Cloud mobile learning system based on artificial intelligence

Abstract

The invention provides a cloud mobile learning system based on artificial intelligence, and the system comprises the steps: obtaining the learning data of a student through a mobile device, and the learning data at least comprises the exercise record and a test result of the student; according to the learning data, an artificial intelligence algorithm is adopted to evaluate the learning level of the student, and the knowledge mastering condition of the student is determined; a personalized learning scheme is generated according to the knowledge mastering condition, and the learning scheme at least comprises targeted practice content and learning path planning; the personalized learning scheme is pushed to the student through the mobile device; and tracking the learning progress of the student, and dynamically adjusting the personalized learning scheme according to the learning progress.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and is a cloud-based mobile learning system based on artificial intelligence. Background Art

[0002] In current classroom teaching, due to individual differences among students, each student may have different knowledge points that they don't understand or are not firmly grasped. How to solve these knowledge points after class is a common problem. There are currently two solutions, as follows:

[0003] 1. Active Learning: Students use learning assistance apps to choose which videos to watch, when to watch them, which exercises to do, how many exercises to do, and when to do them. This is a self-selected learning model, similar to a supermarket. The drawbacks and problems of active learning are that it can lead to poor or even no learning outcomes. This is because students struggle to find resources that suit them, are overwhelmed by resources, and waste time on useless work.

[0004] 2. Passive learning: Following a study plan that you don't know where you came from, you follow a fixed set of steps, such as watching a video, then doing five exercises, then reviewing a PowerPoint presentation, and then completing a set of assessments. This is a passive training model. The drawbacks and problems with passive learning are that some students learn well, while others don't, or even have no effect at all, wasting time on useless work. This is due to a lack of professional intervention, research, and pedagogical knowledge. Furthermore, given the individual differences, background knowledge, and stage of development among students, using a fixed set of prescribed actions, the same resource types, and the same resources for all students is inappropriate.

[0005] Therefore, how to solve students’ problems in understanding a certain knowledge point efficiently, personally, accurately and effectively is a technical problem that needs to be solved in current education and teaching. Summary of the Invention

[0006] To solve the above problems, a cloud-based mobile learning system based on artificial intelligence is proposed. The system can customize learning plans and practice questions according to students' knowledge mastery level, and use mobile devices such as mobile phones for online learning.

[0007] To achieve the above object, the technical solution adopted by the present invention is:

[0008] An artificial intelligence-based cloud-based mobile learning system, comprising:

[0009] Acquiring the student's learning data through a mobile device, wherein the learning data at least includes the student's practice records and test results;

[0010] Based on the learning data, an artificial intelligence algorithm is used to evaluate the student's learning level and determine the student's knowledge mastery;

[0011] Generate a personalized learning plan based on the knowledge mastery, the learning plan at least including targeted practice content and learning path planning;

[0012] The personalized learning plan is pushed to the student via the mobile device, and the personalized learning plan is dynamically adjusted according to the learning progress.

[0013] Preferably, the use of an artificial intelligence algorithm to assess a student's learning level includes:

[0014] Extracting students' practice records and test results from the learning data, and analyzing students' performance data on different knowledge points;

[0015] The performance data is classified using a pre-trained machine learning model to obtain the student's mastery level of each knowledge point, determine the knowledge points that the student has mastered and the knowledge points that the student has not mastered, and output the mastered and unmastered knowledge points as evaluation results for subsequent generation of the personalized learning plan. If the proportion of unmastered knowledge points exceeds a preset threshold, the practice content of the relevant knowledge points in the learning plan is preferentially adjusted;

[0016] Based on the evaluation results, a special improvement plan targeting students' weak points is generated.

[0017] Preferably, generating a personalized learning plan based on the knowledge mastery includes:

[0018] Obtaining the unmastered knowledge points and error-prone knowledge points in the knowledge mastery status, extracting related knowledge content from a preset knowledge point map, and matching corresponding exercises and learning resources;

[0019] By analyzing the student's historical learning data, determining the learning pace and difficulty level suitable for the student, generating a personalized learning path including exercises and learning resources, and integrating the personalized learning path into the personalized learning plan;

[0020] The personalized learning plan is pushed in real time via the mobile device.

[0021] Preferably, the pushing of the personalized learning plan to the student via the mobile device includes:

[0022] Obtaining the exercise content and learning path planning in the personalized learning plan;

[0023] According to the students' study schedule, the exercise content and learning path planning are broken down into daily tasks in stages;

[0024] Sending a notification of the daily task to the student via the mobile device, and sending a reminder message via the mobile device if the student fails to complete the daily task on time;

[0025] Based on the student's task completion, record the learning progress data, analyze the student's execution effect of the personalized learning plan, and adjust the priority and frequency of subsequent push content based on the execution effect.

[0026] Preferably, the tracking of the student's learning progress and the dynamic adjustment of the personalized learning plan according to the learning progress include:

[0027] collecting the student's learning progress data in real time via the mobile device, the learning progress data including at least exercise completion rate and test scores, analyzing the student's changes in knowledge point mastery under the current learning plan, and if the changes in knowledge point mastery do not meet preset targets, extracting specific knowledge points that did not meet the target from the learning progress data, and adjusting the exercise content and learning resources in the personalized learning plan based on the specific knowledge points that did not meet the target;

[0028] New learning tasks are generated through the adjusted personalized learning plan, and the new learning tasks are pushed through the mobile device. The completion status of the students on the new learning tasks is continuously tracked as a basis for subsequent adjustments.

[0029] Preferably, the method of using an artificial intelligence algorithm to evaluate the student's learning level based on the learning data includes:

[0030] Extracting students' wrong answer records from the learning data and analyzing the distribution of knowledge points corresponding to the wrong answers;

[0031] Using a pre-established error analysis model, the system determines the classification of the causes of the errors, which include at least unfamiliarity with knowledge points and incorrect problem-solving ideas. Based on the classification of the causes, the system generates targeted error analysis content, identifies students' weaknesses in various knowledge points, and outputs a learning level assessment report.

[0032] The learning level assessment report provides data support for the subsequent generation of the personalized learning plan to ensure that the learning plan is consistent with the students' actual needs.

[0033] Preferably, the learning plan includes at least targeted practice content and learning path planning, including:

[0034] Obtaining course outlines and textbook content, and constructing a knowledge point network, wherein the knowledge point network at least includes associations between knowledge points;

[0035] Based on the knowledge mastery, extract the knowledge points that students need to focus on learning from the knowledge point network and generate corresponding exercise content;

[0036] By analyzing students' learning habit data, we determine the learning path planning method suitable for students, generate personalized learning plans including time arrangements and content sequence, and ensure that the personalized learning plan meets the students' learning needs.

[0037] The beneficial effects of using the present invention are:

[0038] The present invention discloses a personalized education method based on learning data. By obtaining the completion status of students' exercises, wrong question records and evaluation results, the current learning level is determined, and the degree of mastery is analyzed using a knowledge point map model to construct a personalized learning path. Matching exercises and preview tasks are generated according to the path and intelligently pushed through a cloud platform. A machine learning algorithm is used to dynamically evaluate the learning effect, and the path is optimized and adjusted when it is below a threshold. Learning behavior data is continuously tracked, and a real-time learning status report is generated. Potential weak links are predicted in combination with a deep learning model, and targeted reinforcement exercises are generated in advance. The present invention realizes a comprehensive analysis, personalized guidance and dynamic adjustment of the student learning process, effectively improves the teaching pertinence and learning effect, and provides technical support for teaching students in accordance with their aptitude. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flow chart of the artificial intelligence-based cloud-based mobile learning system of the present invention.

[0040] Figure 2 Schematic diagram of the connection of the cloud-based mobile learning system based on artificial intelligence of the present invention.

[0041] Figure 3 This is an overall flow chart of the cloud-based mobile learning system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of this technical solution more clear, the following technical solution is further described in detail in conjunction with specific implementation methods. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of this technical solution.

[0043] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] like Figure 1 In this embodiment, a cloud-based mobile learning system based on artificial intelligence may specifically include:

[0045] Obtain the student's learning data, which includes the completion status of exercises, records of wrong questions, and homework evaluation results, and determine the student's current learning level through preliminary processing of the learning data.

[0046] The system collects students' answer data in exercises and homework to obtain their completion status, error records, and assessment results. The collected answer data is initially cleaned and formatted to determine structured learning data. An intelligent diagnostic algorithm is used to analyze structured learning data to determine the student's current learning level. Through in-depth mining of error records, the student's easy-to-error questions, questions they already know how to answer, and questions they don't know how to answer are classified. The classification results are matched against the knowledge point database to determine which knowledge points the student has mastered and which have not. If the number of knowledge points that the student has not mastered exceeds a preset threshold, the supplementary exercise generation mechanism is triggered to generate targeted exercises. By analyzing the student's answer data in the supplementary exercises, the updated learning level and error records are obtained. A real-time learning status report is generated based on the updated learning data to determine the student's performance progress curve. A dynamic evaluation algorithm is used to continuously track the learning status report to determine the direction of correction of the learning plan.

[0047] Specifically, the system collects data on students' responses to exercises and homework assignments. For example, the system records the time, accuracy, and error numbers for 20 math questions. This information is then used to determine the student's completion status, error history, and assessment results. The collected data is then cleaned and formatted, such as removing duplicate data, filling in missing values, and converting the data into JSON format to create structured learning data.

[0048] In the above process, intelligent diagnostic algorithms are used to analyze structured learning data. For example, a K-Means clustering algorithm is used to categorize students into high, medium, and low learning level groups to determine their current learning level. By deeply mining the student's error records, for example, analyzing a student's error rate in geometry as high as 60%, the student's classification results are divided into easy-to-error questions, mastered questions, and difficult questions. This classification result is matched against a knowledge point database, for example, by associating the error question number with the knowledge point ID to determine which knowledge points the student has mastered and which have not. If the number of knowledge points a student has not mastered exceeds a preset threshold, for example, if the number of unmastered knowledge points exceeds 5, a supplementary exercise generation mechanism is triggered to generate targeted exercises, such as 10 geometry questions related to the unmastered knowledge points. By analyzing the student's response data in the supplementary exercises, for example, recording a student's accuracy rate in the supplementary exercises increasing to 80%, an updated learning level and error record are obtained. Based on the updated learning data, a real-time learning report is generated, such as a student's performance progress graph, to determine the student's performance progress. Dynamic evaluation algorithms are used to continuously track academic reports. For example, time series analysis is used to predict students' future performance trends and determine the direction of revisions to learning plans.

[0049] After determining the student's current learning level, a pre-established knowledge point map model is used to analyze the student's mastery of each knowledge point based on the current learning level, and a classification result of the student's mastery of knowledge points and unmastered knowledge points is obtained.

[0050] By assessing students' practice progress, we obtain their answer data and assess their current learning level. Based on their answer data, we analyze the number of incorrect answers in each assignment to determine the distribution of incorrect answers and their mastery of the relevant knowledge points. Using a pre-established knowledge point graph model, we match incorrect answers with knowledge points, generating a corresponding result. Using the knowledge point graph model, we analyze students' performance on each knowledge point and determine which knowledge points they have mastered and which have not. Based on the classification of mastered and unmastered knowledge points, we generate a report on their mastery. We obtain students' historical homework data, analyze the causes of incorrect answers and trends in their mastery of knowledge points, and assess their learning dynamics. Based on these learning dynamics assessments, we update the student's mastery data in the knowledge point graph and determine a personalized knowledge point recommendation sequence. Based on this personalized knowledge point recommendation sequence, we match the corresponding textbook content and exercises to generate a dynamic learning plan for the student. Using this dynamic learning plan, we adjust the question distribution and difficulty level for the next assessment, generating new student answer data.

[0051] Specifically, by assessing students' practice, we obtain their answer data. For example, in a math assessment, a student completes 20 questions, with 15 correct answers and 5 incorrect answers, resulting in a 75% accuracy rate for the student's current learning level. Based on the student's answer data, we analyze the number of incorrect answers in each assignment. For example, if 3 of the 5 incorrect answers involve functions and 2 involve geometry, we can determine the distribution of incorrect answers and the student's mastery of the relevant knowledge points. Using a pre-established knowledge point graph model, we match incorrect answers with knowledge points. For example, we associate functions with incorrect answers numbered 1, 3, and 5, and geometry with incorrect answers numbered 2 and 4, to obtain the correspondence between knowledge points and incorrect answers. Using the knowledge point graph model, we analyze the student's performance on each knowledge point. For example, if the correct answer rate for functions is 40% and for geometry is 50%, we can determine that the student has mastered geometry and has not yet mastered functions. Based on the classification of mastered and unmastered knowledge points, we generate a report on the student's mastery of the knowledge points. For example, the report may indicate that functions require improvement and geometry has been mastered. Obtain students' historical homework data, analyze the reasons for wrong questions and the changing trends in the mastery of knowledge points. For example, the error rate of function knowledge points in the past three homework assignments has dropped from 60% to 40%, and the evaluation result of students' learning dynamics is that function knowledge points have gradually improved. Based on the evaluation results of learning dynamics, update the students' mastery data in the knowledge point map, for example, update the mastery of function knowledge points from 40% to 50%, and determine the personalized knowledge point recommendation sequence to prioritize the review of function knowledge points. According to the personalized knowledge point recommendation sequence, match the corresponding textbook content and practice questions, such as recommending function-related textbook chapters and 10 function practice questions, and generate a dynamic learning plan for students. Through the dynamic learning plan, adjust the distribution and difficulty of the questions in the next assessment, for example, increase the proportion of function questions from 25% to 40%, and obtain new student answer data.

[0052] Based on the classification results, a personalized learning path is constructed, which includes reducing the frequency of practice for knowledge points that have been mastered and increasing targeted practice content for knowledge points that have not been mastered and knowledge points that are easy to make mistakes.

[0053] Student learning data is collected and machine learning algorithms are used to classify student response records to identify mastered knowledge points, unmastered knowledge points, and key points prone to error. Based on the classification results, data mining techniques are used to analyze the correlations between knowledge points and construct a knowledge point map. The knowledge point map is used to determine the learning weight of each knowledge point and prioritize unmastered and key points prone to error. If the priority exceeds a set threshold, the corresponding knowledge point content from the textbook is extracted through the cloud computing platform to produce refined learning materials.

[0054] Based on the learning material, an intelligent association algorithm is used to match relevant practice questions and generate a targeted practice question bank. Through big data analysis, the student's historical learning path and practice frequency are analyzed to determine the reduction rate of practice frequency for mastered knowledge points. If the reduction rate meets pre-set rules, a personalized learning path is constructed by combining the practice question bank with unmastered knowledge points and knowledge points prone to error. A dynamic evaluation algorithm is used to track students' performance in the learning path, generating real-time evaluation data on learning effectiveness. Based on this evaluation data, a machine learning model is used to adjust the content and frequency of exercises in the learning path to generate an optimized personalized learning plan.

[0055] Specifically, student learning data was collected and the K-means clustering method, a machine learning algorithm, was used to classify student answer records. Knowledge points were divided into three categories: mastered, not mastered, and prone to error, with a classification accuracy of 95%. Based on the classification results, the Apriori algorithm, a data mining algorithm, was used to analyze the correlations between knowledge points and construct a knowledge point graph. The graph contains correlation strength values between knowledge points. For example, the correlation strength for "Functions and Derivatives" was 0.85. The learning weight of each knowledge point was obtained from the knowledge point graph. The PageRank algorithm was used to calculate the priority of not mastered and prone to error points. For example, the priority for "Probability and Statistics" was 0.92. If the priority was above the set threshold of 0.9, the corresponding knowledge point content from the textbook was extracted using a cloud computing platform. The text was refined using the BERT model, a natural language processing technology, to produce refined learning content. For example, the "Probability and Statistics" chapter was condensed into a 500-word core content. Based on the learning content, the collaborative filtering method, a smart association algorithm, was used to match relevant practice questions and generate a targeted practice question bank containing 100 questions related to "Probability and Statistics." Through big data analysis, students' historical learning paths and practice frequencies are captured. Regression analysis is used to determine if the frequency of practicing mastered knowledge points should be reduced by 50%. If this reduction meets pre-set criteria, a personalized learning path is constructed, combining a database of exercises for unmastered and error-prone knowledge points. 70% of these exercises should include "Probability and Statistics" exercises. Using time series analysis within a dynamic assessment algorithm, students' performance along the learning path is tracked, generating real-time assessment data on learning outcomes. For example, if mastery of "Probability and Statistics" increased from 60% to 80%, the learning path's practice content and frequency are adjusted using a gradient boosting decision tree within the machine learning model. This generates an optimized personalized learning plan, with the "Probability and Statistics" exercise frequency adjusted to three times per week.

[0056] Through the personalized learning path, matching exercises and preview tasks are generated, and the cloud computing platform is used to intelligently push the exercises and preview tasks to obtain students' completion feedback data.

[0057] Obtain student historical learning data and course content data to determine a personalized learning path. Generate matching exercises and pre-study tasks based on the personalized learning path. Use a cloud computing platform to push exercises and pre-study tasks to students' mobile devices.

[0058] Collect feedback data from students completing exercises and pre-study tasks. Analyze this feedback data using machine learning algorithms to determine the student's knowledge mastery. If the knowledge mastery level falls below a set threshold, generate supplementary exercises and update the learning path. Based on the updated learning path, adjust the content of subsequent exercises and pre-study tasks. Obtain the adjusted task completion feedback data and compile it into a real-time learning status report. This learning status report is pushed to the student's mobile device to provide learning progress tracking data.

[0059] Specifically, the system collects student historical learning data and course content data. By analyzing student homework completion rates, test scores, and course engagement over the past three months, combined with the distribution of knowledge points in the course syllabus, it determines a personalized learning path. Based on the personalized learning path, a recommendation algorithm based on collaborative filtering is used to generate matching exercises and pre-study tasks. For example, for a calculus course, 10 moderately difficult integral calculation problems and five pre-study videos are generated. Using a cloud computing platform, the exercises and pre-study tasks are pushed to students' mobile devices via message queue technology, ensuring real-time delivery. Student feedback on the exercises and pre-study tasks, including time spent completing them, accuracy rate, and video viewing time, is collected and stored in a cloud database. A decision tree model within a machine learning algorithm analyzes this feedback data to determine student knowledge mastery. If the accuracy rate falls below 70%, additional exercises are generated and the learning path is updated. Based on the updated learning path, subsequent exercises and pre-study tasks are adjusted, such as adding five lower-difficulty exercises and two videos explaining fundamental concepts. The adjusted task completion feedback data is compiled into a real-time learning report using data visualization tools, showing student progress curves and knowledge point mastery. Learning status reports are pushed to students' and teachers' mobile devices through the cloud computing platform, and real-time updates are achieved using WebSocket technology to obtain learning progress tracking data.

[0060] Based on the completion feedback data, a machine learning algorithm is used to dynamically evaluate the student's learning effect to obtain an evaluation index of the learning effect. The evaluation index is used to determine whether the learning path needs to be adjusted.

[0061] Feedback data on students completing learning tasks is collected via mobile devices to determine accuracy, time consumption, and question type distribution. Data preprocessing techniques are used to clean and standardize the collected feedback data, generating a structured dataset of student learning behavior. Machine learning algorithms are used to extract features from this structured dataset to determine the characteristics of students' mastery of different knowledge points. Based on the extracted mastery characteristics, a supervised learning model is used to train and predict students' learning outcomes, generating preliminary assessments of learning outcomes.

[0062] The initial evaluation results are further analyzed using a deep learning model to identify dynamic trends in learning outcomes. Based on these dynamic trends, a clustering algorithm is used to group students' learning performance and identify groups with different learning abilities. By comparing the degree of match between these groups and historical learning data, the suitability of the current learning path is determined, providing preliminary indicators for path adjustment. Based on these preliminary indicators, a reinforcement learning algorithm is used to optimize the learning path and determine the adjusted personalized learning plan. The optimized learning plan is stored in the system database, and content for the next learning task is pushed to the system.

[0063] Specifically, mobile devices are used to collect student feedback data on completed learning tasks, capturing accuracy, time spent, and question type distribution. For example, a student's accuracy on calculus questions is 75%, with an average answer time of 90 seconds. Question types include limits, derivatives, and integrals. Based on the collected feedback data, data preprocessing techniques are used to clean and standardize the data. For example, Z-score normalization is used to normalize answer time data to eliminate dimensionality effects, resulting in a structured dataset of student learning behavior. Machine learning algorithms are used to extract features from this structured dataset. For example, a random forest algorithm is used to analyze students' error rates across different knowledge points, identifying weaker mastery of derivatives. Based on the extracted mastery features, supervised learning models are used to train and predict student learning outcomes. For example, a logistic regression model is used to predict students' accuracy on the next test, providing a preliminary assessment of learning outcomes. This preliminary assessment is further analyzed using deep learning models, such as using LSTM networks to capture temporal changes in students' accuracy rates and capture dynamic trends in learning outcomes. Based on dynamic trends, clustering algorithms are used to group students' learning performance. For example, the K-means algorithm is used to divide students into three mastery groups: high, medium, and low, to identify student groups with different learning abilities. By comparing the degree of match between student groups and historical learning data, for example, calculating the cosine similarity between the current student group and the historical high-performing group, it is determined whether the current learning path is suitable for the students and obtain preliminary indicators for path adjustment. Based on these preliminary indicators, reinforcement learning algorithms are used to optimize the learning path. For example, the Q-learning algorithm is used to adjust the order of question push and determine the adjusted personalized learning plan. The optimized learning plan is stored in the system database. For example, the adjusted question push strategy is written to the MySQL database to obtain the push content for the next learning task.

[0064] In the process of determining whether the learning path needs to be adjusted based on the evaluation indicators, if the evaluation indicators are lower than the preset threshold, the personalized learning path is optimized and adjusted, the exercise content and difficulty are updated, and a new learning plan is generated and pushed to the student end.

[0065] Obtain the completion data of the exercise content and homework submission data pushed by the student end, extract the accuracy and completion time of each exercise through the system analysis module, and obtain the quantitative indicators of the student's current learning performance. According to the quantitative indicators, call the pre-trained machine learning model to classify the student's learning performance. If the classification result is lower than the preset threshold, it is judged that the personalized learning path needs to be optimized and adjusted. By analyzing the error distribution of students' historical learning data and current exercises, a deep learning model is used to predict the weak links in students' knowledge points and determine the scope of learning content that needs to be adjusted. According to the determined scope of learning content, extract exercise questions that match the students' weak knowledge points from the question bank, and use the algorithm to sort and filter out a set of questions with increasing difficulty to obtain preliminary optimized exercise content. Obtain preliminary optimized exercise content, combine students' historical learning progress and preference data, use the recommendation algorithm to adjust the order and type of question presentation, and generate a personalized exercise content sequence.

[0066] The difficulty assessment model is called through the cloud computing platform to analyze the generated exercise content sequence. If the difficulty coefficient exceeds the student's current ability range, the difficulty of the questions is adjusted to obtain exercise content that meets the student's ability. Based on the adjusted exercise content, combined with the course outline and learning objectives, a new personalized learning plan is regenerated using the learning path planning algorithm to determine the structure and time schedule of the learning plan. The new learning plan and exercise content sequence are converted into a visual data format through the front-end interface, and a learning task package is generated and pushed to the student end to obtain task data that can be received by the student end. The task data feedback received by the student end is obtained, and the student's interaction data on the new learning plan is recorded. The student learning behavior database is updated through the system log analysis module to obtain optimized learning effect tracking data.

[0067] Specifically, the system obtains data on exercise completion and homework submission pushed from the student end. The system analyzes the accuracy and completion time of each exercise, e.g., a 65% accuracy rate and 30-minute completion time for an exercise, to obtain quantitative indicators of the student's current learning performance. Based on these quantitative indicators, a pre-trained machine learning model is used to classify the student's learning performance, such as a support vector machine (SVM) model. If the classification result falls below a preset threshold of 70%, the personalized learning path is determined to require optimization and adjustment. By analyzing the error distribution of students' historical learning data and current exercises, a deep learning model is used to predict weak areas in the student's knowledge. For example, an LSTM model is used to analyze the error distribution and determine the scope of learning content that requires adjustment: "Probability Theory and Mathematical Statistics." Based on the determined learning content, exercises are extracted from the question bank that match the student's weak points. An algorithm is used to sort and select a set of questions of increasing difficulty. For example, 10 questions are extracted from the question bank with difficulty coefficients of 0.3, 0.5, and 0.7, respectively, to obtain a preliminary set of optimized exercises. Initially optimized exercise content is obtained. Based on the student's historical learning progress and preference data, a recommendation algorithm is used to adjust the order and type of questions presented. For example, a collaborative filtering algorithm is used to reorder questions by type and difficulty, generating a personalized exercise content sequence. A difficulty assessment model is invoked through the cloud computing platform to analyze the generated exercise content sequence. If the difficulty coefficient exceeds the student's current ability, the difficulty coefficient is adjusted, for example, from 0.7 to 0.6, to obtain exercise content that matches the student's ability. Based on the adjusted exercise content and in conjunction with the course syllabus and learning objectives, a learning path planning algorithm is used to regenerate a new personalized learning plan. For example, a learning path is planned, with the structure and schedule determined as "three 30-minute exercises per week." The new learning plan and exercise content sequence are converted into a visual data format through a front-end interface, generating a learning task package that is pushed to the student end. For example, the data is converted into JSON format to obtain task data that the student can receive. Student feedback on the task data is collected, and student interaction data on the new learning plan is recorded. The system log analysis module updates the student learning behavior database, recording, for example, that the student completed an exercise in 25 minutes with a 75% accuracy rate, generating optimized learning outcome tracking data.

[0068] With respect to the optimized and adjusted learning plan, the student's learning behavior data is continuously tracked, wherein the learning behavior data includes the completion status of each homework and the analysis results of wrong questions, so as to determine the improvement effect of the learning plan.

[0069] The system collects the completion data and error records of each student's homework assignment to obtain the original data set of student learning behavior data. Based on the original data set, the wrong questions in each homework assignment are classified and sorted to determine the knowledge points corresponding to the wrong questions and the analysis results of the causes of the errors. Using the error analysis results, combined with the knowledge point mastery assessment algorithm, the student's learning weakness data for each knowledge point is obtained. By learning the weak link data, the system matches the preset learning plan optimization rules to determine the adjustment strategy for the student's personalized learning plan. Based on the adjustment strategy, the content and exercise recommendations of the student's current learning plan are updated to obtain an optimized personalized learning plan. The system continuously tracks the student's homework completion status under the optimized learning plan to obtain new learning behavior data.

[0070] If the error rate in the new learning behavior data is lower than the preset threshold, the learning plan improvement effect is confirmed to be positive, and the improvement effect evaluation result is obtained. If the error rate is higher than or equal to the preset threshold, it is marked as requiring further adjustment, and the improvement effect evaluation result is obtained. Based on the improvement effect evaluation result, the weight parameters of the learning plan optimization rule are adjusted to determine the updated optimization rule. Using the updated optimization rule, the student learning behavior data is reanalyzed to obtain the adjustment strategy for the next round of learning plans.

[0071] Specifically, the system automatically collects student completion data for each assignment, such as accuracy rate, time distribution, and wrong question numbers, to form a structured dataset containing question IDs, answer records, and grading results. Natural language processing is performed on the wrong question records to extract key words from the question stem and match them with the knowledge point database. For example, a wrong question "Trigonometric function simplification error" is classified as the "Trigonometric identity transformation" knowledge point, and the cause of the error is marked as "poor formula memory." Based on the wrong question analysis results, a Bayesian knowledge tracking algorithm is used to calculate the mastery probability of each knowledge point. For example, if the mastery of "Trigonometric identity transformation" is 65%, which is below the 80% threshold, it is determined to be a weak link. The system calls a preset rule engine. If the mastery of a certain knowledge point is below the threshold for three consecutive assignments, an adjustment strategy is triggered to increase specific exercises for that knowledge point, such as increasing the proportion of exercises related to "Trigonometric identity transformation" from 15% to 30%. The adjusted learning plan is pushed to the student's terminal, including five new trigonometric transformation exercises and three accompanying micro-course links. The system monitors students' homework data under the new plan and calculates changes in error rates. For example, if the error rate for trigonometric transformations drops from 40% to 25%, falling below the 30% improvement threshold, the optimization plan is deemed effective. Based on the magnitude of the decrease, the rule engine dynamically adjusts the weighting coefficients for recommended exercises, such as adjusting the correlation between mastery and the number of exercises from 0.5 to 0.7. Based on the updated weighting coefficients, the exercise allocation ratio for each knowledge point is recalculated, generating a new round of adjustment plans consisting of four trigonometric transformation exercises and two derivative application exercises.

[0072] A real-time learning status report is generated through the learning behavior data. The learning status report includes a student's performance change curve and knowledge point mastery progress, which is used to assist in adjusting subsequent learning content.

[0073] The system collects data on each student's exercises and homework, generating a learning behavior dataset containing answer records, accuracy rates, and timestamps. Based on this learning behavior dataset, a machine learning algorithm is used to classify the answer records and determine the corresponding knowledge point labels for each question. Using these labels, combined with the syllabus and course materials, the student's progress in mastering each knowledge point is determined, generating a knowledge point mastery chart. If the accuracy rate for a particular knowledge point in the knowledge point mastery chart falls below a set threshold, data mining techniques are used to analyze the characteristics of the relevant questions and identify a set of weak knowledge points. Based on this set of weak knowledge points, big data analytics techniques are used to extract related learning paths from historical learning data to generate a personalized learning content recommendation sequence. Time series analysis techniques are used to generate a student's performance curve based on the student's accuracy rate for each homework and exercise. Based on the performance curve and the knowledge point mastery chart, cloud computing is used to integrate the data to generate a real-time learning status report containing the curve and progress. Based on the weak knowledge points and recommended sequences in the learning status report, a dynamic programming algorithm is used to adjust subsequent learning content, generating an updated learning plan. Based on the updated learning plan, a machine learning model is used to evaluate learning outcomes and determine the correction parameters for the learning plan.

[0074] Specifically, the system obtains data on each student's exercises and homework in the system, generating a learning behavior dataset containing answer records, accuracy rates, and timestamps. For example, a student may complete a math homework assignment on October 1, 2023, with an accuracy rate of 75%. Based on the learning behavior dataset, a machine learning algorithm is used to classify the answer records. For example, the K-nearest neighbor algorithm is used to label the question "Solving Quadratic Equations" as the "Algebra" knowledge point. Using the knowledge point labels, combined with the syllabus and course materials, the student's progress in mastering each knowledge point is determined. For example, the average accuracy rate of students in the "Algebra" knowledge point is 70%, resulting in a knowledge point mastery progress chart. If the accuracy rate of a knowledge point in the knowledge point mastery progress chart falls below a set threshold, such as a 60% accuracy rate for the "Geometry" knowledge point, data mining techniques are used to analyze the characteristics of the relevant questions. For example, association rule mining is used to identify students with a high error rate in the "Triangle Area Calculation" question, thereby determining a set of weak knowledge points. Based on a set of weak points, big data analytics techniques are used to extract related learning paths from historical learning data. For example, this analysis reveals that students need to learn "properties of triangles" before mastering "calculating the area of a triangle," resulting in a personalized learning content recommendation sequence. Time series analysis techniques are used to analyze the changes in students' accuracy across assignments and exercises. For example, an ARIMA model is used to generate a student performance curve, showing a gradual improvement from 70% to 85%. Based on the performance curve and the knowledge point mastery progress chart, cloud computing is used to integrate the data and generate a real-time learning report containing the curve and progress. For example, the report shows that a student's mastery of "Algebra" is 80% and that of "Geometry" is 60%. Based on the weak points and recommended sequences identified in the learning report, dynamic programming algorithms are used to adjust subsequent learning content. For example, the learning path is optimized to prioritize learning "properties of triangles," resulting in an updated learning plan. Based on the updated learning plan, machine learning models are used to evaluate learning outcomes. For example, a random forest model is used to predict a 90% probability of a student's performance improvement in the next stage, thus determining the revised parameters of the learning plan.

[0075] The data in the learning report is used in combination with a pre-trained deep learning model to predict the student's potential weaknesses in future learning and obtain a prediction result.

[0076] Student data from learning reports is extracted, including analysis results for each exercise and assignment, classification of incorrect questions, and performance improvement curves, to create a quantitative dataset of student learning behavior. The intelligent diagnostic module processes this quantitative dataset to analyze the student's mastered knowledge points, key error points, and incompetent knowledge points, determining the student's current learning level and distribution of knowledge points. Based on the knowledge point map, the student's incompetent knowledge points and key error points are associated with the knowledge points in the textbook to create a knowledge point subgraph corresponding to the student's weak links. If there are cross-disciplinary knowledge point associations in the knowledge point subgraph, an intelligent association algorithm is used to match related lessons and questions to obtain a set of exercises related to the weak points.

[0077] Using a pre-trained deep learning model, the quantitative data set and knowledge point subgraph of the academic report are input to predict the weak knowledge points that students may encounter in their future studies and obtain preliminary prediction results. Through the intelligent analysis module, the preliminary prediction results are compared with the wrong question data in the historical academic report to determine whether the predicted weak knowledge points are consistent with the trend of students' easy-to-make mistakes. If the prediction results are consistent with the trend of easy-to-make mistakes, the priority of the weak knowledge points is calculated based on the weight relationship of the knowledge point map, and a list of high-priority weak knowledge points is determined. Based on the list of high-priority weak knowledge points, a matching set of exercises is extracted, and targeted learning task recommendations are generated to obtain a personalized learning path. Through the real-time update mechanism of the academic report, the execution results of the learning path are fed back to the quantitative data set to update the students' learning level and knowledge point mastery distribution.

[0078] Specifically, data such as students' exercise accuracy, error types, and answer time are extracted from their learning reports. Clustering algorithms (such as K-means, with K=3) are used to classify the incorrect answers into three categories: mastered questions, easy-to-error questions, and difficult questions, forming a structured dataset. The intelligent diagnosis module calculates the student's ability value θ based on IRT (Item Response Theory). This is combined with association rules in the knowledge point graph (such as the Apriori algorithm, with support ≥ 0.3) to map a subgraph of weak knowledge points. If interdisciplinary associations exist within the subgraph (such as vector knowledge points in mathematics and physics), a collaborative filtering algorithm (with a similarity threshold of 0.7) is used to match 20 highly relevant questions from the associated exercise database. The historical learning data (such as the change in the error rate in the past three months) is input into the LSTM model (64 hidden units), and the weak points that may appear in the next two weeks are predicted (output probability ≥ 0.8). The data is then compared with the current trend of easy-to-error questions (such as 5 consecutive similar errors) and a chi-square test (p < 0.05) is performed. If there is a significant correlation, the priority of the knowledge points is calculated based on the PageRank algorithm (weight α = 0.85), and the TOP5 weak points are screened to generate a learning path. Finally, the mastery indicators in the learning database are updated through the real-time API (such as the accuracy rate improvement Δ ≥ 15%).

[0079] Based on the prediction results, targeted reinforcement exercise content is generated in advance. The reinforcement exercise content is related to potential weak links and is pushed to the student end through the mobile phone system to complete the learning closed loop.

[0080] Data from each exercise and homework completed by students through the mobile learning system is collected and analyzed for accuracy, error types, and knowledge point distribution, revealing students' learning behavior and knowledge mastery. Based on this analysis of knowledge mastery and the system's stored course knowledge point network, the student's weak points and potential weaknesses are identified. If the potential weakness is related to a specific knowledge point, the intensive practice question bank associated with that knowledge point is extracted to create a targeted set of practice questions. A machine learning algorithm is used to extract features from the student's historical learning data and weak points to predict the student's learning difficulties in the weak points and generate personalized learning needs predictions. Based on these predictions, questions matching the learning needs are selected from the intensive practice question bank to generate targeted intensive practice content. The generated intensive practice content is formatted using a cloud computing platform and converted into an interactive practice module suitable for mobile display, resulting in a push-ready content package. The formatted intensive practice content package is then transmitted to the student's application via the mobile learning system's push interface, completing content distribution. Data from students completing intensive practice on their mobile phones is collected, and the accuracy rate and changes in knowledge point mastery are analyzed to evaluate the effectiveness of the practice. Based on the evaluation results of the practice effects, the students' learning reports are updated, the mastery data in the knowledge point network is adjusted, and a new learning progress curve is generated.

[0081] Specifically, data from each exercise and homework completed by a student through the mobile learning system is collected and analyzed for accuracy, error types, and knowledge point distribution. For example, a student in an advanced mathematics course completes 10 exercises with a 70% accuracy rate, with errors primarily concentrated in calculus. This data then provides insights into the student's learning behavior and knowledge mastery. Based on this analysis of knowledge mastery, combined with the system's stored network of course knowledge points (e.g., the association between calculus and sub-knowledge points such as limits and derivatives), the student's weak point is determined to be limit calculus, identifying potential weaknesses. If the potential weakness is associated with a specific knowledge point, a database of intensive practice questions associated with that knowledge point is extracted. For example, 20 limit calculus-related questions are selected from the database to create a targeted set of practice questions. Machine learning algorithms are used to extract features from the student's historical learning data and weak points. For example, a random forest algorithm is used to analyze common error patterns in limit calculus and predict that the student's learning difficulty in limit calculus is infinitesimal processing, resulting in personalized learning needs prediction results. Based on the prediction results, the intensive practice question bank is screened for questions that match learning needs. For example, five limit calculation questions involving infinitesimal quantities are selected to generate targeted intensive practice content. The generated intensive practice content is formatted using a cloud computing platform, for example, by using AWS Lambda to convert the questions into JSON format. This is then adapted for display in the interactive practice module on the mobile device, resulting in a push-ready content package. The formatted intensive practice content package is then transferred to the student's mobile app via the mobile learning system's push interface, for example, using Firebase Cloud Messaging to push the content package to the student's mobile app, completing content distribution. The student's mobile intensive practice data is collected. For example, if a student completes five questions with an 80% accuracy rate, the accuracy rate and changes in knowledge point mastery are analyzed to generate an evaluation of the practice effectiveness. Based on the evaluation results, the student's learning report is updated, for example, by increasing their mastery of limit calculation from 60% to 75%. The mastery data in the knowledge point network is adjusted to generate a new learning progress curve.

[0082] The above content is only a preferred embodiment of the present invention. For ordinary technicians in this field, many changes can be made in the specific implementation methods and application scope based on the ideas of the present technical content. As long as these changes do not deviate from the concept of the present invention, they all fall within the scope of protection of this patent.

Claims

1. A cloud-based mobile learning system based on artificial intelligence, characterized by: include: Acquiring the student's learning data through a mobile device, wherein the learning data at least includes the student's practice records and test results; Based on the learning data, an artificial intelligence algorithm is used to evaluate the student's learning level and determine the student's knowledge mastery; Generate a personalized learning plan based on the knowledge mastery, the learning plan at least including targeted practice content and learning path planning; The personalized learning plan is pushed to the student via the mobile device, and the personalized learning plan is dynamically adjusted according to the learning progress.

2. The artificial intelligence-based cloud-based mobile learning system according to claim 1, wherein: The use of artificial intelligence algorithms to assess students' learning levels includes: Extracting students' practice records and test results from the learning data, and analyzing students' performance data on different knowledge points; The performance data is classified using a pre-trained machine learning model to obtain the student's mastery level of each knowledge point, determine the knowledge points that the student has mastered and the knowledge points that the student has not mastered, and output the mastered and unmastered knowledge points as evaluation results for subsequent generation of the personalized learning plan. If the proportion of unmastered knowledge points exceeds a preset threshold, the practice content of the relevant knowledge points in the learning plan is preferentially adjusted; Based on the evaluation results, a special improvement plan targeting students' weak points is generated.

3. The artificial intelligence-based cloud-based mobile learning system according to claim 1, wherein: Generating a personalized learning plan based on the knowledge mastery includes: Obtaining the unmastered knowledge points and error-prone knowledge points in the knowledge mastery status, extracting related knowledge content from a preset knowledge point map, and matching corresponding exercises and learning resources; By analyzing the student's historical learning data, determining the learning pace and difficulty level suitable for the student, generating a personalized learning path including exercises and learning resources, and integrating the personalized learning path into the personalized learning plan; The personalized learning plan is pushed in real time via the mobile device.

4. The artificial intelligence-based cloud-based mobile learning system according to claim 1, wherein: Pushing the personalized learning plan to the student via the mobile device includes: Obtaining the exercise content and learning path planning in the personalized learning plan; According to the students' study schedule, the exercise content and learning path planning are broken down into daily tasks in stages; Sending a notification of the daily task to the student via the mobile device, and sending a reminder message via the mobile device if the student fails to complete the daily task on time; Based on the student's task completion, record the learning progress data, analyze the student's execution effect of the personalized learning plan, and adjust the priority and frequency of subsequent push content based on the execution effect.

5. The artificial intelligence-based cloud-based mobile learning system according to claim 1, wherein: Tracking the student's learning progress and dynamically adjusting the personalized learning plan based on the learning progress includes: collecting the student's learning progress data in real time via the mobile device, the learning progress data including at least exercise completion rate and test scores, analyzing the student's changes in knowledge point mastery under the current learning plan, and if the changes in knowledge point mastery do not meet preset targets, extracting specific knowledge points that did not meet the target from the learning progress data, and adjusting the exercise content and learning resources in the personalized learning plan based on the specific knowledge points that did not meet the target; New learning tasks are generated through the adjusted personalized learning plan, and the new learning tasks are pushed through the mobile device. The completion status of the students on the new learning tasks is continuously tracked as a basis for subsequent adjustments.

6. The artificial intelligence-based cloud-based mobile learning system according to claim 1, wherein: The method of using an artificial intelligence algorithm to evaluate the student's learning level based on the learning data includes: Extracting students' wrong answer records from the learning data and analyzing the distribution of knowledge points corresponding to the wrong answers; Using a pre-established error analysis model, the system determines the classification of the causes of the errors, which include at least unfamiliarity with knowledge points and incorrect problem-solving ideas. Based on the classification of the causes, the system generates targeted error analysis content, identifies students' weaknesses in various knowledge points, and outputs a learning level assessment report. The learning level assessment report provides data support for the subsequent generation of the personalized learning plan to ensure that the learning plan is consistent with the students' actual needs.

7. The artificial intelligence-based cloud-based mobile learning system according to claim 1, wherein: The learning plan shall at least include targeted practice content and learning path planning, including: Obtaining course outlines and textbook content, and constructing a knowledge point network, wherein the knowledge point network at least includes associations between knowledge points; Based on the knowledge mastery, extract the knowledge points that students need to focus on learning from the knowledge point network and generate corresponding exercise content; By analyzing students' learning habit data, we determine the learning path planning method suitable for students, generate personalized learning plans including time arrangements and content sequence, and ensure that the personalized learning plan meets the students' learning needs.

Images