A lesson plan generation method and system based on multi-agent cooperation

Abstract

The application relates to the technical field of teaching plan generation, in particular to a teaching plan generation method and system based on multi-agent cooperation. The method comprises the following steps: obtaining teaching task parameters, constructing a global teaching task description, and parsing the global teaching task description into a plurality of teaching content generation sub-tasks with associated dependency relationships to form a task scheduling sequence; a plurality of agents cooperatively complete each sub-task in sequence to generate a teaching plan draft, and implement cooperative evaluation on the teaching plan draft to form evaluation feedback; according to the evaluation feedback, the content of the teaching plan is directionally reconstructed and optimized, the optimization result is written into accumulated context to constrain the subsequent generation process, the global reflection evaluation of the complete teaching plan is finally carried out, and the final teaching plan generation result is output. The application realizes efficient generation and global consistency optimization of the content of the teaching plan, significantly improves the quality, coherence and automation degree of the teaching plan generation, effectively reduces the cost of manual compilation, and improves the overall quality of the teaching plan.

Description

Technical Field

[0001] This application relates to the field of lesson plan generation technology, and in particular to a method and system for generating lesson plans based on multi-agent collaboration. Background Technology

[0002] In related technologies, with the continuous development of artificial intelligence technology and intelligent education applications, the generation of teaching content based on generative artificial intelligence has gradually become an important technical means to assist teachers in lesson preparation and instructional design. This type of technology typically uses large-scale pre-trained language models to semantically analyze the teaching task parameters input by teachers, such as grade level, subject, and topic, and generates lesson plan texts containing teaching objectives, teaching processes, and arrangements for teaching activities. This, to some extent, improves the efficiency of lesson plan generation and reduces the burden on teachers in repetitive lesson preparation work.

[0003] However, most existing lesson plan generation methods rely on introspective reasoning and one-time text generation mechanisms using a single model. These methods are significantly inadequate when dealing with instructional design, a task with high structure, stages, and constraints imposed by educational principles. On one hand, single-model generation lacks multi-perspective negotiation and independent verification capabilities. When deviations occur in early stages such as student learning analysis, setting learning objectives, or designing instructional activities, subsequent content often continues to extrapolate along predetermined assumptions, making it difficult to promptly identify and correct potential cognitive or logical problems. On the other hand, existing technologies often treat teaching content as independent fragments during lesson plan generation, lacking effective characterization of the sequential dependencies and logical connections between sub-tasks. This results in repetitive content, logical jumps, or hierarchical confusion, making it difficult to form structured lesson plans that conform to the actual teaching process. Furthermore, existing lesson plan generation methods generally lack multi-subject collaboration and systematic evaluation mechanisms. The generated results typically rely on a single generation process, making it difficult to promptly identify and correct problems such as deviations in learning objectives, unreasonable content, or mismatched teaching strategies. This often requires extensive manual intervention for repeated modifications and verification, resulting in low efficiency and inconsistent quality.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The main objective of this application is to propose a teaching plan generation method and system based on multi-agent collaboration, so as to achieve efficient generation and global consistency optimization of teaching plan content, improve the quality, coherence and automation of teaching plan generation, effectively reduce manual compilation costs and improve the overall quality of teaching plans.

[0006] To achieve the above objectives, one aspect of this application proposes a lesson plan generation method based on multi-agent collaboration, the method comprising the following steps:

[0007] Obtain teaching task parameters;

[0008] Based on the aforementioned teaching task parameters, a global teaching task description is constructed;

[0009] The global teaching task description is parsed into multiple teaching content sub-tasks with sequential dependencies, forming a task scheduling sequence;

[0010] According to the task scheduling sequence, multiple agents are invoked to execute the teaching content generation sub-task to generate the corresponding lesson plan content draft;

[0011] Based on the multi-agent system, collaborative review processing is performed on the draft lesson plan to generate evaluation feedback information.

[0012] Based on the evaluation feedback information, the draft lesson plan content is reconstructed and optimized to obtain the optimized lesson plan content result;

[0013] Write the lesson plan content results into the cumulative context, and continue to schedule the generation of lesson plans for subsequent subtasks under the constraints of the cumulative context until all subtasks are completed;

[0014] After completing all sub-tasks, a global reflection and evaluation process is performed on the completed lesson plan content to generate the final lesson plan result.

[0015] In some embodiments, the teaching task parameters include characteristics of the teaching object, teaching stage, subject type, teaching theme, and class hour constraints.

[0016] In some embodiments, parsing the global teaching task description into multiple sub-tasks that have sequential dependencies on teaching content generation, forming a task scheduling sequence, includes:

[0017] Based on the global teaching task description, teaching objective constraints, student learning constraints, and teaching process constraints are extracted to construct a set of teaching task constraints.

[0018] Based on the set of teaching task constraints, the lesson plan generation process is divided into multiple teaching content generation sub-tasks according to the teaching logic sequence.

[0019] Based on the teaching content, generate logical dependencies between subtasks and construct a directed dependency graph of subtasks to determine the execution order of tasks with sequential dependency constraints.

[0020] A task scheduling sequence is generated based on the task execution order.

[0021] In some embodiments, the teaching content generation subtask includes a teaching objective generation subtask, a teaching scenario introduction subtask, a teaching activity design subtask, a teaching evaluation design subtask, and a teaching reflection generation subtask.

[0022] In some embodiments, the step of invoking multiple agents to execute the teaching content generation subtask according to the task scheduling sequence to generate a corresponding lesson plan draft includes:

[0023] Based on the task scheduling sequence, sub-tasks are generated for each teaching content, and multi-agents with different teaching functions are invoked respectively.

[0024] Based on the multi-agent system, the task types and constraints of the sub-tasks generated from the teaching content are used to independently generate corresponding teaching content segments.

[0025] The aforementioned teaching content segments are compiled into a draft lesson plan content corresponding to the sub-tasks of generating the aforementioned teaching content.

[0026] In some embodiments, the multi-agent system includes an instructional planning agent, a content generation agent, a collaborative evaluation agent, a reflection and optimization agent, and an assessment and monitoring agent.

[0027] In some embodiments, writing the lesson plan content result into the accumulation context, and continuing to schedule the generation of lesson plans for subsequent subtasks under the constraints of the accumulation context until all subtasks are completed, includes:

[0028] Write the results of the lesson plan content into the cumulative context;

[0029] Based on the accumulated context, the subtasks for generating teaching content to be executed are judged with context awareness. Subtasks with strong causal dependencies are determined as serially executed subtasks, and relatively independent subtasks are determined as parallelly executed subtasks.

[0030] In the serial execution of subtasks, the tasks are executed sequentially according to the task scheduling sequence. After each subtask is completed, the corresponding optimized lesson plan content is written into the cumulative context to constrain the generation of lesson plans for subsequent subtasks.

[0031] In the parallel execution subtask, the lesson plan generation of the corresponding subtask is performed in parallel while keeping the cumulative context read-only, and the generation result is written to the preset lesson plan structure position;

[0032] After the serial execution subtask and the parallel execution subtask are completed, a consistency check is performed on the generated lesson plan content, and the cumulative context is updated after the check passes.

[0033] Based on the updated cumulative context, continue scheduling unfinished teaching content to generate subtasks until all subtasks in the task scheduling sequence are completed.

[0034] In some embodiments, the step of performing a global reflection and evaluation process on the completed lesson plan content after all sub-tasks are completed to generate a final lesson plan result includes:

[0035] After completing all sub-tasks, a multi-dimensional quantitative evaluation of the complete lesson plan is performed based on the TOPIC teaching quality evaluation model to obtain the overall evaluation results.

[0036] Determine whether the preset quality threshold is met based on the global evaluation results;

[0037] When the quality threshold is not met, the multi-agent is triggered to reconstruct the complete lesson plan content based on the global evaluation result, and the TOPIC teaching quality evaluation model is re-executed to form an iterative process of reflection and reconstruction.

[0038] When the quality threshold is met, the corresponding complete lesson plan content is determined as the final lesson plan generation result.

[0039] In some embodiments, the construction process of the TOPIC teaching quality evaluation model includes:

[0040] Based on a pre-defined teaching evaluation theory system, multiple evaluation dimensions for assessing lesson plan quality are determined. The teaching evaluation theory system includes the correctness of teaching content, the operability of teaching process, the rationality of teaching strategies, the adaptability of learning cognition, and the consistency of teaching objectives.

[0041] Based on the aforementioned multiple evaluation dimensions, quantitative scores for each evaluation dimension are calculated for the complete lesson plan content, forming a TOPIC dimension score vector;

[0042] A weighted fusion operation is performed on the TOPIC dimension scoring vector to obtain a comprehensive scoring result that represents the overall teaching quality;

[0043] Based on the TOPIC dimension scoring vector and the comprehensive scoring results, a TOPIC teaching quality evaluation model is trained and constructed.

[0044] To achieve the above objectives, another aspect of this application proposes a lesson plan generation system based on multi-agent collaboration, the system comprising:

[0045] The data acquisition module is used to obtain teaching task parameters;

[0046] A global construction module is used to construct a global teaching task description based on the teaching task parameters;

[0047] The task parsing module is used to parse the global teaching task description into multiple teaching content sub-tasks with sequential dependencies, forming a task scheduling sequence.

[0048] The agent execution module is used to call multiple agents to execute the teaching content generation sub-task according to the task scheduling sequence, so as to generate the corresponding lesson plan content draft;

[0049] The collaborative evaluation module is used to perform collaborative evaluation processing on the draft lesson plan based on the multi-agent system and generate evaluation feedback information.

[0050] The content optimization module is used to reconstruct and optimize the draft lesson plan content based on the evaluation feedback information to obtain the optimized lesson plan content result.

[0051] The context accumulation module is used to write the lesson plan content results into the accumulation context, and continue to schedule the generation of lesson plans for subsequent subtasks under the constraints of the accumulation context until all subtasks are completed.

[0052] The global reflection and evaluation module is used to perform a global reflection and evaluation process on the completed lesson plan after all sub-tasks are completed, and generate the final lesson plan result.

[0053] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0054] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0055] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0056] The embodiments of this application include at least the following beneficial effects: This application provides a method and system for generating lesson plans based on multi-agent collaboration. This invention obtains teaching task parameters, performs structured modeling of these parameters, and constructs a global teaching task description, achieving a unified expression of lesson plan generation goals, teaching constraints, and content structure, thus improving the overall consistency and relevance of lesson plan generation from the source. By parsing the global teaching task into multiple sub-tasks with sequential dependencies and forming a task scheduling sequence, the generation process of teaching content has a clear logical order and hierarchical relationship, avoiding the problems of fragmented lesson plan content and loose connections in existing technologies. This invention introduces a multi-agent collaboration mechanism to generate lesson plan drafts in parallel for different teaching content sub-tasks. Through collaborative evaluation and feedback fusion among multiple agents, it achieves multi-dimensional verification of the matching of teaching goals, the completeness of content, and the rationality of teaching strategies, thereby significantly improving the professionalism and teaching adaptability of lesson plan content. Furthermore, through reconstruction optimization based on evaluation feedback and a continuous generation mechanism under cumulative contextual constraints, the generated content can effectively guide subsequent sub-tasks, enhancing the coherence and consistency of the overall lesson plan. After completing all sub-tasks, a comprehensive reflection and evaluation process is performed on the complete lesson plan. This allows for further optimization from the perspectives of overall teaching goal achievement and content structure rationality, effectively reducing the cost of repeated manual modifications and improving the efficiency and quality of lesson plan generation. It is suitable for the need for high-quality, scalable automatic generation of lesson plans in complex teaching scenarios. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating a lesson plan generation method based on multi-agent collaboration provided in an embodiment of this application.

[0058] Figure 2 yes Figure 1 A flowchart illustrating step S8 in the process;

[0059] Figure 3 This is a schematic diagram of a lesson plan generation system based on multi-agent collaboration provided in an embodiment of this application. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0061] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0062] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0064] Before providing a detailed description of the embodiments of this application, some of the nouns and terms used in the embodiments of this application will be explained first. The nouns and terms used in the embodiments of this application shall be interpreted as follows:

[0065] TOPIC is an evaluation model used to quantitatively assess the teaching quality of lesson plan content in multiple dimensions. Its evaluation dimensions include the correctness of teaching content, the operability of teaching process, the rationality of teaching strategy, the adaptability of learning cognition, and the consistency of teaching objectives.

[0066] TaskParam is a data object that provides a structured representation of teaching task parameters. It is used to uniformly encapsulate information such as teaching object characteristics, teaching stage, subject type, teaching theme, and class hour constraints, and serves as the basic input condition for the lesson plan generation process.

[0067] The Rubric evaluation framework is a standardized evaluation system used to evaluate teaching activities or lesson plan content in a multi-dimensional and multi-level manner. It achieves standardized assessment of teaching quality through preset evaluation indicators and level descriptions.

[0068] e-ECD theory refers to electronic evidence-centered design theory, which is oriented towards achieving educational goals. It is used to guide the collaborative design of teaching activities and assessment processes to ensure consistency between teaching objectives, teaching activities and assessment evidence.

[0069] UBD refers to the "Understanding by Design" framework for curriculum and instructional design. It involves planning instructional content and activities backward from learning objectives and expected outcomes to enhance the goal orientation and overall consistency of instructional design.

[0070] Bloom's Taxonomy of Instructional Objectives is a theoretical framework that systematically categorizes instructional objectives according to cognitive levels. It describes learning objectives at different cognitive levels, such as knowledge, comprehension, application, analysis, evaluation, and creation.

[0071] ICADMAS refers to an intelligent collaborative and adaptive decision-making agent system (ICAMAS) used to support the generation and quality optimization of teaching content. It is a teaching plan generation method based on multi-agent collaboration. By comprehensively analyzing teaching task parameters, task dependencies, and evaluation feedback information, it realizes adaptive collaboration and dynamic scheduling of multiple agents in the teaching plan generation process.

[0072] Multi-agent refers to a collection of collaborative execution units consisting of multiple agents with independent decision-making capabilities and functional divisions. Under unified task objectives and constraints, these agents collaborate to complete complex lesson plan generation tasks.

[0073] This application provides a method and system for generating lesson plans based on multi-agent collaboration. This invention acquires teaching task parameters, performs structured modeling of these parameters, and constructs a global teaching task description. This achieves a unified expression of lesson plan generation goals, teaching constraints, and content structure, fundamentally improving the overall consistency and relevance of lesson plan generation. By parsing the global teaching task into multiple sub-tasks with sequential dependencies and forming a task scheduling sequence, the generation process of teaching content has a clear logical order and hierarchical relationship, avoiding the fragmented and loosely connected problems of lesson plan content in existing technologies. This invention introduces a multi-agent collaboration mechanism to generate lesson plan drafts in parallel for different teaching content sub-tasks. Through collaborative evaluation and feedback fusion among multiple agents, it achieves multi-dimensional verification of the matching of teaching goals, the completeness of content, and the rationality of teaching strategies, thereby significantly improving the professionalism and teaching adaptability of the lesson plan content. Furthermore, through reconstruction optimization based on evaluation feedback and a continuous generation mechanism under cumulative contextual constraints, the generated content can effectively guide subsequent sub-tasks, enhancing the coherence and consistency of the overall lesson plan. After completing all sub-tasks, a comprehensive reflection and evaluation process is performed on the complete lesson plan. This allows for further optimization from the perspectives of overall teaching goal achievement and content structure rationality, effectively reducing the cost of repeated manual modifications and improving the efficiency and quality of lesson plan generation. It is suitable for the need for high-quality, scalable automatic generation of lesson plans in complex teaching scenarios.

[0074] This application provides a method for generating lesson plans based on multi-agent collaboration, relating to the field of lesson plan generation technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or in-vehicle terminal, but is not limited thereto. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing a method for generating lesson plans based on multi-agent collaboration, but is not limited to the above forms.

[0075] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0076] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0077] Figure 1 This is an optional flowchart of a lesson plan generation method based on multi-agent collaboration provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S1 to S8:

[0078] S1: Obtain teaching task parameters; among which, teaching task parameters include teaching object characteristics, teaching stage, subject type, teaching theme and class hour constraints.

[0079] In this embodiment, the method first performs the step of obtaining teaching task parameters. These parameters are input by the user once before lesson plan generation begins, specifying the basic teaching conditions required for the current lesson plan generation. The teaching task parameters include at least the characteristics of the learning target, the grade level, the subject type, the teaching theme, and class time constraints. The characteristics of the learning target are used to characterize the students' grade level and cognitive foundation, while the class time constraints limit the overall duration of the teaching activities.

[0080] After obtaining the above teaching task parameters, this embodiment analyzes and standardizes each parameter. Specifically, the teaching stage is analyzed into the corresponding cognitive development stage to limit the language complexity and thinking depth in subsequent instructional design; the subject type and teaching theme are analyzed into identifiable subject semantic tags to limit the scope of knowledge that the teaching content is allowed to cover; and the class hour constraint information is converted into time constraints that can be used for the allocation of subsequent teaching links.

[0081] Furthermore, the parsed teaching task parameters are uniformly coded to form a complete set of teaching task parameters, which is then written into the initial context of the current lesson plan generation process. This set of teaching task parameters is continuously referenced as a constraint in subsequent steps to ensure that the generated lesson plan content always meets the established grade level requirements, subject standards, and time constraints.

[0082] Specifically, the teaching task parameters may include:

[0083] Characteristics of the students: such as "third-grade students who are active thinkers but have weak calculation skills";

[0084] Teaching stage: Third grade of primary school;

[0085] Subject type: Mathematics;

[0086] Teaching topic: Introduction to fractions;

[0087] Class time limit: 40 minutes (which can be further broken down into time limits for introduction / new teaching / practice / summary).

[0088] Structure the parameter set into TaskParam (such as JSON), and write the optional learning information field as a key constraint for personalized teaching.

[0089] S2: Construct a global teaching task description based on teaching task parameters;

[0090] In this embodiment, after obtaining the teaching task parameters, a global teaching task description is constructed based on these parameters. The global teaching task description serves as a unified constraint on the teaching objectives, main teaching line, and teaching design direction of the entire lesson plan, and is the overall basis for each subsequent step in lesson plan generation.

[0091] When constructing a comprehensive teaching task description, the first step is to determine the cognitive progression path and teaching strategy orientation that the instructional design should follow, based on the characteristics of the teaching stage and the students. For example, for teaching tasks in primary school, the comprehensive teaching task description should clearly define that teaching activities should focus on concrete operations, intuitive representations, and gradual concept construction, avoiding abstract reasoning requirements that exceed students' cognitive level, thereby controlling the difficulty of teaching from the outset.

[0092] Subsequently, the teaching theme, subject type, and class hour constraints are integrated to form a comprehensive description of the teaching objective hierarchy, the organization of teaching segments, and the arrangement of teaching pace. This global teaching task description is written into and stored within the context of the current lesson plan generation process and is continuously referenced as a unified constraint in subsequent steps to ensure that the sub-tasks generated from the subsequently decomposed teaching content remain consistent in their objective orientation and teaching logic.

[0093] S3: Parse the global teaching task description into multiple teaching content sub-tasks with sequential dependencies, forming a task scheduling sequence;

[0094] Specifically, the global teaching task description is parsed into multiple sub-tasks with sequential dependencies, forming a task scheduling sequence, including:

[0095] S31: Based on the global teaching task description, extract teaching objective constraints, student learning constraints, and teaching process constraints to construct a set of teaching task constraints;

[0096] S32: Based on the set of teaching task constraints, the lesson plan generation process is divided into multiple teaching content generation sub-tasks according to the teaching logic sequence; among them, the teaching content generation sub-tasks include teaching objective generation sub-tasks, teaching situation introduction sub-tasks, teaching activity design sub-tasks, teaching evaluation design sub-tasks, and teaching reflection generation sub-tasks.

[0097] S33: Generate logical dependencies between subtasks based on each teaching content, construct a directed dependency graph of subtasks, and determine the execution order of tasks with sequential dependency constraints;

[0098] S34: Generate a task scheduling sequence based on the task execution order.

[0099] In this embodiment, the global teaching task description is parsed to extract key information that constrains the lesson plan generation process. This key information includes at least teaching objective constraints, student learning constraints, and teaching process constraints. Teaching objective constraints limit the level of teaching objectives, the scope of core knowledge points, and the focus on key points and difficulties. Student learning constraints limit the difficulty of teaching activities, language complexity, and cognitive scaffolding requirements. Teaching process constraints limit the composition, sequence, and pace of teaching steps. These constraints are uniformly organized to form a set of teaching task constraints, which are used for subsequent subtask division and execution control.

[0100] After constructing the set of teaching task constraints, the lesson plan generation process is broken down according to the general logical sequence of instructional design, dividing the overall generation process into multiple sub-tasks for generating teaching content. These sub-tasks include sub-tasks for generating teaching objectives, sub-tasks for introducing teaching scenarios, sub-tasks for designing teaching activities, sub-tasks for designing teaching evaluation, and sub-tasks for generating teaching reflections. Each sub-task corresponds to a specific teaching segment in the lesson plan, and its generated content is uniformly constrained by the aforementioned set of teaching task constraints, thus ensuring consistency in teaching objectives, cognitive levels, and teaching processes across the generated results of each sub-task.

[0101] Subsequently, the logical dependencies between the subtasks generated from each teaching content are analyzed, and a directed dependency graph is constructed accordingly. In this dependency graph, the order of subtasks is set based on the teaching logic. For example, the subtask for generating teaching objectives serves as a prerequisite for subsequent subtasks; the subtask for designing teaching activities depends on the subtasks for generating teaching objectives and introducing teaching scenarios; the subtask for designing teaching evaluation depends on the subtask for designing teaching activities; and the subtask for generating teaching reflection depends on the output results of the aforementioned subtasks. This approach makes the causal relationships in the instructional design explicit, avoiding a chaotic generation order.

[0102] After constructing the directed dependency graph of subtasks, the execution order of tasks satisfying the sequential dependency constraints is determined based on the dependencies, and a task scheduling sequence is generated accordingly. The task scheduling sequence clearly defines the execution order of the subtasks generated for each teaching content and serves as the basis for the execution of subsequent lesson plan generation steps. This transforms the lesson plan generation process from a holistic description into multiple sequentially executable subtasks, thereby improving the structural stability and logical consistency of the generated lesson plan.

[0103] S4: Based on the task scheduling sequence, call the multi-agent to execute the teaching content generation sub-task to generate the corresponding lesson plan content draft;

[0104] This includes invoking a multi-agent task to generate teaching content based on a task scheduling sequence, in order to generate a corresponding draft lesson plan, including:

[0105] S41: Based on the task scheduling sequence, generate sub-tasks for each teaching content and call upon multiple agents with different teaching functions respectively; among them, the multiple agents include teaching planning agent, content generation agent, collaborative evaluation agent, reflection and optimization agent, and evaluation and supervision agent;

[0106] S42: Based on the multi-agent system, generate sub-tasks with task types and constraints according to the teaching content, and independently generate corresponding teaching content segments.

[0107] S43: Compile the various teaching content segments to form a draft lesson plan content corresponding to the teaching content generation sub-task.

[0108] In this embodiment, after obtaining the task scheduling sequence, each teaching content generation sub-task is triggered sequentially according to the task scheduling sequence, and multiple agents with different teaching functions are invoked for different sub-tasks. Specifically, based on the task type of the sub-task, a teaching planning agent, a content generation agent, a collaborative evaluation agent, a reflection and optimization agent, and an evaluation and supervision agent are matched to it, so that different teaching functions form a division of labor and cooperation relationship in the execution of the same sub-task, thereby avoiding the deviation of teaching logic caused by a single generation perspective.

[0109] When executing a subtask to generate specific teaching content, the task type and constraints corresponding to that subtask are input into the invoked multi-agent system. These constraints include at least teaching objective constraints, student learning constraints, teaching process constraints, and contextual constraints formed by preceding subtasks. Within their respective functional areas, the multi-agent system independently generates corresponding teaching content segments around the same subtask. This includes targeted generation of content such as teaching objectives, teaching contexts, teaching activities, or teaching evaluations. During the generation process, the agents mutually verify whether the generated content conforms to the established constraints to ensure the relevance and consistency of the teaching content.

[0110] After completing the teaching content generation sub-task, the teaching content fragments generated by each multi-agent are collected and organized to form a draft lesson plan corresponding to that sub-task. The draft lesson plan is organized according to the pre-set lesson plan structure, ensuring that the generated results are structurally complete and logically coherent. It serves as the input for subsequent collaborative review and restructuring optimization steps, thus providing a stable foundation for improving the quality of lesson plans in the next stage.

[0111] Specifically, the multi-agent system is as follows:

[0112] Instructional planning intelligent agent ( ):

[0113] Responsibilities: Simulate a teaching and research group leader, responsible for overall task decomposition, status management, context maintenance, and hybrid workflow scheduling.

[0114] Decision basis: User's feature tuples and domain knowledge base .

[0115] Execution logic: Using retrieval-enhanced generation technology, the system retrieves historical excellent lesson plan templates from the database, breaks down macro-level teaching objectives into a sequence of sub-tasks that conform to standard operating procedures, and determines the dependencies between tasks.

[0116] Output: Structured instruction set, JSON format task queue (Includes task ID, task type, and dependent prerequisite task ID).

[0117] Content-generating intelligent agents ( ):

[0118] Responsibilities: To serve as a key member of the discipline, to use ZPD theory to build an initial cognitive scaffold, and to complete the production of real and specific content.

[0119] Input: Current cumulative context Specific generation instructions.

[0120] Execution logic: Employs a zero-shot thought chain strategy, based on accumulated contextual information. Derive the chain of teaching intentions for the current module, and then fill in the specific content.

[0121] Output: Structured initial draft It includes the name of the teaching segment, the design intention, and a description of the specific teaching activities.

[0122] Collaborative evaluation agent ( ):

[0123] Responsibilities: To simulate a review expert, introduce the perspective of an outsider, and scan the draft based on Bloom's Taxonomy and curriculum standards.

[0124] Input: Initial draft Curriculum standards constraints.

[0125] Assessment criteria: Bloom's Taxonomy of Educational Objectives is used as an external tool.

[0126] Execution logic: For existing text Perform logical verification and locate logical breakpoints through CoT multi-step reasoning (such as "the inquiry activity lacks support from higher-order thinking").

[0127] Output: Critical Vector (Includes a list of logical vulnerabilities, warnings of cognitive biases, and suggestions for improvement).

[0128] Reflecting on and optimizing intelligent agents ( ):

[0129] Responsibilities: To emulate the role of a reflective practitioner in processing feedback, resolving cognitive conflicts, and implementing in-depth restructuring and refinement work.

[0130] Decision-making strategy: dialectical weighing.

[0131] enter: , .

[0132] Execution logic: Synthesis and Integrate and discard conflicting opinions to reconcile cognitive contradictions. If differing opinions point to opposing teaching objectives, then yield to the teaching objectives; if logical fallacies are pointed out, then revise accordingly.

[0133] Output: The locally optimal solution after logic hardening .

[0134] Evaluation and supervision agents ( ):

[0135] Responsibilities: To simulate the role of an expert in quality monitoring, to use the TOPIC teaching quality evaluation model to conduct quantitative analysis and evaluation, and to determine whether to trigger global iteration work based on this.

[0136] Execution logic: Based on the TOPIC teaching quality evaluation model, a reflection trajectory (scoring reasons, reasoning judgments) is generated through CoT.

[0137] Output: TOPIC scoring matrix Global optimization commands.

[0138] S5: Based on multi-agent technology, collaborative review and processing of lesson plan content drafts are performed to generate evaluation feedback information;

[0139] In this embodiment, after completing a sub-task of generating a certain teaching content and obtaining the corresponding draft lesson plan, the process proceeds to the collaborative review step. The collaborative review process is used to check the draft lesson plan for consistency and rationality from multiple perspectives before formally entering the optimization and restructuring phase, in order to identify potential deviations in teaching logic, cognitive mismatches, or structural defects.

[0140] In practice, the draft lesson plan is analyzed item by item based on pre-defined evaluation dimensions. These dimensions include at least the degree of alignment with teaching objectives, the rationality of teaching activities, the suitability for cognitive levels, and the coherence of the teaching process. Specifically, the degree of alignment with teaching objectives is used to determine whether each teaching activity in the draft serves the established teaching objectives; and the suitability for cognitive levels is used to determine whether the activity design aligns with the cognitive characteristics of students at the corresponding grade level.

[0141] During the review process, the draft lesson plan is scanned and logically verified section by section to identify potential problems, such as overly difficult teaching activities, unclear transitions between activities, and evaluation methods that are disconnected from the teaching objectives. For each type of problem, this embodiment generates corresponding evaluation feedback information, which records the problem type, location, and suggested directions for modification in a structured format.

[0142] Furthermore, the multi-dimensional evaluation results are summarized and organized to form a complete set of evaluation feedback information. This set of evaluation feedback information does not directly modify the original draft lesson plan content, but rather serves as the input basis for the next step of reconstruction and optimization, thereby avoiding the introduction of new teaching deviations during the evaluation stage and ensuring the objectivity and traceability of the evaluation results.

[0143] S6: Based on the evaluation feedback, the draft lesson plan content is restructured and optimized to obtain the optimized lesson plan content result;

[0144] In this embodiment, after obtaining the evaluation feedback information set, the lesson plan content is restructured and optimized. Restructuring and optimization is not simply polishing the draft, but rather, based on the evaluation feedback information, targeted adjustments and restructuring are made to problematic teaching structures, activity designs, or logical sequences.

[0145] In practice, the evaluation feedback information is first analyzed to distinguish different types of feedback. When the evaluation feedback indicates that the teaching objectives and teaching activities are inconsistent, the design or sequence of teaching activities is adjusted first to align them with the established teaching objectives. When the evaluation feedback indicates that the cognitive load is too high or does not conform to the characteristics of the learning stage, the content is revised by simplifying task requirements, introducing intuitive operations, or adding transitional teaching scaffolds.

[0146] During the restructuring process, the decision-making principle of "goal priority and logical consistency" is followed. When there are potential conflicts between different evaluation feedback information, the teaching objectives and cognitive paths defined in the overall teaching task description are used as the basis for adjudication, and conflicting content is selected or integrated to avoid local optimization from destroying the overall teaching structure.

[0147] After the refactoring and optimization are completed, the method generates optimized lesson plan content and performs integrity and consistency checks on it to ensure that the optimized content meets the constraints in terms of teaching logic, cognitive level, and teaching process. The optimized lesson plan content, as a reliable output, is used as the basis for subsequent cumulative context updates and scheduling of the next teaching content generation subtask.

[0148] Specifically, Receive user input ,and By combining semantic augmentation, a global context vector for the entire dialogue is created. This vector serves as the initial "a priori axiom" of the system, thereby constraining the generated content to remain within the students' cognitive level.

[0149] ;

[0150] For example, "third grade" can be mapped to "the concrete operational stage, which requires the use of more visual teaching aids".

[0151] Initialize task queue The task of generating lesson plans is broken down into... Sub-tasks, such as: (like: Set goals, :Design import...).

[0152] Initialize the cumulative context container .

[0153] Generate subtasks for each teaching content. At that time, the atomic "generate-adversarial-optimize" ternary protocol is executed to resolve the logic breakage problem:

[0154] generate: Based on the current context Generate a first draft .

[0155] ;

[0156] in, This represents the draft lesson plan content obtained in the t-th round of lesson content generation; t represents the iteration round index in the lesson plan generation process. This represents a function that generates lesson plan content.

[0157] confrontation: Check if the initial draft conforms to the curriculum standards and whether there is a disconnect between "teaching" and "learning," and output a critique vector. .

[0158] ;

[0159] in, This represents the evaluation feedback information or evaluation result vector obtained after performing collaborative review on the draft lesson plan content in the t-th generation round; This represents the collaborative review processing function; This represents the set of teaching evaluation criteria used to constrain collaborative review processes.

[0160] optimization: Based on the initial draft and feedback, a dialectical restructuring was carried out, resulting in the revised module content. This step simulates the process of reaching consensus through social negotiation.

[0161] ;

[0162] in, This represents the optimized lesson plan content obtained after refactoring and optimizing the draft lesson plan content based on evaluation feedback information in the t-th generation round; This represents the refactoring and optimization processing function.

[0163] S7: Write the lesson plan content results into the cumulative context, and continue to schedule the generation of lesson plans for subsequent subtasks under the constraints of the cumulative context until all subtasks are completed;

[0164] This involves writing the lesson plan content into a cumulative context, and then, under the constraints of the cumulative context, continuing to schedule the generation of lesson plans for subsequent subtasks until all subtasks are completed, including:

[0165] S71: Write the lesson plan content results into the cumulative context;

[0166] S72: Based on the cumulative context, perform context-aware judgment on the subtasks generated from the teaching content to be executed, identify subtasks with strong causal dependencies as serial execution subtasks, and identify relatively independent subtasks as parallel execution subtasks.

[0167] S73: In the serial execution of subtasks, they are executed sequentially according to the task scheduling sequence, and after each subtask is completed, the corresponding optimized lesson plan content is written into the cumulative context to constrain the generation of lesson plans for subsequent subtasks;

[0168] S74: In parallel execution of subtasks, the lesson plan generation of the corresponding subtask is performed in parallel while keeping the cumulative context read-only, and the generation result is written to the preset lesson plan structure position.

[0169] S75: After the serial and parallel execution of subtasks are completed, perform a consistency check on the generated lesson plan content, and update the cumulative context after the check passes;

[0170] S76: Based on the updated cumulative context, continue to schedule unfinished teaching content to generate subtasks until all subtasks in the task scheduling sequence are completed.

[0171] In this embodiment, the lesson plan content is written into the cumulative context. During the writing process, the method categorizes the generated results according to the preset lesson plan structure and simultaneously extracts key teaching information from the results, including key teaching objectives, core concept expressions, main teaching activity logic, and time constraints. This forms the core semantic information in the cumulative context that can constrain subsequent generation, rather than simply stacking the entire text content.

[0172] After the lesson plan content is written, context-aware judgment is performed on the subtasks generated from the teaching content that have not yet been executed, based on the accumulated context. The judgment is used to identify the degree of dependence of each subtask on the existing lesson plan content. When the generation of a subtask must strictly depend on the semantic output of the preceding teaching content, it is identified as a serially executed subtask with a strong causal dependency; when a subtask only depends on the completed core content and there is no direct causal influence between it and other subtasks to be executed, it is identified as a relatively independent parallel execution subtask.

[0173] For subtasks determined to be executed sequentially, they are executed strictly according to the task scheduling sequence. During the execution of each sequential subtask, the latest cumulative context is used as a constraint, ensuring that the subtask inherits the target audience, conceptual scope, and teaching pace of the preceding teaching content during its generation. Whenever a sequential subtask is completed and generates an optimized lesson plan, the result is immediately written into the cumulative context to continuously update the constraints on the generation of subsequent subtasks.

[0174] For subtasks determined to be executed in parallel, the corresponding lesson plan generation is performed in parallel while maintaining read-only access to the cumulative context. During parallel execution, each parallel subtask is only allowed to read the same cumulative context snapshot, without writing to or modifying it, thus avoiding issues of mutual overwriting or semantic pollution during parallel generation. After parallel generation is completed, the generation results of each parallel subtask are written to the preset lesson plan structure location, ensuring that the outputs of different subtasks do not conflict with each other at the structural level.

[0175] After both serial and parallel execution of subtasks are completed, a consistency check is performed on the currently generated lesson plan content. The consistency check includes at least checking the consistency between teaching objectives and activities, teaching evaluations, the consistency of key concept expressions, and whether the time allocation conforms to the class hour constraints. When the check passes, the key summary of the parallel-generated content is integrated with the latest summary of the serially generated content, and the cumulative context is updated; when the check fails, conflicting content is marked for subsequent adjustments.

[0176] After completing the cumulative context update, the remaining teaching content is scheduled to generate subtasks based on the updated cumulative context. The process of context-aware judgment, serial or parallel execution, and consistency verification is repeated until all subtasks in the task scheduling sequence are completed. In this way, the lesson plan generation process ensures the consistency of the overall teaching logic while taking into account the generation efficiency, ultimately resulting in a lesson plan with a complete structure and coherent content.

[0177] Specifically, a predefined causal linked list based on teaching business logic divides tasks into two categories.

[0178] Execute subtasks sequentially ( The content generation of this type of task strictly depends on the semantic output of the preceding steps, and belongs to the module with strong causal dependence (e.g., teaching objectives → teaching activities → teaching evaluation).

[0179] Parallel execution of subtasks These types of tasks are usually relatively independent auxiliary modules or extensions of the same core content in different dimensions. For example, both "blackboard design" and "homework assignment" depend on the core "new teaching segment," but there is no direct causal relationship between them. Therefore, after the "new teaching segment" is completed, the two can be executed in parallel.

[0180] S8: After completing all sub-tasks, perform a global reflection and evaluation process on the completed lesson plan content to generate the final lesson plan result.

[0181] Among them, reference Figure 2 As shown, after completing all sub-tasks, a global reflection and evaluation process is performed on the completed lesson plan content to generate the final lesson plan result, including:

[0182] S81: After completing all sub-tasks, based on the TOPIC teaching quality evaluation model, perform a multi-dimensional quantitative assessment of the complete lesson plan content to obtain the overall evaluation results;

[0183] S82: Determine whether the preset quality threshold is met based on the global evaluation results;

[0184] S83: When the quality threshold is not met, the multi-agent is triggered to reconstruct the complete lesson plan content based on the global evaluation results, and the TOPIC teaching quality evaluation model is re-executed to form an iterative process of reflection and reconstruction.

[0185] S84: When the quality threshold is met, the corresponding complete lesson plan content is determined as the final lesson plan generation result.

[0186] Specifically, the construction process of the TOPIC teaching quality evaluation model includes:

[0187] Based on a pre-defined teaching evaluation theory system, multiple evaluation dimensions for assessing lesson plan quality are determined. The teaching evaluation theory system includes the correctness of teaching content, the operability of teaching process, the rationality of teaching strategies, the adaptability of learning cognition, and the consistency of teaching objectives.

[0188] Based on multiple evaluation dimensions, quantitative scores for each evaluation dimension are calculated for the complete lesson plan content, forming a TOPIC dimension score vector;

[0189] A weighted fusion operation is performed on the TOPIC dimension score vector to obtain a comprehensive score result that represents the overall teaching quality;

[0190] Based on the TOPIC dimension scoring vector and the comprehensive scoring results, a TOPIC teaching quality evaluation model is trained and constructed.

[0191] In this embodiment, after all the teaching content generation subtasks in the task scheduling sequence have been executed, a complete lesson plan is obtained. The complete lesson plan includes at least the following components: teaching objectives, introduction of the teaching context, arrangement of teaching activities, design of teaching evaluation, and key points for teaching reflection. Each part has maintained logical consistency through accumulated contextual constraints in previous steps. Subsequently, the method enters a global reflection and evaluation phase to uniformly verify the teaching quality of the entire lesson plan.

[0192] During the overall reflection and evaluation phase, a multi-dimensional quantitative evaluation of the complete lesson plan content is conducted based on the TOPIC teaching quality evaluation model. In specific implementation, according to the pre-set teaching evaluation theoretical system, the lesson plan content is comprehensively scanned and analyzed from multiple evaluation dimensions, such as the correctness of teaching content, the operability of teaching process, the rationality of teaching strategy, the adaptability of learning cognition, and the consistency of teaching objectives, thereby obtaining a global evaluation result reflecting the overall quality of the lesson plan.

[0193] The overall evaluation results should include at least quantitative scores for each evaluation dimension and corresponding evaluation conclusions, indicating the strengths and weaknesses of the lesson plan across different teaching quality dimensions. When generating the overall evaluation results, the overall teaching task description and the main constraints within the cumulative context should be considered to determine whether the lesson plan experienced issues such as goal deviation, cognitive overload, or imbalance in teaching processes during the generation process, thereby improving the reliability of the evaluation results.

[0194] After obtaining the overall evaluation results, a pass / fail determination is made based on preset quality thresholds. These quality thresholds may include an overall comprehensive score threshold and minimum requirements for key evaluation dimensions, ensuring that the lesson plan meets overall quality standards while avoiding critical defects that could affect classroom implementation. When the determination result shows that the current lesson plan content meets the quality thresholds, the method directly confirms that complete lesson plan content as the final lesson plan generation result.

[0195] When the evaluation results indicate that the quality threshold is not met, a targeted restructuring of the complete lesson plan is triggered based on the global evaluation results. This targeted restructuring focuses on the weaknesses explicitly identified in the evaluation results, making specific adjustments to the relevant teaching content rather than completely rewriting the entire lesson plan. This avoids ineffective modifications that could disrupt the overall teaching logic.

[0196] After completing the targeted reconstruction, the updated complete lesson plan content is re-evaluated using the TOPIC teaching quality assessment, and global evaluation results and threshold judgment results are generated again, thus forming an iterative process of "reflection-reconstruction-reassessment". To ensure the stability of the generation process, the method can set a maximum limit on the number of iterations. If the quality threshold is not met after reaching the maximum number of iterations, the current optimal version and corresponding improvement prompts are output.

[0197] When the evaluation results of a certain round meet the preset quality threshold, the corresponding complete lesson plan content is determined as the final lesson plan generation result and output. Through the above-mentioned global reflection evaluation and iterative optimization process, the quality of lesson plans in terms of correctness, operability, teaching rationality, and overall consistency can be effectively improved, avoiding structural defects caused by one-time generation, thereby obtaining lesson plan generation results that better meet actual teaching needs.

[0198] Specifically, the TOPIC teaching quality evaluation model incorporates mainstream teaching theories such as the Rubric evaluation framework, e-ECD (evidence-based instructional design), UBD (backward instructional design for understanding), and Bloom's taxonomy of educational objectives, reflecting the educational scientific nature of the evaluation standards. The logical connections between the various scoring dimensions are as follows:

[0199] T(Truth): Check if the knowledge points retrieved from the knowledge base are correct.

[0200] O (Operability): The feasibility of evaluating the duration and steps involved.

[0201] P (Pedagogy): Assess whether the teaching strategies used align with students' cognitive development (e.g., Piaget's stages theory).

[0202] I (Insight): This assesses whether there is insight into students' cognition and whether guidance is pre-planned to address students' misconceptions.

[0203] C (Cohesion): Combining vector computation with logical verification. The system calculates the cosine similarity between the teaching objective vector and the teaching activity vector as an objective reference. Simultaneously, prompts require the model to qualitatively judge whether the evaluation task truly reflects the requirements of the teaching objectives based on the UBD (Understanding by Design) reverse engineering concept, ensuring the integration of teaching, learning, and assessment.

[0204] Calculate the weighted total score .

[0205] ;

[0206] in, This represents the scoring threshold parameter used to determine whether the content of the lesson plan meets the output requirements. The scoring threshold is the minimum passing standard preset by the system. When the overall score is lower than this threshold, further reflection and reconstruction of the lesson plan content is triggered to avoid directly outputting the generated results. This represents the iteration count value in the current lesson plan generation and reflection optimization process, used to record the number of reflection iterations that have been executed in the multi-agent collaborative generation and optimization process; This represents the maximum number of iterations allowed for reflective optimization. When the current iteration reaches this maximum value, further reflective optimization will be terminated even if the overall score has not yet reached the score threshold, in order to prevent excessive iteration or semantic drift in the generation process.

[0207] Generate global feedback matrix It includes specific modification instructions (such as "The import process takes too long, compress it to 5 minutes").

[0208] The adaptive reconstruction is as follows:

[0209] Analysis Rescheduled The weak modules in the lesson plan were analyzed in detail, restructured, and then rewritten and updated in a targeted manner. This process is repeated until the maximum number of iterations is reached. .

[0210] In this embodiment, the default maximum number of iterations is 3. Experimental results show that too many iterations (such as more than 5) can easily cause the large model to gradually deviate from the original user intent after multiple modifications, resulting in overfitting and semantic drift. Therefore, setting the number of iterations to 3 is the best compromise between maintaining the original intent and optimizing the results.

[0211] In this embodiment, taking the People's Education Press primary school mathematics textbook as an example, five core themes, such as the meaning of decimals and the area of ​​parallelograms, are selected for testing.

[0212] 1. The experimental setup is as follows:

[0213] ChatGPT-5.0 (OpenAI): Represents the most powerful general-purpose monolithic model currently available.

[0214] Gemini-3.0 (Google): Represents a strong multimodal model with long-term contextual understanding capabilities.

[0215] DeepSeek-R1: A representative of open-source models with strong chained reasoning capabilities.

[0216] The configuration in this embodiment is as follows:

[0217] Basic model: GPT-4o-mini (as the core of the intelligent agent, balancing cost and speed).

[0218] Reflection Rounds: Maximum Number of Iterations .

[0219] Knowledge Base: Loads data from the People's Education Press primary school mathematics textbooks and curriculum standards.

[0220] The evaluation indicators are as follows:

[0221] Objective dimension: Based on the LLM-as-a-Judge paradigm, the TOPIC teaching quality evaluation model is used for five-dimensional quantitative scoring (with a maximum score of 100).

[0222] Subjective dimension: 30 experienced front-line mathematics teachers were invited to conduct double-blind reviews.

[0223] 2. The experimental results and analysis are as follows:

[0224] Based on the experiments, the overall performance of different models is compared as follows:

[0225] Refer to Table 1: Overall performance comparison of different models;

[0226]

[0227] Table 1 shows the comparison results of the lesson plan generation method proposed in this invention with several mainstream latest baseline methods in terms of lesson plan generation quality. The comparison methods include ChatGPT-5.0, Gemini-3.0, and DeepSeek-R1, which represent the current typical general generative model, long context understanding model, and reinforced chain inference model, respectively; the method of this invention is compared using ICADMAS as the system version identifier.

[0228] As shown in Table 1, the evaluation adopted the TOPIC teaching quality evaluation model, which quantitatively scored the generated lesson plans from five dimensions: accuracy (T), operability (O), pedagogy (P), interactivity (I), and consistency (C), and calculated the average score to reflect the overall teaching quality level. Among them, accuracy is used to measure the degree of consistency between teaching content and subject knowledge and curriculum standards; operability is used to measure the feasibility of teaching process in real classroom environment; pedagogy is used to measure whether teaching strategies and instructional design conform to educational principles; interactivity is used to measure the adequacy of teacher-student interaction and learning participation design; and consistency is used to measure the overall coordination between teaching objectives, teaching activities, and teaching evaluation.

[0229] The comparison results show that the three baseline methods all achieve high levels across all evaluation dimensions, indicating that the general-purpose model has a certain basic capability in lesson plan generation tasks. However, these methods still have significant shortcomings in terms of consistency and pedagogical relevance. These shortcomings are mainly manifested in weak alignment or loose structure between teaching objectives, teaching activities, and evaluation processes, and the difficulty in maintaining a consistent teaching thread in long-term lesson plan design.

[0230] In comparison, the proposed method (ICADMAS) achieved significantly higher scores across all five evaluation dimensions, with particularly noticeable improvements in instructional (94.0), interactivity (90.5), and consistency (93.8) dimensions. The final average score reached 92.96, significantly higher than the comparative methods. This result demonstrates that the proposed method, through task-dependent modeling, context-aware scheduling, multi-agent collaborative generation, and cumulative context constraint mechanisms, effectively alleviates the structural dissonance problem that traditional single-agent generation methods are prone to in long-term instructional design.

[0231] By further integrating a TOPIC-driven global reflection evaluation and iterative reconstruction mechanism, this invention enables multi-dimensional quantitative evaluation of lesson plans after generation and targeted optimization of weak points. This significantly improves the pedagogical rationality and overall consistency of lesson plans without sacrificing generation efficiency. Experimental results verify the effectiveness of the multi-agent collaborative mechanism and reflection closed-loop mechanism in lesson plan generation tasks, demonstrating that the method of this invention outperforms existing mainstream technologies in terms of generation quality, structural stability, and pedagogical usability.

[0232] It should be noted that, in order to verify the necessity of multi-agent collaboration and global reflective evaluation, an ablation experiment was conducted in this embodiment.

[0233] Refer to Table 2: Comparison results of different system variants in ablation experiments;

[0234]

[0235] Table 2 presents the results of the ablation experiment conducted to verify the necessity of the multi-agent collaboration mechanism and the reflective optimization mechanism. The experiment involved removing or weakening key technical mechanisms in the method of this invention one by one while keeping other conditions constant, observing the changes in the average score of the generated lesson plans under the TOPIC teaching quality evaluation model, thereby analyzing the contribution of each technical feature to the overall quality of the generated lesson plans.

[0236] Experimental variant B1 corresponds to the scenario of "no multi-agent collaboration," which removes role division and retains only a single generating agent to complete the lesson plan generation. The experimental results show that the average score dropped to 81.50, a decrease of 11.46 points compared to the complete method, representing the largest drop. This result indicates that a multi-agent collaboration mechanism is a core factor in improving the logical depth and multi-perspective rationality of lesson plans. A single generating agent struggles to simultaneously handle multiple teaching functions such as instructional planning, content generation, evaluation, and optimization, easily leading to lesson plans with a simplistic structure and insufficient logical depth.

[0237] Experimental variant B2 corresponds to the "no-reflection optimization" scenario, which removes the iterative loop based on the TOPIC teaching quality evaluation model, retaining only the one-time generated results. The results show that the average score decreased to 85.20, a decrease of 7.76 points compared to the complete method. This result indicates that the global reflective assessment and targeted reconstruction mechanism plays a crucial role in correcting potential knowledge biases, logical omissions, and teaching mismatches during the generation process. Without a reflective loop, although the lesson plan may have a complete structure, it is prone to hidden defects at the detail level.

[0238] Experimental variant B3 corresponds to the "single-agent reflection" scenario. It retains the reflection process but does not introduce multi-role collaboration; only a single agent completes self-reflection and optimization. The experimental results show an average score of 88.40, a decrease of 4.56 points compared to the complete method. This result indicates that introspective reflection relying solely on a single agent is insufficient to fully expose potential problems, easily falls into a self-verification loop, has limited optimization scope, and fails to achieve the effects of multi-perspective collaborative review.

[0239] Experiment B4 corresponds to the complete method of this invention, retaining the multi-agent collaboration mechanism, context-aware scheduling mechanism, and TOPIC-based reflection-reconstruction closed-loop mechanism. The final average score reached 92.96, the highest among all experimental variants. This result indicates that the multi-agent collaboration and reflection mechanisms are not simply superimposed, but rather produce a significant non-linear gain effect, jointly ensuring high-quality output of the lesson plan in terms of accuracy, operability, teaching rationality, and overall consistency.

[0240] In summary, the above ablation experiment results experimentally verify the necessity and synergy of the present invention in the lesson plan generation task, proving that the present invention, through the combination of multi-agent collaboration and reflection loop, can significantly outperform existing single-agent generation or weak reflection generation technical solutions, and has outstanding technical effects and practical value.

[0241] Please see Figure 3 This application also provides a lesson plan generation system based on multi-agent collaboration, the system comprising:

[0242] The data acquisition module is used to obtain teaching task parameters;

[0243] The global construction module is used to build a global teaching task description based on teaching task parameters;

[0244] The task parsing module is used to parse the global teaching task description into multiple teaching content subtasks with sequential dependencies, forming a task scheduling sequence.

[0245] The agent execution module is used to call multiple agents to execute sub-tasks for generating teaching content according to the task scheduling sequence, so as to generate corresponding lesson plan content drafts;

[0246] The collaborative review module is used to perform collaborative review of draft lesson plans based on multiple agents and generate evaluation feedback information.

[0247] The content optimization module is used to reconstruct and optimize the draft lesson plan content based on evaluation feedback information, resulting in an optimized lesson plan content.

[0248] The context accumulation module is used to write the lesson plan content results into the accumulation context, and continue to schedule the generation of lesson plans for subsequent subtasks under the constraints of the accumulation context until all subtasks are completed.

[0249] The global reflection and evaluation module is used to perform a global reflection and evaluation process on the completed lesson plan after all sub-tasks are completed, and generate the final lesson plan result.

[0250] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0251] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0252] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0253] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0254] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0255] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0256] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0257] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0258] This application provides a method and system for generating lesson plans based on multi-agent collaboration. This invention acquires teaching task parameters, performs structured modeling of these parameters, and constructs a global teaching task description. This achieves a unified expression of lesson plan generation goals, teaching constraints, and content structure, fundamentally improving the overall consistency and relevance of lesson plan generation. By parsing the global teaching task into multiple sub-tasks with sequential dependencies and forming a task scheduling sequence, the generation process of teaching content has a clear logical order and hierarchical relationship, avoiding the fragmented and loosely connected problems of lesson plan content in existing technologies. This invention introduces a multi-agent collaboration mechanism to generate lesson plan drafts in parallel for different teaching content sub-tasks. Through collaborative evaluation and feedback fusion among multiple agents, it achieves multi-dimensional verification of the matching of teaching goals, the completeness of content, and the rationality of teaching strategies, thereby significantly improving the professionalism and teaching adaptability of the lesson plan content. Furthermore, through reconstruction optimization based on evaluation feedback and a continuous generation mechanism under cumulative contextual constraints, the generated content can effectively guide subsequent sub-tasks, enhancing the coherence and consistency of the overall lesson plan. After completing all sub-tasks, a comprehensive reflection and evaluation process is performed on the complete lesson plan. This allows for further optimization from the perspectives of overall teaching goal achievement and content structure rationality, effectively reducing the cost of repeated manual modifications and improving the efficiency and quality of lesson plan generation. It is suitable for the need for high-quality, scalable automatic generation of lesson plans in complex teaching scenarios.

[0259] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0260] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0261] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0262] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0263] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0264] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for generating lesson plans based on multi-agent collaboration, characterized in that, The method includes the following steps: Obtain teaching task parameters; Based on the teaching task parameters, a global teaching task description is constructed; The global teaching task description is parsed into multiple teaching content sub-tasks with sequential dependencies, forming a task scheduling sequence; According to the task scheduling sequence, multiple agents are invoked to execute the teaching content generation sub-task to generate the corresponding lesson plan content draft; Based on the multi-agent system, collaborative review processing is performed on the draft lesson plan to generate evaluation feedback information. Based on the evaluation feedback information, the draft lesson plan content is reconstructed and optimized to obtain the optimized lesson plan content result; Write the lesson plan content results into the cumulative context, and continue to schedule the generation of lesson plans for subsequent subtasks under the constraints of the cumulative context until all subtasks are completed; After completing all sub-tasks, a global reflection and evaluation process is performed on the completed lesson plan content to generate the final lesson plan result; The step of writing the lesson plan content result into the cumulative context, and continuing to schedule the generation of lesson plans for subsequent subtasks under the constraints of the cumulative context until all subtasks are completed, includes: Write the results of the lesson plan content into the cumulative context; Based on the accumulated context, the subtasks for generating teaching content to be executed are judged with context awareness. Subtasks with strong causal dependencies are determined as serially executed subtasks, and relatively independent subtasks are determined as parallelly executed subtasks. In the serial execution subtasks, the tasks are executed sequentially according to the task scheduling sequence, and after each subtask is completed, the corresponding optimized lesson plan content is written into the cumulative context to constrain the generation of lesson plans for subsequent subtasks. In the parallel execution subtask, the lesson plan generation of the corresponding subtask is performed in parallel while keeping the cumulative context read-only, and the generation result is written to the preset lesson plan structure position; After the serial execution subtask and the parallel execution subtask are completed, a consistency check is performed on the generated lesson plan content, and the cumulative context is updated after the check passes. Based on the updated cumulative context, continue scheduling unfinished teaching content to generate subtasks until all subtasks in the task scheduling sequence are completed.

2. The method according to claim 1, characterized in that, The teaching task parameters include the characteristics of the teaching object, the teaching stage, the subject type, the teaching theme, and the time constraints.

3. The method according to claim 1, characterized in that, The step of parsing the global teaching task description into multiple sub-tasks with sequential dependencies to form a task scheduling sequence includes: Based on the global teaching task description, teaching objective constraints, student learning constraints, and teaching process constraints are extracted to construct a set of teaching task constraints. Based on the set of teaching task constraints, the lesson plan generation process is divided into multiple teaching content generation sub-tasks according to the teaching logic sequence. Based on the teaching content, generate logical dependencies between subtasks and construct a directed dependency graph of subtasks to determine the execution order of tasks with sequential dependency constraints. A task scheduling sequence is generated based on the task execution order.

4. The method according to claim 3, characterized in that, The teaching content generation subtasks include teaching objective generation subtasks, teaching scenario introduction subtasks, teaching activity design subtasks, teaching evaluation design subtasks, and teaching reflection generation subtasks.

5. The method according to claim 1, characterized in that, The step of invoking multiple agents to execute the teaching content generation subtask according to the task scheduling sequence to generate the corresponding lesson plan content draft includes: Based on the task scheduling sequence, sub-tasks are generated for each teaching content, and multi-agents with different teaching functions are invoked respectively. Based on the multi-agent system, the task types and constraints of the sub-tasks generated from the teaching content are used to independently generate corresponding teaching content segments. The teaching content segments are compiled to form a draft lesson plan content corresponding to the sub-tasks of generating the teaching content.

6. The method according to claim 5, characterized in that, The multi-agent system includes an instructional planning agent, a content generation agent, a collaborative evaluation agent, a reflection and optimization agent, and an assessment and monitoring agent.

7. The method according to claim 1, characterized in that, After completing all sub-tasks, a global reflection and evaluation process is performed on the completed lesson plan content to generate the final lesson plan result, including: After completing all sub-tasks, a multi-dimensional quantitative evaluation of the complete lesson plan is performed based on the TOPIC teaching quality evaluation model to obtain the overall evaluation results. Determine whether the preset quality threshold is met based on the global evaluation results; When the quality threshold is not met, the multi-agent is triggered to reconstruct the complete lesson plan content based on the global evaluation result, and the TOPIC teaching quality evaluation model is re-executed to form an iterative process of reflection and reconstruction. When the quality threshold is met, the corresponding complete lesson plan content is determined as the final lesson plan generation result.

8. The method according to claim 7, characterized in that, The construction process of the TOPIC teaching quality evaluation model includes: Based on a pre-defined teaching evaluation theory system, multiple evaluation dimensions for assessing lesson plan quality are determined. The teaching evaluation theory system includes the correctness of teaching content, the operability of teaching process, the rationality of teaching strategies, the adaptability of learning cognition, and the consistency of teaching objectives. Based on the aforementioned multiple evaluation dimensions, quantitative scores for each evaluation dimension are calculated for the complete lesson plan content, forming a TOPIC dimension score vector; A weighted fusion operation is performed on the TOPIC dimension scoring vector to obtain a comprehensive scoring result that represents the overall teaching quality; Based on the TOPIC dimension scoring vector and the comprehensive scoring results, a TOPIC teaching quality evaluation model is trained and constructed.

9. A lesson plan generation system based on multi-agent collaboration, characterized in that, The system is applied to the method as described in any one of claims 1-8, the system comprising: The data acquisition module is used to obtain teaching task parameters; A global construction module is used to construct a global teaching task description based on the teaching task parameters; The task parsing module is used to parse the global teaching task description into multiple teaching content sub-tasks with sequential dependencies, forming a task scheduling sequence. The agent execution module is used to call multiple agents to execute the teaching content generation sub-task according to the task scheduling sequence, so as to generate the corresponding lesson plan content draft; The collaborative evaluation module is used to perform collaborative evaluation processing on the draft lesson plan based on the multi-agent system and generate evaluation feedback information. The content optimization module is used to reconstruct and optimize the draft lesson plan content based on the evaluation feedback information to obtain the optimized lesson plan content result. The context accumulation module is used to write the lesson plan content results into the accumulation context, and continue to schedule the generation of lesson plans for subsequent subtasks under the constraints of the accumulation context until all subtasks are completed. The global reflection and evaluation module is used to perform a global reflection and evaluation process on the completed lesson plan after all sub-tasks are completed, and generate the final lesson plan result.

Images