Human resource master data management entry method based on multi-modal large language model

Abstract

The invention relates to a human resource master data management input method based on a multi-mode large language model. The method comprises the steps that text conversion and data cleaning processing are conducted on an obtained PDF text through the optical character recognition (OCR) technology; performing category division on the acquired text data by using a four-color prototype method so as to create a knowledge domain model for subsequent analysis processing; performing information extraction processing according to information category definition logic, and constructing and deploying an open source large language model; performing compliance verification on the acquired data according to a preset requirement; analyzing the data meeting the requirements, and integrating the main data into a unified table to serve as a data warehouse for storage and analysis; performing data security management by adopting an authentication encryption technology; and the data service platform is used for human resource professionals to access and process and analyze the main data. By adopting the method provided by the invention, the optimal practice of data governance is carried out in combination with knowledge modeling, and efficient, accurate and safe management of the human resource main data is realized.

Description

Technical Field

[0001] The present invention relates to the field of big data processing, in particular to the field of data management and entry, and specifically refers to a human resources master data management and entry method based on a multimodal large language model. Background Art

[0002] In the context of the big data era, the concept and content of human resources master data entry management technology are constantly evolving. Human Resources Master Data Management refers to the centralized, standardized and systematic management of human resources-related master data in an enterprise. These master data include employee information, organizational structure, job information, etc. They are indispensable key information in enterprise operations, have high business value and are reused in multiple business systems. The role of human resources master data is mainly reflected in the following aspects: First, it is the basis for cross-departmental collaboration and meets the needs of different departments for consistent and accurate data; second, master data entry management helps to eliminate data redundancy and improve data processing efficiency; in addition, through unified data standards, it can improve the company's strategic synergy and provide support for corporate decision-making.

[0003] The latest progress in human resources master data entry management includes the use of OCR technology to automatically identify and extract text from images, and the use of large models for deeper data analysis and understanding. Among them, CN202311734269.8 discloses a large model fine-tuning model that can generate corresponding instructions based on the natural language provided by the user, and the development of corresponding operation instructions technology has improved the efficiency and accuracy of data collection; CN202311331376.6 discloses a method of using hobby information tags to calculate scores to assist in screening high-quality resumes and low-quality resumes in the initial screening resume library. Then, since the existing technologies are mainly based on historically processed data set examples, manual master data governance is still required in different steps.

[0004] Due to the complexity and diversity of human resource master data, resume data lacks unified planning due to the independence of needs during construction, resulting in inconsistent data rules, poor quality and integrity, and difficulty in resumé data interaction. Current technical methods lack systematic knowledge modeling to understand the information in human resource master data. OCR technology is usually based on specific fonts and can only extract text. It cannot recognize some encrypted texts, and the recognition accuracy of special texts may be affected. Using information tags to calculate scores can only assist in screening high-quality and low-quality resumes in the resume library. In different processing steps, manual intervention is still required for master data governance. Label-based score calculation cannot fully reflect the needs of master data management and cannot be fully automated. It still cannot completely replace manual meticulous review and decision-making, especially when dealing with complex data management and analysis needs. Therefore, in order to improve the accuracy and efficiency of resume screening, it is necessary to further optimize the algorithm so that it can understand and process master data more comprehensively, while reducing manual intervention and achieving a higher degree of automation. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a human resources master data management and entry method based on a multimodal large language model.

[0006] In order to achieve the above purpose, the human resources master data management and input method based on the multimodal large language model of the present invention is as follows:

[0007] The main feature of the human resources master data management and input method based on the multimodal large language model is that the method comprises the following steps:

[0008] (1) Use optical character recognition (OCR) technology to convert the acquired PDF text and perform data cleaning;

[0009] (2) Use the four-color primitive method to classify the acquired text data into categories, thereby creating a knowledge domain model for subsequent analysis and processing;

[0010] (3) Use information category definition logic to extract information and build and deploy an open source large language model;

[0011] (4) Verify the compliance of the acquired data according to preset requirements;

[0012] (5) Analyze the data that meets the requirements in step (4) and integrate the master data into a unified table as a data warehouse for storage and analysis;

[0013] (6) Use authentication and encryption technology to manage data security to ensure data security;

[0014] (7) Use the data service platform to enable human resources professionals to access, process and analyze master data.

[0015] Preferably, the step (1) specifically comprises the following steps:

[0016] (1.1) Use the PyPDF2 library to read the PDF text content and determine whether the corresponding text is successfully converted into an editable txt text format. If so, go directly to step (1.4), otherwise, go to step (1.2);

[0017] (1.2) Use the pdf2image library to render the PDF page into an image format, and then use optical character recognition technology (OCR) to extract resume information text from the current image;

[0018] (1.3) Use the TesseractOCR engine to recognize text in the image and convert it into a machine-readable resume information text format;

[0019] (1.4) Perform resume information text cleaning to remove text content that does not meet the requirements in the currently acquired resume information text;

[0020] (1.5) Format the text data after text cleaning into a structured format to facilitate subsequent data processing and analysis.

[0021] Preferably, the step (2) is specifically as follows:

[0022] The resume text information extracted by optical character recognition technology (OCR) is classified according to educational background, work experience, resume author and project experience using the four-color prototype method, and a structured knowledge domain model is created based on the classification results to complete the corresponding data storage and processing.

[0023] Preferably, the step (3) specifically comprises the following steps:

[0024] (3.1) According to the classification results obtained in step (2), information is extracted using information category definition logic, and an open source large language model GLM-4-9b and an open source multimodal large language model GLM-4V are constructed and deployed;

[0025] (3.2) Integrate the extracted information into data tables of corresponding categories;

[0026] (3.3) Integrate the currently extracted and classified information into a unified data warehouse for further analysis and processing.

[0027] Preferably, the information category definition logic includes using NLP tools / prompt engineering to extract the following information:

[0028] Personal information: including name, gender, age;

[0029] Educational background: including school name, education, major, and study time;

[0030] Skill information: including language proficiency and other professional skills.

[0031] Preferably, the step (4) is specifically as follows:

[0032] Use the regular expression algorithm of the re tool library to verify the compliance of the obtained personal information, including email address, mobile phone number and date of birth, to determine whether the corresponding string conforms to the regular expression rules and output the data in JSON format.

[0033] Preferably, the step (5) specifically comprises the following steps:

[0034] (5.1) Parse the obtained JSON string format data into Python dictionary format so as to access each field separately;

[0035] (5.2) Create a new blank CSV file using Python’s csv module;

[0036] (5.3) Write all the currently acquired field names into the header of the CSV file;

[0037] (5.4) Extract the field values ​​obtained in step (5.1) from the JSON dictionary data and write them into the CSV file in the order of the CSV header;

[0038] (5.5) Integrate the scattered data into a unified table and save it as a CSV file for storage and further analysis as a data warehouse.

[0039] Preferably, the step (6) is specifically as follows:

[0040] Deploy the data application platform interface and use authentication encryption technology to generate a unique key X-key to verify the identity of each requester, where the key X-key is a pass to access the API to monitor data security.

[0041] Preferably, the step (7) is specifically as follows:

[0042] By providing a data service interface, human resources professionals and other stakeholders are allowed to access and analyze the personal data of resume providers under the premise of possessing the key X-key.

[0043] The human resources master data management entry method based on the multimodal large language model of the present invention solves the data integration problem existing in the existing human resources master data entry management, solves the limitations of optical character recognition (OCR) and large model technology, the complexity of data governance and the continuous operation of human resources master data management, not only improves the efficiency and accuracy of data entry, but also optimizes the data governance process, and ensures the consistency of system functions with business needs, in order to promote the digital transformation of human resources master management, solve the problem of master data in the context of big data, achieve the purpose of fine and flexible management, efficient and full use of data, improve the flexibility and adaptability of master data management, can meet the master data management needs of multiple industries and fields, and effectively enhance the value of data assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The present invention is a flowchart of the human resources master data management and input method based on the multimodal large language model.

[0045] Figure 2 The present invention provides a flowchart of master data knowledge modeling for the human resources master data management and entry method based on a multimodal large language model.

[0046] Figure 3 It is a schematic diagram of a specific embodiment of the human resources master data management and entry method based on a multimodal large language model of the present invention.

[0047] Figure 4 The figure is a schematic diagram of data integration of the human resources master data management and input method based on the multimodal large language model of the present invention. DETAILED DESCRIPTION

[0048] In order to more clearly describe the technical content of the present invention, further description is given below in conjunction with specific embodiments.

[0049] Before describing in detail embodiments according to the present invention, it should be noted that, hereinafter, the terms "comprises", "includes" or any other variations are intended to cover non-exclusive inclusion, whereby a process, method, article or apparatus comprising a series of elements includes not only these elements, but also other elements not explicitly listed or inherent to such process, method, article or apparatus.

[0050] See also Figures 1 to 4 As shown, the human resources master data management and input method based on the multimodal large language model is developed based on the Python language environment, using commercially available Python open source tools and a large language model, wherein the method comprises the following steps:

[0051] (1) Use optical character recognition (OCR) technology to convert the acquired PDF text and perform data cleaning;

[0052] (2) Use the four-color primitive method to classify the acquired text data into categories, thereby creating a knowledge domain model for subsequent analysis and processing;

[0053] (3) Use information category definition logic to extract information and build and deploy an open source large language model;

[0054] (4) Verify the compliance of the acquired data according to preset requirements;

[0055] (5) Analyze the data that meets the requirements in step (4) and integrate the master data into a unified table as a data warehouse for storage and analysis;

[0056] (6) Use authentication and encryption technology to manage data security to ensure data security;

[0057] (7) Use the data service platform to enable human resources professionals to access, process and analyze master data.

[0058] As a preferred embodiment of the present invention, the step (1) specifically comprises the following steps:

[0059] (1.1) Use the PyPDF2 library to read the PDF text content and determine whether the corresponding text is successfully converted into an editable txt text format. If so, go directly to step (1.4), otherwise, go to step (1.2);

[0060] (1.2) Use the pdf2image library to render the PDF page into an image format, and then use optical character recognition technology (OCR) to extract resume information text from the current image;

[0061] (1.3) Use the TesseractOCR engine to recognize text in the image and convert it into a machine-readable resume information text format;

[0062] (1.4) Perform resume information text cleaning to remove text content that does not meet the requirements in the currently acquired resume information text;

[0063] (1.5) Format the text data after text cleaning into a structured format to facilitate subsequent data processing and analysis.

[0064] As a preferred embodiment of the present invention, the step (2) is specifically as follows:

[0065] The resume text information extracted by optical character recognition technology (OCR) is classified according to educational background, work experience, resume author and project experience using the four-color prototype method, and a structured knowledge domain model is created based on the classification results to complete the corresponding data storage and processing.

[0066] As a preferred embodiment of the present invention, the step (3) specifically comprises the following steps:

[0067] (3.1) According to the classification results obtained in step (2), information is extracted using information category definition logic, and an open source large language model GLM-4-9b and an open source multimodal large language model GLM-4V are constructed and deployed;

[0068] (3.2) Integrate the extracted information into data tables of corresponding categories;

[0069] (3.3) Integrate the currently extracted and classified information into a unified data warehouse for further analysis and processing.

[0070] In practical applications, the data processing layer is the core of this solution. This layer includes optical character recognition (OCR) and multimodal large language model technologies to process images and scanned documents, as well as data cleaning and standardization technologies to ensure data quality. Multimodal large language models are used to process and analyze preprocessed data. The preprocessed data is subject to complex data verification and conversion tasks performed by the knowledge model to ensure data consistency and accuracy. The data application layer provides the processed data to various resource applications and is also responsible for monitoring data usage and optimizing data services. In the data processing stage, this solution uses multimodal large language models to integrate and analyze data from different sources. Data integration into a unified format is a key task in this stage. Multimodal large models can integrate data from different systems and formats into a unified data warehouse to ensure data uniqueness and consistency. Multimodal large models use machine learning algorithms to identify and correct errors and inconsistencies in data. For example, the model can identify logical errors in employee information (such as date of birth before the company's founding date) or format errors (such as missing numbers in phone numbers).

[0071] As a preferred embodiment of the present invention, the information category definition logic includes using NLP tools / prompt engineering to extract the following information:

[0072] Personal information: including name, gender, age;

[0073] Educational background: including school name, education, major, and study time;

[0074] Skill information: including language proficiency and other professional skills.

[0075] As a preferred embodiment of the present invention, the step (4) is specifically:

[0076] Use the regular expression algorithm of the re tool library to verify the compliance of the obtained personal information, including email address, mobile phone number and date of birth, to determine whether the corresponding string conforms to the regular expression rules and output the data in JSON format.

[0077] As a preferred embodiment of the present invention, the step (5) specifically comprises the following steps:

[0078] (5.1) Parse the obtained JSON string format data into Python dictionary format so as to access each field separately;

[0079] (5.2) Create a new blank CSV file using Python’s csv module;

[0080] (5.3) Write all the currently acquired field names into the header of the CSV file;

[0081] (5.4) Extract the field values ​​obtained in step (5.1) from the JSON dictionary data and write them into the CSV file in the order of the CSV header;

[0082] (5.5) Integrate the scattered data into a unified table and save it as a CSV file for storage and further analysis as a data warehouse.

[0083] As a preferred embodiment of the present invention, the step (6) is specifically as follows:

[0084] Deploy the data application platform interface and use authentication encryption technology to generate a unique key X-key to verify the identity of each requester, where the key X-key is a pass to access the API to monitor data security.

[0085] As a preferred embodiment of the present invention, the step (7) is specifically as follows:

[0086] By providing a data service interface, human resources professionals and other stakeholders are allowed to access and analyze the personal data of resume providers under the premise of possessing the key X-key.

[0087] The following will further explain in detail the various processing steps of this technical solution:

[0088] Step 1: Deploy Optical Character Recognition (OCR) technology using PyPDF2 and pdf2image libraries as data preprocessing tools to process the collected data and convert pdf format into txt and jpg.

[0089] PDF file reading uses the PyPDF2 library. First, read the content of the PDF file and convert the text in the PDF into an editable text format (txt), which is very important for subsequent text analysis and processing. Image conversion: Use the pdf2image library to convert PDF pages into images. This step usually involves rendering the PDF page into an image format (such as JPEG), so that OCR technology can be used to extract text from the image. OCR processing: Apply OCR technology to the converted image file to recognize the text in the image. Using the Tesseract OCR engine, you can recognize the text in the image and convert it into a machine-readable text format. Text cleaning: The text obtained from the OCR processing may contain errors or formatting problems. Therefore, text cleaning is required, including removing extra spaces, line breaks, correcting recognition errors, etc. Format the cleaned text data into a structured format, such as JSON or CSV, for subsequent data processing and analysis.

[0090] Step 2: Master data knowledge modeling, such as Figure 2 As shown in the figure, the knowledge domain model is created using the four-color primitives method. The four-color primitives method is a method for classifying and organizing information to help understand and process complex data sets. In the context of human resources master data management, this method can be used to classify resume information into different categories for better management and analysis.

[0091] The following is a detailed description of how to model based on the data in step 1. The Four Color Primitives is a classification technique that divides information into four categories, each represented by a color.

[0092] Based on the text data extracted by OCR technology in step 1, the resume information is classified into these four categories.

[0093] Blue: Educational background (school name, major, degree).

[0094] Green: Work experience (company name, role).

[0095] Yellow: Author of the resume (name, contact information).

[0096] Red: Project experience (project name, description).

[0097] Using the four color primitives, create a knowledge domain model that organizes resume information into the following sections:

[0098] ·personal information

[0099] cv_id: unique identifier, such as cv_id.

[0100] Name: Such as name.

[0101] Year and month of birth: such as birth_year and birth_month.

[0102] Phone and email: such as phone and email.

[0103] Work starting year: such as work_starting_year.

[0104] Skill Information

[0105] Resume ID: associated with personal information.

[0106] Skill information: such as skill_info.

[0107] Educational background

[0108] Resume ID: associated with personal information.

[0109] School name: such as school_name.

[0110] Starting and ending years: such as start_year and end_year.

[0111] Major and degree: such as major and degree.

[0112] Work experience

[0113] Resume ID: associated with personal information.

[0114] Company name: such as company_name.

[0115] Starting and ending years: such as start_year and end_year.

[0116] Job role: such as role.

[0117] Project experience

[0118] Resume ID: associated with personal information.

[0119] Project name: such as project_name.

[0120] Starting and ending years: such as start_year and end_year.

[0121] Project description: such as work_or_project_descriptions.

[0122] 1.Person(cv_id,name,birth_year,birth_month,phone,email,work_starting_year)

[0123] 2.Skill(cv_id,skill_info)

[0124] 3.Education(cv_id,school_name,start_year,start_month,end_year,end_month,major,degree)

[0125] 4.Work(cv_id,company_name,start_year,end_year,role)

[0126] 5.Project(cv_id,project_name,start_year,end_year,work_or_project_descriptions)

[0127] Step 2-1: Use OCR technology to extract text and image data from PDF documents.

[0128] Step 2-2: Classify the extracted data into four categories of the four color primitive method.

[0129] Step 2-3: Based on the classification results, create a structured data model, including personal information, skill information, educational background, work experience, and project experience as described above.

[0130] Step 2-4: Store these structured data in a database for further analysis and processing.

[0131] This method can effectively organize and process large amounts of resume data, making it easier to manage

[0132] The master data resume information is divided into five categories: personal information, skill information, learning experience, work experience and project experience.

[0133] Step 3: Based on the table information in step 2, define the logic of information extraction rules with information categories, and use NVIDIA's H800 graphics processor to deploy the encoder based on the Transformer model in the CUDA 12.4 environment to extract text features. Build and deploy the open source large language model GLM-4-9b and the open source multimodal large language model GLM-4V to enable them to process and analyze the human resources master data in step 1 and classify it into detailed information in step 3.

[0134] Personal information: name, gender, age.

[0135] Educational background: school name, academic qualifications, major, and study time.

[0136] Skills: Language skills and other professional skills.

[0137] Use NLP tools / prompt engineering to identify school name, academic qualifications, major, and study time.

[0138] Identify skill-related keywords, such as "language ability", "College English Test Band 4", etc.

[0139] Data integration uses the Transformer self-attention mechanism to capture complex relationships in the text, automatically identify and extract personal information such as name, gender, and age, and fully utilize the parallel computing capabilities of the graphics processor to integrate the extracted information into the corresponding table.

[0140] Step 3: Classify the information extracted by the large model into the corresponding tables in step 2, such as personal information table, education background table, skill information table, etc. Integrate all the extracted and classified information into a unified data warehouse for further analysis and processing.

[0141] Through this process, big models are used to automatically process and analyze human resources master data, and the information is effectively classified into structured tables, thereby improving the efficiency and accuracy of data processing.

[0142] Step 4: Implement data quality assessment measures. The RE tool library regular algorithm is used to control the model output JSON format and verify that the data meets the system compliance. A series of personal information is extracted from the human resources master data, and it is necessary to ensure that this information meets specific compliance requirements in both format and content. The goal is to verify the following information:

[0143] The email address must contain the @ symbol and the suffix must be a valid domain name.

[0144] Phone numbers must begin with a country code, such as +86, followed by 11 digits.

[0145] The date must strictly follow the format of YYYY-MM-DD.

[0146] Writing regular expressions

[0147] Email: ^[a-zA-Z0-9_.+-]+@[a-zA-Z0-9-]+\.[a-zA-Z0-9-.]+$

[0148] Phone number: ^\+86\d{11}$

[0149] Date: ^\d{4}-\d{2}-\d{2}$ ...

[0151] Python uses the re module to verify whether a string conforms to the rules of a regular expression.

[0152] Step 5: Parse the data from step 4 and save it in CSV format, and integrate the master data into a unified table as a data warehouse. In the process of data integration and conversion, start with the JSON format data obtained in step 4. This data has passed compliance verification and contains detailed information such as personal information, skills, education background, work experience, and project experience. Integrate these scattered data into a unified table and save it in CSV format for storage and further analysis as a data warehouse.

[0153] Data integration and transformation steps

[0154] Step 5-1 Parse JSON data: Parse the JSON string format data into Python dictionary format so that each field can be accessed separately.

[0155] Step 5-2 Create CSV file: Create a new blank CSV file using Python’s csv module.

[0156] Step 5-3 Write CSV header: Write the header of the CSV file, including the names of all fields, such as cv_id, name, birth_year, skill_info, school_name, company_name, project_name, etc.

[0157] Step 5-4 Data conversion: Extract the values ​​of the fields in step 5-1 from the JSON dictionary data and write them into the CSV file in the order of the CSV header.

[0158] Step 5-5 Save CSV and name the file: Save the integrated data as a CSV file and name it Resume_data.csv.

[0159] Step 6: Data security management includes protecting data from unauthorized access and leakage. Deploy the data application platform interface (API) and use the pycryptodome library of Python cryptography to implement zero-knowledge proof authentication encryption technology by generating a unique key (X-key) to verify the identity of each requester. This key is the pass to access the API. Only users with the correct key can interact with the data through the API. Control the API access mechanism to ensure data security.

[0160] Step 6-1 Password generation: The system generates a unique key for each registered user. This key is stored in the server's secure database and is only known to the user.

[0161] Step 6-2 API request: The user initiates a request through a client (such as an application or website), and the request contains the data and key to be queried.

[0162] Step 6-3 Key Verification: After receiving the request, the server first verifies the validity of the key. If the key is invalid or expired, the server will reject the request and return an error message.

[0163] Step 6-4 Data Query: If the key verification is successful, the server will retrieve the corresponding data from the database according to the data query conditions in the request.

[0164] Step 6-5 Response return: The server returns the query result to the user. If the query is successful, the data is returned; if the query fails or no data is found, the corresponding error message is returned.

[0165] Step 7: Use a data services platform such as Postman to enable HR professionals to access and process analytical master data.

[0166] In practice, when a user attempts to access data through an API, they are asked to provide an X-key. This key is used to check the user's permissions to ensure they have access to the requested data. If key verification fails, access is denied, thus preventing unauthorized data access. This protects data from unauthorized access, ensures data integrity and confidentiality, and provides a secure and reliable data environment.

[0167] In the service application phase, the processed data is used to support various HR decisions and operations. This solution provides a data service interface that allows HR professionals and other stakeholders to access and analyze data.

[0168] Through these steps, the technical solution of the present invention provides a comprehensive human resources master data solution, which improves the efficiency, accuracy and compliance of data management. In addition, through continuous monitoring and optimization, the solution can adapt to the ever-changing business needs and technical environment, and provide continuous data management support for enterprises.

[0169] In a specific embodiment of the present invention, please refer to Figure 3 As shown, this section further illustrates the implementation of the technical solution by using an automated processing method for one resume in a resume database of approximately 300,000 resumes of different formats (pdf).

[0170] First, you need to convert the document into a text character file (txt / docx).

[0171] Step 1: Convert the PDF document into a picture jpg using OCR / multimodal large model and give a text txt;

[0172] Input such as Figure 3 The image information shown;

[0173] Output:

[0174] "content":"Personal strengths: Steady work, comprehensive consideration, easier to concentrate on work, good at listening to others' ideas and respecting others' opinions. Willing to explore new things, curious about unknown areas, able to quickly master work in different fields. Optimistic and persistent in work, dare to face difficulties and challenges; have good business ability and psychological quality, good communication and stress resistance. Zhou Can Educational experience Shanghai Zhongqiao Vocational and Technical University College of Applied English 2021-2024 Female | 20 years old | Non-party member Professional skills Contact information Language ability: Passed College English Test Band 4, Mandarin level test"

[0175] "height":4097

[0176] }

[0177] Step 2: Process the output of step 1 and classify the information according to the following logic

[0178] ·personal information

[0179] Resume ID: cv_id

[0180] Name: name

[0181] Year of birth: birth_year

[0182] Birth month: birth_month

[0183] Phone: phone

[0184] Email: email

[0185] Work starting year: work_starting_year (earliest year of work or project participation)

[0186] Work skills

[0187] Resume ID: cv_id

[0188] Skill information: skill_info (using Markdown format)

[0189] Educational background

[0190] Resume ID: cv_id

[0191] School name: school_name

[0192] Start year: start_year

[0193] Start month: start_month

[0194] End year: end_year

[0195] End month: end_month

[0196] Major: major

[0197] Degree: degree

[0198] Work experience

[0199] Resume ID: cv_id

[0200] Company Name: company_name

[0201] Start year: start_year

[0202] End year: end_year

[0203] Job role: role (describe job responsibilities in a few words)

[0204] Project experience

[0205] Resume ID: cv_id

[0206] Project name: project_name

[0207] Start year: start_year

[0208] End year: end_year

[0209] Project description: work_or_project_descriptions (description of the work or project) ...

[0211] Step 3:

[0212] "output":"{\n\"Person\":{\n \"cv_id\":\"630704\",\n \"name\":\"Can Zhou\",\n\"birth_year\":\"2003\",\n \"birth_month\":\"02\",\n \"phone\":\"181...908\",\n\"email\":\"\",\n \"work_starting_year\":\"2021\"\n},\n\"Skill\":{\n \"cv_id\":\"630704\",\n\"skill_info\":\"Passed CET-4 and Putonghua Proficiency Test.\\nComputer Skills: Passed the Shanghai Computer Proficiency Test and proficient in daily office software such as word, excel, and PPT.\"\n},\n\"Education\":{\n \"cv_id\":\"630704\",\n \"school_name\":\"Shanghai Zhongqiao Vocational and Technical University\",\n\"start_year\":\"2021\",\n \"start_month\":\"02\",\n \"end_year\":\"2024\",\n\"end_month\":\"02\",\n\"major\":\"Applied English\",\n\"degree\":\"Associate Degree\"\n},\n\"Work\":{\n\"cv_id\":\"630704\",\n\"company_name\":\"(Specific company name not provided)\",\n\"start_year\":\"2021\",\n\"end_year\":\"(Specific end year not provided)\",\n\"role\":\"(Specific job role not provided)\"\n},\n\"Project\":{\n\"cv_id\":\"630704\",\n\"project_name\":\"(Specific project name not provided)\",\n\"start_year\":\"(Specific start year not provided)\",\n\"end_year\":\"(Specific end year not provided)\",\n\"work_or_project_descriptions\":\"(Specific project description not provided)\"\n}\n}"

[0213] Step 4: As Figure 4 shown. Integrate into the database table;

[0214] Step 5: API monitoring interface and encryption processing deployment service. For deployment, you only need to access the server in an encrypted manner and upload the API application development interface (Application Programming Interface).

[0215] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0216] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device.

[0217] A person of ordinary skill in the art may understand that all or part of the steps of the method for implementing the above-mentioned embodiment may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one of the steps of the method embodiment or a combination thereof.

[0218] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0219] In the description of this specification, the description with reference to the terms "an embodiment", "some embodiments", "example", "specific example", or "embodiment" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0220] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

[0221] The human resources master data management entry method based on the multimodal large language model of the present invention solves the data integration problem existing in the existing human resources master data entry management, solves the limitations of optical character recognition (OCR) and large model technology, the complexity of data governance and the continuous operation of human resources master data management, not only improves the efficiency and accuracy of data entry, but also optimizes the data governance process, and ensures the consistency of system functions with business needs, in order to promote the digital transformation of human resources master management, solve the problem of master data in the context of big data, achieve the purpose of fine and flexible management, efficient and full use of data, improve the flexibility and adaptability of master data management, can meet the master data management needs of multiple industries and fields, and effectively enhance the value of data assets.

[0222] In this specification, the present invention has been described with reference to specific embodiments thereof. However, it is apparent that various modifications and variations may be made without departing from the spirit and scope of the present invention. Therefore, the specification and drawings should be regarded as illustrative rather than restrictive.

Claims

1. A method for human resources master data management and input based on a multimodal large language model, characterized in that: The method comprises the following steps: (1) Use optical character recognition (OCR) technology to convert the acquired PDF text and perform data cleaning; (2) Use the four-color primitive method to classify the acquired text data into categories, thereby creating a knowledge domain model for subsequent analysis and processing; (3) Use information category definition logic to extract information and build and deploy an open source large language model; (4) Verify the compliance of the acquired data according to preset requirements; (5) Analyze the data that meets the requirements in step (4) and integrate the master data into a unified table as a data warehouse for storage and analysis; (6) Use authentication and encryption technology to manage data security to ensure data security; (7) Use the data service platform to enable human resources professionals to access, process and analyze master data.

2. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 1, characterized in that: The step (1) specifically comprises the following steps: (1.1) Use the PyPDF2 library to read the PDF text content and determine whether the corresponding text is successfully converted into an editable txt text format. If so, go directly to step (1.4), otherwise, go to step (1.2); (1.2) Use the pdf2image library to render the PDF page into an image format, and then use optical character recognition technology (OCR) to extract resume information text from the current image; (1.3) Use the TesseractOCR engine to recognize text in the image and convert it into a machine-readable resume information text format; (1.4) Perform resume information text cleaning to remove text content that does not meet the requirements in the currently acquired resume information text; (1.5) Format the text data after text cleaning into a structured format to facilitate subsequent data processing and analysis.

3. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 2, characterized in that: The step (2) is specifically as follows: The resume text information extracted by optical character recognition technology (OCR) is classified according to educational background, work experience, resume author and project experience using the four-color prototype method, and a structured knowledge domain model is created based on the classification results to complete the corresponding data storage and processing.

4. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 3 is characterized in that: The step (3) specifically comprises the following steps: (3.1) According to the classification results obtained in step (2), information is extracted using information category definition logic, and an open source large language model GLM-4-9b and an open source multimodal large language model GLM-4V are constructed and deployed; (3.2) Integrate the extracted information into data tables of corresponding categories; (3.3) Integrate the currently extracted and classified information into a unified data warehouse for further analysis and processing.

5. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 4 is characterized in that: The information category definition logic includes the following information extraction using NLP tools / prompt engineering: Personal information: including name, gender, age; Educational background: including school name, education, major, and study time; Skill information: including language proficiency and other professional skills.

6. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 5 is characterized in that: The step (4) is specifically as follows: Use the re tool library regular algorithm to verify the compliance of the obtained personal information, including email address, mobile phone number and date of birth, to determine whether the corresponding string conforms to the regular expression rules and output the data in JSON format.

7. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 6 is characterized in that: The step (5) specifically comprises the following steps: (5.1) Parse the obtained JSON string format data into Python dictionary format so as to access each field separately; (5.2) Create a new blank CSV file using Python’s csv module; (5.3) Write all the currently acquired field names into the header of the CSV file; (5.4) Extract the field values ​​obtained in step (5.1) from the JSON dictionary data and write them into the CSV file in the order of the CSV header; (5.5) Integrate the scattered data into a unified table and save it as a CSV file for storage and further analysis as a data warehouse.

8. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 6, characterized in that: The step (6) is specifically as follows: Deploy the data application platform interface and use authentication encryption technology to generate a unique key X-key to verify the identity of each requester, where the key X-key is a pass to access the API to monitor data security.

9. The method for managing and inputting human resources master data based on a multimodal large language model according to claim 8, characterized in that: The step (7) is specifically as follows: By providing a data service interface, human resources professionals and other stakeholders are allowed to access and analyze the personal data of resume providers under the premise of possessing the key X-key.

Images