A course standard alignment teaching plan generation method and system

Abstract

This invention provides a method and system for generating lesson plans aligned with curriculum standards. The method includes: processing original lesson plan files using a lesson plan data processing framework based on question retrieval and replacement to obtain processed lesson plan data; constructing lesson plan generation training data based on the processed lesson plan data and lesson plan generation instructions; fine-tuning the lesson plan generation model using the training data; and generating curriculum standard-aligned lesson plans based on the fine-tuned model. This invention can generate lesson plans that are structurally complete, accurate in content, and logically clear.

Description

Technical Field

[0001] This invention belongs to the field of teaching management technology, specifically relating to a method and system for generating lesson plans that align with curriculum standards. Background Technology

[0002] Lesson plans are crucial for teachers' teaching activities, helping to clarify teaching objectives, rationally organize teaching content, and thus improve overall teaching quality. Traditional lesson plan writing relies heavily on manual design by teachers, resulting in long writing cycles, inconsistent structures, and outdated content. With the rise of large language model technology, research has begun on lesson plan generation methods based on these models. Most existing methods generate lesson plans by writing instructions to prompt a general-domain large language model; however, these methods are merely secondary developments on general models, failing to deeply integrate educational data and fully adhere to the new curriculum standards, thus lacking educational professionalism.

[0003] Existing lesson plan generation methods rely on pre-written prompts and user-defined lesson plan requirements, utilizing a general-domain language model to generate lesson plans. Some methods further incorporate actual textbook content to assist the language model in generating lesson plans based on user needs. In general, existing methods are all based on prompt technology and a general-domain language model.

[0004] Existing lesson plan generation methods involve writing prompts and attaching teachers' lesson plan requirements to these prompts. The prompts then generate lesson plans that meet the teachers' needs using a general language model. This method relies almost entirely on the knowledge and capabilities of the general language model itself. The general language model has not been specifically fine-tuned on professional educational data or lesson plan data, and has failed to deeply integrate data from the education field. The generated lesson plans cannot fully comply with the guidance of the new curriculum standards, making it difficult to guarantee educational professionalism and meet actual teaching needs. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method and system for generating lesson plans that align with curriculum standards. It aims to follow the guidance of the new curriculum standards, deeply integrate professional knowledge in the field of education, and automatically generate lesson plans that are structurally complete, accurate in content, and logically clear.

[0006] To achieve the above objectives, the present invention provides the following solution: A method for generating lesson plans aligned with curriculum standards, the method comprising: The original lesson plan files were processed using a lesson plan data processing framework based on question retrieval and replacement to obtain processed lesson plan data. Based on the processed lesson plan data and combined with the lesson plan generation instructions, lesson plan generation training data is constructed. The lesson plan generation training data was used to fine-tune the lesson plan generation model; Based on the finely tuned lesson plan generation model, lesson plans aligned with curriculum standards are generated.

[0007] Preferably, the lesson plan data processing framework based on question retrieval and replacement includes: a lesson plan file preprocessing component and a question retrieval and replacement component; The lesson plan file preprocessing component is used to convert the original lesson plan file into a text format lesson plan using an image modality processing method. The question retrieval and replacement component is used to retrieve and replace questions based on the weight of knowledge points from the perspective of the lesson plan, based on the text format lesson plan, and obtain the processed lesson plan data.

[0008] Preferably, the lesson plan file preprocessing component employs an image modality processing method to uniformly convert the original lesson plan files into text format lesson plans, including: Convert the original lesson plan file into a lesson plan image; The lesson plan images are input into a multimodal large language model, and the information contained in the lesson plan illustrations is combined to generate a text-formatted lesson plan in Markdown format.

[0009] Preferably, the question retrieval and replacement component, based on text-formatted lesson plans, performs question retrieval and replacement based on the weight of knowledge points from the perspective of the lesson plan, and the method for obtaining processed lesson plan data includes: The content of the questions in the text-formatted lesson plans is annotated and extracted to obtain the question stem, analysis and additional information for each question; Based on the question stem, analysis, and additional information of each question, calculate the weight of knowledge points from the perspective of the lesson plan; Based on the knowledge point weights from the perspective of lesson plans, question retrieval and replacement are performed to obtain processed lesson plan data.

[0010] Preferred methods for calculating the weight of knowledge points from a lesson plan perspective, based on the question stem, explanation, and additional information for each question, include: ; in, For the first i The weight of knowledge points from the perspective of each topic's lesson plan. This refers to the number of times a knowledge point is repeated in the lesson plan. For predefined coefficients, This represents the number of knowledge points within the cluster to which the knowledge point belongs.

[0011] Preferred methods for obtaining processed lesson plan data by retrieving and replacing questions based on the knowledge point weights from the perspective of lesson plans include: Using the weight of knowledge points from the perspective of lesson plans as the sorting index, the knowledge points of the questions to be replaced are sorted in descending order to obtain an ordered list of knowledge points. Based on an ordered list of knowledge points, question retrieval and replacement are performed to obtain processed lesson plan data.

[0012] Preferred methods for obtaining processed lesson plan data by retrieving and replacing questions based on an ordered list of knowledge points include: S1. Let the set of questions to be replaced in this lesson plan be... ;No. i One question to be replaced Ordered list of knowledge points Its knowledge point set is , For the i-th question to be replaced The number of knowledge points, and the set of candidate replacement questions. Initialize the set of problems already used. ; S2, dealing with replacement questions Search for knowledge point sets in the question bank Completely identical set of candidate topics ,like Then, randomly select one question from them and add it. and ; S3, dealing with replacement questions ,according to The knowledge points are sequentially traversed step by step, and a candidate set is constructed by searching the question bank: ; in For the question bank, express A collection of knowledge points This refers to the questions retrieved from the question bank. This indicates the number of iterations. If the intersection is empty during the iteration, the non-empty candidate set from the previous round is retained and the iteration stops; if... Then, randomly select one question from them and add it. and ; S4. Dealing with Replacement Questions Its candidate replacement question set is If there is only one question, then that question will be replaced. Otherwise, use a multimodal large language model in Choose the closest The question has been replaced; S5. Repeat S2-S4 until all questions to be replaced have been retrieved and replaced, and the processed lesson plan data is obtained.

[0013] Preferably, the lesson plan generation model is built on the Qwen2.5-14B-instruct platform.

[0014] The present invention also provides a lesson plan generation system for aligning with curriculum standards. The system is used to implement the aforementioned method and includes: a lesson plan data processing module, a training data construction module, a model fine-tuning module, and a lesson plan generation module. The lesson plan data processing module is used to process the original lesson plan file using a lesson plan data processing framework based on question retrieval and replacement to obtain the processed lesson plan data. The training data construction module is used to construct training data for generating lesson plans based on the processed lesson plan data and in conjunction with the lesson plan generation instructions. The model fine-tuning module is used to fine-tune the lesson plan generation model using the training data generated from the lesson plan. The lesson plan generation module is used to generate lesson plans that are aligned with curriculum standards based on the fine-tuned lesson plan generation model.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a method and system for generating lesson plans aligned with curriculum standards. It can achieve high-quality conversion of existing lesson plan data to obtain high-quality plain text lesson plan data. Following the guidance of the new curriculum standards, it deeply integrates professional knowledge in the field of education, automatically generating lesson plans that are structurally complete, accurate in content, and logically clear. The lesson plans generated by this invention not only emphasize knowledge transfer and curriculum instruction but also take into account guidance at the competency level, meeting the needs of educational applications. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the lesson plan generation method for aligning curriculum standards according to an embodiment of the present invention; Figure 2 This is an example diagram of a text-formatted lesson plan obtained after preprocessing, as described in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1 like Figure 1 As shown, this invention provides a method for generating lesson plans aligned with curriculum standards, comprising: The original lesson plan files were processed using a lesson plan data processing framework based on question retrieval and replacement to obtain processed lesson plan data. Based on the processed lesson plan data and combined with the lesson plan generation instructions, lesson plan generation training data is constructed. The lesson plan generation training data was used to fine-tune the lesson plan generation model; Based on the finely tuned lesson plan generation model, lesson plans aligned with curriculum standards are generated.

[0021] The specific implementation process of this invention is as follows: The technical solution of this invention consists of three parts: a lesson plan data processing framework based on question retrieval and replacement, a lesson plan generation instruction template and a method for constructing lesson plan generation training data, and instruction fine-tuning training of the lesson plan generation model. Based on this technical solution, it is possible to (1) convert existing lesson plan data into high-quality plain text lesson plan data; (2) construct lesson plan generation training data; and (3) train the lesson plan generation model.

[0022] First, a batch of publicly available lesson plan data was collected online, and training data for lesson plan generation was constructed. Finally, a lesson plan generation model was fine-tuned and trained using instructions, as detailed below: The lesson plan data processing framework based on the knowledge points of the new curriculum standards includes: a lesson plan file preprocessing component and a question retrieval and replacement component; The lesson plan file preprocessing component is used to convert the original lesson plan files into text format lesson plans using image modality processing methods. The question retrieval and replacement component is used to retrieve and replace questions based on the weight of knowledge points from the perspective of the lesson plan, based on the text format lesson plan, and obtain the processed lesson plan data.

[0023] Lesson plan document preprocessing component: The original lesson plan files have the following problems: (1) The original lesson plan files have inconsistent formats and structures, making them difficult to process uniformly; (2) Some lesson plans contain illustrations, which the large language model cannot process; (3) The formulas in the lesson plans are diverse, and directly extracting formula objects or formula text is prone to errors. The lesson plan file preprocessing component can uniformly convert any form of lesson plan file into a Markdown plain text format that the large language model can directly process, while ensuring the integrity of the lesson plan content.

[0024] To ensure the integrity of key information such as the structure, images, and formulas in the original lesson plan file, this invention uses an image modality processing method to process the original lesson plan file. Specifically, the original lesson plan file is first converted into a lesson plan image, and then the lesson plan image is input into a multimodal large language model, prompting it to output the lesson plan content as Markdown text. For lesson plan illustrations in the lesson plan, the multimodal large language model is prompted to identify their necessity: (1) For lesson plan illustrations with no information content, they are directly deleted; (2) For lesson plan illustrations that are relatively simple and have little information content, the multimodal large language model is used to convert them into text descriptions; (3) For complex lesson plan illustrations, the multimodal large language model is used to extract the relevant context of the lesson plan illustration, and analyze the teaching objectives and knowledge points involved in that part, and finally convert them into approximate content that does not require images based on the analysis.

[0025] The processing of illustrations in lesson plans involves operations such as image deletion and transformation of images and their contexts, which may disrupt the logic of the original lesson plan. Therefore, it is necessary to input the plain text lesson plan into a multimodal large language model to uniformly check the logical integrity and content correctness, and make appropriate repairs.

[0026] Example of a preprocessed text-formatted lesson plan: Figure 2 As shown.

[0027] Title search and replacement component: The overall structure of a lesson plan is composed of interconnected and organized questions, including specific questions and examples, as well as simple inquiries from the teacher. The order in which the questions are organized reflects the core teaching process design of the lesson plan: the teaching process in the lesson plan is often introduced by questions, and the progressive explanation of knowledge points is also connected by questions. For example, in a lesson plan teaching the Pythagorean theorem, questions about right triangles are often used as the introduction. For low-quality original lesson plan files, i.e., files with missing images or formulas, or messy formatting, although they cannot be used directly, the text information of each question still retains the teaching ideas of the lesson plan. In summary, low-quality lesson plans can be reconstructed by identifying the questions in them and replacing them with approximate complete questions. The question retrieval and replacement is based on the new curriculum standard knowledge point system built by this invention, specifically composed of the following: Question Annotation and Extraction: The questions in the lesson plan consist of three parts: stem, explanation, and additional information. Each question must have a stem, while the explanation and additional information are optional. To enable question retrieval and replacement, these three parts of the original question need to be annotated and extracted. This process is implemented using a multimodal large language model: First, three types of tags are defined:<statement_i>< / statement_i> ,<soluiton_i>< / solution_i> ,<extra_info_i>< / extra_info_i> The terms represent the three parts of a question: the question stem, the explanation, and the additional information. The multimodal large language model is prompted to use three types of labels to mark the corresponding parts of the question, with 'i' in the label used to identify the question. Then, regular expression matching is used to extract the question stem, explanation, and additional information of each question, thus completing the question extraction. Knowledge Point Weight Calculation from a Lesson Plan Perspective: The knowledge point space comprised of all questions in a lesson plan reflects the teaching content of that lesson plan. Compared to considering and retrieving each question individually, retrieving questions from the overall perspective of the lesson plan better aligns with the teaching content. Based on this, this invention designs a method for calculating the weight of each knowledge point from the perspective of the entire lesson plan. These knowledge point weights are used for question retrieval. Let the set of questions extracted from a lesson plan using the aforementioned question content annotation and extraction method be denoted as [example of question set]. Where n is the number of questions, this invention first uses private knowledge point data to train a knowledge point labeling model, as follows: First, a knowledge point labeling training dataset is constructed, which can be represented as a set of binary tuples. ,in The title text, For the corresponding knowledge point tags, This indicates the number of training samples in the group. Then, the title text... Input, knowledge point tags For output, the pre-trained model Qwen2.5-7B-Instruct is fine-tuned, with the optimization objective being: in, For model parameters, This represents the conditional probability distribution of the model. For the first i The first sample t One target token, The target number of tokens, For the first t All target tokens before the given position. By optimizing the above objective function, a knowledge point labeling model is obtained that can output corresponding knowledge point tags based on the question text.

[0028] Use the above knowledge point marking model to mark the knowledge points for each question, and obtain the knowledge point set of each question; assume the question 's knowledge point set is , is the number of knowledge points of the i-th question , that is ; the knowledge point set of the entire teaching plan is , where represents the number of knowledge points in the teaching plan, that is ; at the same time, the number of times each knowledge point repeats in the teaching plan is recorded as . Then for each knowledge point in , use the Python Chinese word segmentation component Jieba for word segmentation; then use the preset stop word list to filter the stop words of the segmented knowledge points (such as removing meaningless content such as "of" and "le"), and use the "Mathematical Terms" published by the Mathematics Nomenclature Committee as the protected word dictionary during the filtering process to prevent mathematical proper nouns from being misfiltered; after the above process, the segmented is obtained, and then is embedded as a TF-IDF vector , so as to realize the vectorization of the knowledge points in text form; then define the distance and between , where represents the cosine similarity between vectors; finally, use the distance as the distance metric between knowledge points, and perform hierarchical clustering on the knowledge points in , and the number of knowledge points within the cluster where the knowledge point is located is recorded as . Finally, the weight of the knowledge point in the perspective of the teaching plan is obtained , is a predefined coefficient; the knowledge point weight is calculated from the overall teaching plan and reflects the contribution degree of the knowledge point to the overall teaching process of the teaching plan. The present invention uses the knowledge point weight in the perspective of the teaching plan to guide the retrieval and replacement of questions.

[0029] Question retrieval and replacement based on the knowledge point weight in the perspective of the teaching plan: For the knowledge point set in a teaching plan, after calculating the knowledge point weight in the perspective of the teaching plan, the weight of each knowledge point can be obtained. Use the knowledge point weight as the sorting index to sort the knowledge points in descending order to obtain an ordered list The ranking of each knowledge point in this list can be seen as a ranking of its contribution to the overall teaching process of the lesson plan. Similarly, using the weight of knowledge points as a ranking criterion, the questions to be replaced... Sort the knowledge points in descending order to obtain an ordered list of knowledge points. The question retrieval and replacement process of this invention is based on an ordered list of knowledge points, as detailed below: S1. Let the set of questions to be replaced in this lesson plan be... ,in The number of questions to be replaced, i.e. Each question to be replaced Ordered list of knowledge points Its knowledge point set is The candidate replacement question set is Initialize the set of problems already used. .

[0030] S2, dealing with replacement questions Search for knowledge point sets in the question bank Completely identical set of candidate topics ,like Then, randomly select one question from them and add it. and .

[0031] S3, dealing with replacement questions ,according to The knowledge points are sequentially traversed step by step, and a candidate set is constructed by searching the question bank: in For the question bank, express A collection of knowledge points. Among them... This refers to the questions retrieved from the question bank. Indicates the number of iterations: if the intersection is empty during the iteration, the non-empty candidate set from the previous round is retained and the iteration stops; if... Then, randomly select one question from them and add it. and .

[0032] S4. Dealing with Replacement Questions Its candidate replacement question set is If there is only one question, then that question will be replaced. Otherwise, use a large language model in Choose the closest The question has been replaced.

[0033] S5. Repeat S2-S4 until all items to be replaced have been retrieved and replaced.

[0034] After replacing the title in the lesson plan, the overall content of the lesson plan is rewritten and polished using a large language model to ensure that the logic is coherent and the teaching content is accurate, thus obtaining high-quality lesson plan data after processing.

[0035] Furthermore, to construct training data for lesson plan generation, not only high-quality lesson plan data is needed, but also corresponding lesson plan generation instructions. Based on feedback from frontline teachers and education experts, this invention designs multi-dimensional instruction generation prompts covering key elements such as knowledge points (topic names), teaching focus and difficulties, and teaching objectives, centered around the core competencies orientation of the new curriculum standards. It utilizes a large language model to generate diverse instructions from existing lesson plan data. Each lesson plan corresponds to several instructions, and each instruction and its corresponding lesson plan constitute a training sample. All instructions involve at least one knowledge point or teaching focus and difficulty, and are labeled based on the new curriculum standards' knowledge system to ensure coverage of core knowledge across all subjects. The instruction formats are both professional and practical, including both clearly structured and specific standardized expressions, as well as colloquial expressions relevant to teaching scenarios, aiming to enhance the model's understanding and generation capabilities for competency-oriented teaching needs.

[0036] Based on existing lesson plan data and the instruction construction process, this invention constructs a competency-oriented lesson plan generation training dataset, totaling 45,313 data entries, covering mathematics, physics, and chemistry subjects in junior high school. Each training data entry is in the following format:<Q,A> The dataset consists of question-and-answer pairs, where Q is the input during training, i.e., the instructions for generating lesson plans; and A is the expected output corresponding to Q during training, i.e., the lesson plan data. This dataset closely integrates with the new curriculum standards' knowledge system, taking into account both key teaching points and difficulties, as well as the cultivation of core competencies. It aims to provide high-quality data support for model training in lesson plan generation tasks, thereby contributing to personalized teaching and improving teaching quality. Examples of training data are shown in Table 1.

[0037] Table 1 Furthermore, using Qwen2.5-14B-instruct as a foundation, and based on the lesson plan generation training data constructed in the aforementioned part, Qwen2.5-14B-instruct is trained using instruction fine-tuning to obtain a lesson plan generation model.

[0038] Instruction fine-tuning is based on a supervised learning paradigm. Each training data point consists of an input instruction and an expected output. It uses the standard cross-entropy loss function to optimize model parameters by maximizing the log-likelihood probability, thereby guiding the model to learn task-related language generation patterns. Specifically, given a set of training samples... ,in Indicates the first Input command, Represents the target sequence. Indicates the target number of tokens. This represents the number of training samples in the group. The training objective is to maximize the likelihood probability of the target sequence under the given instructions. Expanding each target sequence at the token level, it can be represented as: in, This is the conditional probability distribution of the model, and the model parameters involved in the optimization are: ; It is the first The first sample Target tokens; loss function It is the standard cross-entropy loss.

[0039] Furthermore, the lesson plan generation capability of the fine-tuned model was evaluated from the following five dimensions: structural completeness, content accuracy, content consistency, language logic, and competency orientation. The overall evaluation results are shown in Table 2, where each dimension represents the weighted score for mathematics, physics, and chemistry. The lesson plan generation model developed in this invention performed excellently in all dimensions, achieving a higher overall score than all other models in the table, including the most advanced models such as Qwen2.5-14B-instruct and DeepSeek-R1-Distill-Qwen-14B. The lesson plan generation model developed in this invention scored highest in both structural completeness (83.76) and content accuracy (71.53), indicating that the model's lesson plan content is clearly organized, comprehensively covers key points, and possesses strong knowledge acquisition and communication capabilities. The lesson plan generation model developed in this invention has leading performance in both language logic (87.00) and content consistency (88.20), demonstrating its advantage in generating coherent and consistent lesson plan content. Its competency orientation (84.75) is higher than that of most other models, indicating that the lesson plan generation model developed in this invention not only emphasizes knowledge transfer and curriculum teaching, but also takes into account competency guidance, which meets the needs of educational applications.

[0040] Table 2 In summary, this invention proposes a lesson plan data processing framework based on question retrieval and replacement. This framework can generate high-quality plain text lesson plan data from existing lesson plan files of varying quality, file formats, and organizational structures, which can then be used to construct training data for lesson plan generation tasks. Based on feedback from frontline teachers and education experts, this invention has constructed a set of lesson plan generation instruction templates and, combined with the question retrieval and replacement-based lesson plan data processing framework, created a batch of high-quality lesson plan generation training data that conforms to the new curriculum standards. A lesson plan generation model was then fine-tuned using these self-built data instructions. The fine-tuned lesson plan generation model covers the knowledge points of the new curriculum standards, follows the guidance of the new curriculum standards, and deeply integrates professional knowledge in the field of education. It can automatically generate lesson plans that are structurally complete, accurate in content, and logically clear.

[0041] Example 2 The present invention also provides a lesson plan generation system for aligning with curriculum standards, used to implement the method described in the foregoing embodiments. The system includes: a lesson plan data processing module, a training data construction module, a model fine-tuning module, and a lesson plan generation module. The lesson plan data processing module is used to process the original lesson plan file using a lesson plan data processing framework based on question retrieval and replacement to obtain the processed lesson plan data. The training data construction module is used to construct training data for generating lesson plans based on the processed lesson plan data and in conjunction with the lesson plan generation instructions. The model fine-tuning module is used to fine-tune the lesson plan generation model using the training data generated from the lesson plan. The lesson plan generation module is used to generate lesson plans that are aligned with curriculum standards based on the fine-tuned lesson plan generation model.

[0042] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made by those skilled in the art to the technical solutions of the present invention without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for generating lesson plans aligned with curriculum standards, characterized in that, The method includes: The original lesson plan files were processed using a lesson plan data processing framework based on question retrieval and replacement to obtain processed lesson plan data. Based on the processed lesson plan data and combined with the lesson plan generation instructions, lesson plan generation training data is constructed. The lesson plan generation training data was used to fine-tune the lesson plan generation model; Based on the finely tuned lesson plan generation model, lesson plans aligned with curriculum standards are generated.

2. The method according to claim 1, characterized in that, The lesson plan data processing framework based on question retrieval and replacement includes: a lesson plan file preprocessing component and a question retrieval and replacement component; The lesson plan file preprocessing component is used to convert the original lesson plan file into a text format lesson plan using an image modality processing method. The question retrieval and replacement component is used to retrieve and replace questions based on the weight of knowledge points from the perspective of the lesson plan, based on the text format lesson plan, and obtain the processed lesson plan data.

3. The method according to claim 2, characterized in that, The lesson plan file preprocessing component employs image modality processing to uniformly convert the original lesson plan files into text format lesson plans. The methods include: Convert the original lesson plan file into a lesson plan image; The lesson plan images are input into a multimodal large language model, and the information contained in the lesson plan illustrations is combined to generate a text-formatted lesson plan in Markdown format.

4. The method according to claim 2, characterized in that, The question retrieval and replacement component, based on text-formatted lesson plans, performs question retrieval and replacement based on the weight of knowledge points from the perspective of the lesson plan. The method for obtaining processed lesson plan data includes: The content of the questions in the text-formatted lesson plans is annotated and extracted to obtain the question stem, analysis and additional information for each question; Based on the question stem, analysis, and additional information of each question, calculate the weight of knowledge points from the perspective of the lesson plan; Based on the knowledge point weights from the perspective of lesson plans, question retrieval and replacement are performed to obtain processed lesson plan data.

5. The method according to claim 4, characterized in that, Based on the question stem, explanation, and additional information for each question, methods for calculating the weight of knowledge points from the perspective of lesson plans include: ； in, For the first i The weight of knowledge points from the perspective of each topic's lesson plan. This refers to the number of times a knowledge point is repeated in the lesson plan. For predefined coefficients, This represents the number of knowledge points within the cluster to which the knowledge point belongs.

6. The method according to claim 4, characterized in that, Methods for obtaining processed lesson plan data by retrieving and replacing questions based on the knowledge point weights from a lesson plan perspective include: Using the weight of knowledge points from the perspective of lesson plans as the sorting index, the knowledge points of the questions to be replaced are sorted in descending order to obtain an ordered list of knowledge points. Based on an ordered list of knowledge points, question retrieval and replacement are performed to obtain processed lesson plan data.

7. The method according to claim 6, characterized in that, Methods for obtaining processed lesson plan data by retrieving and replacing questions based on an ordered list of knowledge points include: S1. Let the set of questions to be replaced in this lesson plan be... ;No. i One question to be replaced Ordered list of knowledge points Its knowledge point set is , For the i-th question to be replaced The number of knowledge points, and the set of candidate replacement questions. Initialize the set of problems already used. ; S2, dealing with replacement questions Search for knowledge point sets in the question bank Completely identical set of candidate topics ,like Then, randomly select one question from them and add it. and ; S3, dealing with replacement questions ,according to The knowledge points are sequentially traversed step by step, and a candidate set is constructed by searching the question bank: ； in For the question bank, express A collection of knowledge points This refers to the questions retrieved from the question bank. This indicates the number of iterations. If the intersection is empty during the iteration, the non-empty candidate set from the previous round is retained and the iteration stops; if... Then, randomly select one question from them and add it. and ; S4. Dealing with Replacement Questions Its candidate replacement question set is If there is only one question, then that question will be replaced. Otherwise, use a multimodal large language model in Choose the closest The question has been replaced; S5. Repeat S2-S4 until all questions to be replaced have been retrieved and replaced, and the processed lesson plan data is obtained.

8. The method according to claim 1, characterized in that, The lesson plan generation model is built on the Qwen2.5-14B-instruct platform.

9. A lesson plan generation system aligned with curriculum standards, the system being used to implement the method described in any one of claims 1-8, characterized in that, The system includes: a lesson plan data processing module, a training data construction module, a model fine-tuning module, and a lesson plan generation module; The lesson plan data processing module is used to process the original lesson plan file using a lesson plan data processing framework based on question retrieval and replacement to obtain the processed lesson plan data. The training data construction module is used to construct training data for generating lesson plans based on the processed lesson plan data and in conjunction with the lesson plan generation instructions. The model fine-tuning module is used to fine-tune the lesson plan generation model using the training data generated from the lesson plan. The lesson plan generation module is used to generate lesson plans that are aligned with curriculum standards based on the fine-tuned lesson plan generation model.

Images