Teaching behavior pattern recognition method based on big data technology

Abstract

The invention relates to a teaching behavior pattern recognition method based on a big data technology, and the method comprises the steps: obtaining a high-concurrency data stream, carrying out the fragmentation processing of the high-concurrency data stream through a distributed processing technology, and obtaining a preliminary fragmentation data stream; for the preliminary fragmented data stream, determining a data stream state by adopting a real-time judgment technology to obtain an optimized fragmented data stream; extracting features from the optimized fragmented data stream, and training a behavior recognition model through a machine learning algorithm to obtain a feature vector set; according to the feature vector set, a pattern recognition algorithm is adopted to generate a behavior recognition result, and a prediction result set is obtained; processing the prediction result set through a real-time control technology and an event-driven mechanism, and generating an adjustment instruction set; monitoring the adjustment instruction set by adopting a closed-loop control system to obtain a synchronous adjustment parameter set; optimizing the behavior recognition model through a reinforcement learning algorithm, and generating a feedback loop data stream; and extracting a performance index from the feedback loop data stream, and determining a system optimization result.

Description

Technical Field

[0001] The present application relates to the field of electrical data processing technology, and in particular to a teaching behavior pattern recognition method based on big data technology. Background Art

[0002] Currently, many solutions are still at the level of static data analysis, relying on post-statistics or offline processing, and are unable to respond to learners' dynamic needs in a timely manner. This lag causes teaching adjustments to be out of touch with students' actual status, limiting the potential effectiveness of intelligent teaching systems. The limitations of existing methods are mainly reflected in insufficient real-time performance and low accuracy in behavior recognition. Although many systems can collect large amounts of data, it is difficult to quickly analyze and convert them into effective interventions during the teaching process, and critical teaching opportunities are often missed. In addition, the complexity of teaching behaviors makes pattern recognition algorithms less adaptable in diverse scenarios, and is prone to misjudgment or omissions. These defects make it difficult for intelligent teaching environments to truly achieve dynamic adjustment and instant feedback.

[0003] The core challenge focuses on how to use digital data and big data technology to solve the real-time and accuracy problems in teaching behavior pattern recognition. First, real-time analysis of learning behavior data streams requires efficient data processing capabilities, and current technologies often face delay bottlenecks when facing high-concurrency data. Second, the accuracy of pattern recognition depends on a deep understanding of complex teaching scenarios, including subtle differences in attention distraction or learning breakthroughs. Existing algorithms find it difficult to fully capture these dynamic features. These two technical factors have not been resolved, resulting in the system's inability to accurately intervene at critical moments, which in turn affects the adaptive adjustment capabilities of the teaching process.

[0004] Therefore, how to seamlessly combine teaching behavior recognition with instant feedback mechanism through real-time control technology to achieve adjustment of teaching content and resources based on the dynamic state of learners has become a key issue in building an intelligent teaching environment. This issue not only involves the rapid processing of data streams, but also requires improving the ability of pattern recognition to capture key teaching moments to ensure that closed-loop control of the teaching process can be efficiently achieved. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the purpose of this application is to provide a teaching behavior pattern recognition method based on big data technology.

[0006] The teaching behavior pattern recognition method based on big data technology described in this application includes the following steps: S101, obtaining a high-concurrency data stream from the learning behavior data, sharding the data using distributed stream processing technology, using a preset threshold based on latency and throughput to determine whether the data stream meets the real-time requirements, and obtaining the data stream shards after preliminary processing; S102: for the data stream fragments after preliminary processing, an incremental calculation method is used to update the data processing capacity. When a high-concurrency data stream arrives, computing resources are allocated through a parallel computing framework to determine the data stream state after the real-time bottleneck is alleviated. S103, obtaining action and time feature sets related to behavior recognition accuracy from the data flow state after the real-time bottleneck is relieved, training multi-dimensional features in complex teaching scenarios through a deep learning model, and obtaining feature vectors of key teaching moments; S104, extracting identification information from feature vectors of key teaching moments, using an attention mechanism to enhance the adaptability of the pattern recognition algorithm to complex teaching scenarios, and obtaining a prediction result set with improved behavior recognition accuracy; S105, obtaining identification information of key teaching moments from the prediction result set after the behavior recognition accuracy is improved, triggering dynamic adjustment capabilities through the combination of real-time control technology and event-driven architecture, and generating an instruction set for adjusting teaching content; S106, extracting specific parameters from the instruction set for adjusting the teaching content, monitoring the change trend of the learning behavior data through a closed-loop control system, and obtaining an adjustment parameter set synchronized with the high-concurrency data stream; S107, extracting the execution state of the dynamic adjustment capability from the adjustment parameter set synchronized with the high concurrent data stream, optimizing the accuracy of the pattern recognition algorithm using a reinforcement learning algorithm, and generating a feedback loop data stream after the closed-loop control system is optimized; S108, extracting delay and accuracy indicators from the feedback loop data stream after the closed-loop control system is optimized, and obtaining a verification data stream after the system performance is improved by analyzing and verifying the execution effect of the dynamic adjustment capability; S109. Obtain the operating status of the closed-loop control system from the verification data stream after the system performance is improved, seamlessly combine the teaching content adjustment with the instant feedback mechanism through the event-driven architecture, and generate a teaching optimization parameter set after precise intervention at key teaching moments.

[0007] Preferably, in step S101, the slicing of the high-concurrency data stream by using a distributed processing technology comprises: acquiring a sensor data stream, slicing the sensor data stream by using a distributed stream processing technology, and obtaining an initial slicing data stream; For the initial fragmented data stream, a preset delay threshold is used for judgment, and if the delay exceeds the delay threshold, it is marked as a non-real-time data stream; According to the judgment result, the throughput data of the initial sharded data flow is obtained, and a preset throughput threshold is used for judgment. If the throughput is lower than the throughput threshold, the sharding strategy is adjusted to obtain an optimized sharded data flow; The optimized sharded data stream is redistributed through a consistent hashing algorithm to obtain a balanced sharded data stream.

[0008] Preferably, in step S102, the adopting of real-time judgment technology to determine the data stream state includes: obtaining an initial sharded data stream, and updating the initial sharded data stream using an incremental calculation method to obtain a data stream with improved processing capability; When high-concurrency data streams arrive, computing resources are allocated through the parallel computing framework to obtain optimized data streams; Determine whether the optimized data flow exceeds a preset concurrency threshold, and if so, adjust the resource allocation strategy; According to the adjustment results, the stream processing technology is used to detect the bottleneck relief status and obtain the data flow after the bottleneck is relieved; The data flow after the bottleneck is alleviated is analyzed by a status monitoring tool to obtain a final data flow status.

[0009] Preferably, in step S103, the step of training the behavior recognition model by a machine learning algorithm to obtain a feature vector set includes: obtaining action features and time features to obtain an initial feature set; Using a convolutional neural network to extract multidimensional features from the initial feature set to obtain a multidimensional feature matrix; If the matching degree between the action features and the time features in the multidimensional feature matrix is ​​higher than a preset threshold, the multidimensional feature matrix is ​​trained by a long short-term memory network to obtain a behavior recognition model; The key moment candidate set is grouped by clustering algorithm to obtain a set of feature vectors; If there are outliers in the feature vector set, the outliers are removed by a filtering tool to obtain an optimized feature vector set.

[0010] Preferably, in step S104, the use of a pattern recognition algorithm to generate a behavior recognition result includes: acquiring teaching moment data, extracting identification information according to preset rules, and obtaining a preliminary feature set; For the preliminary feature set, an attention mechanism is used to assign weights to obtain an enhanced feature representation; Inputting the enhanced feature representation into a pattern recognition algorithm to obtain a behavior feature; Determine whether the behavior feature matches a preset category. If so, generate a behavior recognition result through a classifier to obtain a behavior classification set. According to the behavior classification set, a statistical tool is used to calculate the recognition accuracy to obtain accuracy improvement data; The behavior classification set and the precision improvement data are integrated to generate a prediction result set.

[0011] Preferably, in step S105, the processing of the prediction result set by real-time control technology and event-driven mechanism includes: obtaining the prediction result set, extracting identification information of key moments by behavior recognition technology, and obtaining preliminary classification data; According to the preliminary classification data, real-time control technology is used to match a preset event-driven architecture to determine whether a trigger condition is met, and if the trigger condition is met, a dynamic adjustment signal is determined; According to the dynamic adjustment signal, an adjustment capability parameter is generated through an event-driven mechanism to obtain an adjustment requirement; According to the adjustment requirement, an instruction set is generated in combination with the adjustment capability parameter to obtain an adjustment plan; The integrity of the instruction set is judged by a preset threshold to determine the final output sequence.

[0012] Preferably, in step S106, the use of a closed-loop control system to monitor the adjustment instruction set includes: obtaining an initial parameter set, and using a closed-loop control method to obtain a real-time change trend; Determine whether the real-time change trend exceeds a preset threshold, and if so, determine an updated value of the adjustment parameter through a parameter extraction algorithm; According to the state of the high concurrent data stream, synchronously process the adjustment parameters and the monitoring data to obtain synchronously processed data; Using a support vector machine algorithm to classify the synchronously processed data to obtain an optimized parameter set; According to the optimized parameter set, updating the adjustment instruction set to obtain an updated instruction set; The updated instruction set is monitored by the closed-loop control system to obtain a final adjustment parameter set.

[0013] Preferably, in step S107, optimizing the behavior recognition model by using a reinforcement learning algorithm includes: acquiring high-concurrency data, and obtaining an adjustment parameter set and an initial execution state by synchronously processing the high-concurrency data; Extracting dynamic adjustment capability features from the adjustment parameter set to determine the change trend of the execution state; Using a reinforcement learning algorithm, the pattern recognition algorithm is optimized according to the change trend to obtain parameters with improved accuracy; According to the parameters with improved accuracy, a closed-loop control system is adjusted to generate an optimized feedback data stream; According to the optimized feedback data flow, determining the system adjustment direction and obtaining the adjusted execution state; If the adjusted execution state does not reach a preset threshold, the feedback data stream is re-optimized to obtain a stable state.

[0014] The teaching behavior pattern recognition method based on big data technology described in the present application has the advantages that the method performs real-time analysis on learning behavior data through distributed stream processing technology, adopts deep learning models to identify key moments in complex teaching scenarios, and combines attention mechanisms to improve behavior recognition accuracy. The present invention utilizes real-time control technology and event-driven architecture to dynamically adjust teaching content according to recognition results, and continuously monitors changes in learning behavior through a closed-loop control system. At the same time, the present invention applies reinforcement learning algorithms to optimize pattern recognition accuracy, and realizes a seamless combination of teaching content adjustment and instant feedback. This method can realize accurate identification and timely intervention of teaching scenarios in a high-concurrency environment, effectively improving teaching quality and learning outcomes. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is the process of a teaching behavior pattern recognition method based on big data technology described in this application Figure 1 ; Figure 2 This is the process of a teaching behavior pattern recognition method based on big data technology described in this application Figure 2 . DETAILED DESCRIPTION

[0016] like Figure 1-Figure 2 As shown, the present application describes a teaching behavior pattern recognition method based on big data technology.

[0017] The teaching behavior pattern recognition method based on big data technology described in this application includes the following steps: S101, obtaining a high-concurrency data stream from the learning behavior data, sharding the data using distributed stream processing technology, using a preset threshold based on latency and throughput to determine whether the data stream meets the real-time requirements, and obtaining the data stream shards after preliminary processing; S102: for the data stream fragments after preliminary processing, an incremental calculation method is used to update the data processing capacity. When a high-concurrency data stream arrives, computing resources are allocated through a parallel computing framework to determine the data stream state after the real-time bottleneck is alleviated. S103, obtaining action and time feature sets related to behavior recognition accuracy from the data flow state after the real-time bottleneck is relieved, training multi-dimensional features in complex teaching scenarios through a deep learning model, and obtaining feature vectors of key teaching moments; S104, extracting identification information from feature vectors of key teaching moments, using an attention mechanism to enhance the adaptability of the pattern recognition algorithm to complex teaching scenarios, and obtaining a prediction result set with improved behavior recognition accuracy; S105, obtaining identification information of key teaching moments from the prediction result set after the behavior recognition accuracy is improved, triggering dynamic adjustment capabilities through the combination of real-time control technology and event-driven architecture, and generating an instruction set for adjusting teaching content; S106, extracting specific parameters from the instruction set for adjusting the teaching content, monitoring the change trend of the learning behavior data through a closed-loop control system, and obtaining an adjustment parameter set synchronized with the high-concurrency data stream; S107, extracting the execution state of the dynamic adjustment capability from the adjustment parameter set synchronized with the high concurrent data stream, optimizing the accuracy of the pattern recognition algorithm using a reinforcement learning algorithm, and generating a feedback loop data stream after the closed-loop control system is optimized; S108, extracting delay and accuracy indicators from the feedback loop data stream after the closed-loop control system is optimized, and obtaining a verification data stream after the system performance is improved by analyzing and verifying the execution effect of the dynamic adjustment capability; S109. Obtain the operating status of the closed-loop control system from the verification data stream after the system performance is improved, seamlessly combine the teaching content adjustment with the instant feedback mechanism through the event-driven architecture, and generate a teaching optimization parameter set after precise intervention at key teaching moments.

[0018] like Figure 1-Figure 2 As shown, in step S101, a high-concurrency data stream is obtained from the learning behavior data, the data is sharded through distributed stream processing technology, and a preset threshold based on latency and throughput is used to determine whether the data stream meets the real-time requirements, thereby obtaining the data stream shards after preliminary processing.

[0019] Furthermore, in step S101, a high-concurrency data stream is obtained from the learning behavior data, and the data is sharded by using a distributed stream processing technology to obtain an initial sharded data stream; For the initial sharded data stream, a preset delay threshold is used for judgment. If the delay exceeds the delay threshold, it is marked as a non-real-time data stream, and a preliminary real-time judgment result is obtained; According to the preliminary real-time judgment result, the throughput data of the initial shard data flow is obtained, and the throughput is judged by the preset throughput threshold to obtain the judgment result of whether the throughput meets the requirement; If the throughput data is lower than the throughput threshold, the sharding strategy is adjusted, and the sharding operation is performed on the adjusted strategy through the distributed system to obtain optimized data stream sharding; The distributed system performs stream processing operations on the optimized data stream shards, and uses the consistent hashing algorithm to redistribute the processed data to obtain a balanced shard data stream; Extract key features from the balanced sharded data stream, analyze the key features using the support vector machine algorithm, determine whether the real-time performance meets the requirements, and obtain the final real-time performance judgment result; According to the final real-time judgment result, the sharding results are stored through the logging tool, and the data processing capacity is updated using the incremental calculation method to obtain a persistent sharding data stream; For persistent sharded data streams, computing resources are allocated through a parallel computing framework. If the high-concurrency data stream exceeds the preset concurrency threshold when it arrives, the resource allocation strategy is adjusted to obtain an optimized data stream. The bottleneck relief status of the optimized data flow is detected through stream processing technology, and the data status is analyzed using status monitoring tools to obtain the final data flow status.

[0020] Specifically, in step S101, a high-concurrency data stream of 100,000 pieces per second is collected from the learning behavior data, and the KeyBy operator of Apache Flink is used to perform sharding processing according to the user ID to generate an initial sharded data stream; A sliding window is used for the initial shard data stream, including a window size of 1 second and a sliding step of 500 milliseconds to calculate the processing delay. If the 95th percentile delay exceeds the 200 millisecond threshold, the shard is marked as a non-real-time data stream; Based on the real-time marking results, the throughput indicators of each shard are counted. When the throughput of a shard is lower than 5,000 records / second, the dynamic sharding strategy adjustment is triggered, and the original shard is split into two sub-shards according to the hash modulus method; Flink's Rebalance operator is used to perform streaming aggregation calculations on the optimized sharded data, and then the consistent hashing algorithm is used to redistribute the data partitions. The number of virtual nodes is set to 1000 to ensure that the load difference between each node does not exceed 15%; 12-dimensional features such as timestamp interval and data volume volatility are extracted from the balanced shard data and input into the SVM classifier, including the kernel function using RBF, the penalty factor C=1.0 for real-time judgment, and the classification confidence threshold is set to 0.85; The judgment results are written into the Elasticsearch log cluster, and the system processing capacity index is updated based on the exponentially weighted moving average method, where the smoothing coefficient α=0.3; When the instantaneous concurrent volume exceeds 80,000 per second, the YARN resource scheduler is automatically triggered to expand the computing nodes to 1.5 times the original number; The network IO utilization is detected through Flink's back pressure monitoring mechanism. When it exceeds the 75% threshold for three consecutive cycles, the data localization optimization strategy is started, and finally a stable data stream is output.

[0021] like Figure 1-Figure 2As shown, in step S102, for the data stream shards after preliminary processing, an incremental calculation method is used to update the data processing capacity. When high-concurrency data streams arrive, computing resources are allocated through a parallel computing framework to determine the data stream state after the real-time bottleneck is alleviated.

[0022] Furthermore, in step S102, the data stream fragments after preliminary processing are obtained, and the data processing capacity is updated using an incremental calculation method; Allocate computing resources through a parallel computing framework to obtain data streams with improved processing capabilities; Determine whether the data flow after the processing capacity is improved exceeds the preset concurrency threshold. If so, adjust the resource allocation strategy; According to the adjustment result of the resource allocation strategy, the optimized data flow is obtained; Detect the bottleneck relief state of the optimized data flow through stream processing technology to obtain the data flow after the bottleneck is relieved; Use status monitoring tools to analyze the data flow after the bottleneck is alleviated to obtain the final data flow status; Extract key indicators from the final data flow status to determine whether the system operation is stable; According to the judgment result of system operation stability, the data stream is redistributed through the consistent hashing algorithm to obtain a balanced sharded data stream; Real-time features are extracted from the balanced sharded data stream, and the data stream status after the real-time bottleneck is alleviated is determined by the support vector machine algorithm.

[0023] Specifically, in step S102, the data stream fragments after preliminary processing are obtained, and the data processing capacity is updated using an incremental calculation method, including dynamically adjusting the computing node load based on a sliding window mechanism, with the window size set to 5 seconds and the incremental factor set to 0.2; Computing resources are allocated through the parallel computing framework, and the Spark dynamic resource scheduling strategy is adopted. The minimum number of executors is set to 4 and the maximum is set to 16. The memory allocation is 8GB / executor. The data flow with improved processing power is obtained. Determine whether the data flow after the processing capacity is improved exceeds the preset concurrency threshold, including QPS ≥ 5000. If it exceeds, adjust the resource allocation strategy and use the Kubernetes automatic expansion and contraction mechanism to increase the number of Pod copies from 4 to 8; According to the adjustment results of the resource allocation strategy, the optimized data flow is obtained, and the delay is reduced from 200ms to 80ms; The bottleneck relief status of the optimized data flow is detected through stream processing technology. Flink's back pressure monitoring mechanism is used to determine that the bottleneck is relieved when the buffer usage rate is lower than 70%, and the data flow after the bottleneck is relieved is obtained; Use the status monitoring tool Prometheus to analyze the data flow after the bottleneck is relieved, collect indicators such as CPU utilization ≤ 60%, network IO ≤ 1 Gbps, and obtain the final data flow status; Extract key indicators from the final data flow status, including throughput fluctuation rate ≤ 5%, to determine whether the system operation is stable; According to the judgment result of system operation stability, the data flow is redistributed through the consistent hashing algorithm, the number of virtual nodes is set to 200, and the shard offset rate is ensured to be ≤3%, so as to obtain a balanced shard data flow; Real-time features including end-to-end delay and processing rate are extracted from the balanced sharded data stream. The classification model is trained through the support vector machine algorithm, including the kernel function as RBF and the penalty factor C=1.0, with an accuracy rate of ≥95%, to determine the data stream status after the real-time bottleneck is alleviated.

[0024] like Figure 1-Figure 2 As shown, in step S103, the action and time feature sets related to behavior recognition accuracy are obtained from the data flow state after the real-time bottleneck is alleviated, and the multi-dimensional features in the complex teaching scene are trained through the deep learning model to obtain the feature vectors of the key teaching moments.

[0025] Further, in step S103, action features and time features are extracted from the data flow state after the real-time bottleneck is relieved to obtain an initial feature set; A convolutional neural network is used to extract multi-dimensional features from the initial feature set to obtain a multi-dimensional feature matrix; If the matching degree between the action features and the time features in the multidimensional feature matrix is ​​higher than the preset threshold, the multidimensional feature matrix is ​​trained through the long short-term memory network to obtain the behavior recognition model of the teaching scene; According to the behavior recognition model, the changing trend of action features in the multi-dimensional feature matrix is ​​analyzed to obtain the candidate set of key moments; The key moment candidate set is grouped by clustering algorithm to obtain the feature vector set of the key moment; If there are outliers in the feature vector set that are inconsistent with the time characteristics, the outliers are removed through the filtering tool to obtain the optimized feature vector set; Verify the accuracy of behavior recognition for the optimized feature vector set to obtain the final feature vector; Extract identification information from the final feature vector, use the attention mechanism to enhance the pattern recognition algorithm, and obtain a set of prediction results with improved behavior recognition accuracy; The multi-dimensional feature changes in complex teaching scenarios are analyzed through the prediction result set to determine the feature vectors of key teaching moments.

[0026] Specifically, in step S103, action features and time features are extracted from the data stream state after the real-time bottleneck is relieved, the OpenPose algorithm is used to detect the coordinates of the key points of the human body, and the initial feature set is constructed in combination with the timestamp information. The feature dimension is 25×3, including 25 key points, each of which contains x, y coordinates and confidence. A convolutional neural network is used to extract multidimensional features from the initial feature set, and the ResNet-50 architecture is used to extract spatial features, output a 128-dimensional feature vector, and form a multidimensional feature matrix; If the matching degree between the action features and the time features in the multidimensional feature matrix is ​​higher than the preset threshold, including the cosine similarity > 0.85, the multidimensional feature matrix is ​​trained through a bidirectional LSTM network with 64 hidden layer units, and the behavior recognition model of the teaching scene is output; According to the behavior recognition model, the change trend of action features in the multidimensional feature matrix is ​​analyzed, and the mean square error of feature differences between adjacent frames is calculated. Frames with MSE>0.1 are marked as key moment candidate sets; The key moment candidate set is grouped by DBSCAN clustering algorithm, the neighborhood radius ε=0.5 and the minimum number of samples min_samples=3 are set, and the feature vector set of the key moment is obtained; If there are outliers in the feature vector set that are inconsistent with the time characteristics, including isolated points with a time interval of more than 5 seconds, the Kalman filter is used to remove the outliers to obtain the optimized feature vector set; Verify the accuracy of behavior recognition for the optimized feature vector set, and use cross-validation (k=5) to calculate the feature vector with F1-score>0.9 as the final output; Extract identification information from the final feature vector, use a multi-head attention mechanism (head number = 8) to enhance the pattern recognition algorithm, with an output dimension of 256, and obtain a prediction result set with improved behavior recognition accuracy; The multi-dimensional feature changes in complex teaching scenarios are analyzed by predicting the result set, the dynamic time warping (DTW) distance of the feature vector is calculated, and the feature vector of the key teaching moment is determined.

[0027] like Figure 1-Figure 2 As shown, in step S104, identification information is extracted from the feature vector of the key teaching moment, and the attention mechanism is used to enhance the adaptability of the pattern recognition algorithm to complex teaching scenarios, so as to obtain a prediction result set with improved behavior recognition accuracy.

[0028] Further, in step S104, identification information is extracted from the feature vector of the key teaching moment, and the attention mechanism is used to enhance the adaptability of the pattern recognition algorithm to complex teaching scenarios, so as to obtain a prediction result set with improved behavior recognition accuracy; According to the identification information in the prediction result set, the action and time feature sets related to the behavior recognition accuracy are obtained, and the preliminary classification data of the multi-dimensional features are determined; The action and time feature sets are trained through deep learning models to obtain feature vectors of key teaching moments in complex teaching scenarios; Extract the identification information of key moments from the feature vector, and use the data flow status after the real-time bottleneck is alleviated to determine the degree of improvement in behavior recognition accuracy; If the behavior recognition accuracy is determined to meet the preset threshold, the event-driven architecture is matched through real-time control technology to determine the start signal of dynamic adjustment; According to the start signal, an event-driven mechanism is used to generate adjustment capability parameters and obtain the instruction set corresponding to the adjustment requirements of the teaching content; Through the core fields of the instruction set, the integrity of the instruction set is judged using a preset threshold to obtain the final output sequence; After obtaining the final output sequence, a machine learning algorithm is used to optimize the accuracy of the behavior recognition technology to obtain an updated set of prediction results; According to the updated prediction result set, the identification information of key moments is processed cyclically to obtain the next round of teaching content adjustment instructions.

[0029] Specifically, in step S104, identification information is extracted from the feature vector of the key teaching moment, and the Transformer model based on the multi-head attention mechanism is used to enhance the features of the teaching video frame sequence. The number of attention heads is set to 8, the hidden layer dimension is set to 512, and the weight distribution is calculated by Softmax normalization, and the prediction result set with the behavior recognition accuracy improved to 92.3% is obtained; According to the timestamp in the prediction result set, the sliding window algorithm is used to extract the action features at intervals of 0.5 seconds. Combined with the time series features output by the LSTM network, a 128-dimensional feature vector containing joint angles and movement speed is formed to determine the preliminary classification data. The feature set was trained using the ResNet-50 deep learning model, with a batch size of 32, a learning rate of 0.001, and a cross entropy loss function for 100 iterations, outputting a 2048-dimensional feature vector of key teaching moments. The TOP-10 significant area markers are extracted from the feature vector, and the Kalman filter algorithm is used to process the data stream delay. When the recognition frame rate is stabilized at 25FPS, the improvement in behavior recognition accuracy exceeds the preset threshold by 15%; If the accuracy meets the requirement, the event trigger frequency is adjusted through the PID controller, matching the event-driven architecture based on the Kafka message queue, and generating a start signal when the message throughput reaches 5,000 messages per second; According to the signal strength parameters, the decision tree model is used to generate adjustment weights, and the learning concentration threshold is set at 70%. 12 types of instruction templates including knowledge point reorganization and speech speed adjustment are output; By parsing the keyword frequency in the instruction, the TF-IDF algorithm is used to calculate the field weight. When the integrity score exceeds 0.85, the final JSON format output sequence is generated; The sequence was input into the XGBoost model for incremental training, and the feature selection parameters were adjusted to the optimal split point, which increased the F1 value of behavior recognition to 94.6%; Based on the updated confusion matrix results, key frame identifiers with timestamp errors less than 0.2 seconds are re-extracted to trigger a new round of dynamic adjustment instruction generation.

[0030] like Figure 1-Figure 2 As shown, in step S105, identification information of key teaching moments is obtained from the prediction result set after the behavior recognition accuracy is improved, and the dynamic adjustment capability is triggered by combining real-time control technology with event-driven architecture to generate an instruction set for adjusting the teaching content.

[0031] Furthermore, in step S105, a prediction result set is obtained, and the prediction result set is processed using behavior recognition technology to extract identification information of key moments to obtain preliminary classification data; Based on the preliminary classification data, real-time control technology is used to match the preset event-driven architecture to determine whether the preset trigger conditions are met and determine the start signal for dynamic adjustment; If the start signal of dynamic adjustment is determined, the adjustment capability parameters are generated through the event-driven mechanism to obtain the adjustment requirements of the teaching content; According to the adjustment requirements, the corresponding instruction set is generated in combination with the adjustment capability parameters to obtain the adjustment plan of the teaching content; Extract the core fields of the instruction set from the adjustment plan of the teaching content, judge the integrity of the instruction set through the preset threshold, and determine the final output sequence; After obtaining the final output sequence, a machine learning algorithm is used to optimize the accuracy of the behavior recognition technology to obtain an updated set of prediction results; Based on the updated prediction result set, the attention mechanism is used to enhance the pattern recognition algorithm to process complex teaching scenarios, extract the feature vectors of key teaching moments, and obtain the prediction result set with improved behavior recognition accuracy; Extract key moment identification information from the prediction result set after the behavior recognition accuracy is improved, and generate adjustment capability parameters by combining real-time control technology with event-driven architecture; According to the adjustment ability parameters, the teaching optimization parameter set is generated in combination with the operating status of the closed-loop control system, and the instruction set after precise intervention at the key teaching moment is obtained.

[0032] Specifically, in step S105, after obtaining the set of prediction results, the teaching behavior in the video stream is analyzed using the behavior recognition model based on YOLOv5, the confidence threshold is set to 0.85 to extract key frames, and the behavior data is divided into 5 types of teaching scenes through the time series clustering algorithm to obtain preliminary classification data with timestamps; According to the teacher-student interaction frequency parameters in the classified data, the PID control algorithm is used to match the rule engine in the event-driven architecture. When the interaction frequency is detected to be lower than the preset threshold of 0.6 for three consecutive times, a dynamic adjustment signal is triggered; The adjustment instructions are issued through the event bus, and the weighted sliding average algorithm is used to calculate the change rate of the behavioral characteristics in the last 10 seconds, generating a capability parameter matrix containing the adjustment range (±15%) and direction (content difficulty / presentation speed); Combined with the metadata index of the teaching resource library, the Dijkstra algorithm is used to search for the optimal adjustment path in the knowledge graph and generate an instruction set containing three alternative solutions; Extract the core fields in the instruction set, including knowledge point ID and presentation duration, calculate the semantic completeness score through the BERT model, and output the final sequence when the score exceeds 0.92; Adopting the incremental random forest algorithm, the recognition model was updated with 2,000 newly collected behavior data, increasing the accuracy of key behavior recognition from 86% to 91%. Apply a multi-head attention mechanism (8 heads) to the updated prediction set to compress the feature vector dimension (512 dimensions → 128 dimensions) and enhance the recognition ability of complex scenarios such as group discussions; Extract key segments with a time window of 5 seconds from the optimized feature vector, correct the behavior trajectory prediction in real time through the Kalman filter, and generate adjustment parameters including confidence intervals; Finally, 30 operating indicators in the closed-loop system are integrated, including latency <50ms and throughput >200QPS, and fuzzy logic reasoning is used to generate a teaching optimization parameter set containing priority weights (0-1).

[0033] like Figure 1-Figure 2 As shown, in step S106, specific parameters are extracted from the instruction set for adjusting the teaching content, and the changing trend of the learning behavior data is monitored through a closed-loop control system to obtain an adjustment parameter set synchronized with the high-concurrency data stream.

[0034] Further, in step S106, specific parameters are extracted from the instruction set for adjusting the teaching content to obtain an initial parameter set; The learning behavior data is collected in real time through a closed-loop control system to obtain real-time change trends; Determine whether the real-time change trend exceeds the preset threshold and whether to trigger parameter adjustment; If parameter adjustment is triggered, the parameter extraction algorithm is used to calculate the updated value to obtain the adjustment parameter; According to the state synchronization of high concurrent data streams, adjustment parameters and system monitoring data are processed to obtain synchronized data; The support vector machine algorithm is used to classify the synchronously processed data to obtain the optimized parameter set; Update the instruction set of the teaching content through the optimized parameter set, and obtain the updated instruction set; The updated instruction set is monitored by a closed-loop control system to obtain a final adjustment parameter set synchronized with the data stream; Based on the final adjustment parameter set combined with behavior recognition technology, key moment identification information is extracted to generate the instruction set for the next round of adjustment.

[0035] Specifically, in step S106, specific parameters are extracted from the instruction set for adjusting the teaching content, including parsing parameters such as learning time, knowledge point difficulty coefficient, and interaction frequency from the instruction set to obtain an initial parameter set; The closed-loop control system collects learning behavior data in real time, including the accuracy of students' answers, distribution of learning time, and number of interactions, to obtain real-time change trends. Determine whether the real-time change trend exceeds the preset threshold, including setting the correct answer rate below 60% or the learning time exceeding 120 minutes as abnormal, and determine whether to trigger parameter adjustment; If parameter adjustment is triggered, the parameter extraction algorithm is used to calculate the updated value, including optimizing the difficulty coefficient of the knowledge point through the gradient descent method to obtain the adjustment parameter; Synchronize adjustment parameters and system monitoring data based on the status of high-concurrency data streams, including matching adjustment parameters with real-time monitoring data to obtain synchronized data when processing 1,000 data streams per second; The support vector machine algorithm is used to classify the synchronously processed data, including classifying the data into three categories of high, medium and low learning effects, and obtaining the optimized parameter set; The instruction set of the teaching content is updated through the optimized parameter set, including adjusting the difficulty coefficient of the knowledge point from 0.8 to 0.6, and obtaining the updated instruction set; The updated instruction set is monitored through a closed-loop control system, including real-time monitoring of whether the correct rate of students' answers has increased to 70%, and the final adjustment parameter set synchronized with the data stream is obtained; Based on the final adjustment parameter set combined with behavior recognition technology, key moment identification information is extracted, including identifying that students' learning efficiency peaks at the 15th and 45th minutes, and generating an instruction set for the next round of adjustments.

[0036] like Figure 1-Figure 2As shown, in step S107, the execution state of the dynamic adjustment capability is extracted from the adjustment parameter set synchronized with the high-concurrency data stream, and the precision of the pattern recognition algorithm is optimized by using a reinforcement learning algorithm to generate a feedback loop data stream after the closed-loop control system is optimized.

[0037] Further, in step S107, high-concurrency data is obtained, and an adjustment parameter set and an initial execution state are obtained by synchronously processing the high-concurrency data; Extracting dynamic adjustment capability characteristics from the adjustment parameter set to determine a change trend of the execution state; Using a reinforcement learning algorithm, the pattern recognition algorithm is optimized according to the change trend to obtain parameters with improved accuracy; According to the parameters with improved accuracy, a closed-loop control system is adjusted to generate an optimized feedback data stream; According to the optimized feedback data flow, determining the system adjustment direction and obtaining the adjusted execution state; If the adjusted execution state does not reach a preset threshold, re-optimizing the feedback data stream according to the adjustment parameter set to obtain an updated feedback data stream; According to the updated feedback data stream, determining whether the system has reached a stable state, and obtaining a stable state determination result; According to the stable state determination result, updating the synchronization processing process of the high concurrent data to generate a new data stream of the closed-loop control system; The adjusted dynamic adjustment capability feature is extracted from the new data stream to determine the optimization execution state of the closed-loop control system.

[0038] Specifically, in step S107, when acquiring high-concurrency data, the distributed message queue Kafka is used to process 100,000 real-time data streams per second, and the data timestamps are aligned through the time window synchronization algorithm to generate an adjustment parameter set including a delay threshold, a throughput indicator, and an initial execution state; When extracting the dynamic adjustment capability features from the adjustment parameter set, calculate whether the standard deviation of data delay in the sliding window exceeds 50ms, and determine the change trend based on the fluctuation range of CPU utilization in the execution state; The DQN-based reinforcement learning algorithm was used, and the reward function was set as the percentage of pattern recognition accuracy improvement. The number of convolution kernels and the learning rate of the convolutional neural network were optimized through Q-learning iterations, which increased the classification accuracy from 92% to 96%. According to the optimized parameters, the proportional coefficient of the PID controller of the closed-loop control system is adjusted to generate a feedback data stream containing an error correction value; Analyze the slope of the error convergence curve in the feedback data stream, determine that the system adjustment direction is to increase the weight of the integral link, and obtain the adjusted execution state; If the steady-state error of the execution state is still higher than the preset threshold of 0.5%, a genetic algorithm is used to perform multi-objective optimization on the feedback data stream, with the crossover probability set to 0.8 and the mutation probability set to 0.1, to generate an updated feedback data stream; The variance value of the updated data stream is calculated through Kalman filtering, and the system is judged to be in a stable state when the variance is less than 0.01 for 5 consecutive iterations; Based on the stable state determination result, the data synchronization processing thread pool is reconstructed, the number of threads is dynamically adjusted from 200 to 150, and a data stream containing new sampling period parameters is generated; When extracting the adjusted dynamic adjustment capability features from the new data stream, the LSTM network is used to predict the execution status of the next three cycles and output the response time distribution histogram after system optimization.

[0039] like Figure 1-Figure 2 As shown, in step S108, delay and accuracy indicators are extracted from the feedback loop data stream after the closed-loop control system is optimized, and the execution effect of the dynamic adjustment capability is verified by analysis to obtain a verification data stream after the system performance is improved.

[0040] Further, in step S108, a data stream in the optimized closed-loop system feedback loop is obtained, and a delay index and an accuracy index in the data stream are extracted to obtain a preliminary quantification result; Calculating the distribution characteristics of the delay index and the accuracy index in the preliminary quantification result by using a statistical method to determine the execution effect of the dynamic adjustment; If the delay indicator exceeds a preset threshold, an optimized data stream is obtained by adjusting the parameters of the feedback loop; Extracting new delay indicators and accuracy indicators from the optimized data stream, and using a machine learning algorithm to determine the changing trend of system performance; According to the change trend, by comparing the difference between the new and old accuracy indicators, the verification basis for the improvement of system performance is obtained; After obtaining the verification basis, a preset classification algorithm is used to judge the stability of the performance improvement to obtain a final quantitative data stream; By analyzing the final quantized data flow, the optimization control capability of the closed-loop system is determined to obtain a verification result of the execution effect; Extracting the operating state of the closed-loop control system from the verification results, and generating a teaching optimization parameter set through an event-driven architecture; According to the teaching optimization parameter set, the teaching content is adjusted through an instant feedback mechanism to obtain a data stream after precise intervention at key teaching moments.

[0041] Specifically, in step S108, the data stream in the optimized closed-loop system feedback loop is obtained, and the delay indicators in the data packets are extracted at intervals of 10 ms using a sliding window algorithm, including an average delay of 15 ms and accuracy indicators, including a classification accuracy of 95%, to obtain a preliminary quantitative result including a timestamp and a numerical value; Statistical methods are used to calculate the distribution characteristics of latency and accuracy indicators, including using normal distribution to test the mean of latency indicators (μ=18ms, σ=3ms) and the confidence interval of accuracy indicators (95%±2%), to determine whether the execution effect of dynamic adjustment meets expectations; If the delay indicator is detected to exceed the preset threshold, including 20ms, the proportional gain parameter of the feedback loop is adjusted through the PID controller (Kp is reduced from 1.2 to 0.8) to generate an optimized data stream; Extract updated latency metrics (12ms) and accuracy metrics (96%) from the new data stream, and use the LSTM model to predict performance trends over the next five cycles (error rate <1%). Based on the trend analysis results, the difference between the new and old accuracy indicators (+1%) was compared, and the t-test (p<0.05) was used to verify the significance of the performance improvement; The verification basis is input into the random forest classifier, and the stability level (Grade A) is determined based on the feature importance ranking, including the delay weight of 0.6 and the accuracy weight of 0.4, and the final quantitative data stream is output; The final data stream was processed by principal component analysis (PCA) to reduce the dimension, and the first two principal components (cumulative contribution rate 85%) were extracted to determine the optimization control capability boundary of the closed-loop system (delay <15ms, accuracy >94%). Analyze the system operation status from the verification results, including the load rate of 70%, trigger the teaching parameter generation module through the event-driven architecture, and output the teaching optimization parameter set including the adjustment range (±5%) and response time (200ms); Based on the parameter set, the reinforcement learning strategy (ε-greedy algorithm, ε=0.1) is called to dynamically adjust the frequency of teaching content push (from 1 time / minute to 1.2 times / minute) to generate the intervened data stream for the next round of closed-loop optimization.

[0042] like Figure 1-Figure 2 As shown, in step S109, the operating status of the closed-loop control system is obtained from the verification data stream after the system performance is improved, and the teaching content adjustment and the instant feedback mechanism are seamlessly combined through the event-driven architecture to generate a teaching optimization parameter set after precise intervention at key teaching moments.

[0043] Further, in step S109, the operating status data of the closed-loop control system is extracted from the verification data stream after the system performance is improved, and the system performance index after the data stream is parsed is obtained; According to the dynamic change characteristics in the running status data, determine whether to trigger the key teaching moment; If a key teaching moment is triggered, the event-driven architecture is used to process the dynamic change characteristics and determine the adjustment direction of the teaching content; Integrate the immediate feedback mechanism according to the adjustment direction to generate an intervention strategy set; Analyze the execution effect of the intervention strategy set at key moments and obtain a preliminary set of optimization parameters; The preliminary set is classified and processed by random forest algorithm to obtain the teaching optimization parameter set; Determine whether the matching degree of the teaching optimization parameter set is lower than a preset threshold, and obtain a matching degree analysis result; If the matching degree is lower than the preset threshold, the teaching optimization parameter set is iteratively adjusted through the gradient descent algorithm to obtain the adjusted parameter set; The adjusted parameter set is monitored through a closed-loop control system to obtain the final teaching optimization parameter set synchronized with the data stream.

[0044] Specifically, in step S109, the operating status data of the closed-loop control system is extracted from the verification data stream after the system performance is improved, and the data stream parsing algorithm is used, including sliding window analysis, with a window size of 5 seconds and a sampling frequency of 100Hz to calculate the system performance indicators, including response delay, average delay ≤50ms and throughput ≥1000 items / second; Based on the dynamic change characteristics in the running status data, including the fluctuation range of learning behavior data exceeding ±15%, the threshold judgment model is used to set the confidence level of the threshold to 0.7 to detect whether the key teaching moment is triggered; If a critical teaching moment is triggered, the event-driven architecture, including the Kafka message queue to handle dynamic change characteristics, combined with LSTM time series analysis, determines the direction of teaching content adjustment, including reducing the difficulty coefficient by 0.2; Integrate the instant feedback mechanism according to the adjustment direction, including real-time collection of students' correct answer rate, with a sampling interval of 2 seconds, and generate an intervention strategy set containing 3 strategies, including adjusting the explanation speed, increasing the number of examples, and switching the media format; For the intervention strategy set, an effect evaluation model was used, including weighted scores based on accuracy, time consumption, and interaction rate, with weights of 0.5, 0.3, and 0.2, respectively, to analyze the execution effect at key moments and output a preliminary set of optimization parameters, including a parameter range of [0.1, 0.9]; Using the random forest algorithm, 100 decision trees with a maximum depth of 10 layers were set up to classify the preliminary set and select the teaching optimization parameter set, including the optimal parameter combination of 0.6 / 0.8 / 0.4; The matching degree of the teaching optimization parameter set is determined by cosine similarity calculation with a threshold of 0.85. If the matching degree is lower than the threshold, the gradient descent algorithm is applied with a learning rate of 0.01 and the parameter set is adjusted for 50 iterations until the error rate is ≤5%. Finally, the adjusted parameter set was monitored through the PID controller of the closed-loop control system with a proportional coefficient of 1.2 and an integral time of 0.5 seconds, and the final teaching optimization parameter set synchronized with the data stream was achieved, including a dynamic adjustment frequency of 1 Hz.

[0045] For those skilled in the art, various other corresponding changes and deformations can be made according to the technical solutions and concepts described above, and all of these changes and deformations should fall within the protection scope of the claims of this application.

Claims

1. A teaching behavior pattern recognition method based on big data technology, characterized in that: The following steps are involved: S101. Slice the high-concurrency data stream of learning behavior data through distributed stream processing technology, judge the real-time performance based on the preset delay threshold and throughput threshold, and generate an optimized slicing data stream; S102. Using an incremental computing method to dynamically adjust the processing capacity of the sharded data stream, and allocate computing resources through a parallel computing framework to obtain a balanced sharded data stream; S103. Extracting action features and time features from the balanced sharded data stream, training a behavior recognition model using a deep learning model, and generating a set of feature vectors for key teaching moments; S104. Performing pattern recognition enhancement on the feature vector set based on the attention mechanism to improve the accuracy of behavior recognition and outputting a prediction result set; S105. Combining real-time control technology with event-driven architecture, triggering dynamic adjustment instructions according to the prediction result set, and generating an instruction set for adjusting the teaching content.

2. According to claim 1, a teaching behavior pattern recognition method based on big data technology is characterized in that: Monitoring the execution status of the instruction set through a closed-loop control system for synchronously adjusting parameters to match the changing trend of the high concurrent data stream; Reinforcement learning algorithm is used to optimize the accuracy of the behavior recognition model and obtain feedback loop data flow; Extract latency and accuracy metrics from the feedback loop data stream to verify the system performance improvement effect; A teaching optimization parameter set is generated based on the verification results and used for intervention at key teaching moments through an event-driven architecture.

3. According to the teaching behavior pattern recognition method based on big data technology as described in claim 1, it is characterized in that: In the steps S101-S102, the initial sharded data stream is redistributed by a consistent hashing algorithm, and if the delay exceeds a preset threshold or the throughput is lower than a threshold, the sharding strategy is dynamically adjusted; The support vector machine algorithm is used to classify the fragmented data stream in real time, and the classification confidence threshold is set to 0.85; The parallel computing framework adopts a dynamic resource scheduling strategy, including: When the instantaneous concurrent volume exceeds 80,000 per second, the computing node is triggered to expand to 1.5 times the original resources; Adjust the data localization strategy through the stream processing backpressure monitoring mechanism to ensure that the network IO utilization is less than 75%.

4. According to claim 1, a teaching behavior pattern recognition method based on big data technology is characterized in that: In step S103, the deep learning model includes: Convolutional neural networks are used to extract multidimensional spatial features, and bidirectional LSTM networks are used to analyze temporal features; When the cosine similarity between action features and time features exceeds 0.85, it is marked as a candidate set of key teaching moments; The DBSCAN clustering algorithm is used to group the candidate sets, and the outliers are removed through Kalman filtering.

5. According to claim 1, a teaching behavior pattern recognition method based on big data technology is characterized in that: In step S104, the attention mechanism is a multi-head Transformer model, including: Set 8 attention heads to assign weights to feature vectors; The enhanced feature representation is generated through Softmax normalization, and the accuracy of behavior recognition is improved to 92.3%.

6. According to claim 1, a teaching behavior pattern recognition method based on big data technology is characterized in that: In step S105, the triggering conditions of the dynamic adjustment instruction include: When the teacher-student interaction frequency is lower than 0.6 for three consecutive times, instructions are generated including knowledge point reorganization and speech speed adjustment; The decision tree model is used to calculate the adjustment weights and output the instruction sequences with integrity scores exceeding 0.85 in JSON format.

7. According to claim 2, a teaching behavior pattern recognition method based on big data technology is characterized in that: The closed-loop control system adopts a PID controller, including: Collect the changing trend of learning behavior data in real time. If the correct answer rate is lower than 60% or the learning time exceeds 120 minutes, the parameter adjustment will be triggered. The parameters after synchronization processing are classified through the support vector machine algorithm, and the optimized adjustment instruction set is output.

8. According to claim 2, a teaching behavior pattern recognition method based on big data technology is characterized in that: The reinforcement learning algorithm is a deep Q network DQN, including: The percentage increase in pattern recognition accuracy is used as the reward function to optimize the number of convolution kernels and learning rate of the convolutional neural network; When the steady-state error exceeds 0.5%, a genetic algorithm is used to perform multi-objective optimization on the feedback data stream.

9. According to claim 2, a teaching behavior pattern recognition method based on big data technology is characterized in that: The system performance verification includes: The LSTM model is used to predict the latency and accuracy indicator trends in the next 5 cycles; The t-test with p<0.05 is used to verify the significance of the performance improvement, and a quantitative data stream with a stability level of A is output.

10. According to claim 2, a teaching behavior pattern recognition method based on big data technology is characterized in that: The generation of the teaching optimization parameter set includes: The optimal parameter combination was screened by the random forest algorithm, and the matching threshold was set to 0.85; If the matching degree does not reach the threshold, the gradient descent algorithm is used to iteratively adjust the parameters until the error rate is ≤5%.

Images