A multi-source teaching data processing method based on deep learning

Abstract

The present application relates to a multi-source teaching data processing method based on deep learning, comprising acquiring multimodal data, the multimodal data including text data, audio data and video data; preprocessing the multimodal data using a modal separation technique to obtain a plurality of independent feature sets; generating a fused feature set based on the plurality of independent feature sets using a feature fusion algorithm; determining a feature extraction range for the fused feature set using a convolution adjustment algorithm; processing the feature extraction range using a multi-scale feature fusion network to obtain a high-dimensional feature space representation; generating a clear feature set based on the high-dimensional feature space representation using a noise suppression algorithm; determining a difficulty gradient division for the clear feature set using a difficulty perception model; processing the difficulty gradient division using a network optimization technique to obtain a real-time optimized hierarchical division scheme; and generating a final output based on the real-time optimized hierarchical division scheme.

Description

Technical Field

[0001] The present application relates to the technical field of electronic digital data processing, and in particular to a multi-source teaching data processing method based on deep learning. Background Art

[0002] When using deep learning technology to process multi-source teaching data to achieve dense layered design of teaching content, a hidden and unique technical contradiction gradually emerges: when convolutional neural networks are used to extract subtle features from multi-source teaching data, the system needs to accurately divide the microscopic difficulty gradient in the high-dimensional feature space. However, this process faces the dual constraints of data heterogeneity and model computational complexity.

[0003] Multi-source teaching data usually contains multiple modalities such as text, audio, and video. The feature distribution of each modality varies significantly, including the difficulty in aligning the semantic density of text data and the inter-frame temporal correlation of video data in terms of magnitude and structure. This heterogeneity causes convolutional neural networks to easily ignore certain modality-specific micro-difficulty clues due to the fixed convolution kernel during cross-modal fusion, such as changes in speaking speed in short-term audio or fluctuations in sentence complexity in text. At the same time, in order to capture these subtle features, the network layer needs to be deepened to improve the resolution ability, but deepening the layer makes the computational complexity increase exponentially. Especially when dividing dense layers in real time, the system may not be able to meet the immediacy requirements of teaching scenarios due to excessive latency.

[0004] In addition, noise interference from multi-source data further exacerbates the contradiction: background noise may mask key difficulty signals in the audio, while non-standard expressions in the text may mislead semantic parsing, making the dividing boundary of the difficulty gradient unclear.

[0005] This contradiction manifests itself in business scenarios as follows: when the system attempts to break down teaching content into dozens or even hundreds of dense levels, it must ensure the complete extraction of cross-modal features while avoiding the loss of real-time performance of the hierarchical division due to excessive computational burden, forming a deep contradiction between technology and scenario requirements. Summary of the Invention

[0006] In order to solve the problems existing in the above-mentioned prior art, the purpose of this application is to provide a multi-source teaching data processing method based on deep learning.

[0007] The present application discloses a method for processing multi-source teaching data based on deep learning, comprising the following steps:

[0008] Step S101: After acquiring multi-source teaching data, pre-process the text, audio, and video data using modality separation technology, extract independent feature sets from each modality, and obtain preliminary mapping results of modality feature distribution;

[0009] Step S102: Based on the preliminary mapping results of the modal feature distribution, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size according to the differences in the modal feature distribution to determine the feature extraction range before cross-modal fusion;

[0010] Step S103: Processing the adjusted feature extraction range through a multi-scale feature fusion network, fusing text semantic density, audio speech rate variation, and temporal correlation between video frames at different scales to obtain a unified high-dimensional feature space representation;

[0011] Step S104: Based on the unified high-dimensional feature space representation, a noise suppression algorithm is used to filter data segments with large background noise or non-standard expression noise interference. If the noise detected exceeds a preset threshold, the corresponding feature weight is attenuated to obtain a clear feature set;

[0012] Step S105: After obtaining a clear feature set, the feature set is scanned layer by layer using a micro-difficulty perception model. Micro-difficulty clues are matched using a difficulty gradient template based on semantic complexity or speech rate changes to determine the initial boundary of the difficulty gradient.

[0013] Step S106: Optimize the convolutional layer or fully connected layer using lightweight network compression technology based on the initial boundary of the difficulty gradient, thereby reducing computational complexity while maintaining resolution, and obtaining a hierarchical partitioning solution with real-time optimization.

[0014] Step S107: Using a real-time optimized hierarchical partitioning scheme and a dynamic time window algorithm, the dense hierarchical partitioning is iteratively adjusted. If the fuzziness of the partitioning boundary calculated based on the feature overlap rate or confidence level is higher than a preset threshold, contextual semantics or audio spectrum features are introduced for recalibration to obtain the final dense hierarchical partitioning result.

[0015] Step S108: Based on the final dense hierarchical division result, a cross-modal fusion effect evaluation mechanism based on accuracy or recall rate is used. If the fusion accuracy is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated and recalculated to obtain an optimized hierarchical output of teaching content.

[0016] Preferably, in step S101, the multimodal data is preprocessed using a modal separation technique to obtain a plurality of independent feature sets, including: for the text data, a text feature extraction algorithm is used to obtain a text independent feature set;

[0017] For the audio data, obtain an audio independent feature set by audio data processing, and determine a feature distribution of the audio independent feature set;

[0018] For the video data, using video feature extraction technology to obtain a video independent feature set, and determining the modal features of the video independent feature set;

[0019] fusing the text-independent feature set, the audio-independent feature set, and the video-independent feature set to obtain a multimodal feature set;

[0020] A clustering algorithm is used for the multimodal feature set to obtain a modal feature distribution mapping result.

[0021] Preferably, in step S102, for the fused feature set, a convolution adjustment algorithm is used to determine a feature extraction range, including: obtaining feature distribution data of the fused feature set, and obtaining a quantitative representation of distribution differences through a preset feature distribution analysis method;

[0022] According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain adjusted convolution kernel parameters;

[0023] Based on the adjusted convolution kernel parameters, a feature extraction algorithm is used to extract feature data before cross-modal fusion from the fusion feature set to determine a feature extraction range;

[0024] If the modal consistency within the feature extraction range is lower than a preset threshold, the convolution kernel size is readjusted according to the distribution difference to obtain an updated feature extraction range.

[0025] Preferably, in the step S103, the feature extraction range is processed by a multi-scale feature fusion network to obtain a high-dimensional feature space representation, including: extracting text semantic density features for the text data;

[0026] Extracting audio speech rate change features from the audio data;

[0027] Extracting temporal correlation features between video frames from the video data;

[0028] fusing the text semantic density feature, the audio speech rate variation feature, and the video frame temporal correlation feature to obtain a preliminary multi-scale feature representation;

[0029] Processing the preliminary multi-scale feature representation through a convolutional neural network, combining it with an attention mechanism to process inter-frame relationships, and generating a comprehensive spatiotemporal feature representation;

[0030] A unified high-dimensional feature space representation is generated based on the comprehensive spatiotemporal feature representation.

[0031] Preferably, in step S104, generating a clear feature set based on the high-dimensional feature space representation using a noise suppression algorithm includes: detecting interference data formed by background noise and non-standard expressions using a noise suppression algorithm for the high-dimensional feature space representation, and determining interference data segments;

[0032] Extracting noise intensity from the interference data segment, and if the noise intensity exceeds a preset threshold, adjusting the weight of the corresponding feature using a weight decay algorithm to obtain an adjusted feature set;

[0033] According to the adjusted feature set, a clustering algorithm is used to obtain a clear feature subset;

[0034] The data segment is reconstructed in the feature space by using the clear feature subset to obtain a feature representation after filtering.

[0035] Preferably, in step S105, determining the difficulty gradient division for the clear feature set using a difficulty perception model includes: scanning the clear feature set layer by layer using a microscopic difficulty perception model to obtain a feature level sequence after scanning;

[0036] Matching microscopic clues in the feature hierarchy sequence through a preset gradient template to determine the initial trend of semantic complexity or speech rate change;

[0037] Based on the preliminary change trend, semantic complexity analysis is used to determine the difficulty gradient fluctuation and obtain the fluctuation detection result;

[0038] If the fluctuation detection result exceeds a preset threshold, the division boundary is adjusted through template matching to determine an optimized difficulty gradient boundary.

[0039] Preferably, in step S106, the difficulty gradient division is processed by network optimization technology to obtain a real-time optimized hierarchical division scheme, including: obtaining a convolutional neural network model to be optimized, and processing the convolution layer by using lightweight network compression technology to obtain an optimized convolution layer structure;

[0040] According to the optimized convolutional layer structure, the parameters of the fully connected layer are adjusted to obtain streamlined parameters of the fully connected layer;

[0041] Evaluate the computational complexity of the simplified fully connected layer parameters to obtain a network model with reduced complexity;

[0042] According to the network model with reduced complexity, a preliminary hierarchical structure is divided. If the resolution capability does not meet the preset requirements, the division boundary is adjusted to obtain an optimized hierarchical division result.

[0043] Preferably, in step S107, generating a final output according to the real-time optimized hierarchical division scheme includes: iteratively adjusting the real-time optimized hierarchical division scheme using a dynamic time window algorithm to obtain a preliminary dense hierarchical structure;

[0044] Calculating a feature overlap ratio of a dividing boundary in the preliminary dense hierarchical structure to obtain a fuzziness evaluation value;

[0045] If the fuzziness evaluation value is higher than a preset threshold, calibrating the division boundary according to the contextual semantic features and the audio spectrum features to obtain a calibrated dense hierarchical structure;

[0046] According to the calibrated dense hierarchical structure, the confidence is calculated to generate a final dense hierarchical partitioning result.

[0047] The multi-source teaching data processing method based on deep learning described in the present application has the advantages that, in order to solve the problem of large differences in the distribution of different modal features such as text, audio, and video in multi-source teaching data, the present invention adopts an adaptive convolution kernel adjustment algorithm to dynamically determine the feature extraction range, and realizes the unified representation of cross-modal information through a multi-scale feature fusion network. In order to solve the problems of noise interference and inaccurate difficulty division in teaching content, the present invention utilizes a noise suppression algorithm and a microscopic difficulty perception model to perform feature optimization and difficulty gradient division. Through lightweight network compression and a dynamic time window algorithm, the present invention realizes dense hierarchical division with real-time optimization. Finally, based on the cross-modal fusion effect evaluation mechanism, the present invention can output high-precision teaching content stratification results, effectively improving the intelligent processing and refined management level of multimodal teaching resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is the process of a multi-source teaching data processing method based on deep learning described in this application Figure 1 ;

[0049] Figure 2 This is the process of a multi-source teaching data processing method based on deep learning described in this application Figure 2 . DETAILED DESCRIPTION

[0050] like Figure 1-Figure 2 As shown, the multi-source teaching data processing method based on deep learning described in this application includes the following steps:

[0051] Step S101: After acquiring multi-source teaching data, pre-process the text, audio, and video data using modality separation technology, extract independent feature sets from each modality, and obtain preliminary mapping results of modality feature distribution;

[0052] Step S102: Based on the preliminary mapping results of the modal feature distribution, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size according to the differences in the modal feature distribution to determine the feature extraction range before cross-modal fusion;

[0053] Step S103: Processing the adjusted feature extraction range through a multi-scale feature fusion network, fusing text semantic density, audio speech rate variation, and temporal correlation between video frames at different scales to obtain a unified high-dimensional feature space representation;

[0054] Step S104: Based on the unified high-dimensional feature space representation, a noise suppression algorithm is used to filter data segments with large background noise or non-standard expression noise interference. If the noise detected exceeds a preset threshold, the corresponding feature weight is attenuated to obtain a clear feature set;

[0055] Step S105: After obtaining a clear feature set, the feature set is scanned layer by layer using a micro-difficulty perception model. Micro-difficulty clues are matched using a difficulty gradient template based on semantic complexity or speech rate changes to determine the initial boundary of the difficulty gradient.

[0056] Step S106: Optimize the convolutional layer or fully connected layer using lightweight network compression technology based on the initial boundary of the difficulty gradient, thereby reducing computational complexity while maintaining resolution, and obtaining a hierarchical partitioning solution with real-time optimization.

[0057] Step S107: Using a real-time optimized hierarchical partitioning scheme and a dynamic time window algorithm, the dense hierarchical partitioning is iteratively adjusted. If the fuzziness of the partitioning boundary calculated based on the feature overlap rate or confidence level is higher than a preset threshold, contextual semantics or audio spectrum features are introduced for recalibration to obtain the final dense hierarchical partitioning result.

[0058] Step S108: Based on the final dense hierarchical division result, a cross-modal fusion effect evaluation mechanism based on accuracy or recall rate is used. If the fusion accuracy is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated and recalculated to obtain an optimized hierarchical output of teaching content.

[0059] like Figure 1-Figure 2 As shown, in step S101, after obtaining multi-source teaching data, the text, audio, and video data are preprocessed using modality separation technology, and independent feature sets are extracted from each modality to obtain a preliminary mapping result of the modality feature distribution.

[0060] Furthermore, in step S101, multi-source teaching data is obtained, wherein the multi-source teaching data includes text data, audio data and video data;

[0061] Use modal separation technology to process multi-source teaching data to obtain text modal data, audio modal data and video modal data;

[0062] For text modal data, a text feature extraction algorithm is used to obtain a text-independent feature set;

[0063] For audio modal data, extract an audio independent feature set through audio data processing, and determine the feature distribution of the audio independent feature set;

[0064] For video modal data, video feature extraction technology is used to extract video independent feature sets and determine the modal features of the video independent feature sets;

[0065] Fusing text-independent feature sets, audio-independent feature sets, and video-independent feature sets to obtain a multimodal feature set;

[0066] For the multimodal feature set, a clustering algorithm is used to obtain the modal feature distribution mapping result;

[0067] Extract key features from the modal feature distribution mapping results and generate preliminary modal feature distribution results;

[0068] Based on the preliminary modal feature distribution results, the feature space unification technology is used to generate feature outputs suitable for downstream tasks.

[0069] Specifically, in step S101, multi-source teaching data is obtained, and the multi-source teaching data includes text data, audio data and video data;

[0070] Use modality separation technology to process multi-source teaching data. If the data is in mixed format, use FFmpeg tool to separate text modality data, audio modality data and video modality data.

[0071] For text modal data, the TF-IDF algorithm is used to extract text-independent feature sets, and the feature dimension is set to 300;

[0072] For audio modal data, the audio independent feature set is extracted through Mel spectrum analysis, and the K-means clustering algorithm is used to determine its feature distribution, with the number of cluster centers being 5;

[0073] For video modal data, the OpenCV library is used to extract the optical flow features between video frames, and a convolutional neural network is used to determine the modal features of the video independent feature set. The network has three layers.

[0074] The text-independent feature set, audio-independent feature set, and video-independent feature set are integrated and weighted averaged to obtain a multimodal feature set with weights of 0.4, 0.3, and 0.3, respectively.

[0075] For the multimodal feature set, the DBSCAN clustering algorithm is used to obtain the modal feature distribution mapping results, and the neighborhood radius is set to 0.5;

[0076] Extract key features from the modal feature distribution mapping results, use PCA dimensionality reduction technology to generate preliminary modal feature distribution results, and retain 80% of the variance;

[0077] Based on the preliminary modal feature distribution results, the feature space unification technology is adopted to generate feature outputs suitable for downstream tasks through linear discriminant analysis, with a feature dimension of 100.

[0078] like Figure 1-Figure 2 As shown, in step S102, based on the preliminary mapping results of the modal feature distribution, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size according to the difference in the modal feature distribution to determine the feature extraction range before cross-modal fusion.

[0079] Furthermore, in step S102, preliminary data of modal features are obtained, and a quantitative representation of distribution differences is obtained through a preset feature distribution analysis method;

[0080] According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain the adjusted convolution kernel parameters;

[0081] Based on the adjusted convolution kernel parameters, a feature extraction algorithm is used to extract feature data before cross-modal fusion from the modal features and determine the feature extraction range;

[0082] Obtain data within the feature extraction range and generate a fusion feature set through a cross-modal fusion algorithm;

[0083] If the modal consistency of the fused feature set is lower than the preset threshold, the convolution kernel size is readjusted according to the distribution difference to obtain the updated feature extraction range;

[0084] Through the updated feature extraction range, the feature extraction algorithm is used to process the modal features to obtain the optimized fusion feature set;

[0085] Based on the optimized fusion feature set, the consistency of cross-modal fusion features is judged through a preset verification method to obtain the final feature representation;

[0086] Obtain cross-modal fusion data from the dense hierarchical partitioning results, use precision evaluation and recall evaluation to calculate the initial fusion accuracy value;

[0087] If the initial fusion accuracy value is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated, and the optimized feature extraction data is obtained.

[0088] Specifically, in step S102 , preliminary data of modal features are obtained, and the feature distribution is analyzed using a Gaussian mixture model (GMM) to obtain a quantitative representation of the distribution difference, including calculating the inter-modal Kullback-Leibler divergence (KL divergence) of 0.75;

[0089] According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is adopted, including using a dynamic window adjustment strategy to adjust the convolution kernel size from 3x3 to 5x5, and the adjusted convolution kernel parameters are obtained;

[0090] Based on the adjusted convolution kernel parameters, a convolutional neural network (CNN) feature extraction algorithm is used to extract feature data before cross-modal fusion from the modal features, and the feature extraction range is determined to be a 128-dimensional vector;

[0091] Obtain data within the feature extraction range and generate a fused feature set using a cross-modal fusion algorithm based on the attention mechanism, including a feature dimension of 256;

[0092] If the modality consistency of the fused feature set is lower than the preset threshold of 0.9, the convolution kernel size is readjusted according to the distribution difference, including adjusting the convolution kernel size from 5x5 to 7x7 to obtain the updated feature extraction range;

[0093] Through the updated feature extraction range, the CNN feature extraction algorithm is used to process the modal features to obtain the optimized fusion feature set, including a feature dimension of 512;

[0094] Based on the optimized fusion feature set, the cross-modal fusion feature consistency is judged by the cross-validation method to obtain the final feature representation, including a consistency score of 0.95;

[0095] Obtain cross-modal fusion data from the dense hierarchical partitioning results, use precision evaluation and recall evaluation to calculate the initial fusion accuracy values, including precision of 0.85 and recall of 0.80;

[0096] If the initial fusion accuracy value is lower than the preset threshold of 0.9, the adaptive convolution kernel parameters are adjusted, including adjusting the convolution kernel size from 7x7 to 9x9, updating the feature extraction range, and obtaining optimized feature extraction data, including a feature dimension of 1024.

[0097] like Figure 1-Figure 2 As shown, in step S103, the adjusted feature extraction range is processed by a multi-scale feature fusion network, and the text semantic density, audio speech rate change and temporal correlation between video frames are fused at different scales to obtain a unified high-dimensional feature space representation.

[0098] Furthermore, in step S103, multimodal data is acquired, wherein the multimodal data includes text data, audio data, and video data;

[0099] Extracting text semantic density features from the text data;

[0100] Extracting audio speech rate change features from the audio data;

[0101] Extracting temporal correlation features between video frames from the video data;

[0102] fusing the text semantic density feature, the audio speech rate variation feature, and the video frame temporal correlation feature to obtain a preliminary multi-scale feature representation;

[0103] Using a convolutional neural network to process the text semantic density features to obtain an enhanced semantic density representation;

[0104] fusing the enhanced semantic density representation and the audio speech rate variation feature, analyzing the speech rate variation sequence through a recurrent neural network to obtain a time-series enhanced semantic feature;

[0105] Combining the temporal correlation features between video frames, an attention mechanism is used to process the relationship between frames to obtain a comprehensive spatiotemporal feature representation;

[0106] generating a unified high-dimensional feature space representation through network processing based on the comprehensive spatiotemporal feature representation;

[0107] Determine whether the feature distribution represented by the high-dimensional feature space meets a preset threshold, and if so, output a final feature representation;

[0108] Through feature space unification technology, feature output suitable for downstream tasks is generated.

[0109] Specifically, in step S103, text data, audio data, and video data are acquired through the data acquisition interface, where the text data is a data set containing 1000 sentences, the audio data is a speech clip with a sampling rate of 44.1kHz, and the video data is a video clip with a frame rate of 30fps;

[0110] The BERT model was used to process the text data, extracting the semantic density features of each sentence, and the calculated lexical density value was 0.85;

[0111] Short-time Fourier transform was used to extract speech rate variation features from the audio data, and the average speech rate variation rate per second was 0.12;

[0112] The optical flow method is used to extract the temporal correlation features between frames of video data, and the inter-frame motion vector is 0.35;

[0113] The text semantic density features, audio speech rate change features, and video frame temporal correlation features are input into the multi-scale feature fusion network and initially fused using the weighted average method to obtain a multi-scale feature representation vector.

[0114] A convolutional neural network is used to process the semantic density features of text. The convolution kernel size is 3×3 and the output is an enhanced semantic density representation vector with a dimension of 128.

[0115] The enhanced semantic density representation and audio speech rate variation features are input into the LSTM network, the number of hidden layer units is set to 64, and the time-series enhanced semantic feature vector is output;

[0116] Combined with the temporal correlation features between video frames, a multi-head attention mechanism is used to calculate the inter-frame relationship weights and output a comprehensive spatiotemporal feature representation vector with a dimension of 256;

[0117] The comprehensive spatiotemporal feature representation is input into the fully connected network, and a unified high-dimensional feature space representation vector with a dimension of 512 is output;

[0118] The difference between the feature distribution represented in the high-dimensional feature space and the preset threshold is calculated by KL divergence. If the difference value is less than 0.01, the final feature representation is output;

[0119] The principal component analysis method is used to reduce the dimensionality of the final feature representation to generate a 128-dimensional feature output suitable for downstream tasks.

[0120] like Figure 1-Figure 2 As shown, in step S104, based on the unified high-dimensional feature space representation, a noise suppression algorithm is used to filter the data segments with large background noise or non-standard expression noise interference. If the noise is detected to exceed the preset threshold, the corresponding feature weight is attenuated to obtain a clear feature set.

[0121] Furthermore, in step S104, input data is obtained and mapped into a high-dimensional feature set;

[0122] Use noise suppression algorithms to detect background noise and interference data formed by non-standard expressions in high-dimensional feature sets and determine interference data segments;

[0123] Extracting noise intensity from the interference data segment and determining whether the noise intensity exceeds a preset threshold;

[0124] If it exceeds, the corresponding feature is marked as a feature to be processed;

[0125] The weight of the feature to be processed is adjusted by the weight decay algorithm to obtain the adjusted feature set;

[0126] Clustering algorithm is used to group the adjusted feature set to obtain clear feature subsets;

[0127] Reconstruct the data segment based on the clear feature subset in the feature space to obtain the feature representation after filtering;

[0128] Extract the final output from the filtered feature representation and determine whether the final output meets the preset clarity condition;

[0129] If not, return to the noise suppression algorithm step and reprocess;

[0130] Obtain preliminary data of modal characteristics and obtain a quantitative representation of distribution differences through a preset characteristic distribution analysis method;

[0131] According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain the adjusted convolution kernel parameters.

[0132] Specifically, in step S104, input data is obtained and mapped into a high-dimensional feature set using a principal component analysis (PCA) algorithm to generate a feature space containing 128-dimensional feature vectors;

[0133] A noise suppression algorithm based on wavelet transform is used to detect background noise and interference data formed by non-standard expressions in high-dimensional feature sets. The interference data segments with a signal-to-noise ratio below 10dB are determined by calculating the signal-to-noise ratio (SNR).

[0134] Extract the noise intensity from the interference data segment and use the fast Fourier transform (FFT) to calculate the noise spectrum. If the peak value in the noise spectrum exceeds the preset threshold of -20dB, the corresponding feature is marked as a feature to be processed.

[0135] The features to be processed are adjusted through the weight decay algorithm, and the feature weight is decayed from 1.0 to 0.3 using the exponential decay function to obtain the adjusted feature set;

[0136] The K-means clustering algorithm is used to group the adjusted feature set, and the number of clusters is set to 5 to obtain a clear feature subset;

[0137] Reconstruct the data segment based on the subset of clear features in the feature space and use the inverse Fourier transform to generate the filtered feature representation;

[0138] Extract the final output from the filtered feature representation and use the Structural Similarity Index (SSIM) to determine whether the final output meets the preset clarity condition of 0.9. If not, return to the noise suppression algorithm step and reprocess;

[0139] Obtain preliminary data on modal features, perform feature distribution analysis using the Gaussian mixture model (GMM), obtain a quantitative representation of distribution differences, and calculate the KL divergence value to be 0.5;

[0140] According to the quantitative representation of the distribution difference, the adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size, and the convolution kernel is adjusted from 3x3 to 5x5 to obtain the adjusted convolution kernel parameters.

[0141] like Figure 1-Figure 2 As shown, in step S105, after obtaining a clear feature set, the feature set is scanned layer by layer using a micro-difficulty perception model, and the preliminary division boundary of the difficulty gradient is determined by matching the micro-difficulty clues with a difficulty gradient template based on semantic complexity or speech speed changes.

[0142] Furthermore, in step S105, audio data is acquired and hierarchically divided into layers using a dynamic time window algorithm to obtain a preliminary dense hierarchical structure;

[0143] For the preliminary dense hierarchical structure, the feature overlap rate of each partition boundary is calculated to obtain the fuzziness evaluation value of each partition boundary;

[0144] Obtaining contextual semantic features of each partition boundary in the preliminary dense hierarchical structure; if the fuzziness evaluation value is higher than a preset threshold, adjusting the partition boundary according to the contextual semantic features to obtain an adjusted dense hierarchical structure;

[0145] Extracting spectral features of the audio data, and calibrating the adjusted dense hierarchical structure according to the spectral features to obtain a calibrated dense hierarchical structure;

[0146] For each partition boundary in the calibrated dense hierarchical structure, calculate its confidence and obtain the stability confirmation value;

[0147] According to the stability confirmation value, the window algorithm is used to iteratively update the real-time optimization solution to obtain the final dense hierarchical structure;

[0148] Extract the matching relationship between dense layers and real-time optimization from the final dense hierarchical structure and determine the dense layer division scheme;

[0149] Obtain the feature set in the dense hierarchical partitioning scheme, scan the feature set layer by layer using the microscopic difficulty perception model, and obtain the scanned feature hierarchical sequence;

[0150] By matching the microscopic clues in the feature hierarchy sequence with a preset gradient template, the initial trend of semantic complexity or speech rate change is determined, and the preliminary dividing boundary of the difficulty gradient is obtained.

[0151] Specifically, in step S105, after acquiring the audio data, the audio data is hierarchically divided using a dynamic time window algorithm with 50 milliseconds as the basic unit, and frequency features are extracted through FFT (Fast Fourier Transform) to obtain a preliminary dense hierarchical structure;

[0152] For the preliminary dense hierarchical structure, the feature overlap rate of each partition boundary is calculated and evaluated using the Jaccard similarity coefficient. If the overlap rate exceeds 0.7, it is marked as high ambiguity, and the ambiguity evaluation value of each partition boundary is generated;

[0153] Obtain the contextual semantic features of each partition boundary in the preliminary dense hierarchical structure. If the fuzziness evaluation value is higher than the preset threshold of 0.8, analyze the contextual semantics through the BERT model, adjust the partition boundary, and obtain the adjusted dense hierarchical structure;

[0154] Extract the spectral features of the audio data and use Mel-Frequency Cepstral Coefficients (MFCC) to calibrate the adjusted dense hierarchical structure to ensure that the frequency distribution is consistent with the semantic features, thus obtaining a calibrated dense hierarchical structure.

[0155] For each partition boundary in the calibrated dense hierarchical structure, its confidence is calculated, and the boundary stability is evaluated using a Bayesian probability model. If the confidence is lower than 0.9, it is marked as unstable and the stability confirmation value is obtained;

[0156] Based on the stability confirmation value, a sliding window algorithm is used to iteratively update the real-time optimization solution in units of 10 milliseconds to ensure the real-time and accuracy of boundary division and obtain the final dense hierarchical structure;

[0157] Extract the matching relationship between dense levels and real-time optimization from the final dense hierarchical structure, use the KNN algorithm (K nearest neighbor algorithm) for feature matching, and determine the dense hierarchical division scheme;

[0158] Obtain the feature set in the dense hierarchical partitioning scheme, scan the feature set layer by layer using the micro-difficulty perception model, extract the hierarchical sequence through the LSTM (long short-term memory network), and obtain the scanned feature hierarchical sequence;

[0159] By matching the microscopic clues in the feature hierarchy sequence with a preset gradient template, the DTW (Dynamic Time Warping) algorithm is used to analyze the initial change trend of semantic complexity or speech rate. If the change exceeds 20%, it is marked as a difficulty gradient boundary, and the preliminary division boundary of the difficulty gradient is obtained.

[0160] like Figure 1-Figure 2As shown, in step S106, for the preliminary division boundary of the difficulty gradient, the convolution layer or the fully connected layer is optimized by lightweight network compression technology, so as to reduce the computational complexity while maintaining the resolution ability, and obtain a hierarchical division scheme with real-time optimization.

[0161] Furthermore, in step S106, a convolutional neural network model to be optimized is obtained, where the model includes at least one convolutional layer and at least one fully connected layer;

[0162] The convolution layer is processed using a lightweight network compression technology to obtain an optimized convolution layer structure;

[0163] According to the optimized convolutional layer structure, the structure of the fully connected layer is adjusted to obtain streamlined fully connected layer parameters;

[0164] Evaluate the computational complexity of the streamlined fully connected layer parameters and obtain a network model with reduced complexity.

[0165] If the network model after complexity reduction meets the preset complexity threshold, the preliminary hierarchical structure is divided according to the combination of convolutional layers and fully connected layers to obtain the preliminary hierarchical division results;

[0166] Based on the preliminary hierarchical division results, analyze its resolution ability to determine whether it meets the preset real-time optimization requirements;

[0167] If the preset real-time optimization requirements are not met, the optimized hierarchical division results are obtained by adjusting the division boundaries in the preliminary hierarchical division results;

[0168] Determine the optimized hierarchical division result as the final real-time optimization solution;

[0169] After obtaining the final real-time optimization plan, the rationality of the final hierarchical division results is confirmed by verifying the difficulty gradient of each level in the plan.

[0170] Specifically, in step S106, a convolutional neural network model to be optimized is obtained, wherein the model includes at least one convolutional layer and at least one fully connected layer, including a VGG network including 5 convolutional layers and 3 fully connected layers;

[0171] The convolutional layer is processed using a lightweight network compression technique, including using a pruning technique to reduce the number of parameters of the convolutional layer by 30%, thereby obtaining an optimized convolutional layer structure;

[0172] According to the optimized convolutional layer structure, the structure of the fully connected layer is adjusted, including reducing the number of neurons in the fully connected layer from 4096 to 1024, thus obtaining streamlined fully connected layer parameters;

[0173] We evaluated the computational complexity of the streamlined fully connected layer parameters, including floating-point operations (FLOPs), and found that the computational effort was reduced by 40%, resulting in a network model with reduced complexity.

[0174] If the network model after complexity reduction meets the preset complexity threshold, including FLOPs reduced to less than 50% of the original model, then the preliminary hierarchical structure is divided according to the combination of convolutional layers and fully connected layers, including dividing the network into input layer, feature extraction layer and classification layer, to obtain the preliminary hierarchical division results;

[0175] Based on the preliminary hierarchical classification results, the resolution capability was analyzed, including classification accuracy assessment. The accuracy was found to be above 95%, which was used to determine whether the preset real-time optimization requirements were met.

[0176] If the preset real-time optimization requirements are not met, including accuracy below 90%, the optimized hierarchical division results are obtained by adjusting the division boundaries in the preliminary hierarchical division results, including redefining the number of convolution kernels in the feature extraction layer;

[0177] Determine the optimized hierarchical division result as the final real-time optimization solution;

[0178] After obtaining the final real-time optimization plan, the rationality of the final hierarchical division result is confirmed by verifying the difficulty gradient of each level in the plan, including using the gradient descent method to evaluate the weight update speed of each layer.

[0179] like Figure 1-Figure 2 As shown, in step S107, the dense hierarchical division is iteratively adjusted through the real-time optimized hierarchical division scheme combined with the dynamic time window algorithm. If the division boundary fuzziness calculated based on the feature overlap rate or confidence is higher than the preset threshold, the contextual semantics or audio spectrum features are introduced for recalibration to obtain the final dense hierarchical division result.

[0180] Furthermore, in step S107, audio data is acquired and hierarchically divided into layers using a dynamic time window algorithm to obtain a preliminary dense hierarchical structure;

[0181] For the preliminary dense hierarchical structure, the feature overlap rate of each partition boundary is calculated to obtain the fuzziness evaluation value of each partition boundary;

[0182] Obtaining contextual semantic features of each partition boundary in the preliminary dense hierarchical structure; if the fuzziness evaluation value is higher than a preset threshold, adjusting the partition boundary according to the contextual semantic features to obtain an adjusted dense hierarchical structure;

[0183] Extracting spectral features of the audio data, and calibrating the adjusted dense hierarchical structure according to the spectral features to obtain a calibrated dense hierarchical structure;

[0184] For each partition boundary in the calibrated dense hierarchical structure, calculate its confidence and obtain the stability confirmation value;

[0185] According to the stability confirmation value, the window algorithm is used to iteratively update the real-time optimization solution to obtain the final dense hierarchical structure;

[0186] Extract the matching relationship between dense layers and real-time optimization from the final dense hierarchical structure and determine the dense layer division scheme;

[0187] Obtain a feature set, and use a microscopic difficulty perception model to scan the feature set layer by layer to obtain a feature level sequence after scanning;

[0188] By matching the microscopic clues in the feature hierarchy sequence with a preset gradient template, the preliminary change trend of semantic complexity or speech speed is determined. Based on the preliminary change trend, it is judged whether there is a significant fluctuation in the difficulty gradient to obtain the fluctuation detection result.

[0189] Specifically, in step S107, after obtaining the audio data, the audio data is hierarchically divided using a dynamic time window algorithm with an initial window size of 50 milliseconds to obtain a preliminary dense hierarchical structure;

[0190] For the preliminary dense hierarchical structure, the feature overlap rate of each partition boundary is calculated, and the cosine similarity algorithm is used. If the overlap rate is lower than 0.8, it is marked as a high fuzziness boundary, and the fuzziness evaluation value of each partition boundary is obtained;

[0191] Obtain the contextual semantic features of each partition boundary in the preliminary dense hierarchical structure. If the fuzziness evaluation value is higher than the preset threshold of 0.7, extract the contextual semantic features based on the BERT model and adjust the partition boundary to obtain the adjusted dense hierarchical structure.

[0192] Extracting spectral features of the audio data, extracting spectral features in the frequency range of 0-8000 Hz through fast Fourier transform, calibrating the adjusted dense hierarchical structure according to the spectral features, and obtaining a calibrated dense hierarchical structure;

[0193] For each partition boundary in the calibrated dense hierarchical structure, the support vector machine is used to calculate its confidence. If the confidence is lower than 0.9, it is marked as an unstable boundary to obtain the stability confirmation value.

[0194] Based on the stability confirmation value, the real-time optimization scheme is iteratively updated using a sliding window algorithm with a step size of 20 milliseconds to obtain the final dense hierarchical structure;

[0195] Extract the matching relationship between dense hierarchy and real-time optimization from the final dense hierarchy structure, divide the matching relationship into three categories using the K-means clustering algorithm, and determine the dense hierarchy division scheme;

[0196] Obtain a feature set, and use a microscopic difficulty perception model to scan the feature set layer by layer at 0.1 second intervals to obtain a feature level sequence after scanning;

[0197] The microscopic clues in the feature hierarchy sequence are matched by a preset gradient template, and the dynamic time warping algorithm is used to calculate the preliminary change trend of semantic complexity or speech rate. If the change exceeds 15%, it is marked as a significant fluctuation to obtain the fluctuation detection result.

[0198] like Figure 1-Figure 2 As shown, in step S108, according to the final dense hierarchical division result, a cross-modal fusion effect evaluation mechanism based on accuracy or recall rate is used. If the fusion accuracy is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated and recalculated to obtain an optimized hierarchical output of teaching content.

[0199] Furthermore, in step S108, the cross-modal fusion data in the dense hierarchical partitioning result is obtained, and the accuracy evaluation and recall evaluation are used to calculate the initial fusion accuracy value;

[0200] If the initial fusion accuracy value is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated, and the optimized feature extraction data is obtained;

[0201] The optimized feature extraction data is subjected to cross-modal fusion processing to obtain an updated fusion accuracy value;

[0202] Determine whether the updated fusion accuracy value is lower than the preset threshold. If so, repeat the adaptive convolution kernel parameter adjustment and feature extraction range update to obtain optimized feature extraction data again.

[0203] Based on the re-optimized feature extraction data, the cross-modal fusion results are recalculated to obtain a fusion accuracy value that meets the preset threshold;

[0204] Extract teaching content from the fusion accuracy value that meets the preset threshold, combine it with the dense hierarchical division structure, and generate preliminary hierarchical output data;

[0205] Format the preliminary hierarchical output data and adopt the hierarchical output mechanism to obtain the final optimized hierarchical output results of teaching content;

[0206] Obtain preliminary data of modal characteristics and obtain a quantitative representation of distribution differences through a preset characteristic distribution analysis method;

[0207] According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain the adjusted convolution kernel parameters.

[0208] Specifically, in step S108, the cross-modal fusion data in the dense hierarchical partitioning result is obtained, and the accuracy and recall evaluation are used to calculate the initial fusion accuracy value, including by comparing the matching degree of text and image, the accuracy is 85% and the recall is 78%;

[0209] If the initial fusion accuracy value is lower than the preset threshold of 90%, the adaptive convolution kernel parameters are adjusted and the feature extraction range is updated, including adjusting the convolution kernel size from 3x3 to 5x5 to obtain optimized feature extraction data;

[0210] The optimized feature extraction data is subjected to cross-modal fusion processing, including fusing text and image features through weighted averaging, to obtain an updated fusion accuracy value, which is calculated to be 92%;

[0211] Determine whether the updated fusion accuracy value is lower than the preset threshold. If so, repeat the adaptive convolution kernel parameter adjustment and feature extraction range update, including adjusting the convolution kernel size from 5x5 to 7x7, to obtain optimized feature extraction data again.

[0212] Based on the re-optimized feature extraction data, the cross-modal fusion results are recalculated, including dimensionality reduction of fusion features through principal component analysis, to obtain a fusion accuracy value that meets the preset threshold of 95%;

[0213] Extracting teaching content from the fusion accuracy values ​​that meet the preset threshold, combined with a dense hierarchical partitioning structure, including mapping the fusion features to knowledge graph nodes, to generate preliminary hierarchical output data;

[0214] Format the preliminary hierarchical output data and adopt a hierarchical output mechanism, including describing the hierarchical relationship of teaching content through XML structured language, to obtain the final optimized hierarchical output results of teaching content;

[0215] Obtain preliminary data on modal features, and use a preset feature distribution analysis method, including calculating the KL divergence of text and image features, to obtain a quantitative representation of the distribution difference of 0.15;

[0216] According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size, including adjusting the convolution kernel step size from 1 to 2, to obtain the adjusted convolution kernel parameters.

[0217] Those skilled in the art can make various other corresponding changes and deformations based on the technical solutions and concepts described above, and all of these changes and deformations should fall within the scope of protection of the claims of this application.

Claims

1. A multi-source teaching data processing method based on deep learning, characterized in that: The following steps are involved: Step S101: After acquiring multi-source teaching data, pre-process the text, audio, and video data using modality separation technology, extract independent feature sets from each modality, and obtain preliminary mapping results of modality feature distribution; Step S102: Based on the preliminary mapping results of the modal feature distribution, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size according to the differences in the modal feature distribution to determine the feature extraction range before cross-modal fusion; Step S103: Processing the adjusted feature extraction range through a multi-scale feature fusion network, fusing text semantic density, audio speech rate variation, and temporal correlation between video frames at different scales to obtain a unified high-dimensional feature space representation; Step S104: Based on the unified high-dimensional feature space representation, a noise suppression algorithm is used to filter data segments with large background noise or non-standard expression noise interference. If the noise detected exceeds a preset threshold, the corresponding feature weight is attenuated to obtain a clear feature set; Step S105: After obtaining a clear feature set, the feature set is scanned layer by layer using a micro-difficulty perception model to obtain a scanned feature hierarchy sequence. Micro-cues in the feature hierarchy sequence are matched using a preset gradient template to determine preliminary trends in semantic complexity or speech rate, thereby obtaining preliminary boundaries for the difficulty gradient. Step S106: Optimize the convolutional layer or fully connected layer using lightweight network compression technology based on the initial boundary of the difficulty gradient, thereby reducing computational complexity while maintaining resolution, and obtaining a hierarchical partitioning solution with real-time optimization. Step S107: Using a real-time optimized hierarchical partitioning scheme and a dynamic time window algorithm, the dense hierarchical partitioning is iteratively adjusted. If the fuzziness of the partitioning boundary calculated based on the feature overlap rate or confidence level is higher than a preset threshold, contextual semantics or audio spectrum features are introduced for recalibration to obtain the final dense hierarchical partitioning result. Step S108: Based on the final dense hierarchical division result, a cross-modal fusion effect evaluation mechanism based on accuracy or recall rate is used. If the fusion accuracy is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated and recalculated to obtain an optimized hierarchical output of teaching content.

2. The multi-source teaching data processing method based on deep learning according to claim 1 is characterized in that: In step S101, multi-source teaching data is obtained, and the multi-source teaching data includes text data, audio data and video data; Use modal separation technology to process multi-source teaching data to obtain text modal data, audio modal data and video modal data; For text modal data, a text feature extraction algorithm is used to obtain a text-independent feature set; For audio modal data, extract an audio independent feature set through audio data processing, and determine the feature distribution of the audio independent feature set; For video modal data, video feature extraction technology is used to extract video independent feature sets and determine the modal features of the video independent feature sets; Fusing text-independent feature sets, audio-independent feature sets, and video-independent feature sets to obtain a multimodal feature set; For the multimodal feature set, a clustering algorithm is used to obtain the modal feature distribution mapping result; Extract key features from the modal feature distribution mapping results and generate preliminary modal feature distribution results; Based on the preliminary modal feature distribution results, the feature space unification technology is used to obtain feature outputs suitable for downstream tasks.

3. The multi-source teaching data processing method based on deep learning according to claim 1 is characterized in that: In step S102, preliminary data of modal features are obtained, and a quantitative representation of distribution differences is obtained through a preset feature distribution analysis method; According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain the adjusted convolution kernel parameters; Based on the adjusted convolution kernel parameters, a feature extraction algorithm is used to extract feature data before cross-modal fusion from the modal features and determine the feature extraction range; Obtain data within the feature extraction range and generate a fusion feature set through a cross-modal fusion algorithm; If the modal consistency of the fused feature set is lower than the preset threshold, the convolution kernel size is readjusted according to the distribution difference to obtain the updated feature extraction range; Through the updated feature extraction range, the feature extraction algorithm is used to process the modal features to obtain the optimized fusion feature set; Based on the optimized fusion feature set, the consistency of cross-modal fusion features is judged through a preset verification method to obtain the final feature representation; Obtain cross-modal fusion data from the dense hierarchical partitioning results, use precision evaluation and recall evaluation to calculate the initial fusion accuracy value; If the initial fusion accuracy value is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated, and the optimized feature extraction data is obtained.

4. The multi-source teaching data processing method based on deep learning according to claim 1 is characterized in that: In step S103, multimodal data is acquired, where the multimodal data includes text data, audio data, and video data; Extracting text semantic density features from the text data; Extracting audio speech rate change features from the audio data; Extracting temporal correlation features between video frames from the video data; Fusing the text semantic density feature, the audio speech rate change feature, and the video frame temporal correlation feature to obtain a preliminary multi-scale feature representation; Using a convolutional neural network to process the text semantic density features to obtain an enhanced semantic density representation; fusing the enhanced semantic density representation and the audio speech rate variation feature, analyzing the speech rate variation sequence through a recurrent neural network to obtain a time-series enhanced semantic feature; Combining the temporal correlation features between video frames, an attention mechanism is used to process the relationship between frames to obtain a comprehensive spatiotemporal feature representation; generating a unified high-dimensional feature space representation through network processing based on the comprehensive spatiotemporal feature representation; Determine whether the feature distribution represented by the high-dimensional feature space meets a preset threshold, and if so, output a final feature representation; Through feature space unification technology, feature output suitable for downstream tasks is generated.

5. The multi-source teaching data processing method based on deep learning according to claim 1 is characterized in that: In step S104, input data is obtained and mapped into a high-dimensional feature set; Use noise suppression algorithms to detect background noise and interference data formed by non-standard expressions in high-dimensional feature sets and determine interference data segments; Extracting noise intensity from the interference data segment and determining whether the noise intensity exceeds a preset threshold; If it exceeds, the corresponding feature is marked as a feature to be processed; The weight of the feature to be processed is adjusted by the weight decay algorithm to obtain the adjusted feature set; Clustering algorithm is used to group the adjusted feature set to obtain clear feature subsets; Reconstruct the data segment based on the clear feature subset in the feature space to obtain the feature representation after filtering; Extract the final output from the filtered feature representation and determine whether the final output meets the preset clarity condition; If not, return to the noise suppression algorithm step and reprocess; Obtain preliminary data of modal characteristics and obtain a quantitative representation of distribution differences through a preset characteristic distribution analysis method; According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain the adjusted convolution kernel parameters.

6. The multi-source teaching data processing method based on deep learning according to claim 1 is characterized in that: In step S105, audio data is obtained and hierarchical division is performed on the audio data using a dynamic time window algorithm to obtain a preliminary dense hierarchical structure; For the preliminary dense hierarchical structure, the feature overlap rate of each partition boundary is calculated to obtain the fuzziness evaluation value of each partition boundary; Obtaining contextual semantic features of each partition boundary in the preliminary dense hierarchical structure; if the fuzziness evaluation value is higher than a preset threshold, adjusting the partition boundary according to the contextual semantic features to obtain an adjusted dense hierarchical structure; Extracting spectral features of the audio data, and calibrating the adjusted dense hierarchical structure according to the spectral features to obtain a calibrated dense hierarchical structure; For each partition boundary in the calibrated dense hierarchical structure, calculate its confidence and obtain the stability confirmation value; According to the stability confirmation value, the window algorithm is used to iteratively update the real-time optimization solution to obtain the final dense hierarchical structure; Extract the matching relationship between dense layers and real-time optimization from the final dense hierarchical structure and determine the dense layer division scheme; Obtain the feature set in the dense hierarchical partitioning scheme, scan the feature set layer by layer using the microscopic difficulty perception model, and obtain the scanned feature hierarchical sequence; By matching the microscopic clues in the feature hierarchy sequence with a preset gradient template, the initial trend of semantic complexity or speech rate change is determined, and the preliminary dividing boundary of the difficulty gradient is obtained.

7. The multi-source teaching data processing method based on deep learning according to claim 1, characterized in that: In step S106, a convolutional neural network model to be optimized is obtained, where the model includes at least one convolutional layer and at least one fully connected layer; The convolution layer is processed using a lightweight network compression technology to obtain an optimized convolution layer structure; According to the optimized convolutional layer structure, the structure of the fully connected layer is adjusted to obtain streamlined fully connected layer parameters; Evaluate the computational complexity of the streamlined fully connected layer parameters and obtain a network model with reduced complexity. If the network model after complexity reduction meets the preset complexity threshold, the preliminary hierarchical structure is divided according to the combination of convolutional layers and fully connected layers to obtain the preliminary hierarchical division results; Based on the preliminary hierarchical division results, analyze its resolution ability to determine whether it meets the preset real-time optimization requirements; If the preset real-time optimization requirements are not met, the optimized hierarchical division results are obtained by adjusting the division boundaries in the preliminary hierarchical division results; Determine the optimized hierarchical division result as the final real-time optimization solution; After obtaining the final real-time optimization plan, the rationality of the final hierarchical division results is confirmed by verifying the difficulty gradient of each level in the plan.

8. The multi-source teaching data processing method based on deep learning according to claim 1 is characterized in that: In step S107, audio data is obtained and hierarchical division is performed on the audio data using a dynamic time window algorithm to obtain a preliminary dense hierarchical structure; For the preliminary dense hierarchical structure, the feature overlap rate of each partition boundary is calculated to obtain the fuzziness evaluation value of each partition boundary; Obtaining contextual semantic features of each partition boundary in the preliminary dense hierarchical structure; if the fuzziness evaluation value is higher than a preset threshold, adjusting the partition boundary according to the contextual semantic features to obtain an adjusted dense hierarchical structure; Extracting spectral features of the audio data, and calibrating the adjusted dense hierarchical structure according to the spectral features to obtain a calibrated dense hierarchical structure; For each partition boundary in the calibrated dense hierarchical structure, calculate its confidence and obtain the stability confirmation value; According to the stability confirmation value, the window algorithm is used to iteratively update the real-time optimization solution to obtain the final dense hierarchical structure; Extract the matching relationship between dense layers and real-time optimization from the final dense hierarchical structure and determine the dense layer division scheme; Obtain a feature set, and use a microscopic difficulty perception model to scan the feature set layer by layer to obtain a feature level sequence after scanning; By matching the microscopic clues in the feature hierarchy sequence with a preset gradient template, the preliminary change trend of semantic complexity or speech speed is determined. Based on the preliminary change trend, it is judged whether there is a significant fluctuation in the difficulty gradient to obtain the fluctuation detection result.

9. The multi-source teaching data processing method based on deep learning according to claim 1, characterized in that: In step S108, cross-modal fusion data in the dense hierarchical partitioning result is obtained, and an initial fusion accuracy value is calculated using precision evaluation and recall evaluation; If the initial fusion accuracy value is lower than the preset threshold, the adaptive convolution kernel parameters are adjusted, the feature extraction range is updated, and the optimized feature extraction data is obtained; Perform cross-modal fusion processing on the optimized feature extraction data to obtain an updated fusion accuracy value; Determine whether the updated fusion accuracy value is lower than the preset threshold. If so, repeat the adaptive convolution kernel parameter adjustment and feature extraction range update to obtain optimized feature extraction data again. Based on the re-optimized feature extraction data, the cross-modal fusion results are recalculated to obtain a fusion accuracy value that meets the preset threshold; Extract teaching content from the fusion accuracy value that meets the preset threshold, and generate preliminary hierarchical output data by combining it with the dense hierarchical division structure; Format the preliminary hierarchical output data and adopt the hierarchical output mechanism to obtain the final optimized hierarchical output results of teaching content; Obtain preliminary data of modal characteristics and obtain a quantitative representation of distribution differences through a preset characteristic distribution analysis method; According to the quantitative representation of the distribution difference, an adaptive convolution kernel adjustment algorithm is used to dynamically adjust the convolution kernel size to obtain the adjusted convolution kernel parameters.

Images