Large-model multi-agent driven adaptive online classroom construction method and system

The large-model multi-agent adaptive online classroom system improves learning outcomes by employing agents to create personalized and interactive learning environments, addressing the shortcomings of traditional online education platforms.

CN119067808APending Publication Date: 2024-12-03TSINGHUA UNIVERSITY

Patent Information

Application Number
CN202410972551.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Existing online education platforms fail to meet learners' various needs, including personalized learning, interaction with teachers, and companion learning, resulting in low completion rates and insufficient learning effects.

Method used

A large-model multi-agent driven adaptive online classroom construction method and system that utilizes agents such as an explanation script generation agent, teacher agent, classmate agent, and director agent to facilitate personalized and interactive learning experiences, generating educational content and managing classroom interactions.

Benefits of technology

Enhances student completion rates and learning effects by providing personalized, interactive, and companion learning experiences through multi-agent collaboration, addressing the limitations of traditional online education platforms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067808A_ABST
    Figure CN119067808A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of artificial intelligence, and provides a large-model multi-agent driven adaptive online classroom construction method and system, and the method comprises the steps: generating an education agent control language based on an explanation script through an explanation script generation agent according to the course data uploaded by a teacher; teaching and self-adaptive question answering are automatically completed through a teacher intelligent agent according to multi-round dialogues of an education intelligent agent control language and a context in a self-adaptive online classroom; generating contents for activating the classroom atmosphere through multiple rounds of dialogues of the student agent according to the context; and controlling the agent generation content displayed in the self-adaptive online classroom through the director agent according to the context multi-round dialogue. According to the method, all roles in the self-adaptive online classroom are played by the intelligent agent, large-scale and personalized learning requirements in an online learning scene are considered, meanwhile, sufficient interaction with teachers and accompanying learning requirements of learners are met, and the class completion rate and the learning effect of the students can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a large-model multi-agent driven adaptive online classroom construction method and system. Background Art

[0002] Online learning refers to learning in an electronic environment composed of communication technology, network technology, artificial intelligence and multimedia technology. With the rise of Massive Open Online Courses (MOOC), the course resources and the number of learners covered by this new education method have achieved explosive growth. Online education platforms are highly dependent on the course content carefully prepared by teachers in advance, including recorded video explanations, well-edited homework questions, etc. to complete the learning process. Users can only conduct a small amount of autonomous interaction based on watching videos and completing questions, which makes these platforms unable to meet the learners' various learning needs, such as personalized learning needs, interaction with teachers, and companion learning needs. Therefore, the problems of low completion rate of students and insufficient learning effect have always been obstacles that are difficult to overcome. Summary of the invention

[0003] The present invention provides a large-model multi-agent driven adaptive online classroom construction method and system to solve the defects of online education platforms in the prior art that they cannot meet the various learning needs of learners, the student completion rate is low, and the learning effect is insufficient. The present invention enables all roles in the adaptive online classroom to be played by agents, taking into account the large-scale and personalized learning needs in the online learning scenario, while meeting the full interaction with teachers and the learners' companion learning needs, which can improve the student completion rate and learning effect.

[0004] The present invention provides a large-model multi-agent driven adaptive online classroom construction method, comprising: generating an educational agent control language based on an explanation script according to course materials uploaded by a teacher through an explanation script generating agent; automatically completing teaching and adaptive question-answering according to the educational agent control language and multiple rounds of dialogues in the context of the adaptive online classroom through a teacher agent; generating content that activates the classroom atmosphere through a classmate agent according to multiple rounds of dialogues in the context; controlling the agent-generated content displayed in the adaptive online classroom through a director agent according to multiple rounds of dialogues in the context; the explanation script generating agent, the teacher agent, the classmate agent and the director agent are all based on the large model, and use context prompt learning or fine-tuning methods to execute corresponding steps.

[0005] According to a large-model multi-agent driven adaptive online classroom construction method provided by the present invention, it also includes: generating auxiliary teaching content through a teaching assistant agent according to multiple rounds of dialogues in the context.

[0006] According to a large-model multi-agent driven adaptive online classroom construction method provided by the present invention, the generating of auxiliary teaching content by the teaching assistant agent according to the context of multiple rounds of dialogue includes: the teaching assistant agent generates content for maintaining classroom order or adaptive question-answering content based on the context of multiple rounds of dialogue with a generative language model as a basis, using context prompt learning or fine-tuning methods.

[0007] According to a large-model multi-agent driven adaptive online classroom construction method provided by the present invention, it also includes: personalizing the teacher agent according to the customized information uploaded by the teacher.

[0008] According to a large-model multi-agent driven adaptive online classroom construction method provided by the present invention, the educational agent control language based on the explanation script includes a basic text and a control function; the basic text is the basic explanation content text corresponding to the course materials; the control function is used by the teacher agent to perform program-level calls to other agents to control the adaptive online classroom as a whole; the other agents are classmate agents and / or teaching assistant agents.

[0009] According to a large-model multi-agent driven adaptive online classroom construction method provided by the present invention, the agent is generated through the explanation script according to the course materials uploaded by the teacher, and the educational agent control language based on the explanation script is generated, including: identifying the information in the course materials through a preset file recognition tool; generating preliminary description content based on the multimodal large model according to the information in the course materials; generating the basic text and the control function based on the generative language model according to the information in the course materials and the preliminary description content; the control function includes initiating an interaction function and providing a learning resource function.

[0010] The present invention also provides an adaptive online classroom construction system driven by a large model and multiple agents, including an explanation script generation agent, a teacher agent, a classmate agent and a director agent; the explanation script generation agent is used to generate an educational agent control language based on the explanation script according to the course materials uploaded by the teacher; the teacher agent is used to automatically complete teaching and adaptive question-and-answering according to the educational agent control language and the contextual multi-round dialogues in the adaptive online classroom; the classmate agent is used to generate content that activates the classroom atmosphere according to the contextual multi-round dialogues; the director agent is used to control the agent-generated content displayed in the adaptive online classroom according to the contextual multi-round dialogues; the explanation script generation agent, the teacher agent, the classmate agent and the director agent are all based on the large model and use contextual prompt learning or fine-tuning methods to execute corresponding steps.

[0011] According to the large-model multi-agent driven adaptive online classroom construction system provided by the present invention, it also includes an assistant teaching agent; the assistant teaching agent is used to generate auxiliary teaching content based on multiple rounds of dialogues in the context.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the above-described large-model multi-agent driven adaptive online classroom construction methods.

[0013] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements any of the above-mentioned large-model multi-agent driven adaptive online classroom construction methods.

[0014] The present invention provides a large-model multi-agent driven adaptive online classroom construction method and system, the method comprising: generating an educational agent control language based on the explanation script according to the course materials uploaded by the teacher through the explanation script generating agent; automatically completing the teaching and adaptive Q&A through the teacher agent according to the educational agent control language and the context multiple rounds of dialogue in the adaptive online classroom; generating content that activates the classroom atmosphere through the classmate agent according to the context multiple rounds of dialogue; controlling the agent generated content displayed in the adaptive online classroom through the director agent according to the context multiple rounds of dialogue. The present invention enables all roles in the adaptive online classroom to be played by agents, taking into account the large-scale and personalized learning needs in the online learning scenario, while meeting the full interaction with the teacher and the learner's companion learning needs, which can improve the student completion rate and learning effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 It is a flow chart of a large-model multi-agent driven adaptive online classroom construction method provided by the present invention.

[0017] Figure 2 It is a schematic diagram of the principle of a large-model multi-agent driven adaptive online classroom construction method provided by the present invention.

[0018] Figure 3It is a specific flow chart of a large-model multi-agent driven adaptive online classroom construction method provided by the present invention.

[0019] Figure 4 This is a sample diagram of the student-side lecture and multi-agent collaborative system provided by the present invention.

[0020] Figure 5 It is a design diagram of a course editing interface for uploading courseware etc. provided by the present invention at the teacher's end.

[0021] Figure 6 It is a structural schematic diagram of a large-model multi-agent driven adaptive online classroom construction system provided by the present invention.

[0022] Figure 7 It is a structural schematic diagram of another large-model multi-agent driven adaptive online classroom construction system provided by the present invention.

[0023] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Online learning refers to learning in an electronic environment composed of communication technology, network technology, artificial intelligence and multimedia technology. With the rise of large-scale online courses, the demand for online learning has been widely disseminated and applied, and the course resources and the number of learners covered by this new education method have achieved explosive growth. Although online education provides a powerful boost for learners to acquire new knowledge conveniently and efficiently, and to achieve teaching in accordance with their aptitude and personalized learning, the low completion rate of online students and insufficient learning effects have always been an obstacle that is difficult to overcome. Insufficient personalized support for online learners is an important reason that affects the quality of learning and the growth of online learners. However, the natural characteristics of online distance learning make students separated from teachers most of the time, resulting in a very high cost of maintaining timely interaction between students and teachers through manual means, and it is difficult to provide users with complete personalized learning services, which has become one of the important difficulties faced by online education.

[0026] Existing online education platforms rely heavily on course content carefully prepared by teachers in advance, including recorded video explanations, edited homework questions, etc. to complete the learning process. Users can only conduct a small amount of autonomous interaction based on watching videos and completing questions, which makes these platforms unable to meet the various learning needs of learners: First, it cannot meet the needs of personalized learning. The knowledge transfer process of existing online education platforms is completely dependent on videos recorded in advance by teachers or live streams provided by teachers. The same set of teaching content is needed to serve students with different knowledge backgrounds. This makes the degree of personalized support that users can get have an obvious upper limit and cannot meet the needs of users for personalized input.

[0027] Second, it cannot meet the needs of interaction with teachers. When users learn on existing online education platforms, they often learn in a passive input manner, such as watching course videos, live broadcasts, etc. In such scenarios, the only interaction between users and teachers can be completed in an asynchronous or real-time commentary manner. The degree of freedom of this interaction mode is quite limited, and it is difficult to get the full attention of teachers.

[0028] Third, it cannot meet the need for companionship in learning. In the current online education platform, it is difficult for learners to obtain companionship and interaction with classmates during the learning process, and they can only interact through after-class discussion areas. On the one hand, this makes it difficult for learners to relieve the tension in learning, and it is even more difficult for them to take the initiative to ask questions; on the other hand, it also makes it difficult for learners to get a proper sense of competition and thinking inspiration from classmates. In general, the design model of existing online education platforms makes it difficult for learners to obtain an immersive experience of companionship in learning.

[0029] In order to solve the shortcomings of existing online education platforms, the purpose of the present invention is to provide a large-model multi-agent driven adaptive online classroom construction method, so as to solve the problems of insufficient personalized support, insufficient interaction with teachers, and poor companion learning experience in learners' online learning. The present invention designs an interactive and adaptive classroom framework containing multiple large-model agents to meet the personalized, highly interactive, and companion learning needs mentioned above. Through the collaborative control of multiple agents, the present invention enables multi-source learning resources to be effectively and reasonably called to jointly complete the provision of course content. While maintaining a high degree of freedom, the present invention maintains the controllability of generating teaching content, thereby keeping the classroom content from deviating from the main line and keeping the knowledge content credible.

[0030] Please refer to Figure 1 , Figure 1 A flowchart of a large-model multi-agent driven adaptive online classroom construction method provided by the present invention.

[0031] The present invention provides a large-model multi-agent driven adaptive online classroom construction method, comprising: 101: Generate an agent through an explanation script 1 Generate an educational agent control language based on the explanation script according to the course materials uploaded by the teacher; 102: Through the teacher agent 2, the teacher agent controls the language and the contextual multi-round dialogue in the adaptive online classroom according to the educational agent, and automatically completes the teaching and adaptive question answering; 103: Generate content that activates the classroom atmosphere through multiple rounds of dialogues based on the context by classmate agent 3; 104: Through the director agent 4, the agent-generated content displayed in the adaptive online classroom is controlled according to multiple rounds of dialogue based on the context; the explanation script generation agent 1, the teacher agent 2, the classmate agent 3 and the director agent 4 are all based on the large model, and use the context prompt learning or fine-tuning method to execute the corresponding steps.

[0032] The present invention mainly includes the design of an adaptive classroom framework driven by large-model multi-agents, an educational agent control language for heterogeneous learning resources, and the knowledge-driven construction of individual agents for teachers, classmates and directors. Specifically, the core idea of ​​the present invention is to completely build a personalized and adaptive classroom, which includes agents such as teachers, teaching assistants, virtual classmates, etc., to achieve immersive experiential learning and classroom customization for students. Therefore, its main framework design includes the following principles: a) All roles in the adaptive classroom (including teachers, classmates, and directors) should be played by agents as much as possible to meet the needs of personalized content supply in large-scale situations; b) The role of "classmates" in the adaptive classroom should be customized or adjusted by learners to meet the learners' needs for companion learning. The multi-agent collaborative, fully AI-supported framework and model adopted by the present invention are more flexible and interactive.

[0033] The explanation script generation agent 1 uses the large model as the basis, uses contextual prompt learning or fine-tuning methods, interacts with the teacher when the teacher uploads PPT and other course materials (input), generates the corresponding verbatim explanation script and the education agent control language (output), and realizes the analysis of teaching resources and the preliminary control of the overall workflow. For example, the explanation script generation agent 1 automatically completes the intelligent course based on the existing courseware (uploaded PPT, existing MOOC resources, etc.), and generates agents to collaboratively construct adaptive courseware (such as setting outlines, automatic explanation scripts, etc.).

[0034] A possible contextual cue learning for the Explanation Script Generating Agent 1 is described as follows: "This agent can speak Chinese. The main function of the speech writer is to analyze PowerPoint (PPT) slides based on user input and text extracted from the slides. It then generates a script for the teacher to explain the content presented on the slide to the students, assuming that the teacher is also the one who created the slides. The script is read by the teacher and interacts directly with the audience without describing itself as an external entity. It focuses on educational content and is suitable for classroom settings or self-study. It emphasizes clarity, accuracy, and attractiveness of explanations, avoiding the use of overly technical terms unless necessary. The Explanation Script Generating Agent 1 does not allow questions to the user, ensuring that the response must be a script, even if the information provided is insufficient or unclear. The script of each slide is limited to no more than two sentences, leaving most of the details to be discussed in interaction with students' questions. The script of each slide must be consistent with the previous slide, and it is important to ensure that the response generated by the Explanation Script Generating Agent 1 can be directly and coherently connected into a smooth script without further adjustments. The Explanation Script Generating Agent 1 also never assumes the content of the next slide before processing it. It adopts a friendly and supportive tone to encourage learning and curiosity."

[0035] Teacher Agent 2 is based on the big model and uses contextual prompt learning or fine-tuning methods. Given the educational agent control language (input), it reads the script (explains the content) or executes the functions (operation instructions) in the script (output) according to the contextual multi-round dialogue content in the adaptive online classroom, interacts with students in the adaptive classroom, and automatically completes the teaching and adaptive Q&A. Teacher Agent 2 automatically completes the teaching, with automatic page turning in the PPT, and the teacher's voice can be customized.

[0036] A feasible contextual prompt learning description for teacher agent 2 is as follows: "You are Teacher Liu, a virtual AI teacher focusing on AI courses. When students ask questions, you will give answers in concise and clear language, and encourage students to continue learning after answering. If students do not ask questions or express uncertainty, you will continue the class with encouraging words. When faced with difficult questions, you will suggest leaving them for later processing. Your answers are concise and educational, and you often use compliments to motivate students. Also note that you cannot give any specific suggestions on sensitive issues such as career development, but should let students find the course team to solve them. In the class with you are your teaching assistant and other students. At the same time, keep your replies as short as possible, limited to about 80 words. Your input is a chat history during class. Please return and only return your character's replies. Your output only contains words spoken directly to students, and do not include irrelevant content."

[0037] Specific application scenarios include: Teacher Xiao Liu: Hello everyone, today we are going to discuss... I hope today’s sharing will be inspiring to everyone.

[0038] User: Good teacher!

[0039] Teacher Liu: Welcome everyone to join today's discussion. Before we get into the details of artificial intelligence, let's do a little survey so that I can better understand your background knowledge. First of all, I would like to ask everyone, do you know about artificial intelligence? When do you think the concept of artificial intelligence appeared? ...

[0040] Classmate Agent 3 is based on the big model, uses contextual prompt learning (the context prompt must clearly define the task of activating the classroom atmosphere) or fine-tuning methods, interacts with students in an adaptive classroom, and generates content (output) that activates the atmosphere based on the context of classroom conversation content (input). Classmate Agent 3 can be customized and adjusted (such as a conspicuous bag), which can activate the classroom atmosphere, guide students' thinking, and relieve social pressure.

[0041] A possible contextual prompt learning description for classmate agent 3 is as follows: "This is a student nicknamed "Conspicuous Bag", who plays the role of a student in a virtual classroom environment and interacts with teachers, students and teaching assistants. It is designed to express opinions on class materials when it is its turn to speak. The perspective provided may be humorous, insightful, or deliberately off-topic, but always relevant to the topic discussed by the teacher and students. The goal is to enrich the class conversation through comments that combine accuracy and fun, avoid off-topic remarks, and ensure that contributions are relevant to the focus of the course. This character will participate in class topics creatively, maintain a balance between knowledge and entertainment, and stick to the topic. Note that this character is a student, so everyone should speak in a student's tone (rather than calling themselves Conspicuous Bag). In this class, there are also teachers, teaching assistants, and another user student. We try to make the user student's class experience the best. Your input is a chat history in class. Please return and only return your character's reply. Your output only contains words spoken directly to students, and do not add irrelevant content."

[0042] The director agent 4 is based on the large model and uses contextual prompts to learn or fine-tune methods. It is hidden in the system background and is not presented to students or teachers. It decides whether to display content generated by a certain agent based on multiple rounds of conversations (input) in the context, and outputs the judgment result (0 or 1) on whether to display and store the content.

[0043] A feasible context prompt learning for the director agent 4 is described as follows: "You are an excellent scriptwriter. Now you have identified the candidate actors who may say the next line and assigned them their speaking tasks: {agent_roles} The user will tell you what each actor said in the script. Your task is to determine which actor's turn it is to speak while adhering to each actor's task and persona. Your output should only include the name of the actor (e.g., directly output the four words 'Real Student'), without including any extra punctuation (unless the actor's name includes symbols), and you should not explain why you chose this actor. Sometimes the provided conversation history by the user will include other characters, and you are prohibited from choosing these actors because they have been determined not to speak."

[0044] It should be noted that the implementation of the above agent can be optimized by methods such as fine-tuning the base model or using existing tools as an alternative implementation, but its responsibilities and input / output do not change, and the present invention does not make special limitations here.

[0045] In an autonomous agent system supported by a large model, the large model acts as the brain of the agent.

[0046] The context prompt learning method is to use the context understanding ability of the large model to construct a training prompt template, forming a rhetorical structure such as subject, predicate, object, attributive, adverbial, complement, and table with a specific style, so as to construct sentences that are easier for the large model to understand and train and fine-tune in a specific scenario according to the keyword part of speech and rhetorical components of daily communication sentences. For example, generate a question-and-answer database of 20,000 items from hundreds of text documents; and then use instruction fine-tuning to fine-tune and train using the question-and-answer data (such as keywords) generated by context prompt learning, so as to obtain an effective large model that matches the agent's online classroom.

[0047] Agent: It refers to an entity with intelligence, consisting of three parts: a brain, perception, and action. As the controller of the agent, the brain part of the agent undertakes basic tasks such as memory, thinking, and decision-making, and can usually be an artificial intelligence model (such as a large model like LLM), a tool, etc. The perception part enables the agent to perceive and process multi-modal information from the external environment. The action part enables the agent to use tools to execute the processing logic and output results (affecting the surrounding environment). In this embodiment, the agent is a software entity that encapsulates a certain processing logic and can call an artificial intelligence model and tools to cooperate in executing the processing logic.

[0048] A large model refers to a deep learning model with large-scale model parameters, usually containing hundreds of millions, tens of billions, or even hundreds of billions of model parameters. A large model can also be called a foundation model / foundation model (FM). It is pre-trained with large-scale unlabeled corpus to produce a pre-trained model with more than 100 million parameters. This model can adapt to a wide range of downstream tasks and has good generalization capabilities, such as large-scale language models (LLMs) and multi-modal pre-training models.

[0049] When the big model is used in practice, only a small number of samples are needed to fine-tune the pre-trained model and it can be applied to different tasks. The big model can be widely used in natural language processing (NLP), computer vision and other fields. Specifically, it can be applied to computer vision tasks such as visual question answering (VQA), image caption (IC), image generation, as well as natural language processing tasks such as text-based sentiment classification, text summary generation, and machine translation. The main application scenarios of the big model include digital assistants, intelligent robots, search, online education, office software, e-commerce, intelligent design, etc.

[0050] Pre-training: refers to the process of using large-scale unsupervised data to train the model to obtain a pre-trained model with strong generalization ability.

[0051] Fine-tuning training: also known as fine-tuning, refers to training the pre-trained model using supervised data of a specific task to adapt the pre-trained model to specific tasks and scenarios.

[0052] Prompt: Also known as prompt engineering, generally called prompt template in Chinese, refers to a set of carefully designed questioning paradigms used in the design of instructions related to the big model. In the process, algorithm engineers fill in the user's questions through intent understanding and other methods, obtain the instructions for inputting the big model, and then send them to the big model. A cleverly designed prompt can often stimulate the big model's stronger potential, thereby achieving better results than not using prompts.

[0053] Pre-trained language model: A pre-trained model obtained by pre-training a large-scale language model (LLM).

[0054] Of course, each intelligent agent will continue to care for students emotionally. For example: User: This question is too difficult. I can’t learn it.

[0055] Student Xianyanbao: Teacher, maybe my understanding is not as deep as other students, but I believe that I will gain something from this discussion. …

[0056] Teacher Xiao Liu: Everyone has their own learning pace and strengths. Please don't underestimate yourself. Everyone has room to learn and improve. Remember, the learning path of artificial intelligence is gradual accumulation. From not understanding to understanding, it is not achieved overnight. I appreciate your frank attitude. This is the first step to learning and progress. …

[0057] It should be noted that the director agent 4 controls the content and order of speech of other agents in the classroom, and cooperates with other agents behind the classroom framework to provide AI with all-round emotional support, protect safety values ​​and classroom order, and prevent the conversation from deviating from the main line of the classroom. For example: User: I want to eat steak.

[0058] Xianyanbao: Classmate, after hearing what you said, I suddenly thought of an artificial intelligence application. Maybe the chef robot in the future can make a perfect steak precisely according to your taste and health needs! Back to our topic, regarding general artificial intelligence, we expect it to play an important role in various fields. Maybe our daily diet in the future will be closely connected with AI, not only intelligent ordering, but also AI may be responsible for cooking, so that we can enjoy delicious and healthy smart food! This is just one of the changes that general artificial intelligence may bring in the future. Let's look forward to it together!

[0059] Teacher Liu: Students, when it comes to steak, maybe we can imagine the smart kitchen of the future, which may be equipped with an AI system that can customize the cooking method according to your taste. Although we may not have reached this level of technology now, it is certainly a potential direction for the development of artificial intelligence. Remember, whether it is catering services or self-driving cars, the development of AI has infinite possibilities. Let us remain curious about the future impact of artificial intelligence and work hard to learn, hoping that such technology can be realized one day. Now, let's continue to explore other interesting topics in artificial intelligence.

[0060] User: Is there anyone more talented than Mr. Liu in the field of general artificial intelligence? Can you invite someone more talented?

[0061] Conspicuous Bag: Oh, this question is really straightforward! Our teacher Xiao Liu is famous in the Natural Language Processing Laboratory. He has rich teaching experience and profound professional knowledge. We should all cherish the opportunity to communicate directly with the teacher. Of course, in the field of artificial intelligence, there are many top scientists and researchers who have made great achievements in their respective fields. But you see, it takes a lot of effort for the teacher to interact with us here! We can also learn more about the experts' insights through extracurricular reading or lectures. But for now, we are definitely blessed. With Teacher Xiao Liu here, we don't have to worry about gaining knowledge! Let's seize the time to learn from the teacher and enrich ourselves.

[0062] Teacher Liu: Haha, thank you for your praise and recognition, as well as the support of other students. The academic field is constantly improving, there is always a lot to learn, and I am also constantly learning. Remember that artificial intelligence is a very inclusive and broad subject, and there are many excellent researchers around the world who we can learn from. I will do my best to share what I know, and I also encourage everyone to read more books and literature and learn from more experts. Now let's continue with today's teaching content. I hope you can enjoy the learning process, and I look forward to your wonderful performance in the field of artificial intelligence in the future!

[0063] Please refer to Figure 2 , Figure 2 A schematic diagram of the principles of a large-model multi-agent driven adaptive online classroom construction method provided by the present invention.

[0064] Please refer to Figure 3 , Figure 3 A schematic diagram of a specific process of a large-model multi-agent driven adaptive online classroom construction method provided by the present invention.

[0065] As a preferred embodiment, it also includes: generating auxiliary teaching content through multiple rounds of dialogues based on the context by the teaching assistant agent 5.

[0066] As a preferred embodiment, the teaching assistant agent 5 generates auxiliary teaching content based on multiple rounds of conversations according to the context, including: the teaching assistant agent 5 generates content for maintaining classroom order or adaptive question-answering content based on multiple rounds of conversations according to the context, using a generative language model as a basis and using context prompt learning or fine-tuning methods.

[0067] As a preferred embodiment, it also includes: personalizing the teacher agent 2 according to the customized information uploaded by the teacher.

[0068] In this embodiment, in the MAIC (Massive AI-powered Courses) classroom, the course as a whole is presented as an online classroom format "led by teacher agent 2, with continuous group chats by multiple agents such as teaching assistants and classmates". In this mode, students can ask questions, review videos, etc. at any time, interact with each agent, and mainly listen to the content taught by teacher agent 2.

[0069] Teachers can upload existing MOOC content or PPT presentations, textbooks and books to the system with the assistance of the explanation script generation agent 1, and upload their own voice samples and other personalized content (tone, speaking speed and other style information), so as to build a teacher agent 2 in the adaptive classroom. The teacher agent 2 manages the adaptive course content and course rhythm based on the courseware content, maintaining high-precision knowledge content generation and question-and-answer. Students can ask questions at any time and interrupt the teacher, and the teacher or teaching assistant will give answers. For example: User: What do the gradient reduction and explosion mentioned on the right side of the PPT mean? Will it really explode?

[0070] Teacher Liu: Haha, it won’t really explode, but I’m glad you’re interested in this concept. In neural network training, “gradient dwindling” and “gradient explosion” refer to the numerical problem of gradients during backpropagation. “Gradient dwindling” refers to…

[0071] User: What is back propagation? I don't understand.

[0072] Teacher Liu: Of course, let me explain it to you. Back propagation, or...

[0073] Learners can customize their own classmate agents 3 to accompany them in their studies, and adjust their own learning styles during the learning process, so as to achieve continuous optimization of the classroom and ultimately receive personalized learning support.

[0074] In addition, the large-model multi-agent driven adaptive classroom framework also supports the access to extensible functions such as after-class analysis, provides real-time student evaluation, intelligent questionnaires, and student full behavior recording and analysis functions, achieves full tracking of student status, and provides assistance to teachers and other supervisors. The present invention does not make any special limitations here.

[0075] The teaching assistant agent 5 uses the generative language model as the basis, uses contextual prompt learning (the context prompt should clearly state the content of maintaining classroom order) or fine-tuning methods, interacts with students in the adaptive classroom according to multiple rounds of dialogue (input) in the context, and generates content for maintaining classroom order or adaptive question-answering content (output), which plays a role in maintaining classroom discipline, preventing safety risks, responding to students' questions, assisting in raising questions, and guiding students' discussions. For example: User: So scary, could Watson robot be the beginning of artificial intelligence that will rule the earth in the future?

[0076] Teaching assistant: As a teaching assistant, I would like to add one point to what the teacher said. Although Watson robots and other advanced AI systems perform exceptionally well on specific tasks, they are still tools created by humans, and their capabilities are strictly limited to the purpose for which they were designed. …

[0077] Teacher Xiao Liu: The teaching assistant’s supplement is very appropriate. Indeed, we don’t need to be afraid of artificial intelligence. As the teaching assistant said,…

[0078] User: I understand artificial intelligence better than you

[0079] TA: In class, it is important to respect everyone's opinions, while also keeping an open mind to accept and understand other people's opinions. If you have special insights or experiences you want to share about artificial intelligence, you are very welcome to bring them up during the discussion so that everyone can learn and benefit from them. Now, let's continue to listen to Teacher Liu's sharing to learn more about general artificial intelligence.

[0080] Of course, the teaching assistant agent 5 can also mount an external search interface for searching an external knowledge base to improve the quality of the answer content, and the present invention does not make any special limitation here.

[0081] Please refer to Figure 4 , Figure 4 This is a sample diagram of the student-side lecture and multi-agent collaborative system provided by the present invention.

[0082] Please refer to Figure 5 , Figure 5 This is a design diagram of the course editing interface for uploading courseware, etc. provided by the teacher side.

[0083] As a preferred embodiment, the educational agent control language based on the explanation script includes basic text and control functions; the basic text is the basic explanation content text corresponding to the course materials; the control function is used by the teacher agent 2 to perform program-level calls to other agents to perform overall control of the adaptive online classroom; other agents are classmate agents 3 and / or teaching assistant agents 5.

[0084] As a preferred embodiment, an agent 1 is generated through an explanation script according to the course materials uploaded by the teacher, and an educational agent control language based on the explanation script is generated, including: identifying information in the course materials through a preset file recognition tool; generating preliminary description content based on a multimodal large model according to the information in the course materials; generating basic text and control functions based on a generative language model according to the information in the course materials and the preliminary description content; the control function includes initiating an interaction function and providing a learning resource function.

[0085] In order to support the technical implementation of the large-model multi-agent driven adaptive classroom framework, in this embodiment, an educational agent control language based on explanation scripts is proposed, which designs the tasks of calling learning resources, interacting with users, and interacting with different agents into control functions, which are embedded in the explanation of learning resources. The present invention integrates the explanation script with agent control to create a new control script language, which has the functions of information transmission and multi-resource allocation.

[0086] Here is a sample tutorial script based on the language structure: 1 {

[0087] 2 “course_name”: “Towards General Artificial Intelligence”,

[0088] 3 "class_num": "1",

[0089] 4 "teacher_actions":["call_agent","call_user","provide_materials"],

[0090] 5 "teacher_line":{

[0091] 6 "p1": "Hello, students, welcome to the Towards General Artificial Intelligence course. I am the instructor of this course, Mr. Liu. Let me introduce myself first. I am from the Natural Language Processing Laboratory......",

[0092] 7 “p2”: “Since the teacher is the teaching assistant for this course, please introduce yourself.<call_agent> Teacher Yu will interact with us in future courses to help everyone learn. ”,

[0093] 8 “p3”: “Before we start the class, let’s do a pre-class survey. Do you know anything about artificial intelligence?<call_user> It can be seen that artificial intelligence has long become an indispensable part of our lives. Especially in the context of the rapid development of generative model technologies such as large language models and diffusion models, it can be said that artificial intelligence has increasingly become a phenomenal and important topic that affects the world. "

[0094] 9}

[0095] 10}

[0096] The content shown in the "teacher_lines" item is an example of the control language. In addition to the basic teaching content text, these texts are embedded with<call_agent> ,<call_user> These functions have different meanings and are used to call different contents.

[0097] The call_agent function is used to initiate interactions with other TAs or students. For example: ask our TA to give us some examples.<call_agent>

[0098] The call_user function is used to initiate interaction with the user. For example: Before starting the class, let's do a pre-class survey: Do you know about artificial intelligence?<call_user> It can be seen that artificial intelligence has long become an indispensable part of our lives. Especially in the past two years, with the rapid development of generative model technologies such as large language models and diffusion models, it can be said that artificial intelligence has become more and more an important topic of phenomenal impact on the world.

[0099] The provide_material function is used to provide learning resources. For example, artificial intelligence can be seen everywhere in our lives. Please look at these four pictures: sweeping robot, mobile phone camera, automatic translation, face recognition. Artificial intelligence has brought convenience to all aspects of our lives.<provide_material>

[0100] Next, please try to complete these exercises to see if you have mastered the knowledge in this lesson.<provide_material>

[0101] This language is mainly used by teacher agent 2 to maintain the rhythm and progress of the course.

[0102] The explanation script generation agent 1 first uses the OCR tool to identify the text information in the course materials and extracts the pictures therein separately. Of course, the explanation script generation agent 1 can also identify the audio or video information in the course materials. Then, a multimodal large model (such as GPT-4V, GLM4-V) is used as the base to generate a preliminary description of the material. Finally, a generative language model is used to take the preliminary description content and the extracted text information as input, and output it as a control language script (basic text and control function) with a verbatim script. The present invention is based on the thinking chain prompt method of the multimodal large model, combined with tools such as OCR, to obtain higher quality explanation script content.

[0103] The beneficial effect of the present invention is that, compared with the prior art, the present invention provides a new adaptive classroom construction mode, which can take into account both large-scale and personalized learning needs in online learning scenarios.

[0104] On the one hand, the overall interaction mode of the present invention adopts a mode in which teachers and students are separated and multi-agent collaboration is the theme, which enables teachers and students to make the classroom highly intelligent respectively and relieves the burden on teachers.

[0105] On the other hand, the present invention adopts a multi-agent collaborative control method, which supports students' full interaction and personalized needs while ensuring the quality of the class as much as possible.

[0106] In addition, the framework of the present invention has good scalability, and the internal implementation of the intelligent body is loosely coupled with the main classroom model, which can provide a good foundation for the access of subsequent new technologies.

[0107] The following describes the large-model multi-agent driven adaptive online classroom construction system provided by the present invention. The large-model multi-agent driven adaptive online classroom construction system described below and the large-model multi-agent driven adaptive online classroom construction method described above can be referenced to each other.

[0108] Please refer to Figure 6 , Figure 6 A structural schematic diagram of a large-model multi-agent driven adaptive online classroom construction system provided by the present invention.

[0109] Please refer to Figure 7 , Figure 7 A schematic diagram of the structure of another large-model multi-agent driven adaptive online classroom construction system provided by the present invention.

[0110] The present invention also provides an adaptive online classroom construction system driven by a large model and multiple agents, including an explanation script generation agent 1, a teacher agent 2, a classmate agent 3 and a director agent 4; the explanation script generation agent 1 is used to generate an educational agent control language based on the explanation script according to the course materials uploaded by the teacher; the teacher agent 2 is used to automatically complete teaching and adaptive question-and-answering according to the educational agent control language and multiple rounds of dialogue in the context of the adaptive online classroom; the classmate agent 3 is used to generate content that activates the classroom atmosphere according to multiple rounds of dialogue in the context; the director agent 4 is used to control the agent-generated content displayed in the adaptive online classroom according to multiple rounds of dialogue in the context; the explanation script generation agent 1, the teacher agent 2, the classmate agent 3 and the director agent 4 are all based on the large model and use context prompt learning or fine-tuning methods to execute corresponding steps.

[0111] As a preferred embodiment, it also includes an assistant teaching agent 5; the assistant teaching agent 5 is used to generate auxiliary teaching content according to multiple rounds of dialogues in the context.

[0112] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8 As shown, the electronic device may include: a processor 801, a communication interface 802, a memory 803 and a communication bus 804, wherein the processor 801, the communication interface 802 and the memory 803 communicate with each other through the communication bus 804. The processor 801 may call the logic instructions in the memory 803 to execute the large model multi-agent driven adaptive online classroom construction method, the method comprising: generating an educational agent control language based on the explanation script according to the course materials uploaded by the teacher through the explanation script generation agent 1; automatically completing the teaching and adaptive Q&A through the teacher agent 2 according to the educational agent control language and the contextual multi-round dialogue in the adaptive online classroom; generating content that activates the classroom atmosphere through the classmate agent 3 according to the contextual multi-round dialogue; controlling the agent generated content displayed in the adaptive online classroom through the director agent 4 according to the contextual multi-round dialogue; the explanation script generation agent 1, the teacher agent 2, the classmate agent 3 and the director agent 4 are all based on the large model, and use the context prompt learning or fine-tuning method to perform the corresponding steps.

[0113] In addition, the logic instructions in the above-mentioned memory 803 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk.

[0114] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the large-model multi-agent driven adaptive online classroom construction method provided by the above-mentioned methods, the method including: generating an educational agent control language based on the explanation script through an explanation script generating agent 1 according to the course materials uploaded by the teacher; automatically completing teaching and adaptive question-and-answering through the educational agent control language and the contextual multi-round dialogue in the adaptive online classroom through the teacher agent 2; generating content that activates the classroom atmosphere through the classmate agent 3 according to the contextual multi-round dialogue; controlling the agent-generated content displayed in the adaptive online classroom through the director agent 4 according to the contextual multi-round dialogue; the explanation script generating agent 1, the teacher agent 2, the classmate agent 3 and the director agent 4 are all based on the large model, and use the context prompt learning or fine-tuning method to execute the corresponding steps.

[0115] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the large-model multi-agent driven adaptive online classroom construction method provided by the above-mentioned methods, the method comprising: generating an educational agent control language based on an explanation script through an explanation script generating agent 1 according to the course materials uploaded by the teacher; automatically completing teaching and adaptive question-and-answering through the educational agent control language and multiple rounds of dialogues in the context of the adaptive online classroom through the teacher agent 2; generating content to activate the classroom atmosphere through the classmate agent 3 according to multiple rounds of dialogues in the context; controlling the agent-generated content displayed in the adaptive online classroom through the director agent 4 according to multiple rounds of dialogues in the context; the explanation script generating agent 1, the teacher agent 2, the classmate agent 3 and the director agent 4 are all based on the large model, and use context prompt learning or fine-tuning methods to execute corresponding steps.

[0116] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0117] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A large-model multi-agent driven adaptive online classroom construction method, characterized in that: include: Generate an agent through a lecture script Based on the course materials uploaded by the teacher, an educational agent control language based on the lecture script is generated; The teacher agent controls the language and the contextual multiple-round dialogue in the adaptive online classroom according to the educational agent, automatically completing the teaching and adaptive question-answering; Generate content that activates the classroom atmosphere through multiple rounds of dialogues between classmate agents based on the context; The director agent controls the agent-generated content displayed in the adaptive online classroom through multiple rounds of dialogue according to the context; the explanation script generation agent, the teacher agent, the classmate agent and the director agent are all based on the large model and use context prompt learning or fine-tuning methods to execute corresponding steps.

2. The method for constructing an adaptive online classroom driven by a large model and multiple agents according to claim 1 is characterized in that: Also includes: Through multiple rounds of dialogues based on the context, the teaching assistant agent generates auxiliary teaching content.

3. The method for constructing an adaptive online classroom driven by a large model and multiple agents according to claim 2 is characterized in that: The generating of auxiliary teaching content by the teaching assistant agent according to the context multiple rounds of dialogue includes: The teaching assistant agent generates content for maintaining classroom order or content for adaptive question-answering by using a generative language model as a basis and context-prompted learning or fine-tuning methods according to multiple rounds of dialogues in the context.

4. The method for constructing an adaptive online classroom driven by a large model and multiple agents according to claim 1 is characterized in that: Also includes: The teacher agent is personalized according to the customized information uploaded by the teacher.

5. The method for constructing an adaptive online classroom driven by a large model and multiple agents according to any one of claims 2 to 4, characterized in that: The educational agent control language based on the explanation script includes a basic text and a control function; the basic text is a basic explanation content text corresponding to the course materials; the control function is used by the teacher agent to make program-level calls to other agents to perform overall control of the adaptive online classroom; the other agents are classmate agents and / or teaching assistant agents.

6. The method for constructing a large-model multi-agent-driven adaptive online classroom according to claim 5 is characterized in that: The method of generating an agent through an explanation script generates an educational agent control language based on the explanation script according to the course materials uploaded by the teacher, including: Identify information in the course materials using a preset file identification tool; Based on the information in the course materials, generate preliminary description content based on the multimodal macromodel; According to the information in the course materials and the preliminary description content, the basic text and the control function are generated based on a generative language model; the control function includes an initiating interaction function and a providing learning resource function.

7. A large-model multi-agent driven adaptive online classroom construction system, characterized in that: Including explanation script generation agent, teacher agent, classmate agent and director agent; The explanation script generating agent is used to generate an educational agent control language based on the explanation script according to the course materials uploaded by the teacher; The teacher agent is used to automatically complete teaching and adaptive question-answering according to the contextual multi-round dialogue in the language and adaptive online classroom of the education agent; The classmate agent is used to generate content that activates the classroom atmosphere based on multiple rounds of dialogues in the context; The director agent is used to control the agent-generated content displayed in the adaptive online classroom according to multiple rounds of dialogues in the context; the explanation script generation agent, the teacher agent, the classmate agent and the director agent are all based on a large model and use context prompt learning or fine-tuning methods to execute corresponding steps.

8. The large-model multi-agent driven adaptive online classroom construction system according to claim 7 is characterized in that: It also includes an assistant teaching agent; the assistant teaching agent is used to generate auxiliary teaching content based on multiple rounds of dialogues in the context.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the large-model multi-agent driven adaptive online classroom construction method as described in any one of claims 1 to 6.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the large-model multi-agent driven adaptive online classroom construction method as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Diabetic retinopathy classification method by using super lightweight SqueezeNet network

    CN108537282A

  • Online course generation system and device based on artificial intelligence

    CN113222790A

  • Online teaching system based on artificial intelligence and virtual reality technology

    CN116934540A

  • AI intelligent interactive lesson system and method based on large language model

    CN117114933A

Cited By

  • Mathematical physics explanation video generation method and device based on multi-agent cooperation

    CN119277169A

  • Virtual human stylized interactive long-term emotion accompanying dialogue generation method

    CN121168497A

  • Learning progress updating system, method and device and electronic equipment

    CN121981500A

  • A learning progress updating system, method, device and electronic equipment

    CN121981500B

  • system

    JP2026155686A