Intelligent classroom interactive teaching software and teaching system

Abstract

The present application relates to the technical field of teaching management, and discloses a kind of smart classroom interactive teaching software and teaching system, comprising: multi-modal data acquisition module, regional feature analysis module, cognitive bias prediction module and teaching strategy scheduling module;Multi-modal data acquisition module obtains the behavior characteristic sequence in classroom environment and the teaching instruction sequence containing knowledge point weight;Regional feature analysis module divides behavior characteristic sequence into regional behavior matrix;Cognitive bias prediction module is aligned sequence by time regularization logic, determines the ratio of response time delay difference of feedback relative to instruction and knowledge point weight, to obtain cognitive transmission factor and generate bias transmission prediction graph;Teaching strategy scheduling module generates teaching intervention signal according to bias transmission prediction graph, the present application quantifies the transmission law of cognitive bias to back sequence node, realizes the evaluation of teaching administrative management liability, and improves the management efficiency of teaching resource scheduling.

Description

Technical Field

[0001] This invention relates to the field of teaching management technology, specifically to a smart classroom interactive teaching software and teaching system. Background Technology

[0002] Current smart classroom management technologies collect audio data from teachers and image data from students, and use multimodal feature analysis to build evaluation models, providing data support for the quantitative supervision of the teaching process. In a normalized, large-scale centralized teaching environment, the propagation of signals in physical space and the audience's response feedback follow a topological evolution law of diffusion from the center to the periphery. Current solutions usually simplify the physical teaching space into homogeneous logical nodes, assuming that the transmission of teaching instructions and behavioral responses are completely synchronized on the time axis. This neglect of the heterogeneous physical attributes of space causes the supervision model to produce logical deviations in the judgment benchmark when dealing with the objective time delay caused by different seating arrangements. Using methods such as increasing the video sampling frequency or adding linear alignment methods makes it difficult to distinguish between normal physical propagation resistance and the actual cognitive disconnect. Such methods trigger a non-linear increase in computing power resource requirements, and because they cannot isolate the background noise generated by the spatial topology, they generate a large number of abnormal alarms, reducing the efficiency of administrative intervention resource allocation.

[0003] The teaching process is a highly logically dependent, sequential flow. The mastery of prior knowledge points directly affects the transmission resistance of subsequent business units. Existing monitoring models focus on threshold judgment of transient behavioral characteristics, lacking quantitative assessment of the transmission of cognitive biases downstream along the logical chain of the teaching syllabus. This lagging monitoring mechanism causes minor deviations to accumulate and amplify at business nodes, ultimately inducing group management risks. Control strategies mostly focus on the physical coverage of hardware acquisition terminals, and the software control logic model construction is not adaptable to complex teaching scenarios. For example, Chinese invention patent CN118071833B discloses a method and system for multimodal positioning of target objects based on a three-dimensional digital classroom. This scheme achieves static binding of student identity and seat through geometric mapping of visual features and physical coordinates, improving the accuracy of attendance and location verification. However, this scheme presupposes an idealized signal propagation environment, ignores the objective physical diffusion phase difference of sound and light signals in non-homogeneous spaces, lacks a propagation damping compensation mechanism in the judgment logic, and is prone to misjudging normal physical response delays as cognitive disconnect when processing feedback from physically distant audiences, inducing misleading scheduling of computing resources. Audience behavior analysis is limited to transient isolated posture feature comparison, and no evolution model covering the logical dependencies of the teaching syllabus has been established. It is unable to predict the cumulative liabilities and risks of cognitive biases throughout the teaching process, and the intervention decision-making in the face of dynamically evolving teaching processes exhibits obvious lag and blindness.

[0004] Therefore, how to compensate for spatial propagation delay based on the topological heterogeneity of physical seating, and combine business logic dependency quantification to measure global execution liability, thereby achieving prediction of the teaching process and optimized scheduling of computing resources, becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention proposes a smart classroom interactive teaching software and system, comprising: The multimodal data acquisition module is used to acquire multimodal behavioral feature sequences and teaching instruction sequences in the classroom environment. The teaching instruction sequence includes multiple teaching instructions and the knowledge point weights corresponding to each teaching instruction. The region feature parsing module is used to divide the multimodal behavioral feature sequence into multiple region behavioral matrices based on the coordinate parameters of the acquisition end; The cognitive bias prediction module is used to align the multimodal behavioral feature sequence and the teaching instruction sequence using the dynamic time warping algorithm, calculate the response delay difference between the behavioral feedback reflected by the multimodal behavioral feature sequence and the teaching instruction sequence, and determine the ratio of the response delay difference to the knowledge point weight of the corresponding teaching instruction to obtain the cognitive transmission influence factor. Based on the logical dependency relationship of the teaching instruction sequence, it generates a cognitive bias transmission prediction map containing the cumulative amount of cognitive bias. The teaching strategy scheduling module is used to generate teaching intervention signals for specific abnormal areas based on the deviation distribution in the cognitive deviation transmission prediction map.

[0006] Preferably, the cognitive bias prediction module uses the Kalman filter algorithm to predict the evolution trajectory of the multimodal behavioral feature sequence. When the multimodal behavioral feature sequence is interrupted by environmental occlusion, resulting in transient loss, interpolation features are generated based on the state transition matrix corresponding to the evolution trajectory. The interpolation features are used to complete the multimodal behavioral feature sequence, so that the data dimension of the input cognitive bias transmission prediction map remains constant during the teaching cycle.

[0007] Preferably, the multimodal behavioral feature sequence acquired by the multimodal data acquisition module includes: posture parameters in the visual modality, feedback signal features in the audio modality, and response latency in the interactive modality.

[0008] Preferably, the cognitive bias prediction module is used to map the unstructured regional behavior matrices to the teaching quality assessment dimension using high-dimensional mapping logic, and calculate the logical deviation of each regional behavior matrix relative to the teaching instruction sequence.

[0009] Preferably, the teaching intervention signals generated by the teaching strategy scheduling module include: weight adjustment parameters for teaching resources in specific abnormal areas and progress retracing instructions for specific teaching paths.

[0010] Preferably, the cognitive bias prediction module establishes a teaching instruction association model. The teaching instruction association model is used to define the transmission influence function between the preceding basic instruction and the subsequent derived instruction. The transmission influence function is used to predict the management impact value of the cumulative amount of cognitive bias on the overall teaching progress.

[0011] Preferably, the teaching strategy scheduling module also includes closed-loop feedback logic, which is used to dynamically correct the prediction gain coefficient of the cognitive bias transmission prediction map based on the real-time status of the regional behavior matrix after the teaching intervention signal is output.

[0012] Preferably, when performing the division, the regional feature analysis module is also used to establish an association matrix between regions. The association matrix is ​​used to characterize the state coupling degree between the behavior matrices of adjacent regions, so as to assist the teaching strategy scheduling module in determining the diffusion characteristics of abnormal regions.

[0013] Preferably, the system also includes a central processing cluster, which stores a cognitive bias assessment model trained by deep learning. The cognitive bias assessment model is used to support the cognitive bias prediction module in generating a cognitive bias transmission prediction map.

[0014] The beneficial effects of this invention are: 1. In the smart classroom interactive teaching software, the system processes group behavior data through a spatial hierarchical association mechanism. By utilizing the phase difference in the response to teaching instructions between different spatial levels, it can accurately distinguish between physical propagation delay and the disconnect between group cognition. This topological phase-locked gradient evaluation method transforms the objective signal diffusion law in physical space into a benchmark scale for supervision and judgment. Thus, when determining the management status, it can identify and eliminate normal behavioral lags caused by differences in student seating. Since the system only generates abnormal signals when the phase difference gradient breaks nonlinearly, it eliminates false alarms caused by physical environmental noise in large-scale teaching conditions, ensuring the accuracy of management intervention instructions.

[0015] 2. The system utilizes a directed acyclic graph structure built based on the semantic dependencies of the teaching syllabus to achieve a logical upgrade from transient behavior correction to full-link management and liability prediction. This mechanism no longer evaluates the deviation value of the current time slice in isolation, but combines business logic dependency weights to calculate the expected liability accumulation of the current cognitive deviation on subsequent related knowledge nodes. When a small deviation is detected to have global transmission and destructive potential, the strategy scheduling unit triggers intervention arbitration and generates management messages in advance. This feedforward monitoring mode based on the logical execution chain effectively intercepts the transmission and amplification of hidden management risks in the business process, avoids the collapse of later administrative execution due to the accumulation of early deviations, and enhances the stability of the implementation of teaching plans.

[0016] 3. The system improves the efficiency of administrative resource scheduling in large-scale concurrent environments through the synergistic effect of cross-modal predictive analysis and spatial topology decomposition. When extracting the temporal tensor of group behavior state, the system orthogonally decomposes the data into multiple sub-region behavior tensors according to the spatial coordinates of the physical acquisition terminal. The strategy scheduling unit then uses the deviation extreme value distribution law in the predictive feature map to achieve targeted intervention for specific abnormal areas. Since the system can accurately lock the topological sub-nodes where gradient breakage occurs, subsequent resource scheduling and high-frequency sampling can be highly focused on local abnormal areas, thereby avoiding the waste of computing power caused by performing indiscriminate high-frequency calculations on the entire data stream. This allows the system to maintain low energy consumption and high response speed when processing large-scale business. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a diagram illustrating the architecture of the smart classroom interactive teaching system with multimodal features according to the present invention. Figure 2 This is a diagram showing the internal logic execution path for the cognitive bias prediction and quantitative evaluation of this invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] A smart classroom interactive teaching software and system includes: The multimodal data acquisition module is used to acquire multimodal behavioral feature sequences and teaching instruction sequences in the classroom environment. The teaching instruction sequence includes multiple teaching instructions and the knowledge point weights corresponding to each teaching instruction. The region feature parsing module is used to divide the multimodal behavioral feature sequence into multiple region behavioral matrices based on the coordinate parameters of the acquisition end; The cognitive bias prediction module is used to align the multimodal behavioral feature sequence and the teaching instruction sequence using the dynamic time warping algorithm, calculate the response delay difference between the behavioral feedback reflected by the multimodal behavioral feature sequence and the teaching instruction sequence, and determine the ratio of the response delay difference to the knowledge point weight of the corresponding teaching instruction to obtain the cognitive transmission influence factor. Based on the logical dependency relationship of the teaching instruction sequence, it generates a cognitive bias transmission prediction map containing the cumulative amount of cognitive bias. The teaching strategy scheduling module is used to generate teaching intervention signals for specific abnormal areas based on the deviation distribution in the cognitive deviation transmission prediction map.

[0021] Preferably, the cognitive bias prediction module uses the Kalman filter algorithm to predict the evolution trajectory of the multimodal behavioral feature sequence. When the multimodal behavioral feature sequence is interrupted by environmental occlusion, resulting in transient loss, interpolation features are generated based on the state transition matrix corresponding to the evolution trajectory. The interpolation features are used to complete the multimodal behavioral feature sequence, so that the data dimension of the input cognitive bias transmission prediction map remains constant during the teaching cycle.

[0022] Preferably, the multimodal behavioral feature sequence acquired by the multimodal data acquisition module includes: posture parameters in the visual modality, feedback signal features in the audio modality, and response latency in the interactive modality.

[0023] Preferably, the cognitive bias prediction module is used to map the unstructured regional behavior matrices to the teaching quality assessment dimension using high-dimensional mapping logic, and calculate the logical deviation of each regional behavior matrix relative to the teaching instruction sequence.

[0024] Preferably, the teaching intervention signals generated by the teaching strategy scheduling module include: weight adjustment parameters for teaching resources in specific abnormal areas and progress retracing instructions for specific teaching paths.

[0025] Preferably, the cognitive bias prediction module establishes a teaching instruction association model. The teaching instruction association model is used to define the transmission influence function between the preceding basic instruction and the subsequent derived instruction. The transmission influence function is used to predict the management impact value of the cumulative amount of cognitive bias on the overall teaching progress.

[0026] Preferably, the teaching strategy scheduling module also includes closed-loop feedback logic, which is used to dynamically correct the prediction gain coefficient of the cognitive bias transmission prediction map based on the real-time status of the regional behavior matrix after the teaching intervention signal is output.

[0027] Preferably, the cognitive bias prediction module uses the following quantification logic when calculating the cognitive transmission influencing factors: ,in, As a factor influencing cognitive transmission, L represents the difference in response delay between the aligned instruction sequence and the behavioral feedback, where L is the weight of the knowledge point corresponding to the instruction in the instruction sequence.

[0028] Preferably, when performing the division, the regional feature analysis module is also used to establish an association matrix between regions. The association matrix is ​​used to characterize the state coupling degree between the behavior matrices of adjacent regions, so as to assist the teaching strategy scheduling module in determining the diffusion characteristics of abnormal regions.

[0029] Preferably, the system also includes a central processing cluster, which stores a cognitive bias assessment model trained by deep learning. The cognitive bias assessment model is used to support the cognitive bias prediction module in generating a cognitive bias transmission prediction map.

[0030] Example 1: In this smart classroom interactive teaching software and system deployed in a lecture hall physical environment accommodating 300 audience members, the audio and visual information emitted by the teaching end has a physical propagation delay when it spreads to the back rows of physical seats. The audience's internalization feedback on specific concepts is constrained by the differences in group cognitive cascade. Traditional monitoring systems divide the physical space into homogeneous data nodes and use a unified fixed time window to compare the front-end teaching instructions with the back-end group behavior. This results in the back-row audience's physical response being delayed, which is judged as cognitive disconnect and triggers administrative intervention instructions. The teaching process is an administrative planning execution chain with pre-dependent relationships. The cognitive deviation of the basic concept nodes is transmitted to the downstream derived concept nodes along the knowledge topology graph. Conventional systems are limited by the feature comparison of transient discrete behaviors and cannot quantify the deviation accumulation process that has not reached the alarm threshold, causing administrative intervention to lag behind the business deviation node and consume management computing power.

[0031] The multimodal data acquisition module acquires multimodal behavioral feature sequences and teaching instruction sequences within the classroom environment. The teaching instruction sequence includes multiple teaching instructions and their corresponding knowledge point weights. The regional feature analysis module divides the multimodal behavioral feature sequences into multiple regional behavioral matrices based on the coordinate parameters of the acquisition end, thereby restoring the topological gradient of the underlying data stream in physical space. The cognitive bias prediction module uses a dynamic time warping algorithm to align the multimodal behavioral feature sequences and teaching instruction sequences, calculating the response delay difference between the behavioral feedback reflected in the multimodal behavioral feature sequences and the teaching instruction sequences. This eliminates the physical time delay lag between the teaching speed and behavioral feedback. The cognitive bias prediction module determines the ratio of the response delay difference to the corresponding knowledge point weight of the teaching instruction to obtain the cognitive transmission influence factor, establishing a data mapping between bias assessment and the administrative level of teaching content. Based on the logical dependencies of the teaching instruction sequences, the cognitive bias prediction module calculates the impact of the initial bias of the current node on subsequent logical relationships. The system calculates the cumulative cognitive bias generated by the nodes and generates a cognitive bias transmission prediction graph containing this cumulative cognitive bias. This prediction graph reveals the global transmission risk caused by basic teaching instructions under specific knowledge point weights, resolving the technical contradiction between local bias tolerance and the overall administrative plan's anti-deviation requirements. The cognitive bias assessment model supporting this prediction process employs a multi-layer graph convolutional neural network (GCN) architecture. Its input layer receives a graph node feature matrix containing the topological out-degree of each node and real-time cognitive transmission influencing factors. Combined with an adjacency matrix derived from teaching logic instruction relationships, two layers of graph convolution operations extract the coupled implicit representation of the local network structure and bias values. In a pre-defined supervised learning process, the model uses mean squared error as the loss function for the difference between the cumulative predicted bias value and the historical test's true distribution. Weights are iteratively updated using a gradient backpropagation algorithm on an offline sample library containing past teaching video recordings and synchronous learning performance annotations.

[0032] The cognitive bias prediction module quantifies the influencing factors of cognitive transmission through mathematical calculations. The specific calculation formula is as follows: ,in As a factor influencing cognitive transmission, In response to the latency difference, where L represents the knowledge point weight corresponding to the instruction in the teaching instruction sequence, the teaching strategy scheduling module generates a teaching intervention signal for a specific abnormal area based on the deviation distribution in the cognitive deviation transmission prediction graph. This intervention signal includes weight adjustment parameters for teaching resources within the specific abnormal area and a progress rewind instruction for a specific teaching path. A management data packet containing operation codes and spatial coordinates is sent to the main control unit of the teaching terminal via the transmission control protocol. This directly calls the teaching terminal's demonstration software application interface, and the current playback interface jumps to the page corresponding to the preceding micro-knowledge point of the progress rewind instruction. Simultaneously, the directional speaker array corresponding to the specific abnormal area is adjusted, increasing the preset decibel output sound pressure. During this process, directional sound... The acoustic physical intervention of the pressure is not intended to directly overcome the intellectual barriers of learners, but rather to serve as a pre-emptive physiological sensory stimulation and arousal mechanism. By transiently increasing the acoustic energy of the local physical environment, it forcibly blocks the detached state in abnormal areas and resets the audience's attention allocation. This, combined with the subsequent dimensionality reduction and retrospective demonstration of the preceding micro-knowledge points with lower cognitive barriers, jointly constructs a closed-loop synergistic path from physical sensory attention capture to business logic reconstruction. The teaching strategy scheduling module redirects local monitoring sampling resources and intervention capabilities according to the teaching intervention signal, blocking the cascading amplification path of deviations along the business logic chain, so that the overall teaching execution state converges within the preset administrative planning envelope.

[0033] Example 2: This smart classroom interactive teaching software and system is deployed in a tiered teaching environment with a building area of ​​400 square meters. The multimodal data acquisition module includes a cluster of 6 visual sensors arranged at the top of the physical space. Each visual sensor has a resolution of 1920×1080 and a sampling frequency of 30Hz. The comparison time window length of the dynamic time warping algorithm needs to be balanced between the completeness of the alignment of the teaching instruction feature sequence and the memory resident load of data processing. When the physical period of a single instruction in the teaching instruction sequence is in the range of 2000ms to 3000ms, in order to avoid truncating the secondary behavioral feedback sequence of the audience in the back row, 1.5 times the physical period of a single instruction is selected as the comparison time window length. The comparison time is determined according to the physical space span of the current tiered teaching environment. The window length was 4500ms. Physical disturbances were introduced into the experimental environment to quantify the management prediction efficiency under heterogeneous physical spaces. Specifically, an environmental darkening condition with an illuminance of less than 50lx was superimposed on the back row area at a horizontal distance greater than 15m from the teaching end. At the same time, the physical seating arrangement in the front row was set to generate 20% visual obstruction area. The experiment selected spatial depth as a variable parameter to characterize the physical propagation resistance and behavioral lag. The test space was orthogonally divided into three gradient regions: the spatial depth of the first region was less than 5m, the spatial depth of the second region was 5m to 15m, and the spatial depth of the third region was 15m to 30m. A control group was set up in the experiment. The control group used a fixed time delay threshold to trigger the alarm and did not include knowledge point weight modulation logic. An experimental group was also set up in the experiment. The experimental group used the aforementioned dynamic time warping alignment and cognitive transmission influence factor calculation logic.

[0034] The test data reflects the mapping relationship between multimodal behavioral feature sequences and teaching instruction sequences under different spatial gradients. In the first region, the initial response delay difference acquired by the acquisition module is less than 300ms. Under this condition, the administrative intervention accuracy rate output by the control group is 92.4%, and the administrative intervention accuracy rate output by the experimental group is 94.1%. When the spatial depth increases to the second region, the physical spatial distance causes an increase in behavioral feedback lag, and the average response delay difference reflected by the multimodal behavioral feature sequences reaches 850ms. The control group judges this physical lag as cognitive disconnect and triggers intervention instructions, and its intervention accuracy rate drops to 71.5%. The cognitive bias prediction module of the experimental group uses a dynamic time warping algorithm to eliminate linear physical delay and, according to the formula... Calculate the cognitive transmission influencing factors, among which As a factor influencing cognitive transmission, The response delay difference after alignment is L, which is the dimensionless knowledge point weight corresponding to the teaching instruction. Under the condition that the knowledge point weight of a specific instruction is set to 0.8 and the measured response delay difference is 850ms, the calculated cognitive transmission influence factor is within the envelope of the conventional administrative planning. Based on this, the experimental group suppressed the intervention signal, and its intervention accuracy reached 91.8%.

[0035] When the test node enters the third region containing visual disturbances, the ambient darkness and visual field obstruction limit multimodal feature extraction, increasing the original response latency difference to 1850ms. The fixed threshold judgment mechanism of the control group triggers a large number of intervention commands, and its intervention accuracy drops to 34.2%. The multimodal data acquisition module reads the ambient light intensity values ​​output by the illuminance sensors deployed in each physical region in real time and the percentage of facial contour pixel integrity extracted by the visual sensor. When the ambient light intensity value is lower than the preset lower limit threshold or the pixel integrity percentage is lower than the preset area percentage threshold, it is determined that the current data stream break is due to pure physical obstruction. The cognitive bias prediction module locks the response latency difference of the last effective physical time period before the obstruction of the corresponding region, suspends the dynamic update calculation of the regional behavior matrix, and removes the input weight of transient outliers in the global prediction until the corresponding sensor measurement parameters return to the normal range. By decoupling physical variables and cognitive variables, the cognitive bias prediction module of the experimental group constructs a cognitive bias transmission prediction map and evaluates it based on logical dependencies. The data on the cumulative cognitive bias of the current latency on the downstream knowledge topology shows that when low-weight instructions do not trigger the disconnection of associated high-weight nodes, the teaching strategy scheduling module does not trigger global reset redirection, and the intervention accuracy of the experimental group remains at 86.5%. At the same time, the data reflects the measurement constraints of physical sensors. When the spatial depth is greater than 28m and the ambient illuminance is less than 30lx, the loss of facial action pixel features collected by the visual sensor exceeds the preset tolerance, resulting in missing bottom-level input data. The intervention accuracy of the experimental group then shows a non-linear decrease to 79.3%. The gradient data measured in the experiment shows the parametric correlation between the response latency difference and the knowledge point weight. Under the combined conditions of increased spatial depth and darkened environment, the experimental group filters out the temporal jitter caused by physical distance based on the mathematical transformation of cognitive transmission influencing factors and the quantitative output of the deviation accumulation map. This process blocks the transmission path of physical lag to administrative alarms, confirming that the system maintains the accuracy of the teaching strategy scheduling instruction output under the conditions of physical spatial heterogeneity and light disturbance.

[0036] Example 3: This smart classroom interactive teaching software and system is deployed in a physical environment with a sequence of micro-knowledge nodes. The cognitive bias prediction module extracts a pre-set directed acyclic graph of the teaching syllabus, calculates the topological out-degree of the micro-knowledge node corresponding to each independent teaching instruction in the directed acyclic graph, and selects the total number of connecting edges from the current micro-knowledge node to the downstream derived knowledge node as the calculation benchmark. It calculates the product of the total number of connecting edges and the corresponding class time allocation ratio, and determines the knowledge point weight of the corresponding teaching instruction. Specifically, before the alignment calculation, the multimodal behavioral feature sequence and the teaching instruction sequence are mapped to a common tensor metric space. The cognitive bias prediction module extracts the issuance time coordinate, continuous physical cycle parameter, and semantic complexity of the teaching instruction to generate a one-dimensional instruction temporal vector. At the same time, it normalizes the facial posture angle change rate and audio feedback energy features collected in the corresponding time window in the physical space to form a behavioral temporal vector with the same dimensional structure as the aforementioned instruction vector, thereby satisfying the subsequent dynamic time... The regularization algorithm has mathematical requirements for the equivalence of input feature dimensions and the computability of spatial distance. The cognitive bias prediction module uses Euclidean distance to measure feature vectors, calculates the local spatial distance between the multimodal behavior feature sequence and the teaching instruction sequence within the comparison time window, and constructs a cumulative cost matrix. Based on the cumulative cost matrix, the cognitive bias prediction module backtracks from the matrix boundary to find the regularization path with the minimum cumulative distance. According to the regularization path, it locates the mapping node between the state transition time coordinates in the multimodal behavior feature sequence and the instruction issuance time coordinates in the teaching instruction sequence, extracts the response delay difference, extracts the posture angle parameters and audio energy parameters in each region behavior matrix to form the initial input tensor, calculates the covariance matrix of the initial input tensor and solves for the eigenvalues ​​and eigenvectors, selects the K principal components with the cumulative contribution rate exceeding the preset proportion threshold to construct a low-dimensional feature space, calculates the Mahalanobis distance between the real-time extracted state coordinates and the preset standard behavior benchmark coordinates in the low-dimensional feature space, and outputs the calculated Mahalanobis distance value as the logical deviation of the region behavior matrix relative to the teaching instruction sequence.

[0037] The cognitive bias prediction module sets the cognitive transmission influence factor, determined based on the response delay difference and knowledge point weight, as the activation attribute of the initial micro-knowledge node in the directed acyclic graph. It uses a breadth-first search algorithm to traverse downstream derived knowledge nodes connected to the initial micro-knowledge node. The module calculates the product of the cognitive transmission influence factor and the corresponding edge attenuation coefficient, outputting the accumulated cognitive bias received by each downstream derived knowledge node. The teaching strategy scheduling module obtains a cognitive bias transmission prediction graph containing the spatial distribution of the accumulated cognitive bias, generates a teaching intervention signal containing spatial coordinate parameters, and outputs teaching strategy scheduling instructions for specific abnormal areas. The cognitive bias prediction module introduces a Kalman filter algorithm to predict the evolution trajectory of multimodal behavioral feature sequences. It constructs a basic state vector from the multimodal behavioral feature sequences extracted within the preceding physical time period, based on preset... The state transition matrix calculates the prior feature state and prior error covariance matrix at the current physical time node. Simultaneously, it reads the real-time observation feature sequence output by the multimodal data acquisition module at the current physical time node. The cognitive bias prediction module calculates the Kalman gain coefficient by combining the preset observation matrix and prior error covariance matrix. It compares the real-time observation feature sequence with the prior feature state to extract feature residuals. It uses the Kalman gain coefficient and feature residuals to determine the numerical compensation amount for the prior feature state. It merges the prior feature state and numerical compensation amount to generate the posterior feature state sequence. The cognitive bias prediction module confirms the continuously output posterior feature state sequence as the smooth evolution trajectory of the multimodal behavioral feature sequence and substitutes it into the dynamic time warping algorithm to extract the response delay difference. This mathematical procedure relies on the recursive calculation of state prediction and measurement update to filter out transient feature jumps caused by physical environmental disturbances in the underlying data acquisition link.

[0038] Example 4: When the system faces the initial deployment of a tiered teaching physical space with unknown parameter gradients, to isolate the acoustic and optical propagation delay interference caused by the spatial structure on the multimodal behavioral characteristic sequence, the system initiates a spatial time delay baseline calibration procedure before the start of normalized teaching activities. The teaching terminal continuously sends a test sequence containing alternating flashing patterns and pulse audio signals into the physical space. The visual sensor cluster and audio receiving array arranged at the top of the physical space synchronously capture the initial stress physical feedback generated by the test group in response to the test sequence within the behavioral matrix of each area. The cognitive bias prediction module calculates the absolute time span between the time coordinate of the test sequence and the time coordinate of the initial stress physical feedback extraction, and divides the physical space into multiple discrete feature grids according to the depth coordinate. Based on the statistical average of the absolute time span within each discrete feature grid, a static spatial time delay compensation matrix corresponding to the current physical space is generated. In subsequent normalized teaching activities... In the teaching process, the cognitive bias prediction module calculates the initial response delay difference including propagation lag using a dynamic time warping algorithm. Then, it subtracts the reference physical delay corresponding to the coordinate parameters of the current acquisition end in the static spatial delay compensation matrix from the initial response delay difference to obtain the actual cognitive response delay difference after removing the physical transmission delay. This operation of subtracting the static reference physical delay is essentially to achieve the starting translation alignment of the coordinate system nodes in the time domain. Since the millisecond-level delay caused by the physical propagation of sound and light constitutes a constant system noise that increases linearly with distance in a large-scale physical space, after statically removing it as a constant term, the system logically establishes a unified relative time zero point across the entire field. This avoids the linear noise of physical distance from masking the actual second-level and minute-level nonlinear cognitive lag phenomena, ensuring that the cognitive internalization index has decoupled extractability after removing the underlying physical variables.

[0039] In defining the connectivity quantification rules of the directed acyclic graph (DAG) within a pre-defined teaching syllabus, to determine the criterion for the attenuation coefficient of the connection edges, the cognitive bias prediction module imports a historical comparison database containing teaching test parameters from previous years. This database contains the original error rate indicators of each micro-knowledge node in the audience sample set. The cognitive bias prediction module identifies any initial micro-knowledge node and derived knowledge node within the DAG that are logically related upstream and downstream. It reads the first test error rate of the initial micro-knowledge node and the second test error rate of the derived knowledge node from the historical comparison database, calculates the relative ratio of the first test error rate to the second test error rate, and substitutes this relative ratio, along with the attenuation base of the target teaching path, into the calculation model. The calculation logic for the attenuation coefficient of the connection edges satisfies the formula... Where μ is the attenuation coefficient of the connecting edge, and α is the dimensionless attenuation base of the corresponding target teaching path. The error rate for the first test, The second test error rate is the same as the first test error rate. Since the second test error rate has the same dimensional structure, the resulting relative ratio is a dimensionless constant, which ensures that the connection edge attenuation coefficient and the dimensionless attenuation base remain consistent in dimensional properties. The cognitive bias prediction module uses the connection edge attenuation coefficient to construct a cognitive bias transmission prediction graph. The teaching strategy scheduling module generates teaching strategy scheduling instructions based on the updated cognitive bias transmission prediction graph.

[0040] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart classroom interactive teaching software and system, characterized in that, The system includes: The multimodal data acquisition module is used to acquire multimodal behavioral feature sequences and teaching instruction sequences in the classroom environment. The teaching instruction sequence includes multiple teaching instructions and the knowledge point weights corresponding to each teaching instruction. The region feature parsing module is used to divide the multimodal behavioral feature sequence into multiple region behavioral matrices based on the coordinate parameters of the acquisition end; The cognitive bias prediction module is used to initiate a spatial time delay baseline calibration procedure before the start of routine teaching activities. It calculates the absolute time span between the time coordinates of the test sequence distribution and the time coordinates of the initial stress physical feedback extraction, and divides the physical space into multiple discrete feature grids according to the depth coordinates. Based on the statistical average of the absolute time spans in each discrete feature grid, it generates a static spatial time delay compensation matrix corresponding to the current physical space. In the subsequent routine teaching process, it uses a dynamic time warping algorithm to align the multimodal behavioral feature sequence and the teaching instruction sequence, calculates the initial response time delay difference between the behavioral feedback reflected by the multimodal behavioral feature sequence and the teaching instruction sequence, and subtracts the baseline physical delay corresponding to the coordinate parameters of the current acquisition end in the static spatial time delay compensation matrix from the initial response time delay difference to obtain the actual cognitive response time delay difference after removing the physical transmission delay. It then determines the ratio of this actual cognitive response time delay difference to the knowledge point weight of the corresponding teaching instruction to obtain the cognitive transmission influence factor, and generates a cognitive bias transmission prediction map containing the cumulative amount of cognitive bias based on the logical dependency relationship of the teaching instruction sequence. The teaching strategy scheduling module is used to generate teaching intervention signals for specific abnormal areas based on the deviation distribution in the cognitive deviation transmission prediction map.

2. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The cognitive bias prediction module uses the Kalman filter algorithm to predict the evolution trajectory of multimodal behavioral feature sequences. When the multimodal behavioral feature sequences are interrupted due to environmental occlusion, interpolation features are generated based on the state transition matrix corresponding to the evolution trajectory. The interpolation features are used to complete the multimodal behavioral feature sequences, so that the data dimension of the input cognitive bias transmission prediction map remains constant during the teaching cycle.

3. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The multimodal behavioral feature sequences acquired by the multimodal data acquisition module include: posture parameters in the visual modality, feedback signal features in the audio modality, and response latency in the interactive modality.

4. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The cognitive bias prediction module uses high-dimensional mapping logic to map unstructured regional behavior matrices to teaching quality assessment dimensions and calculates the logical deviation of each regional behavior matrix from the teaching instruction sequence.

5. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The teaching intervention signals generated by the teaching strategy scheduling module include: weight adjustment parameters for teaching resources in specific abnormal areas and progress retracing instructions for specific teaching paths.

6. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The cognitive bias prediction module establishes a teaching instruction association model, which defines the transmission influence function between the preceding basic instruction and the subsequent derived instruction. The transmission influence function is used to predict the management impact of the cumulative amount of cognitive bias on the overall teaching progress.

7. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The teaching strategy scheduling module also includes closed-loop feedback logic, which is used to dynamically correct the prediction gain coefficient of the cognitive bias transmission prediction map based on the real-time status of the regional behavior matrix after the teaching intervention signal output.

8. The smart classroom interactive teaching software and system according to claim 1, characterized in that, When dividing regions, the regional feature analysis module is also used to establish an association matrix between regions. The association matrix is ​​used to characterize the state coupling degree between the behavior matrices of adjacent regions, so as to assist the teaching strategy scheduling module in determining the diffusion characteristics of abnormal regions.

9. The smart classroom interactive teaching software and system according to claim 1, characterized in that, The system also includes a central processing cluster, which stores a cognitive bias assessment model trained by deep learning. The cognitive bias assessment model is used to support the cognitive bias prediction module in generating a cognitive bias transmission prediction map.

Images