Resume screening method and system applying natural language large model

Abstract

The invention relates to the technical field of computers, in particular to a resume screening method and system applying a natural language large model, and the method comprises the following steps: resume uploading, resume analysis, rigid index scoring and elastic index scoring. The resume screening method and system applying the natural language large model have the advantages that in the resume preliminary screening process, the screening efficiency is greatly improved through the method, the recruitment period is shortened, the human input in the resume screening link is reduced, the recruitment cost is reduced, and the recruitment efficiency is improved. The HR team can have more time to focus on deep evaluation of high-quality candidates; interference of human factors is reduced in the automatic preliminary screening process, and potential talent loss caused by subjective prejudice is effectively avoided.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a resume screening method and system using a natural language large model. Background Art

[0002] Processing resumes is a key task of the human resources department and an important module in the human resources information system. The most traditional way to analyze a resume is for HR to read it manually and draw conclusions based on scenarios and experience. However, when processing a large amount of resume data, manual screening is inefficient and it is impossible to find the most matching resumes in a short time.

[0003] Most current human resource information management systems maintain a resume module. The system will parse the pushed resume file into a corresponding resume data, including age, education background, and work experience. Users can see the resume data display in a fixed format in the system, and realize the conditional screening of simple fields. Compared with resume files with a variety of formats, this will undoubtedly effectively improve the screening efficiency, but users still need to spend a lot of time focusing on the work experience part. Summary of the invention

[0004] The purpose of the present invention is to provide a resume screening method and system using a natural language large model to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a resume screening method using a natural language large model, the method comprising the following steps:

[0006] Upload the resume, and after uploading, the results of the resume file parsing are returned. The user supplements or modifies the parsing results and finally submits the data for scoring;

[0007] Resume parsing: after receiving the resume file uploaded by the user, it is necessary to parse the key field data from the file;

[0008] Rigidity index scoring presents the analysis results in the form of scoring. For rigidity index scoring, there are only two situations: compliance or non-compliance.

[0009] The elasticity index is scored by code logic judgment and natural language big model, which can be optimized and upgraded independently. It uses multiple fine-tuned LoRA layers. The big model has rich evaluation levels for work experience and is highly referenceable.

[0010] Preferably, when uploading a resume, the user chooses whether to modify the default scoring rules and weights of certain indicators, such as age and education level, in the upload interface; except for work experience, the scoring logic of each field remains unchanged as a whole, and points are assigned according to the value range. For example, 0 points are set for those over 45 years old or under 18 years old, 100 points for those between 18 and 25 years old, 90 points for those between 25 and 35 years old, 80 points for those between 35 and 40 years old, and 70 points for those between 40 and 50 years old. In actual use, it is necessary to modify it according to the characteristics of the position.

[0011] Preferably, the specific operations of resume parsing include:

[0012] Use parent-child heading tags to split the resume text as a whole;

[0013] When there is no basic information module title, the first part of the default split is the basic information module;

[0014] After the resume text is split as a whole, age information needs to be extracted in the basic information module, and the "age" keyword needs to be matched. If no match is found, the "date of birth" keyword needs to be matched, and finally the "ID number" keyword needs to be matched. If no match is found, the user will be reminded to supplement.

[0015] After the resume text is split as a whole, the first degree, highest degree and graduation school information are extracted in the education experience module;

[0016] After splitting the resume text as a whole, the project experience module and the work experience module are integrated and used as a work resume for the big model to score and evaluate.

[0017] Preferably, the specific operations of rigidity index scoring include:

[0018] The percentage system is adopted, and the weight ratio of each indicator can be customized;

[0019] Scoring indicators are divided into rigid indicators and flexible indicators. Rigid indicators have only two conditions: satisfied or not satisfied, and correspond to 0 points and 100 points. When the score is 0, the scoring process ends, the overall score is 0, and the resume is directly eliminated.

[0020] Support for re-scoring.

[0021] Preferably, the specific operations of elasticity index scoring include:

[0022] The code logic judgment scoring and the natural language model scoring are independent of each other, and the scoring accuracy and availability are optimized independently;

[0023] The scoring task is executed asynchronously, and the front-end interface will be automatically modified to display the results after the scoring is completed;

[0024] The large model scoring task is called five times, each time using a different LoRA layer, corresponding to the five aspects of work experience evaluation: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability;

[0025] When the big model scores the job relevance, it needs to receive the job type selected when uploading the resume file.

[0026] A resume screening system using a natural language large model, the system comprising:

[0027] The resume upload module returns the parsed results of the resume file after uploading. The user supplements or modifies the parsed results and finally submits the data for scoring;

[0028] The resume parsing module, after receiving the resume file uploaded by the user, needs to parse the key field data from the file;

[0029] The rigid index scoring module presents the analysis results in the form of scores. For rigid index scoring, there are only two situations: compliance or non-compliance.

[0030] The elasticity index scoring module is completed by code logic judgment and natural language big model. It can be optimized and upgraded independently. It uses multiple fine-tuned LoRA layers. The big model has rich evaluation levels for work experience and is highly referenceable.

[0031] Preferably, in the resume upload module, the user can choose whether to modify the default scoring rules and weights of certain indicators, such as age and education level, in the upload interface; the scoring logic of each field except work experience remains unchanged as a whole, and points are assigned according to the value range. For example, 0 points are set for those over 45 years old or under 18 years old, 100 points for those between 18 and 25 years old, 90 points for those between 25 and 35 years old, 80 points for those between 35 and 40 years old, and 70 points for those between 40 and 50 years old. In actual use, it is necessary to modify it according to the characteristics of the position.

[0032] Preferably, the resume parsing module uses parent-child title tags to split the resume text as a whole;

[0033] When there is no basic information module title, the first part of the default split is the basic information module;

[0034] After the resume text is split as a whole, age information needs to be extracted in the basic information module, and the "age" keyword needs to be matched. If no match is found, the "date of birth" keyword needs to be matched, and finally the "ID number" keyword needs to be matched. If no match is found, the user will be reminded to supplement.

[0035] After the resume text is split as a whole, the first degree, highest degree and graduation school information are extracted in the education experience module;

[0036] After splitting the resume text as a whole, the project experience module and the work experience module are integrated and used as a work resume for the big model to score and evaluate.

[0037] Preferably, the rigid indicator scoring module adopts a percentage system, and each indicator can be customized in terms of weight;

[0038] Scoring indicators are divided into rigid indicators and flexible indicators. Rigid indicators have only two conditions: satisfied or not satisfied, and correspond to 0 points and 100 points. When the score is 0, the scoring process ends, the overall score is 0, and the resume is directly eliminated.

[0039] Support for re-scoring.

[0040] Preferably, in the elasticity index scoring module, the code logic judgment scoring and the natural language large model scoring are independent of each other, and the scoring accuracy and availability are independently optimized;

[0041] The scoring task is executed asynchronously, and the front-end interface will be automatically modified to display the results after the scoring is completed;

[0042] The large model scoring task is called five times, each time using a different LoRA layer, corresponding to the five aspects of work experience evaluation: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability;

[0043] When the big model scores the job relevance, it needs to receive the job type selected when uploading the resume file.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The resume screening method and system using a natural language big model proposed in the present invention greatly improve the screening efficiency during the initial screening of resumes, which not only shortens the recruitment cycle, reduces the manpower input in the resume screening link, and reduces the recruitment cost, but also enables the HR team to have more time to focus on the in-depth evaluation of high-quality candidates; the automated initial screening process reduces the interference of human factors and effectively avoids the potential loss of talent due to subjective bias. Both the big model and the code logic are analyzed based on objective data to ensure the fairness and consistency of the screening results; the method adopts a modular design, and the big model and the code logic can be independently optimized and upgraded to meet the recruitment needs of different industries and different positions, bringing more convenience and benefits to the recruitment work. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose and technical solution of the present invention clearly and completely described, and the advantages more clearly understood, the embodiments of the present invention are further described in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, rather than all of the embodiments, and are only used to explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] For example, see Figure 1 The present invention provides a technical solution: a resume screening method using a natural language large model, the method comprising the following steps:

[0049] It is divided into four steps: resume uploading, resume analysis, rigid indicator scoring and flexible indicator scoring.

[0050] In the resume upload step, users are allowed to upload a single or multiple resume files, but they need to select resumes belonging to the same category of positions, such as software development engineer, account manager, etc., so as to guide the subsequent large model to score the work experience in terms of style and focus. After uploading, the results of the resume file parsing will be returned. Users can supplement or modify the parsing results and finally submit the data for scoring. Users can also choose whether to modify the default scoring rules and weights of certain indicators, such as age, education, etc. in the upload interface. Except for work experience, the scoring logic of each field is unchanged as a whole, and most of them are assigned points according to the value range. For example, setting the age to 0 points for those over 45 years old or under 18 years old, 100 points for those between 18 and 25 years old, 90 points for those between 25 and 35 years old, 80 points for those between 35 and 40 years old, and 70 points for those between 40 and 50 years old. Of course, this is just the default rule, and it needs to be modified according to the characteristics of the position in actual use.

[0051] After receiving the resume file uploaded by the user, it is necessary to parse out the key field data from the file, such as age, education, work experience, etc. Here, the method of splitting according to keywords is adopted. Although there are many different resume templates, they can be divided into several sub-modules, such as the basic information module, which includes the applicant's name, phone number and email address; the education experience module can obtain information such as the first degree, highest degree and graduation school. Each sub-module will have a corresponding title, but it is usually different. For example, the title of the education experience module can be "Educational Experience", "Educational Background" or "Learning Experience". The resume content can be split by defining parent-child tags. For example, the title of the education experience module of a resume is "Educational Experience", and the title of the next module is "Internship Experience", then the text part between the two titles can be split out, corresponding to the education experience module. The next step is to extract scoring indicators from the education experience module, such as first degree, highest degree and graduation school.

[0052] After extracting the target fields from the resume file, the scoring phase begins. For rigid indicator scoring, there are only two situations: compliance or noncompliance. For example, for the first degree restriction, if the first degree parsed is a junior college, the resume will be directly eliminated and the score result will be set to 0, without the need to enter the subsequent flexible indicator scoring.

[0053] The elasticity index scoring can be divided into two parts. First, the elasticity index fields other than work experience are scored using code logic, such as the highest degree of education, the university where the student graduated, etc. Finally, the work experience and job type are input into the big model, so that the big model can output a brief evaluation of the work experience and give a score. In order to enable the big model to give a more objective and accurate score and improve its usability, it is necessary to fine-tune the big model to make it more suitable for the scenario of work experience evaluation and scoring. This method uses the LoRA (Low-Rank Adaptation) method to fine-tune the big model, which improves the usability of the scenario without consuming too many resources, including computer hardware resources and human resources required for data preparation. In order to improve the accuracy, the big model is fine-tuned from five aspects: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability, and five LoRA layers are obtained. When scoring the work experience, five scoring operations are performed respectively, and the LoRA layer of the corresponding topic is selected each time, the weight is set, and the total score is calculated and returned.

[0054] Embodiment 2, based on embodiment 1, further includes:

[0055] Parse the resume file and extract the scoring indicator fields; including:

[0056] Use parent-child heading tags to split the resume text as a whole;

[0057] When there is no basic information module title, the first part of the default split is the basic information module;

[0058] After the resume text is split as a whole, age information needs to be extracted in the basic information module, and the "age" keyword needs to be matched. If no match is found, the "date of birth" keyword needs to be matched, and finally the "ID number" keyword needs to be matched. If no match is found, the user will be reminded to supplement.

[0059] After the resume text is split as a whole, it is necessary to extract the first degree, highest degree and graduation school information in the education experience module;

[0060] After splitting the resume text as a whole, the project experience module and the work experience module need to be integrated as a work resume for the big model to score and evaluate.

[0061] The analysis results are presented in the form of scores; including:

[0062] The percentage system is adopted, and the weight ratio of each indicator can be customized;

[0063] Scoring indicators are divided into rigid indicators and flexible indicators. Rigid indicators have only two conditions: satisfied or not satisfied, and correspond to 0 points and 100 points. When the score is 0, the scoring process ends, the overall score is 0, and the resume is directly eliminated.

[0064] Support for re-scoring, including:

[0065] Code logic judgment scoring and natural language large model scoring are independent of each other, and can independently optimize scoring accuracy and availability;

[0066] The scoring task is executed asynchronously, and the front-end interface is actively modified to display the results after the scoring is completed.

[0067] Scoring is done by code logic judgment and natural language big model, which can be optimized and upgraded independently;

[0068] Using multiple fine-tuned LoRA layers, the large model has rich evaluation levels for work experience and is more referenceable;

[0069] Supports customizing the proportion of each scoring indicator and scoring rules (scoring range); including:

[0070] The large model scoring task is called five times, each time using a different LoRA layer, corresponding to the five aspects of work experience evaluation: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability;

[0071] When the big model scores the job relevance, it needs to receive the job type selected when uploading the resume file.

[0072] Embodiment 3, based on embodiment 2, proposes a resume screening system using a natural language large model, the system comprising:

[0073] The resume upload module returns the results of resume file parsing after uploading. The user supplements or modifies the parsing results and finally submits the data for scoring. The user chooses whether to modify the default scoring rules and weights of certain indicators, such as age and education level, on the upload interface. The scoring logic of each field except work experience remains unchanged as a whole, and points are assigned according to the value range. For example, 0 points are set for those over 45 years old or under 18 years old, 100 points for those between 18 and 25 years old, 90 points for those between 25 and 35 years old, 80 points for those between 35 and 40 years old, and 70 points for those between 40 and 50 years old. In actual use, it is necessary to modify it according to the characteristics of the position.

[0074] The resume parsing module, after receiving the resume file uploaded by the user, needs to parse the key field data from the file; use the parent-child title tags to split the resume text as a whole;

[0075] When there is no basic information module title, the first part of the default split is the basic information module;

[0076] After the resume text is split as a whole, age information needs to be extracted in the basic information module, and the "age" keyword needs to be matched. If no match is found, the "date of birth" keyword needs to be matched, and finally the "ID number" keyword needs to be matched. If no match is found, the user will be reminded to supplement.

[0077] After the resume text is split as a whole, the first degree, highest degree and graduation school information are extracted in the education experience module;

[0078] After splitting the resume text as a whole, the project experience module and the work experience module are integrated and used as a work resume for the big model to score and evaluate.

[0079] The rigid indicator scoring module presents the analysis results in the form of scores. For rigid indicator scoring, there are only two situations: compliance or non-compliance. The percentage system is adopted, and the weight ratio of each indicator can be customized.

[0080] Scoring indicators are divided into rigid indicators and flexible indicators. Rigid indicators have only two conditions: satisfied or not satisfied, and correspond to 0 points and 100 points. When the score is 0, the scoring process ends, the overall score is 0, and the resume is directly eliminated.

[0081] Support for re-scoring.

[0082] The elastic index scoring module is completed by code logic judgment and natural language big model, which can be optimized and upgraded independently. It uses multiple fine-tuned LoRA layers. The big model has rich evaluation levels for work experience and is highly referenceable. Code logic judgment scoring and natural language big model scoring are independent of each other, and the scoring accuracy and availability are optimized independently.

[0083] The scoring task is executed asynchronously, and the front-end interface will be automatically modified to display the results after the scoring is completed;

[0084] The large model scoring task is called five times, each time using a different LoRA layer, corresponding to the five aspects of work experience evaluation: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability;

[0085] When the big model scores the job relevance, it needs to receive the job type selected when uploading the resume file.

[0086] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A resume screening method using a natural language large model, characterized in that: The method comprises the following steps: Upload the resume, and after uploading, the results of the resume file parsing are returned. The user supplements or modifies the parsing results and finally submits the data for scoring; Resume parsing: after receiving the resume file uploaded by the user, it is necessary to parse the key field data from the file; Rigidity index scoring presents the analysis results in the form of scoring. For rigidity index scoring, there are only two situations: compliance or non-compliance. The elasticity index is scored by code logic judgment and natural language big model, which can be optimized and upgraded independently. It uses multiple fine-tuned LoRA layers. The big model has rich evaluation levels for work experience and is highly referenceable.

2. The resume screening method using a natural language large model according to claim 1 is characterized in that: When uploading a resume, the user can choose whether to modify the default scoring rules and weights of certain indicators, such as age and education level, on the upload interface. The scoring logic of each field except work experience remains unchanged as a whole, and points are assigned according to the value range. For example, 0 points are set for those over 45 years old or under 18 years old, 100 points for those between 18 and 25 years old, 90 points for those between 25 and 35 years old, 80 points for those between 35 and 40 years old, and 70 points for those between 40 and 50 years old. In actual use, you need to modify it according to the characteristics of the position.

3. The resume screening method using a natural language large model according to claim 1 is characterized in that: The specific operations of resume parsing include: Use parent-child heading tags to split the resume text as a whole; When there is no basic information module title, the first part of the default split is the basic information module; After the resume text is split as a whole, age information needs to be extracted in the basic information module, and the "age" keyword needs to be matched. If no match is found, the "date of birth" keyword needs to be matched, and finally the "ID number" keyword needs to be matched. If no match is found, the user will be reminded to supplement. After the resume text is split as a whole, the first degree, highest degree and graduation school information are extracted in the education experience module; After splitting the resume text as a whole, the project experience module and the work experience module are integrated and used as a work resume for the big model to score and evaluate.

4. The resume screening method using a natural language large model according to claim 1 is characterized in that: The specific operations of rigidity index scoring include: The percentage system is adopted, and the weight ratio of each indicator can be customized; Scoring indicators are divided into rigid indicators and flexible indicators. Rigid indicators have only two conditions: satisfied or not satisfied, and correspond to 0 points and 100 points. When the score is 0, the scoring process ends, the overall score is 0, and the resume is directly eliminated. Support for re-scoring.

5. The resume screening method using a natural language large model according to claim 1 is characterized in that: The specific operations of elasticity index scoring include: The code logic judgment scoring and the natural language model scoring are independent of each other, and the scoring accuracy and availability are optimized independently; The scoring task is executed asynchronously, and the front-end interface will be automatically modified to display the results after the scoring is completed; The large model scoring task is called five times, each time using a different LoRA layer, corresponding to the five aspects of work experience evaluation: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability; When the big model scores the job relevance, it needs to receive the job type selected when uploading the resume file.

6. A resume screening system using a natural language big model according to the resume screening method using a natural language big model according to any one of claims 1 to 5, characterized in that: The system comprises: The resume upload module returns the parsed results of the resume file after uploading. The user supplements or modifies the parsed results and finally submits the data for scoring; The resume parsing module, after receiving the resume file uploaded by the user, needs to parse the key field data from the file; The rigid index scoring module presents the analysis results in the form of scores. For rigid index scoring, there are only two situations: compliance or non-compliance. The elasticity index scoring module is completed by code logic judgment and natural language big model. It can be optimized and upgraded independently. It uses multiple fine-tuned LoRA layers. The big model has rich evaluation levels for work experience and is highly referenceable.

7. The resume screening system using a natural language large model according to claim 6 is characterized by: In the resume upload module, users can choose whether to modify the default scoring rules and weights of certain indicators, such as age and education level, in the upload interface. The scoring logic of each field except work experience remains unchanged as a whole, and points are assigned according to the value range. For example, 0 points are set for those over 45 years old or under 18 years old, 100 points for those between 18 and 25 years old, 90 points for those between 25 and 35 years old, 80 points for those between 35 and 40 years old, and 70 points for those between 40 and 50 years old. In actual use, it is necessary to modify it according to the characteristics of the position.

8. The resume screening system using a natural language large model according to claim 6 is characterized by: Resume parsing module, which uses parent-child title tags to split the resume text as a whole; When there is no basic information module title, the first part of the default split is the basic information module; After the resume text is split as a whole, age information needs to be extracted in the basic information module, and the "age" keyword needs to be matched. If no match is found, the "date of birth" keyword needs to be matched, and finally the "ID number" keyword needs to be matched. If no match is found, the user will be reminded to supplement. After the resume text is split as a whole, the first degree, highest degree and graduation school information are extracted in the education experience module; After splitting the resume text as a whole, the project experience module and the work experience module are integrated and used as a work resume for the big model to score and evaluate.

9. The resume screening system using a natural language large model according to claim 6 is characterized by: The rigid indicator scoring module adopts a percentage system, and each indicator can customize its weight ratio; Scoring indicators are divided into rigid indicators and flexible indicators. Rigid indicators have only two conditions: satisfied or not satisfied, and correspond to 0 points and 100 points. When the score is 0, the scoring process ends, the overall score is 0, and the resume is directly eliminated. Support for re-scoring.

10. The resume screening system using a natural language large model according to claim 6, characterized in that: In the flexible indicator scoring module, the code logic judgment scoring and the natural language large model scoring are independent of each other, and the scoring accuracy and availability are independently optimized; The scoring task is executed asynchronously, and the front-end interface will be automatically modified to display the results after the scoring is completed; The large model scoring task is called five times, each time using a different LoRA layer, corresponding to the five aspects of work experience evaluation: position relevance, work experience richness, responsibilities and achievements, skills and expertise, and learning ability and adaptability; When the big model scores the job relevance, it needs to receive the job type selected when uploading the resume file.

Images