AIGC key content examination system based on NLP and authority text driving

Abstract

The invention relates to the technical field of natural language processing, in particular to an AIGC key content review system based on NLP and authoritative text driving, which comprises an authoritative text acquisition module, a question generation and review module, a data management module and an interactive execution module, the authoritative text acquisition module provides an original text for the topic generation and examination module through a standardized data interface; the question generation and review module outputs the structured data to the data management module through NLP analysis and semantic review; the interaction execution module calls storage resources and carries out front-end and back-end asynchronous interaction; the interactive execution module comprises a front-end visual operation unit and a rear-end data processing unit. According to the invention, the authoritative text collection module is combined with the NLP technology to generate questions, the man-machine cooperation module supports manual correction and result optimization of the questions and results, a Vue + Flask architecture is adopted to realize asynchronous interaction of front and rear ends, JSON structured storage guarantees data transmission efficiency, high concurrent processing is supported, and expansibility and response speed are both achieved.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing technology, and in particular to an AIGC key content review system driven by NLP and authoritative text. Background Art

[0002] In recent years, with the introduction of COT, the development of pre-training technology and the improvement of computing hardware, AIGC has made breakthrough progress, especially the generative dialogue model represented by ChatGPT. With the rapid development of generative artificial intelligence (AIGC) technology, its application in various fields is becoming more and more extensive, but it also exposes significant security risks.

[0003] Currently, AIGC-generated content suffers from systemic deficiencies in its security regarding politics, ethics, and ideology. Existing censorship mechanisms rely heavily on rule-based matching or simple supervised learning, making them ill-equipped to adapt dynamically in complex contexts. Existing domestic censorship products face multiple challenges. Technically, they face issues such as insufficient real-time performance, difficulty identifying deepfakes, and limited adaptability to new technologies. Externally, these challenges are limited by inconsistent censorship standards, privacy risks, and subjective ethical judgments. Public surveys reveal that 37.65% of respondents lack awareness of AIGC security, and nearly half choose to ignore false information when encountering it, highlighting the urgent need to improve the public's ability to discern AI content.

[0004] Therefore, there is an urgent need to design an examination system that can solve the shortcomings of existing technologies in terms of comprehensiveness, real-timeness and standard uniformity. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above problems and provide an AIGC key content review system based on NLP and authoritative text drive. To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] The AIGC key content review system driven by NLP and authoritative text includes an authoritative text collection module, a question generation and review module, a data management module, and an interactive execution module; the authoritative text collection module provides the original text to the question generation and review module through a standardized data interface; the question generation and review module outputs structured data to the data management module after NLP analysis and semantic review; the interactive execution module calls storage resources and performs asynchronous front-end and back-end interaction; the authoritative text collection module is used to obtain legal provisions, administrative regulations, central policy documents and officially designated historical readings to build a standardized text database; the data management module uses a structured storage format to record questions, answers, review conclusions and related metadata; the interactive execution module includes a front-end visual operation unit and a back-end data processing unit.

[0007] Furthermore, the question generation and review module includes a question generation module, a multi-level content review module and a human-computer collaborative optimization module; the question generation module converts text content into standardized test questions based on natural language processing technology; the multi-level content review module conducts security risk assessment on the generated content, and sets three review dimensions of political security, ideology, and factual accuracy; the human-computer collaborative optimization module provides a manual review interface to correct test questions and review results, and uses feedback data to optimize question generation rules.

[0008] Furthermore, the front-end visual operation unit includes a question setting interface, a question modification interface, a question screening interface and a question review interface.

[0009] The question-setting interface provides an authoritative text upload component, calls the back-end sentence segmentation interface to split the text into sentences, and supports manual tagging of key content tags.

[0010] The question modification interface displays a generated question list, allowing manual editing of question content, adjustment of question type and difficulty level.

[0011] The screening interface manages screening conditions through Vuex and dynamically loads questions that meet the requirements.

[0012] The question review interface calls a third-party generative AI model API to generate answers, and outputs a deviation score by comparing the answers with the authoritative text standard.

[0013] Furthermore, the specific implementation method of the front-end visual operation unit is as follows;

[0014] Step S1: Login and register, use the bcrypt algorithm to encrypt and store the password, and require the user to set an 8-digit password containing uppercase and lowercase letters;

[0015] Step S2: Upload and annotate sentences. Users upload authoritative texts and legal documents. The backend automatically segments the text into independent sentences for interface annotation and screening. After confirmation, the Qwen2.5-Instruct model is called through the integrated FlaskAPI to generate fill-in-the-blank questions, and the question preview interface is updated in real time through Vue.

[0016] Step S3: View and modify fill-in-the-blank questions, allowing users to edit or delete questions directly on the interface, and adjust their difficulty and labels;

[0017] Step S4: Filter questions. With the help of Vue's state management tool, users can flexibly select filtering conditions and quickly locate the required questions based on specific assessment needs.

[0018] Step S5: Generate answers. The user selects the AI answer mode and calls the API of the Class B generative artificial intelligence model to automatically generate answers and perform scoring.

[0019] Furthermore, the back-end data processing unit includes a Flask framework building module, a data storage and management module, a scoring system, and a model calling and optimization module; the Flask framework building module receives front-end requests through a RESTful API, passes the text and prompt words uploaded by the user to the model calling and optimization module, and stores the questions generated by the model in the data storage and management module; at the same time, it receives feedback results from the scoring system and updates the question scoring information in the data storage.

[0020] Furthermore, the Flask framework building module builds a RESTful API interface module based on the Python language, receives and processes front-end requests, and interacts with the business logic layer and data layer. It adopts a microservice architecture and is compatible with the Linux system environment, supporting high-concurrency request processing and modular function expansion.

[0021] Furthermore, the data storage and management module stores questions, answers, labels, difficulty and availability information in a structured manner based on the JSON file format, and provides a data addition, deletion, modification and query interface through the FlaskAPI to achieve lightweight transmission of real-time synchronization of front-end and back-end data.

[0022] Furthermore, the scoring system integrates the API call interface of the third-party LLM to score the answers according to the preset evaluation ideas.

[0023] Furthermore, the model call and optimization module integrates Qwen2.5-Instruct, generates questions through the Python interface, combines the Prompt template library designed with authoritative text, and configures a multi-round performance comparison test mechanism to optimize model selection.

[0024] The advantages of the present invention are:

[0025] 1. This invention uses an authoritative text collection module to acquire official texts such as legal provisions and central policy documents, building a standardized database. A question generation module based on NLP technology converts these texts into test questions and establishes a triple review dimension based on political security, ideology, and factual accuracy to ensure the authority of the review basis and the compliance of the content, addressing the limitations of traditional review methods that rely on non-authoritative data.

[0026] 2. This invention leverages the efficient interaction capabilities of a lightweight architecture, employing the Vue front-end framework and the Flask microservices architecture to achieve asynchronous front-end and back-end interaction via a RESTful API. The back-end stores structured data in JSON format, supports high-concurrency request processing, and the front-end dynamically loads questions, ensuring system scalability and responsiveness.

[0027] 3. This invention uses a human-computer collaboration module to provide a manual review interface for question correction and review results, and uses feedback data to optimize generation rules. The front-end visual interface supports manual editing of question content and adjustment of question difficulty, achieving a dynamic balance between algorithmic and human intervention, improving review flexibility and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings constituting a part of this application are used to provide a further understanding of this application and make other features, purposes and advantages of this application more apparent. The drawings and descriptions of the exemplary embodiments of this application are used to explain this application and do not constitute an improper limitation on this application.

[0029] In the attached figure:

[0030] Figure 1 This is a model design block diagram of the AIGC key content review system based on NLP and authoritative text drive in Example 1.

[0031] Figure 2 This is a block diagram of the front-end and back-end interaction design in the interactive execution module in Example 1.

[0032] Figure 3 This is a web page diagram developed by Vue for the front-end visual operation unit in Example 1.

[0033] Figure 4 This is the specific process of the front-end visual operation unit system in Example 1.

[0034] Figure 5 This is the login and registration page for the front-end visual operation unit in Example 1.

[0035] Figure 6 This is the new user registration interface of the front-end visual operation unit in Example 1.

[0036] Figure 7 This is the user login interface of the front-end visual operation unit in Example 1.

[0037] Figure 8 This is the sentence uploading and annotation interface of the front-end visual operation unit in Example 1.

[0038] Figure 9 Modify the title interface of the front-end visual operation unit in Example 1.

[0039] Figure 10 This is the question screening interface of the front-end visual operation unit in Example 1.

[0040] Figure 11 Generate an answer interface for the front-end visual operation unit in Example 1.

[0041] Figure 12 This is a diagram of the Flask framework built by the back-end data processing unit in Example 1.

[0042] Figure 13 This is the design and optimization diagram of the back-end data processing unit Prompt in Example 1.

[0043] Figure 14 This is the testing, maintenance and experimental means framework in Example 1. DETAILED DESCRIPTION

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0045] The present invention is described in detail and specifically below through specific examples to provide a better understanding of the present invention. However, the following examples do not limit the scope of protection of the present invention.

[0046] Example 1

[0047] like Figure 1-14 As shown, the AIGC key content review system driven by NLP and authoritative text includes an authoritative text collection module, a question generation and review module, a data management module and an interactive execution module; the authoritative text collection module provides the original text to the question generation and review module through a standardized data interface; the question generation and review module outputs structured data to the data management module after NLP parsing and semantic review; the interactive execution module calls storage resources and performs front-end and back-end asynchronous interaction; the authoritative text collection module is used to obtain legal provisions, administrative regulations, central policy documents and officially designated historical readings to build a standardized text database; the data management module uses a structured storage format to record questions, answers, review conclusions and related metadata; the interactive execution module includes a front-end visual operation unit and a back-end data processing unit.

[0048] Furthermore, the question generation and review module includes a question generation module, a multi-level content review module, and a human-computer collaborative optimization module; the question generation module converts text content into standardized test questions based on natural language processing technology; the multi-level content review module conducts security risk assessments on the generated content, setting three review dimensions: political security, ideology, and factual accuracy; the human-computer collaborative optimization module provides a manual review interface to correct test questions and review results, and uses feedback data to optimize question generation rules;

[0049] Furthermore, the front-end visual operation unit includes a question setting interface, a question modification interface, a question screening interface, and a question review interface;

[0050] The question setting interface provides an authoritative text upload component, calls the backend sentence segmentation interface to split the text into sentences, and supports manual tagging of key content tags;

[0051] The question modification interface displays a list of generated questions, allowing manual editing of question content, adjustment of question type and difficulty level;

[0052] The screening interface manages screening conditions through Vuex and dynamically loads questions that meet the requirements;

[0053] The question review interface calls a third-party generative AI model API to generate answers, and outputs a deviation score by comparing the answers with the authoritative text standard.

[0054] Furthermore, the specific implementation method of the front-end visual operation unit is as follows;

[0055] Step S1: Login and register. The bcrypt algorithm is used to encrypt and store the password. The user is required to set an 8-digit password containing uppercase and lowercase letters.

[0056] Step S2: Upload and annotate sentences. Users upload authoritative texts and legal documents. The backend automatically divides the text into independent sentences for interface annotation and screening. After confirmation, the Qwen2.5-Instruct model is called through the integrated FlaskAPI to generate fill-in-the-blank questions, and the question preview interface is updated in real time through Vue.

[0057] On this page, users can upload original materials such as authoritative texts, legal and regulatory documents, etc. The webpage backend will automatically read the file, segment the text, and divide the text into several sentences. Then, users can use the interface to select the sentences for which questions need to be generated and mark them for subsequent screening. After marking, users can confirm the generated questions. The frontend integrates the backend FlaskAPI and sends these tags and uploaded content to the backend, and the Qwen2.5-Instruct model generates fill-in-the-blank questions. Vue's responsiveness ensures that the page is updated immediately after the questions are generated, making it convenient for users to immediately view the generated results.

[0058] Step S3: View and modify fill-in-the-blank questions, allowing users to edit or delete questions directly on the interface, and adjust their difficulty and labels.

[0059] Although the backend has achieved automatic generation of questions and answers, and can automatically determine their difficulty and question labels, in order to ensure the accuracy of questions and related data, the system still allows users to make manual modifications. The page presents a list of generated fill-in-the-blank questions, and users can directly edit or delete questions on this interface, and adjust their difficulty and labels. Through the RESTful API provided by Flask, data can flow in both directions, allowing users to manage question content in real time and flexibly change question labels for subsequent screening operations. Compared with traditional server rendering methods, this approach significantly improves the flexibility and response speed of interaction. This feature provides a solid foundation for testing and evaluating the accuracy and consistency of different generative AI.

[0060] Step S4: Filter questions. With the help of Vue's state management tool, users can flexibly select filtering conditions and quickly locate required questions based on specific evaluation requirements.

[0061] After ensuring the correctness of the questions, users can filter questions based on their needs on this page. This page provides a variety of filtering criteria, such as the number of questions, tag percentage, and difficulty percentage. Users can flexibly select appropriate filtering criteria based on their specific assessment needs to quickly locate the required questions. With the help of Vue's state management tools, such as Vuex, filtering operations can be completed efficiently and responsively. This dynamic filtering state management method avoids the drawbacks of traditional form submissions or repeated page loading, significantly improving the convenience of question bank management.

[0062] Step S5: Generate answers. The user selects the AI answer mode and calls the API of the Class B generative artificial intelligence model to automatically generate answers and perform scoring.

[0063] On this page, users can select an AI response mode and call the API of the Class B generative AI model to automatically generate and score answers based on the generated questions. To ensure the comprehensiveness and accuracy of the evaluation, this system provides a detailed set of evaluation dimensions and indicators, according to which users can evaluate the Class B model's answers item by item. This page is designed to simplify the operation process, allowing users to quickly complete the entire operation chain from question generation to answer generation and evaluation.

[0064] The front-end application uses Vue's state management (Vuex) and routing management (Vue Router), enabling smooth page switching and dynamic data updates. On the question filtering page, users can quickly retrieve qualified questions using a variety of filtering criteria. Vuex effectively manages the filtering state, ensuring data consistency and instant presentation of filtering results, improving system responsiveness.

[0065] Furthermore, the back-end data processing unit includes a Flask framework building module, a data storage and management module, a scoring system, and a model calling and optimization module; the Flask framework building module receives front-end requests through a RESTful API, passes the text and prompt words uploaded by the user to the model calling and optimization module, and stores the questions generated by the model in the data storage and management module; at the same time, it receives feedback results from the scoring system and updates the question scoring information in the data storage.

[0066] Furthermore, the Flask framework building module builds a RESTful API interface module based on the Python language, receives and processes front-end requests, and interacts with the business logic layer and data layer. It adopts a microservice architecture and is compatible with the Linux system environment, supporting high-concurrency request processing and modular function expansion.

[0067] The backend, built on Flask, is a RESTful API that receives and processes frontend requests and interacts with the business logic and data layers. As a lightweight Python web framework, Flask is compatible with all Python syntax, greatly facilitating code writing and improving code portability. Compared to more complex frameworks like Django, Flask offers greater flexibility, allowing for the inclusion of extension libraries as needed. Its modularity and high flexibility allow the project team to freely design the backend architecture within a Linux environment and flexibly add functional modules based on project requirements.

[0068] Furthermore, the data storage and management module uses the JSON file format to structuredly store questions, answers, tags, difficulty, and availability information. It also provides data addition, deletion, modification, and query interfaces through the Flask API, enabling lightweight transmission of front-end and back-end data with real-time synchronization. Flask's ease of use makes it ideal for quickly building and testing API interfaces, especially for generative AI project backends that require flexible design. Furthermore, Flask's microservice architecture and compatibility with Linux environments enable it to maintain efficient operation even with limited system resources, helping to optimize system performance and reduce server load. This lightweight design provides a solid technical foundation for the project's subsequent system scalability and maintainability.

[0069] Project information, including questions, answers, tags, difficulty, and availability, is stored in JSON files. The JSON file format is clearly structured and easy to read, while also offering the advantage of being lightweight for data transmission. Furthermore, the backend manages questions through operations related to JSON files, including adding, editing, and deleting questions. Requests made through the Flask API and UI trigger backend responses, synchronously updating the question data in the JSON file.

[0070] Furthermore, the scoring system integrates the API call interface of the third-party LLM to score the answers according to the preset evaluation ideas.

[0071] On the "Generate Answers" page, the backend calls the third-party LLMs API to generate answers and scores them based on a pre-defined evaluation process. During the quiz phase, professionals conduct simultaneous manual evaluations, which are then used to optimize the model's assessments. The model's responses are then optimized based on these evaluations, achieving efficient review.

[0072] Furthermore, the model calling and optimization module integrates Qwen2.5-Instruct, generates questions through the Python interface, combines the prompt template library designed with authoritative texts, and configures a multi-round performance comparison test mechanism to optimize model selection. The back-end uses Python language to call the model. The Qwen2.5-Instruct model has been specially trained for authoritative texts such as the Four Histories to ensure that the generated questions have high accuracy and practicality. The integration process begins with the user uploading the text to be tested through the front-end interactive interface. Subsequently, the system will build and send prompt words (Prompt) containing specific requirements to the Qwen2-Instruct model. When the model receives this information, it will analyze the uploaded text and generate corresponding questions based on the prompts provided. These generated questions will be stored in JSON format for subsequent viewing, modification and management.

[0073] Prompt design is key to ensuring high-quality, compliant questions. Since generated questions are used for security reviews and assessments, Prompt design must be optimized for text characteristics to ensure the generated content is both professional and practical. The following are key design elements:

[0074] (1) The accuracy and professionalism of terminology to ensure the professionalism and standardization of the topic.

[0075] (2) Guidance by contextual information or examples can be used to clearly instruct the model in the prompt words to "extract key concepts from the following legal provisions to generate fill-in-the-blank questions", thereby strengthening the model's understanding ability and enabling it to generate questions that are more in line with needs and expectations.

[0076] (3) Diversity and challenge: Prompt design needs to consider the diversity and difficulty of generated questions. By setting different prompt combinations, the generation of repetitive and low-difficulty questions can be effectively avoided.

[0077] This system utilizes a separate front-end and back-end architecture. The front-end is based on the Vue framework, and the back-end uses Flask to build a RESTful API. Vue is responsible for presenting the user interface and data interaction, while Flask handles core business logic and model call requests. This separation not only improves development efficiency but also supports independent deployment of the front-end and back-end, facilitating future maintenance and expansion. Furthermore, the front-end communicates with the back-end in real time via the API, achieving efficient data transmission and rapid interface response, ensuring system stability and flexibility.

[0078] It should be noted that this embodiment also includes testing and maintenance and experimental means.

[0079] 1. Front-end page rendering test

[0080] (1) Performance testing

[0081] Evaluate front-end loading speed and interactive experience to ensure a smooth user experience when viewing and modifying authoritative text-generated titles. Use tools such as Lighthouse to monitor key performance indicators (KPIs) such as First Paint (FP), Largest Contentful Paint (LCP), and Time to Interactive (TTI). Compare data across different browsers and devices to ensure system compatibility in various environments, especially ensuring efficient response when processing legal texts.

[0082] (2) User Interface Testing (UIT)

[0083] Verify the effectiveness of front-end design and functionality, and optimize user interaction with generative AI. Through user interviews and automated testing (such as Selenium), simulate user interaction with the test interface to ensure that functionality is accurate and easy to use.

[0084] 2. Backend server function test

[0085] (1) Functional testing: Ensure that the backend API is stable and reliable, and supports efficient question generation and screening. Use Postman or Swagger to test the API, verify the correctness and stability of the response data, and ensure that requests from the frontend can be processed smoothly.

[0086] (2) Performance testing: Evaluate the backend's performance under high concurrency to ensure the system remains stable under a large number of requests. Use tools such as LoadRunner or JMeter to simulate high-concurrency requests, monitor response time and system load, and optimize processing performance.

[0087] 3. Testing of Generative Models

[0088] (1) Model call test: Verify the call success rate of the generated model and the accuracy of the data returned to ensure that the generated questions meet expectations. Conduct multiple rounds of testing, record the success rate and response time of each call, and ensure that the output is consistent with the legal text content.

[0089] (2) Answer generation testing: Ensure the accuracy and compliance of generated answers, and verify the legal compliance of generated content through expert review and automated testing.

[0090] (3) System security testing: Identify potential vulnerabilities through static and dynamic security analysis, and conduct penetration testing, such as SQL injection and XSS attacks, to ensure the security of data transmission.

[0091] (4) Model performance comparison test: Compare the performance and quality of different generative models, evaluate the consistency and efficiency of their generated answers, and select the best performing model.

[0092] (5) Scoring criteria testing: Ensure the fairness and accuracy of the scoring criteria, design a standardized scoring framework, collect review results, conduct statistical analysis, and ensure the consistency between automatic scoring and manual scoring.

[0093] (6) System Iteration Optimization Testing: Based on user feedback and test results, the system functions are continuously optimized, including the front-end interface, back-end processing, and model updates, to ensure that the system can flexibly adapt to changes in legal texts and continuously improve overall performance.

[0094] While the specific embodiments of the present invention have been described in detail above, they are merely exemplary, and the present invention is not equivalent to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions of the present invention are also within the scope of the present invention. Therefore, equivalent changes and modifications made without departing from the spirit and scope of the present invention are intended to be encompassed within the scope of the present invention.

Claims

1. AIGC key content review system based on NLP and authoritative text, characterized by: It includes an authoritative text collection module, a question generation and review module, a data management module and an interactive execution module; the authoritative text collection module provides the original text to the question generation and review module through a standardized data interface; the question generation and review module outputs structured data to the data management module after NLP analysis and semantic review; the interactive execution module calls storage resources and performs asynchronous front-end and back-end interaction; the authoritative text collection module is used to obtain legal provisions, administrative regulations, central policy documents and officially designated historical readings to build a standardized text database; the data management module uses a structured storage format to record questions, answers, review conclusions and related metadata; the interactive execution module includes a front-end visual operation unit and a back-end data processing unit.

2. The AIGC key content review system based on NLP and authoritative text drive according to claim 1 is characterized by: The question generation and review module includes a question generation module, a multi-level content review module and a human-computer collaborative optimization module; the question generation module converts text content into standardized test questions based on natural language processing technology; the multi-level content review module conducts security risk assessment on the generated content, and sets three review dimensions of political security, ideology, and factual accuracy; the human-computer collaborative optimization module provides a manual review interface to correct test questions and review results, and uses feedback data to optimize question generation rules.

3. The AIGC key content review system based on NLP and authoritative text drive according to claim 1 is characterized by: The front-end visual operation unit includes a question setting interface, a question modification interface, a question screening interface, and a question review interface: The question setting interface provides an authoritative text upload component, calls the backend sentence segmentation interface to split the text into sentences, and supports manual tagging of key content tags; The question modification interface displays a list of generated questions, allowing manual editing of question content, adjustment of question type and difficulty level; The screening interface manages screening conditions through Vuex and dynamically loads questions that meet the requirements; The question review interface calls a third-party generative AI model API to generate answers, and outputs a deviation score by comparing the answers with the authoritative text standard.

4. The AIGC key content review system based on NLP and authoritative text drive according to claim 3 is characterized by: The specific implementation method of the front-end visual operation unit is as follows: Step S1: Login and register, use the bcrypt algorithm to encrypt and store the password, and require the user to set an 8-digit password containing uppercase and lowercase letters; Step S2: Upload and annotate sentences. Users upload authoritative texts and legal documents. The backend automatically segments the text into independent sentences for interface annotation and screening. After confirmation, the Qwen2.5-Instruct model is called through the integrated FlaskAPI to generate fill-in-the-blank questions, and the question preview interface is updated in real time through Vue. Step S3: View and modify fill-in-the-blank questions, allowing users to edit or delete questions directly on the interface, and adjust their difficulty and labels; Step S4: Filter questions. With the help of Vue's state management tool, users can flexibly select filtering conditions and quickly locate the required questions based on specific assessment needs. Step S5: Generate answers. The user selects the AI answer mode and calls the API of the Class B generative artificial intelligence model to automatically generate answers and perform scoring.

5. The AIGC key content review system based on NLP and authoritative text drive according to claim 4 is characterized by: The back-end data processing unit includes a Flask framework building module, a data storage and management module, a scoring system, and a model calling and optimization module; the Flask framework building module receives front-end requests through a RESTful API, passes the text and prompt words uploaded by the user to the model calling and optimization module, and stores the questions generated by the model in the data storage and management module, while receiving feedback results from the scoring system and updating the question scoring information in the data storage.

6. The AIGC key content review system based on NLP and authoritative text drive according to claim 5 is characterized by: The Flask framework building module builds a RESTful API interface module based on the Python language, receives and processes front-end requests, and interacts with the business logic layer and data layer. It adopts a microservice architecture and is compatible with the Linux system environment, supporting high-concurrency request processing and modular function expansion.

7. The AIGC key content review system based on NLP and authoritative text drive according to claim 6 is characterized by: The data storage and management module stores questions, answers, tags, difficulty and availability information in a structured manner based on the JSON file format, and provides a data addition, deletion, modification and query interface through the Flask API to achieve lightweight transmission of front-end and back-end data in real-time synchronization.

8. The AIGC key content review system based on NLP and authoritative text drive according to claim 7 is characterized by: The scoring system integrates the API call interface of the third-party LLM and scores the answers according to the preset evaluation ideas.

9. The AIGC key content review system based on NLP and authoritative text drive according to claim 8 is characterized by: The model calling and optimization module integrates Qwen2.5-Instruct, generates questions through a Python interface, combines a prompt template library designed with authoritative text, and configures a multi-round performance comparison test mechanism to optimize model selection.

Images