A teaching style optimization method based on teaching quality feedback

Abstract

The application discloses a teaching style optimization method based on teaching quality feedback, comprising: capturing classroom environment data in real time based on a sensing device; decoupling the time-frequency domain characteristics of unstructured teaching behavior data based on an adaptive wavelet packet decomposition algorithm, and obtaining a multi-dimensional feature vector; constructing a teaching effect evaluation matrix to obtain a nonlinear mapping relationship between teaching behavior characteristics and a preset teaching target; generating a generative adversarial network to obtain a teaching behavior adjustment scheme, and verifying the effectiveness of the scheme based on a historical teaching case library; dynamically compensating the scheme based on an adjustment scheme, optimizing the scheme parameters according to secondary feedback data after the scheme is executed; and constructing a teaching style evolution map to form a traceable teaching style optimization trajectory for the iterative path of the teaching scheme. The application has the advantages that teaching strategies can be accurately adjusted, classroom interaction and teaching effect can be improved, and a dynamic optimization mechanism and a traceable evolution map help teachers to continuously improve the personalized teaching style.

Description

TECHNICAL FIELD

[0001] The present application relates to parameter optimization technology, in particular to a teaching style optimization method based on teaching quality feedback. BACKGROUND

[0002] With the diversification and individualization of education needs, the traditional "one-size-fits-all" teaching mode has been difficult to meet the learning needs of different students. The core goal of teaching style optimization is to design a teaching program that meets the individual needs of students according to their cognitive characteristics, learning styles and interest preferences. In recent years, with the progress of information technology, especially the application of artificial intelligence, big data and learning analysis technology, educators can more accurately understand students' learning behavior, performance and learning path, thus providing a scientific basis for the optimization of teaching style.

[0003] Current teaching style optimization methods on the market mainly focus on interactive teaching and blended learning. Interactive teaching emphasizes interaction between teachers and students, as well as between students, through question guidance, discussion and group cooperation to stimulate students' thinking and creativity, and enhance students' sense of participation and learning motivation. Blended learning combines the advantages of online and offline learning, combining traditional teaching and modern technology to ensure the interactivity of face-to-face teaching and the convenience and rich resources of online platforms to improve learning effectiveness. However, traditional teaching style optimization methods lack real-time feedback and flexible adjustment, making it difficult to meet the individual needs of students. In contrast, optimization methods based on teaching quality feedback can achieve precise adjustment through data-driven methods to improve teaching effectiveness. SUMMARY

[0004] To improve existing teaching style optimization methods, a teaching style optimization method based on teaching quality feedback is provided. This method captures multi-dimensional teaching behavior data in real time and uses advanced algorithms for analysis, enabling precise adjustment of teaching strategies and improving classroom interaction and teaching effectiveness.

[0005] To achieve the above objectives, the technical solution adopted by the present application is as follows:

[0006] A teaching style optimization method based on teaching quality feedback, characterized by comprising:

[0007] Real-time capture of teaching behavior data in the classroom environment, including voice tone spectrum features, board writing trajectory pressure values, and teacher-student interaction response time delays, based on sensing devices;

[0008] Decoupling of unstructured teaching behavior data in time-frequency domain features based on adaptive wavelet packet decomposition algorithm to obtain multi-dimensional feature vectors;

[0009] Based on the obtained multi-dimensional feature vector, a teaching effect evaluation matrix is constructed, and a nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target is obtained;

[0010] Based on the nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target, a generative adversarial network is generated, a teaching behavior adjustment scheme is obtained, and the effectiveness of the scheme is verified based on a historical teaching case library;

[0011] Based on the adjustment scheme dynamic compensation mechanism, the scheme parameters are optimized according to the secondary feedback data after the scheme is executed;

[0012] A teaching style evolution map is constructed to form a traceable teaching style optimization trajectory for the iteration path of the teaching scheme.

[0013] Preferably, the self-adaptive wavelet packet decomposition algorithm decouples the time-frequency domain characteristics of unstructured teaching behavior data to obtain a multi-dimensional feature vector, which specifically includes:

[0014] Based on the obtained teaching behavior data, a synchronization time sequence signal is obtained ;

[0015] Based on the signal characteristics, the wavelet basis is adaptively selected by minimizing the reconstruction error, and the formula is: ,

[0016] wherein, is a candidate wavelet basis function, is a candidate set of wavelet bases, , is the inner product of the signal and the wavelet basis, is the wavelet basis function with scale j and translation k, is the L2 norm;

[0017] Based on the Shannon entropy criterion, the decomposition layer L is adaptively determined, and the sub-band with energy ratio exceeding the threshold is retained;

[0018] The multi-dimensional feature vector specifically includes an emotional intensity index E, a knowledge transfer efficiency coefficient K, and a classroom attention concentration degree A;

[0019] Based on the energy and variance of the high-frequency sub-band, the emotional intensity index E is obtained, and the formula is: ,

[0020] wherein, is a set of high-frequency sub-bands, is the square of the L2 norm of the sub-band wavelet coefficient, is the variance, is the weight coefficient;

[0021] Based on the fundamental frequency stability and semantic information entropy of the teacher's voice, the knowledge transfer efficiency coefficient K is obtained, and the formula is: ,

[0022] wherein, is the teacher's voice fundamental frequency, is the semantic sentiment score, is the total length of the class, is the inverse of semantic information entropy, the lower the entropy, the more concentrated the knowledge delivery is;

[0023] Based on the student posture signal Low-frequency energy ratio and eye movement trajectory entropy to obtain classroom attention concentration A, the formula is: ,

[0024] wherein, is the low-frequency wavelet packet coefficient of the student posture signal , is the eye movement trajectory entropy, is the total fixation point number, is the weight parameter for balancing low-frequency energy and eye movement entropy.

[0025] Preferably, based on the obtained multi-dimensional feature vector, a teaching effect evaluation matrix is constructed to obtain a nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target, specifically including:

[0026] Based on the obtained multi-dimensional feature vector and the preset teaching target vector T, a teaching effect evaluation matrix is constructed: ,

[0027] wherein, is the Gaussian kernel function, which measures the nonlinear similarity between the multi-dimensional feature vector and the preset teaching target vector, is the preset teaching target weight vector, is the Gaussian kernel bandwidth, which controls the similarity decay rate;

[0028] Based on the deep neural network architecture, the mapping function from the multi-dimensional feature vector V to the preset teaching target vector T is obtained.

[0029] Preferably, the nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target is generated by the generative adversarial network, the teaching behavior adjustment scheme is obtained, and the effectiveness of the scheme is verified based on the historical teaching case library, specifically including:

[0030] Based on the historical teaching quality data in the historical case library, it is input into the generative adversarial network for alternating training, and the generator minimizes the generator loss, and the discriminator minimizes the discriminator loss.

[0031] The generator part adopts conditional variational autoencoder (CVAE), and the latent space embeds the teaching outline constraint condition;

[0032] The discriminator part includes a rule-based knowledge graph verification module for connecting subject knowledge ontology library to verify the logical consistency of teaching strategies, and a statistical feature comparison module for calculating the Wasserstein distance between the generated strategy and the excellent teaching case;

[0033] A gradient penalty term is added to the discriminator loss to obtain stable training results;

[0034] The multi-dimensional feature vector group under the optimal adjustment is obtained through iterative optimization of the generator;

[0035] Based on the generated adjustment scheme, the similarity between it and the historical teaching quality feedback higher case is calculated to obtain the effectiveness of the generated adjustment scheme.

[0036] Preferably, the similarity between the generated adjustment scheme and the historical teaching quality feedback higher case is calculated to obtain the effectiveness of the generated adjustment scheme, which specifically includes:

[0037] The similarity calculation formula is: ,

[0038] Wherein, the threshold value 0.8 represents high similarity, and the pass rate needs to be greater than 70%.

[0039] Referring to Figure 5 Based on the adjustment scheme dynamic compensation mechanism, the scheme parameters are optimized according to the secondary feedback data after the scheme is executed, which specifically includes:

[0040] Based on each generated adjustment scheme, real-time teaching quality feedback data and teaching behavior data are obtained, and the error vector between it and the teaching target vector is calculated ;

[0041] The compensation amount is calculated based on the PID control update formula, and the formula is: ,

[0042] Wherein, The size of the control error, The influence of the cumulative error, The control error change rate, The integral term of the error, which represents the cumulative error in the past, The error rate, which represents the rate of error;

[0043] The new parameter feature vector E, K, A obtained by calculating based on the PID control update formula is applied to the new teaching scheme to obtain secondary feedback data, and a new error vector between it and the teaching target vector is calculated ;

[0044] Based on the secondary feedback error vector Further adjust the teaching plan. If the error vector If it is small enough, stop adjusting; if it is still large, repeat the above process to further optimize the parameters.

[0045] Preferably, the construction of the teaching style evolution map to form a traceable teaching style optimization trajectory for the iterative path of the teaching plan specifically includes:

[0046] Based on each iteration of the teaching plan, data for each dimension is recorded;

[0047] Use a graph database to store nodes and edges. Nodes include teaching behaviors and student feedback, while edges include optimization paths and effect associations.

[0048] If the current solution improves students’ higher-order thinking, the weight of the edge will be increased; otherwise, the lesson plan will be refined and the solution parameters will be optimized;

[0049] Based on the iteration records, the timeline shows the style evolution, the heat map shows the effectiveness of the strategy, and the trajectory of each style optimization is drawn;

[0050] The decision basis for a specific iteration is located through SPARQL queries, and an early warning is triggered when the solution deviates from the historical optimal range.

[0051] Compared with the prior art, the advantages of the present invention are:

[0052] By comprehensively utilizing multiple sensor devices to capture real-time classroom teaching behavior data, a comprehensive and accurate reflection of teachers' teaching activities and classroom effectiveness can be achieved. By decoupling the time-frequency domain features of these unstructured data using an adaptive wavelet packet decomposition algorithm, multidimensional feature vectors can be extracted, which are then used to construct a teaching effectiveness evaluation matrix. This clarifies the nonlinear mapping relationship between teaching behavior and pre-set teaching objectives. A generative adversarial network is then used to design a teaching behavior adjustment plan, which is then validated against a database of historical teaching cases. This secondary feedback data optimization, based on a dynamic compensation mechanism, enables continuous iteration and adjustment of the plan, improving its accuracy and flexibility. Finally, by constructing a teaching style evolution graph and forming a traceable optimization path, continuous evolution and personalized adjustment of teaching style are achieved. This approach can refine teaching effectiveness, optimize teaching strategies, and provide teachers with more scientific, data-driven teaching improvement plans. It offers significant advantages, particularly in enhancing the quality of classroom interaction and personalized teaching. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of the method proposed by the present invention;

[0054] Figure 2A multi-dimensional feature vector acquisition schematic diagram of the method proposed in the application;

[0055] Figure 3 A nonlinear mapping relationship schematic diagram of the method proposed in the application;

[0056] Figure 4 A scheme adjustment schematic diagram of the method proposed in the application;

[0057] Figure 5 A scheme parameter optimization schematic diagram of the method proposed in the application;

[0058] Figure 6 A scheme management schematic diagram of the method proposed in the application. DETAILED DESCRIPTION

[0059] The following description is used to disclose the application so that those skilled in the art can implement the application. The preferred embodiments in the following description are only used as examples, and other obvious variants can be thought of by those skilled in the art.

[0060] Referring to Figure 1 As shown in the figure, a teaching style optimization method based on teaching quality feedback includes:

[0061] Step one: based on a sensing device, real-time capture of teaching behavior data in a classroom environment, including voice tone spectrum features, blackboard writing trajectory pressure values, and teacher-student interaction response time delays;

[0062] Step two: based on an adaptive wavelet packet decomposition algorithm, time-frequency domain feature decoupling of unstructured teaching behavior data, to obtain a multi-dimensional feature vector;

[0063] Step three: based on the obtained multi-dimensional feature vector, construct a teaching effectiveness evaluation matrix, and obtain a nonlinear mapping relationship between teaching behavior features and preset teaching goals;

[0064] Step four: based on the nonlinear mapping relationship between teaching behavior features and preset teaching goals, generate a generative adversarial network, obtain a teaching behavior adjustment scheme, and verify the effectiveness of the scheme based on a historical teaching case library;

[0065] Step five: based on an adjustment scheme dynamic compensation mechanism, optimize scheme parameters according to secondary feedback data after scheme execution;

[0066] Step six: construct a teaching style evolution map to form a traceable teaching style optimization trajectory for the iteration path of the teaching scheme.

[0067] Referring to Figure 2 As shown in the figure, based on an adaptive wavelet packet decomposition algorithm, time-frequency domain feature decoupling of unstructured teaching behavior data, to obtain a multi-dimensional feature vector specifically includes:

[0068] Based on the acquired teaching behavior data, a synchronization timing signal is acquired ;

[0069] Based on the signal characteristics, a wavelet basis is adaptively selected by minimizing the reconstruction error, and the formula is: ,

[0070] Wherein, is a candidate wavelet basis function, is a candidate set of wavelet bases, , is the inner product of the signal and the wavelet basis, is a wavelet basis function with scale j and translation k, is the L2 norm;

[0071] Based on the Shannon entropy criterion, the decomposition layer number L is adaptively determined, and the sub-band with energy ratio exceeding the threshold is retained;

[0072] The multi-dimensional feature vector specifically includes an emotional intensity index E, a knowledge transfer efficiency coefficient K, and a classroom attention concentration degree A;

[0073] Based on the energy and variance of the high-frequency sub-band, the emotional intensity index E is obtained, and the formula is: ,

[0074] Wherein, is a set of high-frequency sub-bands, is the square of the L2 norm of the sub-band wavelet coefficient, is the variance, is the weight coefficient;

[0075] Based on the fundamental frequency stability and semantic information entropy of the teacher's voice, the knowledge transfer efficiency coefficient K is obtained, and the formula is: ,

[0076] Wherein, is the fundamental frequency of the teacher's voice, is the semantic sentiment score, is the total length of the class, is the inverse of the semantic information entropy, and the lower the entropy, the more concentrated the knowledge transfer is;

[0077] Based on the low-frequency energy ratio of the student posture signal and the eye movement trajectory entropy, the classroom attention concentration degree A is obtained, and the formula is: ,

[0078] Wherein, is the low-frequency wavelet packet coefficient of the student posture signal , is the eye movement trajectory entropy, is the total number of fixation points, Balancing the low-frequency energy and the weight parameter of the eye movement entropy.

[0079] It can be understood that the adaptive determination of the decomposition layer number L depends on the Shannon entropy criterion, but different signal characteristics can cause large changes in the entropy value, thereby affecting the selection of the L value, causing insufficient or excessive feature decoupling. Multi-scale analysis can be introduced, combined with the energy distribution, variance and other characteristics of the signal, and appropriate constraints can be introduced to ensure that the selection of the L value is reasonable and to avoid excessive decomposition or insufficient decomposition.

[0080] At the same time, the energy and variance of the high-frequency sub-band, the proportion of low-frequency energy, and the eye movement entropy may be easily affected by noise or environmental factors, leading to unstable calculation, especially in complex environments. Through multi-modal data such as heart rate and expression recognition, comprehensive monitoring of student status is increased to ensure that the attention concentration evaluation is more accurate.

[0081] Referring to Figure 3 Based on the obtained multi-dimensional feature vector, a teaching effect evaluation matrix is constructed, and a nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target is obtained, which specifically includes:

[0082] Based on the obtained multi-dimensional feature vector and the preset teaching target vector T, a teaching effect evaluation matrix is constructed: ,

[0083] wherein, is a Gaussian kernel function, which measures the nonlinear similarity between the multi-dimensional feature vector and the preset teaching target vector, is a preset teaching target weight vector, is a Gaussian kernel bandwidth, which controls the similarity decay rate;

[0084] Based on the deep neural network architecture, a mapping function of the multi-dimensional feature vector V to the preset teaching target vector T is obtained ;

[0085] Specifically, the deep neural network architecture includes an input layer, multiple hidden layers and an output layer. The input layer receives the learning feature vector, and the output layer predicts the teaching target vector. A suitable loss function (such as mean square error or cross entropy) is selected to measure the error between the predicted output and the actual teaching target. A suitable loss function (such as mean square error or cross entropy) is selected to measure the error between the predicted output and the actual teaching target. The trained neural network is evaluated using a validation set to check the generalization ability and prediction accuracy of the model. According to the training result, the achievement of the student in each teaching target dimension is output, and the teaching effect evaluation result is generated.

[0086] Referring to Figure 4As shown, the nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target is used to generate a generative adversarial network, and a teaching behavior adjustment scheme is obtained, and the effectiveness of the scheme is verified based on a historical teaching case library, and the specific steps include:

[0087] Based on the historical teaching quality data in the historical case library, the data is input into the generative adversarial network for alternating training, and the generator minimizes the generator loss, and the discriminator minimizes the discriminator loss.

[0088] The generator part adopts a conditional variational autoencoder (CVAE), and the latent space embeds the teaching program constraint condition.

[0089] The discriminator part includes a rule-based knowledge graph verification module for connecting a subject knowledge ontology library to verify the logical consistency of the teaching strategy, and a statistical feature comparison module for calculating the Wasserstein distance between the generated strategy and the excellent teaching case.

[0090] A gradient penalty term is added to the discriminator loss to obtain stable training results.

[0091] A multi-dimensional feature vector group under the optimal adjustment is obtained through iterative optimization of the generator.

[0092] Based on the generated adjustment scheme, the similarity between the generated adjustment scheme and the historical teaching quality feedback is calculated to obtain the effectiveness of the generated adjustment scheme.

[0093] Based on the generated adjustment scheme, the similarity between the generated adjustment scheme and the historical teaching quality feedback is calculated to obtain the effectiveness of the generated adjustment scheme, and the specific steps include:

[0094] The similarity calculation formula is: ,

[0095] Wherein, the threshold value 0.8 represents a high similarity, and the pass rate needs to be greater than 70%.

[0096] Referring to Figure 5 As shown, based on the adjustment scheme dynamic compensation mechanism, the scheme parameters are optimized according to the secondary feedback data after the scheme is executed, and the specific steps include:

[0097] Based on each generated adjustment scheme, real-time teaching quality feedback data and teaching behavior data are obtained, and an error vector between the teaching target vector and the error vector is calculated .

[0098] The compensation amount is calculated based on the PID control update formula, and the formula is: ,

[0099] Wherein, The size of the control error, The influence of the cumulative error, The control error change rate, is the integral term of error, representing the accumulation of past errors, is the rate of change of error, representing the rate of error;

[0100] Based on the PID control update formula, the new parameter feature vector E, K, A obtained is applied to the new teaching plan to obtain secondary feedback data, and a new error vector between the teaching target vector and the secondary feedback data is calculated ;

[0101] Based on the secondary feedback error vector Further adjustment is made to the teaching plan. If the error vector is small enough, the adjustment is stopped; if it is still large, the above process is repeated to further optimize the parameters.

[0102] Referring to Figure 6 , a teaching style evolution map is constructed to form a traceable teaching style optimization trajectory for the iteration path of the teaching plan. Specifically, it includes:

[0103] Based on each iteration of the teaching plan, data of each dimension is recorded;

[0104] A graph database is used to store nodes and edges, including teaching behaviors and student feedback, and edges including optimization paths and effect correlations;

[0105] If the current plan improves the students' high-order thinking, the weight of the edge is increased, otherwise the teaching plan is refined and the plan parameters are optimized;

[0106] Based on the iteration records, the style evolution is displayed on the time axis and the strategy effectiveness is displayed on the heat map to draw a trajectory graph of each style optimization;

[0107] Through SPARQL query, the decision basis of a specific iteration is located, and when the plan deviates from the historical optimal interval, an early warning is triggered.

[0108] Specifically, in each iteration of the teaching plan, first, the data of each dimension is recorded, including students' classroom performance, participation, emotional response, and academic achievement, etc. All data are stored in the graph database in the form of nodes, each node representing a teaching behavior or student feedback. The edge represents the optimization path and effect correlation between different teaching plans. When a certain plan effectively improves the students' high-order thinking ability, the weight of the related edge in the graph database will increase, indicating the effectiveness of the plan in this field. Each refinement records the specific adjustment path and improvement effect through the update of nodes and edges in the graph database, providing data support and decision basis for subsequent teaching plan optimization.

[0109] It should be noted that the above-mentioned embodiments of the application are merely intended to describe the application and are not intended to limit the application. In addition, the above-mentioned embodiments of the application are described in the order of the description. The order of the embodiments is not intended to represent the advantages of the embodiments. In some embodiments, multi-task processing and parallel processing are possible or advantageous.

[0110] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments.

[0111] The above-mentioned embodiments are only the preferred embodiments of the application, and are not intended to limit the application. Any modification, equivalent replacement, improvement, etc. made within the principles of the application shall be included in the protection scope of the application.

Claims

1. A teaching style optimization method based on teaching quality feedback, characterized by, The application relates to a teaching behavior adjustment method based on a generative adversarial network (GAN) and a teaching behavior adjustment system. The application comprises the following steps: Real-time capturing of teaching behavior data in a classroom environment based on a sensing device, including the spectral characteristics of speech intonation, the writing trajectory pressure value of blackboard writing, and the response time delay of teacher-student interaction; Decoupling of time-frequency domain features of unstructured teaching behavior data based on an adaptive wavelet packet decomposition algorithm to obtain a multi-dimensional feature vector; Based on the obtained multi-dimensional feature vector, a teaching effect evaluation matrix is constructed to obtain a nonlinear mapping relationship between the teaching behavior characteristics and the preset teaching target. Specifically, based on the obtained multi-dimensional feature vector and the preset teaching target vector T, a teaching effect evaluation matrix is constructed as follows: , wherein, is a Gaussian kernel function, measuring the non-linear similarity of the multi-dimensional feature vector and the preset teaching target vector, is a preset teaching target weight vector, is a Gaussian kernel bandwidth, controlling the similarity decay rate; Based on the deep neural network architecture, a mapping function of a multi-dimensional feature vector V to a preset teaching target vector T is obtained ; The multi-dimensional feature vector specifically comprises an emotional intensity index E, a knowledge transmission efficiency coefficient K, and a classroom attention concentration degree A; Generation of a GAN based on a nonlinear mapping relationship between teaching behavior features and preset teaching goals to obtain a teaching behavior adjustment scheme, and verification of the effectiveness of the scheme based on a historical teaching case library; Based on each generated adjustment scheme, real-time teaching quality feedback data and teaching behavior data are obtained, and an error vector between the teaching quality feedback data and the teaching target vector is calculated ; The compensation amount is calculated based on a PID control update formula, which is: , wherein, controlling the size of the error, accumulating the effect of the error, controlling the rate of change of the error, is an integral term of the error, representing the accumulation of past errors, is a rate term of the error, representing the rate of the error; Based on the PID control update formula, the new parameter characteristic vector E, K, A is obtained, which is applied to the new teaching scheme to obtain secondary feedback data, and a new error vector between the secondary feedback data and the teaching target vector is calculated ; Based on a quadratic feedback error vector The teaching scheme is further adjusted, if the error vector is small enough, the adjustment is stopped; if it is still large, the above process is repeated to further optimize the parameters; Dynamic compensation mechanism based on the adjustment scheme, optimization of scheme parameters according to secondary feedback data after scheme execution, specifically comprising the following steps: 2.The teaching style optimization method based on teaching quality feedback of claim 1, wherein, Construction of a teaching style evolution map to form a traceable teaching style optimization trajectory for the iterative path of the teaching scheme. Based on the acquired teaching behavior data, a synchronization timing signal is acquired ; Based on the signal characteristics, the wavelet basis is adaptively selected by minimizing the reconstruction error, which is given by , wherein, is a candidate wavelet basis function, is a candidate set of wavelet basis functions, , is an inner product of a signal and a wavelet basis, is a wavelet basis function at scale j and translation k, is an L2 norm; The step of decoupling time-frequency domain features of unstructured teaching behavior data based on an adaptive wavelet packet decomposition algorithm to obtain a multi-dimensional feature vector specifically comprises the following steps: The emotional intensity index E is obtained based on the energy and variance of the high-frequency sub-band, and the formula is: , wherein is a set of high frequency subbands, is the L2 norm square of the subband wavelet coefficients, is the variance, is a weight coefficient; The knowledge transmission efficiency coefficient K is obtained based on the fundamental frequency stability of the teacher voice and semantic information entropy, and the formula is: , wherein, is the teacher's voice fundamental frequency, is the semantic sentiment score, is the total length of the class, is the inverse of the semantic information entropy, the lower the entropy, the more concentrated the knowledge transfer; Based on student posture signals The low-frequency energy proportion and eye movement trajectory entropy of the student posture signals are used to obtain the degree of concentration A in the classroom, and the formula is: , wherein, is the low-frequency wavelet packet coefficient of the student posture signal is the low-frequency wavelet packet coefficient of the student posture signal is the eye movement trajectory entropy, is the total fixation point number, is the weight parameter balancing the low-frequency energy and the eye movement entropy. 3.The teaching style optimization method based on teaching quality feedback of claim 1, wherein, Adaptive determination of the decomposition layer number L based on a Shannon entropy criterion, and reservation of sub-bands with an energy ratio exceeding a threshold value; The step of generation of a GAN based on a nonlinear mapping relationship between teaching behavior features and preset teaching goals to obtain a teaching behavior adjustment scheme, and verification of the effectiveness of the scheme based on a historical teaching case library specifically comprises the following steps: Input of historical teaching quality data in the historical case library into the GAN for alternating training, minimization of a generator loss by a generator, and minimization of a discriminator loss by a discriminator; The generator part adopts a conditional variational autoencoder, and a latent space of the conditional variational autoencoder is embedded with a teaching outline constraint condition; The discriminator part comprises a rule-based knowledge graph verification module for connecting a subject knowledge ontology library to verify the logical consistency of a teaching strategy, and a statistical feature comparison module for calculating the Wasserstein distance between the generated strategy and an excellent teaching case; A gradient penalty term is added to the discriminator loss to obtain a stable training result; Optimization of the multi-dimensional feature vector group under the optimal adjustment through the generator; 4.The teaching style optimization method based on teaching quality feedback of claim 3, wherein, Calculation of the similarity between the generated adjustment scheme and historical teaching cases to obtain the effectiveness of the generated adjustment scheme. The step of calculation of the similarity between the generated adjustment scheme and historical teaching cases to obtain the effectiveness of the generated adjustment scheme specifically comprises the following steps: 。 5.The teaching style optimization method based on teaching quality feedback of claim 1, wherein, The similarity calculation formula is: The step of construction of a teaching style evolution map to form a traceable teaching style optimization trajectory for the iterative path of the teaching scheme specifically comprises the following steps: Data recording of each dimension based on each teaching scheme iteration; A graph database is used to store nodes and edges, the nodes comprising teaching behavior and student feedback, and the edges comprising optimization paths and effect correlations; If the current scheme improves the student's high-order thinking, the weight of the edge is increased, otherwise, the scheme parameter optimization is triggered by triggering the teaching plan refinement; Based on the iteration record, a time axis displays the style evolution, a heat map displays the strategy effectiveness, and a trajectory graph of each style optimization is drawn; A specific iteration decision basis is located through SPARQL query, and a warning is triggered when the scheme deviates from the historical optimal interval.

Images