A data name recommendation and review assistance system of a large language model

Abstract

The application provides a data name recommendation and review auxiliary system of a large language model, and the steps include data collection and collection, model deployment, API development, front-end integration and display, and realization of review auxiliary functions. The application is customized for the application name data of the public data operation and management system, and a large language model fine-tuning method is used. Through fine-tuning of the large language model for specific data, the model has higher precision and accuracy in understanding application names.

Description

Technical Field

[0001] This invention relates to the field of computer algorithms, and more specifically, to a data name recommendation and verification assistance system based on the Chinese-Vicuna large language model. Background Technology

[0002] Currently, various data naming recommendation and verification systems exist in the market. However, most algorithms lack iterative processing of the acquired data, relying on conventional practices. This often leads to inaccuracies in data processing and application, such as low naming quality. Different departments use different naming methods and standards for data directories, resulting in spelling errors, grammatical errors, or unclear terminology, leading to inconsistent naming styles and formats. This hinders cross-departmental collaboration, making it difficult for other users to understand the meaning of the data directories, reducing data utilization efficiency, and increasing understanding and communication costs. Secondly, there is a lack of unified naming conventions: the absence of clear naming specifications and standards can lead to naming chaos, which is detrimental to data management and maintenance. Furthermore, verification is difficult; manual verification may be subjective and inconsistent, resulting in varying verification results. Summary of the Invention

[0003] To address this problem, this invention provides a data name recommendation and verification assistance system based on the Chinese-Vicuna large language model, the specific technical solution of which is as follows:

[0004] A data name recommendation and review assistance system for a large language model includes: Step 1, data collection, using Python's requests library to simulate network requests and scrape data directory names from a public data operation and maintenance management system; Step 2, data processing, using the language model Chinese-Vicuna, and performing model selection and fine-tuning on the scraped data directory names, including data preprocessing, using the HuggingFaceTransformers library to process the structured dataset and convert it into the input format required by the large language model; Step 3, model deployment, using containerization technology to deploy the fine-tuned large language model Chinese-Vicuna to a server; Step 4, API development, using a framework protocol to generate interfaces for intelligent recommendation and review assistance functions and designing the input and output formats of the API; Step 5, front-end integration and display, introducing intelligent recommendation functionality into the React front-end of the public data operation and maintenance management system, using a library to send requests to the back-end API in real time and obtain and display the recommendation results; Step 6, implementing review assistance functionality, collecting the system's effectiveness and problems in review assistance, and using the collected feedback data to further fine-tune the large language model.

[0005] Furthermore, the data preprocessing includes the following specific operations: loading the pre-trained Chinese-Vicuna model and corresponding word segmenter using the HuggingFaceTransformers library; segmenting the data directory name text and converting the segmented text into the input tensor required by the model; fine-tuning the Chinese-Vicuna model: setting fine-tuning training parameters according to the training requirements of the Chinese-Vicuna model, and using the data directory name data from the public data operation and maintenance management system to fine-tune the large language model.

[0006] Furthermore, the captured data directory names are preprocessed. The preprocessing steps include: removing null values, deleting null values ​​and invalid data; deduplication, removing duplicate data directory names; text cleaning, i.e., removing special characters and correcting spelling errors; and data cleaning, using Python's pandas and xlwings libraries to organize the preprocessed data into a structured dataset.

[0007] Furthermore, the model fine-tuning steps include configuring training parameters using the training tools provided by the HuggingFaceTransformers library, and passing the training set and validation set into the Chinese-Vicuna model.

[0008] Furthermore, the steps for organizing the structured dataset include first using the pandas library to read the preprocessed data and convert it into a DataFrame format; then using the xlwings library to export the processed data to a CSV file.

[0009] Furthermore, the API's input and output formats, including data directory name queries, recommendation results, and review suggestions, define the RESTful API path and JSON data format.

[0010] Furthermore, an audit assistance function has been added to the display interface.

[0011] As can be seen from the above technical solution, the beneficial effects of the present invention are: 1) A customized large language model fine-tuning method for application name data in the public data operation and maintenance management system, which improves the accuracy and precision of the model in understanding application names by fine-tuning specific data of the large language model; 2) The design and implementation of an intelligent recommendation and review assistance system based on the large language model, which integrates the fine-tuned model into the public data operation and maintenance management system to provide users with intelligent recommendation and review assistance functions; 3) Combining the front-end and back-end technology stacks of the public data operation and maintenance management system, such as React and Java, an easy-to-integrate model deployment and API development scheme is designed to achieve seamless integration with the existing operation and maintenance system; 4) By continuously collecting user feedback and system usage, the model performance is further optimized to improve the accuracy and practicality of intelligent recommendation and review assistance functions. Attached Figure Description

[0012] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate corresponding parts.

[0013] Figure 1 This is a flowchart illustrating the data name recommendation and review assistance system for a large language model according to an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to its embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.

[0015] It should be noted that when an element is referred to as being "fixed to" another element, it can be directly attached to the other element or there may be an intervening element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or there may be an intervening element. The terms "vertical," "horizontal," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0016] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0017] In this invention, "first" and "second" do not represent a specific quantity or order, but are merely used to distinguish names.

[0018] like Figure 1 As shown, an embodiment of the present invention provides a data name recommendation and review assistance system for a large language model, comprising: Step 1, data collection, using Python's requests library to simulate network requests and scrape data directory names from a public data operation and maintenance management system, i.e., first sending an HTTP request to the target website, obtaining the response content, and then using regular expressions to extract the required data directory names; Step 2, data processing to improve data quality and accuracy, using the language model Chinese-Vicuna, and performing model selection and fine-tuning on the scraped data directory names. This model has been pre-trained for Chinese text and can better understand Chinese context and semantic information, thereby improving the model's performance on related tasks. This step involves data preprocessing, using the HuggingFaceTransformers library to process the structured dataset into the input format required by the large language model; Step 3, model deployment, using containerization technology to deploy the fine-tuned large language model Chinese-Vicuna to a server. Here, the containerization technology can be Docker or Kubernetes. The first step involves an open-source container orchestration system for automating application deployment, scaling, and management. It runs in various environments and is highly compatible with Docker. The second step, API development, utilizes backend frameworks such as Spring Boot and Node.js's Express.js to handle HTTP requests, process business logic, and interact with databases or other backend services. This generates interfaces for intelligent recommendation and review assistance functions and designs the API's input and output formats. The third step, frontend integration and display, introduces intelligent recommendation functionality into the React frontend of the public data operation and maintenance management system. It uses a library to send requests to the backend API in real time, retrieve recommendation results, and display them. This primarily involves displaying the intelligently recommended text suggestions, adding a text display module, and introducing WebSockets to achieve bidirectional communication with the server, presenting review suggestions in a streaming format. The fourth step, implementing review assistance, adds text content of review suggestions to the review confirmation interface to help reviewers notice overlooked points during the review process, enhance review accuracy, and collect data on the system's effectiveness and issues in review assistance.

[0019] In one embodiment, the preprocessing of the captured data directory names includes the following steps: removing null values: deleting null values ​​and invalid data to ensure data integrity; deduplication: removing duplicate data directory names to ensure data uniqueness; text cleaning: cleaning the data directory names, including removing special characters and correcting spelling errors, to improve data readability and accuracy; when using Python's pandas and xlwings libraries to organize the preprocessed data into a structured dataset, the pandas library is first used to read the preprocessed data and convert it into a DataFrame format, and then the xlwings library is used to export the processed data to a CSV file for easier subsequent data analysis and visualization.

[0020] In one embodiment, the data processing involves using the Chinese-Vicuna language model for model selection and fine-tuning of the crawled data directory names. This includes data preprocessing, where the structured data directory name dataset is converted into the input format required by the large language model using the Hugging Face Transformers library. The input text is then divided into a training set and a validation set. The validation set is used to evaluate the model's performance during training and to adjust model parameters or change training strategies. The validation set data and training set data are separate, but they typically come from the same overall dataset to prevent overfitting. Next, the pre-trained Chinese-Vicuna model and its corresponding word segmenter are loaded using the Hugging Face Transformers library. The input text is segmented, and the segmented text is converted into the input tensor required by the model, which is then provided to the Chinese-Vicuna model for fine-tuning. Based on the training requirements of the Chinese-Vicuna model, fine-tuning training parameters are set, and the large language model is fine-tuned using data directory names from the public data operation and maintenance management system.

[0021] In one embodiment, the front-end integration and display involves introducing an intelligent recommendation function into the React front-end of the public data operation and maintenance management system. A library is used to send requests to the back-end API in real time and obtain and display the recommendation results. When the user enters a data directory name, Axios or other HTTP libraries are used to send requests to the back-end API in real time and obtain the recommendation results. The recommendation results are then displayed to the user in a user-friendly manner, such as through drop-down lists or auto-completion.

[0022] In one embodiment, an audit assistance function is introduced into the audit personnel interface. Based on the audit suggestions returned by the backend API, the function provides audit personnel with references, such as highlighting problematic parts. At the same time, the function can collect the system's effectiveness and problems in audit assistance, and use the collected feedback data to perform 4-bit quantization fine-tuning on the large language model.

[0023] In one embodiment, API development utilizes a backend framework to generate interfaces for intelligent recommendation and review assistance functions and designs the API's input and output formats. The API's input and output formats include data directory name queries, recommendation results, and review suggestions, and define RESTful API paths and JSON data formats.

[0024] In one embodiment, the API path for retrieving recommended data directories creates the path `server_domain / recommendations`, while the API for retrieving audit suggestions creates the path `server_domain / audit-suggestions`. The JSON data format design considers the data structure and type. For a request to `server_domain / recommendations`, the response is designed to return an array containing a series of recommendations, while for a request to `server_domain / audit-suggestions`, an object containing a `suggestion` string is returned.

[0025] In one embodiment, the system further fine-tunes the large language model using collected feedback data by updating the Chinese-Vicuna model deployed on the server via automated scripts or a continuous deployment process. The system utilizes the continuous integration / continuous deployment (CI / CD) system Jenkins to monitor the source code management system and automatically execute update and deployment tasks when new model files are detected; and the containerized application Docker running on the server is used to deploy and run the machine learning model. API performance is regularly checked and optimized, and server resource usage is monitored.

[0026] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0027] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A data name recommendation and verification auxiliary system for a large language model, characterized in that, include: Step 1, data collection: Use Python's requests library to simulate network requests and retrieve data directory names from the public data operation and maintenance management system. That is, first send an HTTP request to the target website, get the response content, and then use regular expressions to extract the required data directory names. Step two, data processing, using the Chinese-Vicuna language model, and performing model selection and fine-tuning on the names of the crawled data directories, including data preprocessing, using the HuggingFaceTransformers library to process the structured dataset into the input format required by the large language model; In data processing, the HuggingFaceTransformers library is used to load the pre-trained Chinese-Vicuna model and the corresponding word segmenter; the input text is segmented into words, and the segmented text is converted into the input tensor required by the model, which is then provided to the Chinese-Vicuna model for fine-tuning training; according to the training requirements of the Chinese-Vicuna model, fine-tuning training parameters are set, and the large language model is fine-tuned using the data directory name data of the public data operation and maintenance management system. Step 3, Model Deployment: Use containerization technology to deploy the fine-tuned large language model Chinese-Vicuna to the server; Step 4, API Development: Use the backend framework to generate interfaces for intelligent recommendation and review assistance functions and design the input and output formats of the API; The API's input and output formats include data directory name queries, recommendation results, and review suggestions, defining the RESTful API path and JSON data format; The API path retrieves the recommended data directory name and creates the path server_domain / recommendations. For the API that retrieves audit suggestions, the path server_domain / audit-suggestions is created. The JSON data format is designed considering the structure and type of the data. For a request to server_domain / recommendations, an array containing a series of recommendations is returned. For a request to server_domain / audit-suggestions, an object containing a suggestion string is returned. Step 5: Front-end integration and display. Introduce intelligent recommendation functionality into the React front-end of the public data operation and maintenance management system. Use a library to send requests to the back-end API in real time, retrieve recommendation results, and display them. When a user enters a data directory name, use Axios to send requests to the back-end API in real time and retrieve recommendation results; then display the recommendation results to the user in a user-friendly manner. Step Six: Implement the review assistance function. Introduce the review assistance function into the reviewer interface. Based on the review suggestions returned by the backend API, provide reviewers with references. At the same time, collect the system's effectiveness and problems in review assistance. Use the collected feedback data to fine-tune the large language model using 4-bit quantization. Update the Chinese-Vicuna model deployed on the server through automated scripts or continuous deployment processes. Utilize the continuous integration or continuous deployment system Jenkins to monitor the source code management system and automatically execute update and deployment tasks when new model files are detected. There is also a containerized application Docker running on the server for deploying and running machine learning models. Regularly check and optimize API performance and monitor server resource usage.

2. The data name recommendation and verification auxiliary system for large language models according to claim 1, characterized in that, The data preprocessing includes the following steps: loading the pre-trained Chinese-Vicuna model and its corresponding word segmenter using the HuggingFaceTransformers library; segmenting the data directory name text and converting the segmented text into the input tensor required by the model; and fine-tuning the Chinese-Vicuna model by setting the fine-tuning training parameters according to the training requirements of the Chinese-Vicuna model.

3. The data name recommendation and verification auxiliary system for large language models according to claim 1, characterized in that, The steps for fine-tuning the model include configuring training parameters using the training tools provided by the HuggingFaceTransformers library, and passing the training and validation sets into the Chinese-Vicuna model.

4. The data name recommendation and verification auxiliary system for large language models according to claim 3, characterized in that, The steps for organizing a structured dataset include first using the pandas library to read the preprocessed data and convert it into a DataFrame format; then using the xlwings library to export the processed data to a CSV file.

5. The data name recommendation and verification auxiliary system for large language models according to claim 1, characterized in that, The API's input and output formats include data directory name queries, recommendation results, and review suggestions, defining the RESTful API path and JSON data format.

Images