A cloud platform-based course achievement monitoring and analysis system

Abstract

The application discloses a course achievement monitoring and analyzing system based on a cloud platform, relates to the technical field of educational big data analysis and intelligent decision-making, and comprises a course data collecting module, a course feature extracting module, a prediction correlating module, a decision scheduling module and an analyzing report module; course data is collected, a course achievement feature library is constructed, the prediction correlating module outputs a knowledge point mastering probability and a collaborative influence weight matrix, the decision scheduling module divides a confidence level and allocates teaching resources, and the analyzing report module performs cloud early warning and pushes a behavior analyzing report and a PDF diagnostic report. The application realizes comprehensive monitoring and analyzing of course achievements through the cloud platform, effectively improves the timeliness and accuracy of data processing, provides a scientific basis for teaching decision-making, promotes personalized learning, timely early warns and optimizes the allocation of teaching resources, significantly improves the teaching quality and learning effectiveness, and has high practicability and intelligent level.

Description

Technical Field

[0001] This invention relates to the field of educational big data analysis and intelligent decision-making technology, specifically to a cloud-based system for monitoring and analyzing course achievement. Background Technology

[0002] Course achievement monitoring and analysis is a systematic evaluation method designed to measure and ensure the effective achievement of educational curriculum goals. It involves the continuous tracking and in-depth analysis of various performance data of students during the learning process. This data includes, but is not limited to, exam scores, classroom participation, homework completion, and project report quality. Through this analysis process, educators can clearly understand the extent to which students have mastered the course content at each learning stage, as well as the gap between course goals and actual learning outcomes. This not only helps to identify problems and deficiencies in teaching in a timely manner, but also provides a strong basis for adjusting teaching strategies and optimizing course content. In short, course achievement monitoring and analysis is an important link in improving teaching quality and ensuring the effective achievement of course goals. It requires educators to have keen data insight, be able to accurately interpret the analysis results, and make scientific and reasonable decisions accordingly.

[0003] To address the issues of lagging knowledge point correlation analysis and intervention measures in course achievement monitoring and analysis, existing technologies rely on traditional manual data collection and periodic analysis reports. However, these methods suffer from untimely data updates and insufficient analytical accuracy, leading to educational decisions lagging behind students' actual needs and feedback on learning outcomes. Traditional monitoring and analysis methods often depend on teachers' personal experience and judgment, lacking real-time and comprehensive data support, making it difficult to accurately capture differences in students' mastery of different knowledge points and potential learning obstacles. Furthermore, the implementation of intervention measures is often passive and delayed due to the lack of an immediate feedback mechanism, failing to respond quickly to changes in students' learning, thus affecting the improvement of teaching effectiveness and the promotion of personalized learning. Therefore, a cloud-based course achievement monitoring and analysis system is proposed to overcome these limitations. Summary of the Invention

[0004] The purpose of this invention is to provide a cloud-based course achievement monitoring and analysis system to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a course achievement monitoring and analysis system based on a cloud platform, including a course data acquisition module, a course feature extraction module, a prediction and correlation module, a decision scheduling module, and an analysis report module; The course data acquisition module collects and preprocesses students' course objective data and learning behavior data. The course feature extraction module extracts features from the preprocessed course target data and learning behavior data to build a feature library of course achievement. The prediction and association module, in conjunction with the feature library of course achievement, constructs a mastery trend prediction model and a knowledge point association analysis topology map, and outputs the probability prediction results of knowledge point mastery and the synergistic influence weight matrix of related knowledge point groups, respectively. The decision scheduling module divides the confidence level according to the probability prediction results, performs binarization processing on the collaborative influence weight matrix, constructs a decision rule base, drives the early warning level decision, allocates teaching resources, and outputs a list of related knowledge points to be reinforced. The analysis report module implements early warnings based on a cloud-based rule engine and pushes relevant student behavior analysis reports, generating a heatmap of course goal achievement and a PDF diagnostic report.

[0006] A further improvement to the technical solution of this invention lies in that: the collection and preprocessing process of student course objective data and learning behavior data in the course data acquisition module includes: The cloud data synchronization node is deployed in the virtual private network of the cloud platform, and connects to the academic affairs system database through a secure interface. Automatic synchronization tasks are configured to obtain teaching syllabus metadata from the academic affairs system database and store it in the cloud relational database. The teaching syllabus metadata includes knowledge point number, chapter weight and target threshold. A lightweight proxy service is deployed on the online teaching platform to intercept chapter test submission requests in real time, extract fields including student identifier, knowledge point number, number of correct answers and total number of questions, and collect the original record data of chapter tests through the proxy service; IoT attendance terminals are deployed in physical classrooms to record student attendance data based on facial recognition technology. Distributed log agents are deployed on the server cluster of the online learning platform to capture the start and end timestamps of video viewing events in real time, collect video viewing completeness data, and capture document download records. The number of downloads within 24 hours is aggregated by student identifier to collect document download frequency statistics. For chapter test original records that were submitted interrupted, the mean of adjacent time periods was used to fill in the missing data. Chapter test original records with a single knowledge point test achievement rate exceeding ±3 times the standard deviation were marked as invalid. Discrete behavior logs were divided into time-series segments by 5-minute windows, and the missing window data was filled with the mean of the preceding window. Attendance count and download frequency statistics were mapped to the [0,1] interval.

[0007] A further improvement to the technical solution of this invention lies in that: the process of extracting features from the preprocessed course target data in the course feature extraction module includes: The weights of each chapter are obtained from the metadata of the teaching syllabus, and the test achievement rates under the same knowledge point are weighted and summed to obtain the mastery of a single knowledge point. The time series is divided into natural weeks, and the average achievement rate is calculated weekly. The average achievement rate results of four consecutive weeks are used to form a trend vector. The influence of the dimension is eliminated by normalization to obtain the weekly dimension achievement rate change trend. When the data of a new week arrives, the data of the earliest week is deleted and the trend vector is recalculated. Based on the achievement rate sequence of knowledge points in the test, the Pearson correlation coefficient of the achievement rate is calculated pairwise. An n×n knowledge point correlation matrix C, reflecting the statistical correlation between knowledge points, is generated. The calculation process is as follows: ; ; in, and Let A and B be the achievement rates of the i-th student on knowledge points A and B, respectively. and These represent the average achievement rates of students on knowledge points A and B, respectively.

[0008] A further improvement to the technical solution of this invention lies in that: the process of extracting features from the preprocessed learning behavior data in the course feature extraction module includes: Constructing a three-dimensional matrix The dataset contains N students, K knowledge points, and each element includes the mean video viewing completion rate and document download frequency density. A matrix is ​​used to map the mastery of related knowledge points. This forms a mapping relationship between behavior and goals; Perform high-order tensor decomposition on the behavior matrix B to extract latent factor vectors for students, knowledge points, and behavior patterns. Based on the decomposition results, calculate the weighted similarity between the knowledge point behavior factors and the mastery vector. The correlation between resource access and control is obtained, a loss function L is designed, and the optimal factor importance weights are iteratively solved using the gradient descent method. To make the correlation degree approach the average of the actual mastery degree, a regularization coefficient is introduced, the calculation process of which is as follows: ; ; in, The number of dimensions of the latent factors. Let k be the projection of knowledge point k onto the r-th factor. This is a column vector representing the students' mastery of knowledge point k. The regularization coefficient is used. By integrating the extracted features of single knowledge point mastery, weekly achievement rate trends, knowledge point correlation matrix, and correlation between resource access and mastery, a feature library of course achievement is constructed.

[0009] A further improvement to the technical solution of this invention lies in the following: the process of constructing a trend prediction model and outputting a probability prediction result of knowledge point mastery in the prediction association module includes: The missing values ​​of the single knowledge point mastery sequence are filled by aligning them by day, and the weekly achievement rate change trend vector is matched with the daily data to construct an input feature vector containing historical single knowledge point mastery and weekly achievement rate change trend components. Based on a temporal convolutional network architecture, dilated causal convolutional kernels are used to process historical feature sequences, and the receptive field is expanded layer by layer to capture long-term and short-term dependencies. A trend prediction model is constructed, and the feature map output of the l-th layer is calculated. Attention weights of each component of the weekly trend vector The weighted fusion is then concatenated with the convolution output to obtain the fused feature vector. Output the future through a fully connected layer Daily success rate The confidence interval is calculated as follows: ; ; ; ; in, Let be the weight matrix of the l-th convolutional kernel. The input feature map is the (l-1)th layer. Let l be the bias vector of the l-th layer. To modify the linear unit activation function, q is a learnable query vector. Let be the embedding representation of the trend component of the achievement rate change in the i-th week. This is a vector concatenation operation. To understand the output characteristics of the last layer of the trend prediction model, This is the weight matrix of the output layer. This is the bias vector for the output layer. This is a normalized exponential function that maps the output to a probability distribution; We design a weighted cross-entropy loss function, prioritize assigning weights to long-term predictions, and dynamically adjust the learning rate using a cosine annealing strategy to optimize the trend prediction model.

[0010] A further improvement to the technical solution of this invention lies in the following: the process of constructing a knowledge point association analysis topology graph and outputting the synergistic influence weight matrix of the associated knowledge point group in the prediction association module includes: Based on the knowledge point correlation matrix features, the correlation features of resource access and mastery, and the graph convolutional network algorithm, a topology graph structure is initialized, with each knowledge point as a node in the topology graph, and the total number of nodes n corresponding to the total number of knowledge points. Edges are defined. The initial weights are Add a resource access correlation attribute to each node k. Construct a topology diagram for knowledge point association analysis; Dynamically adjust the edges in the topology graph of knowledge point association analysis initial weights By integrating the resource correlation attributes of the two endpoints, the corrected edge weights are globally normalized to obtain the optimized edge weights. The system performs neighborhood aggregation on node features, captures high-order relationships between knowledge points through stacked graph convolutional layers, and constructs a collaborative influence weight matrix based on optimized edge weights. , among which, element Remove Weakly related edges with a value below 0.3 are identified, and the collaborative influence weight matrix is ​​symmetrically processed.

[0011] A further improvement to the technical solution of this invention lies in the following: In the decision scheduling module, the process of dividing confidence levels according to probability prediction results, binarizing the collaborative influence weight matrix, and constructing a decision rule base includes: Based on the future Daily success rate Calculate and based on the mean of the probability of meeting the standard with standard deviation Distinguish between different confidence levels for the probability of meeting the standard. Then the probability of meeting the standard is determined to be in the high confidence interval. Then the probability of meeting the standard is determined to be within the middle confidence interval. Then the probability of meeting the standard is determined to be in the high confidence interval; The collaborative influence weight matrix is ​​binarized, and a weight threshold of 0.5 times the median of the historical edge weights is set. ,like Then mark the corresponding strongly related edge as 1, if Then the corresponding weakly correlated edges are marked as 0, and the collaborative influence weight matrix is ​​forced to be symmetric, generating a binary matrix containing only 0 and 1 elements; Construct a decision rule base, define early warning rules, and determine the probability of predicting that knowledge point k meets the target. It is in the low confidence interval and exists If the core related knowledge point j is identified, a red alert is issued, freezing the learning path for the related chapter and pushing customized exercises and video resources to the student. It falls within the low confidence interval and the medium confidence interval, but the related knowledge points satisfy... If it is, it will be judged as a yellow alert, requiring the completion of related tests within a limited time, triggering a learning progress reminder. It is within the middle confidence interval and the related knowledge points satisfy If it is determined to be a blue alert, the corresponding knowledge point is marked as the observation object, and a list of related resource recommendations is generated. It is in the high confidence interval and the related knowledge points satisfy If the condition is met, it will be considered a green alert, unlocking advanced learning content, awarding a learning achievement badge, allowing students to choose additional resources, and dynamically optimizing rule weights and threshold parameters through historical feedback. Input the real-time achievement probability prediction result and the binarized matrix, traverse the knowledge point-association pairs, match the rule base conditions, and execute the intervention action according to the priority of red, yellow, blue and green after matching the rule conditions. If multiple rules are triggered, the highest priority is executed. For cyclically dependent associated knowledge point groups, the subgraph segmentation algorithm is used to implement joint intervention.

[0012] A further improvement to the technical solution of this invention lies in the following: the process of allocating teaching resources and outputting a list of related knowledge points for reinforcement in the decision-making and scheduling module includes: Based on the early warning level matching resource pool, the resource pool is set with a semantic similarity weight of 0.7 and a historical effect weight of 0.3. A resource score F is calculated based on the semantic similarity weight and the historical effect weight to select the optimal resources. Quota limits are set according to the warning level: a maximum of 5 teaching resources are allocated to a red warning, a maximum of 3 teaching resources to a yellow warning, and a maximum of 1 teaching resource to a blue or green warning. High-priority needs are given priority access to highly relevant teaching materials. The calculation process is as follows: ; in, Let be the semantic similarity between resource j and knowledge point k. For the historical usage efficiency of resource j; Extract strongly correlated subgraphs from the synergistic influence matrix, subtract the normalized mastery degree from 1, and add the rate of decrease of the normalized pass probability to obtain the weakness index of each knowledge point in the strongly correlated subgraph. Sort the knowledge points in the top 5 strongly correlated subgraphs from high to low according to the weakness index, generate a reinforcement list, and recommend the most effective exercises, videos and documents in history. A formatted resource list is pushed to student terminals via a message queue. The formatted resource list includes frozen chapters, recommended resources, and priorities. The system continuously monitors changes in knowledge mastery over 3 days. If the improvement is less than 5%, the warning level is automatically raised and resource allocation is retried.

[0013] A further improvement to the technical solution of this invention lies in the following: In the analysis report module, the process of implementing early warnings and pushing related student behavior analysis reports based on the cloud-based rule engine includes: Based on the warning level, the list of affected knowledge points and associated student identifiers are extracted. Student behavior logs, time series of single knowledge point mastery, and subgraph data of synergistic influence weight matrix are aggregated from the course achievement feature library to form an analysis input set. The student behavior logs include video viewing completeness, document download frequency, and attendance rate. By combining the mastery of resources invested per unit of time to improve efficiency, detecting abnormal behavior patterns, matching inefficient learning labels through a decision rule base, quantifying the Pearson correlation coefficient between behavior and performance, and identifying key weaknesses, the abnormal behavior patterns include inefficient practice and discrepancies between classroom learning and performance. Match abnormal behavior patterns with the decision rule base to generate a structured report. The structured report includes a heatmap of weak knowledge points, a behavior-performance correlation matrix, and a resource efficiency ranking. It encapsulates a list of weak knowledge points, inefficient behavior tags, conversion rate indicators, and corresponding suggested measures in JSON format. Based on the warning level, relevant student behavior analysis reports are pushed out. Red warnings are notified in real time, yellow warnings are summarized daily, and blue warnings are reported to classes weekly. Data on report review rate and the improvement in the probability of achieving the target after intervention are collected.

[0014] A further improvement to the technical solution of this invention lies in the following: the process of generating a heatmap of course objective achievement and a PDF diagnostic report in the analysis report module includes: Based on the knowledge point mastery and relevance matrix, the comprehensive popularity value of each knowledge point is analyzed, normalized and mapped to the red-yellow-green color space, a two-dimensional matrix is ​​constructed, and an interactive goal achievement heatmap is generated by class and student groups to show the achievement status and relevance strength of each knowledge point. Integrating global goal achievement rate, goal achievement heatmap, abnormal behavior analysis, and weekly dimension achievement rate trend prediction, the text, charts, and formulas are formatted by chapter, and a PDF diagnostic report is output. The PDF diagnostic report includes a summary, data tables, and customized suggestions, supporting multi-dimensional teaching assessment. The latest collected data is extracted from the course achievement feature database, injected into the PDF diagnostic report template according to priority, the data snapshot version number is recorded, the content of the PDF diagnostic report is synchronized with the real-time teaching status, and the analysis results of each stage are retained.

[0015] Due to the adoption of the above technical solution, the technical progress achieved by this invention compared to the prior art is as follows: 1. This invention provides a cloud platform-based course achievement monitoring and analysis system that can collect and preprocess student data in real time, build an accurate feature database of course achievement, provide timely and comprehensive data support for educational decision-making, and significantly improve the efficiency and scientific nature of teaching management.

[0016] 2. This invention provides a cloud platform-based course achievement monitoring and analysis system. By constructing a mastery trend prediction model and a knowledge point association analysis topology map through a prediction and correlation module, it predicts students' mastery of knowledge points, provides teachers with targeted teaching strategy adjustment suggestions, and promotes the realization of personalized learning.

[0017] 3. This invention provides a cloud-based course achievement monitoring and analysis system. It uses a decision scheduling module to intelligently allocate teaching resources, outputs a list of related knowledge points to be reinforced, and generates detailed diagnostic reports through an analysis report module. This helps students and teachers to identify problems in a timely manner, optimize learning paths, and improve learning outcomes. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a block diagram of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Examples, such as Figure 1 As shown, the present invention provides a cloud platform-based course achievement monitoring and analysis system, including a course data acquisition module, a course feature extraction module, a prediction and correlation module, a decision scheduling module, and an analysis report module; The course data acquisition module collects and preprocesses students' course objective data and learning behavior data. It deploys cloud data synchronization nodes within a cloud platform's virtual private network, connects to the academic affairs system database via a secure interface, configures automatic synchronization tasks, retrieves syllabus metadata from the academic affairs system database, and stores it in a cloud relational database. This syllabus metadata includes knowledge point numbers, chapter weights, and target thresholds. A lightweight proxy service is deployed on the online teaching platform to intercept chapter test submission requests in real time, extracting fields including student identifier, knowledge point number, number of correct answers, and total number of questions. The proxy service collects raw chapter test record data. An IoT check-in terminal is deployed in the physical classroom, based on human... Facial recognition technology records student attendance data. A distributed log agent is deployed on the online learning platform server cluster to capture the start and end timestamps of video viewing events in real time, collect video viewing completeness data, and capture document download records. The number of downloads within 24 hours is aggregated by student identifier, and document download frequency statistics are collected. For chapter test original records that were submitted interrupted, the mean of adjacent time periods is used to fill in the missing data. Chapter test original records with a single knowledge point test achievement rate exceeding ±3 times the standard deviation are marked as invalid. Discrete behavior logs are divided into time series segments by 5-minute windows, and the missing window data is filled with the mean of the preceding window. The attendance count and download frequency statistics are mapped to the [0,1] interval. The course feature extraction module extracts features from preprocessed course objective data and learning behavior data, constructs a feature library of course achievement, obtains the weights of each chapter from the syllabus metadata, and performs a weighted summation of test achievement rates for the same knowledge point to obtain the mastery level of a single knowledge point. It then segments the time series by natural weeks, calculates the weekly average achievement rate, and constructs a trend vector from the weekly average achievement rates over four consecutive weeks. Normalization is then used to eliminate the influence of dimensions, revealing the weekly trend of achievement rate changes. When new week's data arrives, the earliest week's data is deleted, and the trend vector is recalculated. Based on the achievement rate sequence of knowledge points in tests, the Pearson correlation coefficient is calculated for each pair of achievement rates. An n×n knowledge point correlation matrix C, reflecting the statistical correlation between knowledge points, is generated. The calculation process is as follows: ; ; in, and Let A and B be the achievement rates of the i-th student on knowledge points A and B, respectively. and Construct a three-dimensional matrix based on the average achievement rates of students on knowledge points A and B, respectively. The dataset contains N students, K knowledge points, and each element includes the mean video viewing completion rate and document download frequency density. A matrix is ​​used to map the mastery of related knowledge points. This process establishes a mapping relationship between behavior and goals. A higher-order tensor decomposition is performed on the behavior matrix B to extract latent factor vectors representing students, knowledge points, and behavioral patterns. Based on the decomposition results, a weighted similarity between the knowledge point behavioral factors and the mastery vector is calculated. The correlation between resource access and control is obtained, a loss function L is designed, and the optimal factor importance weights are iteratively solved using the gradient descent method. To make the correlation degree approach the average of the actual mastery degree, a regularization coefficient is introduced, the calculation process of which is as follows: ; ; in, The number of dimensions of the latent factors. Let k be the projection of knowledge point k onto the r-th factor. This is a column vector representing the students' mastery of knowledge point k. As the regularization coefficient, the extracted features of single knowledge point mastery, weekly achievement rate change trend, knowledge point correlation matrix and resource access and mastery correlation are integrated to construct a course achievement feature library; The prediction and association module, combined with a feature library of course achievement, constructs a mastery trend prediction model and a knowledge point association analysis topology map. It outputs the probability prediction results of knowledge point mastery and the synergistic influence weight matrix of associated knowledge point groups. For single knowledge point mastery sequences, it fills in missing values ​​by day-alignment. It matches the weekly achievement rate change trend vector with daily data to construct an input feature vector containing historical single knowledge point mastery and weekly achievement rate change trend components. Based on a temporal convolutional network architecture, it uses dilated causal convolutional kernels to process historical feature sequences, progressively expanding the receptive field to capture long-term and short-term dependencies, constructing a mastery trend prediction model, and calculating the feature map output from layer l. Attention weights of each component of the weekly trend vector The weighted fusion is then concatenated with the convolution output to obtain the fused feature vector. Output the future through a fully connected layer Daily success rate The confidence interval is calculated as follows: ; ; ; ; in, Let be the weight matrix of the l-th convolutional kernel. The input feature map is the (l-1)th layer. Let l be the bias vector of the l-th layer. To modify the linear unit activation function, q is a learnable query vector. Let be the embedding representation of the trend component of the achievement rate change in the i-th week. This is a vector concatenation operation. To understand the output characteristics of the last layer of the trend prediction model, The weight matrix of the output layer. This is the bias vector for the output layer. To normalize the exponential function, the output is mapped to a probability distribution. A weighted cross-entropy loss function is designed, prioritizing weights for long-term predictions. A cosine annealing strategy is used to dynamically adjust the learning rate, optimizing the mastery trend prediction model. Based on the knowledge point correlation matrix features, the correlation features between resource access and mastery, and a graph convolutional network algorithm, a topology graph structure is initialized, with each knowledge point as a node. The total number of nodes, n, corresponds to the total number of knowledge points. Edges are defined. The initial weights are Add a resource access correlation attribute to each node k. Construct a knowledge point association analysis topology graph and dynamically adjust the edges in the knowledge point association analysis topology graph. initial weights By integrating the resource correlation attributes of the two endpoints, the corrected edge weights are globally normalized to obtain the optimized edge weights. The system performs neighborhood aggregation on node features, captures high-order relationships between knowledge points through stacked graph convolutional layers, and constructs a collaborative influence weight matrix based on optimized edge weights. , among which, element Remove Weakly related edges with a value below 0.3, and the collaborative influence weight matrix is ​​symmetrically processed; The decision-making and scheduling module categorizes confidence levels based on probability prediction results, binarizes the synergistic influence weight matrix, constructs a decision rule base, drives early warning level decisions, allocates teaching resources, outputs a list of related knowledge points for reinforcement, and is based on future... Daily success rate Calculate and based on the mean of the probability of meeting the standard with standard deviation Distinguish between different confidence levels for the probability of meeting the standard. Then the probability of meeting the standard is determined to be in the high confidence interval. Then the probability of meeting the standard is determined to be within the middle confidence interval. If the probability of meeting the standard is determined to be in the high confidence interval, the collaborative influence weight matrix is ​​binarized, and a weight threshold of 0.5 times the median of the historical edge weights is set. ,like Then mark the corresponding strongly related edge as 1, if Then, the corresponding weakly correlated edges are marked as 0, and the collaborative influence weight matrix is ​​forced to be symmetric, generating a binary matrix containing only 0 and 1 elements. A decision rule base is constructed, and early warning rules are defined. If the predicted probability of knowledge point k meeting the standard is... It is in the low confidence interval and exists If the core related knowledge point j is identified, a red alert is issued, freezing the learning path for the related chapters and pushing customized exercises and video resources to the student. It falls within the low confidence interval and the medium confidence interval, but the related knowledge points satisfy... If it is, it will be judged as a yellow alert, requiring the completion of related tests within a limited time, triggering a learning progress reminder. It is within the middle confidence interval and the related knowledge points satisfy If it is determined to be a blue alert, the corresponding knowledge point is marked as the observation object, and a list of related resource recommendations is generated. It is in the high confidence interval and the related knowledge points satisfy If a warning is triggered, it is considered a green alert, unlocking advanced learning content, awarding a learning achievement badge, allowing students to choose extended resources, and dynamically optimizing rule weights and threshold parameters through historical feedback. The system inputs real-time achievement probability prediction results and a binary matrix, traverses knowledge point-association pairs, matches rule base conditions, and executes intervention actions according to red, yellow, blue, and green priorities. If multiple rules are triggered, the highest priority rule is executed. For cyclically dependent related knowledge point groups, a subgraph segmentation algorithm is used for joint intervention. A resource pool is matched based on the warning level, with a semantic similarity weight of 0.7 and a historical effect weight of 0.3. A resource score F is calculated based on the semantic similarity weight and historical effect weight, and the optimal resource is selected. Quota limits are set according to the level: a maximum of 5 teaching resources are allocated for a red alert, a maximum of 3 for a yellow alert, and a maximum of 1 for both blue and green alerts. High-priority needs are given priority access to highly relevant teaching materials. The calculation process is as follows: ; in, Let be the semantic similarity between resource j and knowledge point k. To determine the historical effectiveness of resource j, a strongly correlated subgraph is extracted from the synergistic influence matrix. The normalized mastery level is subtracted from 1, and the rate of decrease of the normalized pass rate is added to obtain the weakness index of each knowledge point within the strongly correlated subgraph. Knowledge points are sorted from high to low according to their weakness index, and the top 5 strongly correlated subgraphs are selected to generate a reinforcement list. This list is then associated with and recommended exercises, videos, and documents with the highest historical effectiveness. A formatted resource list is pushed to student terminals via a message queue. This formatted resource list includes frozen chapters, recommended resources, and priorities. The change in knowledge point mastery level is continuously monitored over 3 days. If the improvement is less than 5%, the warning level is automatically raised, and resource allocation is retried. The analysis report module, based on a cloud-based rule engine, implements early warnings and pushes related student behavior analysis reports, generating a course goal achievement heatmap and a PDF diagnostic report. Based on the warning level, it extracts a list of affected knowledge points and associated student identifiers. It aggregates student behavior logs, single knowledge point mastery time-series sequences, and synergistic influence weight matrix subgraphs from a course achievement feature library to form an analysis input set. The student behavior logs include video viewing completion rate, document download frequency, and attendance rate. It combines the mastery efficiency improvement based on unit time resource investment, detects abnormal behavior patterns, matches inefficient learning tags through a decision rule library, quantifies the Pearson correlation coefficient between behavior and performance, and identifies key weaknesses. Abnormal behavior patterns include inefficient practice questions and discrepancies between classroom learning and performance. Matching abnormal behavior patterns with the decision rule library generates a structured report. This structured report includes a knowledge point weakness heatmap, a behavior-performance correlation matrix, and a resource efficiency ranking. It encapsulates the list of weak knowledge points, inefficient behavior tags, conversion rate indicators, and corresponding data in JSON format. The recommended measures include: graded push notifications of related student behavior analysis reports based on warning levels; real-time notifications for red warnings; daily summaries for yellow warnings; and weekly class reports for blue warnings. Data on report review rates and the improvement in the probability of achieving goals after intervention are collected. Based on a matrix of knowledge point mastery and relevance, the comprehensive popularity value of each knowledge point is analyzed, normalized, and mapped to a red-yellow-green color space. A two-dimensional matrix is ​​constructed, and interactive goal achievement heatmaps are generated by class and student groups, displaying the achievement status and relevance strength of each knowledge point. Global goal achievement rate, goal achievement heatmaps, abnormal behavior analysis, and weekly achievement rate trend predictions are integrated. Text, charts, and formulas are formatted by chapter, and a PDF diagnostic report is output. This PDF diagnostic report includes a summary, data tables, and customized suggestions, supporting multi-dimensional teaching evaluation. The latest collected data is extracted from the course achievement feature library, injected into the PDF diagnostic report template according to priority, and the data snapshot version number is recorded. The content of the PDF diagnostic report is synchronized with the real-time teaching status, and the analysis results of each stage are retained.

[0022] First, the course data acquisition module collects and preprocesses students' course objectives and learning behavior data. Next, the course feature extraction module extracts features from this preprocessed data to build a feature library of course achievement. Then, the prediction and association module uses this feature library to build a mastery trend prediction model and a knowledge point association analysis topology map, outputting the probability prediction of knowledge point mastery and the synergistic influence of related knowledge point groups. The decision scheduling module divides the confidence level according to the prediction results, binarizes the weight matrix, builds a decision rule library, and makes early warning level decisions based on this, allocates teaching resources, and generates a list of related knowledge point reinforcement. Finally, the analysis report module sends a student behavior analysis report based on the cloud rule engine, and generates a course objective achievement heatmap and PDF diagnostic report for teachers to refer to and adjust their teaching strategies.

[0023] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A cloud-based course achievement monitoring and analysis system, characterized in that: It includes a course data acquisition module, a course feature extraction module, a prediction and correlation module, a decision scheduling module, and an analysis report module; The course data acquisition module collects and preprocesses students' course objective data and learning behavior data. The course feature extraction module extracts features from the preprocessed course target data and learning behavior data to build a feature library of course achievement. The prediction and association module, in conjunction with the feature library of course achievement, constructs a mastery trend prediction model and a knowledge point association analysis topology map, and outputs the probability prediction results of knowledge point mastery and the synergistic influence weight matrix of related knowledge point groups, respectively. The decision scheduling module divides the confidence level according to the probability prediction results, performs binarization processing on the collaborative influence weight matrix, constructs a decision rule base, drives the early warning level decision, allocates teaching resources, and outputs a list of related knowledge points to be reinforced. The analysis report module implements early warnings based on a cloud-based rule engine and pushes relevant student behavior analysis reports, generating a heatmap of course goal achievement and a PDF diagnostic report.

2. The cloud-based course achievement monitoring and analysis system according to claim 1, characterized in that: The course data acquisition module includes the following processes for collecting and preprocessing student course objective data and learning behavior data: The cloud data synchronization node is deployed in the virtual private network of the cloud platform and connected to the academic affairs system database through a secure interface to obtain the teaching syllabus metadata from the academic affairs system database. The teaching syllabus metadata includes knowledge point number, chapter weight and target threshold. A lightweight proxy service is deployed on the online teaching platform to intercept chapter test submission requests in real time, extract fields including student identifier, knowledge point number, number of correct answers and total number of questions, and collect the original record data of chapter tests through the proxy service; IoT attendance terminals are deployed in physical classrooms to record student attendance data based on facial recognition technology. Distributed log agents are deployed on the server cluster of the online learning platform to capture the start and end timestamps of video viewing events in real time, collect video viewing completeness data, and capture document download records. The number of downloads within 24 hours is aggregated by student identifier to collect document download frequency statistics. For chapter test original records that were submitted interrupted, the mean of adjacent time periods was used to fill in the missing data. Chapter test original records with a single knowledge point test achievement rate exceeding ±3 times the standard deviation were marked as invalid. Discrete behavior logs were divided into time-series segments by 5-minute windows, and the missing window data was filled with the mean of the preceding window. Attendance count and download frequency statistics were mapped to the [0,1] interval.

3. The cloud-based course achievement monitoring and analysis system according to claim 2, characterized in that: The course feature extraction module includes the following process for extracting features from the preprocessed course target data: The weights of each chapter are obtained from the metadata of the teaching syllabus, and the test achievement rates under the same knowledge point are weighted and summed to obtain the mastery of a single knowledge point. The time series is divided into natural weeks, and the average achievement rate is calculated weekly. The average achievement rate results of four consecutive weeks are used to form a trend vector. The influence of the dimension is eliminated by normalization to obtain the weekly dimension achievement rate change trend. When the data of a new week arrives, the data of the earliest week is deleted and the trend vector is recalculated. Based on the achievement rate sequence of knowledge points in the test, the Pearson correlation coefficient of the achievement rate is calculated pairwise. Generate an n×n knowledge point correlation matrix C that reflects the statistical correlation between knowledge points.

4. The cloud-based course achievement monitoring and analysis system according to claim 3, characterized in that: The course feature extraction module includes the following process for extracting features from the preprocessed learning behavior data: Constructing a three-dimensional matrix The dataset contains N students, K knowledge points, and each element includes the mean video viewing completion rate and document download frequency density. A matrix is ​​also used to map the mastery of related knowledge points. This forms a mapping relationship between behavior and goals; Perform high-order tensor decomposition on the behavior matrix B to extract latent factor vectors for students, knowledge points, and behavior patterns. Based on the decomposition results, calculate the weighted similarity between the knowledge point behavior factors and the mastery vector. The correlation between resource access and control is obtained, a loss function L is designed, and the optimal factor importance weights are iteratively solved using the gradient descent method. This makes the correlation degree approach the average of the actual degree of control, and introduces a regularization coefficient; By integrating the extracted features of single knowledge point mastery, weekly achievement rate trends, knowledge point correlation matrix, and correlation between resource access and mastery, a feature library of course achievement is constructed.

5. The cloud-based course achievement monitoring and analysis system according to claim 4, characterized in that: In the prediction association module, the process of constructing a mastery trend prediction model and outputting the probability prediction result of knowledge point mastery includes: The missing values ​​of the single knowledge point mastery sequence are filled by aligning them by day, and the weekly achievement rate change trend vector is matched with the daily data to construct an input feature vector containing historical single knowledge point mastery and weekly achievement rate change trend components. Based on a temporal convolutional network architecture, dilated causal convolutional kernels are used to process historical feature sequences, and the receptive field is expanded layer by layer to capture long-term and short-term dependencies. A trend prediction model is constructed, and the feature map output of the l-th layer is calculated. Attention weights of each component of the weekly trend vector The weighted fusion is then concatenated with the convolution output to obtain the fused feature vector. Output the future through a fully connected layer Daily success rate and confidence interval; We design a weighted cross-entropy loss function, prioritize assigning weights to long-term predictions, and dynamically adjust the learning rate using a cosine annealing strategy to optimize the trend prediction model.

6. The cloud-based course achievement monitoring and analysis system according to claim 5, characterized in that: In the prediction association module, the process of constructing a knowledge point association analysis topology graph and outputting the synergistic influence weight matrix of associated knowledge point groups includes: Based on the knowledge point correlation matrix features, the correlation features of resource access and mastery, and the graph convolutional network algorithm, a topology graph structure is initialized, with each knowledge point as a node in the topology graph, and the total number of nodes n corresponding to the total number of knowledge points. Edges are defined. The initial weights are Add a resource access correlation attribute to each node k. Construct a topology diagram for knowledge point association analysis; Dynamically adjust the edges in the topology graph of knowledge point association analysis initial weights By integrating the resource correlation attributes of the two endpoints, the corrected edge weights are globally normalized to obtain the optimized edge weights. The system performs neighborhood aggregation on node features, captures high-order relationships between knowledge points through stacked graph convolutional layers, and constructs a collaborative influence weight matrix based on optimized edge weights. , among which, element Remove Weakly related edges with a value below 0.3 are identified, and the collaborative influence weight matrix is ​​symmetrically processed.

7. A cloud-based course achievement monitoring and analysis system according to claim 6, characterized in that: In the decision scheduling module, the process of classifying confidence levels based on probability prediction results, binarizing the collaborative influence weight matrix, and constructing a decision rule base includes: Based on the future Daily success rate Calculate and based on the mean of the probability of meeting the standard with standard deviation Distinguish between different confidence levels for the probability of meeting the standard. Then the probability of meeting the standard is determined to be in the high confidence interval. Then the probability of meeting the standard is determined to be within the middle confidence interval. Then the probability of meeting the standard is determined to be in the high confidence interval; The collaborative influence weight matrix is ​​binarized, and a weight threshold of 0.5 times the median of the historical edge weights is set. ,like Then mark the corresponding strongly related edge as 1, if Then the corresponding weakly correlated edges are marked as 0, and the collaborative influence weight matrix is ​​forced to be symmetric, generating a binary matrix containing only 0 and 1 elements; Construct a decision rule base, define early warning rules, and determine the probability of predicting that knowledge point k meets the target. It is in the low confidence interval and exists If the core related knowledge point j is identified, a red alert is issued, freezing the learning path for the related chapters and pushing customized exercises and video resources to the student. It falls within the low confidence interval and the medium confidence interval, but the related knowledge points satisfy... If it is, it will be judged as a yellow alert, requiring the completion of related tests within a limited time, triggering a learning progress reminder. It is within the middle confidence interval and the related knowledge points satisfy If it is determined to be a blue alert, the corresponding knowledge point is marked as the observation object, and a list of related resource recommendations is generated. It is in the high confidence interval and the related knowledge points satisfy If the condition is met, it will be considered a green alert, unlocking advanced learning content, awarding a learning achievement badge, allowing students to choose additional resources, and dynamically optimizing rule weights and threshold parameters through historical feedback. Input the real-time achievement probability prediction result and the binarized matrix, traverse the knowledge point-association pairs, match the rule base conditions, and execute the intervention action according to the priority of red, yellow, blue and green after matching the rule conditions. If multiple rules are triggered, the highest priority is executed. For cyclically dependent associated knowledge point groups, the subgraph segmentation algorithm is used to implement joint intervention.

8. The cloud-based course achievement monitoring and analysis system according to claim 7, characterized in that: The decision-making and scheduling module includes the following process: Allocating teaching resources and outputting a list of related knowledge points for reinforcement. Based on the warning level matching resource pool, the resource pool is set with a semantic similarity weight of 0.7 and a historical effect weight of 0.

3. The resource score F is calculated based on the semantic similarity weight and the historical effect weight, the best resources are selected, and quota limits are set according to the level. A maximum of 5 teaching resources are allocated for red warning, a maximum of 3 teaching resources are allocated for yellow warning, and a maximum of 1 teaching resource is allocated for blue and green warning. High priority needs are given priority to obtain highly relevant teaching materials. Extract strongly correlated subgraphs from the synergistic influence matrix, subtract the normalized mastery degree from 1, and add the rate of decrease of the normalized pass probability to obtain the weakness index of each knowledge point in the strongly correlated subgraph. Sort the knowledge points in the top 5 strongly correlated subgraphs from high to low according to the weakness index, generate a reinforcement list, and recommend the most effective exercises, videos and documents in history. A formatted resource list is pushed to student terminals via a message queue. The formatted resource list includes frozen chapters, recommended resources, and priorities. The system continuously monitors changes in knowledge mastery over 3 days. If the improvement is less than 5%, the warning level is automatically raised and resource allocation is retried.

9. A cloud-based course achievement monitoring and analysis system according to claim 8, characterized in that: The analysis report module includes the following process: implementing early warnings and pushing related student behavior analysis reports based on the cloud-based rule engine. Based on the warning level, the list of affected knowledge points and associated student identifiers are extracted. Student behavior logs, time series of single knowledge point mastery, and subgraph data of synergistic influence weight matrix are aggregated from the course achievement feature library to form an analysis input set. The student behavior logs include video viewing completeness, document download frequency, and attendance rate. By combining the mastery of resources invested per unit of time to improve efficiency, detecting abnormal behavior patterns, matching inefficient learning labels through a decision rule base, quantifying the Pearson correlation coefficient between behavior and performance, and identifying key weaknesses, the abnormal behavior patterns include inefficient practice and discrepancies between classroom learning and performance. Match abnormal behavior patterns with the decision rule base to generate a structured report. The structured report includes a heatmap of weak knowledge points, a behavior-performance correlation matrix, and a resource efficiency ranking. It encapsulates a list of weak knowledge points, inefficient behavior tags, conversion rate indicators, and corresponding suggested measures in JSON format. Based on the warning level, relevant student behavior analysis reports are pushed out. Red warnings are notified in real time, yellow warnings are summarized daily, and blue warnings are reported to classes weekly. Data on report review rate and the improvement in the probability of achieving the target after intervention are collected.

10. A cloud-based course achievement monitoring and analysis system according to claim 9, characterized in that: The process of generating a heatmap of course objective achievement and a PDF diagnostic report in the analysis report module includes: Based on the knowledge point mastery and relevance matrix, the comprehensive popularity value of each knowledge point is analyzed, normalized and mapped to the red-yellow-green color space, a two-dimensional matrix is ​​constructed, and an interactive goal achievement heatmap is generated by class and student groups to show the achievement status and relevance strength of each knowledge point. Integrating global goal achievement rate, goal achievement heatmap, abnormal behavior analysis, and weekly dimension achievement rate trend prediction, the text, charts, and formulas are formatted by chapter, and a PDF diagnostic report is output. The PDF diagnostic report includes a summary, data tables, and customized suggestions, supporting multi-dimensional teaching assessment. The latest collected data is extracted from the course achievement feature database, injected into the PDF diagnostic report template according to priority, the data snapshot version number is recorded, the content of the PDF diagnostic report is synchronized with the real-time teaching status, and the analysis results of each stage are retained.

Images