Multi-mode human resource analysis method and system based on large model

Abstract

The invention relates to the field of human resources, in particular to a multi-mode human resource analysis system and method based on a large model, and the system comprises a user interaction module which is used for inputting an instruction and displaying an analysis result; the mode selection module is used for providing selection of an analysis mode and transmitting a user instruction to a locally deployed large model inference engine; the data loading module is used for loading staff related information and investigation detailed rule related information from local storage equipment; the large model reasoning module is used for calling a locally deployed large model to generate a comprehensive analysis result according to the employee related information and the investigation detailed rule related information; the design does not need to depend on an external network or cloud service, data privacy and safety are ensured, the large model reasoning speed is remarkably increased, large-scale employee data and complex rule files can be processed, and enterprise-level application requirements can be met.

Description

Technical Field

[0001] The present invention is a multi-modal human resource analysis method and system based on large models, belonging to the field of human resources. Background Art

[0002] In the human resource management of modern enterprises, employee evaluation and decision support are core aspects. Traditional human resource management systems usually rely on manual analysis or simple rule engines, and it is difficult to handle complex assessment details and large-scale employee data. In recent years, with the development of artificial intelligence technology, the natural language processing ability based on large models has gradually been applied to the field of employee evaluation. However, the existing solutions have the following problems: 1. Strong cloud dependence: Most systems need to upload data to the cloud for processing, resulting in difficult-to-guarantee data privacy and security. 2. Slow response speed: Due to dependence on external networks and cloud services, the system is prone to delays or even interruptions under high load or unstable network conditions. 3. Insufficient hardware performance: The computing power of ordinary servers is limited and cannot efficiently run complex large model inference tasks. 4. Lack of multi-modal support: Existing systems usually only support single-mode analysis and cannot meet diverse scenario requirements. Summary of the Invention

[0003] Aiming at the problems in the prior art, the present invention provides a multi-modal human resource analysis method and system based on large models.

[0004] The technical solution adopted by the present invention to solve its technical problems is: A multi-modal human resource analysis system based on large models, the analysis system includes: A user interaction module for inputting instructions and displaying analysis results; A mode selection module for providing the selection of analysis modes and passing user instructions to a locally deployed large model inference engine; A data loading module for loading employee-related information and assessment detail-related information from a local storage device; A large model inference module for calling a locally deployed large model to generate comprehensive analysis results according to employee-related information and assessment detail-related information; A result storage module for saving the analysis results to a storage device.

[0005] Further, the analysis modes include: A full-staff analysis mode for batch analysis of all employees; An individual precise analysis mode for generating a detailed report for a specific employee; A directional analysis mode for in-depth mining and quantitative scoring of unstructured data; A customized analysis mode is used to dynamically adjust the weights of the scoring model according to the specific requirements in the project plan, and provide scientific personnel matching suggestions.

[0006] Further, when the directional analysis mode is selected, the large model inference module combines NLP technology and topic modeling algorithms to extract key information from unstructured text and convert it into structured data, and generates a multi-dimensional ability quantification score in combination with the weight assignment formula; When the customized analysis mode is selected, the large model inference module analyzes the project plan, extracts key requirements, converts them into specific scoring dimensions, and generates a quantification score in combination with the business capabilities and management capabilities of employees.

[0007] Further, the large model inference module includes a prompt construction module, a large model API call module, an inference result reading module, an exception handling and retry module, and a final result return module; The prompt construction module constructs structured prompts for employee-related information and inspection rules-related information by using string concatenation; The large model API call module calls the locally deployed large model API through an HTTP request to complete the inference task; The inference result reading module is used to read data and display the inference result; The exception handling and retry module is used to capture exceptions and adopts a timeout retry mechanism; The final result return module pushes the complete inference result to the result storage module and the user interaction module.

[0008] Further, the system further includes a pre-analysis module; The pre-analysis module is used to start after the set pre-analysis time is reached and perform task processing; The data loading module pre-loads specific data into the high-speed memory for caching before the set pre-analysis time; The mode selection module passes the instruction sent by the pre-analysis module to the large model inference module; The large model inference module caches the inference result of the pre-analysis into the memory.

[0009] A multi-mode human resource analysis method based on a large model includes: S1, receiving a user instruction through the user interaction module; S2, providing a selection of analysis modes through the mode selection module, and passing the user instruction to the locally deployed large model inference engine; S3, loading employee-related information and inspection rules-related information from a local storage device through the data loading module, and performing necessary preprocessing; S4, based on employee-related information and inspection details, calls the locally deployed large model to generate comprehensive analysis results; S5, saving the analysis results to the result storage module; S6, outputting the analysis results to the user interaction module and displaying them in real time.

[0010] Furthermore, the analysis modes include full-staff analysis mode, individual precision analysis mode, targeted analysis mode and customized analysis mode; For the full-staff analysis mode, the data loaded through the data loading module includes the basic information and assessment details of all employees, and the output analysis results include the comprehensive score and assessment result classification of each employee; For the individual precision analysis mode, the data loaded through the data loading module includes basic information and assessment details of each employee. The output analysis results include an in-depth analysis of the employee's professional qualities and development potential, as well as more detailed assessment conclusions and improvement suggestions. For the targeted analysis mode, the data loaded by the data loading module includes employees' unstructured data and assessment details. The output analysis results include quantitative scores, comprehensive scores, and visual presentations of employees' professional skills, execution, and innovation capabilities. For the customized analysis mode, the data loaded through the data loading module includes project plans and employee capability data. The employee capability data includes structured and unstructured data. The output analysis results include employee scores in the dimensions of market development capability and team collaboration capability, recommended candidates, their score distribution, and reasons for recommendation.

[0011] Beneficial effects of the present invention: 1. Local deployment: All data and large models run on local servers without relying on external networks or cloud services, ensuring data privacy and security and avoiding system unavailability caused by network instability or cloud service interruptions.

[0012] 2. High-performance computing: Leveraging the powerful computing power provided by a 96-thread processor and 256GB of RAM, the system significantly improves the inference speed of large models, capable of processing large-scale employee data and complex rule files, meeting enterprise-level application requirements.

[0013] 3. Multi-mode analysis: Supports multiple analysis modes including full-staff analysis, personal precision analysis, targeted analysis, and customized analysis to meet the needs of different scenarios. Users can switch analysis modes through simple commands, which is easy and efficient to operate.

[0014] 4. Streaming output: With the support of high-performance hardware, real-time streaming output is achieved, and analysis results are gradually displayed. The real-time feedback mechanism is particularly suitable for scenarios that require rapid response.

[0015] 5. Enhanced reliability: A timeout retry mechanism is used to ensure the reliability of the inference process, and the low latency characteristics of high-performance hardware significantly improve the system's fault tolerance.

[0016] 6. Long-term storage: Personal precision analysis results are saved to the history file in append mode to avoid overwriting existing data. The append-only saving mechanism ensures the integrity of the historical records and facilitates long-term storage and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments with reference to the following drawings: Figure 1 This is a system architecture diagram of a multi-mode human resources analysis method based on a large model according to the present invention; Figure 2 This is a flow chart of a multi-mode human resources analysis method based on a large model of the present invention.

[0018] Figure 3 This is an overall architecture diagram of a multi-mode human resources analysis method based on a large model in the present invention.

[0019] Figure 4 This is a hardware architecture diagram of a multi-mode human resources analysis method based on a large model in the present invention. DETAILED DESCRIPTION

[0020] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0021] The present invention provides a multi-modal human resource analysis system based on a large model, which aims to achieve efficient, accurate and flexible human resource analysis and provide strong support for corporate decision-making; Example 1: Figure 1-Figure 4 As shown, the system mainly includes a user interaction module, a mode selection module, a data loading module, a large model inference module, and a result storage module; The data loading module is used to load employee information (in CSV or other formats), assessment details (in DOCX or other formats), and project plans from the local NVMe SSD. Working in conjunction with the high-speed NVMe SSD and 256GB of RAM, it enables rapid data loading and caching, ensuring efficient data transfer. Frequently used data is cached in memory to avoid repeated read operations.

[0022] The data loading module uses the Pandas module to process CSV files and extract the basic information of employees. It uses the python_docx module to parse DOCX files and extract the text content of the inspection rules. The data loading module captures exceptions during data loading through the exception handling mechanism of the Python language itself combined with the print function, and displays detailed error messages in the terminal.

[0023] The user interaction module is used to input instructions and display analysis results. As the interface for users to interact with the system, users can input analysis instructions, select analysis modes, view analysis results, etc. through this module. It provides an intuitive graphical interface to facilitate the operation of users with different technical levels. For example, users can input the ID or department information of a specific employee on the interface to initiate a targeted analysis request.

[0024] The mode selection module is used to provide options for analysis modes. Users can input instructions through terminal devices to select different analysis modes. The available analysis modes include the all-staff analysis mode, the individual precise analysis mode, the targeted analysis mode, and the customized analysis mode. These four modes are designed for different demand scenarios respectively, forming a complete functional matrix that covers multi-level requirements from macro decision-making to micro guidance, and from structured data processing to unstructured data mining.

[0025] The all-staff analysis mode focuses on batch processing. It efficiently processes large-scale data sets through the Pandas module and gradually displays analysis results using streaming output, significantly improving the user experience. It can quickly generate comprehensive inspection reports for all employees to meet the needs of efficient decision-making, and is applicable to macro scenarios such as annual or quarterly performance appraisals, large-scale employee promotions or dismissals, and departmental adjustments. The individual precise analysis mode focuses on individuals, providing detailed personalized analysis reports for individual employees to help management formulate targeted career development plans or improvement plans, such as employee career development plans, key cultivation of high-potential employees, or improvement guidance for problem employees, and supports long-term tracking and comparative analysis. The targeted analysis mode introduces NLP technology and topic modeling algorithms to deeply mine the unstructured data of employees, such as project completion reports and superior evaluations. It extracts key information from unstructured text and converts it into structured data, and generates multi-dimensional, comprehensive, and intuitive ability quantification scores and visualization reports in combination with the weight assignment formula, breaking through the limitations of traditional analysis systems and providing a scientific basis for promotion evaluations and job suitability analyses. The customized analysis mode dynamically adjusts the weight of the scoring model according to the specific requirements in the project plan, calls the large model to generate scores and recommendation results, and screens out the employees most suitable for completing the project, supporting users to customize scoring rules and weight assignments. This technology not only improves the efficiency and accuracy of analysis but also enhances the flexibility and adaptability of the system.

[0026] After receiving the user's selected mode command, the mode selection module passes it to the locally deployed large model inference engine. It then filters and organizes the data obtained by the data loading module based on the selected mode, ensuring that the data input to the large model meets the analysis requirements. A user-friendly command-line interface allows users to switch modes by simply entering commands, and color-coded prompts are used to distinguish different types of information.

[0027] The large model reasoning module is used to call the locally deployed large model to generate comprehensive analysis results based on employee-related information and inspection details. For common problems, the large model reasoning module pre-generates templated prompts to reduce reasoning complexity.

[0028] The large model inference module includes a prompt word construction module, a large model API call module, an inference result reading module, an exception handling and retry module, and a final result return module; The prompt word construction module uses string concatenation to construct structured prompt words for employee-related information and inspection details-related information; the large model API call module calls the locally deployed large model API through HTTP requests to complete the reasoning task; the reasoning result reading module is used to use the response.iter_lines() method to read streaming data line by line and gradually display the reasoning results; the exception handling and retry module is used to capture exceptions and adopt a timeout retry mechanism with an exponential backoff strategy. When the large model reasoning module uses an exponential backoff strategy for timeout retry, the upper limit of the number of retries is N times, N is an integer greater than 1, and N can be set to 5. The waiting time for each retry increases by a power of 2. This design can avoid task failures caused by network or hardware failures, and record log information for each reasoning to facilitate subsequent problem investigation. The module that returns the final result pushes the complete reasoning result to the result storage module and the user interaction module. The result storage module is used to save the analysis results to the high-speed NVMe SSD storage. The results are classified and stored according to different analysis modes and user needs. The analysis results of all employees are saved as independent CSV files, including the assessment results of each employee. The individual precision analysis results are saved in the history record file in append mode to facilitate long-term tracking and comparison. The storage module works together through the NVMe SSD and 256GB RAM to ensure the efficiency of data writing. In terms of technical implementation, the Pandas module is used to save the analysis results as UTF-8 encoded CSV files, which supports the correct display of Chinese characters. If the file does not exist, a new file is created and the table header is written; if the file already exists, the data is written in append mode.

[0029] Preferably, the AES algorithm is used to encrypt and store sensitive information. Before writing data to a file, the data is encrypted using a cryptography library. When reading the data, the same key is used for decryption to ensure data security.

[0030] Preferably, use the rsync tool (under Linux system) or the Robocopy tool (under Windows system) to regularly synchronize the analysis results to cloud storage or other external storage devices.

[0031] The user interaction module includes a command input module and a result output module. The result output module outputs the analysis results to the user in various forms. In addition to being displayed intuitively on the interface of the user interaction module, it also supports exporting to files in formats such as PDF and Excel, making it convenient for users to further process and share. When outputting results, key information will be highlighted and visualized, such as supporting color to distinguish different types of information, such as using charts to show the changing trends of employee performance and relationship diagrams of team collaboration, making the analysis results easier to understand and interpret, thereby facilitating management decision-making. In terms of technical implementation, the colorama module is used to implement color output and display to improve the readability of information. When exporting reports, the pandas.DataFrame.to_excel() method is used to generate Excel files, or the reportlab module is used to generate PDF files.

[0032] The result output module provides a concise and clear result display interface. Users can quickly locate target information through paging or search functions, and supports custom output formats (such as displaying only key indicators or complete reports).

[0033] Preferably, the result output module also provides data visualization functions to generate bar charts, pie charts and other charts. Specifically, Python's Matplotlib and Seaborn libraries can be used to generate charts, and the Plotly library can be introduced to implement the interactive function of the charts and intuitively display the analysis results. Language packages can be used to achieve multi-language output support (such as Chinese and English) to meet internationalization needs.

[0034] In terms of hardware configuration, the system hardware configuration is designed to meet the needs of large-scale data processing and large model inference. A 96-thread processor is selected and paired with 256GB ECC memory to provide powerful power for data processing and operations, accelerating the large model inference process. In terms of storage, high-speed nvme SSDs are adopted, and their fast read and write capabilities meet the system's efficient requirements for data storage and retrieval. At the same time, 2 SSDs with U2 interfaces are configured to store the operating system images of each virtual machine; 2 nvme SSDs with a capacity of 2T are deployed in the form of PCIe expansion cards and allocated to the "Ollama service virtual machine" in a PCIe passthrough manner, specifically for storing large model parameters (weights) to facilitate quick loading into the GPU video memory; 3 mechanical hard disks with a capacity of 12T are used to form a RAID5 array for storing general data documents that have low requirements for disk read and write speeds. In terms of graphics processing capabilities, the server is configured with 8 graphics cards, each with 24GB of video memory. These GPUs can be allocated to different virtual machines that require GPU acceleration as needed in a PCIe passthrough manner, providing powerful parallel computing capabilities for large model inference. In addition, the system is also equipped with an efficient heat dissipation module to effectively avoid performance degradation or system failures caused by overheating.

[0035] Regarding the software environment, the system can run on Ubuntu or Windows operating systems, and Python is selected as the programming language. With the help of the Ollama API inference framework and combined with third-party libraries such as Pandas, Requests, and Colorama, efficient data processing, interaction with large models, and user interaction functions are achieved. When using the open-source PVE virtualization platform, the open-source agreement is strictly adhered to to ensure legal use.

[0036] In terms of the system architecture, such as Figure 4As shown in the figure, the human resource analysis system is configured with 1 4U rack-mounted server, and an open-source PVE virtualization platform is installed on the 4U rack-mounted server, which is responsible for managing and allocating hardware resources. Multiple virtual machines are deployed on the virtual machine platform. Among them, the "Ollama service virtual machine" undertakes the storage and operation tasks of the large model, and provides the large model call service for the upper-layer application through the Ollama API interface to achieve decoupling from other services. In the "Ollama service virtual machine", large language models such as qwen, deepseek, and glm are copied to the specified storage location (such as the 2 2T nvme SSDs mentioned above). The Ollama service virtual machine manages and calls these large language models through the Ollama API to provide the large model inference service for the upper-layer application. At the same time, a dedicated knowledge database is built, and data such as human resource-related regulations and personnel information are stored in it. When the data loading module is working, it will extract corresponding data from the dedicated knowledge database according to different analysis task requirements, preprocess it in combination with the data obtained from other data sources, and then pass it to the inference calculation module, so that during the large model inference process, these data can work together with the large language model to provide strong support for the analysis results.

[0037] Embodiment 2: On the basis of Embodiment 1, the multi-mode human resource analysis system based on the large model further includes a pre-analysis module; The pre-analysis module runs in the system background and is used to start when the set pre-analysis time is reached or the inference volume of the inference module is lower than the set threshold, and send instructions to the data loading module, the mode selection module, and the big data inference module; the data loading module is used to pre-load specific data into the high-speed memory for caching before the set pre-analysis time; the mode selection module optimizes the instructions sent by the pre-analysis module and passes them to the large model inference module; the large model inference module caches the inference results of the pre-analysis in the memory.

[0038] (1) The pre-analysis module uses the apscheduler module to set a timed task to make a determined task request at a certain determined time. For example, it can be set to automatically run the full-staff pre-analysis task at 2:00 am every day, or automatically run the full-staff pre-analysis task at 2:00 am on the 1st of each month. The pre-analysis results are stored as CSV files, and when loading, it is preferred to check whether the file exists and whether it needs to be updated. Through this design, it is possible to run the full-staff pre-analysis task during off-peak hours (such as at night), generate the analysis results in advance and store them in the local file. At this time, when the user queries, the pre-analysis results are directly called from the cache without re-inference, significantly improving the response speed.

[0039] Preferably, the pre-analysis module supports dynamically adjusting the analysis scope (such as screening employees by department or rank), and the user can adjust the analysis scope by sending instructions.

[0040] (2) The pre-analysis module supports the advance scheduling of tasks and the utilization of system idle periods for analysis. Specifically, the user can send multiple task requests to the system at one time and specify the time when the results need to be returned for each task. The system assigns priorities based on the time when the tasks need to return results. When there is a certain time difference between the time when the tasks need to return results and the current time, the system comprehensively detects the usage of resources such as CPU, memory, and disk I / O. If the CPU usage rate is lower than 50%, the memory usage rate is lower than 60%, and the disk I / O is in an idle state, the pre-analysis is started; if any one of the resources does not meet the conditions, the pre-analysis is turned off. During the pre-analysis process, the system will record the task progress in real time. If an abnormal situation occurs (such as system crashes, task interruptions, etc.), the system will automatically save the current task progress and continue to execute the task from the position where it was interrupted last time when the pre-analysis start conditions are met next time. For example, log files can be used to record the execution steps of the tasks and the completed parts for subsequent recovery. The system will cache the pre-analysis results. When the user makes a request in advance, the system first checks whether the corresponding results exist in the cache. If they exist, they are directly called from the cache; if not, it is decided whether to start the analysis immediately or wait for a suitable time to conduct the analysis based on the task priorities and system resource conditions. If the user does not make a request in advance, the system will return the results to the corresponding user at the time preset by the user. Through the above design, the pre-analysis module can make more reasonable use of the system's analysis capabilities and resources, and conduct efficient inference analysis on delayed tasks without affecting immediate tasks.

[0041] A multi-mode human resource analysis method based on a large model, adopting the method of the above human resource analysis system, has the following operating steps: S1, Receive user instructions through the user interaction module; S2, Provide the selection of analysis modes through the mode selection module and pass the user instructions to the locally deployed large model inference engine; S2, Load employee-related information and assessment criteria-related information from the local storage device through the data loading module and perform necessary preprocessing; S4, Invoke the locally deployed large model to infer and generate comprehensive analysis results based on the assessment criteria-related information and employee-related information; S5, Save the analysis results to the result storage module; S6, Output the analysis results to the user interaction module and display them in real time, supporting color differentiation of different types of information.

[0042] Example 1: When the analysis mode selected by the user is the full-staff analysis mode or the individual precise analysis mode, the specific steps are as follows: S11, receiving and passing user instructions to the locally deployed large model inference engine; S12, loading employee-related information and inspection rule-related information, and performing necessary preprocessing; Preprocessing involves data cleaning and formatting to provide structured input for subsequent reasoning. The system uses the Pandas module to read CSV files and extract basic employee information, including fields such as employee ID, name, rank, actual years of service, minimum years of service, training completion status, KPI completion rate, professionalism score, and development potential score. Simultaneously, the python-docx module is used to parse DOCX files (such as assessment details.docx) and extract the text content of the assessment details to ensure that the five evaluation dimensions and their weightings are covered. During the data loading process, the system performs strict checks on file paths, field integrity, and outliers. For example, KPI completion rates are limited to the range of 0-1 to prevent data quality issues from affecting analysis results. Employee information and assessment details are then converted into structured string format to facilitate the construction of prompts.

[0043] 2.1 The code for loading employee information is: df=pd.read_csv("dataset / employee_information_data.csv") Field Description: Employee ID: Uniquely identifies each employee. Name: Employee name. Rank: Current rank (e.g., P1 to P7). Actual Years of Service: Actual years of service at the current rank. Minimum Years of Service: The company's minimum number of years of service. Training Completion Status: A Boolean value (True or False) indicating whether the specified training has been completed. KPI Completion Rate: Key performance indicator completion rate (range 0-1). Professional Competency Score: Scored from 0 to 100. Development Potential Score: Scored from 0 to 100.

[0044] Parameter verification: Check whether the file path exists to ensure that the data is loaded successfully; verify field integrity to ensure that all required fields exist and there are no missing values; handle abnormal values (such as negative numbers or out-of-range values), for example, limit the KPI completion rate to the range of 0-1.

[0045] 2.2 The code for loading the inspection details is: oc=Document("dataset / Inspection Details.docx") criteria_text="\n".join([para.textforparaindoc.paragraphs]) Content Description: The assessment rules include scoring criteria in five dimensions (tenure at the current level, completion of training requirements, work performance, professional qualities, and development potential) and their weight distribution; the rules also stipulate the assessment cycle for different levels and the standards for promotion, reappointment, and dismissal.

[0046] Parameter verification: Check whether the document format is correct to ensure that the complete text content can be extracted; verify the logical consistency of the details, such as whether the total weight is 100%.

[0047] 2.3 The data preprocessing code is: employee_info=(f"Employee ID:{row['Employee ID']}\n";f"Name:{row['Name']}\n";f"Position:{row['Position']}\n";f"Actual Years of Service:{row['Actual Years of Service']}years\n";f"Minimum Years of Service:{row['Minimum Years of Service']}years\n";f"Training Completion Status:{'Completed'ifrow['Training Completion Status']else'Not Completed'}\n";f"KPI Completion Rate:{row['KPI Completion Rate']*100:.2f}%\n";f"Professional Competence Score:{row['Professional Competence Score']}points\n";f"Development Potential Score:{row['Development Potential Score']}points\n" ) Output format: Employee information is presented in string format, which is convenient for constructing prompts; assessment details are presented in paragraph format, which can be directly used for inference input.

[0048] S13, large model reasoning and comprehensive evaluation, large model reasoning is the core module of this system, responsible for generating analysis results based on inspection details and employee data.

[0049] 3.1 Constructing prompts: The system converts the inspection details and employee information into structured input, guiding the large model to generate high-quality analysis reports. Implementation: Use string concatenation to construct prompts containing the inspection details and employee information. The code snippet is as follows: prompt=(f"Based on the following evaluation criteria and employee information, comprehensively analyze the employee's performance.\n";f"Evaluation criteria:\n{evaluation_criteria}\n\n";f"Employee information:\n{employee_info}") Parameter description: evaluation_criteria: The text content of the evaluation criteria extracted from the DOCX file. employee_info: The employee information extracted from the CSV file, including fields such as name, job level, actual tenure, and KPI completion rate.

[0050] It should be noted that: The design of the prompt follows the principles of clarity and conciseness, includes clear instructions (such as "comprehensive analysis" or "judge the evaluation results"), avoids vague expressions, and the length of the prompt should be appropriate to avoid exceeding the maximum input limit of the model (such as 8192 tokens).

[0051] 3.2 Call the large model API. Complete the inference task by making an HTTP request to call the locally deployed large model API (such as Ollama API). Implementation method: Use the requests.post() method to send a POST request and pass the Prompt and other parameters.

[0052] The code snippet is: data = {"model": MODEL, # Model name, such as "Qwen2.5:72b"; "prompt": prompt, # Input prompt; "stream": True, # Enable streaming output} response = requests.post(BASE_URL, json = data, stream = True, timeout = timeout) Parameter description: BASE_URL: The path of the large model API (such as "http: / / 127.0.0.1:11434 / api / generate").

[0053] MODEL: Specify the model name to be used. prompt: The constructed prompt. stream: Whether to enable streaming output (True means returning results step by step). timeout: The request timeout (in seconds, the default value is 200 seconds).

[0054] 3.3 Stream read the inference results, gradually display the inference results, improve the user experience, and avoid long waiting times. Implementation method: Use the response.iter_lines() method to read the streaming data line by line and parse the JSON-formatted response. The code snippet is: full_response = ""; for line in response.iter_lines(decode_unicode=True): if line: json_data = json.loads(line) # Parse the JSON data; token = json_data.get("response", "") # Get the token generated by the model; print(token, end="", flush=True) # Gradually print the token; full_response += token # Accumulate the complete response Parameter description: response: HTTP response object. line: Streaming data for each line (in JSON string format). json_data: Parsed JSON data. token: Each character or word generated by the model (the content output step by step). full_response: Accumulated complete inference result.

[0055] 3.4 Handle exceptions and retry mechanisms to ensure the reliability of the inference process, and continue to run even in case of network fluctuations or hardware failures.

[0056] Implementation method: Catch exceptions (such as timeouts, connection failures, etc.) and adopt an exponential backoff strategy for retries. The code snippet is: retry_count = 0; while retry_count < max_retries: try: response = requests.post(BASE_URL, json=data, stream=True, timeout=timeout); if response.status_code == 200: break # Exit the loop successfully except Exception as e: retry_count += 1; sleep_time = 2 ** retry_count # Exponential backoff strategy sleep(sleep_time) # Wait for a period of time and then retry Parameter description: max_retries: The maximum number of retries (default value is 5). retry_count: The current number of retries. sleep_time: The waiting time for each retry (in seconds, increasing exponentially).

[0057] 3.5 Return the final result and return the complete inference result to the user.

[0058] Implementation method: Clean full_response (such as removing extra spaces or line breaks), and return the final result. The code snippet is: return full_response.strip() # Return the complete inference result.

[0059] During the above large model inference process, multiple calculation formulas are involved for quantitative scoring and comprehensive evaluation. The following is a detailed description: (1) Single item scoring formula Score for the length of service in the current position:

[0060] Parameter description: : Score for the length of service (percentage system). Actual length of service: The actual working years of the employee in the current position. Minimum length of service: The minimum length of service stipulated by the company.

[0061] Score for the completion of training conditions:

[0062] Parameter description: Score for the completion of training (percentage system). Completion of training: Boolean value (True or False).

[0063] Score for work performance:

[0064] Parameter description: : Score for work performance (percentage system). KPI completion rate: The completion rate of the employee's key performance indicators (range 0 - 1).

[0065] Score for professional qualities:

[0066] Parameter description: : Score for professional qualities (directly use the score, range 0 - 100).

[0067] Score for development potential:

[0068] Parameter description: : Development potential score (directly use the score, range 0 - 100).

[0069] (2) Comprehensive scoring formula After the above reasoning is completed, the system calculates the comprehensive score of each employee according to the weight distribution formula in the inspection rules. For example, the tenure at this position level accounts for 10%, the completion of training conditions accounts for 20%, work performance accounts for 40%, professional qualities account for 20%, and development potential accounts for 10%, and finally the total score is obtained. According to the comprehensive score, the system determines the inspection results (promotion, renewal or dismissal) of the employees and generates a detailed analysis report. This stage fully combines the powerful reasoning ability of the large model and the scientific scoring mechanism to ensure the objectivity and reliability of the analysis results.

[0070] Total score calculation formula:

[0071] Parameter description: : Comprehensive score (percentage system).

[0072] Weight distribution: Tenure at this position level: 10%; Completion of training conditions: 20%; Work performance: 40%; Professional qualities: 20%; Development potential: 10%.

[0073] (3) Determine the inspection results. According to the comprehensive score, determine the inspection results of the employees according to the following criteria. Inspection result determination criteria: Promotion: ; Renewal: ; Dismissal: .

[0074] Output format: Return a clear conclusion (such as "Promotion", "Renewal" or "Dismissal"), and attach a detailed analysis report.

[0075] S14, save the analysis result to the result storage module; Use the Pandas module to save the analysis result as a CSV file encoded in UTF-8. The code snippet is: df.to_csv(output_file,index=False,encoding="utf-8-sig") File format: The analysis results of all employees are saved as an independent CSV file; The personal precise analysis results are saved to the historical record file in append mode.

[0076] S15. Output the analysis results to the user interaction module for real-time display, supporting color differentiation for different types of information (e.g., green for promotion, yellow for renewal, and red for dismissal). The terminal color output is implemented through the colorama module to enhance the readability of the information. In addition, the system also supports the function of exporting reports, generating files in PDF or Excel format for the convenience of management decision-making. The design of this stage not only focuses on the security and usability of data but also meets the needs of different users through diverse output methods, thus enhancing the practicality and flexibility of the overall system.

[0077] The code snippet is: print(Fore.GREEN + f"Assistant: The inspection result of {row['Name']} is as follows: {result}") In terms of output format, the terminal color output is implemented through the colorama module to enhance the readability of the information. In addition, the system also supports the function of exporting reports, generating files in PDF or Excel format for the convenience of management decision-making. The design of this stage not only focuses on the security and usability of data but also meets the needs of different users through diverse output methods, thus enhancing the practicality and flexibility of the overall system.

[0078] Embodiment 2: When the analysis mode selected by the user is the targeted analysis mode, its specific steps are refined as follows: S21. Data collection and preparation. Load the unstructured data of employees from the local storage device; this data includes project completion reports, supervisor evaluations, and work summaries. These documents are usually stored in the form of text files, and the file naming rules can be standardized according to the employee ID, such as {employee ID}_project_report.txt, {employee ID}_supervisor_feedback.txt, etc., so that the system can quickly locate and read the data of specific employees. To ensure the efficiency of data loading, the system will preferentially use high-speed storage devices (such as nvme SSD) to read files and cache frequently used data in memory to reduce repeated read operations. During the data loading process, the system will strictly verify the file path, encoding format, and content integrity. For example, UTF-8 encoding is uniformly adopted to support multilingual characters, and at the same time, the maximum size limit of a single file (such as 5MB) is set to avoid affecting performance due to loading oversized files. In addition, if a file is missing or the format does not match, the system will capture the exception and display a detailed error prompt on the terminal to remind the user to supplement the complete data. The above design improves the reliability of data loading and lays a solid foundation for subsequent analysis.

[0079] S22, Data Preprocessing and Cleaning: Preprocess and clean unstructured data to remove meaningless content and extract key information. First, the system removes punctuation marks, stop words (such as "of", "is"), and redundant expressions through regular expressions to ensure that the text content is more concise and standardized. At the same time, all text is converted to lowercase for consistency in subsequent analysis. For special content such as dates and numbers, the system standardizes them to a unified format, for example, converting "January 2023" to "2023-01". Next, the system splits the document into multiple paragraphs based on line breaks or specific delimiters (such as blank lines or heading markers) and marks the theme of each paragraph. For example, a paragraph containing "Project Goals" is marked as "Project Background", and a paragraph containing "Completion Status" is marked as "Task Execution". During this process, the system also detects whether the document lacks key content (such as project outcome descriptions) and prompts the user to supplement it in a timely manner. To improve the accuracy of keyword extraction, the system combines a domain dictionary (such as professional terms in the human resources field) to expand the custom stop word list and adjusts the paragraph segmentation rules and cleaning strategies according to actual needs. The above can be optimized according to specific application scenarios.

[0080] S23, Data Structuring and Competency Scoring: Convert unstructured data into structured information for quantitative competency scoring. The core of this stage is to use natural language processing (NLP) techniques and large model inference capabilities to extract keywords related to evaluation dimensions from the text and generate scores. For example, the system can use spaCy or HuggingFace Transformers to extract keywords related to dimensions such as professional skills, execution ability, and innovation ability, and calculate scores in combination with the weight distribution formula in the inspection rules. To further improve the accuracy of analysis, the system also performs sentiment analysis on the text to determine whether the tone is positive, negative, or neutral, and extracts key statements of positive evaluations (such as "performed excellently") and negative evaluations (such as "communication skills need improvement"). In addition, topic modeling algorithms (such as LDA or BERTopic) are used to identify the core themes in the document and map them to the evaluation dimensions. When constructing the prompt, the system combines the inspection rules and the data structuring results to clearly require the large model to generate high-quality scores and analysis reports. For example, the prompt will detail the evaluation dimensions and their weight distributions and control the length within the maximum input limit of the model (such as 2048 tokens). Finally, the system generates a comprehensive score based on the scores of each dimension. For example, professional skills account for 25%, execution ability accounts for 25%, innovation ability accounts for 20%, teamwork accounts for 20%, and work attitude accounts for 10%, and the total score is calculated through weighted average. The selection of these parameters needs to be dynamically adjusted according to the actual needs of the enterprise to ensure the scientificity and objectivity of the scoring results.

[0081] (1)Scoring Model Design Evaluation Dimensions and Weights: Professional Skills (25%): Extract keywords related to technical capabilities from the project completion report.

[0082] Execution (25%): Score according to indicators such as project completion time and task completion rate.

[0083] Innovation Ability (20%): Extract keywords related to innovation from the text.

[0084] Team Collaboration (20%): Extract keywords related to teamwork from the supervisor's evaluation and work summary.

[0085] Work Attitude (10%): Extract keywords related to work attitude from the supervisor's evaluation and work summary.

[0086] (2)Large Model Inference Prompt Construction: Combine the inspection rules and unstructured data to construct structured prompt words.

[0087] prompt=(f"According to the following inspection rules and the employee's unstructured data, quantitatively score the employee's capabilities (out of 100 points).\n"; f"Inspection rules:\n{evaluation_criteria}\n"; f"Project completion report:\n{project_report}\n"; f"Supervisor's evaluation:\n{supervisor_feedback}\n"; f"Work summary:\n{work_summary}\n"; f"Please output the score for each dimension (e.g., Professional Skills: 20 / 25).") Call the large model: Use streaming output to gradually display the analysis results.

[0088] result=generate_response(prompt) (3)Parameter Selection Prompt Length: Control the length of the Prompt within the maximum input limit of the model (e.g., 2048 tokens).

[0089] Scoring Range: The scoring range for each dimension is 0 - 100 points, and the final comprehensive score is the weighted average.

[0090] S21, Report Generation and Output. After completing the quantitative competency scoring, the system generates a detailed directional analysis report in various formats for easy review and archiving. The report includes each employee's score, comprehensive evaluation, and improvement suggestions. The score distribution is visually displayed using bar charts and radar charts. For example, a bar chart can display a comparison of scores across competencies, while a radar chart clearly illustrates an employee's overall performance across different assessment dimensions. To enhance the user experience, the system uses color coded information on the terminal, such as green for high scores, yellow for medium scores, and red for low scores. Furthermore, the system supports saving analysis results as a separate CSV file (employee_directional_analysis_results.csv) for long-term tracking and comparison of all employee data. Individual precision analysis results are saved in append mode to a history file (person_directional_analysis_results.csv) to ensure data persistence and traceability. Reports can be exported in PDF and Excel formats to meet the needs of different users. To further enhance performance, the system incorporates the distributed computing framework Spark. Using Spark's Resilient Distributed Datasets (RDDs) and Datasets, data is distributed across multiple nodes in the cluster for parallel processing. Furthermore, a caching mechanism, such as in-memory caching (e.g., Spark's in-memory caching strategy), stores frequently used query results in memory, reducing disk I / O operations. GPUs are also used to accelerate inference on large models, significantly reducing analysis time. This design enhances system practicality and provides users with diverse output methods to meet the needs of diverse scenarios.

[0091] The core effectiveness of targeted analytics lies in its powerful data processing capabilities and intelligent scoring mechanism. Traditional human resources analytics systems typically rely on structured data (such as KPI completion rates and professional competency scores). Targeted analytics overcomes this limitation by incorporating large models (Qwen2.5:72b) and NLP technology to transform employees' project completion reports, supervisor evaluations, and work summaries into quantifiable scores and insights. Combined with the weighting formulas in the assessment criteria, it generates multi-dimensional quantitative scores for competencies (such as professional skills, execution, and innovation). This automated analysis significantly reduces the time and cost of manual intervention while improving the objectivity and accuracy of assessment results.

[0092] Example 3: When the analysis mode selected by the user is a customized analysis mode, the specific steps are as follows: S31, Data Preparation and Analysis: Collect and analyze project plans and employee capability data; The project plan, the core input for analysis, includes specific project requirements and goals, such as market development capabilities, technical background, or teamwork skills. The system uses natural language processing (NLP) technology to parse the plan content, extract key requirements, and convert them into scoring dimensions. For example, if the plan mentions "strong market development capabilities," the system will identify "market development capabilities" as a core scoring dimension and assign it a higher weight. The system also loads employee structured data (such as KPI completion rates and professionalism scores) and unstructured data (such as project completion reports, supervisor evaluations, and work summaries). This data is loaded using file path management to ensure quick location and access for each employee. To improve efficiency, the system prioritizes high-speed storage devices (such as NVMe SSDs) for file reading and caches frequently used data in memory to reduce duplication. Furthermore, the system performs strict validation of file paths, encoding formats, and content integrity. It uniformly uses UTF-8 encoding to support multilingual characters and sets a maximum file size limit (such as 5MB) to prevent performance degradation caused by loading extremely large files. If a file is missing or its format doesn't match, the system will catch the exception and display a detailed error message on the terminal, prompting the user to complete the data. This design not only improves the reliability of data loading but also lays a solid foundation for subsequent analysis.

[0093] S32, Data Preprocessing and Capability Model Construction. After data loading is complete, the system preprocesses the unstructured data to remove meaningless content and extract key information. First, the system uses regular expressions to remove punctuation, stop words (such as "的" and "是"), and redundant expressions to ensure concise and standardized text content. Next, the system splits the document into paragraphs based on line breaks or specific delimiters (such as blank lines or heading tags) and tags each paragraph with its topic. For example, a paragraph containing "market development" would be tagged as "business capability," while a paragraph containing "team coordination" would be tagged as "management capability." During this process, the system also detects if the document is missing key content (such as project achievement descriptions or team collaboration examples) and prompts the user to complete it. To improve the accuracy of keyword extraction, the system expands a custom stop word list by integrating domain dictionaries (such as professional terminology in the human resources field) and adjusts segmentation rules and cleaning strategies based on actual needs. Furthermore, the system uses topic modeling algorithms (such as LDA or BERTopic) to identify core topics in the document and map them to scoring dimensions. For example, the "Market Development" theme corresponds to "Market Development Ability" within the business capability, and the "Team Collaboration" theme corresponds to "Team Leadership" within the management capability. The selection of these parameters can be optimized based on the specific application scenario to ensure the accuracy and scientific nature of subsequent analysis.

[0094] (1) The scoring dimensions of business capability scoring are: Market development capabilities: Extract relevant information from historical performance and unstructured data.

[0095] Technical expertise: Extract technology-related keywords from project completion reports.

[0096] Data analysis ability: Extract keywords such as "data modeling" and "trend forecasting" from work summaries.

[0097] Customer Relationship Management: Extract relevant information from customer satisfaction ratings and unstructured data.

[0098] Scoring formula: defcalculate_business_score(employee_data,weights): market_expansion = employee_data["Market Development Capability"]*weights["Market Development"] technical_expertise=employee_data["Technical Expertise"]*weights["Technical Expertise"] data_analysis=employee_data["Data Analysis Ability"]*weights["Data Analysis"] customer_management = employee_data["Customer Relationship Management"]*weights["Customer Relationship"] returnmarket_expansion+technical_expertise+data_analysis+customer_management Scoring dimensions: market development capabilities (M); technical expertise (T); data analysis capabilities (D); customer relationship management (C).

[0099] Weights: Market development weight: w_M; Technical expertise weight: w_T; Data analysis weight: w_D; Customer relationship weight: w_C.

[0100] Scoring formula: (2) The scoring dimensions of management ability are: Team Leadership: Extract relevant information from team member productivity improvements and unstructured data.

[0101] Resource coordination capability: Extract keywords such as “resource integration” and “cross-departmental collaboration” from project on-time completion rates and unstructured data.

[0102] Goal achievement ability: Extract keywords such as "phased goal" and "strong execution ability" from the goal completion rate and unstructured data.

[0103] Decision-making ability: Extract keywords such as "quick response" and "risk assessment" from the success rate of major decisions and unstructured data.

[0104] Scoring formula: def calculate_management_score(employee_data, weights): team_leadership = employee_data["Team leadership"] * weights["Team leadership"] resource_coordination = employee_data["Resource coordination ability"] * weights["Resource coordination"] goal_achievement = employee_data["Goal achievement ability"] * weights["Goal achievement"] decision_making = employee_data["Decision-making ability"] * weights["Decision-making"] return team_leadership + resource_coordination + goal_achievement + decision_making Scoring dimensions: Team leadership (L); Resource coordination ability (R); Goal achievement ability (G); Decision-making ability (K).

[0105] Weights: Team leadership weight: w_L; Resource coordination weight: w_R; Goal achievement weight: w_G; Decision-making ability weight: w_K.

[0106] Scoring formula: The formula for the comprehensive score is: def calculate_comprehensive_score(employee_data, business_weights, management_weights): business_score = calculate_business_score(employee_data, business_weights) management_score = calculate_management_score(employee_data, management_weights) return business_score + management_score

[0107] S33, Customized Analysis and Recommendation Mechanism (1)Dynamic weight allocation, dynamically adjust the weights of business capabilities and management capabilities according to the requirements in the project plan; for example: if the project plan mentions "employees with strong market development capabilities", the weight of market development capabilities can be increased to 40%, if the project plan mentions "employees with strong teamwork capabilities", the weights of team leadership and resource coordination capabilities can be increased to 30% and 20% respectively. This dynamic weight allocation mechanism ensures that the scoring model can flexibly adapt to the requirements of different projects.

[0108] Implementation method: def assign_weights(key_requirements): weights = { "Market Development": 0.25, "Technical Expertise": 0.20, "Data Analysis Ability": 0.15, "Customer Relationship Management": 0.10, "Team Leadership": 0.15, "Resource Coordination Ability": 0.10, "Goal Achievement Ability": 0.05, "Decision-making Ability": 0.05 } for req in key_requirements: if req == "Market Development": weights["Market Development"] += 0.15 elif req == "Team Coordination": weights["Team Leadership"] += 0.10 weights["Resource Coordination Ability"] += 0.10 Return weights (2) Call a large model (such as Qwen2.5:72b) to conduct an in-depth analysis of each employee's capabilities; combine the project plan and the employee's unstructured data to construct prompt words to explicitly require the model to output the score of each dimension (such as "market development capability: 20 / 25") as well as the comprehensive evaluation and total score.

[0109] The prompt could be something like: "Based on the following project plan and the employee's unstructured data, conduct a customized analysis of the employee's capabilities. Please output the score for each dimension and provide a comprehensive evaluation and total score." The scoring results generated by the large model are automatically saved in the system and used for subsequent sorting and recommendation.

[0110] (3) Sorting and recommendation: Rank employees according to their comprehensive scores and recommend employees with the highest scores.

[0111] Implementation method: defrecommend_employee(employees,business_weights,management_weights): sorted_employees = sorted( employees, key=lambdax:calculate_comprehensive_score(x,business_weights,management_weights), reverse=True) recommended_employee=sorted_employees[0] returnrecommended_employee Output recommended_employee=recommend_employee(employees,business_weights,management_weights) print(f"Recommended employee:{recommended_employee['name']},comprehensive score:{calculate_comprehensive_score(recommended_employee,business_weights,management_weights)}")") To further improve the accuracy of recommendations, the system will sort employees based on their comprehensive scores and select the candidates with the highest scores. For example, if an employee performs outstandingly in "market development capabilities" and "team leadership", the system will give priority to recommending this employee to complete the project. In addition, the system supports user-defined scoring rules and weight distribution, such as allowing users to manually adjust the weights of certain dimensions or add new scoring criteria. The selection of these parameters needs to be dynamically adjusted according to the actual needs of the enterprise to ensure the scientificity and objectivity of the recommendation results. Finally, the system will display the employee's score distribution in the form of a chart and generate a detailed recommendation report, listing the reasons for the recommendation (such as "this employee has outstanding performance in market development capabilities and team leadership"). This design improves the practicality of the system and provides enterprises with a variety of output methods to meet the needs of different scenarios.

[0112] Although this specification is described according to implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A multi-modal human resource analysis system based on a large model, characterized by: The analysis system includes: User interaction module, used to input commands and display analysis results; The mode selection module is used to provide analysis mode selection and pass user instructions to the locally deployed large model inference engine; Data loading module, used to load employee-related information and inspection details from local storage devices; The large model inference module is used to call the locally deployed large model to generate comprehensive analysis results based on employee-related information and assessment details; The result storage module is used to save the analysis results to a storage device.

2. The multi-modal human resource analysis system based on a large model according to claim 1, characterized in that, The analysis modes include: All-staff analysis mode, used to perform batch analysis on all employees; Personal precision analysis mode for generating detailed reports for specific employees; Directed analysis mode for deep mining and quantitative scoring of unstructured data; Customized analysis mode is used to dynamically adjust the scoring model weights based on the specific needs of the project plan and provide scientific personnel matching recommendations.

3. The multi-modal human resource analysis system based on a large model according to claim 2, characterized in that: When the targeted analysis mode is selected, the large model reasoning module combines NLP technology and topic modeling algorithms to extract key information from unstructured text and convert it into structured data, and combines the weight distribution formula to generate multi-dimensional capability quantitative scores; When the customized analysis mode is selected, the large model reasoning module analyzes the project plan, extracts key requirements, and converts them into specific scoring dimensions, generating quantitative scores based on the employees' business capabilities and management capabilities.

4. The multi-modal human resource analysis system based on a large model according to claim 2, characterized in that: The system uses high-speed memory and storage devices to work together, uses Python as the programming language, and uses the Ollama API reasoning framework in combination with third-party libraries to achieve efficient data processing, interaction with large models, and user interaction functions.

5. The multi-modal human resource analysis system based on a large model according to claim 2, characterized in that: The large model reasoning module includes a prompt word construction module, a large model API calling module, an inference result reading module, an exception handling and retry module, and a final result return module; The prompt word construction module constructs structured prompt words of employee-related information and inspection rule-related information by string concatenation; The large model API calling module calls the locally deployed large model API through HTTP requests to complete the inference task; The inference result reading module is used to read data and display the inference result; The exception handling and retry module is used to capture exceptions and adopt a timeout retry mechanism; The final result return module pushes the complete reasoning result to the result storage module and the user interaction module.

6. The multi-modal human resource analysis system based on a large model according to claim 5, characterized in that: When the large model inference module adopts the exponential backoff strategy for timeout retry, the upper limit of the number of retries is N times, N is an integer greater than 1, and the waiting time for each retry increases by a power of 2.

7. The multi-modal human resource analysis system based on a large model according to claim 2, characterized in that: The system also includes a pre-analysis module; The pre-analysis module is used to start after the set pre-analysis time is reached to perform task processing; The data loading module pre-loads specific data into the high-speed memory cache before the set pre-analysis time; The mode selection module transmits the instructions sent by the pre-analysis module to the large model inference module; The large model reasoning module caches the pre-analyzed reasoning results in memory.

8. A multi-modal human resource analysis method based on a large model, characterized in that, include: S1, receiving user instructions through the user interaction module; S2. Provide the selection of analysis modes through the mode selection module and pass the user instructions to the large model inference engine deployed locally; S3. Load the employee-related information and the inspection rules-related information from the local storage device through the data loading module and perform necessary preprocessing; S4. Call the large model deployed locally to generate comprehensive analysis results based on the employee-related information and the inspection rules-related information; S5. Save the analysis results to the result storage module; S6. Output the analysis results to the user interaction module and display them in real time.

9. The multi-modal human resource analysis method based on a large model according to claim 8, wherein: The analysis modes include the all-staff analysis mode, the individual precise analysis mode, the targeted analysis mode, and the customized analysis mode; For the all-staff analysis mode, the data loaded through the data loading module includes the basic information of all employees and the inspection rules, and the output analysis results include the comprehensive scores of each employee and the classification of inspection results; For the individual precise analysis mode, the data loaded through the data loading module includes the basic information of a single employee and the inspection rules, and the output analysis results include in-depth analysis of the professional qualities and development potential dimensions of the employee, as well as more detailed evaluation conclusions and improvement suggestions; For the targeted analysis mode, the data loaded through the data loading module includes the unstructured data of employees and the inspection rules, and the output analysis results include the quantitative scores, comprehensive scores, and visual display charts of employees in the dimensions of professional skills, execution ability, and innovation ability; For the customized analysis mode, the data loaded through the data loading module includes the project plan and the ability data of employees. The ability data of employees includes structured and unstructured data, and the output analysis results include the scores of employees in the dimensions of market development ability and teamwork ability, the recommended candidates and their score distributions, and the reasons for recommendation.

Images