Large language model training method for fresh graduate resume screening and its application method

Abstract

The present application discloses a large language model training method for screening fresh graduates' resumes and its application method, the method comprising: for any j-th function under any i-th industry, inputting the corresponding input data into the large language model, the large language model is used to adjust the scoring range and / or weight of multiple recruitment dimensions according to key recruitment demand data and HR screening preference data; extracting the first feature data of each recruitment dimension in the job description data and the second feature data of each recruitment dimension in the sample fresh graduates' resume data; matching the first feature data and the second feature data, giving scores to each recruitment dimension according to the matching results, and performing weighted summation in combination with the weights of each recruitment dimension to obtain a predicted total score; when the prediction accuracy of the large language model is less than a preset threshold, readjusting the scoring range and / or weight until the prediction accuracy is ≥ the preset threshold. The present application is conducive to improving the accuracy and efficiency of fresh graduates' resume screening.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a large language model training method for screening fresh graduate resumes, a fresh graduate resume screening method based on a large language model, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] In current campus recruitment, recruiting companies are faced with the problem of screening a large number of fresh graduates' resumes. Since fresh graduates' resumes often lack rich work experience and the content is mostly concentrated on educational background, internship experience and extracurricular activities, the differences between resumes are small, so it is difficult to accurately evaluate the potential and ability of candidates through traditional resume screening methods. Specifically, traditional resume screening methods mostly rely on keyword matching and cannot effectively mine deep-level information in resumes, especially when judging fresh graduates' soft skills such as innovation, leadership, and teamwork ability. The resume screening effect of traditional resume screening methods is very limited. Especially in popular industries such as technology research and development, finance, and medical care, traditional resume screening methods find it difficult to identify the most promising and outstanding candidates who meet job requirements from a large number of similar fresh graduates' resumes. Summary of the invention

[0003] In view of this, the embodiments of the present application provide a large language model training method for screening fresh graduate resumes, a method for screening fresh graduate resumes based on a large language model, an electronic device, a computer-readable storage medium and a computer program product, which are used to solve at least one of the above technical problems.

[0004] In a first aspect, an embodiment of the present application provides a large language model training method for screening resumes of fresh graduates, including: obtaining input data of the large language model, the input data including key recruitment demand data of different functions in multiple industries, HR screening preference data of recruiters of different functions in multiple industries, job description data of recruitment positions of different functions in multiple industries, and multiple sample resume data of fresh graduates; preprocessing the input data to obtain preprocessed input data; for any j-th function under any i-th industry, inputting the preprocessed input data corresponding to the j-th function under the i-th industry into the large language model, the large language model is used to adjust the score range and / or weight of multiple preset recruitment dimensions at the j-th function under the i-th industry according to the preprocessed key recruitment demand data and HR screening preference data corresponding to the j-th function under the i-th industry, where i and j are positive integers; wherein the multiple recruitment dimensions include at least school characteristics, academic background, social activities, academic awards, and personal abilities;

[0005] The large language model is also used to extract the first feature data of each recruitment dimension in the preprocessed job description data corresponding to the jth function under the ith industry, and extract the second feature data of each recruitment dimension in the preprocessed sample fresh graduate resume data; match the first feature data and the second feature data of each recruitment dimension; assign corresponding scores to each recruitment dimension according to the matching results of each recruitment dimension, and perform weighted summation according to the scores corresponding to each recruitment dimension and the weights of each recruitment dimension to obtain the predicted total score of the sample fresh graduate resume data; determine the prediction accuracy of the large language model according to the comparison results of the predicted total scores of multiple sample fresh graduate resume data with the predetermined actual total scores; when the prediction accuracy is less than the preset threshold, return to the step of adjusting the score range and / or weight of multiple recruitment dimensions in the jth function under the ith industry until the prediction accuracy is greater than or equal to the preset threshold;

[0006] After the preprocessed input data corresponding to different functions in multiple industries are input into the large language model, and the prediction accuracy corresponding to each function under each industry is greater than or equal to the preset threshold, a trained large language model is obtained.

[0007] According to some embodiments of the present application, optionally, the multiple recruitment dimensions also include at least one of school ranking, professional background, internship experience, project experience, skill certificates, career interest matching and career development potential, and personal abilities include at least one of innovation ability, leadership, foreign language ability, programming ability, data analysis ability, mastery of professional tools, teamwork ability, problem-solving ability and communication ability.

[0008] According to some embodiments of the present application, optionally, the key recruitment demand data corresponding to the j-th function under the i-th industry includes n key recruitment dimensions of the j-th function under the i-th industry determined from multiple recruitment dimensions based on expert experience, and the n key recruitment dimensions are sorted according to the importance of each key recruitment dimension determined based on expert experience, and n is a positive integer;

[0009] The HR screening preference data corresponding to the j-th function under the ith industry includes m key recruitment dimensions of the j-th function under the ith industry determined from multiple recruitment dimensions based on HR historical screening preferences, and the m key recruitment dimensions are sorted according to the importance of each key recruitment dimension determined based on HR historical screening preferences, and m is a positive integer;

[0010] The large language model is specifically used to adjust the weights of multiple recruitment dimensions in the j-th function under the ith industry to their respective corresponding basic weights according to the n key recruitment dimensions of the j-th function under the ith industry; and to fine-tune the basic weights of at least some of the recruitment dimensions according to the m key recruitment dimensions of the j-th function under the ith industry, where the magnitude of the fine-tuning is smaller than the size of the basic weights before the fine-tuning.

[0011] According to some embodiments of the present application, optionally, the HR screening preference data corresponding to the j-th function under the ith industry is determined based on historical behavior data of multiple HRs when screening resumes of fresh graduates for the j-th function under the ith industry.

[0012] According to some embodiments of the present application, the input data is optionally preprocessed to obtain preprocessed input data, including: performing text cleaning on the input data to remove noise information in the input data; performing information extraction on the input data after text cleaning to extract key fields related to the recruitment dimension in the input data; and converting the key fields into structured data in a preset format based on natural language processing.

[0013] According to some embodiments of the present application, optionally, the large language model is specifically used to divide the preprocessed multiple sample resume data of recent graduates into different data sets, and for any j-th function under any ith industry, the large language model is trained multiple times using different data sets to obtain the prediction accuracy of the large language model for each data set; when the prediction accuracy of a preset number of data sets is greater than or equal to a preset threshold, the training process of the large language model for the j-th function under the ith industry is terminated.

[0014] In a second aspect, an embodiment of the present application provides a method for screening fresh graduate resumes based on a large language model, comprising: obtaining a trained large language model based on the large language model training method for screening fresh graduate resumes as provided in the first aspect; after preprocessing the target job description data of the target recruitment position and multiple fresh graduate resume data, inputting the preprocessed target job description data and the preprocessed multiple fresh graduate resume data into the trained large language model, the trained large language model is used to determine the target function under the target industry corresponding to the preprocessed target job description data, and calculating the predicted total score of each fresh graduate resume data based on the scoring range and weight of multiple recruitment dimensions for the target function under the target industry; sorting the multiple fresh graduate resume data based on the predicted total score of each fresh graduate resume data, and pushing the sorted multiple fresh graduate resume data to the target user.

[0015] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the steps of the method described above are implemented.

[0016] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which computer program instructions are stored, and the computer program instructions implement the steps of the method described above when executed by a processor.

[0017] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes computer program instructions, and when the computer program instructions are executed by a processor, the steps of the method described above are implemented.

[0018] The large language model training method for screening fresh graduate resumes, the fresh graduate resume screening method based on the large language model, the electronic device, the computer-readable storage medium and the computer program product provided by the embodiments of the present application, on the one hand, compared with the traditional resume screening method, the present application uses the large language model to deeply analyze the content of fresh graduate resumes, and can automatically extract key information that matches the recruitment position from the fresh graduate resumes, understand the potential and ability of fresh graduates, and improve the accuracy of resume scoring; on the other hand, compared with the resumes of working people, the present application selects more detailed and more suitable multiple different recruitment dimensions (such as school characteristics, academic background, social activities, academic awards and personal abilities, etc.) for the screening of fresh graduate resumes based on the characteristics of fresh graduate resumes, not only focusing on the hard skills of fresh graduates, but also effectively identifying the soft skills of fresh graduates. By scoring and weighted summing multiple different recruitment dimensions, the scoring of fresh graduate resumes can be made more comprehensive and multi-angle, it is easier to distinguish the differences between similar fresh graduate resumes, and candidates can be evaluated more comprehensively, improving the fairness and comprehensiveness of screening. On the other hand, for different industries and different functions, by combining key recruitment demand data and HR screening preference data, the scoring range and / or weight of multiple recruitment dimensions in different industries and different functions can be adjusted. This can enable the scoring and screening of fresh graduate resumes to be dynamically matched according to different industries and different functions, making it easier to distinguish fresh graduate resumes suitable for different industries and different functions, or to screen out fresh graduate resumes suitable for various industries and various functions. It can also more accurately imitate HR's screening ideas, making the scoring and screening of fresh graduate resumes more in line with HR's real needs and improving user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the embodiments of the present application, the following briefly introduces the drawings in the embodiments of the present application.

[0020] Figure 1A flowchart of a large language model training method for screening resumes of fresh graduates provided in an embodiment of the present application.

[0021] Figure 2 A flowchart of a method for screening resumes of fresh graduates based on a large language model provided in an embodiment of the present application.

[0022] Figure 3 A schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] The features and exemplary embodiments of various aspects of the present application will be described in detail below. In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain the present application, rather than to limit the present application. For those skilled in the art, the present application can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present application by illustrating the examples of the present application.

[0024] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0025] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0026] It is obvious to those skilled in the art that various modifications and changes can be made in the present application without departing from the spirit or scope of the present application. Therefore, the present application is intended to cover modifications and changes of the present application that fall within the scope of the corresponding claims (technical solutions for protection) and their equivalents. It should be noted that the implementation methods provided in the embodiments of the present application can be combined with each other without contradiction.

[0027] Before describing the technical solutions provided by the embodiments of the present application, in order to facilitate the understanding of the embodiments of the present application, the present application first specifically describes the problems existing in the related art:

[0028] In current campus recruitment, recruiting companies are faced with the problem of screening a large number of fresh graduates' resumes. Since fresh graduates' resumes often lack rich work experience and the content is mostly concentrated on educational background, internship experience and extracurricular activities, the differences between resumes are small, so it is difficult to accurately evaluate the potential and ability of candidates through traditional resume screening methods. Specifically, traditional resume screening methods mostly rely on keyword matching and cannot effectively mine deep-level information in resumes, especially when judging fresh graduates' soft skills such as innovation, leadership, and teamwork ability. The resume screening effect of traditional resume screening methods is very limited. Especially in popular industries such as technology research and development, finance, and medical care, traditional resume screening methods find it difficult to identify the most promising and outstanding candidates who meet job requirements from a large number of similar fresh graduates' resumes.

[0029] In view of the above research findings of the inventors, the embodiments of the present application provide a large language model training method for screening fresh graduate resumes, a method for screening fresh graduate resumes based on a large language model, an electronic device, a computer-readable storage medium and a computer program product, which can solve at least one of the above technical problems existing in the related art.

[0030] The following first introduces the large language model training method for screening resumes of fresh graduates provided in the embodiment of the present application.

[0031] Figure 1 A flow chart of a large language model training method for fresh graduate resume screening provided in an embodiment of the present application. Figure 1 As shown, the method may include the following steps S101 to S104.

[0032] S101: Obtain input data for a large language model, where the input data includes key recruitment demand data for different functions in multiple industries, HR screening preference data for recruiters with different functions in multiple industries, job description data for recruitment positions with different functions in multiple industries, and resume data for multiple sample fresh graduates.

[0033] Large Language Model (LLM) is an artificial intelligence system based on deep learning technology that can understand and generate natural language text. Large language models can be trained using large amounts of text data to learn the structure, grammar, semantics, and context of a language.

[0034] Among them, one industry can include at least one function. For example, in the financial industry, it can include multiple functions such as banking services, asset management, insurance and securities. For example, in the medical industry, it can include multiple functions such as clinical medicine, public health, pharmacy and medical insurance. Recruitment needs may be different for different industries and different functions, that is, different industries and different functions may have different requirements and priority characteristics for candidates. For example, the technology research and development industry pays more attention to programming ability, project experience and technical depth. For fresh graduates majoring in computer science, software engineering, etc., they may be more inclined to check whether they have actual project experience, programming ability, and mastery of technology stacks. For example, in the field of artificial intelligence, more attention may be paid to the candidate's mastery of cutting-edge technologies such as deep learning and machine learning, and whether there are related scientific research projects or technical publications.

[0035] For example, the financial industry often requires candidates to have solid data analysis skills, a good math background, and some financial internship experience. For fresh graduates majoring in finance, they may tend to be evaluated for their skills in financial instruments, financial analysis, etc. In this process, whether the candidate has advanced Excel operation skills, Python programming skills (for data analysis), and understanding of the financial market and relevant internship experience are key evaluation criteria.

[0036] For example, the medical industry places more emphasis on the candidate's academic background, scientific research ability, and whether he or she has participated in relevant internships or medical projects. For fresh graduates in related majors such as medicine and pharmacy, their scientific research project experience, academic paper publication, clinical internship experience, and other factors may be examined.

[0037] In S101, key recruitment demand data for different functions in multiple industries can be obtained, including but not limited to key recruitment demand data for different functions in industries such as technology research and development, finance, medical care, education, and manufacturing. Key recruitment demand data refers to the more critical or more important recruitment needs under the function of the industry. For example, for a certain function in the technology research and development industry, its key recruitment demand data may be actual project experience, programming ability, and mastery of the technology stack. For example, for a certain function in the financial industry, its key recruitment demand data may be data analysis ability, mathematical background, and financial internship experience.

[0038] In some embodiments, according to the recruitment needs of different functions in different industries, expert opinions and suggestions from various industries can be collected to obtain key recruitment demand data for different functions in multiple industries. First, in response to the recruitment needs of various functions in various industries, experts in various functions in various industries are surveyed to obtain the key standards and requirements of various functions in various industries when screening fresh graduates' resumes. The survey content may include the hard skills (such as programming skills, data analysis skills) and soft skills (such as communication skills, leadership) required for various functions in various industries, as well as preferences for candidate backgrounds (such as school rankings, internship experience, etc.). By combing and summarizing the recruitment needs of different functions in multiple industries, key recruitment demand data for different functions in multiple industries are obtained, which are converted into customized inputs for the large language model, so that the large language model can better adapt to the recruitment needs of various functions in various industries.

[0039] In this way, by deeply understanding the recruitment standards and requirements of each industry and each function, it is possible to convert the specific screening standards of each industry and each function into input data that can be used for large language model training based on the special requirements of each industry and each function. By feeding these requirements back to the large language model, the large language model can understand and evaluate the key capabilities and qualities related to each industry and each function when analyzing the resumes of fresh graduates. For example, the various functions under the technology research and development industry attach much more importance to programming skills than non-technical positions, while the various functions under the financial industry have higher requirements for mathematical modeling capabilities, and the various functions under the medical industry focus more on academic background and scientific research capabilities.

[0040] The screening preference data of recruiters (Human Resources, HR for short) in different industries and different functions may also be different. For example, for a certain function in the technology research and development industry, a large number of HRs may prefer or pay more attention to the actual project experience, programming ability, technology stack mastery, and teamwork ability of fresh graduates. For example, for a certain function in the financial industry, a large number of HRs may prefer or pay more attention to the school ranking, data analysis ability, mathematical background, and financial internship experience of fresh graduates. Therefore, in S101, the screening preference data of HRs in various industries and functions, or the common screening preference data of multiple HRs in various industries and functions, can be collected.

[0041] Different industries and different functions can correspond to at least one recruitment position. For example, under a certain function in the technology research and development industry, there is a recruitment position of "software development engineer". In S101, the job description data of recruitment positions under various functions in various industries can be obtained. The job description data describes the recruitment requirements of the recruitment position, such as but not limited to school requirements, major requirements, internship experience requirements and / or skill certificate requirements. In addition, in S101, multiple sample fresh graduate resume data can be obtained in batches from the recruitment platform. The sample fresh graduate resume data is the fresh graduate resume data used for training the large language model.

[0042] S102: Preprocess the input data to obtain preprocessed input data.

[0043] In S102, based on natural language processing (NLP), key recruitment demand data of different functions in multiple industries, HR screening preference data of different functions in multiple industries, job description data of recruitment positions of different functions in multiple industries, and multiple sample resume data of fresh graduates can be pre-processed to facilitate large language model processing.

[0044] S103: For any j-th function under any i-th industry, the pre-processed input data corresponding to the j-th function under the i-th industry is input into the large language model.

[0045] The i-th industry can be any industry. The j-th function can be any function under the i-th industry, and i and j are positive integers. For the j-th function under the i-th industry, the pre-processed key recruitment demand data, HR screening preference data, job description data, and sample fresh graduate resume data corresponding to the j-th function under the i-th industry can be input into the large language model.

[0046] The large language model can be used to adjust the scoring range and / or weight of multiple preset recruitment dimensions for the j-th function under the ith industry based on the pre-processed key recruitment demand data and HR screening preference data corresponding to the j-th function under the ith industry.

[0047] According to the characteristics of fresh graduates' resumes, multiple different recruitment dimensions that are more refined and more suitable for fresh graduates' resume screening can be selected. For example, multiple recruitment dimensions can at least include school characteristics, academic background, social activities, academic awards and personal abilities. In different industries and different functions, the scoring ranges and / or weights of multiple recruitment dimensions can be different, and can be adjusted specifically according to the pre-processed key recruitment demand data and HR screening preference data of each industry and each function. For example, when the pre-processed key recruitment demand data and HR screening preference data corresponding to the jth function under the i-th industry pay more attention to N recruitment dimensions, the scoring range and / or weight of the N recruitment dimensions in the jth function under the i-th industry can be increased, and / or the scoring range and / or weight of other recruitment dimensions except the N recruitment dimensions in the jth function under the i-th industry can be lowered, so that the scoring and screening of fresh graduates' resumes can be dynamically matched according to different industries and different functions, making it easier to distinguish fresh graduates' resumes suitable for different industries and different functions, or to screen fresh graduates' resumes suitable for different industries and different functions.

[0048] The large language model can also be used to extract the first feature data of each recruitment dimension in the preprocessed job description data corresponding to the j-th function under the ith industry, and extract the second feature data of each recruitment dimension in the preprocessed sample fresh graduate resume data; match the first feature data and the second feature data of each recruitment dimension; give each recruitment dimension a corresponding score according to the matching results of each recruitment dimension, and perform weighted summation according to the corresponding score of each recruitment dimension and the weight of each recruitment dimension to obtain the predicted total score of the sample fresh graduate resume data.

[0049] Specifically, for any job position of the jth function under the i-th industry, the large language model can extract the first feature data of each recruitment dimension, i.e., recruitment requirements, such as school characteristics requirements, academic background requirements, social activities requirements, and academic award requirements, from the preprocessed job description data of the job position. For example, in some examples, natural language processing techniques (such as text classification, keyword extraction, etc.) can be used to extract information such as core skills, educational requirements, and professional background required for the job position from the job description data. For example, for the job position of data analysis, the job description data may mention "familiarity with Python programming language" and "experience in data analysis", and the large language model can extract the first feature data of these recruitment dimensions from it.

[0050] At the same time, for any sample fresh graduate resume data, the large language model can extract the second feature data of each recruitment dimension from the preprocessed sample fresh graduate resume data, such as the school characteristics, academic background, social activities and academic awards of the sample fresh graduate. For example, in some examples, based on natural language processing, the second feature data of each recruitment dimension can be extracted from the preprocessed sample fresh graduate resume data, such as but not limited to school background, major, internship experience, project experience, skill certificates, language ability, etc. Specifically, the large language model can use information extraction technology (such as named entity recognition, relationship extraction, syntactic analysis, etc.) to extract key fields from the sample fresh graduate resume data. For example, for a sample fresh graduate resume data, the large language model can automatically identify the school name, degree, graduation time, project experience, technology stack used, whether there are relevant certificates, etc.

[0051] Then, the large language model can match the first feature data and the second feature data of each recruitment dimension one by one, evaluate whether the second feature data of each recruitment dimension meets the recruitment requirements of the corresponding recruitment dimension, and obtain the matching results of each recruitment dimension. Then, according to the matching results of each recruitment dimension and the scoring range of each recruitment dimension in the jth function under the i-th industry, the corresponding score of each recruitment dimension is given. For example, the score corresponding to recruitment dimension 1 is 3 points, the score corresponding to recruitment dimension 2 is 5 points, the score corresponding to recruitment dimension 3 is 10 points, and so on.

[0052] For any recruitment dimension, the score corresponding to the recruitment dimension is selected from the scoring range of the recruitment dimension for the jth function under the ith industry. For example, if the scoring range of the recruitment dimension for the jth function under the ith industry is 0-10 points, then the score corresponding to the recruitment dimension can be any value between 0-10 points. It should be noted that the scoring range of a recruitment dimension can be divided into at least one scoring level. For example, taking the recruitment dimension as school ranking and the scoring range of 0-10 points as an example, for example, if the school ranking is not in the top 100 in the country, 0 points will be given, the school ranking is in the top 100 in the country, 3 points will be given, the school ranking is in the top 50 in the country, 5 points will be given, and the school ranking is in the top 10 in the country. 10 points are given. The specific details can be flexibly adjusted according to the actual situation, and this application does not limit this.

[0053] After obtaining the scores corresponding to each recruitment dimension, the large language model can perform weighted summation based on the scores corresponding to each recruitment dimension and the weights of each recruitment dimension to obtain the predicted total score of the sample fresh graduate resume data.

[0054] In some examples, for example, the predicted total scores of fresh graduate resume data and sample fresh graduate resume data can be calculated according to the following expression:

[0055]

[0056] Among them, S represents the total predicted score, w i represents the weight of the i-th recruitment dimension, s i It represents the score corresponding to the i-th recruitment dimension, and N represents the number of recruitment dimensions.

[0057] The large language model can also be used to determine the prediction accuracy of the large language model based on the comparison results of the predicted total scores of multiple sample fresh graduate resume data with the predetermined actual total scores; when the prediction accuracy is less than a preset threshold, return to the step of adjusting the score range and / or weight of multiple recruitment dimensions for the jth function under the i-th industry until the prediction accuracy is greater than or equal to the preset threshold.

[0058] Specifically, the actual total score of each sample fresh graduate resume data can be predetermined, that is, known. The predicted total score of each sample fresh graduate resume data can be compared with the actual total score of each sample fresh graduate resume data one by one. For any sample fresh graduate resume data, if the difference between the predicted total score of the sample fresh graduate resume data and the actual total score of the sample fresh graduate resume data is less than or equal to the preset error, the prediction is considered accurate; otherwise, the prediction is considered inaccurate. The proportion of sample fresh graduate resume data with accurate predictions can be counted to obtain the prediction accuracy. When the prediction accuracy is less than the preset threshold, the step of adjusting the scoring range and / or weight of multiple recruitment dimensions for the jth function under the i-th industry can be returned until the prediction accuracy is greater than or equal to the preset threshold. The size of the preset error and the preset threshold can be flexibly adjusted according to actual conditions, and this application does not limit this.

[0059] S104: After the pre-processed input data corresponding to different functions in multiple industries are input into the large language model, and the prediction accuracy corresponding to each function under each industry is greater than or equal to a preset threshold, a trained large language model is obtained.

[0060] The preprocessed input data corresponding to different functions in multiple industries can be input into the large language model in sequence, so that the large language model can be trained for each function in each industry respectively until the prediction accuracy of the large language model corresponding to each function in each industry is greater than or equal to a preset threshold, thereby obtaining a trained large language model.

[0061] The large language model training method for screening fresh graduate resumes provided in the embodiments of the present application, on the one hand, compared with traditional resume screening methods, the present application uses a large language model to deeply analyze the content of fresh graduate resumes, and can automatically extract key information that matches the recruitment position from fresh graduate resumes, understand the potential and ability of fresh graduates, and improve the accuracy of resume scoring; on the other hand, compared with resumes of working people, the present application selects more detailed and more suitable multiple different recruitment dimensions (such as school characteristics, academic background, social activities, academic awards and personal abilities, etc.) for the screening of fresh graduate resumes based on the characteristics of fresh graduate resumes. It not only focuses on the hard skills of fresh graduates, but also can effectively identify the soft skills of fresh graduates. By scoring and weighted summing multiple different recruitment dimensions, the scoring of fresh graduate resumes can be made more comprehensive and multi-angle, and the differences between similar fresh graduate resumes can be distinguished. Candidates can be evaluated more comprehensively, and the fairness and comprehensiveness of screening can be improved. On the other hand, for different industries and different functions, by combining key recruitment demand data and HR screening preference data, the scoring range and / or weight of multiple recruitment dimensions for different functions in different industries can be adjusted, so that the scoring and screening of fresh graduate resumes can be dynamically matched according to different industries and different functions, making it easier to distinguish fresh graduate resumes suitable for different industries and different functions, or to screen fresh graduate resumes suitable for various industries and various functions, and it can more accurately imitate HR's screening ideas, so that the scoring and screening of fresh graduate resumes are more in line with HR's real needs and improve user experience. In addition, the screening of fresh graduate resumes based on a large language model helps to improve recruitment efficiency and reduce the burden of manual screening.

[0062] According to some embodiments of the present application, the multiple recruitment dimensions may also include at least one of school ranking, professional background, internship experience, project experience, skill certificate, career interest matching and career development potential. Personal ability may include at least one of innovation ability, leadership, foreign language ability, programming ability, data analysis ability, professional tool mastery, teamwork ability, problem-solving ability and communication ability.

[0063] The large language model can deeply analyze the content of fresh graduates' resumes. For example, if the fresh graduate's resume data shows that the fresh graduate is a core member of the student union, a core member of a club, and / or has organized and planned some activities or projects, then the fresh graduate is considered to have leadership. For example, if the fresh graduate's resume data shows that the fresh graduate has participated in a team and completed some activities or projects with others, then the fresh graduate is considered to have teamwork ability.

[0064] School ranking and school characteristics: The large language model can score graduates based on the ranking, reputation, and academic characteristics of the school they graduated from. The higher the school ranking, the higher the score. For some specific industries, such as the financial industry, graduates from certain well-known schools may receive higher scores.

[0065] Professional and academic background: The large language model can analyze whether the candidate's major meets the recruitment requirements and whether he has relevant academic background (such as scientific research projects, paper publications, etc.). For example, in the medical industry, a background in biomedicine will receive a higher score.

[0066] Internship experience and project experience: For the technology R&D industry, project experience is particularly important. The big language model can evaluate the project experience listed in the resume of fresh graduates, especially their role and actual contribution in the project, such as programming ability, technology stack usage, etc.

[0067] Skill certificates and mastery of professional tools: If graduates have skill certificates related to the recruitment position or have mastered professional tools commonly used in the industry (such as advanced Excel operations, Python programming, data analysis tools, etc.), they will be given extra points based on the value and scarcity of the certificates.

[0068] Leadership, teamwork and problem-solving skills: If the fresh graduate's resume mentions that the fresh graduate has played an important role in the team or demonstrated leadership, the big language model can score based on dimensions such as leadership, teamwork, and problem-solving skills.

[0069] In some examples, the resumes of recent graduates can be evaluated based on 20 recruitment dimensions, such as school ranking, school characteristics, professional background, academic background, internship experience, project experience, skill certificates, innovation ability, leadership, social activities, academic awards, foreign language ability, programming ability, data analysis ability, mastery of professional tools, teamwork ability, problem-solving ability, communication ability, career interest matching and career development potential.

[0070] In this way, this application selects more detailed and more suitable multiple recruitment dimensions for the screening of fresh graduates' resumes based on the characteristics of fresh graduates' resumes. It not only focuses on the hard skills of fresh graduates (such as programming ability, data analysis ability, etc.), but also can effectively identify the soft skills of fresh graduates (such as communication ability, leadership, etc.). By scoring and weighted summing up multiple different recruitment dimensions, the scoring of fresh graduates' resumes can be made more comprehensive and multi-angle, which can make it easier to distinguish the differences between similar fresh graduates' resumes, and can more comprehensively evaluate candidates, thereby improving the fairness and comprehensiveness of screening.

[0071] According to some embodiments of the present application, optionally, the key recruitment demand data corresponding to the j-th function under the ith industry may include n key recruitment dimensions for the j-th function under the ith industry determined from multiple recruitment dimensions based on expert experience, the n key recruitment dimensions are sorted according to the importance of each key recruitment dimension determined based on expert experience, and n is a positive integer.

[0072] As mentioned above, for different industries and different functions, expert opinions and suggestions of each industry and each function can be collected. For the jth function under the ith industry, expert opinions and suggestions of the jth function under the ith industry can be collected. Based on expert experience, n key recruitment dimensions of the jth function under the ith industry can be determined from multiple recruitment dimensions, and the n key recruitment dimensions can be ranked according to the importance of each key recruitment dimension determined based on expert experience.

[0073] For example, for a certain function in the technology R&D industry, the n key recruitment dimensions may be project experience, programming ability, and professional tool mastery. According to the importance of each key recruitment dimension, project experience ranks first, programming ability ranks second, and professional tool mastery ranks third.

[0074] Correspondingly, the HR screening preference data corresponding to the j-th function under the i-th industry may include m key recruitment dimensions for the j-th function under the i-th industry determined from multiple recruitment dimensions based on HR historical screening preferences. The m key recruitment dimensions may be sorted according to the importance of each key recruitment dimension determined based on HR historical screening preferences, and m is a positive integer.

[0075] HR screening preference data for different functions in different industries may also be different. For example, for a certain function in the technology research and development industry, a large number of HRs may prefer or pay more attention to the actual project experience, programming skills, professional tool mastery, and teamwork skills of fresh graduates. For the jth function in the ith industry, HR historical screening preference information can be obtained from the recruitment platform. Based on HR historical screening preferences, m key recruitment dimensions for the jth function in the ith industry can be determined from multiple recruitment dimensions. The m key recruitment dimensions can be sorted according to the importance of each key recruitment dimension determined based on HR historical screening preferences.

[0076] For example, for a certain function in the technology R&D industry, the m key recruitment dimensions may be project experience, programming ability, professional tool mastery, and teamwork ability. According to the importance of each key recruitment dimension, project experience ranks first, programming ability ranks second, professional tool mastery ranks third, and teamwork ability ranks fourth.

[0077] It should be noted that the n key recruitment dimensions determined based on expert experience and the m key recruitment dimensions determined based on HR historical screening preferences may be the same or may be different, and this application does not limit this.

[0078] In some embodiments, the large language model can be specifically used to adjust the weights of multiple recruitment dimensions for the j-th function in the ith industry to their respective corresponding basic weights based on the n key recruitment dimensions for the j-th function in the ith industry; and to fine-tune the basic weights of at least some of the recruitment dimensions based on the m key recruitment dimensions for the j-th function in the ith industry, with the fine-tuning amplitude being smaller than the size of the basic weights before the fine-tuning.

[0079] For example, according to the n key recruitment dimensions of the jth function under the ith industry, the basic weights of the n key recruitment dimensions in the jth function under the ith industry can be set relatively large, and the basic weights of the other recruitment dimensions except the n key recruitment dimensions in the jth function under the ith industry can be set relatively small. According to the m key recruitment dimensions of the jth function under the ith industry, the basic weights of at least some recruitment dimensions can be fine-tuned. For example, the weights of the m key recruitment dimensions can be increased again on the basis of their respective basic weights, and the weights of the other recruitment dimensions except the m key recruitment dimensions can be reduced again on the basis of the basic weights. This is fine-tuning, and the fine-tuning amplitude is less than the original size of the basic weight, that is, the fine-tuning amplitude is less than the size of the basic weight before fine-tuning. For example, if the basic weight before fine-tuning is 0.5, the fine-tuning amplitude should be less than 0.5, such as the fine-tuning amplitude can be 0.1, 0.2 or 0.3.

[0080] According to some embodiments of the present application, optionally, each recruitment dimension may adopt a preset standardized scoring range, and the scoring range of each recruitment dimension may be fixed, for example, 0 to 5 points, 0 to 10 points, etc.

[0081] According to other embodiments of the present application, optionally, similar to adjusting weights, the large language model can also be used to adjust the scoring ranges of multiple recruitment dimensions at the jth function in the ith industry to their respective corresponding basic scoring ranges based on the n key recruitment dimensions of the jth function in the ith industry; and fine-tune the basic scoring ranges of at least some recruitment dimensions based on the m key recruitment dimensions of the jth function in the ith industry, with the fine-tuning amplitude being smaller than the size of the basic scoring range before the fine-tuning. The scoring range adjustment process is similar to the weight adjustment process and will not be repeated here.

[0082] In this way, for different industries and different functions, by combining key recruitment demand data and HR screening preference data, the scoring range and / or weight of multiple recruitment dimensions for different functions in different industries can be adjusted, so that the scoring and screening of fresh graduate resumes can be dynamically matched according to different industries and different functions, making it easier to distinguish fresh graduate resumes suitable for different industries and different functions, or to screen fresh graduate resumes suitable for various industries and various functions, and it can more accurately imitate HR's screening ideas, so that fresh graduate resume scoring and screening are more in line with HR's real needs and improve user experience. In addition, the screening of fresh graduate resumes based on a large language model helps improve recruitment efficiency and reduce the burden of manual screening.

[0083] According to some embodiments of the present application, optionally, the HR screening preference data corresponding to the j-th function under the ith industry may be determined based on historical behavior data of multiple HRs when screening resumes of fresh graduates for the j-th function under the ith industry.

[0084] For example, historical behavior data of multiple HRs on the recruitment platform when screening fresh graduates' resumes for the jth function in the ith industry can be collected. The historical behavior data includes but is not limited to the viewing time, click-through rate and / or screening criteria of fresh graduates' resumes. These data will help understand HR's preferences when screening fresh graduates' resumes for the jth function in the ith industry.

[0085] For example, in some embodiments, if some resumes are viewed for a long time and have a high click-through rate, it means that the content of these resumes has a greater appeal to HR. By analyzing multiple resumes that have been viewed for a long time and have a high click-through rate, it is possible to identify which recruitment dimensions (such as school background, project experience, skill certificates, etc.) are more in line with HR's concerns. By selecting recruitment dimensions that are commonly preferred by multiple HRs, the HR screening preference data corresponding to the jth function under the i-th industry can be obtained.

[0086] For another example, in some embodiments, for the jth function under the i-th industry, HR usually uses specific screening criteria in the resume screening process, such as giving priority to graduates from certain specific schools, whether they have relevant certificates or skills, etc. By recording these historical behavior data, HR's screening preferences can be understood and converted into training data for the large language model. For example, if some HR pays more attention to school rankings, the large language model will give priority to school rankings as a recruitment dimension when evaluating fresh graduates' resumes.

[0087] For another example, in some embodiments, during the screening process, HR often scores or marks resumes (such as "qualified", "pending", "not compliant"). By analyzing multiple resumes with high scores and / or marked as qualified, it is possible to identify which recruitment dimensions (such as school background, project experience, skill certificates, etc.) are more in line with HR's concerns. By selecting recruitment dimensions that are commonly preferred by multiple HRs, HR screening preference data corresponding to the jth function under the i-th industry can also be obtained.

[0088] HR screening preference data helps the big language model learn HR's actual screening preferences and understand which recruitment dimensions are more likely to be selected by HR, so as to adjust the scoring range and / or weight of multiple recruitment dimensions for different functions in different industries. This helps the big language model to more accurately imitate HR's screening ideas and further improve the application effect of the big language model in actual recruitment.

[0089] According to some embodiments of the present application, optionally, S102: preprocessing the input data to obtain preprocessed input data may include the following steps 1 to 3.

[0090] Step 1: Perform text cleaning on the input data to remove noise information in the input data.

[0091] The input data may include key recruitment demand data for different functions in multiple industries, HR screening preference data for different functions in multiple industries, job description data for recruitment positions in different functions in multiple industries, and multiple sample fresh graduate resume data. In step 1, the input data may be cleaned to remove noise information in the input data. For example, the noise information includes, but is not limited to, redundant symbols and non-standard format content.

[0092] Step 2: Extract information from the input data after text cleaning and extract key fields related to the recruitment dimension in the input data.

[0093] Step 3: Convert key fields into structured data in a preset format based on natural language processing.

[0094] Taking the sample fresh graduate resume data as an example, the sample fresh graduate resume data is usually presented as free-format text. In order to facilitate subsequent processing, the sample fresh graduate resume data needs to be cleaned and formatted. This process includes removing irrelevant noise information (such as redundant symbols, non-standard formats, etc.), and annotating and classifying the key fields in the sample fresh graduate resume data. For example, the named entity recognition (NER) technology is used to identify the key fields such as school name, major name, company name, etc. in the sample fresh graduate resume data, and convert them into a standardized format.

[0095] Then, through natural language processing technology (such as dependency syntax analysis, information extraction, etc.), the key fields in the sample fresh graduate resume data are converted into structured data. For example, the "participated in a machine learning project" in the sample fresh graduate resume data is converted into "project experience: machine learning project, role: project member, skills: machine learning, Python". This structured data will provide clearer and more standardized input for subsequent large language model analysis.

[0096] By preprocessing the input data, the large language model can efficiently and accurately process each input data (such as each sample fresh graduate resume data), providing a reliable basis for subsequent scoring and sorting.

[0097] In the process of training a large language model, avoiding overfitting is a key challenge. According to some embodiments of the present application, optionally, in order to prevent the model from overfitting the training data and affecting its performance in real data, at least one strategy of cross-validation, A / B testing, regularization and data augmentation can be adopted.

[0098] Cross-Validation: Using the K-fold cross-validation method, multiple sample fresh graduate resume data are divided into K data sets (or subsets). Through multiple training and validation, the performance of the large language model on different data sets is evaluated, thereby reducing the overfitting of the large language model to a specific data set.

[0099] A / B testing: Use A / B testing to compare the performance of different model versions during the score range and / or weight adjustment process of the large language model. By comparing the performance of the new and old versions in real time, select the score range and / or weight adjustment strategy that best meets the recruitment needs.

[0100] Regularization: During the training process, L1 or L2 regularization can be applied to limit excessive changes in model parameters and avoid overfitting the model to the training data.

[0101] Data Augmentation: When the amount of training data is insufficient, data augmentation technology is used to increase the diversity of data and the generalization ability of training by generating different sample variants (such as synonym replacement, text reorganization, etc.).

[0102] Taking cross-validation as an example, the large language model can be specifically used to divide the pre-processed data of multiple sample fresh graduates' resumes into different data sets. For any j-th function under any i-th industry, the large language model is trained multiple times using different data sets to obtain the prediction accuracy of the large language model for each data set. The large language model can also be used to end the training process of the large language model for the j-th function under the i-th industry when the prediction accuracy of a preset number of data sets is greater than or equal to a preset threshold, thereby reducing the overfitting of the large language model to a specific data set. The preset number can be less than or equal to the total number K of data sets, and the size of the preset number can be flexibly adjusted according to actual conditions, and this application does not limit this.

[0103] According to some embodiments of the present application, optionally, in order to make the large language model have better accuracy and real-time performance, input data such as key recruitment demand data for different functions in multiple industries and HR screening preference data for different functions in multiple industries can be obtained regularly, and the large language model can be updated and retrained based on the input data obtained regularly.

[0104] The trained large language model can not only efficiently analyze the resumes of fresh graduates, but also adjust the scoring range and / or weight of multiple recruitment dimensions based on HR screening preference data to achieve personalized resume scoring.

[0105] Based on the large language model training method for screening fresh graduate resumes provided in the above-mentioned embodiment, the present application also provides a method for screening fresh graduate resumes based on a large language model.

[0106] Figure 2 A flow chart of a method for screening fresh graduates' resumes based on a large language model provided in an embodiment of the present application. Figure 2 As shown, the method for screening resumes of fresh graduates based on a large language model provided in an embodiment of the present application may include the following steps S201 to S203.

[0107] S201: Based on the large language model training method for screening resumes of fresh graduates, a trained large language model is obtained.

[0108] S202: After preprocessing the target job description data of the target recruitment position and the resume data of multiple fresh graduates, the preprocessed target job description data and the preprocessed resume data of multiple fresh graduates are input into the trained large language model. The trained large language model is used to determine the target function under the target industry corresponding to the preprocessed target job description data, and calculate the predicted total score of each fresh graduate resume data according to the scoring range and weight of multiple recruitment dimensions for the target function under the target industry.

[0109] Specifically, the target recruitment position can be any recruitment position. In S202, the target position description data and multiple fresh graduate resume data of the target recruitment position can be preprocessed. The preprocessing process is the same as the preprocessing process of the position description data and the sample fresh graduate resume data mentioned above, and will not be repeated here. After preprocessing, the preprocessed target position description data and the preprocessed multiple fresh graduate resume data can be input into the trained large language model. The trained large language model can be used to determine the target function under the target industry corresponding to the preprocessed target position description data, and calculate the predicted total score of each fresh graduate resume data according to the scoring range and weight of multiple recruitment dimensions in the target function under the target industry.

[0110] Specifically, the trained large language model can extract the first feature data of each recruitment dimension in the preprocessed target position description data. For any fresh graduate resume data, the trained large language model can extract the second feature data of each recruitment dimension in the preprocessed fresh graduate resume data. Then, the trained large language model can match the first feature data and the second feature data of each recruitment dimension to obtain the matching results of each recruitment dimension. Then, the trained large language model can give the corresponding score to each recruitment dimension according to the matching results of each recruitment dimension and the scoring range of each recruitment dimension when it is in the target function under the target industry. Then, the trained large language model can perform weighted summation according to the corresponding score of each recruitment dimension and the weight of each recruitment dimension when it is in the target function under the target industry to obtain the predicted total score of the fresh graduate resume data.

[0111] S203: sorting the multiple fresh graduate resume data according to the predicted total scores of the respective fresh graduate resume data, and pushing the sorted multiple fresh graduate resume data to the target user.

[0112] Specifically, fresh graduate resumes with higher predicted total scores will be recommended first, helping HR to quickly screen out the candidates that best fit the target recruitment position. In S203, multiple fresh graduate resume data can be sorted in descending order of predicted total scores, and the sorted multiple fresh graduate resume data can be pushed to the target user. For example, the target user can be the target HR.

[0113] The embodiment of the present application provides a method for screening fresh graduate resumes based on a large language model. On the one hand, compared with traditional resume screening methods, the present application uses a large language model to deeply analyze the content of fresh graduate resumes, and can automatically extract key information that matches the recruitment position from fresh graduate resumes, understand the potential and ability of fresh graduates, and improve the accuracy of resume scoring; on the other hand, compared with resumes of working people, the present application selects more detailed and more suitable multiple different recruitment dimensions (such as school characteristics, academic background, social activities, academic awards and personal abilities, etc.) for screening fresh graduate resumes based on the characteristics of fresh graduate resumes. It not only focuses on the hard skills of fresh graduates, but also can effectively identify the soft skills of fresh graduates. By scoring and weighted summing multiple different recruitment dimensions, the scoring of fresh graduate resumes can be made more comprehensive and multi-angle, and the differences between similar fresh graduate resumes can be distinguished. Candidates can be evaluated more comprehensively, and the fairness and comprehensiveness of screening can be improved. On the other hand, for different industries and different functions, by combining key recruitment demand data and HR screening preference data, the scoring range and / or weight of multiple recruitment dimensions for different functions in different industries can be adjusted, so that the scoring and screening of fresh graduate resumes can be dynamically matched according to different industries and different functions, making it easier to distinguish fresh graduate resumes suitable for different industries and different functions, or to screen fresh graduate resumes suitable for various industries and various functions, and it can more accurately imitate HR's screening ideas, so that the scoring and screening of fresh graduate resumes are more in line with HR's real needs and improve user experience. In addition, the screening of fresh graduate resumes based on a large language model helps to improve recruitment efficiency and reduce the burden of manual screening.

[0114] The electronic device in the embodiment of the present application may be a user terminal device, a server, other computing devices, or a cloud server. Figure 3 This is a hardware structure diagram of an electronic device of an embodiment of the present application. The electronic device may include a processor 301 and a memory 302 storing computer program instructions. When the processor 301 executes the computer program instructions, it implements the process or function of any of the above-mentioned embodiment methods.

[0115] Specifically, the processor 301 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application. The memory 302 may include a large-capacity memory for data or instructions. For example, the memory 302 may be at least one of the following: a hard disk drive (HDD), a read-only memory (ROM), a random access memory (RAM), a floppy disk drive, a flash memory, an optical disk, a magneto-optical disk, a tape, a universal serial bus (USB) drive, or other physical / tangible memory storage devices. For another example, the memory 302 may include a removable or non-removable (or fixed) medium. For another example, the memory 302 may be inside or outside the integrated gateway disaster recovery device. The memory 302 may be a non-volatile solid-state memory. In other words, the memory 302 generally includes a tangible (non-transitory) computer-readable storage medium (such as a memory device) encoded with computer-executable instructions, and when the software is executed (such as executed by one or more processors), the operations described in the method of the embodiments of the present application may be performed. The processor 301 implements the process or function of any method in the above embodiments by reading and executing the computer program instructions stored in the memory 302 .

[0116] In one example, Figure 3 The electronic device shown may also include a communication interface 303 and a bus 310. Among them, the processor 301, the memory 302, and the communication interface 303 are connected through the bus 310 and complete the communication between each other. The communication interface 303 is mainly used to realize the communication between each module, device, unit and / or device in the embodiment of the present application. The bus 310 includes hardware, software or both, and can couple the components of the online data traffic billing device to each other. For example, the bus may include at least one of the following: an accelerated graphics port (AGP) or other graphics bus, an enhanced industrial standard architecture (EISA) bus, a front-end bus (FSB), a hypertransport (HT) interconnect, an industrial standard architecture (ISA) bus, an infinite bandwidth interconnect, a low pin count (LPC) bus, a memory bus, a micro channel architecture (MCA) bus, a peripheral component interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a serial advanced technology attachment (SATA) bus, a video electronics standard association local (VLB) bus or other suitable buses. The bus 310 may include one or more buses. Although the embodiments of the present application describe or illustrate a specific bus, the embodiments of the present application may consider any suitable bus or interconnection method.

[0117] In combination with the method in the above embodiments, an embodiment of the present application also provides a computer-readable storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the process or function of any method in the above embodiments is implemented.

[0118] In addition, an embodiment of the present application further provides a computer program product, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the process or function of any one of the methods in the above embodiments is implemented.

[0119] The above exemplarily describes the flowcharts and / or block diagrams of the methods, devices, systems and computer program products of the embodiments of the present application, and describes the relevant various aspects. It should be understood that each box or combination thereof in the flowchart and / or block diagram can be implemented by computer program instructions, or by dedicated hardware that performs a specified function or action, or by a combination of dedicated hardware and computer instructions. For example, these computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable data processing device to form a machine that enables these instructions executed by such a processor to enable the implementation of the functions / actions specified in each box or combination thereof in the flowchart and / or block diagram. Such a processor can be a general-purpose processor, a special-purpose processor, a special application processor or a field programmable logic circuit.

[0120] The functional blocks shown in the structured block diagram of the embodiment of the present application can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application-specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc.; when implemented in software, it is a program or code segment used to perform the required task. The program or code segment can be stored in a memory, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier. The code segment can be downloaded via a computer network such as the Internet, an intranet, etc.

[0121] It should be noted that the present application is not limited to the specific configurations and processes described above or shown in the figures. The above is only a specific implementation mode of the present application. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the described system, device, module or unit can refer to the corresponding process in the method embodiment without further description. It should be understood that the scope of protection of the present application is not limited to this. Any technician familiar with the technical field can think of various equivalent modifications or substitutions within the technical scope disclosed in the present application, and these modifications or substitutions should be included in the scope of protection of the present application.

Claims

1. A large language model training method for fresh graduate resume screening, characterized in that: include: Obtain input data for a large language model, wherein the input data includes key recruitment demand data for different functions in multiple industries, HR screening preference data for recruiters for different functions in multiple industries, job description data for recruitment positions for different functions in multiple industries, and resume data for multiple sample fresh graduates; Preprocessing the input data to obtain preprocessed input data; For any j-th function under any i-th industry, the pre-processed input data corresponding to the j-th function under the i-th industry is input into the large language model, and the large language model is used to adjust the score range and / or weight of the preset multiple recruitment dimensions for the j-th function under the i-th industry according to the pre-processed key recruitment demand data and HR screening preference data corresponding to the j-th function under the i-th industry, where i and j are positive integers; wherein the multiple recruitment dimensions at least include school characteristics, academic background, social activities, academic awards and personal abilities; The large language model is also used to extract the first feature data of each recruitment dimension in the preprocessed job description data corresponding to the jth function under the ith industry, and extract the second feature data of each recruitment dimension in the preprocessed sample fresh graduate resume data; match the first feature data and the second feature data of each recruitment dimension; assign corresponding scores to each recruitment dimension according to the matching results of each recruitment dimension, and perform weighted summation according to the scores corresponding to each recruitment dimension and the weights of each recruitment dimension to obtain the predicted total score of the sample fresh graduate resume data; determine the prediction accuracy of the large language model according to the comparison results of the predicted total scores of multiple sample fresh graduate resume data with the predetermined actual total scores; when the prediction accuracy is less than the preset threshold, return to the step of adjusting the score range and / or weight of multiple recruitment dimensions in the jth function under the ith industry until the prediction accuracy is greater than or equal to the preset threshold; After the preprocessed input data corresponding to different functions in multiple industries are input into the large language model, and the prediction accuracy corresponding to each function under each industry is greater than or equal to the preset threshold, a trained large language model is obtained.

2. The method according to claim 1, characterized in that The multiple recruitment dimensions also include at least one of school ranking, professional background, internship experience, project experience, skill certificates, career interest matching and career development potential, and personal abilities include at least one of innovation ability, leadership, foreign language ability, programming ability, data analysis ability, mastery of professional tools, teamwork ability, problem-solving ability and communication ability.

3. The method according to claim 1, characterized in that The key recruitment demand data corresponding to the j-th function under the i-th industry includes n key recruitment dimensions of the j-th function under the i-th industry determined from multiple recruitment dimensions based on expert experience, and the n key recruitment dimensions are ranked according to the importance of each key recruitment dimension determined based on expert experience, and n is a positive integer; The HR screening preference data corresponding to the j-th function under the ith industry includes m key recruitment dimensions of the j-th function under the ith industry determined from multiple recruitment dimensions based on HR historical screening preferences, and the m key recruitment dimensions are sorted according to the importance of each key recruitment dimension determined based on HR historical screening preferences, and m is a positive integer; The large language model is specifically used to adjust the weights of multiple recruitment dimensions in the j-th function under the ith industry to their respective corresponding basic weights according to the n key recruitment dimensions of the j-th function under the ith industry; and to fine-tune the basic weights of at least some of the recruitment dimensions according to the m key recruitment dimensions of the j-th function under the ith industry, where the magnitude of the fine-tuning is smaller than the size of the basic weights before the fine-tuning.

4. The method according to claim 3, characterized in that The HR screening preference data corresponding to the j-th function under the i-th industry is determined based on the historical behavior data of multiple HRs when screening resumes of fresh graduates for the j-th function under the i-th industry.

5. The method according to claim 1, characterized in that Preprocess the input data to obtain preprocessed input data, including: Perform text cleaning on the input data to remove noise information in the input data; Perform information extraction on the input data after text cleaning to extract key fields related to recruitment dimensions in the input data; Convert key fields into structured data in a preset format based on natural language processing.

6. The method according to claim 1, characterized in that The large language model is specifically used to divide the pre-processed sample resume data of multiple fresh graduates into different data sets. For any j-th job in any i-th industry, the large language model is trained multiple times using different data sets to obtain the prediction accuracy of the large language model for each data set. When the prediction accuracy of a preset number of data sets is greater than or equal to a preset threshold, the training process of the large language model of the j-th function under the ith industry is terminated.

7. A method for screening resumes of fresh graduates based on a large language model, characterized in that: include: Based on the large language model training method for fresh graduate resume screening according to any one of claims 1 to 6, a trained large language model is obtained; After preprocessing the target job description data of the target recruitment position and the resume data of multiple fresh graduates, the preprocessed target job description data and the preprocessed resume data of multiple fresh graduates are input into the trained large language model, and the trained large language model is used to determine the target function under the target industry corresponding to the preprocessed target job description data, and calculate the predicted total score of each fresh graduate resume data according to the scoring range and weight of the target function under the target industry of multiple recruitment dimensions; According to the predicted total scores of each fresh graduate resume data, multiple fresh graduate resume data are sorted, and the sorted multiple fresh graduate resume data are pushed to target users.

8. An electronic device, characterized in that: The electronic device comprises: a processor and a memory storing computer program instructions; when the electronic device executes the computer program instructions, the method according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, and when the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer program product, characterized in that It comprises computer program instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 7.

Images