Teaching auxiliary system based on large model Agent and database multi-technology fusion

Abstract

The invention discloses a teaching assistance system based on large model Agent and database multi-technology fusion, which at least comprises a large model application design module, a database information management module and a system interaction design module, and is characterized in that an RAG (Retrieval Augmented Generation) technology and a LoRA (Low-Rank Adaptation, LoRA) technology are applied to teaching assistance, knowledge transfer and intelligent interaction; especially for intelligent processing of educational resources and implementation of personalized teaching, the system can be widely applied to the educational field, and is beneficial to improving the teaching quality and efficiency in a more efficient and convenient manner, helping teachers and students to overcome the limitation of a traditional teaching mode, better performing teaching and learning interaction, and improving the teaching quality and efficiency. Knowledge spreading and skill improvement are promoted, education modernization is promoted, and the system and the method jointly stride towards the future of intelligent education.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and is particularly suitable for intelligent teaching and learning assistance in the education industry. It mainly relates to a teaching assistance system based on the fusion of large model Agent and database multiple technologies. Background Art

[0002] Although the advent of the information age has brought abundant learning resources, it has also increased the difficulty of screening effective information and placed higher demands on students' information processing capabilities.

[0003] Through intelligent means, teachers' workload can be reduced and their work efficiency can be improved; at the same time, personalized learning support and rich learning experience can be used to stimulate students' learning interest and cultivate their ability to learn independently. Such a transformation will help build a more efficient, fair and vibrant education ecosystem.

[0004] The large model, full name is large-scale pre-trained language model, is a deep neural network model obtained by unsupervised or self-supervised learning on massive text data. The model has learned a wealth of language knowledge and patterns through pre-training on massive text data. They have strong natural language understanding and generation capabilities and can handle a variety of complex language tasks, such as text generation, question answering, translation, sentiment analysis, etc. The model has strong adaptive capabilities and can dynamically adjust its output according to the different needs of users to provide personalized services. It can also continuously optimize its own performance by continuously learning user feedback and new data.

[0005] Since big models can analyze user needs, understand user needs and generate content, they are gradually being applied to the field of education. When big models are used in teaching assistance scenarios, they can automatically generate teaching materials such as lesson plans, PPTs, exercises, etc., helping teachers save a lot of time; they can provide personalized learning suggestions and resources based on students' learning history and performance to improve students' learning efficiency; and they can answer various questions encountered by students in the learning process through an intelligent question-and-answer system, providing timely answers and help.

[0006] An agent is a general problem solver based on a large language model. It is a system or program with autonomous learning, decision-making and execution capabilities. It can automatically perform tasks and adapt to changes in a specific environment. An agent based on a large model is an intelligent system built using a large pre-trained model. The model has strong language understanding and generation capabilities and can perform well in a variety of tasks. This type of agent can not only handle complex conversations, but also perform a variety of functions such as information retrieval, content generation, and data analysis. Due to the characteristics of agents, they can play an important role in the field of teaching assistance. They can provide personalized tutoring to help students solve difficult problems and improve learning efficiency; assist teachers in formulating teaching plans and generating teaching materials to reduce teachers' work pressure; by analyzing students' learning data, agents can provide students with valuable learning-related suggestions and increase students' ways to improve themselves.

[0007] Database technology is the key to our realization of knowledge base system and vector database.

[0008] The knowledge base system can provide teachers with a platform for centralized management and sharing of various types of teaching materials (such as Word documents, PDF files, PPT presentations, etc.), solving the inconvenience caused by scattered storage of materials; it can help teachers quickly find teaching resources that meet specific needs and improve work efficiency; it can also promote knowledge exchange and cooperation between teachers and students, allowing the sharing of personal notes and teaching materials; the knowledge base system that supports uploading, reviewing and updating of materials can also ensure that the teaching content can continue to iterate with the development of educational concepts and technologies, and always keep the latest status.

[0009] The vector database processes multimedia content such as text, images, and voice in a vectorized manner to achieve efficient retrieval based on similarity. With the teacher's prompt, the required teaching materials can be automatically found quickly and accurately from massive data. It not only manages data in different forms (such as text, images, and audio) in a unified manner, but also provides an efficient and consistent search experience. Combined with large model technology, the data in the vector database can be used to generate high-quality teaching documents, such as using RAG technology to retrieve relevant information from the knowledge base and generate lesson plans or PPTs that meet teaching needs, thereby significantly improving teaching efficiency and quality. Summary of the invention

[0010] The present invention aims at solving the problems in the prior art and provides a teaching assistance system based on the fusion of large model Agent and database technologies. The solution focuses more on the experience of educators rather than just answering questions for students. The present invention can promote the exchange of resources, reduce the burden on educators, and enrich educational resources with AI, which has social and practical benefits.

[0011] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a teaching auxiliary system based on the fusion of large model Agent and database multi-technology, which at least includes a large model application design module, a database information management module, and a system interaction design module.

[0012] The large model application design module is as follows: a dedicated "Wenwu" large model is designed, which is based on a powerful neural network and processes natural language data or image data. After fine-tuning the relevant data of the teaching materials, it can accurately answer the professional knowledge questions input by teachers and students, and can also organize a large amount of knowledge to generate the content of the teaching materials; the Agent personalizes the call of the "Wenwu" large model, and completes intelligent marking and question judging, and teaching material file generation based on the output of the large model.

[0013] The database information management module is used to manage user identity authority information and teaching material information, maintain the user's identity information and other basic information, the basic information of the teaching material and the review status information, accept the operation information of the system interaction design module to update the data table in real time, and provide the system interaction design module with the URL of the teaching material to be downloaded for subsequent user operations; it is divided into private teaching material library and public teaching material library according to the open permissions of the teaching materials, and stores accessible user information and related teaching material information, and accepts the permission operations of the system interaction design module users to update the library access permissions.

[0014] The system interaction design module designs system visualization and interaction functions, including user application or consent to private library access operations, administrator review of uploaded teaching materials operations, user information display and interface interaction beautification, etc., in order to facilitate user interaction, obtain user operations and transmit operation results to the large model application design module and database information management module.

[0015] As an improvement of the present invention, the large model application design module can be divided into a knowledge question answering submodule, a teaching material generation submodule and a large model adjustment management submodule:

[0016] The knowledge question answering submodule: A user who is a teacher or a student inputs his or her own question, and the "Wenwu" model first determines whether the question is related to the knowledge in the knowledge base. If so, it performs search enhancement generation (RAG) in the knowledge base to output the answer. If not, it directly outputs the answer based on the user's "learned knowledge";

[0017] The teaching material generation submodule: users who are teachers can upload various types of teaching materials and ask the "Wenwu" model to generate other types of corresponding teaching materials according to the files of specific teaching materials. They can also input handwritten mathematical expressions and other contents in the form of pictures to obtain corresponding LaTeX codes for assisting in the editing of teaching materials.

[0018] The large model adjustment management submodule includes preview command function, save training parameter and load training parameter functions, start fine-tuning and interrupt fine-tuning functions, supports large model selection (provides a list of popular open source large models, and the local large model can enter the model path), fine-tuning dataset selection (supports dataset preview, supports constructing new datasets from local), training parameter and LoRA parameter settings, output directory settings, real-time viewing of progress, loss curves and training logs during the fine-tuning process, and supports loading checkpoint paths and conducting question-and-answer tests after fine-tuning.

[0019] As another improvement of the present invention, the database information management module can be divided into a teaching material management submodule and a permission management submodule:

[0020] The teaching material management submodule: receives the result of the content review management submodule passing or failing the operation, and updates the review status of the existing teaching materials in the teaching material library in real time. Users can upload various types of teaching materials to private or public teaching material libraries, and can also download teaching materials from private, shared or authorized private teaching material libraries. The teaching materials generated using the existing materials are automatically stored as new teaching materials in the private teaching material library;

[0021] The authority management submodule receives the identity information of the user requesting access, provides the user requesting access with operation result information according to whether the user of the private library is allowed to access the private library, and updates the relevant teaching material library opening authority.

[0022] As another improvement of the present invention, the system interaction design module specifically includes a content review submodule and a permission operation submodule:

[0023] The content review submodule: a user who is an administrator receives all teaching material contents of the teaching material management module, and reviews the rationality and legality of the material contents by checking downloads, etc., makes a pass or fail operation and returns the operation result to the teaching material management submodule;

[0024] The permission operation submodule: the user can send application information to the private library user through the application access operation; the user who receives the application access information can choose whether to agree, and the access result information will be synchronously transmitted to the user who applied for access.

[0025] As another improvement of the present invention, the large model application design module includes large model LoRA fine-tuning technology and large model Agent technology:

[0026] The large model LoRA fine-tuning technology: The large model LoRA fine-tuning technology is an efficient parameter adjustment method that reduces the number of parameters that need to be trained by decomposing the model weights into the product form of two low-rank matrices. This technology is used to customize the teaching assistant of the large model to adapt to specific teaching tasks, such as generating PPT content from the course outline, converting handwritten mathematical expressions into LaTeX code, and automatically generating and grading subject exercise content. The specific implementation steps include downloading and formatting the data set, selecting a suitable large model, decomposing the model weights into the product form of two low-rank matrices, performing fine-tuning, etc.

[0027] The big model agent technology: Agent calls the big model and uses the output of the big model to provide further support for teachers and students by combining the four major elements of planning, memory, tool use and execution. When a teacher needs to generate a lesson plan or PPT, the agent first understands this demand and breaks it down into a series of steps, such as retrieving relevant materials and generating a document framework. Then, the agent uses the information in the knowledge base and text generation tools to create teaching materials that meet the requirements. In addition, the agent can also process images, such as converting handwritten formulas into LaTeX code, to help teachers prepare course content more efficiently. Throughout the process, the agent will continuously optimize services based on the user's historical interaction records and personal preferences to ensure that the teaching resources provided are more in line with actual needs.

[0028] As another improvement of the present invention, the large model application design module has preset the "Wenwu" test-taking and grading model obtained by fine-tuning the Qwen large model using the MR-Ben data set and the LoRA technology, such as Figure 4 As shown:

[0029] The fine-tuning process of the test-taking and grading model is as follows: First, the MR-Ben dataset specially adapted for test-taking and grading tasks is prepared and parsed, and then based on the LoRA core formula h = W 0 x+ΔWx=W 0 x+BAx, while keeping most of the parameters of the pre-trained model unchanged, only introduces low-rank matrix updates for specific layers. After a carefully designed training process, including data preprocessing, defining LoRA settings, iterative optimization and other steps, the Qwen model is adapted to this complex reasoning evaluation task;

[0030] The structure of the test-taking and grading model: the input format of the model is a structured text sequence, including but not limited to the description of the question, the candidate's answer process and step-by-step instructions. Specifically, each input instance is constructed into a composite structure containing instructions, context, questions and candidate answers. For the output format, the model generates a scoring result and corresponding comments. The scoring result is a quantitative assessment of the accuracy and completeness of the candidate's answers, usually presented in numerical form. The comments are a qualitative analysis of the problem-solving process, pointing out the advantages and disadvantages, and making suggestions for improvement.

[0031] As another improvement of the present invention, in the teaching material management module, the uploading of teaching materials including but not limited to teaching syllabuses, teaching requirements, lesson plans, study guides, teaching PPTs, after-class exercises and other types of teaching materials is supported.

[0032] In order to achieve the above object, the present invention also adopts a technical solution: a method for using the system includes the following steps:

[0033] S1, Registration and Login: When using this system for the first time, the user needs to fill in the necessary system information including identity, user name, account number, password, name, ID number and other information of the national real-name system. When it is not the first time to use this system, the user can log in with the account number and password. The system will switch to different main interfaces according to the identity information of the account.

[0034] S2, intelligent question answering: the user inputs his own questions, and the big model first determines whether the questions are related to the knowledge in the knowledge base. If so, it performs retrieval enhanced generation (RAG) in the knowledge base to output the answer. If not, it directly outputs the answer based on the user's "learned knowledge".

[0035] S3, intelligent generation: when generating teaching material files, the user who is a teacher first uploads a certain type of teaching material as the knowledge base of the large model retrieval enhanced generation (RAG), and then enters the requirements, that is, what type of corresponding teaching material the large model needs to generate; when generating LaTeX code, the user who is a teacher first enters handwritten mathematical expressions and other content in image format, and then previews it after obtaining the corresponding LaTeX code. If necessary, the LaTeX code can be manually modified online.

[0036] S4, Smart Search: Users who are teachers or students can click on the "Smart Search" in the main interface function area to perform search operations. By entering keywords in the input box, the system can retrieve all files containing the keywords from public libraries and accessible private libraries and display them for users to download and review.

[0037] S5, Knowledge base viewing: Users who are teachers or students can click "My Knowledge Base" in the function area of ​​the main interface to enter their personal private knowledge base to view all uploaded files, including the file's review status, upload time and other information. They can also click the "Upload" button to continue uploading local teaching materials to the knowledge base for management and choose whether to make them public as public library files. Users can enter the account number of the user they want to access in the input box to apply for access to the user's private knowledge base. For the knowledge base that has been approved, the user can directly view all file contents in the knowledge base.

[0038] S6, large model tuning: users with the identity of administrators need to first select the large model, then select the fine-tuning dataset (the dataset can be previewed before selection), set the training parameters and LoRA parameters and the output directory, and then click the "Start" button to start fine-tuning. When the fine-tuning is completed, select the output directory of the just fine-tuned model in the checkpoint path, and then enter the chat area for question-and-answer testing.

[0039] S7, Data Review: Users with the identity of administrators can view all files in all libraries, review and judge the contents of the files, and make "pass / fail" operations to change the file review status. The system automatically publishes relevant information to notify the user to whom the file belongs.

[0040] Compared with the prior art, the present invention has the following beneficial effects: the present invention combines the database technology of personal knowledge base and public knowledge base with multiple large models for the first time, and builds a platform for global educators to share educational resources. The system of the present invention can not only create its own knowledge base, but also choose to disclose part of it to the public knowledge base. At the same time, users can retrieve the content disclosed by others to realize the exchange of educational resources. Users can download these files or select these contents. The system can write teaching materials such as lesson plans, PPTs, and test papers based on these contents, reducing the burden on educators.

[0041] In addition, the present invention also introduces an intelligent marking system, which can recognize handwritten content and convert it into characters, and upload it to a large model for judgment. This can greatly facilitate educators and realize the full AI process from question setting to marking. In addition, in order to facilitate user use, the system adds a tool library, which combines commonly used functions to complete all teaching tasks in one system.

[0042] Compared with other large teaching assistance models, the system of the present invention pays more attention to the experience with educators rather than just answering questions for students. The present invention can promote the exchange of resources, reduce the burden on educators, and enrich educational resources with AI, which has social and practical benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1It is a structural architecture diagram of the system of the present invention;

[0044] Figure 2 It is a structural architecture diagram of each submodule of the system of the present invention;

[0045] Figure 3 A flowchart of the steps of the method for using the system of the present invention;

[0046] Figure 4 This is a large model architecture diagram of the system "Wenwu" of the present invention;

[0047] Figure 5 This is a schematic diagram of the interface after a user with the user identity of embodiment 2 of the present invention logs in;

[0048] Figure 6 This is a schematic diagram of the interface after a user with an administrator identity logs in according to Example 3 of the present invention. DETAILED DESCRIPTION

[0049] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.

[0050] Example 1

[0051] The teaching auxiliary system based on the integration of large model Agent and database multi-technology is characterized by: at least including large model application design module, database information management module, system interaction design module,

[0052] The large model application design module is as follows: a dedicated "Wenwu" large model is designed, which is based on a powerful neural network and processes natural language data or image data. After fine-tuning the relevant data of the teaching materials, it can accurately answer the professional knowledge questions input by teachers and students, and can also organize a large amount of knowledge to generate the content of the teaching materials; the Agent personalizes the call of the "Wenwu" large model, and completes intelligent marking and question judging, and teaching material file generation based on the output of the large model.

[0053] The database information management module is used to manage user identity authority information and teaching material information, maintain the user's identity information and other basic information, the basic information of the teaching material and the review status information, accept the operation information of the system interaction design module to update the data table in real time, and provide the system interaction design module with the URL of the teaching material to be downloaded for subsequent user operations; it is divided into private teaching material library and public teaching material library according to the open permissions of the teaching materials, and stores accessible user information and related teaching material information, and accepts the permission operations of the system interaction design module users to update the library access permissions.

[0054] The system interaction design module designs system visualization and interaction functions, including user application or consent to private library access operations, administrator review of uploaded teaching materials operations, user information display and interface interaction beautification, etc., in order to facilitate user interaction, obtain user operations and transmit operation results to the large model application design module and database information management module.

[0055] Example 2

[0056] The teaching assistance system based on the integration of large model Agent and database multi-technology, the user identity in this embodiment is the user, and the function can be realized through the user main interface, the user main interface is as follows Figure 5 This embodiment includes a large model application design module, a database information management module, and a system interaction design module. Figure 1 shown.

[0057] (1) When teachers need to use the intelligent question-and-answer function, they can click to enter the "Intelligent Question-and-Answer" function area, enter the code or question in the question area, and select "Code Modification" or "Knowledge Question-and-Answer" to allow the big model to give more professional and more appropriate answers. If further questions are needed, they can directly "reference" the previous chat records to "activate" the big model's memory and conduct multiple rounds of dialogue. During the process of the big model streaming out answers, they can also stop answering at any time and propose improvement requirements in the question area;

[0058] (2) When teachers need to use the data search function, they can click to enter the "Data Search" function area, enter a complete or partial keyword, and then retrieve a list of all the data related to the keyword in the public knowledge base and their own personal knowledge base, and then download it to the local computer for storage and viewing;

[0059] (3) When teachers need to use the intelligent generation of teaching materials function, they click to enter the "Intelligent Generation" function area, check "Yes" for "Add to Knowledge Base", click the "Upload File" button, select other types of materials related to the teaching materials they want to generate locally, upload them to the system to serve as the knowledge base for large model retrieval enhancement generation, then select "Generate File Type", and finally enter the detailed requirements for the generated materials and click the Send button. After the materials are successfully generated, a file download box will pop up on the page. Click Download to save the materials locally;

[0060] (4) When teachers need to use the intelligent generation of teaching materials function, they can click to enter the "Intelligent Generation" function area, then click the "Convert Handwritten Formula to LaTeX Formula" button, and then select and upload pictures of handwritten mathematical expressions and other contents locally to obtain the corresponding LaTeX code, which can be previewed online. If necessary, the LaTeX code can also be manually modified online;

[0061] (5) When teachers need to use functions related to their personal knowledge base, they can click to enter the "My Knowledge Base" function area to view the review status of the materials they have uploaded. They can click "Download" to obtain the materials in their personal knowledge base that have passed the review. They can click the "Upload File" button to select new materials from their local computer to upload.

[0062] (6) When teachers need to use the tool library, they can click to enter the "My Tool Library" function area and choose tools such as professional calculator, library collection query, CNKI search, Markdown editor, MySQL database tool, idiom dictionary, etc.

[0063] Example 3

[0064] Based on the teaching assistance system of large model agent and database multi-technology integration, the user identity in this embodiment is the administrator, and the functions can be realized through the administrator main interface. Figure 6 This embodiment includes a large model application design module, a database information management module, and a system interaction design module. Figure 1 shown.

[0065] (1) The administrator clicks on the "Home Page" on the left to enter the system management module. In this section, the administrator can see the number of registered users on the platform, the number of knowledge base files, the number of online users on the platform, cumulative revenue, today's revenue, etc.

[0066] (2) The administrator can click "Data Review" on the left to enter the content review management module. The administrator can see the files uploaded by the user for the first time and review whether they violate the rules. The administrator can see all the files marked as "unreviewed" on this interface. The administrator can click the download button and select "pass" or "fail" to mark the file after reading it.

[0067] (3) If the administrator wants to check the knowledge base, he can click "Knowledge Base Management". The administrator can see the file name, uploader, download address, file status, whether the file belongs to a public knowledge base or a private knowledge base, and other related information. The administrator can click the download button to download the file, or click "Pass" or "Fail" to directly modify the file status.

[0068] (4) If the administrator wants to operate the large model, he can click "Large Model Management" to enter the large model management page. The administrator can click "Upload" and select a file or knowledge base from the local computer to upload, so as to adjust the parameters of the large model.

[0069] (5) If you are a root administrator, you can click "Permission Management" to enter the permission management module. The root administrator can see other administrator accounts and click "Audit Permissions", "View Knowledge Base Permissions", and "Large Model Adjustment Permissions" behind the administrator account to assign these permissions to other administrators.

[0070] It should be noted that the above content only illustrates the technical idea of ​​the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.

Claims

1. A teaching assistance system based on the fusion of large model Agent and database technologies, characterized by: At least including large model application design module, database information management module and system interaction design module, The large model application design module: designs a dedicated "Wenwu" large model, based on a powerful neural network, processes natural language data or image data, and can accurately answer professional knowledge questions input by teachers and students after fine-tuning the relevant data of teaching materials, and organizes a large amount of knowledge to generate the content of teaching materials; Agent personalizes the call of the "Wenwu" large model, and completes intelligent examination and judgment, and generates teaching material files based on the output of the large model. The database information management module is used to manage user identity and authority information and teaching material information, maintain the user's identity information and other basic information, basic information of teaching materials and review status information, accept the operation information of the system interactive design module to update the data table in real time, and provide the system interactive design module with the URL of the teaching material to be downloaded for subsequent user operations; it is divided into private teaching material database and public teaching material database according to the open authority of teaching materials, and stores accessible user information and related teaching material information, accepts the authority operation of the system interactive design module user to update the database access authority, The system interaction design module designs system visualization and interaction functions, including user application or consent to private library access operations, administrator review of uploaded teaching materials operations, user information display and interface interaction beautification, so as to obtain user operations for the purpose of facilitating user interaction and transmit the operation results to the large model application design module and the database information management module.

2. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 1 is characterized by: The large model application design module is divided into a knowledge question answering submodule, a teaching material generation submodule and a large model adjustment management submodule. The knowledge question answering submodule: A user who is a teacher or a student inputs his or her own question, and the "Wenwu" model first determines whether the question is related to the knowledge in the knowledge base. If so, it performs search enhancement generation (RAG) in the knowledge base to output the answer. If not, it directly outputs the answer based on the user's "learned knowledge"; The teaching material generation submodule: users with the identity of teachers upload various types of teaching materials, requiring The "Wenwu" model generates other types of teaching materials according to the files of specific teaching materials, or inputs the handwritten mathematical expressions in the form of pictures to obtain the corresponding LaTeX code to assist in the editing of teaching materials; The large model adjustment management submodule includes preview command function, save training parameter and load training parameter functions, start fine-tuning and interrupt fine-tuning functions, supports large model selection, fine-tuning data set selection, training parameter and LoRA parameter settings, output directory settings, real-time viewing of progress, loss curve and training log during fine-tuning, and supports loading checkpoint path and question-and-answer test after fine-tuning.

3. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 1 is characterized by: The database information management module is divided into a teaching material management submodule and a permission management submodule. The teaching material management submodule: receives the result of the content review management submodule passing or failing the operation, and updates the review status of the existing teaching materials in the teaching material library in real time. Users upload various types of teaching materials to private or public teaching material libraries, or download teaching materials from private, shared or authorized private teaching material libraries, wherein the teaching materials generated using the existing materials are automatically stored as new teaching materials in the private teaching material library; The authority management submodule receives the identity information of the user requesting access, provides the user requesting access with operation result information according to whether the user of the private library is allowed to access the private library, and updates the relevant teaching material library opening authority.

4. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 1 is characterized by: The system interaction design module specifically includes a content review submodule and a permission operation submodule. The content review submodule: a user with the identity of an administrator receives all teaching material contents of the teaching material management module, and reviews the rationality and legality of the material contents by checking the download method, makes a pass or fail operation and returns the operation result to the teaching material management submodule; The permission operation submodule: the user sends application information to the private library user through the access application operation; the user who receives the access application information chooses whether to agree, and the access result information is synchronously transmitted to the user who applied for access.

5. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 2 is characterized by: The large model application design module includes large model LoRA fine-tuning technology and large model Agent technology. The large model LoRA fine-tuning technology: The large model LoRA fine-tuning technology is an efficient parameter adjustment method. By decomposing the model weights into the product of two low-rank matrices, the specific implementation steps include downloading and formatting the dataset, selecting a suitable large model, decomposing the model weights into the product of two low-rank matrices, and performing fine-tuning. The big model agent technology: the agent combines the four elements of planning, memory, tool use and execution, calls the big model and uses the output of the big model to provide further support for teachers and students. When the teacher needs to generate a lesson plan or PPT, the agent first understands this demand and breaks it down into a series of steps, including retrieving relevant materials and generating a document framework. Then, the agent uses the information in the knowledge base and text generation tools to create teaching materials that meet the requirements. The agent can also process images, including converting handwritten formulas into LaTeX code, thereby helping teachers prepare course content more efficiently. Throughout the process, the agent will continuously optimize services based on the user's historical interaction records and personal preferences to ensure that the teaching resources provided are more in line with actual needs.

6. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 2 is characterized by: The large model application design module has preset the "Wenwu" test-taking and grading model obtained by fine-tuning the Qwen large model using the MR-Ben data set and LoRA technology. The fine-tuning process of the test-taking and grading model is as follows: first, the MR-Ben dataset specially adapted for the test-taking and grading task is prepared and parsed. Then, based on the LoRA core formula h=W0x+ΔWx=W0x+BAx, while keeping most of the parameters of the pre-trained model unchanged, low-rank matrix updates are introduced only for specific layers. After a carefully designed training process, including data preprocessing, defining LoRA settings, iterative optimization and other steps, the Qwen model is adapted to this complex reasoning and evaluation task. The framework of the test-taking and grading model: The input format of the model is a structured text sequence, including the description of the question, the candidate's answering process and step-by-step instructions. Specifically, each input instance is constructed into a composite structure containing instructions, context, questions and candidate answers. For the output format, the model generates a scoring result and corresponding comments. The scoring result is a quantitative assessment of the accuracy and completeness of the candidate's answer, usually presented in numerical form. The comments are a qualitative analysis of the problem-solving process, pointing out the strengths and weaknesses and making suggestions for improvement.

7. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 3 is characterized by: The teaching material management submodule supports uploading of multiple types of teaching materials including teaching syllabus, teaching requirements, lesson plans, study guides, teaching PPTs, and after-class exercises.

8. The teaching auxiliary system based on large model Agent and database multi-technology integration according to claim 3 is characterized by: The method for using the teaching auxiliary system comprises the following steps: S1, Registration and Login: When using this system for the first time, users need to fill in the necessary system information including identity, user name, account number, password, name, ID number and national real-name system information. When using this system for a non-first time, users can log in with their account number and password. The system will enter different main interfaces according to the identity information of the account. S2, intelligent question answering: users input their own questions, and the big model first determines whether the question is related to the knowledge in the knowledge base. If so, it performs search-enhanced generation (RAG) in the knowledge base to output the answer. If not, it directly outputs the answer based on the user's "learned knowledge". S3, intelligent generation: when generating teaching material files, the user who is a teacher first uploads a certain type of teaching material as the knowledge base of the big model retrieval enhanced generation (RAG), and then enters the requirements, that is, what type of corresponding teaching material the big model needs to generate; when generating LaTeX code, the user who is a teacher first enters the handwritten mathematical expression content in image format, obtains the corresponding LaTeX code, and then previews it. If necessary, the LaTeX code can be manually modified online. S4, Smart Search: Users who are teachers or students can click on the "Smart Search" function area of ​​the main interface to perform search operations. By entering keywords in the input box, all files containing the keywords will be retrieved and displayed in the system public library and accessible private library for users to download and review. S5, Knowledge Base View: Users who are teachers or students can click "My Knowledge Base" in the main interface function area to enter their personal private knowledge base to view all uploaded files, including the file review status and upload time information, or click the "Upload" button to continue uploading local teaching materials to the knowledge base for management and choose whether to make them public. Users enter the user account they want to access in the input box to apply for access to the user's private knowledge base. For the knowledge base that has been approved, the user can directly view all the file contents in the knowledge base. S6, large model tuning: users with the identity of administrators need to first select the large model, then select the fine-tuning dataset, set the training parameters and LoRA parameters and the output directory, and then click the "Start" button to start fine-tuning. At the end of fine-tuning, select the output directory of the just fine-tuned model in the checkpoint path, and then enter the chat area for question-and-answer testing. S7, Data Review: Users with the identity of administrators can view all files in all libraries, review and judge the contents of files, and make "pass / fail" operations to change the file review status. The system automatically publishes relevant information to notify the user to whom the file belongs.

Images