Online teaching quality supervision and management method and system based on large model

Abstract

The invention provides an online teaching quality supervision and management method and system based on a large model, which are used for solving the problems of low network teaching quality supervision and management efficiency, insufficient data privacy protection, difficulty in processing large-scale online teaching data and the like. A large amount of data of online teaching activities are collected and processed, local model training is performed on each device, and the central server updates a global model, so that real-time and accurate evaluation and prediction of teaching quality are realized, and the online teaching quality is improved. Meanwhile, a differential privacy technology is combined, data privacy is protected, student data security is ensured, and the efficiency of processing large-scale online teaching data is improved. And a deep learning model is added and a federal learning concept is utilized, so that the accuracy of teaching quality evaluation is improved, and the data processing capability is also improved. Compared with an existing online teaching quality management method, the online teaching quality management method has higher efficiency and accuracy and is more suitable for large-scale online teaching scenes.

Description

Technical Field

[0001] The present invention belongs to the field of teaching quality management, and more specifically to an online teaching quality supervision management method and system based on a large model. Background Art

[0002] Online teaching plays an important role in today's education field. However, due to its own characteristics, the teaching quality often lacks effective monitoring and guarantee. In addition, the characteristics of online teaching, such as remote, decentralized, large-scale, and heterogeneous, also make it difficult to implement traditional teaching quality monitoring methods. For example, the existence of temporal and spatial dispersion makes real-time interaction difficult; the processing and storage of large amounts of data also brings technical challenges; heterogeneity makes the data formats and types of teaching activities different, making it difficult to manage and analyze them uniformly. Therefore, how to effectively monitor and manage teaching quality in this context is an urgent problem to be solved.

[0003] In recent years, with the development of deep learning and big data analysis technologies, these technologies have been widely used in online teaching quality management. By introducing machine learning methods, a large amount of online teaching data can be automatically analyzed to mine valuable information, thereby helping to improve teaching quality. However, the above traditional centralized model training methods often have problems such as low efficiency and difficulty in ensuring data privacy when dealing with large-scale, distributed online teaching data.

[0004] Federated learning, as an emerging distributed machine learning method, can effectively solve the above problems. It trains the model locally on the client device and then sends the model parameters to the server for aggregation, thereby updating the global model. This method can achieve massive data processing across devices while protecting data privacy, and has great application potential.

[0005] However, the actual application of federated learning in online education is still relatively small, and there is no mature online teaching quality supervision and management system. Therefore, how to comprehensively utilize large models, federated learning and other technologies to design a quality monitoring system for online teaching is a very challenging topic. In response to this problem, the present invention proposes a large-scale model-based online teaching quality supervision and management method and system, hoping to use this technology to create a teaching quality monitoring system that can effectively guarantee the quality of online teaching, fully protect student privacy, and cope with large-scale online teaching scenarios. Summary of the invention

[0006] The present invention aims to solve the current problems of low efficiency in online teaching quality supervision and management, insufficient data privacy protection, and difficulty in processing large-scale online teaching data, and further proposes a new online teaching quality supervision management method and system based on a large model, which can effectively monitor and analyze online teaching activities in real time and accurately, while also ensuring the privacy security of student data.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the method comprises:

[0008] Clarify tasks, supervise and manage the teaching quality of online teachers, evaluate teaching strategies, and analyze student feedback;

[0009] Data collection: collect data from online teaching, including interaction records between teachers and students, and data on various teaching behaviors;

[0010] Data preprocessing, in addition to cleaning and normalization, uses natural language processing techniques to analyze text data, and feature selection is performed through automated tools and expert knowledge;

[0011] Divide the data to ensure that the data is divided properly and each device can effectively perform local training;

[0012] Global model architecture design: Select convolutional neural network as the basic model and use hybrid model to combine different types of neural networks;

[0013] Global model update and distribution: The central server sends the global model back to each device. The device uses the global model and its own data for retraining, and then uploads the model parameters again. This process is repeated until the model converges.

[0014] In one embodiment, in the data preprocessing, natural language processing of text data includes:

[0015] Text cleaning: remove punctuation, stop words, and other irrelevant characters;

[0016] Tokenization and stemming: breaking text into words or stems;

[0017] Feature extraction: Use TF-IDF to represent text data.

[0018] In one embodiment, the feature selection uses LASSO regression or the like to perform feature importance assessment.

[0019] In one solution, the partitioned data applies differential privacy technology to further protect data privacy; differential privacy protects the privacy of individual data by adding noise to data or model updates; before data processing on each device, noise is added to mask individual data and Gaussian noise is added when calculating gradients;

[0020] For a function f, the output after adding noise is:

[0021] f′(x)=f(x)+N(0,σ 2 )

[0022] Among them, N(0,σ 2 ) has a mean of 0 and a variance of σ 2 Gaussian noise.

[0023] In one embodiment, the global model architecture design includes:

[0024] Choose the basic model: Select convolutional neural network (CNN) for processing image data and sequence data; for comprehensive data, use a hybrid model that combines different types of neural networks together;

[0025] Local training: select the loss function for regression tasks and cross entropy loss for classification tasks; use the gradient descent algorithm to update the model parameters on the local device;

[0026] Model update and aggregation: After each client completes a certain round of training locally, it uploads the model update to the central server; the server performs a weighted average of the model updates uploaded by all clients to update the global model;

[0027] Federated average aggregation formula:

[0028]

[0029] Among them, w k is the model update for the kth device, n k is the data volume of the kth device, and n is the total data volume of all devices;

[0030] When uploading model updates, add noise to achieve differential privacy protection to ensure that individual data cannot be inferred from model updates; use differential privacy mechanism to add Gaussian noise: w k′ =w k +N(0,σ 2 )

[0031] Among them, N(0,σ 2 ) has a mean of 0 and a variance of σ 2 Gaussian noise.

[0032] Iterative training: The entire process is iterated in multiple communication rounds, gradually improving the accuracy and generalization ability of the model; after each round, the updated global model will be sent to each client for the next round of local training.

[0033] In one embodiment, the global model updating and distribution includes:

[0034] Global model update: At the end of each round of training, each client device sends its locally trained model update to the central server.

[0035] The server aggregates the local updates received and performs a weighted average of the model updates of each client based on the amount of data from each client to generate a new global model.

[0036] Global model distribution: The updated global model parameters are distributed to all participating client devices. After receiving the new global model, each device uses it as the initial model for the next round of local training.

[0037] Local retraining: Initialize the local model. The device uses the received global model parameters to initialize its local model. The device continues to train the model using its own local data. This process includes several small batches of training to improve the local adaptability of the model.

[0038] Iteration: The above process is continuously iterated. After each round of training, the client device uploads the updated model parameters to the server again, and the server performs a new round of global update and distribution.

[0039] In one aspect, a large model-based online teaching quality supervision and management system, the system is applicable to the method, and the system includes:

[0040] Data collection module: used to collect data from online teaching, including interaction records between teachers and students, and data on various teaching behaviors;

[0041] Data preprocessing module: used for data cleaning and normalization, using natural language processing technology to analyze text data, and feature selection through automated tools and expert knowledge;

[0042] Data partitioning module: used to reasonably partition data to ensure that data can be effectively trained locally on each device, while using differential privacy technology to protect data privacy;

[0043] Global model architecture design module: responsible for selecting basic models, such as convolutional neural networks, and designing other hybrid models for local training;

[0044] Global model update and distribution module: The central server sends the global model back to each device. The device uses the global model and its own data for retraining, and then uploads the model parameters again. This process is repeated until the model converges. The global model update and distribution module can also add noise to protect differential privacy.

[0045] Model monitoring module: used to monitor the performance and effectiveness of the global model to ensure the effectiveness of online teaching quality supervision and management.

[0046] Beneficial effects of the present invention:

[0047] By collecting a large amount of data from online teaching activities and using large models for in-depth analysis and training, it is possible to evaluate and predict teaching quality in real time and accurately, thereby improving the quality and effectiveness of online teaching.

[0048] Utilizing the concept of federated learning, the data of each device is processed and updated locally, and then the updated model parameters are fed back to the central server. This not only ensures the privacy of the data, but also greatly improves the efficiency of processing large-scale online teaching data.

[0049] In the design of the global model architecture, the selected convolutional neural network can effectively process image data, and the hybrid model can process more complex comprehensive data, so that the model can effectively learn and predict various types of online teaching behaviors. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the method of the present invention;

[0051] Figure 2 Design model flow chart for the present invention;

[0052] Figure 3 A flow chart is provided for updating and distributing the global model of the present invention. DETAILED DESCRIPTION

[0053] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0054] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used in the present invention in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0055] Embodiment 1:

[0056] like Figure 1 As shown in the figure, the online teaching quality supervision and management method based on the big model has the following implementation steps:

[0057] Step 1: Clarify the tasks. It is necessary to supervise and manage the teaching quality of online teachers, which involves the evaluation of teaching strategies, analysis of student feedback, etc.

[0058] The core of this stage is to clarify the system's goals and functional requirements to ensure the direction and pertinence of subsequent development work.

[0059] First, we need to clarify the basic goal of the system, which is to supervise and manage the teaching quality of online teachers. This involves real-time monitoring and evaluation of teachers' teaching strategies, and analyzing students' feedback to obtain an objective evaluation of teaching effectiveness. In order to achieve this goal, the system needs to have a variety of functional modules, including automated analysis of teaching behavior, monitoring of student interactions, and quality evaluation of teaching content.

[0060] Secondly, to help teachers continuously improve their teaching strategies, the system should integrate an efficient feedback mechanism. Through this mechanism, the system can collect and analyze students’ feedback and convert it into actionable suggestions. These suggestions can be presented to teachers through an automated suggestion system to help them identify strengths and weaknesses in their teaching. For example, the system can automatically generate personalized teaching guidance based on students’ engagement and understanding, prompting teachers to adjust the teaching rhythm or change teaching methods.

[0061] When designing a feedback mechanism, the timeliness and personalization of feedback are required. The system should be able to collect data in real time during the teaching process and quickly generate feedback reports after class. In addition, the feedback content should be customized according to the characteristics of different teachers and students to ensure its pertinence and effectiveness. For example, for some teachers, the system can focus on the depth and breadth of teaching content, while for other teachers, the system can focus more on the frequency and quality of teaching interactions.

[0062] To achieve the above functions, the system needs to rely on powerful data analysis and processing capabilities. Natural language processing technology can be used to analyze students' text feedback, and machine learning algorithms can be used to identify patterns and trends in teaching. In addition, the system can also use deep learning models to predict teaching effects and provide teachers with data-driven decision support.

[0063] Step 2: Data collection: On the basis of ensuring data privacy and security, collect data from online teaching, including interaction records between teachers and students, and data on various teaching behaviors.

[0064] In the process of realizing the online teaching quality supervision and management system based on large models, data collection is a crucial link. Its success or failure directly affects the training effect of subsequent models and the overall performance of the system. Therefore, in the process of data collection, the diversity and comprehensiveness of the data as well as user privacy and security must be fully considered.

[0065] First, the scope of data collection needs to be clear. In an online teaching environment, the types of data involved include but are not limited to teachers' lecture videos and audio recordings, teaching materials, student interaction records (such as questions, discussions, and homework submissions), student feedback and evaluation results, etc. In order to ensure the comprehensiveness of the data, the system needs to be able to obtain data from multiple channels and platforms, involving integration with learning management systems (LMS), video conferencing platforms, and online interactive tools.

[0066] Data diversity is key to supporting model training. Collect data of different types and formats as much as possible to train models that can handle complex teaching scenarios. This includes structured data (such as ratings and attendance records) and unstructured data (such as text feedback and video content). With diverse data sources, the system can have a more comprehensive understanding of all aspects of teaching activities, thereby improving the generalization ability of the model.

[0067] During the data collection process, user privacy and data security must be prioritized. First, data anonymization technology can be used to remove or obfuscate personally identifiable information to protect user privacy. This includes processing the names, contact information, and other sensitive information of students and teachers so that they cannot be easily identified. Secondly, encryption technology should be used during data transmission and storage to prevent unauthorized access and data leakage. Common encryption methods include symmetric encryption and asymmetric encryption, and the specific choice can be determined based on the performance requirements and security level of the system.

[0068] Step 3: Data preprocessing,Data preprocessing is an important step to improve model performance. In addition to cleaning and normalization, natural language processing techniques are used to analyze text data (such as student feedback). Feature selection can be performed by automated tools or expert knowledge to ensure that the most meaningful features are selected.

[0069] In the process of building an online teaching quality supervision and management system based on a large model, data preprocessing is an indispensable step. After the data collection stage, the data obtained is usually disorganized and in different formats, so systematic preprocessing is required to improve the training effect and prediction performance of the model. The following is a detailed implementation process of data preprocessing, which combines key steps such as data cleaning, normalization, natural language processing, and feature selection.

[0070] Data cleaning

[0071] Data cleaning is the first step in data preprocessing, which aims to remove or correct noise and errors in the data. The specific steps include:

[0072] Missing value handling: Missing values ​​can be handled in a variety of ways, such as deleting samples with missing values, filling missing values ​​with the mean or median, or using more complex interpolation methods.

[0073] Formula: For missing value filling, mean filling can be used:

[0074]

[0075] Among them, x i are non-missing values, and N is the number of non-missing values.

[0076] Outlier detection: Use statistical methods (such as the standard deviation method) or machine learning algorithms (such as isolation forests) to detect and handle outliers to prevent them from adversely affecting the model.

[0077] Deduplication: Identify and remove duplicate records in a data set to reduce data redundancy.

[0078] Data Normalization

[0079] Data normalization is to convert data of different dimensions to the same scale. Commonly used methods include minimum-maximum normalization

[0080]

[0081] Among them, x min and x max are the minimum and maximum values ​​of feature x, respectively.

[0082] Natural Language Processing,For textual data (such as student feedback), natural language processing (NLP) techniques can be used to extract valuable information.

[0083] Text cleaning: remove punctuation, stop words, and other irrelevant characters.

[0084] Tokenization and stemming: Breaking text into words or stems for easier analysis.

[0085] Feature extraction: Use bag-of-words, TF-IDF (term frequency-inverse document frequency), or word vectors (such as Word2Vec, BERT) to represent text data.

[0086] Formula: TF-IDF calculation:

[0087] TF-IDF(t,d)=TF(t,d)×IDF(t)

[0088] Among them, TF(t,d) is the term frequency of word t in document d, and IDF(t) is the inverse document frequency, which is calculated as:

[0089]

[0090] N is the total number of documents, and the denominator is the number of documents containing word t.

[0091] Feature selection is to extract the most informative features from the data to improve the efficiency and performance of the model. Feature importance is evaluated using LASSO regression, etc.

[0092] Combined with the experience and knowledge of domain experts, certain features are manually selected or excluded.

[0093] Formula: Objective function of LASSO regression:

[0094]

[0095] Among them, ‖w‖1 is the L1 norm and λ is the regularization parameter.

[0096] Through the above steps, data preprocessing provides clean, standardized, and information-rich data input for model training, thus laying a solid foundation for subsequent model training and prediction.

[0097] Step 4: Divide the data. The core of federated learning is the distributed processing of data. Ensure that the data is divided reasonably so that each device can effectively perform local training. Use differential privacy technology to further protect the data.

[0098] In the process of building an online teaching quality supervision and management system based on large models, data partitioning is a key step in achieving federated learning. Federated learning allows local model training on multiple devices without centralizing data on a central server, thereby improving data privacy and security. In this context, reasonable data partitioning and privacy protection strategies are crucial.

[0099] Data partitioning strategy

[0100] In federated learning, data is usually distributed across multiple client devices, such as students' and teachers' personal computers or mobile devices. With natural partitioning, data can be directly leveraged because it exists naturally on each user's device. For example, each student's learning records, feedback, and interaction data can be saved on their personal device without the need to upload them centrally.

[0101] To further protect data privacy, differential privacy technology is applied in federated learning. Differential privacy protects the privacy of individual data by adding noise to data or model updates.

[0102] Local differential privacy: Add noise to mask individual data before processing it on each device. For example, add Laplace noise or Gaussian noise when computing gradients.

[0103] Formula: For a function f, the output after adding noise is:

[0104] f′(x)=f(x)+N(0,σ 2 )

[0105] Among them, N(0,σ 2 ) has a mean of 0 and a variance of σ 2 Gaussian noise.

[0106] In the framework of joint learning, the specific form of data division is as follows:

[0107] Local data set: Each device retains its own local data set, including structured data (such as grades, attendance records) and unstructured data (such as text feedback, audio and video recordings). This data is used to train local models.

[0108] Local model update: In each training round, the device trains the model based on the local dataset and calculates model updates (such as gradients).

[0109] Model aggregation: Each device sends local model updates to a central server, which aggregates these updates to update the global model. Common aggregation methods include Federated Averaging, which is a weighted average of model updates from all devices.

[0110] Formula: The update formula of the federated average is:

[0111]

[0112] Among them, w k is the model update for the kth device, n k is the data volume of the kth device, and n is the total data volume of all devices.

[0113] Through reasonable data partitioning and the application of differential privacy technology, federated learning can achieve efficient distributed model training while protecting data privacy. This provides a safe and efficient solution for the online teaching quality supervision and management system.

[0114] Step 5: Design the model and select a hybrid model, Federated Neural Network.

[0115] In building an online teaching quality supervision and management system based on a large model, model design is the core step to achieve system functions and performance. Combined with the previous steps of data collection, preprocessing, and partitioning, a hybrid model suitable for joint learning can be designed: the Federated Neural Network (FNN). This model can be effectively trained in a distributed environment while protecting data privacy.

[0116] like Figure 2 As shown in the figure, the design and implementation of the federated neural network:

[0117] Federated neural network is a deep learning model suitable for federated learning. Its main feature is that it allows local training on each client device and aggregates model updates on the server side through the federated learning framework. The following is the detailed implementation process of the federated neural network:

[0118] S501. Model architecture design: The basic model selects convolutional neural network (CNN) for processing image data and sequence data (such as text feedback). For comprehensive data, a hybrid model is used to combine different types of neural networks.

[0119] According to the characteristics of the data and the task requirements, design a suitable network hierarchy, including input layer, hidden layer and output layer. Components such as fully connected layer, convolutional layer, pooling layer and activation function (such as ReLU) can be used.

[0120] S502, local training: Select an appropriate loss function, such as mean square error (MSE) for regression tasks and cross entropy loss for classification tasks. Use a gradient descent algorithm (such as SGD, Adam) to update model parameters on the local device.

[0121] Formula: For a sample (x i ,y i ), the loss function is L(f(x i ),y i ), the gradient descent update formula is:

[0122]

[0123] Where η is the learning rate, is the gradient of the loss function with respect to the parameter w.

[0124] S503, model update and aggregation: After each client completes a certain round of training locally, it uploads the model update (such as gradient or model parameters) to the central server. The server performs a weighted average of the model updates uploaded by all clients to update the global model.

[0125] Formula: Federated Averaging aggregation formula:

[0126]

[0127] Among them, w k is the model update for the kth device, n k is the data volume of the kth device, and n is the total data volume of all devices.

[0128] S504, Differential privacy protection: When uploading model updates, add noise to achieve differential privacy protection to ensure that individual data cannot be inferred from model updates. Using the differential privacy mechanism, add Gaussian noise:

[0129] w k′ =w k +N(0,σ 2 )

[0130] Among them, N(0,σ 2 ) has a mean of 0 and a variance of σ 2 Gaussian noise.

[0131] S505, iterative training: The whole process is iterated in multiple communication rounds, gradually improving the accuracy and generalization ability of the model. After each round, the updated global model will be sent to each client for the next round of local training.

[0132] Through the design of this federated neural network, it is possible to use distributed data for effective model training while ensuring data privacy. This model is suitable for various tasks in the online teaching quality supervision and management system, such as student behavior analysis, learning effect prediction, and feedback sentiment analysis.

[0133] Step 6: Global model update and distribution. The central server sends the global model back to each device. The device uses the global model and its own data for retraining and then uploads the model parameters again. This process is repeated until the model converges.

[0134] In the framework of federated learning, the updating and distribution of the global model is a key step to ensure that the model is gradually optimized and eventually converges. Combined with the preparation and design of the previous steps 1 to 5, step 6 describes in detail the updating and distribution process of the global model, as well as how the device uses the global model for retraining.

[0135] like Figure 3 As shown in Figure 2, the implementation process of global model update and distribution

[0136] S601, Global Model Update: At the end of each round of training, each client device sends the model update (such as gradient or model parameters) obtained from its local training to the central server. The server aggregates the received local updates, usually using the Federated Averaging method. Specifically, the server performs a weighted average of the model updates of each client based on the amount of data to generate a new global model.

[0137] Formula: The federated average update formula is:

[0138]

[0139] in, is the model parameter of the kth client in round t, n k is the data volume of the kth client, and n is the total data volume of all clients.

[0140] S602, global model distribution: updated global model parameters It is sent to all participating client devices. After each device receives the new global model, it uses it as the initial model for the next round of local training.

[0141] S603, local retraining: Initialize the local model. The device uses the received global model parameters to initialize its local model. Local data training: The device continues to train the model using its own local data. This process may include training of several mini-batches to improve the local adaptability of the model.

[0142] Formula: For local training, the model parameters are updated using gradient descent:

[0143]

[0144] Where η is the learning rate, is the gradient of the loss function.

[0145] S604, loop iteration: The above process continues to iterate, and the client device uploads the updated model parameters to the server again after each round of training, and the server performs a new round of global update and distribution. This loop process continues until the performance of the global model reaches the expected convergence standard on the validation set, or the number of training rounds reaches the preset maximum value.

[0146] Through this global model update and distribution mechanism, federated learning can effectively use distributed data for model training while maintaining the local privacy of the data. This process ensures the consistency and collaborative optimization of the model on multiple devices, ultimately achieving efficient distributed learning.

[0147] The trained federated neural network model is embedded into the online teaching quality supervision system for real-time teaching quality monitoring.

[0148] Embodiment 2:

[0149] Based on the online teaching quality supervision management method based on a large model in Example 1, a large model-based online teaching quality supervision management system is constructed. The system is applicable to the method described, and the system includes:

[0150] Data collection module: used to collect data from online teaching, including interaction records between teachers and students, and data on various teaching behaviors;

[0151] Data preprocessing module: used for data cleaning and normalization, using natural language processing technology to analyze text data, and feature selection through automated tools and expert knowledge;

[0152] Data partitioning module: used to reasonably partition data to ensure that data can be effectively trained locally on each device, while using differential privacy technology to protect data privacy;

[0153] Global model architecture design module: responsible for selecting basic models, such as convolutional neural networks, and designing other hybrid models for local training;

[0154] Global model update and distribution module: The central server sends the global model back to each device. The device uses the global model and its own data for retraining, and then uploads the model parameters again. This process is repeated until the model converges. The global model update and distribution module can also add noise to protect differential privacy.

[0155] Model monitoring module: used to monitor the performance and effectiveness of the global model to ensure the effectiveness of online teaching quality supervision and management.

[0156] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0157] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for supervising and managing the quality of online teaching based on a large model, characterized by: The method includes: Clarify tasks, supervise and manage the teaching quality of online teachers, evaluate teaching strategies, and analyze student feedback; Data collection: collect data from online teaching, including interaction records between teachers and students, and data on various teaching behaviors; Data preprocessing, in addition to cleaning and normalization, uses natural language processing techniques to analyze text data, and feature selection is performed through automated tools and expert knowledge; Divide the data to ensure that the data is divided properly and each device can effectively perform local training; Global model architecture design: Select convolutional neural network as the basic model and use hybrid model to combine different types of neural networks; Global model update and distribution: The central server sends the global model back to each device. The device uses the global model and its own data for retraining, and then uploads the model parameters again. This process is repeated until the model converges.

2. The online teaching quality supervision and management method based on a large model according to claim 1 is characterized by: In the data preprocessing, natural language processing of text data includes: text cleaning: removing punctuation marks, stop words and other irrelevant characters; Tokenization and stemming: breaking text into words or stems; Feature extraction: Use TF-IDF to represent text data.

3. The online teaching quality supervision and management method based on a large model according to claim 1 is characterized by: The feature selection described uses LASSO regression to evaluate feature importance.

4. The online teaching quality supervision and management method based on a large model according to claim 1 is characterized by: In order to further protect data privacy, the partitioned data applies differential privacy technology; differential privacy protects the privacy of individual data by adding noise to data or model updates; before data processing on each device, noise is added to mask individual data and Gaussian noise is added when calculating gradients; For a function f, the output after adding noise is: f′(x)=f(x)+N(0,σ 2 ) Among them, N(0,σ 2 ) has a mean of 0 and a variance of σ 2 Gaussian noise.

5. The online teaching quality supervision and management method based on a large model according to claim 1 is characterized by: The global model architecture design includes: Choose the basic model: Select convolutional neural network (CNN) for processing image data and sequence data; for comprehensive data, use a hybrid model that combines different types of neural networks together; Local training: select the loss function for regression tasks and cross entropy loss for classification tasks; use the gradient descent algorithm to update the model parameters on the local device; Model update and aggregation: After each client completes a certain round of training locally, it uploads the model update to the central server; the server performs a weighted average of the model updates uploaded by all clients to update the global model; Federated average aggregation formula: Among them, w k is the model update for the kth device, n k is the data volume of the kth device, and n is the total data volume of all devices; When uploading model updates, add noise to achieve differential privacy protection to ensure that individual data cannot be inferred from model updates; use differential privacy mechanism to add Gaussian noise: w k′ =w k +N(0,σ 2 ) Among them, N(0,σ 2 ) has a mean of 0 and a variance of σ 2 Gaussian noise; Iterative training: The entire process is iterated in multiple communication rounds, gradually improving the accuracy and generalization ability of the model; after each round, the updated global model will be sent to each client for the next round of local training.

6. The method for supervising and managing the quality of online teaching based on a large model according to claim 1 is characterized by: The global model update and distribution includes: Global model update: At the end of each round of training, each client device sends its locally trained model update to the central server; The server aggregates the local updates received and performs a weighted average of the model updates of each client based on the amount of data to generate a new global model. Global model distribution: The updated global model parameters are distributed to all participating client devices. After receiving the new global model, each device uses it as the initial model for the next round of local training. Local retraining: Initialize the local model. The device uses the received global model parameters to initialize its local model. The device continues to train the model using its own local data. This process includes several small batches of training to improve the local adaptability of the model. Iteration: The above process is continuously iterated. After each round of training, the client device uploads the updated model parameters to the server again, and the server performs a new round of global update and distribution.

7. An online teaching quality supervision and management system based on a large model, wherein the system is applicable to the method according to any one of claims 1 to 6, characterized in that: The system comprises: Data collection module: used to collect data from online teaching, including interaction records between teachers and students, and data on various teaching behaviors; Data preprocessing module: used for data cleaning and normalization, using natural language processing technology to analyze text data, and feature selection through automated tools and expert knowledge; Data partitioning module: used to reasonably partition data to ensure that data can be effectively trained locally on each device, while using differential privacy technology to protect data privacy; Global model architecture design module: responsible for selecting basic models, such as convolutional neural networks, and designing other hybrid models for local training; Global model update and distribution module: The central server sends the global model back to each device. The device uses the global model and its own data for retraining, and then uploads the model parameters again. This process is repeated until the model converges. The global model update and distribution module can also add noise to protect differential privacy. Model monitoring module: used to monitor the performance and effectiveness of the global model to ensure the effectiveness of online teaching quality supervision and management.

Images