CN-121683726-B - Teaching plan generation method and system based on multi-agent cooperation

CN121683726BCN 121683726 BCN121683726 BCN 121683726BCN-121683726-B

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 steps of obtaining teaching task parameters, constructing global teaching task description, analyzing the teaching task description into a plurality of teaching content generation subtasks with association dependency relations to form a task scheduling sequence, completing each subtask in sequence by a plurality of agents to generate a teaching plan draft, performing collaborative assessment on the draft to form evaluation feedback, performing directional reconstruction and optimization on the teaching plan content according to the evaluation feedback, writing an optimization result into an accumulated context to restrict a subsequent generation process, performing global negative evaluation on the complete teaching plan, and outputting a final teaching plan generation result. The application realizes the efficient generation and global consistency optimization of the content of the teaching plan, remarkably improves the quality, consistency and automation degree of the generation of the teaching plan, effectively reduces the cost of manual programming and improves the overall quality of the teaching plan.

Inventors

Huang Qionghao
HU CAN
HUANG CHANGQIN
Lu Lingnuo

Assignees

浙江师范大学

Dates

Publication Date: 20260508
Application Date: 20260212

Claims (9)

1. A teaching plan generation method based on multi-agent cooperation is characterized by comprising the following steps: Acquiring teaching task parameters; constructing a global teaching task description based on the teaching task parameters; Analyzing the global teaching task description into a plurality of teaching content generation subtasks with precedence-order dependency relationships to form a task scheduling sequence; invoking a plurality of agents to execute the teaching content generation subtasks according to the task scheduling sequence so as to generate a corresponding teaching plan content draft; Based on the multi-agent, executing cooperative evaluation processing on the teaching plan content draft to generate evaluation feedback information; Based on the evaluation feedback information, reconstructing and optimizing the draft of the draft content to obtain an optimized draft content result; Writing the teaching plan content result into an accumulated context, and continuing to schedule the teaching plan generation of the subsequent subtasks under the constraint of the accumulated context until all the subtasks are completed; after all subtasks are completed, global thinking-back evaluation processing is carried out on the completed complete teaching plan content, and a final teaching plan generation result is generated; Writing the teaching plan content result into an accumulated context, and continuing to schedule teaching plan generation of subsequent subtasks under the constraint of the accumulated context until all the subtasks are completed, wherein the method comprises the following steps of: Writing the teaching plan content result into an accumulated context; Based on the accumulated context, performing context awareness judgment on the teaching content generation subtasks to be executed, determining the subtasks with strong causal dependency relationship as serial execution subtasks, and determining the relatively independent subtasks as parallel execution subtasks; In the serial execution sub-tasks, the tasks are sequentially executed according to the task scheduling sequence, and after each sub-task is completed, the corresponding optimized teaching plan content result is written into the accumulated context so as to restrict the teaching plan generation of the subsequent sub-tasks; in the parallel execution subtasks, generating teaching plan of the corresponding subtasks in parallel under the condition of keeping the accumulated context read-only, and writing the generated result into a preset teaching plan structure position; After the serial execution subtask and the parallel execution subtask are completed, consistency verification is carried out on the generated teaching plan content, and the accumulated context is updated after verification is passed; And continuously scheduling the incomplete teaching content to generate subtasks based on the updated accumulated context until all the subtasks in the task scheduling sequence are completed.
2. The method of claim 1, wherein the teaching task parameters include teaching object features, teaching segments, discipline types, teaching topics, and time-of-day constraint information.
3. The method of claim 1, wherein parsing the global teaching task description into a plurality of teaching content generation subtasks having precedence-dependent relationships forms a task scheduling sequence, comprising: Based on the global teaching task description, extracting teaching target constraint, learning condition constraint and teaching flow constraint, and constructing a teaching task constraint set; Dividing a teaching plan generating process into a plurality of teaching content generating subtasks according to a teaching logic sequence based on the teaching task constraint set; Generating a logic dependency relationship among subtasks based on the teaching contents, and constructing a subtask directed dependency relationship graph to determine a task execution sequence with sequential dependency constraint; and generating a task scheduling sequence according to the task execution sequence.
4. The method of claim 3, wherein the teaching content generation subtask includes a teaching target generation subtask, a teaching context import subtask, a teaching activity design subtask, a teaching assessment design subtask, and a teaching jeopardy generation subtask.
5. The method of claim 1, wherein invoking the multi-agent to perform the tutorial content generation subtask according to the task schedule sequence to generate a corresponding tutorial content draft comprises: generating subtasks for each teaching content according to the task scheduling sequence to respectively call multiple agents with different teaching functions; According to the multi-agent, generating a task type and a constraint condition of a subtask based on the teaching content, and independently generating a corresponding teaching content segment; and collecting the teaching content fragments to form a teaching plan content draft corresponding to the teaching content generation subtask.
6. The method of claim 5, wherein the multi-agent comprises a teaching planning agent, a content generation agent, a collaborative assessment agent, an disfigurement optimization agent, and an assessment supervision agent.
7. The method of claim 1, wherein after completing all sub-tasks, performing a global disbelief evaluation process on the completed complete teaching plan content to generate a final teaching plan generation result, comprising: After all subtasks are completed, based on TOPIC teaching quality evaluation models, multidimensional quantitative evaluation is carried out on the complete teaching plan content, and a global evaluation result is obtained; Judging whether a preset quality threshold is met or not according to the global evaluation result; when the quality threshold is not met, triggering the directional reconstruction of the complete teaching plan content by multiple intelligent agents based on the global evaluation result, and re-executing the TOPIC teaching quality evaluation model to form an anti-thinking-reconstruction iterative process; and when the quality threshold is met, determining the corresponding complete teaching plan content as a final teaching plan generation result.
8. The method of claim 7, wherein the TOPIC teaching quality assessment model construction process includes: Determining a plurality of evaluation dimensions for evaluating the quality of a teaching plan based on a preset teaching evaluation theory system, wherein the teaching evaluation theory system comprises teaching content correctness, teaching process operability, teaching strategy rationality, learning cognition suitability and teaching target consistency; based on the multiple evaluation dimensions, respectively calculating quantitative scores of all the evaluation dimensions for the complete teaching plan content to form TOPIC-dimension score vectors; Performing weighted fusion operation on the TOPIC-dimension scoring vector to obtain a comprehensive scoring result representing the overall teaching quality; training and constructing TOPIC teaching quality evaluation model based on the TOPIC-dimension grading vector and the comprehensive grading result.
9. A teaching plan generation system based on multi-agent collaboration, characterized in that the system is applied to the method of any of claims 1-8, the system comprising: the data acquisition module is used for acquiring teaching task parameters; the global construction module is used for constructing global teaching task description based on the teaching task parameters; The task analysis module is used for analyzing the global teaching task description into a plurality of teaching content generation subtasks with sequential dependency relationships to form a task scheduling sequence; The intelligent agent execution module is used for calling a plurality of intelligent agents to execute the teaching content generation subtasks according to the task scheduling sequence so as to generate a corresponding teaching plan content draft; the collaborative review module is used for executing collaborative review processing on the teaching plan content draft based on the multi-agent to generate evaluation feedback information; the content optimization module is used for carrying out reconstruction optimization on the teaching plan content draft based on the evaluation feedback information to obtain an optimized teaching plan content result; the context accumulation module is used for writing the teaching plan content result into an accumulation context and continuing to schedule the teaching plan generation of the subsequent subtasks under the constraint of the accumulation context until all the subtasks are completed; and the global thinking-back evaluation module is used for executing global thinking-back evaluation processing on the completed complete teaching plan content after all the subtasks are completed, and generating a final teaching plan generation result.

Description

Teaching plan generation method and system based on multi-agent cooperation Technical Field 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. Background In the related art, with the continuous development of artificial intelligence technology and intelligent education application, the generation of teaching contents based on generated artificial intelligence gradually becomes an important technical means for assisting teachers in preparing lessons and teaching designs. The technology generally carries out semantic analysis on teaching task parameters such as grades, subjects and subjects input by teachers through a large-scale pre-training language model, and generates teaching plan texts containing teaching targets, teaching processes, teaching activity arrangement and the like according to the teaching task parameters, so that the teaching plan generation efficiency is improved to a certain extent, and the burden of the teachers in repeated teaching work is reduced. However, most of the existing teaching plan generation methods rely on introspective reasoning and one-time text generation mechanisms of a single model, and have obvious defects when dealing with the task of teaching design, which has high constraints of structural, stepwise and educational rules. On one hand, the single model lacks multi-view negotiation and independent verification capability in the generation process, when deviation occurs in early links such as learning condition analysis, teaching target setting or teaching activity design, the follow-up content is always continuously deduced along the established assumption, potential cognitive or logic problems are difficult to discover and correct in time, on the other hand, in the process of generating a teaching plan, teaching content is always treated as mutually independent fragments in the prior art, effective description of sequential dependency relationship and logic association between subtasks is lacking, and therefore repetition, logic jump or level confusion of the generated content is caused, and a structured teaching plan which accords with an actual teaching flow is difficult to form. In addition, the existing teaching plan generation method generally lacks a multi-main-body cooperation and systematic evaluation mechanism, the generation result generally depends on a single generation process, the problems of teaching target deviation, unreasonable content, mismatching of teaching strategies and the like are difficult to discover and correct in time, a large amount of manpower is often required to participate in repeated modification and verification, and the efficiency and quality are low. In summary, the technical problems in the related art are to be improved. Disclosure of Invention The embodiment of the application mainly aims to provide a teaching plan generation method and system based on multi-agent cooperation, so that efficient generation of teaching plan content and global consistency optimization are realized, the quality, consistency and automation degree of teaching plan generation are improved, the labor programming cost is effectively reduced, and the overall quality of the teaching plan is improved. In order to achieve the above object, an aspect of the embodiments of the present application provides a scenario generation method based on multi-agent collaboration, the method including the following steps: Acquiring teaching task parameters; constructing a global teaching task description based on the teaching task parameters; Analyzing the global teaching task description into a plurality of teaching content generation subtasks with precedence-order dependency relationships to form a task scheduling sequence; invoking a plurality of agents to execute the teaching content generation subtasks according to the task scheduling sequence so as to generate a corresponding teaching plan content draft; Based on the multi-agent, executing cooperative evaluation processing on the teaching plan content draft to generate evaluation feedback information; Based on the evaluation feedback information, reconstructing and optimizing the draft of the draft content to obtain an optimized draft content result; Writing the teaching plan content result into an accumulated context, and continuing to schedule the teaching plan generation of the subsequent subtasks under the constraint of the accumulated context until all the subtasks are completed; And after all the subtasks are completed, global thinking-back evaluation processing is carried out on the completed complete teaching plan content, and a final teaching plan generation result is generated. In some embodiments, the teaching task parameters include teaching object features, teaching segments, discipline types, teaching topics, and lesson constraint information. In some embodiments, the parsing the g