Teaching objective achievement degree automatic discrimination method based on classroom behavior intelligent analysis

Abstract

The invention discloses a teaching objective achievement degree automatic discrimination method based on classroom behavior intelligent analysis, and relates to the technical field of education technology and artificial intelligence, and the method comprises the steps: S1, obtaining multi-dimensional indexes represented by students, carrying out the standardization processing of each index, and obtaining a preliminary achievement degree analysis data set; s2, according to the preliminary achievement degree analysis data set, carrying out weight distribution on each index, determining the contribution degree of each index in the overall achievement degree, and forming a weighted achievement degree comprehensive score; s3, aiming at the weighted achievement degree comprehensive score, generating a contribution explanation report of each index to a final score, and obtaining a traceable scoring basis; according to the teaching objective achievement degree automatic discrimination method based on classroom behavior intelligent analysis, accurate evaluation and targeted improvement of the teaching objective achievement degree are realized, and the teaching quality and effect are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of education technology and artificial intelligence, and particularly relates to a teaching goal achievement degree automatic discrimination method based on classroom behavior intelligent analysis. BACKGROUND

[0002] In the field of education technology, the evaluation of teaching goal achievement degree is a key link to improve teaching quality and personalized teaching. The research in this field has an irreplaceable value for optimizing the allocation of educational resources and improving student learning outcomes, especially under the background of informatization, automated evaluation tools are widely used in teaching scenarios to provide data support for teachers and students.

[0003] However, although the related technology is constantly developing, its role in practical application is still limited. At present, many solutions often overlook the actual needs and psychological acceptance of teachers as the core users in design, resulting in a lack of sufficient transparency and explainability in the output results of these tools. Teachers often face difficulties in judging the rationality of the results when using automated evaluation systems, making it difficult to combine them with their own teaching experience and intuition. This lack of trust not only affects the acceptance of teachers to the tools, but also limits the deep application of technology in teaching improvement. The deeper challenge is how to integrate the teacher's trust mechanism into the automated discrimination process and ensure the understandability of the results. On the one hand, the generation logic of automatic discrimination results is often a "black box" for teachers, lacking intuitive explanation paths, making it difficult for teachers to understand the basis behind the results; on the other hand, this opacity further leads teachers to have no starting point when manually reviewing, and cannot link the discrimination results to specific teaching improvement directions. These two factors are interrelated, and the opacity of the former directly exacerbates the difficulties of the latter in practical application, making the technology tools unable to truly serve teaching practice. SUMMARY

[0004] The purpose of the present application is to provide a teaching goal achievement degree automatic discrimination method based on classroom behavior intelligent analysis, which can not only enhance the trust of teachers in automatic discrimination results, but also provide clear and understandable review basis and point out the direction for teaching improvement.

[0005] To achieve the above object, the present application provides the following technical solutions: a teaching target achievement degree automatic discrimination method based on classroom behavior intelligent analysis, comprising S1, collecting multi-dimensional indexes of student classroom behavior, and performing unified standardization processing on each index to form an initial achievement degree analysis data set; S2, based on the initial analysis data set, assigning weights to each index, determining the contribution proportion of each index to the overall achievement degree, and obtaining a weighted comprehensive score of the achievement degree; S3, around the weighted comprehensive score, generating a contribution explanation for each index to form traceable scoring basis and explanation content; S4, using the scoring basis, corresponding and associating the distribution details of the index with the contribution explanation to determine a clear and intuitive logical explanation path; S5, according to the explanation path, outputting a feedback report for teachers, and embedding the corresponding relationship between each index and the corresponding teaching improvement direction to obtain clear improvement guidance; S6, when the contribution degree of an index is lower than a preset threshold, retrieving associated cases from historical teaching data to determine a referenceable improvement scheme; S7, according to the reference scheme, generating a personalized teaching adjustment suggestion for a specific situation to form a final improvement direction output; S8, based on the final improvement direction, constructing a teacher interaction interface to determine a presentation mode of the discrimination result that can be reviewed by teachers; S9, periodically updating the parameters of the teaching target achievement degree evaluation model according to the reviewable presentation mode, and extracting teacher labeled data from the feedback report as the basis for continuous optimization of the model.

[0006] Preferably, the S1 comprises extracting multi-dimensional indexes of student performance from teaching data through a preset teaching target achievement degree evaluation model; performing standardization processing on the extracted multi-dimensional indexes using the Z-score method to obtain a standardized index set; calculating the average value of each index according to the standardized index set to obtain an average standardized index; if the average standardized index is higher than a preset threshold, determining that the index is a high achievement degree to obtain a high achievement degree index subset; grouping the indexes using the K-means clustering algorithm through the high achievement degree index subset to obtain a clustering grouping result; obtaining the maximum clustering group from the clustering grouping result, judging the consistency of the internal indexes to obtain a consistency evaluation value; and according to the consistency evaluation value, fusing the preliminary achievement degree analysis data set to obtain a comprehensive achievement degree data set.

[0007] Preferably, the S2 comprises obtaining a weight distribution set by assigning weights to each index using a random forest algorithm through preliminary achievement degree analysis of the data set; obtaining a preliminary comprehensive score by calculating a weighted average value according to the weight distribution set and fusing the standardized values of each index; if the preliminary comprehensive score is higher than a preset threshold, fusing high-contribution indexes to obtain a high-contribution index group; obtaining a low-contribution classification result by classifying a low-contribution part using a support vector machine algorithm through the high-contribution index group; judging the consistency of the low-contribution classification result with the weight distribution set, if the consistency is lower than a threshold, adjusting the weight distribution set to obtain an adjusted weight set; obtaining updated index values from the adjusted weight set, calculating a new weighted average, and determining a final comprehensive score; and predicting a future trend using a linear regression algorithm for the final comprehensive score to obtain a trend prediction value.

[0008] Preferably, the S3 comprises obtaining an index distribution information by a weighted comprehensive score, fusing a contribution proportion, calculating a proportion value of each index in the whole to obtain a proportion set; obtaining a fitting model by fitting the relationship between the index distribution information and the contribution proportion using a linear regression algorithm according to the proportion set; obtaining a regression coefficient of each index from the fitting model, fusing a trace basis to determine a coefficient set; if the value in the coefficient set is higher than a preset threshold, adjusting the index distribution information to obtain an adjusted distribution set; obtaining an updated contribution set by calculating an updated value of the contribution proportion through the adjusted distribution set; generating an explanation report for the updated contribution set, fusing a text description in the generated report to determine a draft report; extracting a key basis from the draft report, judging the consistency of the key basis with the trace basis, and obtaining a final report.

[0009] Preferably, the S4 comprises obtaining discrimination logic data by traceable score basis, grouping data using a clustering algorithm to obtain a grouping set; determining a statistical set by calculating distribution statistical values according to the grouping set and fusing index distribution details; if the value in the statistical set is higher than a preset threshold, adjusting the proportion in the contribution explanation report to obtain an adjusted report; obtaining a logical explanation path from the adjusted report, constructing an associated link of the visualization module to obtain a link set; performing path optimization for the link set, classifying the path using a decision tree algorithm to determine an optimized path set; integrating intuitive logical explanation through the optimized path set to generate module display data to obtain a display set; verifying the association consistency according to the display set, judging the path completeness, and obtaining a complete path.

[0010] Preferably, the S5 comprises extracting key data through a logical path, grouping indicators by using a clustering algorithm to obtain a grouping set; fusing a feedback report template according to the grouping set, calculating a correlation strength, and determining a strength set; if the correlation value in the strength set is higher than a preset threshold, adjusting a mapping correlation ratio to obtain an adjustment set; integrating an improvement direction description for the adjustment set, classifying a guide by using a decision tree algorithm to determine a classification set; obtaining guide content details from the classification set, processing information links to obtain a link set; verifying mapping correlation consistency according to the link set, judging completeness, and obtaining a verification set; generating a specific description output through the verification set, embedding teacher indicator details, and obtaining an output set.

[0011] Preferably, the S6 comprises calculating a correlation value by an indicator contribution to obtain a contribution set; if the contribution set is lower than a threshold preset, extracting teaching data from historical data by using data mining to determine a case set; fusing related cases according to the case set, grouping potential solutions by using a clustering algorithm to obtain a solution group; analyzing an improvement reference through the solution group, judging consistency, and obtaining a reference set; if the reference set matches a solution, linking mining technology from teaching data to determine a link set; integrating indicator contributions by using the link set to obtain a fusion description to obtain a description set.

[0012] Preferably, the S7 comprises fusing a teaching scene through a potential reference, processing context fusion by using a data integration technology to obtain a fusion group; if the fusion group meets a preset condition, grouping adjustment suggestions by using a K-means algorithm to determine a grouping set; analyzing an improvement direction according to the grouping set, obtaining a correlation link, and obtaining a link group; integrating individualization by using the link group to judge consistency, and obtaining a generation set; if the generation set exceeds a threshold, extracting features from a scene context to determine a feature group; constructing a suggestion determination by using the feature group to obtain an integrated description to obtain a description group; linking a direction output by using the description group, classifying a final determination by using a decision tree algorithm to determine an output set.

[0013] Preferably, the S8 comprises obtaining original discriminant logic data from a data source, performing preliminary processing on the data using a pre-established classification model to obtain a preliminary logical classification result; visualizing the key features in the classification according to the preliminary logical classification result, generating corresponding graphical display content using a visualization tool, and determining the output form of the visualization display module; obtaining the core logical elements from the output form of the visualization display module, combining the data input of the improvement direction, and matching the logical elements and the improvement direction using a decision tree algorithm to determine the improvement guidance content with a higher matching degree; if the matching degree is higher than a preset threshold, the improvement guidance content is integrated with the output of the visualization display module to generate preliminary interactive interface display data, and the integrated interface data is obtained; according to the integrated interface data, the review result part is obtained, the presentation form of the review result is formatted, and the final presentation style data is determined; the layout requirements of the teacher interactive interface are obtained through the final presentation style data, the preset layout template is adjusted, the final interactive interface display content is generated, and the usability state of the interface is determined; if the usability state of the interface meets the preset standard, the final interactive interface display content is stored in the database, the entire processing process is completed, and the complete business output result is obtained.

[0014] Preferably, the S9 comprises obtaining teacher review discriminant result data from a feedback report, constructing an initial data set for subsequent processing and analysis, and obtaining a preliminary data set; according to the preliminary data set, using data cleaning technology to remove outliers and redundant information, if the proportion of missing data in the cleaning process is found to be higher than a preset threshold, the missing part is completed by an interpolation method, and the cleaned data set is determined; for the cleaned data set, extracting the key features of the teacher labeled data, using a logistic regression algorithm to classify and weight the features, and determining the correlation strength of the features and the teaching target achievement degree; obtaining the classified feature data, combining the evaluation standard of the teaching target, updating the parameters of the evaluation model, if the deviation of the updated parameters and the historical parameters is higher than a preset range, triggering the parameter calibration process, and obtaining the adjusted model parameters; using the adjusted model parameters, performing a regular update process, dynamically adjusting the prediction logic of the evaluation model according to the change trend of the discriminant result, determining the updated model configuration; according to the updated model configuration, analyzing the evaluation result of the teaching target achievement degree, combining the demand for continuous optimization, generating the optimized evaluation logic, and determining the performance of the model on the current data set; obtaining the optimized evaluation logic, iterating and verifying the new data in the feedback report, if the verification result does not meet the preset standard, backtracking to the feature extraction link for reprocessing, and obtaining the final model optimization scheme.

[0015] From the above technical solutions, the present application has the following advantages: The teaching target achievement degree automatic discrimination method based on classroom behavior intelligent analysis obtains multi-dimensional indexes of student performance from teaching data by establishing a teaching target achievement degree evaluation model, performs standardization processing and weight distribution, and forms a weighted comprehensive score of achievement degree. The application also generates a contribution explanation report of each index to the final score, constructs a visual display module of discrimination logic, and determines an intuitive logical explanation path. Based on this, the application generates a feedback report for teachers, provides specific improvement guidance content, and extracts relevant cases from historical teaching data through data mining, generates personalized teaching adjustment suggestions combined with the current teaching scene. The application also constructs a teacher interaction interface, integrates discrimination logic display and improvement direction output, and continuously optimizes the evaluation model through teacher feedback, realizing accurate evaluation and targeted improvement of teaching target achievement degree, effectively improving teaching quality and effect. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The method flowchart of the application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0018] As Figure 1As shown, the present application provides a technical solution: a teaching target achievement degree automatic discrimination method based on classroom behavior intelligent analysis, comprising S1, collecting multi-dimensional indexes of students' classroom behavior, and performing unified standardization processing on each index to form an initial analysis data set of achievement degree; S2, based on the initial analysis data set, assigning weights to each index, determining the contribution proportion of each index to the overall achievement degree, and obtaining a weighted comprehensive score of achievement degree; S3, around the weighted comprehensive score, generating a contribution explanation for each index to form traceable scoring basis and explanation content; S4, using the scoring basis, corresponding and associating the distribution details of the index with the contribution explanation to determine a clear and intuitive logical explanation path; S5, according to the explanation path, outputting a feedback report for teachers, and embedding the corresponding relationship between each index and the corresponding teaching improvement direction to obtain clear improvement guidance; S6, when the contribution degree of a certain index is lower than a preset threshold, retrieving associated cases from historical teaching data to determine a reference improvement scheme; S7, according to the reference scheme, generating a personalized teaching adjustment suggestion for a specific situation to form the final improvement direction output; S8, based on the final improvement direction, constructing a teacher interaction interface to determine a presentation method of the discrimination result that can be reviewed by teachers; S9, periodically updating the parameters of the teaching target achievement degree evaluation model according to the reviewable presentation method, and extracting teacher labeled data from the feedback report as the basis for continuous optimization of the model.

[0019] The core working principle of the method is to automatically evaluate the teaching target achievement degree through data-driven analysis technology. First, the behavior of students is quantified, and through standardization processing, the dimensional difference of different dimensional indexes is removed to obtain a unified scoring system. Then, based on the actual needs of the teaching target, each index is given a reasonable weight, so that the comprehensive score can accurately reflect the contribution of students' performance in different dimensions to the teaching target achievement degree. Through the analysis of the weighted data, the specific contribution report of each index to the final achievement score can be generated, so that the scoring process is transparent and traceable. Further generate a feedback report for teachers, including the teaching improvement direction and improvement suggestion of each index, so as to help teachers effectively adjust the teaching strategy to optimize the teaching effect.

[0020] The method can significantly improve the automation and accuracy of teaching target evaluation. Through multi-dimensional analysis of students' classroom performance, the learning state of students can be fully captured, avoiding the deviation of traditional single-dimensional evaluation method. The weighted comprehensive score can not only provide a more scientific evaluation standard, but also enable teachers to obtain detailed teaching feedback and point out specific aspects that need to be improved in teaching. In addition, by combining historical teaching data, the evaluation model can be continuously optimized, making the achievement degree discrimination of teaching target more accurate and personalized, and improving the teaching effect of teachers and the learning experience of students.

[0021] S1 includes extracting multi-dimension indicators of student performance from teaching data by a preset teaching target achievement evaluation model; for the extracted multi-dimension indicators, adopting Z-score method to perform standardization processing to obtain a standardized indicator set; calculating the average value of each indicator according to the standardized indicator set to obtain an average standardized indicator; if the average standardized indicator is higher than a preset threshold, determining that the indicator is a high achievement, obtaining a high achievement indicator subset; through the high achievement indicator subset, adopting K-means clustering algorithm to group the indicators to obtain a clustering grouping result; obtaining a maximum clustering group from the clustering grouping result, judging the consistency of the internal indicators to obtain a consistency evaluation value; according to the consistency evaluation value, fusing a preliminary achievement analysis data set to obtain a comprehensive achievement data set.

[0022] In a possible implementation, S1 includes the following specific steps: first, performing structural processing on the original data recorded in the teaching process by a preset teaching target achievement evaluation model, the model pre-establishes the corresponding relationship between the teaching target and the student performance dimension, and sets the data extraction logic and interface configuration. The model calls the data interface to obtain the original data containing multiple dimensions such as student attendance, classroom speaking frequency, group cooperation times, homework completion quality, test scores, and classroom behavior video analysis data from the teaching platform, extracts the specific indicator values of each dimension through field mapping and preprocessing rules, and forms an initial set of multi-dimension indicators of student performance, each indicator corresponds to a specific numerical value.

[0023] The preset teaching goal achievement evaluation model is an intelligent evaluation framework for teaching process quality control and student learning effect analysis. The model realizes quantitative analysis and automatic discrimination of student learning achievement by explicitly mapping teaching goals, classroom behavior performance and evaluation indicators through structured configuration. The construction of the model is based on the explicit stage teaching goals listed in the teaching outline, combined with the student behavior data that can be collected in the actual classroom teaching process, and 10 core evaluation indicators including attendance frequency, classroom speaking frequency, cooperation and interaction frequency, homework completion degree, test score, classroom attention, learning attitude, video behavior analysis score, after-school discussion participation and teacher-student interaction frequency are designed. Each indicator corresponds to a clear numerical definition and data collection path, and its dimension and mapping relationship to the teaching goal are set in the model initialization stage. The data preprocessing module in the model cleans, normalizes and processes outliers of the collected data to ensure the accuracy and consistency of the input data. Then, the model standardizes each indicator to ensure data calculation under the same evaluation benchmark. The model presets multiple statistical and algorithmic modules, including Z-score conversion, mean calculation, threshold comparison, clustering analysis, consistency evaluation and weight fusion process modules. The parameters of each module are obtained based on historical teaching data samples and can be dynamically adjusted according to user feedback. During the execution of the model, all processing steps are automatically executed in the specified logical order, and the output result is the comprehensive achievement score and its data support details, which provides teachers with traceable, interpretable and operable teaching improvement basis. The model is deployed on the server side of the teaching platform, supports automatic triggering and execution after each classroom teaching task is completed, and the evaluation results are stored in the database and can be displayed and analyzed through the teacher's exclusive interface.

[0024] Subsequently, the above-mentioned extracted multi-dimensional indicators are standardized by using the Z-score method. The specific implementation process is as follows: for each dimension of the indicator data, first calculate the arithmetic mean of all sample values in the dimension, that is, add up the original scores of all students in the dimension and divide by the total number of samples to obtain the mean value of the indicator; then calculate the square of the difference between each sample value and the mean value, sum it up and divide by the sample size, and take the square root to obtain the standard deviation of the indicator; then subtract the mean value of the indicator from the score of each student on the indicator and divide by the standard deviation of the indicator to obtain the corresponding Z-score. After this processing, all dimension indicator data is unified into a standardized indicator set with no dimension, mean value of zero and standard deviation of one, ensuring the fairness of subsequent comparison. After standardization, each standardized indicator is averaged, specifically: the Z-scores of all students in a certain dimension are added and divided by the sample size to obtain the average standardized value of the indicator. Then, the average value is compared with a preset threshold. The determination of the threshold is based on the statistical analysis results of historical teaching data, which is generally set to zero point five, indicating that if the average standardized value of an indicator exceeds the threshold, it means that the overall performance of the dimension is excellent in the current teaching activity, which can be considered as a high achievement indicator. All indicators higher than this threshold are included in the high achievement indicator subset for subsequent clustering analysis. After obtaining the high achievement indicator subset, the K-means clustering algorithm is used to group the indicators in the subset. The implementation process is as follows: first, set the initial number of clustering centers according to the number of dimensions in the subset, generally set to three groups, representing different types of achievement characteristics; then randomly select three indicators as the initial clustering centers, then calculate the Euclidean distance between each indicator and the clustering center, and classify the indicators into the group represented by the nearest clustering center; update the center value of each group to be the average value of all indicators in the current group, and continue the above process until all clustering centers converge or reach the set iteration times, and finally obtain the stable clustering grouping result. From the above clustering grouping result, select the clustering group with the most indicators as the maximum clustering group, which represents the most concentrated performance trend in the current teaching process. On this basis, the consistency of the indicators in the group is evaluated, specifically: calculate the Pearson correlation coefficient between all indicators in the group, and take the average of all correlation coefficients to obtain the consistency evaluation value. The value is between zero and one, the larger the value, the more consistent the indicators in the group, reflecting the stronger the representativeness of the group indicators for achievement evaluation.

[0025] Finally, according to the above consistency evaluation value, the indicator data of the maximum clustering group is fused with the preliminary achievement analysis data set. The fusion method is to take the consistency evaluation value as a weight factor and add the weighted average value of the indicators in the high consistency group to the preliminary achievement analysis data set to form a comprehensive achievement data set. The comprehensive data set improves the influence of key dimensions on the basis of preserving the original indicator information, providing higher accuracy and representativeness support for subsequent evaluation.

[0026] S2 includes assigning weights to each index by preliminary achievement analysis dataset using random forest algorithm to obtain weight distribution set; according to the weight distribution set, the standardized values of each index are fused, the weighted average value is calculated, and the preliminary comprehensive score is obtained; if the preliminary comprehensive score is higher than the preset threshold, the high contribution index is fused to obtain a high contribution index group; through the high contribution index group, the low contribution part is classified using support vector machine algorithm to obtain a low contribution classification result; according to the low contribution classification result, the consistency with the weight distribution set is judged, if the consistency is lower than the threshold, the weight distribution set is adjusted to obtain an adjusted weight set; the updated index value is obtained from the adjusted weight set, the new weighted average is calculated, and the final comprehensive score is determined; for the final comprehensive score, a linear regression algorithm is used to predict future trend to obtain a trend prediction value.

[0027] In a possible implementation, the specific implementation process of S2 includes the following steps: first, based on the preliminary achievement analysis dataset generated by S1 stage, the configured random forest algorithm model is called, and each standardized behavior index of each student in the dataset is used as input feature to train the model with the preset teaching target achievement label as output variable. The structure of the random forest model is composed of 100 independent decision trees, each tree is generated by randomly sampling samples from the original dataset using the bootstrap method, and 3 indexes are randomly selected from all indexes as candidate features for each node splitting, the feature with the smallest Gini coefficient is selected for splitting, and the process is stopped until the leaf node sample size is not less than 5. After the model training is completed, the average feature importance score of each input index in all decision trees is calculated. The score reflects the frequency and depth of the index in the model splitting node, and the larger the value, the greater the influence of the index on the teaching achievement. The importance scores of all indexes are normalized so that the sum of all scores is 1, and the weight distribution set is constructed, each index corresponds to a determined value as the coefficient for subsequent weighting.

[0028] Then, a weighted sum of the standardized indicator values of each student is performed according to the weight distribution set, specifically, each standardized score is multiplied by its corresponding weight coefficient in turn, and the sum of all product values is the preliminary comprehensive score of the student. This calculation is completed row by row in the data table, and finally a score list corresponding to all students is generated. The score list is used to evaluate the overall achievement of students on the teaching objectives. Then, the preliminary comprehensive score of each student is compared with the set threshold. The threshold is the minimum acceptable performance score set by analyzing the average achievement data in the past 10 teaching cycles, and the specific value is 0.6. If the score of a student is higher than 0.6, the student is identified as having a higher achievement. On this basis, the top 3 indicators with the highest standardized score of the student are extracted, and these indicators are considered as the key factors that promote the score of the student. The key indicators of all such high-score students are summarized to form a high-contribution indicator group. The number of indicators in the indicator group is usually between 5 and 7, which is a representative dimension affecting the achievement.

[0029] For other indicators that do not belong to the above high-contribution indicator group, a support vector machine classification algorithm is called for classification analysis. The specific method is as follows: the features in the high-contribution indicator group are used to construct positive samples, and the remaining indicators are used to construct samples to be classified. A support vector machine model with a radial basis kernel function is used, and the proportion of training samples is 80% and the proportion of test samples is 20% during the training process, and the number of iterations is not more than 500 times. After the model training is completed, a classification label and a confidence score are given to each indicator to be classified. The classification results are compared with the indicator weight ranking generated by the initial random forest, and the consistency of the indicator importance ranking is calculated by using the Spearman rank correlation coefficient method to obtain a consistency value. If the consistency value is less than 0.7, it means that there is a significant difference between the two models in evaluating the contribution of the indicators.

[0030] In the case of insufficient consistency, the original weight distribution set is modified. The modification method is as follows: the indicators classified as high contribution and with a confidence score higher than 0.85 are increased by 1.2 times the original weight proportion in the original weight; the indicators classified as low contribution and with a confidence score higher than 0.85 are adjusted by 0.8 times the original weight. All the adjusted indicator weights are normalized to ensure that the total weight is still 1. The adjusted result is used as the new weight distribution set to replace the original weight. Then, the weight distribution set is used again to calculate the weighted average of the standardized indicator scores of each student, and the process is the same as the preliminary scoring to obtain the updated final comprehensive score. This score represents the true achievement of the student on the teaching objectives at the current stage, and is the core basis for feedback and evaluation.

[0031] Finally, based on the comprehensive score records of each student in the last three courses, the linear regression analysis module is called to make trend prediction. The regression model constructs a simple linear model with time sequence number as the independent variable and score as the dependent variable, and calculates the slope of the straight line fitting by least squares method. If the slope is positive, it means that the score increases with time, and the output is "trend rising"; if the slope is negative, it means that the score decreases, and the output is "trend falling"; if the absolute value of the slope is less than 0.05, the output is "trend stable". The trend result is written into the teacher report database together with the score record, which is used for reference by teachers when formulating teaching strategies.

[0032] S3 includes extracting index distribution information by weighted comprehensive score, fusing contribution proportion, calculating the proportion of each index in the whole, obtaining the proportion set; according to the proportion set, the relationship between the index distribution information and the contribution proportion is fitted by using linear regression algorithm, and the fitting model is obtained; the regression coefficient of each index is obtained from the fitting model, the tracing basis is fused, and the coefficient set is determined; if the value in the coefficient set is higher than the preset threshold, the index distribution information is adjusted to obtain the adjusted distribution set; the updated value of the contribution proportion is calculated through the adjusted distribution set, and the updated contribution set is obtained; the explanation report is generated for the updated contribution set, the text description in the generated report is fused, and the draft report is determined; the key basis is extracted from the draft report, and the consistency with the tracing basis is judged to obtain the final report.

[0033] In a possible implementation, the specific implementation process of S3 is as follows: first, based on the calculated weighted comprehensive score data, the distribution information of each behavior index related to the score is extracted. The specific implementation is to perform frequency statistics on each standardized index in turn to form the score distribution sequence of each index, including the minimum value, the maximum value, the mean value, the standard deviation, the frequency interval number, the frequency of each interval and the proportion it occupies, and to form the distribution information table in a fixed format. All the values in the table come from the data of all students in the actual teaching period, ensuring that the data basis has wide representativeness. Next, the distribution information is fused with the index contribution proportion data generated in the last stage. The fusion method is to take the frequency of each index in all student samples as the weight, and then multiply the frequency by the allocated weight value of the index in the comprehensive score. The sum of all frequency weight products is normalized to form the specific proportion of each index in the score composition, and the sum of the proportions of all indexes is 1, forming the proportion set. The core of this process is to establish the relative relationship of the actual score contribution of the index through the combination of the performance activity degree of the index in the score and its score weight.

[0034] Subsequently, on the basis of the proportion set, a linear regression model is constructed to fit the relationship between the distribution characteristics of each indicator and its contribution proportion in the score. The mean, standard deviation, maximum and minimum values of each indicator are selected as input variables, and the proportion value in the proportion set is selected as the output variable. The least squares method is used to train the linear regression model. After the model is trained, the regression coefficients of each input variable are extracted and recorded as quantitative values representing the strength of the linear relationship between them and the contribution proportion. All regression coefficients are classified by indicators and summarized into a regression coefficient set to describe the causal relationship between the behavior characteristics of the indicators and the role of the score. Next, these regression coefficients are fused with the trace basis in the historical evaluation records. The trace basis refers to the stability of each indicator's performance in the past teaching cycles, the correlation with student learning outcomes, and the degree of teaching intervention response, all expressed as decimals between 0 and 1, with higher values indicating greater stability and greater reliability. The fusion method is to multiply the regression coefficient of each indicator by the trace reliability value of that indicator and standardize the result to obtain the final coefficient set. This set is used to determine whether the indicators need to be adjusted.

[0035] To determine whether to adjust, a coefficient judgment threshold is set. This threshold is determined by analyzing the distribution median of the regression coefficients of all indicators in the past 10 effective sample data and the fusion results of the trace basis, and is specifically set to 0.5. When the value of an indicator in the coefficient set is greater than 0.5, it means that the contribution of the indicator is high and the performance is stable, and it is marked as a need-to-adjust item. Subsequently, the original distribution information of these marked indicators is adjusted. The adjustment method is to redistribute all student scores of the indicator to a structure close to a normal distribution. Specifically, while keeping the mean score unchanged, the standard deviation is controlled within 0.8 times the original standard deviation, the tail frequency is compressed, the frequency distribution is adjusted to make the original frequency dense area more balanced, and the adjusted distribution set is formed. Next, based on the adjusted distribution set, the contribution proportion of each indicator is recalculated. The calculation method is to multiply the frequency distribution value of each indicator by its original weight value, sum all the products, and normalize to form an updated contribution set, with the sum of all items still being 1. This step ensures that the actual score impact changes caused by changes in distribution characteristics can be reflected.

[0036] After obtaining the updated contribution set, the text generation module is automatically called to describe the changes in each indicator from the original contribution to the updated contribution in language. The description structure includes six items: "indicator name, original contribution value, updated value, change reason, data support, and suggestion". Each item is automatically converted into a complete sentence by the value information in the corresponding field through standard template processing. All description contents are combined to form a draft explanation report. Further analysis of the key field information involved in the draft report, including the main changing indicators, adjustment direction, and suggestion categories, is performed. The traceability basis fields recorded in the previous period are compared item by item to determine their consistency in identifying direction, numerical trend, and recommended action. The consistency judgment is performed using the complete matching rate of fields. If the matching item proportion is greater than or equal to 90%, the report is determined to be consistent with the historical basis, and the final report is automatically output. The report content is complete, logically coherent, has sufficient data support and technical rationality, and can be used as an important reference for teachers to improve teaching.

[0037] S4 includes extracting discriminant logic data through traceable score basis, grouping data using clustering algorithm to obtain grouping set; fusing indicator distribution details according to grouping set, calculating distribution statistical value, determining statistical set; if the value in statistical set is higher than preset threshold, adjusting proportion in contribution explanation report, obtaining adjustment report; obtaining logic explanation path from adjustment report, constructing associated link of visualization module, obtaining link set; optimizing path for link set, classifying path using decision tree algorithm, determining optimized path set; integrating intuitive logic explanation through optimized path set, generating module display data, obtaining display set; verifying association consistency according to display set, judging path completeness, obtaining complete path.

[0038] In a possible implementation, the specific implementation process of S4 includes the following detailed steps: first, extract the discriminant logic data from the traceable score basis generated in the previous stage. This operation specifically reads the information involved in each score result, such as index name, score algorithm, weight value used, score time point, and score position, to form a score path data structure. This structure contains five fields: "starting index name - score algorithm type - participating weight - score node identification - final score result". All fields constitute a complete path set. Then, perform K-means clustering on the above score path set. The clustering goal is to find index path combinations with similar score structure and score process. The execution process of the clustering algorithm is as follows: first, set the number of clusters K to 3, which is derived from the median of the number of historical teaching score path structures analyzed; then randomly select 3 paths from the score paths as the initial cluster centers. For each path, calculate the similarity between it and the 3 center paths. The similarity calculation method uses the Jaccard similarity coefficient, that is, the intersection size of the field sets of the two paths divided by the union size to get the similarity rate. To perform K-means clustering, use the value of 1 minus the similarity rate as the "distance" to select the center path with the smallest distance for classification, and complete one grouping. Then recalculate the center path of all paths in each group, that is, the path with the highest frequency of fields as the new center. Repeat the above steps until the change rate of all center paths in two consecutive iterations is less than 0.01, and the clustering terminates, outputting the grouping set.

[0039] Next, perform index distribution detail fusion operation on each cluster group. The specific process is as follows: for all the indexes involved in the score paths in each group, extract their numerical distribution in the score data, including minimum value, maximum value, average value, standard deviation, frequency distribution interval, etc.; then average each item of the corresponding distribution parameters of all indexes in the group, for example, sum all standard deviation values and divide by the number of indexes to form a distribution statistical mean, and so on to form a complete statistical set. Each parameter in the statistical set represents the overall performance of the index distribution characteristics in the group.

[0040] Then, judge the statistical set. If a parameter exceeds the set threshold, trigger the adjustment logic. The threshold here is 0.6, which is derived from the average of the standard deviation and skewness data in the statistical set of 1000 historical score reports identified as "score explanation abnormality". Compare the standard deviation and skewness in the statistical set item by item. If a value is greater than 0.6, it is determined that the volatility of the group score explanation is too high, which may cause distortion of the explanation. Accordingly, adjust the index contribution proportion in the explanation report. The adjustment method is as follows: first, identify the index list involved in the group, extract the proportion of these indexes in the score explanation from the explanation report text, then uniformly reduce these proportions by 15%, and redistribute the reduced part according to the proportion of the original proportion of other group indexes to generate the adjusted report data structure.

[0041] After generating the adjustment report, further extraction of the logical path information in the report is performed. The process is: structural analysis is performed on each explanatory text, the "score result source - index name - impact direction" field is identified, and the causal relationship therebetween is connected by a directed edge. All explanation paths form a directed graph structure, each node in the graph is a score result or index name, and each edge represents a data causal flow. The graph structure is converted into a data interface path used by the visualization module, while binding the corresponding text description position, score algorithm name and numerical change of each edge, and a link set is generated.

[0042] Subsequently, the link set is subjected to path optimization operation, and a decision tree algorithm is used to classify each path. The construction process of the decision tree is: first, set the feature variables to include path length, number of included nodes, number of covered score nodes, and ratio, a total of 4. Use these features to train the path samples labeled as "recommended", "alternative" or "redundant", and use the information gain criterion to divide the nodes, and select the feature with the maximum information gain as the division basis each time. If the length of a path is between 3 and 5 steps, and the number of covered score nodes is not less than 70% of the total score nodes, it is classified as a recommended display path; if the length is less than 3 steps but the number of covered score nodes is less than 50%, it is classified as an alternative path; and if the length is more than 5 steps and the coverage rate is less than 50%, it is a redundant path. All recommended paths and part of the high-quality alternative paths are summarized as an optimized path set.

[0043] Convert all nodes in the optimized path set to a visualization display structure, integrate the score logic, index contribution value and change direction information in each path, and form a module display data structure, i.e. a display set. Perform consistency check on all nodes and edges in the display set. The checking method is: for each score item involved in the explanation path, whether the numerical value and the score basis are consistent, whether the index contribution proportion matches the weighted score model, and whether the score order in the path is consistent with the actual score logic. The matching method is to compare each field and record the ratio of the total number of matched items to the total number of fields. If the matching rate is higher than 0.95, it is determined that the path is a complete path, and it is stored in the final explanation path set for the final score explanation display of the teacher interface.

[0044] S5 includes extracting key data through a logical path, grouping indicators by using a clustering algorithm to obtain a grouping set; fusing a feedback report template according to the grouping set, calculating a correlation strength, and determining a strength set; if the correlation value in the strength set is higher than a preset threshold, adjusting a mapping correlation ratio to obtain an adjustment set; integrating an improvement direction description for the adjustment set, classifying a guide by using a decision tree algorithm to determine a classification set; obtaining guide content details from the classification set, processing information links to obtain a link set; verifying mapping correlation consistency according to the link set, judging completeness, and obtaining a verification set; generating a specific description output through the verification set, embedding teacher indicator details, and obtaining an output set.

[0045] In a possible implementation, the specific implementation process of S5 includes the following detailed steps: first, in the generated scoring logical path, all key fields involved therein are extracted, specifically including the name of each scoring indicator in the path, the type of scoring algorithm triggered by the indicator, the weight value of the indicator in the algorithm, the score change trend direction, and the order of appearance of the indicator. The above information is structured into a structured key data table according to "path number, indicator identifier, scoring algorithm, indicator score change direction, and scoring position order", each record is a row, and all fields are fixed values, without speculative or uncertain semantics.

[0046] Subsequently, K-means clustering processing is performed on all indicator information in the key data table, for classifying indicators according to behavior patterns. The clustering process is as follows: first, the number K of clustering centers is set to 3, which is derived from the median of the classification number of typical suggestions in teacher feedback in course evaluation reports in previous years, representing three common types of indicators, operational indicators, and strategic indicators. Three initial center indicators are randomly selected from the indicator set to construct multi-dimensional vectors, each indicator corresponds to a vector, which includes four dimensions: the frequency of its appearance in the path, the number of algorithms it participates in, the consistency of the impact direction, and the depth level of the participating scoring nodes. Then, the Euclidean distance between all indicators and the three center vectors is calculated, that is, the corresponding dimension values are subtracted item by item, squared, summed, and then squared to obtain the distance value. Each indicator is assigned to the cluster with the smallest distance. After completing one grouping, the mean value of all indicators in each group in each dimension is recalculated, and the new clustering center is updated. This process continues until the change of all center vectors is less than 0.01, and the clustering converges, and the final grouping set is output.

[0047] Next, the index group in each set is matched with the preset feedback report template. A set of improvement suggestions corresponding to a certain type of index behavior is preset in each feedback template. The template structure includes four fields: suggestion type, applicable index range, suggestion operation content, and target expectation. The coincidence degree of each set of indexes with the standard indexes listed in each template is calculated in turn. The calculation method is as follows: for each index, if it appears in the template standard index, it is recorded as 1, otherwise as 0. Finally, the number of indexes recorded as 1 in the set is divided by the total number of indexes in the set to obtain the original matching degree of the set with the template. Then, the matching degree is multiplied by the average weight of each index in the path to obtain the fusion matching strength. After normalization of all matching strength values, a strength set is formed, with a value range of 0 to 1.

[0048] The matching strength threshold is set to 0.7, which is determined by the average matching strength of the suggestion templates and actual indexes in 2000 historical feedback reports with high acceptance by teachers (i.e., the proportion of actively adopted suggestions is greater than 80%). If the matching strength between a set in the strength set and a template is higher than 0.7, the suggestion association ratio between the set and the template is increased by 15%, i.e., the weight of the set in the template suggestion content is multiplied by 1.15 to form a new adjustment set, and the other sets maintain the original values.

[0049] After obtaining the adjustment set, integration operations are performed on each set of suggestion content to construct a structured suggestion guidance text. The text generation module is called to generate language expressions for each suggestion with the core structure of “index name, improvement direction, operation method, and expected effect”. Then, a decision tree algorithm is used to classify all generated suggestions. The decision tree construction process is as follows: set the classification labels to general, operation, and strategy, and select the suggestion text length, verb frequency, whether it contains quantitative values, and whether the target expectation field exists as four classification features. The labeled suggestion samples are input into the model, and information gain is used as the splitting standard for model training. After training, the model is applied to the current suggestion data to output the category of each suggestion, forming a classification set. Further, detailed information is extracted from each category of suggestions in the classification set, including the index identifier bound to the suggestion, the text details of the operation suggestion, the consistency description with the scoring change direction, and the suggestion usage occasion, forming a complete guidance detail field set. To ensure traceability of the data structure, a unique link identifier is generated for each suggestion, and these links and their corresponding text contents form a link set for subsequent interface display and logical backtracking. Then, consistency verification is performed on the above link set, which includes: judging whether the index name in the link is consistent with the original scoring path, whether the improvement direction in the suggestion content is consistent with the scoring trend, whether the operation suggestion has a clear action description, and whether the expected target field exists and conforms to the scoring target structure. If all the above fields exist and the matching field ratio exceeds 95%, the link is recorded as verified and a verification set is formed.

[0050] Based on the passing data in the verification set, the final output content is constructed. Specifically, the suggestion content of each verification passing is integrated into a "teacher suggestion feedback structure" through a fixed template, which includes specific indicator performance involved by the teacher, corresponding improvement suggestion text, suggestion classification label, suggestion execution order, feedback timestamp and expected goal summary. The content is embedded into the teacher's exclusive interface data structure and stored as the output set.

[0051] S6 includes calculating the correlation value by the index contribution to obtain the contribution set; if the contribution set is lower than the threshold preset, the teaching data is extracted from the historical data by data mining to determine the case set; the related cases are fused according to the case set, the clustering algorithm is used to group the potential schemes to obtain the scheme group; the reference set is obtained by analyzing and improving the reference through the scheme group and judging the consistency; if the reference set matches the scheme, the link set is determined by linking the teaching data mining technology; the index contribution is integrated by using the link set to obtain the fusion description to obtain the description set.

[0052] In one possible implementation, the complete technical implementation process of S6 is as follows: first, based on the standardized behavior index score of each item in the current teaching period and the corresponding weight, the actual score contribution of each item is calculated from the final comprehensive score result. The calculation process is: the standardized index value of each student is traversed one by one, and the value is multiplied by the corresponding index weight coefficient to obtain the actual contribution value of each index to the student's comprehensive score; then the contribution values of all students on the index are summed and divided by the number of students to obtain the average contribution value of the index. The average contribution values of all indexes are sequentially summarized to form the contribution set in the format of "key-value pair list of index name-average contribution value". Then a threshold for screening low-contribution indexes is set, specifically 0.05, which is derived from the lower quartile mean value of the index contribution value distribution in the analysis of years of teaching evaluation, which has statistical representativeness and stability. Each average value in the contribution set is compared with the threshold, if less than 0.05, the index is marked as a low-contribution index, and all such indexes constitute the to-be-optimized index set.

[0053] Then the historical data mining module is started, and each index marked as low contribution is sequentially searched for teaching cases. The search method is to find the record row containing the index field in all teaching activity records in the past 3 years in the teaching data main index table, requiring each search result to contain at least 20 historical sample data to ensure analysis stability. Data items must contain teacher number, course number, scoring time, intervention operation record and student feedback score to be included in the analysis range. The data records that meet the conditions are uniformly converted into structured case items, the fields include "index name, course number, teacher action description, intervention time point, student behavior change, score change amplitude", and all items constitute the preliminary case set.

[0054] Then, the clustering grouping analysis is performed on the case set, and the algorithm used is the density clustering algorithm DBSCAN. The algorithm controls the clustering behavior through two key parameters, wherein the minimum sample number is set to 5, that is, each class must contain at least 5 similar cases, and the radius distance is set to 0.3, that is, if two cases have more than 70% similarity in four feature dimensions, they are determined to be the same class. Each case is converted into a four-dimensional vector, and the dimensions include the index change amplitude, the intervention type code, the student response level code and the course type code. The distance relationship between cases is judged based on the Euclidean distance, and a number of clustering groups are formed, each clustering group corresponds to a potential improvement scheme, and the structure is called a scheme group.

[0055] Then, the correlation analysis is performed on the scheme group, and the numerical relationship between the teacher intervention behavior and the index score improvement in all cases in the group is identified. The analysis method is to use the intervention type code in all cases in the group as the independent variable, and the score improvement amplitude as the dependent variable, and calculate using the Pearson correlation coefficient formula. If the correlation coefficient result of a group is greater than 0.6, it means that the teaching intervention mode in the group and the score improvement exist strong positive correlation, and the group is determined as a high-consistency improvement scheme. All schemes that meet the condition are included in the reference set, and each record includes “index name, intervention type, average improvement amplitude, correlation coefficient value, and scheme group number”.

[0056] Then, each record in the reference set is matched and verified to determine whether the scheme is suitable for the current to-be-optimized index. The matching conditions are: first, the current index name is completely consistent with the index name in the reference record; second, the course type in the reference record is consistent with the current course type code; third, the score standard field value is the same. When the three conditions are met at the same time, the reference scheme is marked as an adaptive item. The teaching data linkage mining module is started, and all data records containing the index name, course number, teacher number and score time field in the database main linkage table are retrieved to form a linkage set, and each linkage record identifies a traceable path, and the fields include “index name, source course ID, teacher ID, score node ID, operation log ID, student response ID”.

[0057] After obtaining the link set, all data contents from the links are integrated by combining processing mode. Specifically, the operation log content, student response content and score change field are concatenated in time sequence to construct a "intervention action-student reaction-score change" three-segment behavior chain, and numerical description is attached to each segment, including intervention frequency, student response delay and score improvement value. Each behavior chain is output in "index name, behavior chain structure, score change details, sample size, data source ID" five field structure to form a description set. The description set is a structured summary result of the historical improvement path of the low contribution index, which provides a direct basis for subsequent generation of personalized teaching adjustment suggestions.

[0058] S7 includes fusing the teaching scene by potential reference, processing context fusion by data integration technology to obtain a fusion group; if the fusion group meets the preset condition, grouping the adjustment suggestions by K-means algorithm to determine a grouping set; analyzing the improvement direction according to the grouping set, obtaining associated links to obtain a link group; integrating the personalized generation by the link group, judging consistency to obtain a generation set; if the generation set exceeds the threshold, extracting features from the scene context to determine a feature group; constructing the suggestion determination by the feature group, obtaining integrated description to obtain a description group; outputting by the description group link direction, finally determining by the decision tree algorithm classification to determine an output set.

[0059] In a possible implementation, the complete technical implementation process of S7 is as follows: first, based on the reference set generated in the previous stage, extract the index name, intervention action, improvement effect value and applicable scene field of each suggestion record, and fuse with the specific context information of the current teaching task. The context information includes five structured fields of course type, class time period, teacher experience level, student basic ability level and teaching content difficulty. Each suggestion and its associated scene field form a five-dimensional structured vector, and the same structure vector constructed by the current teaching scene is compared. The similarity value is calculated by using the Euclidean distance method. When the distance is less than 0.3, it is determined that the suggestion matches the current scene, and the suggestion is included in the fusion group. This distance threshold is derived from the median value of the similarity distribution of historical effective suggestions in the actual use scene, to ensure that the screening result has applicability and stability.

[0060] When the number of suggestion records in the fusion group exceeds the set minimum suggestion number threshold of 15, the K-means clustering analysis is continued to be performed on the fusion group for identifying the behavioral structural differences among the suggestions. The clustering parameter K is set to 3, representing that the suggestions are divided into three categories of "operation class suggestions", "feedback class suggestions" and "strategy class suggestions", and the classification is based on the structural feature analysis results derived from the expert annotated sample data. Four dimension values of the suggestion text length, the number of verbs, the number of contained quantitative words and the target description complexity are extracted for each suggestion, and a four-dimensional vector is constructed as input data. The initial clustering center is initialized using the sample mean, the maximum number of iterations is set to 100, and the sample is re-assigned and the center vector is updated according to the current clustering center in each iteration until the center changes less than 0.001, and the final grouping set is output.

[0061] Subsequently, the suggestion action and target fields of the records in each suggestion group are structurally parsed, and the correspondence between the "teacher operation - indicator name - target orientation" is extracted as a structural link to form a triple format. All triples are aggregated to generate a link group, and whether there is a semantic consistency relationship in the link group is analyzed. The semantic consistency determination standard is: if the target fields in two triples have the same orientation type (such as "improve participation rate", "increase accuracy rate", etc.), it is determined to be consistent. All consistent triples are included in the production set, which is used for subsequent suggestion merging to generate operations. The upper threshold of the total number of suggestions in the production set is set to 30, which is determined by the average optimal number of suggestions in the historical teacher use suggestion feedback. If the number of records in the production set exceeds 30, it is determined that the current number of suggestions is overloaded, and the suitability of the suggestions needs to be further filtered through the teaching context. The context feature extraction module is called to read the standardized values of the five structured context fields of the current teaching task, and the feature group is combined. The feature group is matched with the suggestion action field and the target field of each suggestion in the production set at the field level. For example, if the teaching content difficulty level is high and the student foundation level is low, only the suggestion entries with moderate operation frequency and "auxiliary understanding" target orientation are retained. The suggestions that match successfully are assigned an adaptation label, and their suggestion description text, suggestion structure field and reference source are extracted to generate a description group.

[0062] After the completion of the group establishment, the target-oriented keywords of each suggestion are extracted and compared with the standard target category labels to determine the target direction label, such as "participation in promotion", "content mastery" or "behavior specification". These labels are input into the decision tree classification module together with the suggestion operation field, complexity score, and target achievement standard. The final classification of the suggestion is performed using a classification tree model trained on labeled samples, with model parameters including a maximum tree depth of 5, a minimum leaf node sample size of 5, and an information gain ratio split criterion. Each suggestion is assigned to a unique final category, forming an output set. The output set records the number, classification label, suggestion text, adapted context features, target direction, and reference number of each suggestion, constituting a structured feedback result for subsequent presentation to the teacher interface.

[0063] S8 includes obtaining original discriminant logic data from a data source, using a pre-established classification model to preliminarily process the data, and obtaining a preliminary logic classification result; according to the preliminary logic classification result, visualizing the key features in the classification, using a visualization tool to generate corresponding graphical display content, and determining the output form of the visualization display module; through the output form of the visualization display module, obtaining the core logic elements therein, combining the data input of the improvement direction, using a decision tree algorithm to match the logic elements and the improvement direction, and determining the improvement guidance content with a higher matching degree; if the matching degree is higher than a preset threshold, integrating the improvement guidance content with the output of the visualization display module to generate preliminary interactive interface display data, obtaining integrated interface data; according to the integrated interface data, obtaining the review result part thereof, formatting the presentation form of the review result, and determining the final presentation style data; through the final presentation style data, obtaining the layout requirements of the teacher interactive interface, adjusting using a preset layout template, generating the final interactive interface display content, and determining the usability state of the interface; if the usability state of the interface meets the preset standard, storing the final interactive interface display content in the database, completing the entire processing process, and obtaining complete business output results.

[0064] In one possible implementation, the original discriminant logic data is first obtained through the teaching behavior data source interface, which includes each scoring path recorded in the teaching activity, and the field content includes the scoring path number, the behavior index name involved, the scoring value of each index, the order of index appearance, the scoring algorithm name used, the scoring timestamp, and the teacher historical feedback content, etc. These data are constructed into a structured record set, and each record represents a complete scoring path. Then, a trained multi-classification classification model is called to perform preliminary classification processing on the above structured data. The classification model is a random forest model containing one hundred decision trees, and the input features of the model include the scoring path length, the scoring standard deviation, the correlation score between behavior indexes, and the average index scoring value, etc. The model outputs the logic classification label of each path through a voting mechanism, including three categories of compact structure, scoring fluctuation, and weight imbalance. Each path in the classification process is discriminated in each tree, and the category with the most occurrences of the label is counted as the final result. The classification label is attached to the original data set to form a preliminary logic classification result set. Then, the paths in the preliminary classification result belonging to the compact structure category are further processed, and the top three behavior indexes with the highest scoring value and weight value are extracted as the key features. For these three key indexes, their scoring values, scoring times, and causal order information with the previous and next indexes are read, and a visual node graph is constructed. A graph construction engine is called to draw a node graph in an SVG graph canvas, where each behavior index is a node, each causal order relationship is a edge represented by an arrow, and the index name and scoring value are labeled on the node, and the algorithm name is labeled on the edge. Each graph is limited to no more than ten index nodes to ensure clear interface display. Then, a decision tree algorithm is called to match the core logic elements in the above visual graph with the preset teaching improvement direction data. The input features of the decision tree model include the index category, the index scoring change direction, the logical order between indexes, and the classification label, and the output result is the most matched teaching suggestion number. The matching degree is calculated by dividing the number of matched indexes by the total number of indexes in the visual graph, and adding the score of the consistency of the scoring change direction by 0.5. If the matching degree value is greater than the preset threshold of 0.75 (this value is obtained by statistical analysis of a large number of teacher feedback and suggestion adoption ratio, which is the critical point for the significant increase in the accuracy of suggestion adoption), it is considered that the suggestion is effective. The above matching successful suggestion content and the visual graph are integrated. The integration process includes: adding a suggestion display area on the right side of the graph, displaying the suggestion title, suggestion content text, reference index name, and suggestion source algorithm; adding a teacher feedback area below the graph, including a single selection button for whether to adopt the suggestion and a teacher opinion text input box. The integrated structured interactive interface display data includes the graph component content, the suggestion component content, and the interactive component settings.

[0065] Subsequently, the review result part is extracted from the interactive interface data, i.e., whether the teacher adopts the suggestion, whether the feedback opinion is filled in, etc. The fields are formatted. If the teacher's feedback opinion exceeds fifty characters, it is split and displayed according to each paragraph of not more than twenty-five characters. The adoption state field is uniformly converted into three standard labels of "accept", "reject" and "adjust". After formatting, the data is re-embedded in the bottom area of the interactive interface graph to form the final presentation style data. After reading the final style data, the interface layout template is determined according to the platform display resolution, the number of controls, the proportion of the graph, the height of the interactive area, etc. All possible layout template numbers are matched item by item to screen the template numbers with a graph proportion less than eighty percent of the total width, a suggestion content area not more than forty percent of the height, and an interactive area that can be completely displayed. The template with the highest number matching degree is selected as the current template for layout, and the complete interactive interface display content is finally generated. The interface display content is handed over to the interface usability detection module for detection process. The detection content includes the integrity of the graph, the response speed of the interactive control not more than five hundred milliseconds, whether there are component occlusion or misplacement problems in the page, etc. If all detection items pass, the current display content and its version number, suggestion number, teacher feedback number, etc. are written into the teaching interactive database, and the display content state is marked as "published". Thus, the processing flow of the entire data is completed, and the complete business output result is obtained.

[0066] S9 includes obtaining the discrimination result data of the teacher's review from the feedback report, constructing an initial data set for subsequent processing and analysis, and obtaining a preliminary data set; according to the preliminary data set, using data cleaning technology to remove outliers and redundant information, if the proportion of missing data exceeds the preset threshold in the cleaning process, the missing part is completed by interpolation method, and the cleaned data set is determined; for the cleaned data set, extract the key features in the teacher's labeled data, use the logistic regression algorithm to classify and weight the features, and judge the correlation strength of the features and the teaching target achievement degree; obtain the classified feature data, update the parameters of the evaluation model combined with the evaluation standard of the teaching target, if the deviation of the updated parameters and the historical parameters exceeds the preset range, trigger the parameter calibration process, and obtain the adjusted model parameters; execute the regular update process through the adjusted model parameters, dynamically adjust the prediction logic of the evaluation model according to the change trend of the discrimination result, determine the updated model configuration; according to the updated model configuration, analyze the evaluation result of the teaching target achievement degree, generate the optimized evaluation logic combined with the demand of continuous optimization, and judge the performance of the model on the current data set; obtain the optimized evaluation logic, and perform iterative verification on the new data in the feedback report, if the verification result does not reach the preset standard, backtrack to the feature extraction link for reprocessing, and obtain the final model optimization scheme.

[0067] In a possible implementation, the complete technical implementation process of the S9 is as follows: first, the feedback interface is used to extract key fields including review confirmation status, scoring path identification, teacher manual annotation indicators, manual adjustment scoring values, suggestion adoption records, specific suggestion annotations, and teaching scene fields from the feedback report submitted by the teacher, to construct an initial data set. After formatting, the data set contains eight fields of “scoring number, indicator name, teacher annotation value, suggestion number, adoption status, manual scoring value, teaching stage identification, and feedback time”, and all teacher feedback records are arranged to form a preliminary data set. Then, the preliminary data set is cleaned, including three steps. The first step is to remove outliers, and the Z-score standardization method is used to standardize each scoring value field. The difference between each data point and the mean value is divided by the standard deviation. If the absolute value is greater than 3, it is marked as an outlier and deleted. The second step is to process redundant data. The “scoring number” and “feedback time” field combination is de-duplicated, and only the latest record is retained. The third step is to complete the missing value. The missing rate of each field is calculated. If the missing rate of any field is more than 0.1 (i.e., the missing proportion is greater than 10%), the Lagrange interpolation method is used to complete the missing field. The interpolation calculation estimates the missing value by using the fitting curve of the two effective data points before and after. After all the abnormal values and redundant information are processed, the cleaned data set is formed. Then, the teacher annotation indicators are extracted from the cleaned data set as key features, including indicator frequency, teacher adjustment amplitude, and consistency with suggestion adoption, and a logistic regression model is constructed for classification and weight distribution. First, all feature values are normalized to map the numerical values to the interval of 0 to 1. The logistic regression model uses the maximum likelihood estimation method to calculate the prediction probability of each feature on the achievement of the teaching goal. The regression coefficient output is the influence strength of the feature on the achievement. The regression coefficient and classification probability result of each feature are combined to form a classified feature data structure. Each record includes “feature name, regression coefficient value, positive class prediction probability, and whether significant (coefficient greater than 0.3)”. After obtaining the classified features, the model parameters are updated according to the current teaching goal evaluation standard. The update method is as follows: the index scoring weight field used in the existing model is compared item by item. If the classified weight value differs from the corresponding field in the original model by more than 0.2 (the threshold is determined by the performance mutation point in the evaluation accuracy experiment), the weight in the classification result is replaced by the original weight. If the weight difference is between 0.1 and 0.2, the weighted average of the current weight and the historical weight is used as the updated value, and the weighting proportion is 0.7 for the current value and 0.3 for the historical value. After all the parameters are updated, a new parameter configuration set is formed. Then, the parameter deviation detection module is called to evaluate the numerical deviation between the new and old parameter configurations. The parameter change rate is calculated, that is, the absolute value of the difference between the new and old values of each parameter divided by the old value. If the change rate of any item is greater than 0.2, the parameter calibration process is triggered.The calibration process applies the current parameters to the historical 5-period score data, performs a simulation evaluation operation, compares the evaluation output with the historical score results, and if the average deviation exceeds 0.1, performs a callback process, adjusts the new parameters according to the error direction to make them return to the historical mean by 50%.

[0068] After obtaining stable adjusted model parameters, the periodic update process is performed. The performance trend of each scoring indicator in the teacher review in the historical period is read, a time series change graph of each indicator is constructed, and the trend change slope and variance value are calculated using the sliding window method (window size is 3 periods). If the trend slope is continuously positive and the variance is less than 0.05, the "enhanced confidence weight" is set for the indicator scoring logic, otherwise the "weakening scoring influence" is set. At the same time, if a certain type of indicator is not adopted by the teacher for 3 consecutive periods, it is automatically excluded from the main scoring logic. Finally, the updated model configuration set is generated, including all valid indicator fields, weight parameters, judgment logic structure and scoring formula path.

[0069] On this basis, the model effect verification process is performed. The updated model is applied to the samples in the current feedback data that have not participated in training, the evaluation output is generated for each sample and compared with the teacher review result, and the consistency ratio is calculated, that is, the number of accurate prediction records divided by the total number of verification samples. The verification consistency standard is set to 0.85 (this value is derived from the analysis that the teacher's actual acceptance rate curve is smooth when the prediction accuracy is greater than 85%), if the current consistency is lower than the standard, the model is determined as not passed. At this time, the backtracking mechanism is triggered, returning to the feature extraction step, regenerating the feature set and repeating the subsequent processing process until the final verified model optimization scheme is generated. The final optimized model scheme content includes "use of feature fields, classification model type, parameter configuration table, verification accuracy, version number, update timestamp", which is written into the teaching evaluation model database as the basis for continuous optimization.

[0070] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automatically judging the achievement of teaching objectives based on intelligent analysis of classroom behavior, characterized in that, include: S1. Collect multi-dimensional indicators of students' classroom behavior, and standardize each indicator to form an initial dataset for achievement analysis. S2. Based on the initial analysis dataset, assign weights to each indicator, determine the contribution ratio of each indicator to the overall achievement, and obtain the weighted comprehensive achievement score. S3. Based on the weighted comprehensive score, generate a contribution description for each indicator, forming a traceable scoring basis and explanation. S4. Using the scoring criteria, correlate the distribution details of the indicators with the contribution descriptions to determine a clear and intuitive logical explanation path; S5. Based on the explanation path, output a feedback report for teachers and embed the correspondence between each indicator and the corresponding teaching improvement direction to obtain clear improvement guidance; S6. When the contribution of a certain indicator is lower than the preset threshold, retrieve relevant cases from historical teaching data to determine the improvement plan that can be referenced. S7. Based on the aforementioned reference scheme, generate personalized teaching adjustment suggestions for specific situations, and output the final improvement direction; S8. Based on the aforementioned final improvement direction, construct a teacher interaction interface and determine the presentation method of the judgment results that can be reviewed by teachers; S9. According to the verifiable presentation method, periodically update the parameters of the teaching objective achievement assessment model, and extract teacher-annotated data from the feedback report as the basis for continuous model optimization.

2. The method for automatically determining the achievement of teaching objectives based on intelligent classroom behavior analysis according to claim 1, characterized in that: S1 includes extracting multi-dimensional indicators of student performance from teaching data using a preset teaching goal achievement assessment model; standardizing the extracted multi-dimensional indicators using the Z-score method to obtain a standardized indicator set; calculating the average value of each indicator based on the standardized indicator set to obtain the average standardized indicator; determining that the indicator is highly achieved if the average standardized indicator is higher than a preset threshold, thus obtaining a high-achievement indicator subset; and grouping the indicators using the K-means clustering algorithm based on the high-achievement indicator subset to obtain the clustering results. The largest cluster is obtained from the clustering results. The consistency of its internal indicators is judged to obtain the consistency evaluation value. Based on the consistency assessment values, the preliminary achievement analysis dataset is merged to obtain the comprehensive achievement dataset.

3. The method for automatically determining the achievement of teaching objectives based on intelligent classroom behavior analysis according to claim 1, characterized in that: S2 includes analyzing the dataset based on the initial achievement level, assigning weights to each indicator using the random forest algorithm to obtain a weight allocation set; and based on the weight allocation set, integrating the standardized values ​​of each indicator, calculating the weighted average value, and obtaining a preliminary comprehensive score. If the initial comprehensive score is higher than the preset threshold, high contribution indicators are merged to obtain a high contribution indicator group. The low contribution part is classified using the support vector machine algorithm through the high contribution indicator group to obtain the low contribution classification result. Based on the low contribution classification result, its consistency with the weight allocation set is judged. If the consistency is lower than the threshold, the weight allocation set is adjusted to obtain the adjusted weight set. The updated indicator values ​​are obtained from the adjusted weight set, the new weighted average is calculated, and the final comprehensive score is determined. For the final comprehensive score, a linear regression algorithm is used to predict future trends and obtain trend prediction values.

4. The method for automatically determining the achievement of teaching objectives based on intelligent classroom behavior analysis according to claim 1, characterized in that: S3 includes extracting indicator distribution information through weighted comprehensive scoring, integrating contribution ratios, calculating the proportion of each indicator in the whole, and obtaining a proportion set; based on the proportion set, using a linear regression algorithm to fit the relationship between indicator distribution information and contribution ratios to obtain a fitting model; obtaining the regression coefficients of each indicator from the fitting model, integrating traceability evidence, and determining the coefficient set; if the value in the coefficient set is higher than a preset threshold, adjusting the indicator distribution information to obtain an adjusted distribution set. By adjusting the distribution set, the updated value of the contribution ratio is calculated to obtain the updated contribution set; for the updated contribution set, an explanation report is generated, and the text descriptions in the generated report are integrated to determine the draft report; key evidence is extracted from the draft report, and its consistency with the traceability evidence is judged to obtain the final report.

5. The method for automatically determining the achievement of teaching objectives based on intelligent classroom behavior analysis according to claim 1, characterized in that: S4 includes extracting discrimination logic data based on traceable scoring criteria, grouping the data using a clustering algorithm to obtain a group set; calculating distribution statistics based on the distribution details of the fusion indicators in the group set to determine the statistical set; if the value in the statistical set is higher than a preset threshold, adjusting the proportion in the contribution explanation report to obtain an adjustment report; The logical explanation path is obtained from the adjustment report, and the associated links of the visualization modules are constructed to obtain a link set; the link set is optimized by classifying the paths using a decision tree algorithm to determine the optimized path set; the intuitive logical explanation is integrated through the optimized path set to generate module display data and obtain the display set. Verify the consistency of associations based on the display set, determine the completeness of the path, and obtain the complete path.

6. The method for automatically determining the degree of achievement of teaching objectives based on intelligent analysis of classroom behavior according to claim 1, characterized in that: S5 includes extracting key data through logical paths, grouping indicators using clustering algorithms to obtain group sets; calculating association strength based on the group set fusion feedback report template to determine a strength set; if the association value in the strength set is higher than a preset threshold, adjusting the mapping association ratio to obtain an adjustment set; integrating improvement direction descriptions into the adjustment set, using decision tree algorithms to classify guidance and determine a classification set; obtaining guidance content details from the classification set, processing information links to obtain a link set; verifying the consistency of mapping associations and judging completeness based on the link set to obtain a validation set; generating specific descriptive outputs from the validation set, embedding teacher indicator details, to obtain an output set.

7. The method for automatically determining the degree of achievement of teaching objectives based on intelligent analysis of classroom behavior according to claim 1, characterized in that: S6 includes calculating the correlation value through the contribution of indicators to obtain a contribution set; if the contribution set is lower than a preset threshold, data mining is used to extract teaching data from historical data to determine a case set; based on the case set, relevant cases are integrated, and a clustering algorithm is used to group potential schemes to obtain scheme groups; By analyzing and improving the reference through the solution group, we can determine consistency and obtain a reference set. If the reference set matching scheme is determined, the link set is determined using teaching data link mining technology; the contribution of the link set integration indicators is used to obtain the fusion description, thus obtaining the description set.

8. The method for automatically determining the degree of achievement of teaching objectives based on intelligent analysis of classroom behavior according to claim 1, characterized in that: S7 includes processing context fusion using data integration technology through potential reference fusion teaching scenarios to obtain fusion groups; If the fusion group meets the preset conditions, the K-means algorithm is used to group the adjustment suggestions to determine the group set; the improvement direction is analyzed based on the group set, the related links are obtained, and the link group is obtained. The generated set is obtained by integrating personalized data through link groups and judging consistency. If the generated set exceeds the threshold, features are extracted from the scene context to determine the feature group. The feature group is used to construct suggestions and obtain integrated descriptions to obtain description groups; the output direction is output through the link of description groups, and the final classification is determined by the decision tree algorithm to determine the output set.

9. The method for automatically determining the degree of achievement of teaching objectives based on intelligent analysis of classroom behavior according to claim 1, characterized in that: S8 includes obtaining raw discrimination logic data from the data source, performing preliminary processing on the data using a pre-established classification model, and obtaining preliminary logical classification results. Based on the preliminary logical classification results, the key features in the classification are visualized, and the corresponding graphical display content is generated using visualization tools to determine the output format of the visualization display module. Through the output format of the visualization display module, the core logical elements are obtained. Combined with the data input of the improvement direction, the decision tree algorithm is used to match the logical elements with the improvement direction to determine the improvement guidance content with a high degree of matching. If the matching degree is higher than the preset threshold, the improved guidance content will be integrated with the output of the visualization module to generate preliminary interactive interface display data, and the integrated interface data will be obtained. Based on the integrated interface data, the review results are obtained, and the presentation format of the review results is formatted to determine the final presentation style data. Through the final presentation style data, the layout requirements of the teacher's interactive interface are obtained, and adjustments are made using a preset layout template to generate the final interactive interface display content and determine the usability status of the interface. If the usability status of the interface meets the preset standards, the final interactive interface display content is stored in the database, completing the entire processing flow and obtaining a complete business output result.

10. The method for automatically determining the achievement of teaching objectives based on intelligent classroom behavior analysis according to claim 1, characterized in that: S9 includes: obtaining the judgment results data of teachers' review from feedback reports to construct an initial dataset for subsequent processing and analysis, resulting in a preliminary data set; based on the preliminary data set, using data cleaning techniques to remove outliers and redundant information; if the proportion of missing data exceeds a preset threshold during the cleaning process, then using interpolation methods to fill in the missing parts, thus determining the cleaned data set; for the cleaned data set, extracting key features from the teacher-annotated data, using a logistic regression algorithm to classify and weight the features, and determining the correlation strength between the features and the achievement of teaching objectives; obtaining the classified feature data, and updating the parameters of the evaluation model in conjunction with the evaluation criteria of teaching objectives; if the deviation between the updated parameters and historical parameters exceeds a preset range, then triggering a parameter calibration process to obtain the adjusted model parameters; By adjusting the model parameters, a periodic update process is executed. Based on the changing trend of the judgment results, the prediction logic of the evaluation model is dynamically adjusted to determine the updated model configuration. According to the updated model configuration, the evaluation results of the achievement of teaching objectives are analyzed. Combined with the need for continuous optimization, the optimized evaluation logic is generated to judge the model's performance on the current dataset. The optimized evaluation logic is obtained and iteratively verified against the new data in the feedback report. If the verification results do not meet the preset standards, the process is backtracked to the feature extraction stage for reprocessing to obtain the final model optimization scheme.

Images