A smart teaching system and construction method based on a knowledge graph and virtual simulation

Abstract

The application relates to the technical field of teaching management, and discloses a wisdom teaching system and a construction method based on a knowledge graph and virtual simulation, which aims to effectively solve the problem of teaching resource integration and management and provide personalized teaching experience. The system comprises a knowledge graph construction module, which is used for integrating and standardizing teaching resources and constructing a knowledge graph comprising knowledge points and their correlation; a virtual simulation teaching module, which displays knowledge points in the form of virtual simulation by constructing a knowledge presentation model and an interactive learning model and evaluates student learning conditions; an intelligent teaching control module, which adopts a reinforcement learning algorithm to construct a teaching strategy model and dynamically adjusts teaching instructions according to student learning progress and interactive effects; and an execution presentation module, which is responsible for receiving and executing teaching instructions and presenting teaching content in a virtual teaching environment. The system improves the pertinence and effectiveness of teaching and stimulates the learning interest of students.

Description

Technical Field

[0001] This invention relates to the field of teaching management technology, specifically to a smart teaching system and its construction method based on knowledge graphs and virtual simulation. Background Technology

[0002] With the rapid development of information technology, the education field has also ushered in unprecedented changes. Traditional teaching methods often rely on teachers' experience and textbook content, making it difficult to fully meet students' personalized learning needs. At the same time, the integration and utilization of teaching resources also face many challenges, such as unclear relationships between knowledge points and the difficulty in unified management of scattered teaching resources. To solve these problems, smart teaching systems have emerged, among which smart teaching systems combining knowledge graph technology and virtual simulation technology have become an important research direction.

[0003] Knowledge graphs, as a structured knowledge representation method, can effectively integrate and manage teaching resources, including course content, teaching cases, and student feedback. By constructing a knowledge graph containing knowledge points, teaching resources, and their relationships, the inherent connections between knowledge points can be clearly displayed, providing strong support for teaching. However, the current application of knowledge graph technology in education still has some limitations, such as the complexity of the knowledge graph construction process and the difficulty of standardization, which restricts its widespread application in smart teaching systems.

[0004] On the other hand, virtual simulation technology, with its intuitive and interactive features, has shown great potential in the field of education. Through virtual simulation, abstract knowledge points can be presented in a visually appealing way, making them easier for students to understand and master. At the same time, virtual simulation technology can also provide rich interactive learning experiences, stimulating students' interest and enthusiasm for learning. However, current virtual simulation teaching systems often lack intelligent teaching control mechanisms and cannot dynamically adjust according to students' learning progress and interaction effects, thus affecting the improvement of teaching effectiveness. Summary of the Invention

[0005] The purpose of this invention is to provide a smart teaching system and construction method based on knowledge graphs and virtual simulation, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Firstly, this application provides a smart teaching system based on knowledge graphs and virtual simulation, the system comprising:

[0008] The knowledge graph construction module is used to integrate teaching resources, including course content, teaching cases, and student feedback, to build a knowledge graph containing knowledge points, teaching resources, and their relationships; the constructed knowledge graph is then standardized to form a structured knowledge dataset.

[0009] The virtual simulation teaching module is used to construct a knowledge presentation model and an interactive learning model. The knowledge presentation model is used to display knowledge points in a virtual simulation form, and the interactive learning model is used to evaluate students' learning progress and interaction effects. Based on a structured knowledge dataset, deep learning algorithms are used to train the knowledge presentation model and the interactive learning model respectively.

[0010] The intelligent teaching control module is used to generate personalized teaching instructions by utilizing the outputs of the knowledge presentation model and the interactive learning model. Specifically, it uses a reinforcement learning algorithm to construct a teaching strategy model, defines a learning evaluation function, and quantitatively evaluates the teaching results based on students' mastery of knowledge points, learning progress, and interaction effects. Based on the evaluation results, it dynamically adjusts the generated teaching instructions.

[0011] The execution presentation module is used to receive teaching instructions and present teaching content and perform teaching adjustment operations in the virtual teaching environment.

[0012] Preferably, the learning evaluation function is defined as follows:

[0013]

[0014] Where s represents the current state space of the teaching process, including the student's mastery of knowledge points, learning progress, and interaction history; a represents the teaching action taken; p1, p2, and p3 are the weight coefficients of the accuracy of knowledge point mastery, learning progress, and interaction effect, respectively, and p1+p2+p3=1; K(s,a), P(s,a), and I(s,a) represent the evaluation values ​​of the accuracy of knowledge point mastery, learning progress, and interaction effect after taking action a in state s, respectively.

[0015] Preferably, the intelligent teaching control module uses the Proximal Policy Optimization (PPO) algorithm in reinforcement learning to train the teaching strategy model.

[0016] Preferably, the steps for training the teaching strategy model include:

[0017] Step 1: Set the hyperparameters of the reinforcement learning algorithm, including the learning rate, discount factor, policy update step size, and pruning parameter; initialize the policy network and value network, both of which adopt a deep neural network structure. The policy network is used to output the probability distribution of teaching actions in the current state, and the value network is used to estimate the value of the state.

[0018] Step 2: Define an experience replay buffer to store samples of the teaching process and interaction with the environment. Each sample includes the current state, the action taken, the immediate reward received, the next state, and a flag indicating whether the process has ended.

[0019] Step 3: Acquire students' learning status data in real time through the knowledge graph construction module and convert it into a status representation; select actions based on the current policy network and execute them in the virtual simulation teaching environment, observe the feedback from the environment, including new status information and immediate rewards; store the interaction data in the experience replay buffer.

[0020] Step 4: Randomly select a batch of samples from the experience replay buffer; update the value network using the sample data, and optimize the network parameters by minimizing the error between the predicted value and the true value; update the policy network using the sample data and the output of the value network, maximize the expected value of the cumulative reward through the policy gradient method, and apply pruning techniques to control the policy update magnitude.

[0021] Step 5: Repeat steps 3 to 4 until the preset number of training rounds is reached or the performance of the policy network is stable; during the training process, gradually adjust the policy according to the policy update step size.

[0022] Preferably, a knowledge presentation model is constructed using the Graph Convolutional Network (GCN) algorithm. Features in the knowledge graph are extracted through graph convolution operations. Specifically, the GCN model is defined, and the number of graph convolutional layers, the size of each convolutional kernel, the type of activation function, the structure of the fully connected layers, and the representation of the output layer are determined. Among them, the graph convolutional layers are used to aggregate the feature information of neighboring nodes, the fully connected layers are used to map high-dimensional features to the teaching and display space, and the output layer adopts a virtual simulation display form that uses embedded vectors to represent knowledge points.

[0023] Preferably, the steps for training the knowledge presentation model include:

[0024] The standardized structured knowledge dataset is input into the GCN model, and the loss between the model output and the real displayed labels is calculated through the forward propagation algorithm. The loss function adopted is the mean squared error loss function.

[0025] The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm, and the model parameters are updated using the Adam optimization algorithm.

[0026] The performance of the trained GCN model is evaluated on an independent validation set, and the number of layers, kernel size, and learning rate of the GCN model are adjusted based on the evaluation results.

[0027] Preferably, an interactive learning model is constructed by combining an attention mechanism with a recurrent neural network (RNN) algorithm.

[0028] Preferably, the steps for training the interactive learning model include:

[0029] The interactive sequence data in the teaching process, including student operation records, system feedback and learning time, are arranged in chronological order to form interactive learning time sequence data;

[0030] An RNN model incorporating an attention mechanism is constructed, and the number of hidden layers, the number of hidden units in each layer, the activation function type, the learning rate, and the optimizer are determined. The RNN model consists of recurrent units, each of which receives the input at the current time step and the hidden state at the previous time step, and outputs the hidden state at the current time step. The attention mechanism is used to calculate the weighted sum of the hidden states at different time steps to highlight important information.

[0031] The serialized interactive learning data is input into an RNN model with an attention mechanism to learn the patterns and rules in the interactive learning process.

[0032] Based on the discrepancy between the learning progress and interaction effects predicted by the RNN model and the actual teaching data, the model's parameters and structure are adjusted.

[0033] Preferably, the execution presentation module includes a virtual environment rendering submodule and an interactive interface control submodule; the virtual environment rendering submodule is used to render a virtual teaching scene according to teaching instructions, including classroom layout, teaching tools and auxiliary teaching elements; the interactive interface control submodule is used to adjust the layout, display content and interaction mode of the interactive interface according to teaching instructions.

[0034] Preferably, the knowledge graph construction module further includes a data preprocessing submodule, which is used to clean, deduplicate, and normalize the integrated teaching resources.

[0035] Secondly, this application provides a method for constructing a smart teaching system based on knowledge graphs and virtual simulation, including the following steps:

[0036] Step S1: Collect course content, teaching cases and teaching resources, construct a knowledge graph containing knowledge points and their relationships, and standardize the knowledge graph to obtain a structured knowledge dataset;

[0037] Step S2: Use deep learning algorithms to build and train a knowledge presentation model based on a structured knowledge dataset. The knowledge presentation model is used to display knowledge points in a virtual simulation format.

[0038] Step S3: Use deep learning algorithms to build and train an interactive learning model based on a structured knowledge dataset. The interactive learning model is used to evaluate students' learning progress and interaction effects and output teaching result data.

[0039] Step S4: Construct a teaching strategy model based on reinforcement learning algorithm, define a learning evaluation function, and quantitatively evaluate the teaching results based on the teaching result data output by the interactive learning model. The teaching strategy model dynamically generates personalized teaching instructions based on the evaluation results.

[0040] Teaching outcome data includes students' mastery of knowledge points, learning progress, and interaction effectiveness;

[0041] Step S5: Use the knowledge presentation model to present the teaching content. According to the teaching instructions, present the teaching content in the virtual teaching environment. The interactive learning model outputs teaching result data. The teaching strategy model uses the learning evaluation function to quantitatively evaluate the teaching results based on the teaching result data and executes the preset teaching adjustment operations to complete the construction of the smart teaching system.

[0042] As can be seen from the above technical solutions, this application has the following advantages:

[0043] The knowledge graph construction module comprehensively integrates multi-dimensional teaching resources such as course content, teaching cases, and student feedback to construct a structured knowledge graph containing knowledge points, teaching resources, and their relationships. This process not only clarifies the intrinsic connections between knowledge points but also forms a unified and standardized structured knowledge dataset, greatly improving the efficiency of teaching resource integration and the convenience of management.

[0044] The system integrates a virtual simulation teaching module, utilizing deep learning algorithms to train knowledge presentation and interactive learning models. This allows teaching content to be presented in an intuitive and interactive virtual simulation format, while accurately assessing students' learning progress and interaction effectiveness. This personalized teaching approach better adapts to the learning needs and pace of different students, improving the relevance and effectiveness of teaching. The intelligent teaching control module employs reinforcement learning algorithms to construct a teaching strategy model. By defining a learning evaluation function, it quantifies and assesses students' mastery of knowledge points, learning progress, and interaction effectiveness in real time. Based on this evaluation result, the system dynamically adjusts generated teaching instructions to ensure a close match between teaching strategies and students' learning status, thereby continuously optimizing teaching effectiveness and improving students' learning experience and performance.

[0045] The intuitiveness and interactivity of virtual simulation technology, along with the personalized learning experience and dynamically adjusted teaching strategies provided by the system, work together to make learning more vivid and engaging. This teaching method effectively stimulates students' interest and enthusiasm, encouraging them to participate more actively in learning activities, creating a virtuous cycle and promoting the sustainable development of education. Attached Figure Description

[0046] To more clearly illustrate the technical solution of this application, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the working principle of the intelligent teaching system described in this invention;

[0048] Figure 2 A flowchart illustrating the process of training a teaching strategy model using the PPO algorithm with a proximal strategy optimization approach;

[0049] Figure 3 This is a flowchart illustrating the construction of a knowledge presentation model using the Graph Convolutional Network (GCN) algorithm. Detailed Implementation

[0050] To make the purpose, features, and advantages of this application more apparent and understandable, specific embodiments and accompanying drawings will be used to clearly and completely describe the technical solution protected by this application. Obviously, the embodiments described below are only some embodiments of this application, and not all embodiments. Based on the embodiments in this patent, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this patent.

[0051] 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.

[0052] Please see Figures 1-3 This invention provides a technical solution: a smart teaching system and construction method based on knowledge graphs and virtual simulation, the system comprising:

[0053] 1. Knowledge Graph Construction Module:

[0054] Resource Integration and Knowledge Graph Construction: Various teaching resources are collected and integrated, including but not limited to course content, teaching cases, and student feedback. These resources are preprocessed using expert systems or automated tools to extract key knowledge points and their interrelationships. Then, using graph database technology, a knowledge graph containing knowledge points, teaching resources, and their relationships is constructed. Nodes in the graph represent knowledge points or teaching resources, while edges represent the relationships between them.

[0055] Standardization: The constructed knowledge graph undergoes standardization to ensure that the data format is consistent and standardized. This step includes data cleaning, deduplication, and normalization to ultimately form a structured knowledge dataset.

[0056] 2. Virtual simulation teaching module:

[0057] Model Construction: In the virtual simulation teaching module, a knowledge presentation model and an interactive learning model are constructed. The knowledge presentation model is responsible for displaying knowledge points in a virtual simulation format, such as presenting abstract concepts intuitively through 3D animation, virtual reality scenes, and other intuitive methods. The interactive learning model is used to evaluate students' learning progress and interaction effectiveness, such as through data such as students' operation records and answer status.

[0058] Model Training: Based on a structured knowledge dataset, deep learning algorithms are used to train the knowledge presentation model and the interactive learning model. Through iterative training on a large amount of data, the model is able to accurately display knowledge points in a virtual simulation format and accurately assess students' learning progress.

[0059] 3. Intelligent teaching control module:

[0060] Teaching strategy model construction: A teaching strategy model is constructed using reinforcement learning algorithms. This model generates teaching instructions based on students' learning status (such as their mastery of knowledge points, learning progress, and interaction effectiveness).

[0061] Learning Assessment and Instruction Adjustment: A learning assessment function is defined to quantitatively evaluate students' learning outcomes. This function comprehensively considers multiple dimensions, including students' answer accuracy, learning speed, and interaction frequency. Based on the assessment results, the teaching strategy model dynamically adjusts the generated instructional instructions. For example, if a student's grasp of a particular knowledge point is weak, the instruction might be adjusted to provide more related exercises or detailed explanations; if a student's learning progress is fast, the instruction might be adjusted to introduce more advanced knowledge points or increase the learning difficulty.

[0062] 4. Execute the presentation module:

[0063] Command Reception and Presentation: The presentation module receives teaching commands from the intelligent teaching control module and presents the corresponding teaching content in the virtual teaching environment. This includes displaying knowledge points in virtual simulation form and performing teaching adjustment operations (such as switching teaching scenes and adjusting teaching difficulty).

[0064] Interactive feedback processing: During the teaching process, the execution presentation module is also responsible for collecting students' interactive feedback (such as operation records, answer status, etc.) and passing this feedback data to the interactive learning model in the virtual simulation teaching module so that the model can carry out subsequent learning and evaluation.

[0065] The present invention will be further described below with reference to Examples 1 to 5:

[0066] Example 1:

[0067] Step S1: Collect course content, teaching cases and teaching resources, construct a knowledge graph containing knowledge points and their relationships, and standardize the knowledge graph to obtain a structured knowledge dataset;

[0068] Step S2: Use deep learning algorithms to build and train a knowledge presentation model based on a structured knowledge dataset. The knowledge presentation model is used to display knowledge points in a virtual simulation format.

[0069] Step S3: Use deep learning algorithms to build and train an interactive learning model based on a structured knowledge dataset. The interactive learning model is used to evaluate students' learning progress and interaction effects and output teaching result data.

[0070] Step S4: Construct a teaching strategy model based on reinforcement learning algorithm, define a learning evaluation function, and quantitatively evaluate the teaching results based on the teaching result data output by the interactive learning model. The teaching strategy model dynamically generates personalized teaching instructions based on the evaluation results.

[0071] Teaching outcome data includes students' mastery of knowledge points, learning progress, and interaction effectiveness;

[0072] Step S5: Use the knowledge presentation model to present the teaching content. According to the teaching instructions, present the teaching content in the virtual teaching environment. The interactive learning model outputs teaching result data. The teaching strategy model uses the learning evaluation function to quantitatively evaluate the teaching results based on the teaching result data and executes the preset teaching adjustment operations to complete the construction of the smart teaching system.

[0073] In this embodiment, step S1 specifically includes the following steps:

[0074] Step S1-1: Clean, deduplicate, and normalize the integrated teaching resources;

[0075] Step S1-2: Use graph database technology to store and manage knowledge points and their relationships.

[0076] In this embodiment, step S2 includes the following steps:

[0077] Step S2-1: Define the number of graph convolutional layers, kernel size, activation function type, and fully connected layer structure of the knowledge presentation model;

[0078] Step S2-2: Aggregate the features of neighboring nodes in the knowledge graph through graph convolutional layers;

[0079] Steps S2-3: Map high-dimensional features to the teaching and display space through a fully connected layer, output virtual simulation forms of knowledge points with embedded vectors, and complete the knowledge presentation model;

[0080] Step S2-4: Input the structured knowledge dataset into the knowledge presentation model, calculate the loss between the model output and the actual displayed labels through forward propagation, and use the mean squared error loss function;

[0081] Step S2-5: Calculate the loss gradient through backpropagation and update the knowledge presentation model parameters using the Adam optimization algorithm;

[0082] Steps S2-6: Evaluate the model performance using an independent validation set, adjust the number of model layers, convolutional kernel size, and learning rate to complete the training of the knowledge presentation model.

[0083] In this embodiment, step S3 includes the following steps:

[0084] Step S3-1: Arrange the interactive sequence data of the teaching process in chronological order to form interactive learning time series data;

[0085] Step S3-2: Construct an interactive learning model that incorporates an attention mechanism, and determine the number of hidden layers, the number of hidden units, the type of activation function, and the optimizer;

[0086] Step S3-3: Input the interactive learning data into the interactive learning model to predict the learning progress and interaction effect;

[0087] Step S3-4: Adjust the model parameters based on the error between the predicted value and the actual teaching data to complete the interactive learning model training.

[0088] In this embodiment, the learning evaluation function in step S4 is defined as:

[0089]

[0090] Where s represents the teaching state space, a represents the teaching action, p1, p2, and p3 are the weight coefficients of the accuracy of knowledge point mastery, learning progress, and interaction effect, respectively, and p1+p2+p3=1; K(s,a), P(s,a), and I(s,a) represent the evaluation values ​​of the accuracy of knowledge point mastery, learning progress, and interaction effect, respectively.

[0091] In this embodiment, step S4 includes the following steps:

[0092] Step S4-1: Set the learning rate, discount factor, policy update step size, and pruning parameters; initialize the policy network and value network, and construct the teaching policy model;

[0093] Step S4-2: Define an experience replay buffer to store interaction samples;

[0094] Step S4-3: Acquire student learning status data in real time, select and execute actions based on the policy network, and store feedback data;

[0095] Step S4-4: Randomly sample and update the value network and policy network, and optimize the teaching policy model by minimizing the error and maximizing the expected reward.

[0096] Steps S4-5: Repeat the training process until the policy network is stable or reaches the preset number of training rounds, thus completing the training of the teaching policy model.

[0097] In this embodiment, in step S4, the personalized teaching instructions include instructions for the order of knowledge point display, content depth, teaching interaction methods, and difficulty adjustment.

[0098] Example 2:

[0099] This invention achieves intelligent control of the teaching process by defining a specific learning evaluation function and training a teaching strategy model using the Proximal Policy Optimization (PPO) algorithm. The following is a detailed implementation of this system:

[0100] I. Definition of Learning Evaluation Function

[0101] The learning evaluation function L(s,a) is used to quantify the teaching effect after taking teaching action a in a specific teaching state s. It is defined as follows:

[0102] L(s,a)=p1*K(s,a)+p2*P(s,a)+p3*I(s,a)

[0103] in:

[0104] s represents the current state space of the teaching process, including the student's mastery of knowledge points, learning progress, and interaction history.

[0105] 'a' indicates the teaching action taken, such as presenting specific knowledge points, providing practice questions, or adjusting the difficulty of teaching.

[0106] p1, p2, and p3 are the weighting coefficients for the accuracy of knowledge mastery, learning progress, and interaction effect, respectively, and satisfy p1+p2+p3=1. These coefficients are set according to the teaching objectives and key points.

[0107] K(s,a) represents the accuracy assessment value of knowledge mastery after taking action a in state s. It can be measured by the student's answer accuracy rate or knowledge test scores.

[0108] P(s,a) represents the learning progress assessment value after taking action a in state s, which can be measured by the number of learning tasks completed by the student or the amount of learning time invested.

[0109] I(s,a) represents the evaluation value of the interaction effect after taking action a in state s, which can be measured by student participation, frequency of questions, or quality of feedback.

[0110] II. Training of Teaching Strategy Models

[0111] The teaching strategy model is trained using the Proximal Policy Optimization (PPO) algorithm from reinforcement learning, as detailed below:

[0112] Configure the hyperparameters of the reinforcement learning algorithm, including the learning rate, discount factor (used to calculate the cumulative reward), policy update step size, and pruning parameter (used to control the magnitude of policy updates). Initialize the policy network and value network. The policy network uses a deep neural network structure, with the current state s as input and the probability distribution of each taught action to be taken in the current state as output. The value network also uses a deep neural network structure, with the current state s as input and the estimated value of that state as output.

[0113] Define an experience replay buffer to store samples of the teaching process and interactions with the environment. Each sample includes the current state, the action taken, the immediate reward received (calculated according to the learning evaluation function L(s,a)), the next state, and a flag indicating whether the process has terminated.

[0114] The knowledge graph construction module acquires students' learning status data in real time and transforms it into a state representation. An action is selected based on the current policy network, i.e., a teaching action is randomly chosen according to the probability distribution output by the policy network. The selected teaching action is executed in a virtual simulation teaching environment, and the feedback from the environment is observed, including new state information and immediate rewards. The interaction data (current state, action taken, immediate reward, next state, and termination status) is stored in the experience replay buffer.

[0115] A batch of samples is randomly drawn from the experience replay buffer. The value network is updated using the sample data. The parameters of the value network are optimized by minimizing the error between the predicted value and the true value. The true value can be calculated using a discount factor and an immediate reward. The policy network is updated using the sample data and the output of the value network. The expected value of the cumulative reward is maximized using the policy gradient method. Simultaneously, a pruning technique is applied to control the magnitude of the policy update, preventing excessively large policy updates from causing training instability.

[0116] The training process is repeated until the preset number of training rounds is reached or the policy network's performance stabilizes. During training, the policy is gradually adjusted according to the policy update step size, allowing the teaching strategy model to gradually adapt to changes in students' learning needs and the teaching environment.

[0117] Example 3:

[0118] This intelligent teaching system employs the Graph Convolutional Network (GCN) algorithm to construct a knowledge presentation model, effectively extracting features from the knowledge graph and mapping these features to the teaching display space to achieve virtual simulation display of knowledge points. The GCN model is defined as follows:

[0119] Model Structure Determination: The number of graph convolutional layers in the GCN model is determined based on the complexity of the knowledge graph and the required feature depth. For example, simpler knowledge graphs require fewer convolutional layers, while complex graphs require more to capture deeper features. The size of each convolutional kernel is determined, as it determines the range of neighbor nodes that can be aggregated in each convolution operation. Larger kernels capture broader neighbor information but may increase computational complexity. The activation function type, such as ReLU (Rectified Linear Unit) or Sigmoid, is chosen to enhance the model's non-linear expressive power. The structure of fully connected layers is designed to map the high-dimensional features extracted by the graph convolutional layers to the teaching and demonstration space. The number of fully connected layers and the number of neurons in each layer are determined based on specific teaching and demonstration requirements. The representation of the output layer is determined, using embedded vectors to represent knowledge points in a virtual simulation format. The dimension of the embedded vectors is determined based on the dimension of the demonstration space.

[0120] The function of graph convolutional layers: Graph convolutional layers are mainly used to aggregate feature information from neighboring nodes. For each node (i.e., knowledge point) in the knowledge graph, the graph convolutional layer aggregates the feature information of its neighboring nodes onto that node through convolution operations, thereby forming a new node feature representation.

[0121] The role of fully connected layers: Fully connected layers map the high-dimensional features extracted by graph convolutional layers to the teaching and demonstration space. Through the transformation of fully connected layers, the feature representations of nodes can be converted into embedding vectors suitable for teaching and demonstration.

[0122] The training steps for the knowledge presentation model include:

[0123] Data preparation: The structured knowledge dataset is standardized to ensure data consistency and comparability. Standardization includes data cleaning, deduplication, and normalization.

[0124] Model training process:

[0125] The standardized structured knowledge dataset is input into the GCN model. The loss between the model output and the actual displayed labels is calculated using the forward propagation algorithm. The mean squared error (MSE) loss function is used to measure the difference between the model output and the actual displayed labels.

[0126] The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm. The model parameters are then updated using the Adam optimization algorithm, which combines the advantages of the momentum method and the RMSprop method, enabling rapid convergence and finding optimal model parameters. During training, steps S101 and S102 are iterated continuously until the preset number of training epochs is reached or the loss function converges.

[0127] Evaluate the performance of the trained GCN model on an independent validation set. The validation set is used to test the model's generalization ability, ensuring that the model performs well on unseen data. Based on the evaluation results on the validation set, adjust hyperparameters of the GCN model, such as the number of layers, kernel size, and learning rate. If the model performs poorly on the validation set, try increasing or decreasing the number of graph convolutional layers, adjusting the kernel size, or changing hyperparameters such as the learning rate to optimize model performance.

[0128] Example 4:

[0129] This intelligent teaching system employs an attention mechanism combined with a recurrent neural network (RNN) algorithm to construct an interactive learning model. This model captures and analyzes interactive sequence data during the teaching process, thereby learning the patterns and dynamics within the interactive learning process. The following is a detailed implementation of this system:

[0130] Preparation of interactive sequence data:

[0131] Collect interactive sequence data during the teaching process. This data includes student operation records (such as clicks, inputs, selections, etc.), system feedback (such as prompts, explanations, evaluations, etc.), and learning time (the time point when each interactive event occurred). Arrange this interactive data in chronological order to form interactive learning time series data. This data structure can reflect the dynamic changes and temporal relationships in the teaching process.

[0132] Construction of RNN models incorporating attention mechanisms:

[0133] Construct an RNN model consisting of recurrent units. Each recurrent unit receives the input at the current time step and the hidden state from the previous time step, and outputs the hidden state at the current time step. This structure enables RNNs to process sequential data and capture temporal dependencies within it.

[0134] An attention mechanism is introduced into the RNN model. This mechanism is used to compute a weighted sum of hidden states at different time steps to highlight important information. Specifically, for each hidden state at a time step, the attention mechanism assigns a weight based on its importance, and then combines these hidden states through a weighted sum to form the final representation.

[0135] Determine the number of hidden layers, the number of hidden units per layer, the type of activation function, the learning rate, and the optimizer for the RNN model. The choice of these parameters affects the model's performance and training efficiency. For example, the number of hidden layers and hidden units determines the model's complexity, the type of activation function affects the model's non-linear expressiveness, and the learning rate and optimizer affect the model's convergence speed and stability.

[0136] Model input and training:

[0137] The serialized interactive learning data is input into an RNN model incorporating an attention mechanism. The model processes this input data sequentially over time, learning patterns and dynamics. During training, the model predicts learning progress and interaction effectiveness based on the input data. The error between these predictions and the actual teaching data serves as the target for model optimization.

[0138] Adjustment of model parameters and structure:

[0139] Based on the discrepancy between the learning progress and interaction effects predicted by the RNN model and the actual teaching data, the model's parameters and structure are adjusted. Specifically, the gradient of the error with respect to the model parameters can be calculated using the backpropagation algorithm, and the optimizer can be used to update the model parameters.

[0140] Additionally, structural parameters such as the number of hidden layers and the number of hidden units per layer can be adjusted based on the model's performance. If the model performs well on the training set but poorly on the validation set, overfitting may be a problem. In this case, reducing the number of hidden layers or hidden units can be considered to simplify the model. This iterative training and adjustment process continues until the model's performance on the validation set reaches a satisfactory level.

[0141] Example 5:

[0142] The execution presentation module is responsible for presenting the teaching content in the form of virtual simulation, including the virtual environment rendering submodule and the interactive interface control submodule.

[0143] The virtual environment rendering submodule renders virtual teaching scenes according to teaching instructions. These scenes include, but are not limited to, classroom layouts, teaching tools, and auxiliary teaching elements.

[0144] Classroom layout rendering: Based on teaching needs, this submodule can dynamically generate classroom layouts of different styles, such as traditional classrooms, laboratories, and discussion rooms. The layout includes basic facilities such as a podium, desks, and chairs, and their positions and orientations are adjustable.

[0145] Teaching tool rendering: This submodule supports the virtual presentation of various teaching tools, such as blackboards, whiteboards, projectors, and experimental equipment. These tools can be freely placed and used in the virtual scene according to teaching instructions.

[0146] Rendering of supplementary teaching elements: To enhance teaching effectiveness, this submodule can also render supplementary teaching elements, such as wall charts, models, and animations. These elements can be customized according to the teaching content to present knowledge points in an intuitive and vivid way.

[0147] Rendering Technology: Utilizing advanced 3D rendering technology, we ensure the realism and interactivity of virtual scenes. Simultaneously, we optimize rendering performance to ensure smooth operation across various devices.

[0148] The interactive interface control submodule adjusts the layout, displayed content, and interaction methods of the interactive interface according to the teaching instructions.

[0149] Interface layout adjustment: The sub-module can dynamically adjust the layout of the interactive interface according to the teaching content and user needs. For example, when explaining theoretical knowledge, the interface mainly uses text and images; when conducting experiments, the interface highlights the experimental equipment and operating procedures.

[0150] Content display control: This submodule can update the content displayed on the interactive interface in real time to ensure synchronization with the teaching progress. It also supports content customization and expansion to meet the needs of different teaching scenarios.

[0151] Interaction Method Settings: The submodule provides various interaction methods, such as clicking, dragging, and zooming, to support user interaction with the virtual scene. These interaction methods can be set and adjusted according to the teaching content and user habits.

[0152] The knowledge graph construction module is responsible for integrating and processing teaching resources to build the knowledge graph. The data preprocessing submodule is a key component. Specifically, it includes:

[0153] The data preprocessing submodule is used to clean, deduplicate, and normalize the integrated teaching resources to ensure the accuracy and consistency of the knowledge graph.

[0154] Data cleaning: This submodule first cleans the teaching resources, removing noise and irrelevant information. For example, it filters out duplicate, erroneous, or outdated content to ensure the accuracy and reliability of the data.

[0155] Data deduplication: Building upon the data cleaning process, this submodule further deduplicates the teaching resources. By comparing resource similarity or hash values, duplicate resources are identified and merged, avoiding redundant information in the knowledge graph.

[0156] Data normalization: To ensure the consistency of the knowledge graph, this submodule normalizes the teaching resources. Data from different sources and in different formats is converted into a unified format and standard, facilitating subsequent knowledge graph construction and querying.

[0157] The basic concepts have been described above. Obviously, for those skilled in the art, the detailed disclosure above is merely illustrative and does not constitute a limitation of this specification. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to this specification. Such modifications, improvements, and corrections are suggested in this specification and therefore remain within the spirit and scope of the exemplary embodiments described herein.

[0158] Furthermore, unless expressly stated in the claims, the order of processing elements and sequences, the use of numbers and letters, or other names described in this specification are not intended to limit the order of the processes and methods described herein. Although various examples have been discussed in the foregoing disclosure of some embodiments of the invention that are currently considered useful, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments; rather, the claims are intended to cover all modifications and equivalent combinations that conform to the spirit and scope of the embodiments described herein. For example, while the system components described above can be implemented using hardware devices, they can also be implemented solely using software solutions, such as installing the described system on existing servers or mobile devices.

[0159] Similarly, it should be noted that, in order to simplify the description disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments in this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0160] In some embodiments, numbers describing the quantity of components and attributes are used. It should be understood that such numbers used in the description of embodiments are modified in some examples with the terms "approximately," "approximately," or "generally." Unless otherwise stated, "approximately," "approximately," or "generally" indicates that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may be changed depending on the characteristics required by individual embodiments. In some embodiments, numerical parameters should take into account specified significant digits and employ a general method of digit reservation. Although the numerical ranges and parameters used to confirm their breadth of range in some embodiments of this specification are approximate values, in specific embodiments, such values ​​are set as precisely as feasible.

[0161] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A smart teaching system based on a knowledge graph and virtual simulation, characterized in that, The system includes: The knowledge graph construction module is used to integrate teaching resources, including course content, teaching cases, and student feedback, to build a knowledge graph containing knowledge points, teaching resources, and their relationships; the constructed knowledge graph is then standardized to form a structured knowledge dataset. The virtual simulation teaching module is used to construct a knowledge presentation model and an interactive learning model. The knowledge presentation model is used to display knowledge points in a virtual simulation form, and the interactive learning model is used to evaluate students' learning progress and interaction effects. Based on a structured knowledge dataset, deep learning algorithms are used to train the knowledge presentation model and the interactive learning model respectively. A knowledge presentation model is constructed using the Graph Convolutional Network (GCN) algorithm. Features from the knowledge graph are extracted through graph convolution operations. Specifically, the GCN model is defined, determining the number of graph convolutional layers, the size of each convolutional kernel, the type of activation function, the structure of the fully connected layers, and the representation of the output layer. The graph convolutional layers aggregate feature information from neighboring nodes, the fully connected layers map high-dimensional features to the teaching and display space, and the output layer uses embedded vectors to represent knowledge points in a virtual simulation display format. An interactive learning model is constructed by combining an attention mechanism with a recurrent neural network (RNN) algorithm. The steps for training an interactive learning model include: The interactive sequence data in the teaching process, including student operation records, system feedback and learning time, are arranged in chronological order to form interactive learning time sequence data; An RNN model incorporating an attention mechanism is constructed, and the number of hidden layers, the number of hidden units in each layer, the activation function type, the learning rate, and the optimizer are determined. The RNN model consists of recurrent units, each of which receives the input at the current time step and the hidden state at the previous time step, and outputs the hidden state at the current time step. The attention mechanism is used to calculate the weighted sum of the hidden states at different time steps to highlight important information. The serialized interactive learning data is input into an RNN model with an attention mechanism to learn the patterns and rules in the interactive learning process. Based on the error between the learning progress and interaction effects predicted by the RNN model and the actual teaching data, adjust the parameters and structure of the model. The intelligent teaching control module is used to generate personalized teaching instructions by utilizing the outputs of the knowledge presentation model and the interactive learning model. Specifically, it uses a reinforcement learning algorithm to construct a teaching strategy model, defines a learning evaluation function, and quantitatively evaluates the teaching results based on students' mastery of knowledge points, learning progress, and interaction effects. Based on the evaluation results, it dynamically adjusts the generated teaching instructions. The execution presentation module is used to receive teaching instructions and present teaching content and perform teaching adjustment operations in the virtual teaching environment. 2.The wisdom teaching system based on knowledge graph and virtual simulation according to claim 1, characterized in that, The learning evaluation function is defined as follows: Where s represents the current state space of the teaching process, including the student's mastery of knowledge points, learning progress, and interaction history; a represents the teaching action taken; p1, p2, and p3 are the weight coefficients of the accuracy of knowledge point mastery, learning progress, and interaction effect, respectively, and p1+p2+p3=1; K(s,a), P(s,a), and I(s,a) represent the evaluation values ​​of the accuracy of knowledge point mastery, learning progress, and interaction effect after taking action a in state s, respectively. 3.The wisdom teaching system based on knowledge graph and virtual simulation according to claim 2, characterized in that, The intelligent teaching control module uses the Proximal Policy Optimization (PPO) algorithm in reinforcement learning to train the teaching strategy model. 4.The wisdom teaching system based on knowledge graph and virtual simulation according to claim 1, characterized in that, The steps for training an instructional strategy model include: Step 1: Set the hyperparameters of the reinforcement learning algorithm, including the learning rate, discount factor, policy update step size, and pruning parameter; initialize the policy network and value network, both of which adopt a deep neural network structure. The policy network is used to output the probability distribution of teaching actions in the current state, and the value network is used to estimate the value of the state. Step 2: Define an experience replay buffer to store samples of the teaching process and interaction with the environment. Each sample includes the current state, the action taken, the immediate reward received, the next state, and a flag indicating whether the process has ended. Step 3: Acquire students' learning status data in real time through the knowledge graph construction module and convert it into a status representation; select actions based on the current policy network and execute them in the virtual simulation teaching environment, observe the feedback from the environment, including new status information and immediate rewards; store the interaction data in the experience replay buffer. Step 4: Randomly select a batch of samples from the experience replay buffer; update the value network using the sample data, and optimize the network parameters by minimizing the error between the predicted value and the true value; update the policy network using the sample data and the output of the value network, maximize the expected value of the cumulative reward through the policy gradient method, and apply pruning techniques to control the policy update magnitude. Step 5: Repeat steps 3 to 4 until the preset number of training rounds is reached or the performance of the policy network is stable; during the training process, gradually adjust the policy according to the policy update step size. 5.The wisdom teaching system based on knowledge graph and virtual simulation according to claim 1, characterized in that, The steps for training a knowledge presentation model include: The standardized structured knowledge dataset is input into the GCN model, and the loss between the model output and the real displayed labels is calculated through the forward propagation algorithm. The loss function adopted is the mean squared error loss function. The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm, and the model parameters are updated using the Adam optimization algorithm. The performance of the trained GCN model is evaluated on an independent validation set, and the number of layers, kernel size, and learning rate of the GCN model are adjusted based on the evaluation results. 6.The wisdom teaching system based on knowledge graph and virtual simulation according to claim 1, characterized in that, The execution presentation module includes a virtual environment rendering submodule and an interactive interface control submodule. The virtual environment rendering submodule is used to render a virtual teaching scene according to teaching instructions, including classroom layout, teaching tools, and auxiliary teaching elements. The interactive interface control submodule is used to adjust the layout, display content, and interaction mode of the interactive interface according to teaching instructions. 7.The wisdom teaching system based on knowledge graph and virtual simulation according to claim 1, characterized in that, The knowledge graph construction module also includes a data preprocessing submodule, which is used to clean, deduplicate, and normalize the integrated teaching resources.

8. A method for constructing a smart teaching system based on knowledge graphs and virtual simulation, characterized in that, Includes the following steps: Step S1: Collect course content, teaching cases and teaching resources, construct a knowledge graph containing knowledge points and their relationships, and standardize the knowledge graph to obtain a structured knowledge dataset; Step S2: Use deep learning algorithms to build and train a knowledge presentation model based on a structured knowledge dataset. The knowledge presentation model is used to display knowledge points in a virtual simulation format. A knowledge presentation model is constructed using the Graph Convolutional Network (GCN) algorithm. Features from the knowledge graph are extracted through graph convolution operations. Specifically, the GCN model is defined, determining the number of graph convolutional layers, the size of each convolutional kernel, the type of activation function, the structure of the fully connected layers, and the representation of the output layer. The graph convolutional layers aggregate feature information from neighboring nodes, the fully connected layers map high-dimensional features to the teaching and display space, and the output layer uses embedded vectors to represent knowledge points in a virtual simulation display format. Step S3: Use deep learning algorithms to build and train an interactive learning model based on a structured knowledge dataset. The interactive learning model is used to evaluate students' learning progress and interaction effects and output teaching result data. An interactive learning model is constructed by combining an attention mechanism with a recurrent neural network (RNN) algorithm. The steps for training an interactive learning model include: The interactive sequence data in the teaching process, including student operation records, system feedback and learning time, are arranged in chronological order to form interactive learning time sequence data; An RNN model incorporating an attention mechanism is constructed, and the number of hidden layers, the number of hidden units in each layer, the activation function type, the learning rate, and the optimizer are determined. The RNN model consists of recurrent units, each of which receives the input at the current time step and the hidden state at the previous time step, and outputs the hidden state at the current time step. The attention mechanism is used to calculate the weighted sum of the hidden states at different time steps to highlight important information. The serialized interactive learning data is input into an RNN model with an attention mechanism to learn the patterns and rules in the interactive learning process. Based on the error between the learning progress and interaction effects predicted by the RNN model and the actual teaching data, adjust the parameters and structure of the model. Step S4: Construct a teaching strategy model based on reinforcement learning algorithm, define a learning evaluation function, and quantitatively evaluate the teaching results based on the teaching result data output by the interactive learning model. The teaching strategy model dynamically generates personalized teaching instructions based on the evaluation results. Teaching outcome data includes students' mastery of knowledge points, learning progress, and interaction effectiveness; Step S5: Use the knowledge presentation model to present the teaching content. According to the teaching instructions, present the teaching content in the virtual teaching environment. The interactive learning model outputs teaching result data. The teaching strategy model uses the learning evaluation function to quantitatively evaluate the teaching results based on the teaching result data and executes the preset teaching adjustment operations to complete the construction of the smart teaching system.

Images