Student learning effect dynamic analysis and feedback method based on network school

Abstract

The invention discloses a student learning effect dynamic analysis and feedback method based on a network school, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting the multi-modal learning behavior data of students in real time based on a network school platform, and constructing a student learning behavior knowledge base; coupling a student learning behavior knowledge base according to the historical academic records of the students, and dynamically dividing student group types to obtain a preliminary learning behavior feature tag of each student; on the basis of the obtained preliminary learning behavior feature tag of each student, constructing a personalized student dynamic learning state portrait; and based on personalized student dynamic learning state portraits, mapping to a preset subject knowledge graph, automatically identifying student individual and group knowledge vulnerabilities, generating a visual analysis report, and implementing a bidirectional feedback mechanism. The method has the beneficial effects that the accuracy and timeliness of educational intervention are improved, and personalized and differentiated teaching support is realized.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method for dynamic analysis and feedback of student learning outcomes based on online schools. Background Technology

[0002] Existing domestic and international technologies for analyzing student learning outcomes mostly focus on static, single-dimensional data analysis, typically assessing students' learning status solely through historical grades or certain learning behavior characteristics. These traditional methods fail to dynamically capture changes and emotional fluctuations in students during the learning process, resulting in unsatisfactory effects of personalized assessment and intervention. Existing technologies also lack real-time feedback mechanisms, failing to provide effective personalized learning suggestions and emotional support, making it difficult to meet students' personalized and multi-dimensional needs. Traditional learning outcome analysis methods suffer from low accuracy and poor adaptability, failing to provide students with comprehensive and precise learning support and intervention. Summary of the Invention

[0003] To address the aforementioned technical issues, this paper provides a method for dynamic analysis and feedback of student learning outcomes based on online schools. This technical solution solves the problems of low accuracy and poor adaptability of traditional learning outcome analysis methods, which are unable to provide students with comprehensive and accurate learning support and intervention.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A method for dynamic analysis and feedback of student learning outcomes based on online schools includes:

[0006] Based on the online school platform, real-time data on students' multimodal learning behaviors is collected to build a knowledge base of students' learning behaviors;

[0007] Based on students’ historical academic performance, coupled with a knowledge base of students’ learning behavior, students are dynamically categorized into different groups to obtain preliminary learning behavior characteristic labels for each student.

[0008] Based on the initial learning behavior characteristic labels obtained for each student, a personalized dynamic learning status profile of the student is constructed.

[0009] Based on personalized dynamic learning status profiles of students, the system maps these profiles to a pre-defined subject knowledge graph, automatically identifies individual and group knowledge gaps, generates visual analysis reports, implements a two-way feedback mechanism, automatically pushes personalized learning remedial packages to students, provides teachers with early warning dashboards and group learning reports, and conducts targeted teaching interventions.

[0010] Preferably, based on the online school platform, students' multimodal learning behavior data is collected in real time and preprocessed.

[0011] Based on the video viewing data in the students' multimodal learning behavior data, the original viewing duration is obtained, and the effective viewing duration of a single viewing is calculated.

[0012] Calculate the student drag coefficient for both forward and reverse dragging.

[0013] The formula for the student drag coefficient is as follows:

[0014]

[0015] Where A is the student drag coefficient, b is the number of times the student drags forward, c is the number of times the student drags backward, and β and γ are weight coefficients;

[0016] Calculate student focus level for a single video based on student drag-and-drop coefficient;

[0017] Set a sliding window with a fixed time length, and use a weighted average formula to calculate the overall student learning focus, thus obtaining a student learning focus index.

[0018] Based on the homework answer data in the multimodal learning behavior data of students, the natural week is set as the time window. The number of homework submissions and interaction frequency of students are counted. Using the weighted fusion formula, the student learning engagement is calculated to obtain the student learning engagement index.

[0019] Based on homework answer data, all answer records of KU knowledge units within a fixed time period are extracted. Using the Ebbinghaus forgetting curve principle, the time interval between the current moment and each answer is calculated with the timestamp of each answer as the benchmark. The time decay weight of each student's answer record is obtained. Using the weighted summation formula, the student's mastery of KU knowledge units is calculated to obtain the student's overall mastery of learning.

[0020] Based on the interactive discussion data in the multimodal learning behavior data of students, semantic analysis and sentiment judgment are performed on the interactive discussion content of students, discussion topics and positive interaction frequency are extracted, student interactive communication activity is calculated, and student learning interaction quality indicators are obtained.

[0021] By integrating indicators of student learning focus, learning engagement, overall learning mastery, and the quality of learning interaction, a knowledge base of student learning behavior is constructed.

[0022] Preferably, based on students' historical academic performance, K data points are randomly selected as initial cluster centers. For each student data point in the set of students' historical academic performance, the distance between each student data point and the K cluster centers is calculated using the Euclidean distance formula, and the data point is assigned to the nearest cluster center to obtain the initial K cluster centers.

[0023] For each new student's academic performance data point, the Euclidean distance from the student to all K cluster centers is calculated, and the student is assigned to the cluster of the nearest cluster center. The cluster centers are continuously updated iteratively. Based on the mean, variance, and trend characteristics of the scores of each cluster in the clustering results, and combined with prior knowledge in the field of education, students are divided into different group types. The different group types include: high-score stable type, medium-score fluctuating type, and low-score improving type.

[0024] Preferably, based on the division into different group types, the association analysis is performed on the learning focus index, learning engagement index, overall learning mastery, and learning interaction quality index in the student learning behavior knowledge base. Feature analysis is then conducted for each group type to obtain preliminary learning behavior feature labels for each student, as shown in the following formula:

[0025]

[0026] Among them, Salience(e j ,T p Let be the significance of the j-th learning behavior indicator in the p-th student group compared to the total number of students, and let e be the significance of the indicator. j For the j-th learning behavior indicator, T p For the p-th student group, Let F be the mean of the j learning behavior indicators in the p-th student group. G (e j Let be the average value of all students for the j-th learning behavior indicator, and ln(|T) p |+1) is the group size adjustment factor;

[0027] Label(i) is the set of preliminary learning behavior feature labels for student i, where i is a single student, θ is the significance threshold, and α is the individual salience coefficient.

[0028] Preferably, based on homework answering and test data from students' multimodal learning behavior data, the IRT item response theory framework is adopted, with difficulty, discrimination and guessing as item parameters, to train a three-parameter logistic model and construct item feature curves;

[0029] By using the joint maximum likelihood estimation method, the question parameters and student ability parameters are optimized. A likelihood function is constructed based on the specific answer response patterns of students. By iteratively solving the maximum likelihood estimate, the question parameter estimate and student ability value are obtained, and a student ability profile is generated.

[0030] Based on interactive discussion data from students' multimodal learning behavior data, a BERT sentiment analysis model is trained, with students' text statements as input and students' sentiment classification labels and corresponding confidence probabilities as output.

[0031] The student emotion classification tags include: positive / excited, negative / frustrated, confused / questioning, and neutral / statement;

[0032] Set up a sliding window with a fixed time interval to count the frequency and proportion of various emotional statements, and create an emotional profile of students.

[0033] Preferably, the implicit Dirichlet distribution algorithm is used to model the topics of student interaction and discussion, optimize the confusion level to determine the number of topics, extract the keyword distribution and topic-document belonging probability of each topic, explore the hot topics and knowledge areas that students care about, and integrate student emotional profiles to form student psychological profiles.

[0034] Based on the initial learning behavior characteristic tags of students, the student ability portrait and the student psychological portrait are linked and integrated to construct a personalized dynamic learning status profile of students.

[0035] Preferably, based on the student ability portrait and student psychological portrait in the personalized dynamic learning status profile of students, combined with the preset subject knowledge graph, the node relationship and edge weight in the knowledge graph are used to analyze the degree of mastery and the strength of association of students in each knowledge unit.

[0036] By using a threshold-based defect detection algorithm, students' ability scores are compared with the preset difficulty requirements of knowledge nodes, and the gaps and weaknesses in students' mastery of specific knowledge areas are automatically identified.

[0037] The threshold is 0.6;

[0038] Based on the emotional classification tags and hot topic distribution in the student psychological profile, we can identify the emotional obstacles and cognitive biases that students have in the learning process, specifically including: negative attitudes or confusion about certain knowledge points.

[0039] For the identified individual knowledge gaps of students, visualization technology is used to generate personalized learning analysis reports.

[0040] Preferably, for student groups, students with similar knowledge gaps and emotional characteristics are divided into different subgroups, and a group learning report is generated;

[0041] Based on the analysis results of individual students and groups, the system automatically pushes personalized learning remedial packages to students, which include targeted practice questions, micro-lesson videos and learning strategy suggestions. This helps students accurately fill knowledge gaps and adjust their learning status. The system also provides teachers with an early warning dashboard, which displays the distribution of knowledge gaps at the class or group level, the list of students with abnormal emotions, and teaching intervention suggestions in real time, assisting teachers in carrying out differentiated teaching and emotional support.

[0042] Based on the information in the early warning dashboard and the group learning reports, we designed targeted classroom activities, group discussions and individual tutoring, and carried out targeted teaching interventions.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention proposes a method for dynamic analysis and feedback of student learning outcomes based on online schools. This approach comprehensively collects students' multimodal learning behavior data and combines it with historical grades to construct personalized dynamic learning status profiles. This allows for real-time and comprehensive capture of students' learning status and emotional fluctuations. By mapping subject knowledge graphs, the system can accurately identify students' knowledge gaps and weaknesses, providing targeted remedial learning packages. Simultaneously, a two-way feedback mechanism provides teachers with real-time warning dashboards and group learning reports. This method significantly improves the accuracy and timeliness of educational intervention, enabling personalized and differentiated teaching support. Attached Figure Description

[0045] Figure 1 This is a flowchart of a method for dynamic analysis and feedback of student learning outcomes based on online schools. Detailed Implementation

[0046] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0047] Reference Figure 1 As shown, a method for dynamic analysis and feedback of student learning outcomes based on online schools includes:

[0048] S1. Based on the online school platform, collect students' multimodal learning behavior data in real time and construct a student learning behavior knowledge base; the multimodal learning behavior data includes: video viewing, homework answering, interactive discussions and exams;

[0049] Step S1 includes the following:

[0050] Based on the online school platform, students' multimodal learning behavior data are collected in real time and preprocessed.

[0051] Based on the video viewing data in the students' multimodal learning behavior data, the original viewing duration is obtained, and the effective viewing duration of a single viewing is calculated.

[0052] Calculate the student drag coefficient for both forward and reverse dragging.

[0053] The formula for the student drag coefficient is as follows:

[0054]

[0055] Where A is the student drag coefficient, b is the number of times the student drags forward, c is the number of times the student drags backward, and β and γ are weight coefficients;

[0056] Calculate student focus level for a single video based on student drag-and-drop coefficient;

[0057] Set a sliding window with a fixed time length, and use a weighted average formula to calculate the overall student learning focus, thus obtaining a student learning focus index.

[0058] Based on the homework answer data in the multimodal learning behavior data of students, the natural week is set as the time window. The number of homework submissions and interaction frequency of students are counted. Using the weighted fusion formula, the student learning engagement is calculated to obtain the student learning engagement index.

[0059] Based on homework answer data, all answer records of KU knowledge units within a fixed time period are extracted. Using the Ebbinghaus forgetting curve principle, the time interval between the current moment and each answer is calculated with the timestamp of each answer as the benchmark. The time decay weight of each student's answer record is obtained. Using the weighted summation formula, the student's mastery of KU knowledge units is calculated to obtain the student's overall mastery of learning.

[0060] Based on the interactive discussion data in the multimodal learning behavior data of students, semantic analysis and sentiment judgment are performed on the interactive discussion content of students, discussion topics and positive interaction frequency are extracted, student interactive communication activity is calculated, and student learning interaction quality indicators are obtained.

[0061] By integrating indicators of student learning focus, learning engagement, overall learning mastery, and the quality of learning interaction, a knowledge base of student learning behavior is constructed.

[0062] When using it, refer to the steps outlined above.

[0063] Existing domestic and international technologies for multimodal learning behavior data analysis primarily rely on single-data source analysis methods, often neglecting the interactivity and complex relationships between different learning behaviors. Traditional learning behavior analysis often focuses on single-modal data processing such as video viewing and homework completion, failing to effectively integrate multimodal data to comprehensively reflect students' learning status. Existing methods typically lack dynamic analysis and comprehensive evaluation of student behavior data, failing to provide real-time feedback on multi-dimensional indicators such as students' learning focus, engagement, and mastery, and also failing to accurately capture the emotional changes and interaction quality that students experience during the learning process. This step establishes a student learning behavior knowledge base by collecting and comprehensively analyzing students' multimodal learning behavior data (including video viewing, homework completion, interactive discussions, and tests) on online school platforms in real time. This method not only calculates students' learning focus, engagement, mastery, and interaction quality in real time, but also accurately quantifies the multi-dimensional characteristics of students' learning behaviors through indicators such as drag coefficients and learning decay weights, thereby providing a more comprehensive and accurate assessment of students' learning status, effectively compensating for the shortcomings of traditional methods, and improving the comprehensiveness and real-time nature of data analysis.

[0064] S2. Based on students’ historical academic performance, coupled with the student learning behavior knowledge base, students are dynamically divided into group types to obtain preliminary learning behavior characteristic labels for each student.

[0065] Step S2 includes the following:

[0066] Based on students' historical academic performance, K data points are randomly selected as initial cluster centers. For each student data point in the set of students' historical academic performance, the distance between each student data point and the K cluster centers is calculated using the Euclidean distance formula, and the student data point is assigned to the nearest cluster center to obtain the initial K cluster centers.

[0067] For each new student's academic performance data point, the Euclidean distance from the student to all K cluster centers is calculated, and the student is assigned to the cluster of the nearest cluster center. The cluster centers are continuously updated iteratively. Based on the mean, variance, and trend characteristics of the scores of each cluster in the clustering results, and combined with prior knowledge in the field of education, students are divided into different group types. The different group types include: high-score stable type, medium-score fluctuating type, and low-score improving type.

[0068] Step S2 also includes the following:

[0069] Based on the classification into different group types, correlation analysis was conducted on the learning focus index, learning engagement index, overall learning mastery, and learning interaction quality index in the student learning behavior knowledge base. Feature analysis was performed on each group type to obtain preliminary learning behavior characteristic labels for each student, as shown in the following formula:

[0070]

[0071] Among them, Salience(e j ,T p Let be the significance of the j-th learning behavior indicator in the p-th student group compared to the total number of students, and let e be the significance of the indicator. j For the j-th learning behavior indicator, T p For the p-th student group, Let F be the mean of the j learning behavior indicators in the p-th student group. G (e j Let be the average value of all students for the j-th learning behavior indicator, and ln(|T) p |+1) is the group size adjustment factor;

[0072] Label(i) is the set of preliminary learning behavior feature labels for student i, where i is a single student, θ is the significance threshold, and α is the individual salience coefficient.

[0073] When using it, refer to the steps outlined above.

[0074] Traditional methods often overlook the multi-dimensional interaction between students' learning states and behaviors. They tend to focus only on grade fluctuations or specific behavioral patterns, failing to comprehensively analyze aspects such as students' focus, engagement, mastery, and interaction quality during the learning process. This results in traditional methods failing to accurately reflect individual learning differences and trends when segmenting students into groups, easily leading to one-sided student classifications and a lack of targeted and personalized analysis. This step, by combining students' historical academic performance with multi-modal learning behavior data, enables dynamic group segmentation of students from a more comprehensive perspective. By analyzing students' multi-dimensional characteristics such as focus, engagement, mastery, and interaction quality, preliminary learning behavior characteristic labels can be accurately assigned to each student. This method overcomes the shortcomings of traditional techniques in comprehensively assessing students' diverse learning states, improves the accuracy of group segmentation and personalized analysis capabilities, and better reflects students' overall performance and development potential during the learning process.

[0075] S3. Based on the initial learning behavior characteristic labels obtained for each student, construct a personalized dynamic learning status profile of each student;

[0076] Step S3 includes the following:

[0077] Based on homework answering and test data from students' multimodal learning behavior data, we adopted the IRT (Item Response Theory) framework, using difficulty, discrimination, and guessing ability as item parameters, to train a three-parameter logistic model and construct item feature curves.

[0078] By using the joint maximum likelihood estimation method, the question parameters and student ability parameters are optimized. A likelihood function is constructed based on the specific answer response patterns of students. By iteratively solving the maximum likelihood estimate, the question parameter estimate and student ability value are obtained, and a student ability profile is generated.

[0079] Based on interactive discussion data from students' multimodal learning behavior data, a BERT sentiment analysis model is trained, with students' text statements as input and students' sentiment classification labels and corresponding confidence probabilities as output.

[0080] The student emotion classification tags include: positive / excited, negative / frustrated, confused / questioning, and neutral / statement;

[0081] Set up a sliding window with a fixed time interval to count the frequency and proportion of various emotional statements, and create an emotional profile of students.

[0082] Step S3 also includes the following:

[0083] Using the Latent Dirichlet Distribution algorithm, the topic modeling of student interactive discussion content is carried out, the confusion level is optimized to determine the number of topics, the keyword distribution and topic-document belonging probability of each topic are extracted, the hot topics and knowledge areas that students care about are explored, and the student emotional profile is integrated to form a student psychological profile.

[0084] Based on the initial learning behavior characteristic tags of students, the student ability portrait and the student psychological portrait are linked and integrated to construct a personalized dynamic learning status profile of students.

[0085] When using it, refer to the steps outlined above.

[0086] Existing domestic and international technologies typically focus on analyzing student performance or single learning behavior characteristics, neglecting the dynamic changes in students' multi-dimensional learning behaviors. Traditional methods often rely on static, single learning performance data, making it difficult to comprehensively capture students' emotional fluctuations, learning focus, engagement, and other comprehensive aspects of their learning process. This results in simplistic student group segmentation, failing to reveal individual learning differences and intrinsic development potential, and leading to lower accuracy and personalization in the analysis. This step, by comprehensively considering students' multimodal learning behavior data and combining it with historical performance, utilizes advanced analytical methods to more comprehensively and dynamically capture students' learning status and emotional fluctuations. This method not only improves the accuracy of student group segmentation but also delves deeper into students' individual differences and emotional states during the learning process, thus providing a more scientific basis for educational decision-making. Through this multi-dimensional and dynamic analytical approach, the accuracy of personalized student assessment and educational support can be significantly improved.

[0087] S4. Based on personalized dynamic learning status profiles of students, the system maps them to a preset subject knowledge graph, automatically identifies individual and group knowledge gaps of students, generates visual analysis reports, implements a two-way feedback mechanism, automatically pushes personalized learning remedial packages to students, provides teachers with early warning boards and group learning reports, and carries out targeted teaching interventions.

[0088] Step S4 includes the following:

[0089] Based on the student ability portrait and student psychological portrait in the personalized dynamic learning status profile of students, combined with the pre-set subject knowledge graph, the node relationships and edge weights in the knowledge graph are used to analyze the students' mastery and correlation strength in each knowledge unit.

[0090] By using a threshold-based defect detection algorithm, students' ability scores are compared with the preset difficulty requirements of knowledge nodes, and the gaps and weaknesses in students' mastery of specific knowledge areas are automatically identified.

[0091] The threshold is 0.6;

[0092] Based on the emotional classification tags and hot topic distribution in the student psychological profile, we can identify the emotional obstacles and cognitive biases that students have in the learning process, specifically including: negative attitudes or confusion about certain knowledge points.

[0093] For the identified individual knowledge gaps of students, visualization technology is used to generate personalized learning analysis reports.

[0094] Step S4 also includes the following:

[0095] For the student population, students with similar knowledge gaps and emotional characteristics are divided into different subgroups, and a group learning report is generated.

[0096] Based on the analysis results of individual students and groups, the system automatically pushes personalized learning remedial packages to students, which include targeted practice questions, micro-lesson videos and learning strategy suggestions. This helps students accurately fill knowledge gaps and adjust their learning status. The system also provides teachers with an early warning dashboard, which displays the distribution of knowledge gaps at the class or group level, the list of students with abnormal emotions, and teaching intervention suggestions in real time, assisting teachers in carrying out differentiated teaching and emotional support.

[0097] Based on the information in the early warning dashboard and the group learning reports, we designed targeted classroom activities, group discussions and individual tutoring, and carried out targeted teaching interventions.

[0098] When using it, refer to the steps outlined above.

[0099] Analyzing historical grades or single-dimensional learning behavior data lacks a comprehensive focus on students' learning status and emotional fluctuations. Existing systems rely heavily on static data, making it difficult to capture students' dynamic changes and individual differences in real time. This results in inaccurate individualized assessments and a lack of in-depth understanding of students' emotional obstacles and cognitive biases during the learning process. Furthermore, existing systems lack effective feedback mechanisms, often failing to provide timely and personalized intervention suggestions for students and teachers, impacting the accuracy and effectiveness of educational decisions. This step, by constructing personalized dynamic learning status profiles of students, combining student ability and psychological profiles, and utilizing advanced knowledge graph analysis and sentiment analysis technologies, can more accurately identify students' weaknesses and emotional distress during the learning process. Through an automatic push mechanism based on personalized learning profiles, targeted learning remedial packages can be provided to students to help them fill knowledge gaps and adjust their learning status. Simultaneously, teachers can receive real-time alerts and group learning progress reports to assist in conducting precise teaching interventions. This approach not only improves the accuracy and real-time nature of learning support but also enables interaction between students and teachers through a two-way feedback mechanism, optimizing the effectiveness of intervention and guidance during the learning process.

[0100] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for dynamic analysis and feedback of student learning outcomes based on online schools, characterized in that, include: S1. Based on the online school platform, collect students' multimodal learning behavior data in real time and build a knowledge base of students' learning behavior; S2. Based on students’ historical academic performance, coupled with the student learning behavior knowledge base, students are dynamically divided into group types to obtain preliminary learning behavior characteristic labels for each student. S3. Based on the initial learning behavior characteristic labels obtained for each student, construct a personalized dynamic learning status profile of each student; S4. Based on personalized dynamic learning status profiles of students, the system maps them to a preset subject knowledge graph, automatically identifies individual and group knowledge gaps of students, generates visual analysis reports, implements a two-way feedback mechanism, automatically pushes personalized learning remedial packages to students, provides teachers with early warning boards and group learning reports, and carries out targeted teaching interventions.

2. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 1, characterized in that, S1 includes: Based on the online school platform, students' multimodal learning behavior data are collected in real time and preprocessed. Based on the video viewing data in the students' multimodal learning behavior data, the original viewing duration is obtained, and the effective viewing duration of a single viewing is calculated. Calculate the student drag coefficient for both forward and reverse dragging. The formula for the student drag coefficient is as follows: Where A is the student drag coefficient, b is the number of times the student drags forward, c is the number of times the student drags backward, and β and γ are weight coefficients; Calculate student focus level for a single video based on student drag-and-drop coefficient; Set a sliding window with a fixed time length, and use a weighted average formula to calculate the overall student learning focus, thus obtaining a student learning focus index. Based on the homework answer data in the multimodal learning behavior data of students, the natural week is set as the time window. The number of homework submissions and interaction frequency of students are counted. Using the weighted fusion formula, the student learning engagement is calculated to obtain the student learning engagement index. Based on homework answer data, all answer records of KU knowledge units within a fixed time period are extracted. Using the Ebbinghaus forgetting curve principle, the time interval between the current moment and each answer is calculated with the timestamp of each answer as the benchmark. The time decay weight of each student's answer record is obtained. Using the weighted summation formula, the student's mastery of KU knowledge units is calculated to obtain the student's overall mastery of learning. Based on the interactive discussion data in the multimodal learning behavior data of students, semantic analysis and sentiment judgment are performed on the interactive discussion content of students, discussion topics and positive interaction frequency are extracted, student interactive communication activity is calculated, and student learning interaction quality indicators are obtained. By integrating indicators of student learning focus, learning engagement, overall learning mastery, and the quality of learning interaction, a knowledge base of student learning behavior is constructed.

3. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 1, characterized in that, S2 includes: Based on students' historical academic performance, K data points are randomly selected as initial cluster centers. For each student data point in the set of students' historical academic performance, the distance between each student data point and the K cluster centers is calculated using the Euclidean distance formula, and the student data point is assigned to the nearest cluster center to obtain the initial K cluster centers. For each new student's academic performance data point, the Euclidean distance from the student to all K cluster centers is calculated, and the student is assigned to the cluster of the nearest cluster center. The cluster centers are continuously updated iteratively. Based on the mean, variance, and trend characteristics of the scores of each cluster in the clustering results, and combined with prior knowledge in the field of education, students are divided into different group types. The different group types include: high-score stable type, medium-score fluctuating type, and low-score improving type.

4. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 3, characterized in that, S2 further includes: Based on the classification into different group types, correlation analysis was conducted on the learning focus index, learning engagement index, overall learning mastery, and learning interaction quality index in the student learning behavior knowledge base. Feature analysis was performed on each group type to obtain preliminary learning behavior characteristic labels for each student, as shown in the following formula: Among them, Salience(e j ,T p Let be the significance of the j-th learning behavior indicator in the p-th student group compared to the total number of students, and let e be the significance of the indicator. j For the j-th learning behavior indicator, T p For the p-th student group, F Tp (e j Let F be the mean of the j-th learning behavior indicator in the p-th student group. G (e j Let be the average value of all students for the j-th learning behavior indicator, and ln(|T) p |+1) is the group size adjustment factor; Label(i) is the set of preliminary learning behavior feature labels for student i, where i is a single student, θ is the significance threshold, and α is the individual salience coefficient.

5. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 4, characterized in that, S3 includes: Based on homework answering and test data from students' multimodal learning behavior data, we adopted the IRT (Item Response Theory) framework, using difficulty, discrimination, and guessing ability as item parameters, to train a three-parameter logistic model and construct item feature curves. By using the joint maximum likelihood estimation method, the question parameters and student ability parameters are optimized. A likelihood function is constructed based on the specific answer response patterns of students. By iteratively solving the maximum likelihood estimate, the question parameter estimate and student ability value are obtained, and a student ability profile is generated. Based on interactive discussion data from students' multimodal learning behavior data, a BERT sentiment analysis model is trained, with students' text statements as input and students' sentiment classification labels and corresponding confidence probabilities as output. The student emotion classification tags include: positive / excited, negative / frustrated, confused / questioning, and neutral / statement; Set up a sliding window with a fixed time interval to count the frequency and proportion of various emotional statements, and create an emotional profile of students.

6. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 5, characterized in that, S3 further includes: Using the Latent Dirichlet Distribution algorithm, the topic modeling of student interactive discussion content is carried out, the confusion level is optimized to determine the number of topics, the keyword distribution and topic-document belonging probability of each topic are extracted, the hot topics and knowledge areas that students care about are explored, and the student emotional profile is integrated to form a student psychological profile. Based on the initial learning behavior characteristic tags of students, the student ability portrait and the student psychological portrait are linked and integrated to construct a personalized dynamic learning status profile of students.

7. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 6, characterized in that, S4 includes: Based on the student ability portrait and student psychological portrait in the personalized dynamic learning status profile of students, combined with the pre-set subject knowledge graph, the node relationships and edge weights in the knowledge graph are used to analyze the students' mastery and correlation strength in each knowledge unit. By using a threshold-based defect detection algorithm, students' ability scores are compared with the preset difficulty requirements of knowledge nodes, and the gaps and weaknesses in students' mastery of specific knowledge areas are automatically identified. The threshold is 0.6; Based on the emotional classification tags and hot topic distribution in the student psychological profile, we can identify the emotional obstacles and cognitive biases that students have in the learning process, specifically including: negative attitudes or confusion about certain knowledge points. For the identified individual knowledge gaps of students, visualization technology is used to generate personalized learning analysis reports.

8. The method for dynamic analysis and feedback of student learning outcomes based on online schools according to claim 7, characterized in that, S4 further includes: For the student population, students with similar knowledge gaps and emotional characteristics are divided into different subgroups, and a group learning report is generated. Based on the analysis results of individual students and groups, the system automatically pushes personalized learning remedial packages to students, which include targeted practice questions, micro-lesson videos and learning strategy suggestions. This helps students accurately fill knowledge gaps and adjust their learning status. The system also provides teachers with an early warning dashboard, which displays the distribution of knowledge gaps at the class or group level, the list of students with abnormal emotions, and teaching intervention suggestions in real time, assisting teachers in carrying out differentiated teaching and emotional support. Based on the information in the early warning dashboard and the group learning reports, we designed targeted classroom activities, group discussions and individual tutoring, and carried out targeted teaching interventions.

Images