Micro-course generation method and system based on knowledge graph

Abstract

The invention discloses a knowledge graph-based micro-course generation method and system, and relates to the technical field of education science and technology, and the method comprises the steps: inputting a cross-modal knowledge graph into a neural symbol hybrid engine, carrying out the topological sorting of multi-modal knowledge nodes through a graph neural network, and generating a knowledge point sequence; a dynamic teaching strategy is generated by fusing the cognitive weight of the neural network and the logic constraint of the symbol rule; according to the knowledge point sequence and the dynamic teaching strategy, calling a content generator to synthesize a multi-modal micro-course, converting decision logic of the dynamic teaching strategy into a visual traceability identifier, embedding the visual traceability identifier into the multi-modal micro-course, and generating a traceability micro-course; monitoring interaction behaviors of the user and the multi-modal micro-course in real time, calculating an independent synchronization deviation value, and when the independent synchronization deviation value exceeds a cognitive threshold value, updating the multi-modal knowledge node through a feature alignment loss function; according to the method, quantum state fusion feature vectors are generated through parameterized entanglement gate fusion, and unified modeling and optimized expression of image-text-sound information are achieved.

Description

Technical Field

[0001] The present invention relates to the field of educational technology, and in particular to a method and system for generating micro-lessons based on a knowledge graph. Background Art

[0002] With the rapid development of technologies such as artificial intelligence, educational big data, and knowledge engineering, personalized learning and intelligent teaching systems have gradually become important development directions in educational informatization. In particular, in massive online open courses (MOOCs), micro-course platforms, and adaptive learning platforms, how to efficiently build structured, interpretable, and traceable teaching resources has become a key issue in improving teaching quality and learning efficiency. In recent years, knowledge graphs, as a core technology for semantic networks and structured knowledge representation, have been widely used in the education field to model semantic relationships between knowledge points, support learning path planning, and achieve personalized recommendations.

[0003] However, existing technologies still have shortcomings. First, most methods rely on single-modal data input and lack effective mechanisms for integrating multimodal information such as images, text, and audio. This results in incomplete knowledge representation, affecting the integrity and cognitive adaptability of teaching content. Second, traditional knowledge graph construction processes typically use static rules or shallow feature matching, making it difficult to dynamically adapt to user learning behavior feedback and lacking the ability to respond in real time to changes in learners' cognitive states, thereby reducing the personalization level and practical application effectiveness of teaching strategies. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a micro-course generation method based on knowledge graph to solve the problems of the shortcomings of traditional methods in cross-modal knowledge fusion and real-time personalized response.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a micro-course generation method based on a knowledge graph, which includes collecting textbook text data, teaching image data and explanation audio data, and extracting text semantic features, visual features and audio spectrum features through a deep neural network; fusing text semantic features, visual features and audio spectrum features to generate a quantum state fusion feature vector, constructing a multimodal knowledge node after optimization by a feature alignment loss function, and calculating a quantum coherence correlation matrix to construct a weighted relationship edge to generate a cross-modal knowledge graph; inputting the cross-modal knowledge graph into a neural symbolic hybrid engine, topologically sorting the multimodal knowledge nodes through a graph neural network to generate a knowledge point sequence, and generating a dynamic teaching strategy by fusing the cognitive weights of the neural network with the logical constraints of the symbolic rules; calling a content generator to synthesize a multimodal micro-course based on the knowledge point sequence and the dynamic teaching strategy, converting the decision logic of the dynamic teaching strategy into a visual traceability identifier embedded in the multimodal micro-course to generate a traceable micro-course; monitoring the user's interaction behavior with the multimodal micro-course in real time and calculating the independent synchronization deviation value, and when the independent synchronization deviation value exceeds the cognitive threshold, updating the multimodal knowledge node through the feature alignment loss function.

[0007] As a preferred solution of the micro-course generation method based on knowledge graph described in the present invention, the specific steps of constructing multimodal knowledge nodes are as follows: Map text semantic features, visual features, and audio spectrum features to quantum states to generate text quantum states, image quantum states, and audio quantum states; The text quantum state, image quantum state and audio quantum state are fused through parameterized entanglement gates, and quantum circuit measurements are performed to generate quantum state fusion feature vectors. The feature alignment loss function is used to perform regularization optimization on the quantum state fusion feature vector, generate the optimized quantum state fusion feature vector, and construct a multimodal knowledge node.

[0008] As a preferred solution of the micro-course generation method based on knowledge graph described in the present invention, the specific steps of generating a cross-modal knowledge graph are as follows: Based on the quantum state fusion eigenvector, the quantum coherence correlation matrix between multimodal knowledge nodes is calculated; The dynamic standard deviation threshold is used to filter the quantum coherence correlation matrix to generate candidate relationship edge sets; Based on the candidate relationship edge set, weighted relationship edges are constructed between multimodal knowledge nodes, and multimodal knowledge nodes are combined to generate a cross-modal knowledge graph.

[0009] As a preferred solution of the micro-course generation method based on knowledge graph described in the present invention, the specific steps of generating a knowledge point sequence are as follows: Map the multimodal knowledge nodes in the cross-modal knowledge graph into quantum state encoding vectors and inject the symbolic rule projection operator to generate the quantum constraint matrix; Using the quantum constraint matrix as input parameter, a topological sorting Hamiltonian is constructed, and the probability distribution function of the knowledge point sequence is generated by solving it through a quantum-classical hybrid optimizer; According to the probability distribution function of the knowledge point sequence, the knowledge point sequence is generated by sorting in descending order of probability values.

[0010] As a preferred solution of the micro-course generation method based on knowledge graph described in the present invention, the specific steps of generating dynamic teaching strategies are as follows: The quantum state encoding vector is input into the cognitive weight neural network to generate the strategy superposition state vector, and the Prolog rule is compiled through the symbolic rule engine to generate the unitary operator matrix to obtain the symbolic constraint state vector; Input the symbolic constraint state vector into the quantum-neural interference Hamiltonian model to calculate the probability amplitude of the basic teaching strategy; According to the probability amplitude of the basic teaching strategy, a dynamic teaching strategy is generated through dynamic threshold decision-making and real-time cognitive safety verification.

[0011] As a preferred solution of the micro-course generation method based on knowledge graph described in the present invention, wherein: according to the knowledge point sequence and dynamic teaching strategy, the content generator is called to synthesize multimodal micro-courses, the decision logic of the dynamic teaching strategy is converted into a visual traceability identifier embedded in the multimodal micro-course, and a traceable micro-course is generated. The specific steps are as follows: Input the knowledge point sequence and dynamic teaching strategy into the multimodal content generator to generate the original micro-course media stream, and inject the quantum cognitive field to generate a micro-course containing the field; Based on dynamic teaching strategies and the space-time coordinates of field-containing micro-lessons, a quantum decision traceability matrix is ​​constructed; The quantum decision traceability matrix is ​​input into the bionic retinal encoder, which outputs the visual traceability fluctuation signal and is converted into a quantum traceability mark through the triple integral operator; The quantum traceability identifier is dynamically embedded into the corresponding coordinates of the original micro-course media stream to generate a traceable micro-course.

[0012] As a preferred solution of the micro-course generation method based on knowledge graph described in the present invention, wherein: the real-time monitoring of the user's interaction behavior with the multimodal micro-course and the calculation of the independent synchronization deviation value are performed. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge node is updated through the feature alignment loss function. The specific steps are as follows: Collect eye movement trajectories, brain waves, and click event streams of users interacting with traceable micro-courses in real time, and calculate independent synchronization deviation values; When any independent synchronization deviation value exceeds the cognitive threshold, the feature alignment loss function is triggered to update the feature vector of the corresponding multimodal knowledge node in the cross-modal knowledge graph.

[0013] In the second aspect, the present invention provides a micro-course generation system based on a knowledge graph, including a feature extraction module, a graph construction module, a strategy generation module, a micro-course synthesis module and a feedback optimization module. The feature extraction module is used to collect teaching material text data, teaching image data and explanation audio data, and extract text semantic features, visual features and audio spectrum features through a deep neural network; the graph construction module is used to fuse text semantic features, visual features and audio spectrum features to generate a quantum state fusion feature vector, and after optimization by a feature alignment loss function, construct a multimodal knowledge node, and calculate the quantum coherence correlation matrix to construct a weighted relationship edge to generate a cross-modal knowledge graph; the strategy generation module is used to convert The cross-modal knowledge graph is input into the neural-symbolic hybrid engine, which topologically sorts the multimodal knowledge nodes through the graph neural network to generate a knowledge point sequence, and generates a dynamic teaching strategy by fusing the cognitive weight of the neural network with the logical constraints of the symbolic rules; the micro-course synthesis module is used to call the content generator to synthesize multimodal micro-courses based on the knowledge point sequence and dynamic teaching strategy, and convert the decision logic of the dynamic teaching strategy into a visual traceability identifier embedded in the multimodal micro-course to generate a traceable micro-course; the feedback optimization module is used to monitor the user's interaction behavior with the multimodal micro-course in real time and calculate the independent synchronization deviation value. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge node is updated through the feature alignment loss function.

[0014] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the micro-course generation method based on the knowledge graph as described in the first aspect of the present invention.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the micro-course generation method based on the knowledge graph as described in the first aspect of the present invention.

[0016] The beneficial effects of the present invention are: by generating quantum state fusion feature vectors through parameterized entanglement gate fusion, unified modeling and optimized expression of image, text and sound information are realized, which not only breaks through the technical limitations of traditional single modality or shallow fusion, but also utilizes the superposition and entanglement characteristics of quantum states to improve the semantic expression ability of knowledge nodes. Further, the superposition and entanglement characteristics of quantum states are utilized to improve the semantic expression ability of knowledge nodes; ultimately, the integrity and cognitive adaptability of knowledge representation are enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is a flowchart of the micro-course generation method based on knowledge graph.

[0019] Figure 2 Flowchart for generating cross-modal knowledge graph.

[0020] Figure 3 A flowchart for generating dynamic teaching strategies.

[0021] Figure 4 Flowchart for micro-course generation and traceability feedback. DETAILED DESCRIPTION

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0023] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0025] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a method for generating micro-courses based on a knowledge graph, comprising the following steps: S1: Collect textbook text data, teaching image data, and explanation audio data, and extract text semantic features, visual features, and audio spectrum features through deep neural networks.

[0026] It should be noted that the textbook text data is obtained by scanning the textbook documents, a high-definition camera is used to record the teaching video and intercept the teaching image data, and the explanation audio data is recorded through a microphone.

[0027] Furthermore, after collecting textbook text data, teaching image data and explanation audio data, the textbook text data is processed by the text encoder in the deep neural network to generate text semantic features, the teaching image data is processed by the convolutional neural network in the deep neural network to generate visual features, and the explanation audio data is processed by the time-frequency transformation network in the deep neural network to generate audio spectrum features.

[0028] Furthermore, the pre-training process of the deep neural network completes initial training on a large general dataset. The text encoder uses the Transformer architecture for self-supervised pre-training on corpora such as Wikipedia, and learns vocabulary and syntactic representations through the masked language modeling task; the convolutional neural network is pre-trained on the ImageNet dataset through image classification tasks, and the cross-entropy loss function is used to optimize the convolution kernel weights to extract general visual features; the time-frequency transformer network is pre-trained based on audio datasets such as LibriSpeech. After converting the audio into a spectrogram through short-time Fourier transform, the convolutional layer and the recurrent layer are used to jointly optimize the frequency domain feature extraction capability; the Adam optimizer is used for parameter update, the learning rate is set to 3e-5 and the cosine decay strategy is adopted, and the training cycle is no less than 50 epochs to ensure the convergence of the feature extractor; after pre-training is completed, the feature extraction layer weights are retained as initialization parameters, and the fully connected layer weights are randomly initialized to adapt to downstream tasks.

[0029] S2: Fusion of text semantic features, visual features, and audio spectrum features generates a quantum state fusion feature vector. After optimization through the feature alignment loss function, it constructs a multimodal knowledge node and calculates the quantum coherence correlation matrix to construct weighted relationship edges and generate a cross-modal knowledge graph.

[0030] S2.1: Map text semantic features, visual features, and audio spectral features to quantum states to generate text quantum states, image quantum states, and audio quantum states.

[0031] The specific process includes converting the semantic features of the text into complex vectors through a quantum encoder to generate a text quantum state, mapping the visual features to the Hilbert space through a quantum projection operation to generate an image quantum state, and processing the audio spectrum features using quantum Fourier transform to generate an audio quantum state; the quantum encoder is implemented using a parameterized quantum circuit, and after inputting the text semantic features, the text quantum state is output through the rotation gate and entanglement gate operations; the quantum projection operation performs the inner product of the visual features with the preset quantum basis state and then normalizes them to obtain the image quantum state; the quantum Fourier transform performs quantum amplitude encoding on the audio spectrum features to generate an audio quantum state.

[0032] The preset quantum ground state is a quantum state template pre-trained based on the typical visual patterns of the teaching image data. The preset quantum ground state is generated after clustering analysis of high-frequency visual features.

[0033] S2.2: Fusion of text quantum states, image quantum states, and audio quantum states through parameterized entanglement gates, and execution of quantum circuit measurements to generate quantum state fusion feature vectors.

[0034] The specific process includes: a parameterized entanglement gate receives text quantum states, image quantum states, and audio quantum states as inputs, and applies controlled rotation gate and controlled phase gate operations in the quantum circuit to entangle the multimodal quantum states; the entangled quantum states are superimposed through the combined operations of Pauli X-gate and Pauli Y-gate, and finally a quantum measurement operation is performed to project the superimposed quantum states onto the ground state; the ground state is converted into a real vector through quantum amplitude encoding, and then a quantum state fusion feature vector is generated through normalization.

[0035] S2.3: Use the feature alignment loss function to perform regularization optimization on the quantum state fusion feature vector, generate the optimized quantum state fusion feature vector, and construct a multimodal knowledge node.

[0036] The specific process includes: the feature alignment loss function receives the quantum state fusion feature vector as input, and obtains the fidelity difference between the quantum state fusion feature vector and the ideal multimodal quantum state; the complex components of the quantum state fusion feature vector are iteratively adjusted through the gradient descent algorithm to minimize the fidelity difference; the optimized quantum state fusion feature vector meets the orthogonal normalization condition; the optimized quantum state fusion feature vector is converted into a density matrix form; the density matrix is ​​subjected to feature decomposition to generate the quantum state basis vector of the multimodal knowledge node; the quantum state basis vector is bound to the semantic label to complete the multimodal knowledge node construction.

[0037] The ideal multimodal quantum state is a quantum state template pre-constructed based on the optimal cognitive model of the teaching knowledge points, and is generated by analyzing the multimodal feature association patterns that appear frequently in historical teaching data.

[0038] S2.4: Based on the quantum state fusion eigenvector, calculate the quantum coherence correlation matrix between multimodal knowledge nodes, which is expressed as: ; in, represents the quantum coherence correlation matrix, Represents the total number of multimodal knowledge nodes in the cross-modal knowledge graph.

[0039] in, Represents the first Multimodal knowledge nodes and The quantum coherence correlation strength between multimodal knowledge nodes is expressed as: ; It should be explained that and Represents a multimodal knowledge node in a cross-modal knowledge graph, represents the trace of the density matrix, Represents the first The density matrix of multimodal knowledge nodes, Represents the first The density matrix of multimodal knowledge nodes, represents the coherent interference term weight, Represents the first The entangled quantum state vector of multimodal knowledge nodes, Represents the first The density matrix of multimodal knowledge nodes.

[0040] The specific process includes taking the quantum state fusion eigenvector as input, and obtaining the quantum coherence correlation matrix by calculating the joint trace operation of the density matrix and the quantum state inner product; the density matrix is ​​generated by the quantum state fusion eigenvector, and the joint trace operation reflects the degree of quantum state overlap between multimodal knowledge nodes; the quantum state inner product captures the phase interference effect between nodes; each element of the quantum coherence correlation matrix represents the correlation strength of the corresponding multimodal knowledge node.

[0041] S2.5: Use dynamic standard deviation threshold to filter the quantum coherence correlation matrix to generate candidate relationship edge sets.

[0042] The specific process includes: taking the quantum coherence correlation matrix as input, obtaining the global mean and standard deviation of the elements of the quantum coherence correlation matrix; comparing each element in the quantum coherence correlation matrix with the dynamic standard deviation threshold, and retaining elements greater than the dynamic standard deviation threshold as strong correlation items; the multimodal knowledge node pairs corresponding to the strong correlation items constitute the candidate relationship edge set; each edge of the candidate relationship edge set carries the corresponding strength value in the quantum coherence correlation matrix.

[0043] The dynamic standard deviation threshold is obtained by comparing the linear combination of the mean and standard deviation of the elements of the quantum coherence correlation matrix.

[0044] S2.6: Based on the candidate relationship edge set, construct weighted relationship edges between multimodal knowledge nodes, and combine multimodal knowledge nodes to generate a cross-modal knowledge graph.

[0045] The specific process includes: according to the candidate relationship edge set, each candidate relationship edge contains two multimodal knowledge node identifiers and corresponding quantum coherence correlation strength values; the quantum coherence correlation strength values ​​are linearly mapped to relationship edge weights; weighted relationship edge records are added to the storage structure of multimodal knowledge nodes; the weighted relationship edges and multimodal knowledge nodes together constitute the topological structure of the cross-modal knowledge graph; the generated cross-modal knowledge graph supports path search and subgraph matching based on quantum coherence correlation strength.

[0046] S3: Input the cross-modal knowledge graph into the neural-symbolic hybrid engine, perform topological sorting on the multimodal knowledge nodes through the graph neural network, generate a sequence of knowledge points, and generate a dynamic teaching strategy by integrating the cognitive weights of the neural network with the logical constraints of the symbolic rules.

[0047] S3.1: Map the multimodal knowledge nodes in the cross-modal knowledge graph into quantum state encoding vectors and inject the symbolic rule projection operator to generate the quantum constraint matrix.

[0048] The specific process includes: the quantum state fusion feature vector contained in the multimodal knowledge node in the cross-modal knowledge graph is used as input and converted into a quantum state encoding vector through a quantum encoder; the quantum encoder is implemented using a parameterized quantum circuit, which includes rotation gate and entanglement gate operations; the symbolic rule projection operator is compiled and generated by predefined Prolog rules; the symbolic rule projection operator and the quantum state encoding vector are combined through a tensor product operation to generate a mixed state; the mixed state is converted into a real number matrix through Pauli operator measurement; the real number matrix is ​​orthogonally normalized to generate a quantum constraint matrix; each element of the quantum constraint matrix represents the compatibility strength of the corresponding multimodal knowledge node pair under the symbolic rule.

[0049] S3.2: Using the quantum constraint matrix as input, construct a topological sorting Hamiltonian and solve the probability distribution function of the generated knowledge point sequence using a quantum-classical hybrid optimizer. The expression is: ; in, represents a sequence of knowledge points, Represents a sequence of knowledge points The probability distribution function of represents the normalization constant, represents the natural exponential function, represents the temperature parameter ( >0), represents the time step index in the knowledge point sequence, Represents the time interval between adjacent nodes in the knowledge point sequence, Represents the time interval between adjacent nodes in the knowledge point sequence The cognitive rationality constraint function.

[0050] in, represents the topological sorting Hamiltonian, which is expressed as: ; It should be explained that Represents the first Multimodal knowledge nodes and The transfer probability weights between multimodal knowledge nodes, Represents the weight coefficient of cognitive alignment item (0.1≤ ≤0.5), Indicates the index of the multimodal knowledge node currently being calculated, represents the relative entropy difference between two probability distributions, Indicates the first The probability distribution of the position of multimodal knowledge nodes in the generated sequence, represents the cognitive optimal distribution.

[0051] The specific process includes: the quantum constraint matrix is ​​used as the input parameter, and the topological sorting Hamiltonian is constructed by linearly combining the quantum constraint matrix and the cognitive alignment term; the cognitive alignment term calculates the cognitive rationality difference of adjacent nodes in the knowledge point sequence; the quantum-classical hybrid optimizer uses the variational quantum feature solution algorithm to iteratively optimize the ground state energy of the Hamiltonian; the quantum state corresponding to the ground state energy is decoded by probability amplitude to generate the probability distribution function of the knowledge point sequence; the probability distribution function of the knowledge point sequence reflects the rationality weight of different teaching paths.

[0052] S3.3: Generate a knowledge point sequence based on the probability distribution function of the knowledge point sequence and sort it in descending order of probability value.

[0053] The specific process includes taking the probability distribution function of the knowledge point sequence as input, first extracting all possible knowledge point sequences and the probability values ​​corresponding to the knowledge point sequences; sorting the probability values ​​in descending order through a quick sorting algorithm; retaining the mapping relationship between the knowledge point sequence and the probability value during the sorting process; converting the probability value sequence after descending order into an ordered knowledge point sequence list; the first item of the generated knowledge point sequence list is the current optimal teaching path.

[0054] S3.4: Input the quantum state encoding vector into the cognitive weight neural network to generate the strategy superposition state vector, and compile the Prolog rules through the symbolic rule engine to generate the unitary operator matrix to obtain the symbolic constraint state vector.

[0055] The specific process includes taking the quantum state encoding vector as the input of the cognitive weight neural network, processing the quantum state encoding vector through multiple layers of fully connected layers to generate a strategy superposition state vector; the symbolic rule engine parses the predefined Prolog rules to generate logical constraints; the logical constraints are converted into a sequence of quantum gate operations and then synthesized into a unitary operator matrix; the unitary operator matrix acts on the strategy superposition state vector to perform quantum state evolution; the evolution process maintains the quantum coherence of the strategy superposition state vector; the evolution result is projected onto the computational basis state through measurement operations to generate a symbolic constraint state vector; the symbolic constraint state vector retains the cognitive weight characteristics of the original strategy superposition state vector while satisfying the symbolic rule constraints.

[0056] Furthermore, the pre-training method of the cognitive weight neural network first collects a historical teaching strategy dataset, which contains quantum state encoding vectors of multimodal knowledge nodes and corresponding teaching strategy labels; the quantum state encoding vectors are extracted from historical micro-course data through a quantum encoder; the teaching strategy labels are the optimal strategy sequences annotated by the actual teaching effect evaluation results; after the dataset is divided into a training set and a validation set, the cognitive weight neural network uses a cross-entropy loss function for supervised learning; the optimization process uses the Adam optimizer to update the network weights, the learning rate is set to 3e-5 and the cosine decay strategy is adopted; during the training process, the model parameters are saved when the accuracy of each validation set reaches more than 92%; the pre-trained cognitive weight neural network retains the weights of the fully connected layer for subsequent strategy superposition state vector generation.

[0057] The predefined Prolog rules are a symbolic reasoning system generated through the automated logical transformation of the teaching domain knowledge base and the mining of teaching big data.

[0058] S3.5: Input the symbolic constraint state vector into the quantum-neural interference Hamiltonian model and calculate the probability amplitude of the basic teaching strategy. The expression is: ; in, Presentation-based teaching strategies The probability amplitude of Presentation-based teaching strategies The quantum state encoding of represents the quantum-neural interference Hamiltonian, represents the time step of quantum evolution, represents the quantum strategy superposition state after symbol constraint, Presentation-based teaching strategies The cognitive bias measure, Represents the cognitive attenuation temperature coefficient (0.1≤ ≤0.2); in, It represents the adaptability weight function of the basic teaching strategy, and its expression is: ; It should be explained that represents the normalization factor of the complexity adjustment parameter of the basic teaching strategy, represents the complexity adjustment parameter of the basic teaching strategy (2.0≤ ≤3.0), Presentation-based teaching strategies Cognitive resistance coefficient (1.0≤ ≤1.5), Represents a time variable.

[0059] The specific process includes: taking the symbolic constraint state vector as input, the quantum-neural interference Hamiltonian model combines the symbolic constraint state vector with the neural network weight matrix through tensor product operation; the combined mixed state is converted into a real probability amplitude through quantum measurement operation; the real probability amplitude reflects the rationality of the basic teaching strategy under the symbolic constraint; the interference term in the quantum-neural interference Hamiltonian model captures the nonlinear coupling effect between cognitive weights and symbolic rules; the basic teaching strategy probability amplitude of the calculated result meets the input requirements of subsequent dynamic threshold decision-making.

[0060] Furthermore, the pre-training process of the quantum-neural interference Hamiltonian model first collects a historical teaching strategy dataset, which includes symbolic constraint state vectors and corresponding optimal teaching strategy labels; the symbolic constraint state vectors are generated from historical micro-class data through a symbolic rule engine; the optimal teaching strategy labels are annotated by education experts; after the dataset is divided into a training set and a validation set, the quantum-neural interference Hamiltonian model is optimized using a negative log-likelihood loss function; the training process uses a quantum approximate optimization algorithm to iteratively adjust the Hamiltonian parameters; after each round of training, the quantum-neural interference Hamiltonian model parameters are saved when the validation set accuracy reaches more than 89%; the pre-trained quantum-neural interference Hamiltonian model retains the interference term coefficients.

[0061] S3.6: Generate dynamic teaching strategies based on the probability amplitude of the basic teaching strategies through dynamic threshold decision-making and real-time cognitive safety verification.

[0062] The specific process includes: taking the probability amplitude of the basic teaching strategy as input, the dynamic threshold decision first analyzes the probability amplitude distribution characteristics and calculates the dynamic decision threshold of the current cognitive scenario; real-time cognitive safety verification verifies the cognitive matching degree of the candidate strategy based on the user's real-time brain wave data and eye movement trajectory; the verified candidate strategies are sorted according to the probability amplitude weight to generate a dynamic teaching strategy sequence; each strategy node in the dynamic teaching strategy sequence carries a quantum state encoding vector and a cognitive weight label; the generated dynamic teaching strategy sequence meets the topological constraints of the cross-modal knowledge graph and the real-time requirements of micro-course synthesis.

[0063] The real-time requirement means that the delay in the generation and execution of dynamic teaching strategies must be controlled at the millisecond level, which is determined by analyzing the statistical relationship between user cognitive feedback data and the response time of the teaching scenario.

[0064] S4: Based on the knowledge point sequence and dynamic teaching strategy, call the content generator to synthesize multimodal micro-courses, transform the decision logic of the dynamic teaching strategy into a visual traceability identifier and embed it into the multimodal micro-course to generate a traceable micro-course.

[0065] S4.1: Input the knowledge point sequence and dynamic teaching strategy into the multimodal content generator to generate the original micro-course media stream, and inject it into the quantum cognitive field to generate a field-containing micro-course.

[0066] The specific process includes inputting the knowledge point sequence and dynamic teaching strategy into the multimodal content generator, parsing the multimodal knowledge nodes in the knowledge point sequence; the dynamic teaching strategy guides the text generator to synthesize the explanation script, the image renderer to generate the teaching chart, and the speech synthesizer to output the dubbing audio; the text generator uses the pre-trained language model to convert the knowledge points into natural language; the image renderer generates visual charts based on the quantum state coding vector; the speech synthesizer converts the text and emotional tags into spectral features; the three types of media streams are aligned after the timestamps are aligned and mixed into the original micro-course media stream; the quantum cognitive field generator analyzes the quantum state coding vectors in the knowledge point sequence and converts them into a space-time field strength distribution; the field strength distribution is mapped to the timestamp and spatial coordinates of the original micro-course media stream and superimposed with the quantum field information; the generated field-containing micro-course stimulates the interference effect of the quantum cognitive field when it is played.

[0067] S4.2: Construct a quantum decision traceability matrix based on dynamic teaching strategies and the space-time coordinates of field-containing micro-courses.

[0068] The specific process includes taking the dynamic teaching strategy and the space-time coordinates of the field-containing micro-course as input, and the quantum decision traceability matrix generator first extracts the quantum state labels in the dynamic teaching strategy; the space-time coordinates of the field-containing micro-course are tensor-producted with the quantum cognitive field strength distribution; the operation result is converted into a density matrix form to record the quantum state evolution trajectory of each decision point; the density matrix generates eigenvalues ​​and eigenvectors through orthogonal decomposition; the eigenvectors establish a mapping relationship with the multimodal knowledge nodes of the knowledge point sequence to generate a quantum decision traceability matrix.

[0069] S4.3: Input the quantum decision traceability matrix into the bionic retinal encoder, output the visual traceability fluctuation signal, and convert it into a quantum traceability identifier through the triple integral operator.

[0070] The specific process includes taking the quantum decision tracing matrix as input, and the bionic retinal encoder extracting the spatiotemporal pattern features in the quantum decision tracing matrix through the convolution kernel; the spatiotemporal pattern features are filtered by Gaussian difference to generate primary visual tracing fluctuation signals; the primary visual tracing fluctuation signals are fused along the time, space and frequency domain dimensions through the triple integral operator; quantum amplitude modulation is used to maintain the phase coherence between signals during the fusion process; the modulated signal is normalized to generate a quantum traceability mark; the frequency domain features of the quantum traceability mark resonate and match with the quantum cognitive field of the field-containing micro-course; the generated quantum traceability mark meets the spatiotemporal coordinate accuracy requirements of the dynamically embedded micro-course interface.

[0071] S4.4: Dynamically embed the quantum traceability identifier into the corresponding coordinates of the original micro-course media stream to generate a traceable micro-course.

[0072] The specific process includes taking the quantum traceability identifier as input, and matching the timestamp and spatial coordinates of the original micro-course media stream with the spatiotemporal characteristics of the quantum traceability identifier; using a bilinear interpolation algorithm to align the quantum traceability identifier and the pixel position of the media stream during the matching process; superimposing the aligned quantum traceability identifier onto the corresponding frame of the original micro-course media stream through an Alpha blending algorithm; the superposition process retains the frequency domain fluctuation characteristics of the quantum traceability identifier; the generated micro-course media stream contains both audio-visual content and the interference pattern of the quantum traceability identifier; the traceable micro-course stimulates the resonant response of the visual traceability signal through the quantum cognitive field during playback; the traceable micro-course meets the spatiotemporal synchronization accuracy requirements of real-time interactive monitoring.

[0073] S5: Monitor the user's interaction behavior with the multimodal micro-course in real time and calculate the independent synchronization deviation value. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge node is updated through the feature alignment loss function.

[0074] S5.1: Collect eye movement trajectories, brain waves, and click event streams of users interacting with traceable micro-courses in real time, and calculate the independent synchronization deviation value, which is expressed as: ; in, Indicates time Independent synchronization deviation value (0≤ ≤2), Indicates the modal type index variable, A set of types representing multimodal teaching content, Indicates the duration of the sliding integral window, represents the time-integrated variable, Indicates that the user is at time Modal Real-time interactive signals, Indicates modality At the moment The ideal response signal, Indicates time User-to-modal Real-time cognitive matching, Indicates modality Sensitivity coefficient (0< ≤1).

[0075] The specific process includes: when users interact with traceable micro-courses, eye movement trajectories are converted into gaze point coordinate sequences, brain waves are decomposed into power spectrum time series, and click event streams are recorded as timestamp instruction sequences; the gaze point coordinate sequence and text position coordinates are used to calculate the text modal synchronization deviation value, the power spectrum time series and audio spectrum features are used to calculate the audio modal synchronization deviation value, and the timestamp instruction sequence and image switching time series are used to calculate the image modal synchronization deviation value; the independent synchronization deviation values ​​of the text, text, and sound multi-modalities are averaged through sliding time window integration.

[0076] S5.2: When any independent synchronization deviation value exceeds the cognitive threshold, the feature alignment loss function is triggered to update the feature vector of the corresponding multimodal knowledge node in the cross-modal knowledge graph.

[0077] The specific process includes: the independent synchronization deviation value of the multimodal is taken as input and first compared with the cognitive threshold; when any independent synchronization deviation value exceeds the cognitive threshold, the feature alignment loss function is activated; the feature alignment loss function reads the quantum state fusion feature vector of the modality that exceeds the cognitive threshold; the quantum state fusion feature vector adjusts the complex component through the gradient descent algorithm; the optimization process minimizes the fidelity difference between the quantum state fusion feature vector and the ideal multimodal quantum state; the updated quantum state fusion feature vector is re-input into the construction process of the corresponding multimodal knowledge node in the cross-modal knowledge graph; after the feature vector of the multimodal knowledge node is updated, the quantum coherence correlation matrix in the cross-modal knowledge graph is synchronously recalculated.

[0078] The cognitive threshold is preset according to the statistical laws of user cognitive status and interactive behavior in historical teaching data, and the interval is determined by cluster analysis of high-frequency cognitive matching patterns.

[0079] This embodiment also provides a micro-course generation system based on knowledge graph, including: feature extraction module, graph construction module, strategy generation module, micro-course synthesis module and feedback optimization module, feature extraction module, for collecting textbook text data, teaching image data and explanation audio data, and extracting text semantic features, visual features and audio spectrum features through deep neural network; graph construction module, for fusing text semantic features, visual features and audio spectrum features to generate quantum state fusion feature vector, after feature alignment loss function optimization, construct multimodal knowledge nodes, and calculate quantum coherence correlation matrix to construct weighted relationship edges to generate cross-modal knowledge graph; strategy generation module, for fusing cross-modal The dynamic knowledge graph is input into the neural-symbolic hybrid engine, and the multimodal knowledge nodes are topologically sorted through the graph neural network to generate a knowledge point sequence. The dynamic teaching strategy is generated by fusing the cognitive weight of the neural network with the logical constraints of the symbolic rules; the micro-course synthesis module is used to call the content generator to synthesize multimodal micro-courses according to the knowledge point sequence and dynamic teaching strategy, and convert the decision logic of the dynamic teaching strategy into a visual traceability identifier embedded in the multimodal micro-course to generate a traceable micro-course; the feedback optimization module is used to monitor the interaction behavior between users and multimodal micro-courses in real time and calculate the independent synchronization deviation value. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge nodes are updated through the feature alignment loss function.

[0080] This embodiment also provides a computer device, which is suitable for the micro-course generation method based on the knowledge graph, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the micro-course generation method based on the knowledge graph proposed in the above embodiment.

[0081] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0082] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for generating micro-courses based on knowledge graphs as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0083] In summary, the present invention achieves unified modeling and optimized expression of image, text and sound information by: generating quantum state fusion feature vectors through parameterized entanglement gate fusion, which not only breaks through the technical limitations of traditional single modality or shallow fusion, but also utilizes the superposition and entanglement characteristics of quantum states to enhance the semantic expression ability of knowledge nodes. Furthermore, the superposition and entanglement characteristics of quantum states are utilized to enhance the semantic expression ability of knowledge nodes; ultimately, the integrity and cognitive adaptability of knowledge representation are enhanced.

[0084] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A micro-course generation method based on knowledge graph, characterized by: include, Collect textbook text data, teaching image data, and lecture audio data, and extract text semantic features, visual features, and audio spectrum features through deep neural networks; The quantum state fusion feature vector is generated by fusing text semantic features, visual features, and audio spectrum features. After optimization through the feature alignment loss function, multimodal knowledge nodes are constructed. The quantum coherence correlation matrix is ​​calculated to construct weighted relationship edges and generate a cross-modal knowledge graph. The cross-modal knowledge graph is input into the neural-symbolic hybrid engine, and the multimodal knowledge nodes are topologically sorted through the graph neural network to generate a sequence of knowledge points. The dynamic teaching strategy is generated by integrating the cognitive weight of the neural network with the logical constraints of the symbolic rules. Based on the knowledge point sequence and dynamic teaching strategy, the content generator is called to synthesize multimodal micro-courses. The decision logic of the dynamic teaching strategy is converted into a visual traceability mark and embedded into the multimodal micro-course to generate a traceable micro-course. The user's interaction behavior with multimodal micro-courses is monitored in real time and the independent synchronization deviation value is calculated. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge node is updated through the feature alignment loss function.

2. The method for generating a micro-course based on a knowledge graph according to claim 1, wherein: The specific steps of constructing a multimodal knowledge node are as follows: Map text semantic features, visual features, and audio spectrum features to quantum states to generate text quantum states, image quantum states, and audio quantum states; The text quantum state, image quantum state and audio quantum state are fused through parameterized entanglement gates, and quantum circuit measurements are performed to generate quantum state fusion feature vectors. The feature alignment loss function is used to perform regularization optimization on the quantum state fusion feature vector, generate the optimized quantum state fusion feature vector, and construct a multimodal knowledge node.

3. The method for generating a micro-course based on a knowledge graph according to claim 2, wherein: The specific steps of generating a cross-modal knowledge graph are as follows: Based on the quantum state fusion eigenvector, the quantum coherence correlation matrix between multimodal knowledge nodes is calculated; The dynamic standard deviation threshold is used to filter the quantum coherence correlation matrix to generate candidate relationship edge sets; Based on the candidate relationship edge set, weighted relationship edges are constructed between multimodal knowledge nodes, and multimodal knowledge nodes are combined to generate a cross-modal knowledge graph.

4. The method for generating a micro-course based on a knowledge graph according to claim 3, wherein: The specific steps of generating a knowledge point sequence are as follows: Map the multimodal knowledge nodes in the cross-modal knowledge graph into quantum state encoding vectors and inject the symbolic rule projection operator to generate the quantum constraint matrix; Using the quantum constraint matrix as input parameter, a topological sorting Hamiltonian is constructed, and the probability distribution function of the knowledge point sequence is generated by solving it through a quantum-classical hybrid optimizer; According to the probability distribution function of the knowledge point sequence, the knowledge point sequence is generated by sorting in descending order of probability values.

5. The method for generating a micro-course based on a knowledge graph according to claim 4, wherein: The specific steps of generating dynamic teaching strategies are as follows: The quantum state encoding vector is input into the cognitive weight neural network to generate the strategy superposition state vector, and the Prolog rule is compiled through the symbolic rule engine to generate the unitary operator matrix to obtain the symbolic constraint state vector; Input the symbolic constraint state vector into the quantum-neural interference Hamiltonian model to calculate the probability amplitude of the basic teaching strategy; According to the probability amplitude of the basic teaching strategy, a dynamic teaching strategy is generated through dynamic threshold decision-making and real-time cognitive safety verification.

6. The method for generating a micro-course based on a knowledge graph according to claim 5, wherein: According to the knowledge point sequence and dynamic teaching strategy, the content generator is called to synthesize multimodal micro-courses, the decision logic of the dynamic teaching strategy is converted into a visual traceability mark embedded in the multimodal micro-course, and a traceable micro-course is generated. The specific steps are as follows: Input the knowledge point sequence and dynamic teaching strategy into the multimodal content generator to generate the original micro-course media stream, and inject the quantum cognitive field to generate a micro-course containing the field; Based on dynamic teaching strategies and the space-time coordinates of field-containing micro-lessons, a quantum decision traceability matrix is ​​constructed; The quantum decision traceability matrix is ​​input into the bionic retinal encoder, which outputs the visual traceability fluctuation signal and is converted into a quantum traceability mark through the triple integral operator; The quantum traceability identifier is dynamically embedded into the corresponding coordinates of the original micro-course media stream to generate a traceable micro-course.

7. The method for generating a micro-course based on a knowledge graph according to claim 6, wherein: The real-time monitoring of the user's interaction behavior with the multimodal micro-course and the calculation of the independent synchronization deviation value are performed. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge node is updated through the feature alignment loss function. The specific steps are as follows: Collect eye movement trajectories, brain waves, and click event streams of users interacting with traceable micro-courses in real time, and calculate independent synchronization deviation values; When any independent synchronization deviation value exceeds the cognitive threshold, the feature alignment loss function is triggered to update the feature vector of the corresponding multimodal knowledge node in the cross-modal knowledge graph.

8. A knowledge graph-based micro-course generation system, based on the knowledge graph-based micro-course generation method according to any one of claims 1 to 7, characterized in that: Including feature extraction module, graph construction module, strategy generation module, micro-course synthesis module and feedback optimization module, The feature extraction module is used to collect textbook text data, teaching image data, and lecture audio data, and extract text semantic features, visual features, and audio spectrum features through deep neural networks; The graph construction module is used to fuse text semantic features, visual features, and audio spectrum features to generate quantum state fusion feature vectors. After optimization using the feature alignment loss function, multimodal knowledge nodes are constructed, and the quantum coherence correlation matrix is ​​calculated to construct weighted relationship edges to generate a cross-modal knowledge graph. The strategy generation module is used to input the cross-modal knowledge graph into the neural-symbolic hybrid engine, topologically sort the multimodal knowledge nodes through the graph neural network, generate a sequence of knowledge points, and generate dynamic teaching strategies by fusing the cognitive weights of the neural network with the logical constraints of the symbolic rules; The micro-course synthesis module is used to call the content generator to synthesize multimodal micro-courses based on the knowledge point sequence and dynamic teaching strategy, and convert the decision logic of the dynamic teaching strategy into a visual traceability identifier embedded in the multimodal micro-course to generate a traceable micro-course; The feedback optimization module is used to monitor the user's interaction behavior with the multimodal micro-course in real time and calculate the independent synchronization deviation value. When the independent synchronization deviation value exceeds the cognitive threshold, the multimodal knowledge node is updated through the feature alignment loss function.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the micro-course generation method based on the knowledge graph are implemented in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the micro-course generation method based on the knowledge graph according to any one of claims 1 to 7 are implemented.

Images