A cloud-based education management system and method based on big data

Abstract

The present invention discloses a cloud-based education management system and method based on big data, which relates to the field of cloud-based education technology, including: using a dynamic time warping algorithm to fuse MFCC feature vectors and comprehensive eye movement feature vectors according to timestamps to generate a comprehensive feature vector; constructing a dual-stream Transformer model based on a layered processing mechanism and a cross-modal attention mechanism, and predicting the student's cognitive load index based on the comprehensive feature vector; constructing a virtual teaching environment, optimizing the three strategy parameters of explanation speed, interaction frequency and content difficulty in the virtual teaching environment based on an adaptive strategy optimization algorithm combined with cognitive load data, and dynamically generating a teaching plan. The present invention achieves accurate prediction of cognitive load by constructing a dual-stream Transformer model, combining a temporal embedding layer and a spatial embedding layer, and dynamically fusing temporal features and spatial features through a cross-modal attention mechanism, thereby providing a basis for the dynamic optimization of teaching strategies.

Description

Technical Field

[0001] The present invention relates to the field of cloud-based education technology, and in particular to a cloud-based education management system and method based on big data. Background Art

[0002] In recent years, cloud-based education management systems, as a new educational model, have enabled centralized management of educational resources and the dynamic generation of personalized teaching plans by integrating cloud computing, the Internet of Things, and data analysis technologies. Against this backdrop, big data-based education management technologies have become a research hotspot, with significant progress particularly in the collection and analysis of student behavior data. Furthermore, the application of reinforcement learning algorithms in virtual teaching environments enables dynamic adjustments to teaching plans based on real-time student feedback, further enhancing teaching effectiveness.

[0003] However, existing cloud-based education management systems still have some shortcomings. First, existing technologies lack the ability to integrate multimodal data when collecting and processing student behavior data. For example, eye movement data and speech data are often analyzed independently, making it difficult to fully reflect a student's learning status. Second, existing cognitive load prediction models mostly rely on a single feature and fail to fully utilize the complementary nature of multimodal data, resulting in limited prediction accuracy. Summary of the Invention

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

[0005] Therefore, the present invention provides a cloud-based education management method based on big data to solve the problems of insufficient utilization of multimodal data and limited accuracy of cognitive load prediction in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a cloud-based education management method based on big data, which includes collecting students' eye movement data and voice data and preprocessing them, and extracting MFCC features and comprehensive eye movement features at the same time; using a dynamic time warping algorithm to fuse MFCC feature vectors and comprehensive eye movement feature vectors according to timestamps to generate a comprehensive feature vector; based on a hierarchical processing mechanism and a cross-modal attention mechanism, a dual-stream Transformer model is constructed, and based on the comprehensive feature vector, the student's cognitive load index is predicted; a virtual teaching environment is constructed, and based on an adaptive strategy optimization algorithm combined with cognitive load data, three strategy parameters of explanation speed, interaction frequency and content difficulty are optimized in the virtual teaching environment, and a teaching plan is dynamically generated.

[0008] As a preferred solution of the cloud-based education management method based on big data of the present invention, the extraction of MFCC features and comprehensive eye movement features includes the following steps:

[0009] The speech data is divided into frames, each frame signal is subjected to fast Fourier transform to generate a spectrogram, and the MFCC feature vector is extracted through discrete cosine transform;

[0010] The gaze heat map is generated by performing Gaussian kernel density estimation on the gaze point coordinates, the distribution statistics of the gaze duration are identified, the pupil diameter change rate and fluctuation frequency are extracted, and the path complexity is calculated by segmenting the scanning path to generate a comprehensive eye movement feature vector.

[0011] As a preferred solution of the cloud-based education management method based on big data described in the present invention, wherein: the dynamic time warping algorithm is used to fuse the MFCC feature vector and the comprehensive eye movement feature vector according to the timestamp to generate a comprehensive feature vector. The specific steps are as follows:

[0012] Calculate the Euclidean distance between the MFCC feature vector and the eye movement feature vector frame by frame;

[0013] Based on the Euclidean distance between the MFCC feature vector and the eye movement feature vector, the cost matrix is ​​constructed using the global sequence alignment method;

[0014] Based on the cost matrix, a dynamic time planning algorithm is used to find the optimal alignment path according to timestamps. Then, the MFCC features and eye movement features are fused according to the timestamps according to the optimal alignment path to generate a comprehensive feature vector.

[0015] As a preferred solution of the cloud-based education management method based on big data described in the present invention, wherein: based on the hierarchical processing mechanism and the cross-modal attention mechanism, a dual-stream Transformer model is constructed, and the student's cognitive load index is predicted based on the comprehensive feature vector. The specific steps are as follows:

[0016] The temporal embedding layer performs position encoding on the time series of the comprehensive feature vector through Transformer position encoding to generate temporal embedding features;

[0017] The spatial embedding layer performs linear transformation on the comprehensive feature vectors through MLP to generate spatial embedding features;

[0018] The temporal embedding and spatial embedding are spliced ​​together through the cross-modal attention mechanism, and the weights of temporal features and spatial features are adjusted in combination with the Sigmoid gate control mechanism.

[0019] The temporal embedding layer, spatial embedding layer and cross-modal attention mechanism are combined to construct a two-stream Transformer model, and the comprehensive feature vector is input to predict the students' cognitive load index.

[0020] As a preferred solution of the cloud-based education management method based on big data described in the present invention, a virtual teaching environment is constructed based on historical student behavior data and student cognitive load data through the Unity3D engine and combined with the TensorFlow reinforcement learning framework.

[0021] As a preferred solution of the cloud-based education management method based on big data described in the present invention, the construction process of the adaptive strategy optimization algorithm is as follows:

[0022] Define the state space, action space, and reward function;

[0023] Construct a strategy network based on the state space, action space, and students’ cognitive load data;

[0024] Construct a value network based on the state space and reward function;

[0025] Based on the policy network and value network, an adaptive policy optimization algorithm is constructed.

[0026] As a preferred solution of the cloud-based education management method based on big data of the present invention, the specific steps of dynamically generating a teaching plan are as follows:

[0027] Combine the adaptive strategy optimization algorithm with the learner's cognitive load data to optimize strategy parameters;

[0028] According to the optimized strategy parameters, the teaching plan is dynamically generated through RLDP.

[0029] In the second aspect, the present invention provides a cloud-based education management system based on big data, including a feature extraction module, a feature fusion module, a cognitive load prediction module and a teaching plan generation module; the feature extraction module is used to collect students' eye movement data and voice data and pre-process them, and at the same time extract MFCC features and comprehensive eye movement features; the feature fusion module is used to use a dynamic time warping algorithm to fuse MFCC feature vectors and comprehensive eye movement feature vectors according to timestamps to generate a comprehensive feature vector; the cognitive load prediction module is used to construct a dual-stream Transformer model based on a hierarchical processing mechanism and a cross-modal attention mechanism, and predict the student's cognitive load index based on the comprehensive feature vector; the teaching plan generation module is used to construct a virtual teaching environment, based on an adaptive strategy optimization algorithm combined with cognitive load data, optimize the three strategy parameters of explanation speed, interaction frequency and content difficulty in the virtual teaching environment, and dynamically generate a teaching plan.

[0030] 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, any step of the cloud-based education management method based on big data as described in the first aspect of the present invention is implemented.

[0031] 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 cloud-based education management method based on big data as described in the first aspect of the present invention.

[0032] The beneficial effects of the present invention are as follows: the MFCC feature vector and the comprehensive eye movement feature vector are fused according to the timestamp through the dynamic time warping algorithm to generate a comprehensive feature vector, thereby realizing efficient fusion of multimodal data; by constructing a dual-stream Transformer model, combining the temporal embedding layer and the spatial embedding layer, and dynamically fusing the temporal features and spatial features through the cross-modal attention mechanism, accurate prediction of cognitive load is realized, providing a basis for the dynamic optimization of teaching strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] 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.

[0034] Figure 1 This is a flow chart of the cloud-based education management method based on big data in Example 1.

[0035] Figure 2 Schematic diagram of the cloud-based education management system based on big data in Example 1.

[0036] Figure 3 This is a flow chart of data collection and preprocessing in the cloud-based education management method based on big data in Example 1.

[0037] Figure 4 This is a flowchart for predicting the student's cognitive load index in the cloud-based education management method based on big data in Example 1. DETAILED DESCRIPTION

[0038] 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.

[0039] 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.

[0040] 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.

[0041] Example 1, with reference to Figures 1 to 4 , which is the first embodiment of the present invention, provides a cloud-based education management method based on big data, comprising the following steps:

[0042] S1. Collect students' eye movement data and speech data and preprocess them, and extract MFCC features and comprehensive eye movement features.

[0043] Eye movement data includes gaze point coordinates, gaze duration, and pupil diameter changes, and speech data includes fundamental frequency, speech rate, silence intervals, and formant.

[0044] It should be noted that eye movement data is obtained by capturing eye movement trajectories with an infrared camera and calculating changes in gaze point and pupil diameter, and voice data is obtained through a microphone array.

[0045] Preprocessing includes data cleaning, denoising, segmenting fixations, temporal alignment, and normalization, as follows:

[0046] First, we start with data cleaning, ensuring the completeness and accuracy of the dataset by removing outliers in the eye movement and speech data and filling in missing parts; then we perform a denoising step, using filters to smooth the eye movement data, reduce high-frequency noise, and use audio processing tools to remove background noise in the speech data to improve clarity; then, we perform gaze point segmentation on the eye movement data, using algorithms to distinguish different gaze behaviors for subsequent analysis; time alignment ensures that the time axis of different samples, whether eye movement or speech data, remains consistent, laying the foundation for comparative analysis; the last step is standardization, which ensures the consistency of data from different individuals or collection conditions by normalizing the data.

[0047] The speech data is divided into frames, each frame signal is subjected to fast Fourier transform to generate a spectrogram, and the MFCC feature vector is extracted through discrete cosine transform;

[0048] Furthermore, the speech signal is first divided into several frames of fixed duration (e.g., 25 milliseconds), and each frame is windowed (e.g., using a Hamming window) to reduce edge artifacts. Each frame is then subjected to a Fast Fourier Transform (FFT) to convert the time-domain signal into the frequency domain, generating a spectrogram. The spectrogram is then logarithmized and the Mel filter bank energy is calculated to obtain a Mel-spectrum. Finally, the Mel-spectrum is subjected to a Discrete Cosine Transform (DCT) to extract the MFCC feature vector. For example, a 1-second speech signal is divided into 40 frames, each of which is subjected to a FFT to generate a spectrogram, which is then subjected to a DCT to obtain a 13-dimensional MFCC feature vector.

[0049] A gaze heat map is generated by performing Gaussian kernel density estimation on the gaze point coordinates, the distribution statistics of the gaze duration (mean, variance, peak) are identified, the pupil diameter change rate and fluctuation frequency are extracted, and the path complexity (curvature change) is calculated by segmenting the scanning path to generate a comprehensive eye movement feature vector.

[0050] Furthermore, the gaze point coordinates are first smoothed with a Gaussian kernel to generate a gaze heat map to reflect the concentration of the gaze area. Next, the distribution statistics of the gaze duration, including the mean, variance, and peak value, are calculated to describe the concentration and discreteness of the gaze duration. The pupil diameter change rate and fluctuation frequency are then extracted to reflect the contraction and dilation dynamics of the pupil. Finally, the saccade path is segmented, and the curvature change of each segment is calculated to obtain the path complexity, which is used to describe the smoothness of the eye movement trajectory. For example, in a 5-second segment of eye movement data, the gaze point coordinates are estimated using a Gaussian kernel density to generate a gaze heat map. The mean of the gaze duration is 200 milliseconds, the variance is 50 milliseconds, the peak is 300 milliseconds, the pupil diameter change rate is 0.5, the fluctuation frequency is 2 Hz, and the curvature change of the saccade path is 0.2. Finally, a comprehensive eye movement feature vector is generated.

[0051] S2. Use the dynamic time warping algorithm to fuse the MFCC feature vector and the comprehensive eye movement feature vector according to the timestamp to generate a comprehensive feature vector.

[0052] The Euclidean distance between the MFCC feature vector and the eye movement feature vector is calculated frame by frame, and the expression is:

[0053] ;

[0054] in, Indicates the Frame MFCC feature vector and the The Euclidean distance between the frame-integrated eye movement feature vectors, Indicates the The first MFCC feature vector of the frame Dimension values, Indicates the The first frame of the comprehensive eye movement feature vector Dimension values, Represents the dimensions of MFCC feature vectors and eye movement feature vectors;

[0055] It should be noted that for the MFCC feature vector and comprehensive eye movement feature vector of each frame, their dimension values ​​are extracted respectively, the difference between the corresponding dimension values ​​is calculated and squared, the square differences of all dimensions are accumulated and then squared to obtain the Euclidean distance between the feature vectors of the two frames.

[0056] Based on the Euclidean distance between MFCC feature vector and eye movement feature vector , the cost matrix is ​​constructed using the global sequence alignment method;

[0057] Furthermore, an initial cost matrix is ​​first initialized. The number of rows in the initial cost matrix is ​​equal to the number of frames of the MFCC feature vectors, and the number of columns is equal to the number of frames of the comprehensive eye movement feature vectors. Next, the first row and first column of the matrix are initialized to the accumulated Euclidean distances. For example, the first row accumulates the Euclidean distances of each column from the first to the last column, and the first column accumulates the Euclidean distances of each row from the first to the last row. Then, starting from the second row and second column, the cost of the current position is calculated element by element, using the formula of the current Euclidean distance plus the minimum value of the three adjacent positions to the left, top, and upper left. Finally, the complete matrix is ​​filled to obtain the cost matrix. For example, if there are three frames of MFCC feature vectors and four frames of eye movement feature vectors, a matrix with three rows and four columns is initialized. The first row is filled with the accumulated Euclidean distances from the first to the fourth columns, the first column is filled with the accumulated Euclidean distances from the first to the third rows, and the second row and second column is filled with the current Euclidean distance plus the minimum value of the three adjacent positions to the left, top, and upper left. The cost matrix is ​​generated in this way.

[0058] Based on the cost matrix, a dynamic time planning algorithm is used to find the optimal alignment path according to timestamps. The MFCC feature vector and the comprehensive eye movement feature vector are fused according to the timestamps according to the optimal alignment path to generate a comprehensive feature vector.

[0059] Next, starting from the lower right corner of the cost matrix and gradually backtracking toward the upper left corner, the path with the lowest cost among the three adjacent positions of left, top, and upper left is selected as the optimal alignment path. The MFCC feature vector and the neutralized eye movement feature vector are then aligned by timestamp along the optimal alignment path. For example, the MFCC features of the i-th frame are aligned with the eye movement features of the j-th frame. Finally, the aligned MFCC features and eye movement features are concatenated by timestamp to generate a comprehensive feature vector. For example, if the optimal alignment paths of the cost matrix are (1, 1), (2, 2), and (3, 3), the MFCC features of the first frame are aligned with the eye movement features of the first frame, the MFCC features of the second frame are aligned with the eye movement features of the second frame, and the MFCC features of the third frame are aligned with the eye movement features of the third frame. After concatenation, the comprehensive feature vector is generated.

[0060] S3. Based on the hierarchical processing mechanism and cross-modal attention mechanism, a two-stream Transformer model is constructed, and the students' cognitive load index is predicted based on the comprehensive feature vector;

[0061] The hierarchical processing mechanism includes a temporal embedding layer and a spatial embedding layer;

[0062] The time series embedding layer performs position encoding on the time series of the comprehensive feature vector through Transformer position encoding to generate time series embedding features, which are expressed as:

[0063] ;

[0064] in, is in the time step The temporal embedding features of is the comprehensive eigenvector, is the comprehensive feature vector Position encoding of the time series;

[0065] Furthermore, a position code is first generated for each time step of the comprehensive feature vector. The position code is composed of a combination of sine and cosine functions and is used to represent the sequential information of the time step. Then, the comprehensive feature vector is added to the position code of the corresponding time step to generate a temporal embedding feature. For example, the comprehensive feature vector of the first frame is added to the position code of the first frame, the comprehensive feature vector of the second frame is added to the position code of the second frame, and so on. Finally, a temporal embedding feature containing time sequence information is generated.

[0066] The spatial embedding layer performs linear transformation on the comprehensive feature vectors through MLP to generate spatial embedding features, which are expressed as:

[0067] ;

[0068] in, is in the time step The spatial embedding features of is the weight matrix of the MLP, is the bias term of MLP;

[0069] Furthermore, the comprehensive feature vector is first multiplied by the weight matrix of the MLP, and then the bias term is added to obtain the linear transformation result; then the ReLU activation function is applied to the linear transformation result to generate the spatial embedding feature; for example, the time step The comprehensive feature vector of is linearly transformed by the weight matrix and the bias term to obtain the intermediate result, which is then processed by the ReLU activation function to finally generate the spatial embedding feature.

[0070] The temporal embedding and spatial embedding are spliced ​​together through the cross-modal attention mechanism, and the weights of temporal features and spatial features are adjusted in combination with the Sigmoid gate control mechanism;

[0071] It should be noted that first, the temporal embedding features and the spatial embedding features are spliced ​​by dimension to form fusion features; then the fusion features are input into the Sigmoid gate control mechanism to calculate the weights of the temporal features and the spatial features; then the temporal embedding features are multiplied by the temporal feature weights, and the spatial embedding features are multiplied by the spatial feature weights, and the results are added together to generate weighted fusion features.

[0072] Combine the temporal embedding layer, spatial embedding layer and cross-modal attention mechanism to build a two-stream Transformer model;

[0073] The comprehensive feature vector is input into the two-stream Transformer model to predict the student's cognitive load index, which is expressed as:

[0074] ;

[0075] in, It is a weight matrix that converts the features output by the cross-modal attention mechanism into the student’s cognitive load index. is the total number of time steps, is the index of the time step, is the Sigmoid function, is the gating weight matrix, is the element-wise multiplication operator, is the query matrix, is the bond matrix, is the value matrix, is the dimension of the feature, is in the time step The cognitive load index of students, yes function, It is the bias term in the prediction process of students’ cognitive load index.

[0076] Furthermore, we first embed the time series into features and spatial embedding features Add together and get ; Then through the gate weight matrix and Sigmoid function Calculate the gating weights and then embed the time series into features With the query matrix Multiplication, spatial embedding features With the key matrix Multiply, calculate the attention score between the two, and pass the Softmax function Normalize to get the attention weight; then embed the space into features and value matrix Multiply them together, and then multiply them element-by-element with the attention weight to get the weighted feature. Then multiply the weighted feature with the gate weight element-by-element to get the gated weighted feature. Finally, average the gated weighted features of all time steps and pass the weight matrix and bias Linear transformation, generating time steps Cognitive Load Index .

[0077] It should be noted that the comprehensive feature vector is calculated in the form of temporal embedding features in this expression. The temporal embedding features positionally encode the time series of the comprehensive feature vector through Transformer position encoding to capture temporal dynamics; the spatial embedding features linearly transform the comprehensive vectors through MLP to extract spatial semantic information; the two are dynamically fused through feature splicing and Sigmoid gating mechanism.

[0078] S4. Build a virtual teaching environment. Based on an adaptive strategy optimization algorithm combined with cognitive load data, optimize the three strategy parameters of explanation speed, interaction frequency, and content difficulty in the virtual teaching environment to dynamically generate teaching plans.

[0079] Based on historical student behavior data and student cognitive load data, a virtual teaching environment is built using the Unity3D engine combined with the TensorFlow reinforcement learning framework;

[0080] Furthermore, a virtual teaching scene is first designed in the Unity3D engine, including classroom layout, teaching tools and a virtual teacher role. At the same time, historical student behavior data (such as gaze points and voice features) and cognitive load data (such as pupil diameter changes and fundamental frequency fluctuations) are imported into the scene. Then, the state space is defined through the TensorFlow reinforcement learning framework. The state space includes three strategy parameters: explanation speed, interaction frequency and content difficulty. For example, the explanation speed parameter range is 1 to 5, with 1 representing the slowest and 5 representing the fastest. Then, the action space is defined. The action space includes the adjustment amount for the explanation speed, interaction frequency and content difficulty. For example, the adjustment amount for the explanation speed for each action is -1, 0 or 1. Finally, based on the historical student behavior data and cognitive load data, the reinforcement learning model is trained in the virtual teaching environment. For example, when the cognitive load index is lower than the target value, the reinforcement learning model adjusts the action space parameters to improve the explanation speed and content difficulty, generates an optimized teaching strategy, and provides real-time feedback to the virtual teaching scene to complete the construction and dynamic adjustment of the virtual teaching environment.

[0081] Based on the comprehensive eye movement features and speech features, the state space is defined ;

[0082] Further, ;

[0083] is the speed of explanation, which indicates how fast the teaching content is delivered ( ∈[1, 5], 1 is the slowest and 5 is the fastest);

[0084] is the interaction frequency, which indicates the frequency of interaction between teachers and students during the teaching process ( ∈[1, 5], 1 is the lowest and 5 is the highest);

[0085] is the content difficulty, which indicates the complexity of the teaching content ( ∈[1, 5], 1 is the easiest and 5 is the hardest).

[0086] Define action spaces based on strategies for adjusting teaching speed, interaction frequency, and content difficulty in history courses ;

[0087] Further, ;

[0088] It is the adjustment amount of the explanation speed, indicating the increase or decrease of the explanation speed for each action ;

[0089] It is the adjustment amount of the interaction frequency, indicating the increase or decrease of the interaction frequency for each action ;

[0090] It is the adjustment amount of the content difficulty, indicating the increase or decrease of the content difficulty for each action .

[0091] Based on the deviation between the student's cognitive load index and the target cognitive load index and the stability of the strategy parameters, a reward function is defined. , the expression is:

[0092] ;

[0093] in, is the target cognitive load index, is the weight coefficient of the stability of the strategy parameters;

[0094] It should be explained that The weight used to balance the deviation between the student's cognitive load index and the target cognitive load index and the stability of the strategy parameters. Its value is based on the degree of influence of the strategy parameter adjustment on the teaching effect. If the strategy parameter adjustment has a greater impact on the teaching effect, then Take a smaller value (such as 0.1) to prioritize optimizing the cognitive load index; if the adjustment of strategy parameters has little impact on the teaching effect, then take a larger value (such as 0.5) to prioritize maintaining the stability of strategy parameters. The value range is usually between 0 and 1.

[0095] The state space Combined with students’ cognitive load data as input, the action space The probability distribution of is used as output to construct a policy network ;

[0096] Furthermore, the state space s is first combined with the student's cognitive load data as input, and the state space It includes three strategic parameters: explanation speed, interaction frequency, and content difficulty. For example, the explanation speed is 3, the interaction frequency is 2, the content difficulty is 4, and the cognitive load data is 0.8. Then, the input data is processed through a multi-layer neural network to generate an action space. The probability distribution of action space This includes adjustments to the speed of explanation, interaction frequency, and content difficulty. For example, the probability of adjusting the speed of explanation to -1, 0, or 1 is 0.2, 0.6, and 0.2, respectively. Finally, the action space is output. The probability distribution of is used to guide the adjustment of teaching strategies. For example, the state space is [3, 2, 4, 0.8], the policy network The probabilities of outputting explanation speed adjustments of -1, 0, or 1 are 0.2, 0.6, and 0.2, respectively.

[0097] The state space As input, through the reward function The expected cumulative reward of the current state obtained in the state space is used as output to construct the value network ;

[0098] Furthermore, first we transform the state space As input, the state space s includes three policy parameters: explanation speed, interaction frequency, and content difficulty. For example, the explanation speed is 3, the interaction frequency is 2, and the content difficulty is 4. Then, the input data is processed by a multi-layer neural network to calculate the expected cumulative reward of the current state. The reward function Based on the deviation between the cognitive load index and the target cognitive load index and the stability of the strategy parameters, the expected cumulative reward of the current state is output to evaluate the quality of the teaching strategy. For example, the state space For [3, 2, 4], the value network The expected cumulative reward for outputting the current state is 0.75.

[0099] Based on the strategy network and value network, an adaptive strategy optimization algorithm is constructed. Combined with the students' cognitive load data, the three strategy parameters of explanation speed, interaction frequency, and content difficulty are optimized in the virtual teaching environment. The expression is:

[0100] ;

[0101] in, At the current time step After optimization strategy parameters, is in the time step No. strategy parameters, Indicates limiting the value of the policy parameter to [1, 5];

[0102] Next, we first obtain the policy parameter values ​​for the previous time step, for example, if the explanation speed is 3, the interaction frequency is 2, and the content difficulty is 4. We then generate an action space probability distribution through the policy network and calculate the policy parameter adjustment for the current time step, combined with the expected cumulative reward output by the value network. We then add the policy parameter value from the previous time step to the adjustment to obtain the policy parameter value for the current time step. Finally, we restrict the policy parameter value to the range of 1 to 5 to ensure that the parameter value is within a reasonable range. For example, if the explanation speed for the previous time step is 3, the probability of the policy network generating an adjustment of 1 is 0.6, and the expected cumulative reward output by the value network is 0.8, the explanation speed for the current time step is calculated to be 4, and after restriction, it remains 4, completing the optimization of the policy parameters.

[0103] Based on the optimized strategy parameters ,dynamically generate teaching plans through RLDP;

[0104] Based on the optimized policy parameters, the specific process of dynamically generating a teaching plan through RLDP is as follows:

[0105] First, based on the deviation between the real-time cognitive load index and the target value, the RLDP dynamically adjusts the explanation speed. If the current cognitive load is low, the explanation pace is accelerated, for example, the explanation time of each knowledge point is shortened from 5 minutes to 4 minutes, and the voice broadcast speed is increased; otherwise, the pace is slowed down, extended to 6 minutes, and the speaking speed is reduced. Secondly, based on the interaction frequency value in the strategy parameter, the teaching interaction nodes are dynamically set. When the frequency parameter increases, two situational question-and-answer tasks are inserted within 8 minutes, such as having students solve practical problems through virtual scene simulations; when the frequency parameter decreases, the basic multiple-choice test is adjusted to 15 minutes. Finally, based on the content difficulty parameter, matching content is dynamically selected from the pre-set knowledge base. When the difficulty parameter increases, a comprehensive case analysis that crosses multiple disciplines (such as a bridge load calculation that combines mathematics and physics) is selected, and step-by-step guidance prompts are added; when the parameter decreases, it switches to basic concept diagrams and single-step exercises. The entire process uses the dynamic programming mechanism of reinforcement learning to collect cognitive load data every 5 minutes. When students show efficient understanding in case discussions, the difficulty parameters are immediately increased and advanced topics are introduced. If abnormal increases in pupil diameter fluctuations are detected, the speed parameters are immediately lowered and relaxation guidance voice is inserted.

[0106] This embodiment also provides a cloud-based education management system based on big data, including: a feature extraction module, a feature fusion module, a cognitive load prediction module and a teaching plan generation module; the feature extraction module is used to collect students' eye movement data and voice data and pre-process them, and at the same time extract MFCC features and comprehensive eye movement features; the feature fusion module is used to use a dynamic time warping algorithm to fuse MFCC feature vectors and comprehensive eye movement feature vectors according to timestamps to generate a comprehensive feature vector; the cognitive load prediction module is used to construct a dual-stream Transformer model based on a hierarchical processing mechanism and a cross-modal attention mechanism, and predict the student's cognitive load index based on the comprehensive feature vector; the teaching plan generation module is used to construct a virtual teaching environment, and based on the adaptive strategy optimization algorithm combined with cognitive load data, optimize the three strategy parameters of explanation speed, interaction frequency and content difficulty in the virtual teaching environment, and dynamically generate a teaching plan.

[0107] This embodiment also provides a computer device suitable for the cloud-based education management method based on big data, 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 cloud-based education management method based on big data proposed in the above embodiment.

[0108] 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.

[0109] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the cloud-based education management method based on big data 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, magnetic disk or optical disk.

[0110] In summary, the present invention achieves efficient fusion of multimodal data by: fusing MFCC feature vectors and comprehensive eye movement feature vectors according to timestamps using a dynamic time warping algorithm to generate a comprehensive feature vector; by constructing a dual-stream Transformer model, combining the temporal embedding layer and the spatial embedding layer, and dynamically fusing temporal features and spatial features through a cross-modal attention mechanism, accurate prediction of cognitive load is achieved, providing a basis for the dynamic optimization of teaching strategies.

[0111] 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 cloud-based education management method based on big data, characterized by: include, Collect and preprocess the students' eye movement data and speech data, and extract MFCC features and comprehensive eye movement features. The eye movement data includes gaze point coordinates, gaze duration, and pupil diameter changes; the speech data includes fundamental frequency, speech rate, silence intervals, and formants. The dynamic time warping algorithm is used to fuse the MFCC feature vector and the comprehensive eye movement feature vector according to the timestamp to generate a comprehensive feature vector; Based on the hierarchical processing mechanism and cross-modal attention mechanism, a dual-stream Transformer model is constructed, and the students' cognitive load index is predicted based on the comprehensive feature vector. The specific steps are as follows: The temporal embedding layer performs position encoding on the time series of the comprehensive feature vector through Transformer position encoding to generate temporal embedding features; The spatial embedding layer performs linear transformation on the comprehensive feature vectors through MLP to generate spatial embedding features; The temporal embedding and spatial embedding are spliced ​​together through the cross-modal attention mechanism, and the weights of temporal features and spatial features are adjusted in combination with the Sigmoid gate control mechanism. The temporal embedding layer, spatial embedding layer, and cross-modal attention mechanism are combined to construct a two-stream Transformer model, and the comprehensive feature vector is input to predict the students' cognitive load index; Construct a virtual teaching environment, and based on the adaptive strategy optimization algorithm combined with cognitive load data, optimize the three strategy parameters of explanation speed, interaction frequency and content difficulty in the virtual teaching environment, and dynamically generate teaching plans.

2. The cloud-based education management method based on big data according to claim 1, characterized in that: The extraction of MFCC features and comprehensive eye movement features includes the following steps: The speech data is divided into frames, each frame signal is subjected to fast Fourier transform to generate a spectrogram, and the MFCC feature vector is extracted through discrete cosine transform; The gaze heat map is generated by performing Gaussian kernel density estimation on the gaze point coordinates, the distribution statistics of the gaze duration are identified, the pupil diameter change rate and fluctuation frequency are extracted, and the path complexity is calculated by segmenting the scanning path to generate a comprehensive eye movement feature vector.

3. The cloud-based education management method based on big data according to claim 1, characterized in that: The dynamic time warping algorithm is used to fuse the MFCC feature vector and the comprehensive eye movement feature vector according to the timestamp to generate a comprehensive feature vector. The specific steps are as follows: Calculate the Euclidean distance between the MFCC feature vector and the eye movement feature vector frame by frame; Based on the Euclidean distance between the MFCC feature vector and the eye movement feature vector, the cost matrix is ​​constructed using the global sequence alignment method; Based on the cost matrix, a dynamic time planning algorithm is used to find the optimal alignment path according to timestamps. Then, the MFCC features and eye movement features are fused according to the timestamps according to the optimal alignment path to generate a comprehensive feature vector.

4. The cloud-based education management method based on big data according to claim 1, characterized in that: Based on historical student behavior data and students' cognitive load data, a virtual teaching environment is constructed through the Unity3D engine and combined with the TensorFlow reinforcement learning framework.

5. The cloud-based education management method based on big data according to claim 4, characterized in that: The adaptive strategy optimization algorithm is constructed as follows: Define the state space, action space, and reward function; Construct a strategy network based on the state space, action space, and students’ cognitive load data; Construct a value network based on the state space and reward function; Based on the policy network and value network, an adaptive policy optimization algorithm is constructed.

6. The cloud-based education management method based on big data according to claim 5, characterized in that: The specific steps of dynamically generating a teaching plan are as follows: Combine the adaptive strategy optimization algorithm with the learner's cognitive load data to optimize strategy parameters; According to the optimized strategy parameters, the teaching plan is dynamically generated through RLDP.

7. A cloud-based education management system based on big data, based on the cloud-based education management method based on big data according to any one of claims 1 to 6, characterized in that: Including, feature extraction module, feature fusion module, cognitive load prediction module and teaching plan generation module; Feature extraction module, used to collect students' eye movement data and speech data and preprocess them, and extract MFCC features and comprehensive eye movement features; A feature fusion module is used to fuse the MFCC feature vector and the comprehensive eye movement feature vector according to the timestamp using the dynamic time warping algorithm to generate a comprehensive feature vector; The cognitive load prediction module is used to build a two-stream Transformer model based on the hierarchical processing mechanism and cross-modal attention mechanism, and predict the student's cognitive load index based on the comprehensive feature vector; The teaching plan generation module is used to build a virtual teaching environment. Based on the adaptive strategy optimization algorithm combined with cognitive load data, it optimizes the three strategy parameters of explanation speed, interaction frequency and content difficulty in the virtual teaching environment and dynamically generates teaching plans.

8. 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 cloud-based education management method based on big data described in any one of claims 1 to 6 are implemented.

9. 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 cloud-based education management method based on big data described in any one of claims 1 to 6 are implemented.

Images