Presentation analysis model, classroom utterance analysis model and application thereof

Abstract

The invention relates to the field of artificial intelligence and mathematical education, in particular to a PowerPoint analysis model, a classroom speech analysis model and application thereof.The method comprises the steps that the PowerPoint analysis model is constructed, and micro teaching evaluation analysis is conducted on a PowerPoint through the PowerPoint analysis model; and constructing a classroom utterance analysis model, and performing micro teaching evaluation analysis on the classroom utterance through the classroom utterance analysis model. According to the method, the presentation analysis model is constructed, and micro teaching evaluation analysis is carried out on the presentation through the presentation analysis model; and constructing a classroom utterance analysis model, and performing micro teaching evaluation analysis on the classroom utterance through the classroom utterance analysis model. Microteaching evaluation analysis is carried out from the two aspects of the presentation file and the classroom utterance, the analysis is more comprehensive, and the final analysis result can better reflect the real situation of teaching.

Description

Technical Field

[0001] The present invention relates to the fields of artificial intelligence and mathematics education, and more specifically, to a microteaching evaluation and analysis method and system. Background Art

[0002] The evaluation of mathematics micro-teaching requires teachers to analyze and evaluate the micro-teaching materials of normal school students, which consumes a lot of time and resources. Intelligent analysis and evaluation of mathematics micro-teaching can help teachers self-analyze and optimize the micro-teaching evaluation process.

[0003] Analyzing the accuracy of each teaching-related proposition in a teaching presentation allows for real-time monitoring of teaching rigor and optimizes teaching outcomes for both students and teachers. Traditional automated mathematical analysis methods focus on mathematical problem solving. These include software or libraries based on symbolic computation and algebraic systems, such as Mathematica and SymPy, that automatically perform derivatives or factorizations through symbolic derivation; problem-solving systems based on handwritten rules and pattern matching, such as Student and WordPro; and human-like problem-solving systems based on rule engines and logical reasoning, which automatically reason and answer questions through rule, computation, and reasoning engines. These methods rely heavily on predefined rules and algorithms, placing high demands on the format and type of questions. While they perform well on structured problems, they often require extensive manual preprocessing and feature extraction for unstructured or complex natural language problems. Furthermore, their focus on problem solving rather than judgment makes it difficult to evaluate the large number of propositions in microteaching presentations. Recent research on pre-trained large-scale language models is closer to achieving intelligent analysis and evaluation.

[0004] The prior art discloses a teaching behavior analysis system based on the fusion and modeling of teaching features of a knowledge base, comprising: a data acquisition module for collecting classroom teaching data and teaching resource data; a knowledge base construction module for constructing a multimodal knowledge base based on the classroom teaching data and teaching resource data; a feature extraction module for extracting features from the classroom teaching data based on the multimodal knowledge base and generating feature extraction results; a feature fusion module for performing feature mapping and fusion analysis on the feature extraction results to form feature fusion results; and a modeling analysis module for modeling and analyzing classroom behavior based on the feature fusion results and a classroom behavior analysis model to form analysis results. This system can comprehensively analyze the teaching process and classroom behavior based on classroom teaching data and teaching resource data, which is conducive to improving teaching quality. However, this method only analyzes classroom teaching data and teaching resource data, which is not comprehensive enough and cannot fully reflect the teaching situation. Summary of the Invention

[0005] The purpose of the present invention is to disclose a more comprehensive presentation analysis model, classroom discourse analysis model and their applications.

[0006] To achieve the above objectives, the present invention provides a presentation analysis model, a classroom discourse analysis model, and applications thereof, including:

[0007] Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis on presentations;

[0008] Construct a classroom discourse analysis model and use it to conduct micro-teaching evaluation and analysis of classroom discourse.

[0009] Furthermore, a presentation analysis model is constructed to conduct micro-teaching evaluation and analysis on the presentation, including:

[0010] The presentation analysis model includes: visual sub-model, large language sub-model, teacher sub-model, student base model and large language sub-model in the field of mathematics micro-grid;

[0011] Obtain the presentation and extract the statement from the presentation using the visual sub-model;

[0012] Get micro-teaching texts based on the expression sentences through a large language sub-model;

[0013] The teacher sub-model is tuned for prompt words based on the micro-teaching text to obtain the teacher sub-model output. The teacher sub-model output is then subjected to knowledge distillation and data cleaning to obtain a fine-tuning training dataset.

[0014] The student base model is fine-tuned according to the fine-tuning training dataset, and the output of the student base model is used as the input of a large language sub-model in the field of mathematics micro-teaching for micro-teaching evaluation and analysis.

[0015] Further, including:

[0016] The student base submodel is the Qwen-1.5-7B-Chat model;

[0017] The teacher sub-model is the Deepseek-V2 236B model;

[0018] The visual sub-model is the Qwen-2.5-VL-3B model.

[0019] Furthermore, it also includes: combining retrieval enhancement generation technology when conducting microteaching evaluation and analysis in the field of mathematics microteaching using large language sub-models;

[0020] Retrieval-augmented generation is a framework that combines information retrieval and text generation techniques, aiming to improve the relevance and accuracy of the generative model output. The framework consists of a retrieval module and a generation module. The retrieval module efficiently retrieves documents or text fragments related to the input query from a pre-built knowledge base. The generation module usually uses a pre-trained generative language model, combining the user's input information with the external text information returned by the retrieval module as model input to obtain the final generated answer.

[0021] Furthermore, fine-tuning the student base model according to the fine-tuning training data set includes: fine-tuning the base model using a supervised fine-tuning method and a LoRA efficient fine-tuning algorithm according to the fine-tuning training data set.

[0022] Furthermore, a classroom discourse analysis model is constructed to conduct teaching evaluation analysis on classroom discourse, including:

[0023] The classroom discourse analysis model is a context-enhanced dual-channel attention classification model based on a pre-trained model. It includes: a feature extraction module, a multi-semantic channel attention module, and a context feature fusion module based on latent topics.

[0024] Obtain classroom discourse, extract features of the classroom discourse through a feature extraction module, and obtain classroom discourse features;

[0025] The deep features of classroom discourse are extracted through the multi-semantic channel attention module to obtain the deep features of classroom discourse;

[0026] Classroom discourse analysis is conducted based on the characteristics of classroom discourse and the deep characteristics of classroom discourse through the contextual feature fusion module of potential topics.

[0027] Furthermore, the multi-semantic channel attention module includes:

[0028] The multi-semantic channel attention module uses the deep features extracted by multi-channel technology to make up for the shortcomings of the feature extraction module in extracting specific types of features; the features extracted by the dual channels are combined with the features extracted by the pre-trained model through the attention mechanism; specifically, the multi-semantic channel attention module uses the features output by each channel as the query, the classroom discourse features as the key and value, calculates the attention score through the query and key, and then recombine the value with the attention score as the weight.

[0029] Furthermore, the context feature fusion module of the latent topic includes:

[0030] The sentence-level feature vectors extracted by the model pre-training model are modeled to obtain the potential topic information between the previous text, the main text and the following text; the context feature fusion module of the potential topic starts from the perspective of enhancing the potential topic and identifies the classroom discourse type by mining the potential topic information between sentences.

[0031] In addition, the present invention also provides a microteaching evaluation and analysis system, comprising:

[0032] Presentation analysis module: Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis of presentations;

[0033] Classroom discourse analysis module: Construct a classroom discourse analysis model and conduct micro-teaching evaluation and analysis of classroom discourse through the classroom discourse analysis model.

[0034] In addition, the present invention also provides a micro-teaching evaluation and analysis storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned micro-teaching evaluation and analysis method is implemented.

[0035] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0036] This method constructs a presentation analysis model to conduct microteaching evaluation and analysis on presentations, and also constructs a classroom discourse analysis model to conduct microteaching evaluation and analysis on classroom discourse. This microteaching evaluation and analysis from both presentations and classroom discourse is more comprehensive, ensuring that the final analysis results better reflect the actual teaching situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of a microteaching evaluation and analysis method according to Example 1;

[0038] Figure 2 This is a block diagram of a microteaching evaluation and analysis system described in Example 3; DETAILED DESCRIPTION

[0039] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0040] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0041] Example 1:

[0042] This embodiment provides Figure 1 A microteaching evaluation and analysis method is shown, including:

[0043] In order to achieve the above objectives, the present invention provides a microteaching evaluation and analysis method, comprising:

[0044] Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis on presentations;

[0045] Construct a classroom discourse analysis model and use it to conduct micro-teaching evaluation and analysis of classroom discourse.

[0046] In addition, the present invention also provides a microteaching evaluation and analysis system, comprising:

[0047] Presentation analysis module: Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis of presentations;

[0048] Classroom discourse analysis module: Construct a classroom discourse analysis model and conduct micro-teaching evaluation and analysis of classroom discourse through the classroom discourse analysis model.

[0049] This example constructs a presentation analysis model to conduct microteaching evaluation and analysis on presentations, and also constructs a classroom discourse analysis model to conduct microteaching evaluation and analysis on classroom discourse. This microteaching evaluation and analysis from both presentations and classroom discourse is more comprehensive, ensuring that the final analysis results better reflect the actual teaching situation.

[0050] Example 2:

[0051] This embodiment further discloses the following on the basis of the first embodiment:

[0052] Furthermore, a presentation analysis model is constructed to conduct micro-teaching evaluation and analysis on the presentation, including:

[0053] The presentation analysis model includes: visual sub-model, large language sub-model, teacher sub-model, student base model and large language sub-model in the field of mathematics micro-grid;

[0054] Obtain the presentation and extract the statement from the presentation using the visual sub-model;

[0055] Get micro-teaching texts based on the expression sentences through a large language sub-model;

[0056] The teacher sub-model is tuned for prompt words based on the micro-teaching text to obtain the teacher sub-model output. The teacher sub-model output is then subjected to knowledge distillation and data cleaning to obtain a fine-tuning training dataset.

[0057] The student base model is fine-tuned according to the fine-tuning training dataset, and the output of the student base model is used as the input of a large language sub-model in the field of mathematics micro-teaching for micro-teaching evaluation and analysis.

[0058] Introducing mathematical thinking chain training data is a method to guide the model to think step by step, so that the model's response is based on the thinking process (CoT). During the previous knowledge distillation, the teacher model is asked to solve the problem step by step. Therefore, the collected data are all thinking chain answer data. Fine-tuning the student model allows the student model to strengthen its thinking ability and enhance its interpretability.

[0059] The model's thinking ability can be further enhanced by fine-tuning the model.

[0060] Further, including:

[0061] The student base submodel is the Qwen-1.5-7B-Chat model;

[0062] The teacher sub-model is the Deepseek-V2 236B model;

[0063] The visual sub-model is the Qwen-2.5-VL-3B model.

[0064] Furthermore, it also includes: combining retrieval enhancement generation technology when conducting microteaching evaluation and analysis in the field of mathematics microteaching using large language sub-models;

[0065] Retrieval-augmented generation is a framework that combines information retrieval and text generation techniques, aiming to improve the relevance and accuracy of the generative model output. The framework consists of a retrieval module and a generation module. The retrieval module efficiently retrieves documents or text fragments related to the input query from a pre-built knowledge base. The generation module usually uses a pre-trained generative language model, combining the user's input information with the external text information returned by the retrieval module as model input to obtain the final generated answer.

[0066] Furthermore, fine-tuning the student base model according to the fine-tuning training data set includes: fine-tuning the base model using a supervised fine-tuning method and a LoRA efficient fine-tuning algorithm according to the fine-tuning training data set.

[0067] The model's response effects vary under different prompt words. Prompt word tuning means constantly trying different prompt words as question prefixes in the hope of obtaining the desired results.

[0068] Therefore, in the knowledge distillation process of the teacher model, different prefix prompt words are constantly tried, the collected responses are submitted to experts for review, and finally suitable prompt words are selected as the prompt words used for the final distillation.

[0069] Supervised fine-tuning is based on pre-trained language models (such as GPT, BERT, etc.), using labeled task-specific data to further adjust model parameters through supervised learning to adapt them to specific downstream tasks (such as text classification, question answering, dialogue generation, etc.).

[0070] Key Steps

[0071] Pre-trained model initialization: Use pre-trained models (such as LLaMA, GPT-3) as the base model.

[0072] Task data preparation: Collect high-quality labeled data (input-output pairs) related to the target task.

[0073] Fine-tuning training: Calculate the loss (such as cross entropy loss) on the task data.

[0074] Model parameters are updated via backpropagation, typically using a small learning rate to prevent catastrophic forgetting.

[0075] Evaluate and Iterate: Evaluate performance on the validation set and adjust hyperparameters or data quality.

[0076] LoRA is a parameter-efficient fine-tuning (PEFT) method that freezes the pre-trained model parameters and only inserts a trainable low-rank matrix into the model to approximate parameter updates, significantly reducing the amount of training parameters.

[0077] Key steps:

[0078] Freeze the original model parameters: keep the pre-trained weights unchanged.

[0079] Insert low-rank matrices: Add trainable A and B next to the model's attention layer (such as Query, Value matrix).

[0080] Only low-rank parameters are trained: A and B are optimized via backpropagation, and other parameters are fixed.

[0081] Merge weights (optional): Add AB to the original weights, eliminating the need for additional calculations during deployment.

[0082] Furthermore, a classroom discourse analysis model is constructed to conduct teaching evaluation analysis on classroom discourse, including:

[0083] The classroom discourse analysis model is a context-enhanced dual-channel attention classification model based on a pre-trained model. It includes: a feature extraction module, a multi-semantic channel attention module, and a context feature fusion module based on latent topics.

[0084] Obtain classroom discourse, extract features of the classroom discourse through a feature extraction module, and obtain classroom discourse features;

[0085] The deep features of classroom discourse are extracted through the multi-semantic channel attention module to obtain the deep features of classroom discourse;

[0086] Classroom discourse analysis is conducted based on the characteristics of classroom discourse and the deep characteristics of classroom discourse through the contextual feature fusion module of potential topics.

[0087] Furthermore, the multi-semantic channel attention module includes:

[0088] The multi-semantic channel attention module uses the deep features extracted by multi-channel technology to make up for the shortcomings of the feature extraction module in extracting specific types of features; the features extracted by the dual channels are combined with the features extracted by the pre-trained model through the attention mechanism; specifically, the multi-semantic channel attention module uses the features output by each channel as the query, the classroom discourse features as the key and value, calculates the attention score through the query and key, and then recombine the value with the attention score as the weight.

[0089] Furthermore, the context feature fusion module of the latent topic includes:

[0090] The sentence-level feature vectors extracted by the model pre-training model are modeled to obtain the potential topic information between the previous text, the main text and the following text; the context feature fusion module of the potential topic starts from the perspective of enhancing the potential topic and identifies the classroom discourse type by mining the potential topic information between sentences.

[0091] This example constructs a presentation analysis model to conduct microteaching evaluation and analysis on presentations, and also constructs a classroom discourse analysis model to conduct microteaching evaluation and analysis on classroom discourse. This microteaching evaluation and analysis from both presentations and classroom discourse is more comprehensive, ensuring that the final analysis results better reflect the actual teaching situation.

[0092] Example 3:

[0093] This embodiment provides Figure 2 A microteaching evaluation and analysis system is shown, comprising:

[0094] Presentation analysis module: Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis of presentations;

[0095] Classroom discourse analysis module: Construct a classroom discourse analysis model and conduct micro-teaching evaluation and analysis of classroom discourse through the classroom discourse analysis model.

[0096] This example constructs a presentation analysis model to conduct microteaching evaluation and analysis on presentations, and also constructs a classroom discourse analysis model to conduct microteaching evaluation and analysis on classroom discourse. This microteaching evaluation and analysis from both presentations and classroom discourse is more comprehensive, ensuring that the final analysis results better reflect the actual teaching situation.

[0097] Example 4:

[0098] This embodiment also provides a micro-teaching evaluation and analysis storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned micro-teaching evaluation and analysis method is implemented.

[0099] This example constructs a presentation analysis model to conduct microteaching evaluation and analysis on presentations, and also constructs a classroom discourse analysis model to conduct microteaching evaluation and analysis on classroom discourse. This microteaching evaluation and analysis from both presentations and classroom discourse is more comprehensive, ensuring that the final analysis results better reflect the actual teaching situation.

[0100] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A presentation analysis model, characterized in that: include: Visual sub-model, large language sub-model, teacher sub-model, student base model and large language sub-model in the field of mathematics micro-grid; Obtain the presentation and extract the statement from the presentation using the visual sub-model; Get micro-teaching texts based on the expression sentences through a large language sub-model; The teacher sub-model is tuned for prompt words based on the micro-teaching text to obtain the teacher sub-model output. The teacher sub-model output is then subjected to knowledge distillation and data cleaning to obtain a fine-tuning training dataset. The student base model is fine-tuned according to the fine-tuning training dataset, and the output of the student base model is used as the input of a large language sub-model in the field of mathematics micro-teaching for micro-teaching evaluation and analysis.

2. A presentation analysis model according to claim 1, characterized in that: include: The student base submodel is the Qwen-1.5-7B-Chat model; The teacher sub-model is the Deepseek-V2 236B model; The visual sub-model is the Qwen-2.5-VL-3B model.

3. A presentation analysis model according to claim 1, characterized in that: Also includes: The large-scale language sub-model in the field of mathematics micro-teaching is combined with retrieval enhancement generation technology when conducting micro-teaching evaluation and analysis; Retrieval-augmented generation is a framework that combines information retrieval and text generation techniques, aiming to improve the relevance and accuracy of the generative model output. The framework consists of a retrieval module and a generation module. The retrieval module efficiently retrieves documents or text fragments related to the input query from a pre-built knowledge base. The generation module usually uses a pre-trained generative language model, combining the user's input information with the external text information returned by the retrieval module as model input to obtain the final generated answer.

4. A presentation analysis model according to claim 1, characterized in that: Fine-tuning the student base model according to the fine-tuning training data set includes: fine-tuning the base model using a supervised fine-tuning method and a LoRA efficient fine-tuning algorithm according to the fine-tuning training data set.

5. A classroom discourse analysis model, characterized by: include: The classroom discourse analysis model is a context-enhanced dual-channel attention classification model based on a pre-trained model. It includes: a feature extraction module, a multi-semantic channel attention module, and a context feature fusion module based on latent topics. Obtain classroom discourse, extract features of the classroom discourse through a feature extraction module, and obtain classroom discourse features; The deep features of classroom discourse are extracted through the multi-semantic channel attention module to obtain the deep features of classroom discourse; Classroom discourse analysis is conducted based on the characteristics of classroom discourse and the deep characteristics of classroom discourse through the contextual feature fusion module of potential topics.

6. A classroom discourse analysis model according to claim 5, characterized in that: The multi-semantic channel attention module includes: The multi-semantic channel attention module uses the deep features extracted by multi-channel technology to make up for the shortcomings of the feature extraction module in extracting specific types of features; the features extracted by the dual channels are combined with the features extracted by the pre-trained model through the attention mechanism; specifically, the multi-semantic channel attention module uses the features output by each channel as the query, the classroom discourse features as the key and value, calculates the attention score through the query and key, and then recombine the value with the attention score as the weight.

7. A classroom discourse analysis model according to claim 5, characterized in that: The contextual feature fusion module of latent topics includes: The sentence-level feature vectors extracted by the model pre-training model are modeled to obtain the potential topic information between the previous text, the main text and the following text; the context feature fusion module of the potential topic starts from the perspective of enhancing the potential topic and identifies the classroom discourse type by mining the potential topic information between sentences.

8. Application of a presentation analysis model according to any one of claims 1 to 4 and a classroom discourse analysis model according to any one of claims 5 to 7, characterized in that: For microteaching evaluation and analysis, the specific process is as follows: Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis on presentations; Construct a classroom discourse analysis model and use it to conduct micro-teaching evaluation and analysis of classroom discourse.

9. A microteaching evaluation and analysis system, characterized in that: Applying the presentation analysis model described in any one of claims 1 to 4 and the classroom discourse analysis model described in any one of claims 5 to 7 comprises: Presentation analysis module: Construct a presentation analysis model and use it to conduct micro-teaching evaluation and analysis of presentations; Classroom discourse analysis module: Construct a classroom discourse analysis model and conduct micro-teaching evaluation and analysis of classroom discourse through the classroom discourse analysis model.

10. A microteaching evaluation and analysis storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, a micro-teaching evaluation and analysis method according to any one of claims 1 to 8 is implemented.

Images