaluation of job seekers, but the above content is only a basic evaluation of employees, and cannot be optimized for the model, so you can add multiple scoring indicators, such as: skill matching degree, experience matching degree, cultural adaptability, etc., to form a detailed employee evaluation form, and the detailed employee evaluation form will also include the corresponding content filled in the resume information of the previous employee, which can further improve the efficiency and accuracy of information processing during model training.

[0123] After completing the final employee evaluation form in the employee management end, the feedback data collected is used to optimize the person-job matching model based on the final employee evaluation form. This process can be divided into the following steps:

[0124] (1) Recruitment vector model optimization: Use the actual matching situation in the employee evaluation form (such as employee adaptability, job performance) to optimize the recruitment vector model, for example, adjust the keyword weight or semantic vector in the recruitment position and resume, so that the model is more in line with the actual job requirements and candidate performance when generating vector representation of the position and resume;

[0125] (2) Information extraction model optimization: Adjust the rules and strategies of the information extraction model based on the actual feedback in the final employee evaluation form, so that the model can more accurately extract key information from resumes and job descriptions, such as the importance of "work experience" or "professional skills". If some skills or experience are highlighted as important in employee performance, the model will update the weight and pay more attention to the extraction and identification of these features;

[0126] (3) Weight evaluation model optimization: Based on the scores in the final evaluation form, adjust the weights of various features in the position and resume, for example, if the HR or superior's evaluation of certain skills is particularly high, the model will automatically increase the weight of that skill in the matching score, ensuring that the most valued skills and experience of the employer are paid more attention to in the matching process;

[0127] (4) Matching degree scoring model optimization: Use the data of the final employee evaluation form to optimize the matching degree scoring model, so that the score is more in line with the actual job requirements and employee performance. Through regression analysis or classification algorithms, the model will gradually adjust to make the matching degree score better reflect the actual recruitment effect. If the employee evaluation form shows that the matching degree score of some positions is low, the model will adjust through reinforcement learning and other technologies to retrain the scoring system to better predict and match new job seekers.

[0128] Thus, the model optimization result can be obtained. After optimization of the above steps, the updated human-post matching model is finally obtained. The model can more accurately evaluate the matching degree between the candidate and the post, and improve the efficiency and accuracy of recruitment.

[0129] In this embodiment, the human-post matching model is fine-tuned by continuously receiving feedback (final employee evaluation form), and each sub-model (recruitment vector model, information extraction model, weight evaluation model, and matching degree scoring model) is optimized, thereby improving the recruitment efficiency and ensuring that the matching degree of the post and the candidate is more accurate.

[0130] In this embodiment, the above scheme is used, specifically, the recruitment post data and the resume data are received, the recruitment post data and the resume data are matched by the pre-constructed human-post matching model to obtain a matching result, and the human-post matching model is constructed based on the recruitment vector model, the information extraction model, the weight evaluation model, and the matching degree scoring model. Thus, the human-post matching model constructed by the recruitment vector model, the information extraction model, the weight evaluation model, and the matching degree scoring model is used to match the recruitment post data and the resume data to obtain a matching result, which solves the problem that the post requirements and the resume advantages are not aligned in the prior art, resulting in low recruitment quality and talent loss, and improves the accuracy of human-post matching.

[0131] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as the above first embodiment can be referred to the above introduction, and will not be described in detail. On this basis, please refer to Figure 3 , in step S0205, the first deep learning and natural language processing model is knowledge distilled according to the first extraction result to obtain an information extraction model, and the information extraction method further includes steps S02051-S02055:

[0132] In step S02051, the first extraction result is classified by the first deep learning and natural language processing model based on the human-post matching dimension to obtain a first classification result.

[0133] In step S02052, the first extraction result is classified by a classification model based on the human-post matching dimension to obtain a second classification result.

[0134] In step S02053, the first classification result and the second classification result are cross-validated to obtain a dimension classification data set.

[0135] In step S02054, the first deep learning and natural language processing model is knowledge distilled according to the first extraction result to obtain an initial information extraction model.

[0136] Step S02055, according to the dimension classification data set, the sub-network of the initial information extraction model is activated through the gating network, and the initial information extraction model is nonlinearly transformed through the activation function, to obtain an information extraction model.

[0137] In the embodiment, the text segments are reasonably distributed to the expert models corresponding to the respective dimensions, so as to adapt to the discontinuous information extraction from the file. For example, the text content extracted from the PDF format resume file through the OCR technology has many text segments. In order to better train the gating network focusing on content distribution, the embodiment constructs a dimension classification data set. The specific construction method of the dimension classification data set includes:

[0138] The collected resume and post text segments are classified into corresponding dimensions by means of DeepSeek R1. There is a situation that one segment corresponds to multiple dimensions. At the same time, the same classification is performed based on the classification model (i.e. the bag model). Then, the classification results of the two models are cross-validated, and the inconsistent parts in the cross-validation are manually processed, so as to complete the construction of the dimension classification data set.

[0139] Then, the first deep learning and natural language processing model is knowledge distilled according to the first extraction result obtained in the previous step, to obtain an initial information extraction model.

[0140] Subsequently, based on the dimension classification data set, the embodiment aims at the gating network whether the i-th expert model is activated is set to dimension vector space wherein 0 represents inactivation, and 1 represents activation, is an input, is a gating network parameter. Overall, it can be expressed as:

[0141]

[0142] In the embodiment, whether the expert model is activated is regarded as a binary classification task, and the Sigmoid activation function is selected. The output of the gating network is expressed as:

[0143]

[0144] wherein, is a weight parameter, is a bias parameter.

[0145] At the same time, based on the strengthened knowledge distillation data set, the target function is established as follows:

[0146]

[0147] wherein, is the number of input samples, is the number of experts, represents the output of the gating network for the expert when the input is , i.e., the extraction result of the dimension .

[0148] After the above scheme completes the activation and nonlinear transformation of the distilled initial information extraction model, a more accurate information extraction model is obtained, which can obtain more accurate job information and resume information for matching in the recruitment field, thereby solving the problem of inaccurate information extraction, leading to errors in subsequent weight allocation and score matching, and resulting in talent loss.

[0149] More specifically, in the above embodiment, the step S0209 of obtaining a matching degree scoring model by knowledge distillation of the third deep learning and natural language processing model according to the first scoring result includes:

[0150] Step S02091, knowledge distillation of the third deep learning and natural language processing model to obtain an initial matching degree scoring model;

[0151] Step S02092, inputting the first extraction result and evaluation result into the initial matching degree scoring model to obtain a second scoring result;

[0152] Step S02093, difference calculation based on the first scoring result and the second scoring result to obtain a cross-entropy loss;

[0153] Step S02094, difference calculation based on the probability distribution of the first scoring result and the probability distribution of the second scoring result to obtain a divergence loss;

[0154] Step S02095, model optimization of the initial matching degree scoring model according to the cross-entropy loss and the divergence loss to obtain a matching degree scoring model.

[0155] The main task of this step is to transfer the ability of DeepSeek R1 in the six-dimensional information extraction, weight allocation, and matching degree evaluation tasks to the student model through distillation. Although the tasks are different, the overall method is similar. Taking the matching degree scoring model as an example, the process is as follows:

[0156] (1) Knowledge distillation of the DeepSeek R1 model for matching degree evaluation to distill an initial matching degree scoring model;

[0157] (2) The DeepSeek R1 model performing the matching degree evaluation is taken as a teacher model, and an initial matching degree scoring model is taken as a student model. The probability distribution of the teacher model and the probability distribution of the student model are calculated by the following formula:

[0158]

[0159] wherein, DeepSeek R1 model is represented by DeepSeek R1, the student model is represented by Student, the i-th input sample in the task is represented by xi, the output probability distribution of the model for the input xi is represented by P(yi | xi),

[0160] In the knowledge distillation process performed by the embodiment, the student model needs to learn the probability distribution output by the teacher model while learning the target task. Therefore, the joint cross-entropy loss and KL divergence loss are used to comprehensively evaluate the training in the distillation process. The cross-entropy loss is used to evaluate the difference between the output of the student model and the real label, and the KL divergence loss is used to measure the difference between the output distribution of the student model and the teacher model.

[0161] (3) Through the double loss mechanism (i.e., the cross-entropy loss and the KL divergence loss), the student model can better imitate the behavior of the teacher model, thereby improving the performance of the student model. The specific loss calculation is as follows: the calculation formula of the cross-entropy loss is as follows:

[0162]

[0163] The calculation formula of the KL divergence loss is as follows:

[0164]

[0165] The calculation formula of the total distillation loss is as follows:

[0166]

[0167] wherein, the cross-entropy loss is represented by Lce, the divergence loss is represented by Ldiv, the total number of samples is represented by N, the total distillation loss is represented by L, the final output of the teacher model is represented by y, the cross-entropy loss coefficient is represented by λ.

[0168] (4) After obtaining the total loss of the student model and the teacher model, the initial matching degree scoring model can be optimized according to the total loss to obtain the matching degree scoring model.

[0169] ​​In this embodiment, it should be noted that for different tasks, Corresponding adjustments need to be made. For example, in the distillation process of the information extraction expert model in each dimension, according to the characteristics of the extraction dimension, the output of the teacher model is strengthened accordingly, so that the model has stronger domain information mining capability. In this embodiment, the relevant public information of colleges, enterprises, social organizations and the like is collected to construct a corresponding knowledge base, and the RAG method is used to provide more time-effective information for each expert model. Specifically, the information extraction tasks focused on by each dimension expert model are as follows:

[0170] The basic information extraction expert model extracts ranges including mobile phone numbers, emails, ages, work locations, etc. By adding certain reasoning data, the model has certain reasoning ability, which can further complete the information, such as calculating the age based on the date of birth and completing the province based on the city.

[0171] The education experience extraction expert model extracts ranges including education, whether full-time, whether national, school name and national ranking, major information, etc. Combined with the information provided by the knowledge base, the model can identify and complete the information not mentioned.

[0172] The industry background extraction expert model focuses on mining the industry attributes involved in the description information of the resume and the post. Essentially, the model completes a multi-classification task. In actual scenarios, a description may involve multiple industry attributes, such as the project experience description of the wind and power conversion efficiency detection system mentioned in the resume, which involves new energy and software information service industry. For the post, the industry background extraction expert model can obtain more information of the post providing unit based on the knowledge base, thereby extracting the industry background.

[0173] The project experience extraction expert model mainly focuses on the scale, content, complexity, role of the post in the project, etc. Many resumes do not directly mention the project scale, etc. in the project experience description. Therefore, the dimension expert model analyzes the other mentioned content and mines the potential information.

[0174] The professional ability extraction expert model focuses on skill nouns and proficiency. The focus of this dimension includes skill extraction and cleaning. To improve the relevance of matching, the skill nouns extraction ignores irrelevant parts, such as office and PS, which are irrelevant to the expected post. Professional skills are one of the important dimensions of person-post matching. Improving the relevance of the extracted skills and the post can effectively improve the recall rate in the vector retrieval link.

[0175] The general ability extraction expert model focuses on analyzing the comprehensive quality of the candidate in terms of innovation ability, communication and cooperation ability, stress resistance and the like from the self-description, project description and the like in the resume. For a post, the post dimension expert model can analyze the requirements of the post on the general ability of the talent in combination with the company introduction and post description information in the knowledge base.

[0176] That is, by the above-mentioned model distillation, the optimized matching degree scoring model, information extraction model and weight evaluation model can be obtained, the overall precision of the talent-post matching model is improved, and the problem that the talent loss is caused by not considering the advantage content in the resume in the talent-post matching task is solved.

[0177] Further, the step S02 of obtaining the matching result by matching the recruitment post data and the resume data through the pre-constructed talent-post matching model comprises:

[0178] Step S021, information extraction is performed on the recruitment post data and the resume data through the information extraction model to obtain a second extraction result;

[0179] Step S022, weight evaluation is performed on the recruitment post data and the resume data through the weight evaluation model to obtain post dimension weight and resume dimension weight;

[0180] Step S023, the second extraction result is vectorized through the recruitment vector model to obtain a vector representation of the second extraction result;

[0181] Step S024, similarity ranking is performed on the second extraction result based on the post dimension weight and the resume dimension weight to obtain a plurality of candidate information;

[0182] Step S025, matching is performed on the plurality of candidate information through the matching degree scoring model to obtain a matching result.

[0183] The overall scheme of the embodiment is shown in Figure 4 After receiving new resume and post data, weight distribution and dimension information extraction are completed in parallel in an asynchronous manner. The extracted six dimensions of text are respectively vectorized through the recruitment vector model and stored. For a relatively simple post, the requirements of the candidate in terms of project experience may not be mentioned. The embodiment designs a preset value filling method to guide the vector model and the vector database to code and retrieve. Taking a Java development post as an example, when the project experience is not mentioned, the preset value filled is: "It is required to participate in Java development related projects, and to participate in optimization design of complex scenarios such as high concurrency".

[0184] Then, talent-post matching can be directly performed. The specific principle is as follows:

[0185] The vector of each dimension of the resume is The vector of each dimension of the post is Each post and resume has a corresponding weight, the weight of each dimension of the post is The weight of each dimension of the resume is Taking the process of matching resumes according to post information as an example, the calculation method of the initial screening recommendation score can be expressed as:

[0186]

[0187] After sorting based on similarity, the top n candidates can be obtained by initial screening. The information of each dimension of the candidate who passes the initial screening is obtained from the vector database. Based on the matching degree evaluation model, the real HR evaluation principle is simulated for scoring and sorting, so as to obtain the final result of matching the resume according to the person-post matching.

[0188] Similarly, based on The post can be matched according to the talent person-post matching by a similar method.

[0189] In the embodiment, the first classification result is obtained by classifying the first extraction result based on the first deep learning and natural language processing model according to the person-post matching dimension. The second classification result is obtained by classifying the first extraction result based on the classification model according to the person-post matching dimension. The first classification result and the second classification result are cross-validated to obtain a dimension classification data set. The first deep learning and natural language processing model is knowledge distilled according to the first extraction result to obtain an initial information extraction model. The subnetwork of the initial information extraction model is activated by a gating network according to the dimension classification data set, and the initial information extraction model is nonlinearly transformed by an activation function to obtain an information extraction model. Thus, the person-post matching model constructed by the recruitment vector model, the information extraction model, the weight evaluation model and the matching degree scoring model is used to match the post data and the resume data to obtain a matching result, which solves the problem that the post requirements and the resume advantages are not aligned in the prior art, resulting in low recruitment quality and talent loss, and improves the accuracy of person-post matching.

[0190] For example, in order to help understand the implementation process of the person-post matching method obtained after the above embodiment one, please refer to Figure 5 , Figure 5 A brief flowchart of a person-post matching method is provided, specifically:

[0191] Step S1 is the construction of the model, historical post information and historical resume information are preferentially acquired, and then they are cleaned, desensitized, and human-post matching data sets are matched, and finally DeepSeek R1 is selected as the teacher model, a small parameter model is selected as the student model, and distillation is performed on the six-dimensional information extraction, weight allocation, and matching degree evaluation, forming a six-dimensional information extraction MoE model, a weight allocation model, and a matching degree scoring model.

[0192] Step S is the preparation before matching, that is, after receiving new post data and resume data, different parts of information segments are handed over to the MoE model to complete information extraction, and at the same time, the weight allocation model is used to complete weight allocation for each resume and demand, and then the extracted information is dimensionally vectorized, and the dimensional weight information is stored in the vector database.

[0193] Step S3 is the matching stage, that is, the semantic similarity is calculated, the TOP n is selected based on the weight, the preliminary screening result is obtained, the context information of the extracted information is constructed into the resume and the post, and the matching degree evaluation model is used for scoring, the final fine arrangement is performed, and the matching result is given.

[0194] It should be noted that the above examples are only used for understanding the present application and do not constitute a limitation on the human-post matching method of the present application. Based on this technical concept, more forms of simple transformation are within the protection scope of the present application.

[0195] The present application also provides a human-post matching device, please refer to Figure 6 The human-post matching device comprises:

[0196] The receiving module 10 is used for receiving recruitment post data and resume data;

[0197] The matching module 20 is used for performing data matching on the recruitment post data and the resume data through the pre-constructed human-post matching model to obtain a matching result, wherein the human-post matching model is constructed based on a recruitment vector model, an information extraction model, a weight evaluation model, and a matching degree scoring model.

[0198] The human-post matching device provided by the present application adopts the human-post matching method in the above embodiments, and can solve the technical problem that the post requirements and the resume advantages are not aligned in the prior art, resulting in low enterprise recruitment quality and talent loss. Compared with the prior art, the human-post matching device provided by the present application has the same beneficial effects as the human-post matching method provided by the above embodiments, and other technical features in the human-post matching device are the same as the features disclosed in the above embodiments, which will not be repeated here.

[0199] The application provides a person-post matching device, which comprises at least one processor and a memory connected with the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the person-post matching method in Embodiment I.

[0200] Reference will be made to the following description Figure 7 which shows a structural schematic diagram of a person-post matching device suitable for implementing embodiments of the application. The person-post matching device in embodiments of the application can include, but is not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistant), PADs (Portable Application Description), PMPs (Portable Media Player), vehicle-mounted terminals (for example, vehicle-mounted navigation terminals), and the like, and fixed terminals such as digital TVs, desktop computers, and the like. Figure 7 The person-post matching device shown is merely an example and should not bring any limitation to the functions and use range of embodiments of the application.

[0201] As shown in Figure 7 , the person-post matching device can include a processing apparatus 1001 (for example, a central processor, a graphic processor, or the like) that can perform various appropriate actions and processes according to programs stored in a read-only memory 1002 or loaded from a storage apparatus 1003 into a random access memory 1004. In the random access memory 1004, various programs and data required for operation of the person-post matching device are also stored. The processing apparatus 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. An input / output interface 1006 is also connected to the bus. Generally, the following systems can be connected to the input / output interface 1006: input apparatuses 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, and the like; output apparatuses 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, and the like; the storage apparatus 1003 including, for example, a magnetic tape, a hard disk, and the like; and a communication apparatus 1009. The communication apparatus 1009 can allow the person-post matching device to perform wireless or wired communication with other devices to exchange data. Although the person-post matching device with various systems is shown in the figure, it should be understood that all the systems shown are not required to be implemented or provided. More or fewer systems can be alternatively implemented or provided.

[0202] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for executing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiments disclosed in the present application are executed.

[0203] The person-post matching device provided by the present application adopts the person-post matching method in the above-mentioned embodiments, and can solve the technical problem that the post requirements are not aligned with the resume advantages in the prior art, resulting in low quality of enterprise recruitment and talent loss. Compared with the prior art, the beneficial effects of the person-post matching device provided by the present application are the same as those of the person-post matching method provided by the above-mentioned embodiments, and other technical features in the person-post matching device are the same as those disclosed in the previous embodiment method, which will not be repeated here.

[0204] It should be understood that various parts of the present application can be realized by hardware, software, firmware or a combination thereof. In the description of the above-mentioned embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0205] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0206] The present application provides a computer readable storage medium having stored thereon computer readable program instructions (i.e. computer program) for executing the person-post matching method in the above-mentioned embodiments.

[0207] The computer readable storage medium provided in the application may be, for example, a U disk, but is not limited to an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system or device, or any combination thereof. More specific examples of the computer readable storage medium may include, but are not limited to, an electric connection with one or more conductive wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the embodiment, the computer readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system or device. The program code contained on the computer readable storage medium can be transmitted by any suitable medium, including but not limited to an electric wire, an optical cable, an RF (Radio Frequency), and the like, or any suitable combination thereof.

[0208] The computer readable storage medium described above may be contained in the personnel-post matching device, or may exist separately without being assembled into the personnel-post matching device.

[0209] The computer readable storage medium described above carries one or more programs, which, when executed by the personnel-post matching device, cause the personnel-post matching device to: receive recruitment post data and resume data; perform data matching on the recruitment post data and the resume data by using a pre-constructed personnel-post matching model to obtain a matching result, the personnel-post matching model being constructed based on a recruitment vector model, an information extraction model, a weight evaluation model, and a matching degree scoring model.

[0210] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0211] The flow diagrams and the block diagrams in the drawings are illustrations of architectures, functionalities, and operations of possible implementations of systems, methods, and computer program products according to various embodiments of present application. In this regard, each block in the flow diagrams or block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the block can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may

[0212] The modules involved in the embodiments of the present application can be implemented in the form of software or in the form of hardware. In some cases, the name of the module does not constitute a limitation on the module itself.

[0213] The readable storage medium provided by the present application is a computer readable storage medium, which stores computer readable program instructions (i.e. computer program) for executing the above-mentioned person-post matching method, and can solve the technical problem that the post requirements are not aligned with the resume advantages in the prior art, resulting in low quality of enterprise recruitment and talent loss. Compared with the prior art, the computer readable storage medium provided by the present application has the same beneficial effects as the person-post matching method provided by the above-mentioned embodiments, and will not be described here.

[0214] The application also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the method for matching a position with a person as described above.

[0215] The computer program product provided by the application can solve the technical problem that the position requirements are not aligned with the resume advantages in the prior art, resulting in low quality of enterprise recruitment and talent loss. Compared with the prior art, the beneficial effects of the computer program product provided by the application are the same as those of the method for matching a position with a person provided by the above-mentioned embodiments, and will not be repeated here.

[0216] The above only describes some embodiments of the application, and does not limit the patent scope of the application. Any equivalent structural transformation made by using the content of the specification and drawings, or direct / indirect application in other related technical fields under the technical concept of the application is included in the patent protection scope of the application.

Claims

1. A method for matching people to positions, characterized in that, The human position matching method comprises: receiving recruitment position data and resume data; performing data matching on the recruitment position data and the resume data through a pre-constructed human position matching model to obtain a matching result, the human position matching model being constructed based on a recruitment vector model, an information extraction model, a weight evaluation model, and a matching degree scoring model; Before the step of performing data matching on the recruitment position data and the resume data through the pre-constructed human position matching model to obtain the matching result, the method further comprises: obtaining historical position information and historical resume information; performing cluster analysis on the historical position information to obtain a human position matching dimension; performing information extraction on the historical position information and the historical resume information based on the human position matching dimension through a first deep learning and natural language processing model to obtain a first extraction result; constructing a vector model optimization data set according to the first extraction result, and training an initial vector model based on the vector model optimization data set to obtain a recruitment vector model; performing knowledge distillation on the first deep learning and natural language processing model according to the first extraction result to obtain an information extraction model; performing dimension weight evaluation on the human position matching dimension and the historical resume information through a second deep learning and natural language processing model to obtain an evaluation result; performing knowledge distillation on the second deep learning and natural language processing model according to the evaluation result to obtain a weight evaluation model; inputting the first extraction result and the evaluation result into a third deep learning and natural language processing model for matching degree scoring to obtain a first scoring result; performing knowledge distillation on the third deep learning and natural language processing model according to the first scoring result to obtain a matching degree scoring model; The step of performing knowledge distillation on the third deep learning and natural language processing model according to the first scoring result to obtain the matching degree scoring model comprises: performing knowledge distillation on the third deep learning and natural language processing model to obtain an initial matching degree scoring model; inputting the first extraction result and the evaluation result into the initial matching degree scoring model to obtain a second scoring result; performing difference calculation based on the first scoring result and the second scoring result to obtain a cross-entropy loss; The calculation formula of the cross-entropy loss is: wherein, DeepSeek R1 model, student model, DeepSeek R1 model for an input the resulting output probability distribution, student model for an input the resulting output probability distribution; performing difference calculation based on the probability distribution of the first scoring result and the probability distribution of the second scoring result to obtain a divergence loss; The calculation formula of the divergence loss is: performing model optimization on the initial matching degree scoring model according to the cross-entropy loss and the divergence loss to obtain the matching degree scoring model; The calculation formula of the total distillation loss is: wherein, is a cross-entropy loss, is a divergence loss, is a total number of samples, is a total distillation loss, is a final output of the teacher model, is a cross-entropy loss coefficient, which is adjusted according to the recruitment task; constructing the human position matching model based on the recruitment vector model, the information extraction model, the matching degree scoring model, and the weight evaluation model; After the step of performing data matching on the recruitment position data and the resume data through the pre-constructed human position matching model to obtain the matching result, the method further comprises: receiving a position recruitment result; According to the post recruitment result, the resume data is queried to obtain the candidate information; According to the candidate information, an initial employee evaluation form is sent to an employee management end, and the initial employee evaluation form is filled in by the employee management end to obtain a final employee evaluation form; The final employee evaluation form is received, and the human resource matching model is optimized through the final employee evaluation form to obtain a model optimization result, wherein the model optimization includes recruitment vector model optimization, information extraction model optimization, weight evaluation model optimization, and matching degree scoring model optimization.

2. The method of claim 1, wherein, The step of knowledge distillation of the first deep learning and natural language processing model based on the first extraction result to obtain an information extraction model includes: Based on the human resource matching dimension, the first extraction result is classified by the first deep learning and natural language processing model to obtain a first classification result; Based on the human resource matching dimension, the first extraction result is classified by a classification model to obtain a second classification result; The first classification result and the second classification result are cross-validated to obtain a dimension classification data set; According to the first extraction result, the first deep learning and natural language processing model is knowledge distilled to obtain an initial information extraction model; According to the dimension classification data set, the sub-network of the initial information extraction model is activated by a gating network, and the initial information extraction model is nonlinearly transformed by an activation function to obtain an information extraction model.

3. The method of claim 1, wherein, The step of data matching of the recruitment post data and the resume data by the pre-constructed human resource matching model to obtain a matching result includes: The information extraction model is used to extract information from the recruitment post data and the resume data to obtain a second extraction result; The weight evaluation model is used to evaluate the weight of the recruitment post data and the resume data to obtain post dimension weight and resume dimension weight; The second extraction result is vectorized by the recruitment vector model to obtain a vector representation of the second extraction result; Based on the post dimension weight and the resume dimension weight, the second extraction result is sorted by similarity to obtain a plurality of candidate information; The matching degree scoring model is used to match the plurality of candidate information to obtain a matching result.

4. A person-job matching device, characterized in that, The human resource matching device includes: A receiving module for receiving recruitment post data and resume data; A matching module for data matching of the recruitment post data and the resume data by a pre-constructed human resource matching model to obtain a matching result, wherein the human resource matching model is constructed based on a recruitment vector model, an information extraction model, a weight evaluation model, and a matching degree scoring model; The matching module is also used to obtain historical post information and historical resume information; The historical post information is clustered and analyzed to obtain a human resource matching dimension; Based on the human resource matching dimension, the historical post information and the historical resume information are extracted by a first deep learning and natural language processing model to obtain a first extraction result; construct a vector model optimization dataset according to the first extraction result, and train an initial vector model based on the vector model optimization dataset to obtain a recruitment vector model; perform knowledge distillation on the first deep learning and natural language processing model according to the first extraction result to obtain an information extraction model; perform dimension weight evaluation on the person-post matching dimension and historical resume information by a second deep learning and natural language processing model to obtain an evaluation result; perform knowledge distillation on the second deep learning and natural language processing model according to the evaluation result to obtain a weight evaluation model; input the first extraction result and the evaluation result into a third deep learning and natural language processing model to perform matching degree scoring to obtain a first scoring result; perform knowledge distillation on the third deep learning and natural language processing model according to the first scoring result to obtain a matching degree scoring model; The matching module is further configured to perform knowledge distillation on the third deep learning and natural language processing model to obtain an initial matching degree scoring model. input the first extraction result and the evaluation result into the initial matching degree scoring model to obtain a second scoring result; perform difference calculation based on the first scoring result and the second scoring result to obtain a cross-entropy loss; The calculation formula of the cross-entropy loss is: wherein, DeepSeek R1 model, student model, DeepSeek R1 model for input the resulting output probability distribution, student model for input the resulting output probability distribution; perform difference calculation based on the probability distribution of the first scoring result and the probability distribution of the second scoring result to obtain a divergence loss; The calculation formula of the divergence loss is: perform model optimization on the initial matching degree scoring model according to the cross-entropy loss and the divergence loss to obtain a matching degree scoring model; The calculation formula of the total distillation loss is: wherein, is a cross-entropy loss, is a divergence loss, is a total number of samples, is a total distillation loss, is a final output of the teacher model, is a cross-entropy loss coefficient, which is adjusted according to the recruitment task; construct a person-post matching model according to the recruitment vector model, the information extraction model, the matching degree scoring model, and the weight evaluation model; The matching module is further configured to receive a post recruitment result. query the resume data according to the post recruitment result to obtain candidate information; send an initial employee evaluation form to an employee management end according to the candidate information, and fill the initial employee evaluation form by the employee management end to obtain a final employee evaluation form; receive the final employee evaluation form, and perform model optimization on the person-post matching model by the final employee evaluation form to obtain a model optimization result, wherein the model optimization includes recruitment vector model optimization, information extraction model optimization, weight evaluation model optimization, and matching degree scoring model optimization.

5. A person-job matching device characterized by comprising: The device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the person-post matching method according to any one of claims 1 to 3.

6. A storage medium, characterized by The storage medium is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the person-post matching method according to any one of claims 1 to 3.

7. A computer program product, characterised in that, The computer program product comprises a computer program which, when executed by a processor, implements the steps of the person-post matching method according to any one of claims 1 to 3. The computer program product comprises a computer program which, when executed by a processor, implements the steps of the person-post matching method according to any one of claims 1 to 3.

Images