Learning condition analysis method and system based on large language model, terminal and medium

Abstract

The invention relates to a learning condition analysis method and system based on a large language model, a terminal and a medium, and relates to the field of education system technology, the method comprises the following steps: collecting learning data of students, the learning data comprising structured learning data and unstructured learning data, the structured learning data comprising examination scores and homework completion conditions, and the unstructured learning data comprising examination scores and homework completion conditions; the unstructured learning data comprises classroom behavior data, learning process data, student self-evaluation and mutual evaluation data and teacher teaching records; preprocessing the learning data to generate standardized data which can be used for analysis; input interaction information is received, analysis demand parameters are generated, and the interaction information comprises analysis demands of teachers or feedback information of students; based on a large language model, semantic understanding and deep analysis are carried out on the standardized data and the analysis demand parameters, and a learning condition analysis result is obtained; and displaying the study condition analysis result in a visual mode. The application has the effects of improving the teaching quality and meeting the actual teaching demand integrating degree.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of educational system technology, and particularly relates to a learning situation analysis method and system based on a large language model, a terminal and a medium. BACKGROUND

[0002] In a modern education system, a learning situation analysis system is a key technical support for improving teaching quality and meeting actual teaching needs.

[0003] In related technologies, a learning situation analysis system is based on a large language model technology, processes student learning data, matches a teaching standard database, generates a learning situation diagnosis report and teaching improvement suggestions, and finally forms a visual analysis chart.

[0004] In view of the above related technologies, the single analysis method and fixed index system have the problems of insufficient learning situation analysis accuracy and pertinence, which leads to the disconnection between the analysis results and actual teaching needs, and cannot provide effective support for teaching decisions. SUMMARY

[0005] In order to improve teaching quality and meet actual teaching needs, the present application provides a learning situation analysis method and system based on a large language model, a terminal and a medium.

[0006] In a first aspect, the present application provides a learning situation analysis method based on a large language model, which adopts the following technical solution: A learning situation analysis method based on a large language model, comprising: Collecting learning data of students, the learning data including structured learning data and unstructured learning data, the structured learning data including test scores and homework completion, and the unstructured learning data including classroom behavior data, learning process data, student self-evaluation and mutual evaluation data, and teacher teaching records; Preprocessing the learning data to generate standardized data that can be used for analysis; Receiving input interaction information, generating analysis requirement parameters, and the interaction information including teacher analysis requirements or student feedback information; Based on a large language model, performing semantic understanding and deep analysis on the standardized data and the analysis requirement parameters to obtain learning situation analysis results; Displaying the learning situation analysis results in a visual manner.

[0007] By adopting the technical scheme, the structured and unstructured learning data of students are fused, the deep understanding and personalized analysis of learning characteristics are realized by using a large language model, the analysis results are dynamically generated in combination with the teaching needs of teachers and the feedback of students, and the analysis results are presented in a multi-dimensional visual manner, thereby effectively improving the comprehensiveness, accuracy and interactivity of learning analysis, and significantly enhancing the scientificity of teaching decision and the pertinence of learning guidance.

[0008] Optionally, the consistency of the classroom behavior data and the learning process data is detected; If the consistency does not meet a preset standard, when the classroom behavior data is higher than a first data threshold and the learning process data is lower than a second data threshold, it is determined that there is a logical conflict; In response to the logical conflict, effective participation features in the monitoring video data in a conflict time period are analyzed to obtain an effectiveness score, the effective participation features including an angle between a face orientation and a blackboard or a teacher, a relevance between a body movement and a learning task, and an effective speech in group interaction; If the effectiveness score exceeds an effectiveness threshold, a first weight coefficient of the classroom behavior data is increased, and a second weight coefficient of the learning process data is decreased; If the effectiveness score does not exceed the effectiveness threshold, a third weight coefficient of the classroom behavior data is decreased, and a fourth weight coefficient of the learning process data is increased; The classroom behavior data and the learning process data are fused by weighting according to the first weight coefficient, the second weight coefficient, the third weight coefficient and the fourth weight coefficient to generate the standardized data.

[0009] By adopting the technical scheme, the consistency detection mechanism of the classroom behavior data and the learning process data is introduced, and the dynamic weight adjustment is performed in combination with the effective participation features in the monitoring video, so that the real learning state of students can be more accurately reflected, the intelligentization and adaptive ability of the model are improved, the generation of the standardized data is more objective and targeted, and the accuracy and reliability of the subsequent analysis results are improved.

[0010] Optionally, state features of students are collected, the state features including a body movement amplitude feature, a face orientation angle change feature and a group interaction state feature; When the state features exceed a safety threshold, it is determined that there is an abnormal teaching event; If the abnormal teaching event exists, the classroom behavior data is frozen, the classroom behavior data is marked as invalid data segment and removed; The valid data before the generation time of the invalid data segment is taken as regular data; Based on the regular data, missing data and a confidence degree are generated and calculated. if the confidence exceeds a threshold, adding the missing data to the classroom behavior data; if the confidence does not exceed the threshold, freezing the missing data.

[0011] By adopting the above technical solutions, the student state features are intelligently collected and analyzed, combined with the identification of abnormal teaching events and the data effectiveness judgment mechanism, abnormal data can be eliminated and intelligently completed and confidence evaluation can be performed based on historical regular data, the accuracy of the classroom behavior data and the robustness of the analysis model are improved, thereby ensuring the reliability and practicality of the learning situation analysis results, and enhancing the adaptability and intelligent level of the system in complex teaching environments.

[0012] Optionally, a fairness index of the learning situation analysis result is detected, the fairness index including at least one of a score distribution difference degree of different gender student groups, a recommendation similarity of different academic foundation student groups, and an attention deviation value of a special needs student group; if any of the fairness indexes exceeds a deviation threshold, a feature importance distribution is generated according to the fairness index, and the learning situation analysis result is recorded as a learning situation analysis result with potential deviation, the feature importance distribution containing an influence degree score of an input feature of the large language model on a final analysis result; According to the feature importance distribution, a fairness constraint condition is dynamically configured in the large language model, and a corrected learning situation analysis result is regenerated based on the fairness constraint condition; The learning situation analysis result with potential deviation and the corrected learning situation analysis result are simultaneously displayed in a visual interface.

[0013] By adopting the above technical solutions, a fairness index detection mechanism is introduced, combined with feature importance analysis and dynamic constraint configuration, which can identify and correct potential deviations in learning situation analysis results, improve the transparency and explainability of the analysis process, ensure the accuracy of the analysis while enhancing the adaptability and fairness of the system to different student groups, and provide more fair and reliable learning situation support for teaching decisions.

[0014] Optionally, a model accuracy change amount when the fairness constraint condition is applied is measured; When the model accuracy change amount decreases by more than an accuracy threshold, a fairness-accuracy evaluation process is started, and key performance data of the large language model is output, the key performance data including an original model accuracy, a model accuracy under the fairness constraint condition, and a prediction difference degree between sensitive attribute groups; According to the key performance data, a relationship between a fairness improvement degree and the model accuracy change amount is calculated, and a plurality of selectable mediation schemes are generated, the mediation schemes including different combinations of fairness and accuracy; According to the mediation scheme, a bias parameter is selected, an updated constraint strength parameter is dynamically calculated, the constraint strength parameter is fed back to the fairness constraint condition for adaptive adjustment, and the bias parameter is 0, which represents accuracy priority, and the bias parameter is 1, which represents fairness priority.

[0015] By adopting the above technical solutions, model performance detection and multi-objective optimization mechanism are introduced, the fairness constraint strength can be dynamically adjusted under the premise of ensuring the overall accuracy of the model, the adaptive balance of fairness and accuracy is realized, the flexibility and applicability of the system in different application scenarios are improved, the learning situation analysis result is ensured to have high analysis reliability while considering fairness, and the controllability and actual landing value of the model are enhanced.

[0016] Optionally, at least one difference feature parameter between teachers and students is extracted, and the difference feature parameter includes a knowledge point coverage difference, a training intensity difference, or a cognitive load difference. Based on the difference feature parameter, a comprehensive conflict index is calculated. If the comprehensive conflict index exceeds a comprehensive threshold, based on the learning situation analysis result, a first expected effect of a teacher suggestion and a second expected effect of a student feedback scheme are respectively predicted, and a coordination scheme is generated; According to the coordination scheme, the display content and form of the coordination scheme are dynamically configured, and the learning situation analysis result is optimized.

[0017] By adopting the above technical solutions, the knowledge point coverage, training intensity, and cognitive load differences between teachers and students are extracted, the coordination scheme is generated, and the display content and form are dynamically configured, which can effectively identify and alleviate the deviation in learning situation cognition of the teaching parties, improve the pertinence and adaptability of teaching intervention, and enhance the efficiency of teacher-student interaction and the intelligent level of teaching decision-making.

[0018] Optionally, when it is detected that the coordination scheme does not achieve an expected effect for N consecutive times, a traceability process is triggered; Based on the traceability process, a historical feature parameter in the coordination scheme is extracted, and a feature parameter of a dominant conflict is determined; According to the feature parameter, a conflict traceability report is generated, and the conflict traceability report includes a duration change curve of a key conflict parameter and a matching degree analysis of related teaching resources; An alarm signal and the conflict traceability report are pushed to an administrator terminal, and automatic coordination functions on student terminals and teacher terminals are locked; In response to receiving a manual confirmation instruction, unlocking the automatic coordination function on the student terminal and the teacher terminal.

[0019] By adopting the above technical solutions and introducing a continuous monitoring and traceability mechanism for coordination effects, it is possible to automatically trigger traceability analysis when multiple coordination attempts are ineffective, accurately identify the key factors that dominate the conflict, and generate a traceability report. Combined with human intervention, it improves the reliability and security of system decision-making, effectively prevents the repeated execution of invalid mediation, and enhances the intelligent management level and exception handling capabilities of the teaching intervention system.

[0020] In the second aspect, the present application provides a learning situation analysis system based on a large language model, which adopts the following technical solutions: A learning situation analysis system based on a large language model, comprising: Acquisition module, used to obtain learning data and interaction information; A memory, used to store a program of the learning situation analysis method based on a large language model; The program in the processor memory can be loaded and executed by the processor to implement the learning situation analysis method based on the large language model.

[0021] By adopting the above technical solutions, the acquisition module collects student learning data in real time, the processor efficiently executes the learning situation analysis algorithm based on the large language model, and the memory continuously optimizes the analysis model and data feature library, realizing intelligent processing of the entire process from data collection to learning situation diagnosis to teaching intervention. While improving the accuracy and pertinence of analysis, it significantly enhances the efficiency of teaching decision-making, providing efficient and reliable technical support for precise teaching and teaching students in accordance with their aptitude in smart education scenarios.

[0022] In a third aspect, the present application provides a smart terminal that adopts the following technical solution: An intelligent terminal includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any one of the above methods.

[0023] In a fourth aspect, the present application provides a computer storage medium capable of storing corresponding programs, which has the characteristics of facilitating the improvement of teaching quality and meeting the actual teaching needs, and adopts the following technical solutions: A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any of the above-mentioned learning situation analysis methods based on a large language model.

[0024] In summary, this application includes at least one of the following beneficial technical effects: The fusion of structured and unstructured learning data of students, the deep understanding and personalized analysis of learning characteristics by using large language models, the dynamic generation of analysis results combined with the teaching needs of teachers and the feedback of students, and the presentation through multi-dimensional visualization effectively improve the comprehensiveness, accuracy and interactivity of learning analysis, and significantly enhance the scientificity of teaching decision and the pertinence of learning guidance. The consistency detection mechanism of classroom behavior data and learning process data is introduced, and the dynamic weight adjustment is combined with the effective participation features in the monitoring video, which can more accurately reflect the real learning state of students, improve the intelligent and adaptive ability of the model, make the generation of standardized data more objective and targeted, and thus improve the accuracy and reliability of the subsequent analysis results. The fairness index detection mechanism is introduced, combined with feature importance analysis and dynamic constraint configuration, which can identify and correct potential biases in learning analysis results, improve the transparency and explainability of the analysis process, and enhance the adaptability and fairness of the system to different student groups while ensuring the accuracy of the analysis, providing more fair and reliable learning support for teaching decisions. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a flowchart of a learning analysis method based on a large language model provided by an embodiment of the present application.

[0026] Figure 2 is a flowchart of a classroom learning state adaptive analysis method provided by an embodiment of the present application.

[0027] Figure 3 is a flowchart of a classroom behavior data intelligent repair method provided by an embodiment of the present application.

[0028] Figure 4 is a flowchart of a learning analysis fairness correction method provided by an embodiment of the present application.

[0029] Figure 5 is a flowchart of a fairness and accuracy adaptive balancing method provided by an embodiment of the present application.

[0030] Figure 6 is a flowchart of a teacher-student teaching conflict intelligent mediation method provided by an embodiment of the present application.

[0031] Figure 7 is a flowchart of a teaching mediation failure traceability intervention method provided by an embodiment of the present application.

[0032] Figure 8 is a structural diagram of a learning analysis system based on a large language model provided by an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application. Figures 1 to 8 The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application.

[0034] The embodiments of the present application disclose a learning situation analysis method based on a large language model. Referring to the accompanying drawings, Figure 1 The method comprises the following steps: Step S101: collecting learning data of students, the learning data comprising structured learning data and unstructured learning data, the structured learning data comprising examination results and homework completion, and the unstructured learning data comprising classroom behavior data, learning process data, student self-evaluation and mutual evaluation data, and teacher teaching records.

[0035] The learning data refers to a data set reflecting multi-dimensional information such as student learning behavior, knowledge mastery and learning attitude.

[0036] The structured learning data refers to data with fixed format and explicit fields.

[0037] The unstructured data refers to data with no fixed format and difficult to directly quantify and analyze.

[0038] For example, the historical examination results and homework submission records of students in the mathematics subject are obtained, and the classroom speaking frequency and posture change are collected by using a classroom camera.

[0039] Step S102: preprocessing the learning data to generate standardized data that can be used for analysis.

[0040] The preprocessing refers to cleaning, converting and integrating the collected multi-source data.

[0041] The standardized data refers to the unified format data obtained by cleaning, normalizing and feature extraction on the collected original learning data.

[0042] For example, the structured data is filled with missing values and outliers are removed, and the examination results and homework scores are standardized by using a normalization method; the unstructured data is segmented, stop word filtered, semantically annotated and sentiment analyzed, and the key words and feature vectors are extracted to generate standardized data that can be used for model input.

[0043] Step S103: receiving input interaction information, generating analysis requirement parameters, the interaction information comprising analysis requirements of teachers or feedback information of students.

[0044] The analysis requirement parameter refers to a structured control signal generated by parsing the original interaction information.

[0045] For example, the teacher inputs "analyze Zhang San's knowledge mastery of the physical mechanics part" on the terminal interface, and uses natural language processing technology to identify the analysis object as "Zhang San", the analysis dimension parameter as "physical mechanics part", and generate corresponding analysis requirement parameters; if the student feedback "I find it difficult to understand circuit analysis", the system extracts the analysis object as the current student, the analysis dimension parameter as "circuit analysis", and generates corresponding analysis requirement parameters.

[0046] Step S104: Based on the large language model, semantic understanding and in-depth analysis are performed on the standardized data and analysis requirement parameters to obtain learning situation analysis results.

[0047] A large language model refers to a language model with strong semantic understanding and reasoning capabilities, which can perform deep semantic analysis of input standardized data and analysis requirement parameters.

[0048] The learning situation analysis results refer to the comprehensive evaluation conclusions output after multi-dimensional analysis of standardized learning data and analysis requirement parameters through a large language model.

[0049] For example, it is identified that a student performs well in algebraic reasoning but makes frequent errors in geometric proofs, thus judging that the student has a weak link in this knowledge point.

[0050] Step S105: Display the learning situation analysis results in a visual manner.

[0051] Visualization refers to presenting the results of learning situation analysis in the form of charts, maps or graphs.

[0052] For example, a knowledge graph is used to show the relationship between students' mastery of different knowledge points; a heat map is used to show the distribution of students' mastery of each knowledge point; and a line graph is used to show changes in students' learning trends.

[0053] By adopting the above technical solutions, integrating students' structured and unstructured learning data, using large language models to achieve in-depth understanding and personalized analysis of learning characteristics, combining teachers' teaching needs with students' feedback to dynamically generate analysis results, and presenting them through multi-dimensional visualization, it effectively improves the comprehensiveness, accuracy and interactivity of learning situation analysis, and significantly enhances the scientific nature of teaching decisions and the targeted nature of learning guidance.

[0054] The embodiment of the present application discloses a method for adaptively analyzing classroom learning status. Figure 2 , the method comprising: Step S201: Detecting the consistency between classroom behavior data and learning process data.

[0055] The classroom behavior data refers to the behavior characteristics of students in the classroom collected by cameras, sensors and other devices, such as the attendance of students, the performance of classroom interaction.

[0056] The learning process data refers to non-behavioral data generated by students in the learning process, such as the learning duration and learning path on the online learning platform.

[0057] For example, the classroom behavior data of a student frequently speaking in a class and focusing his eyes on the blackboard is collected, and the learning process data of the student not browsing knowledge points for a long time on the learning platform is recorded, and the system determines whether the two are consistent based on this.

[0058] Step S202: If the consistency does not meet the preset standard, when the classroom behavior data is higher than the first data threshold and the learning process data is lower than the second data threshold, it is determined that there is a logical conflict.

[0059] The first data threshold refers to the activity level for measuring classroom behavior.

[0060] The second data threshold refers to the persistence or effectiveness of the learning process.

[0061] The logical conflict refers to a significant difference between the learning state reflected by the classroom behavior data and the learning process data.

[0062] For example, the number of times a student speaks in class and the time of focusing his eyes on the blackboard both exceed the first data threshold, but the number of clicks on knowledge points in the learning platform does not reach the second data threshold, so it is determined that there is a logical conflict between the classroom behavior data and the learning process data of the student.

[0063] Step S203: In response to the logical conflict, analyze the effective participation features in the monitoring video data in the conflict period to obtain an effectiveness score, the effective participation features including the angle between the face and the blackboard or the teacher, the relevance of body movements to learning tasks, and effective speaking in group interaction.

[0064] The effectiveness score refers to the real learning participation of the student in the conflict period.

[0065] Through face recognition and posture detection of the monitoring video data, the position of the student is located and the angle between his face and the blackboard or the teacher is tracked to determine the concentration of attention, the relevance of body movements to learning tasks is identified, and the semantic features of the speaking content in the group discussion are analyzed to determine whether it is around the learning theme, so as to comprehensively obtain the degree of effective participation.

[0066] The effectiveness score is calculated by weighting the face orientation angle, the body movement correlation and the group speech effectiveness, wherein the face orientation weight is 0.4, the body movement weight is 0.3, and the group speech weight is 0.3. After normalization processing, the final effectiveness score is obtained by weighted summation, which is used to judge the actual learning participation of the student.

[0067] For example, the video clip of a student in the conflict period is called, the face always faces the teacher and speaks in the group discussion is analyzed, and the final effectiveness score is 75, indicating that the actual participation is high.

[0068] Step S204: If the effectiveness score exceeds the effectiveness threshold, the first weight coefficient of the classroom behavior data is increased, and the second weight coefficient of the learning process data is decreased.

[0069] When the effectiveness score exceeds the effectiveness threshold, it indicates that the actual learning participation of the student in the conflict period is high, and the monitoring video data analysis result verifies the authenticity and concentration of the classroom behavior data of the student. Therefore, increasing the weight of the classroom behavior data can more accurately reflect the learning state of the student, and relatively decreasing the weight of the learning process data can avoid misleading the analysis result due to short-term low activity.

[0070] Step S205: If the effectiveness score does not exceed the effectiveness threshold, the third weight coefficient of the classroom behavior data is decreased, and the fourth weight coefficient of the learning process data is increased.

[0071] When the effectiveness score does not exceed the effectiveness threshold, it indicates that the actual learning participation of the student in the conflict period is low, and the analysis of the classroom behavior data of the student may exist the condition of surface activity but insufficient learning investment. Therefore, decreasing the weight of the classroom behavior data can avoid misleading the learning situation analysis, and increasing the weight of the learning process data can more accurately reflect the real cognitive state and learning effect of the student, thereby ensuring the objectivity and reliability of the analysis result.

[0072] Step S206: The adjusted classroom behavior data and learning process data are weighted and fused according to the first weight coefficient, the second weight coefficient, the third weight coefficient and the fourth weight coefficient to generate standardized data.

[0073] After multiplying the adjusted classroom behavior data and learning process data by their corresponding weight coefficients, the weighted sum is obtained by adding them, and then the normalized processing is performed to map it to a unified numerical interval.

[0074] By adopting the technical scheme, the consistency detection mechanism of the classroom behavior data and the learning process data is introduced, and the effective participation features in the monitoring video are combined for dynamic weight adjustment, so that the real learning state of the student can be more accurately reflected, the intelligentization and adaptive ability of the model are improved, the generation of the standardized data is more objective and targeted, and therefore the accuracy and reliability of the subsequent analysis result are improved.

[0075] The embodiment of the application discloses a classroom behavior data intelligent repairing method. Referring to Figure 3 The method comprises the following steps. Step S301: collecting state features of students, the state features comprising limb action amplitude features, face orientation angle change features and group interaction state features.

[0076] The state features refer to multi-dimensional features reflecting the behavior performance of students in the classroom.

[0077] Through the camera, the limb action of a student standing frequently and waving hands greatly in a class is collected, the angle change of the face orientation to the blackboard is combined, and the behavior features of speaking in the group discussion are combined, the image recognition algorithm is used to analyze the limb action amplitude and frequency, the posture estimation model is used to calculate the angle change between the face orientation and the blackboard or the teacher, the speech frequency in the group interaction and the content correlation are judged by combining the speech recognition and semantic analysis technology.

[0078] Step S302: when the state features exceed a safety threshold, it is determined that an abnormal teaching event exists.

[0079] The abnormal teaching event refers to an abnormal behavior event occurring in the teaching process, such as emotional agitation or fighting of students.

[0080] By comparing the state features of students in real time, when the limb action amplitude exceeds 1.5 times of the set standard value, the face orientation angle is greater than 60 degrees and the duration is more than 30 seconds, the group interaction frequency is more than 10 times per minute or the speech overlap degree is more than 40% caused by the simultaneous high-frequency speech of multiple people, it is determined that the state features are abnormal, and then it is identified that an abnormal teaching event possibly existing in the teaching order.

[0081] For example, a student continuously and frequently taps the table, stands and walks and the face orientation is far away from the teacher in a class, the limb action amplitude and the face orientation angle change both exceed the set safety threshold, and the system determines that an abnormal teaching event exists in the time period.

[0082] Step S303: if the abnormal teaching event exists, the classroom behavior data is frozen, the classroom behavior data is marked as invalid data segment and removed.

[0083] Freezing classroom behavior data refers to suspending the use of classroom behavior data collected during the abnormal teaching event time period.

[0084] Invalid data segment refers to marking data during the abnormal teaching event time period as abnormal data and not participating in subsequent analysis.

[0085] For example, the student classroom speech frequency, eye movement trajectory and other classroom behavior data collected during the occurrence of the abnormal teaching event are marked as invalid data segments and removed to avoid interference with the subsequent learning situation analysis results.

[0086] Step S304: Take the valid data before the generation time of the invalid data segment as regular data.

[0087] Regular data refers to a data set collected before the invalid data segment and conforming to the normal student state characteristics.

[0088] For example, the classroom behavior data collected within 5 minutes before the invalid data segment is extracted as regular data for subsequent generation and completion of missing data.

[0089] Step S305: Based on the regular data, generate and calculate missing data and confidence.

[0090] Missing data refers to classroom behavior data that is missing due to freezing of abnormal events.

[0091] Using regular data as historical input, a time series prediction algorithm is used to model the trend of classroom behavior data, and then the classroom behavior data that should have appeared during the abnormal time period is predicted based on the model output.

[0092] The confidence formula for calculating missing data is: C = Cmax-k*t, where Cmax is the maximum confidence benchmark value, default 0.95, t is the data missing duration, k is the decay coefficient, default 0.1 / min.

[0093] For example, if an abnormal teaching event causes 2 minutes of behavior data to be missing, the confidence of the missing data is: C = 0.95-0.1x2 = 0.75.

[0094] Step S306: If the confidence exceeds the threshold, add the missing data to the classroom behavior data.

[0095] When the confidence of the missing data is higher than the threshold, it indicates that the data has a high degree of credibility based on historical regular data and can accurately reflect the learning state of students during the abnormal time period. Therefore, the missing data is supplemented to the classroom behavior data, which can improve the data integrity and the reliability of the learning situation analysis results.

[0096] For example, the confidence of a certain missing data segment is 0.75, which is higher than the threshold value 0.7, so it is included in the classroom behavior data set to generate the learning situation analysis result.

[0097] Step S307: If the confidence does not exceed the threshold value, freeze the missing data.

[0098] When the confidence of the missing data is lower than the threshold value, the data is considered unreliable, and the missing data is frozen and does not participate in subsequent analysis.

[0099] For example, the confidence of a certain missing data segment is only 0.6, which is lower than the threshold value 0.7, so the missing data is frozen and does not participate in classroom behavior data analysis.

[0100] By using the above technical solution, the student state characteristics are intelligently collected and analyzed, combined with the recognition of abnormal teaching events and the data validity judgment mechanism, abnormal data can be eliminated and intelligent completion and confidence evaluation can be performed based on historical regular data, the accuracy of classroom behavior data and the robustness of analysis model are improved, thereby ensuring the reliability and practicality of learning situation analysis results, and enhancing the adaptability and intelligent level of the system in complex teaching environment.

[0101] The embodiment of the application discloses a learning situation analysis fairness correction method. Figure 4 The method comprises: Step S401: detecting a fairness index of a learning situation analysis result, the fairness index comprising at least one of a score distribution difference degree of different gender student groups, a suggestion similarity degree of different academic foundation student groups, and an attention deviation value of a special needs student group.

[0102] The fairness index is a quantitative index for measuring whether the learning situation analysis result has systematic bias between different student groups.

[0103] By statistically analyzing the data of different student groups in the learning situation analysis result, the score distribution difference degree of different gender student groups is calculated, the semantic similarity of learning suggestions obtained by different academic foundation students is compared, and the deviation degree of the attention of the special needs student group from the ordinary students in the analysis result is evaluated, so as to detect whether the learning situation analysis result has potential systematic bias.

[0104] Step S402: If any fairness index exceeds a bias threshold value, a feature importance distribution is generated according to the fairness index, and the learning situation analysis result is recorded as a learning situation analysis result with potential bias, and the feature importance distribution comprises an influence degree score of an input feature of a large language model on a final analysis result.

[0105] The feature importance distribution refers to the distribution formed by scoring the importance of each input feature affecting the learning situation analysis result, and is used to identify which features have a greater impact on the generation of bias.

[0106] After detecting that the fairness indicator exceeds the bias threshold, the influence degree of each input feature is analyzed based on the correlation between the input features and the output results of the large language model, and the feature attribution algorithm is used to quantify the contribution of different features to the analysis result, thereby generating a feature importance distribution for identifying key features that cause bias.

[0107] For example, if the gender score difference is detected to exceed the bias threshold, backtracking analysis finds that the "mathematics subject performance" feature has too high a weight in the model, and statistical data shows that the average score of boys in this subject is generally higher than that of girls, resulting in a low overall assessment of the female group by the model, thus systematically underestimating the student group that does not have an advantage. Therefore, this analysis result is marked as having potential bias.

[0108] Step S403: According to the feature importance distribution, dynamically configure fairness constraints in the large language model, and regenerate the corrected learning situation analysis result based on the fairness constraints.

[0109] The fairness constraint refers to the model adjustment rule set based on the feature importance distribution, which is used to limit the influence of certain features on the analysis result to reduce the risk of bias.

[0110] According to the feature importance distribution obtained by backtracking analysis, the weight distribution of related features in the large language model is automatically adjusted, and fairness constraints are dynamically introduced in the process of generating learning situation analysis results. High-impact bias features are inhibited, and features that are beneficial to fairness are enhanced, thereby reducing the impact of the original bias in the corrected learning situation analysis result.

[0111] For example, according to the feature importance distribution, it is found that "classroom speaking frequency" has too great an impact on the learning situation analysis result, so the weight of this feature is dynamically reduced in the large language model, and the influence weight of "assignment completion quality" and "knowledge point mastery degree" is increased, thereby generating a more fair corrected analysis result.

[0112] Step S404: Simultaneously display the learning situation analysis result with potential bias and the corrected learning situation analysis result in the visualization interface.

[0113] Simultaneously displaying the learning situation analysis result with potential bias and the corrected learning situation analysis result in the visualization interface is to clearly compare the differences between the large language model before and after fairness adjustment, and help teachers or students understand the source of bias and the correction effect.

[0114] By adopting the technical scheme, the fairness index detection mechanism is introduced, the feature importance analysis and the dynamic constraint configuration are combined, potential bias in the learning situation analysis result can be identified and corrected, the transparency and the explainability of the analysis process are improved, the adaptability and the fairness of the system to different student groups are enhanced while the analysis accuracy is ensured, and more fair and reliable learning situation support is provided for teaching decision-making.

[0115] The embodiment of the application discloses a fairness and accuracy adaptive balancing method. Figure 5 The method comprises the following steps. Step S501: measuring a model accuracy rate change amount when a fairness constraint condition is applied.

[0116] The model accuracy rate change amount refers to the change value of the model accuracy rate before and after the fairness constraint condition is introduced, and the formula is ΔA = (A0-A1) / A0, wherein A0 represents the original model accuracy rate, and A1 represents the model accuracy rate under the fairness constraint condition.

[0117] For example, when the fairness constraint condition is not applied, the model accuracy rate is 90%, and after the fairness constraint condition is applied, the accuracy rate decreases to 81%, and the model accuracy rate change amount is 10%.

[0118] Step S502: when the model accuracy rate change amount decreases by more than an accuracy rate threshold, a fairness-accuracy evaluation process is started, and key performance data of the large language model is output, the key performance data comprising the original model accuracy rate, the model accuracy rate under the fairness constraint condition, and a prediction difference degree between sensitive attribute groups.

[0119] The fairness-accuracy evaluation process refers to a performance analysis mechanism triggered when the model accuracy rate decreases by more than the accuracy rate threshold, and by comprehensively comparing the key data such as the original model accuracy rate, the model accuracy rate under the fairness constraint condition, and the prediction difference degree between sensitive attribute groups, the performance change of the model after the fairness constraint condition is introduced is evaluated, so as to judge the balance state between the fairness improvement and the accuracy maintenance, and provide a basis for the generation of a subsequent mediation scheme.

[0120] The prediction difference degree between sensitive attribute groups represents the difference degree between the model prediction results for student groups with different sensitive attributes, such as gender and academic foundation, under different model constraint conditions.

[0121] For example, if the model accuracy rate change amount exceeds the accuracy rate threshold, the original model accuracy rate is 90%, the accuracy rate after the fairness constraint is applied is 83%, and the prediction difference degree between sensitive attribute groups decreases from 0.35 to 0.15.

[0122] Step S503: Calculate the relationship between the fairness improvement degree and the model accuracy change according to the key performance data, and generate multiple optional mediation solutions, each of which includes different combinations of fairness and accuracy.

[0123] The degree of fairness improvement is used to measure the model's improvement in fairness. The formula is ΔF = (D0-D1) / D0, where D0 represents the difference in the model's predictions for different sensitive attribute groups without applying fairness constraints, and D1 represents the difference in the model's predictions for different sensitive attribute groups after applying fairness constraints.

[0124] Based on the two key indicators of fairness improvement and model accuracy change, the fairness improvement and model accuracy are taken as optimization targets. A multi-objective optimization algorithm is used to find the optimal solution set between the two and generate multiple mediation schemes. Each mediation scheme corresponds to a specific combination of fairness improvement and model accuracy.

[0125] For example, based on the detected fairness improvement of 40% and the decrease in model accuracy of 8%, the constraint parameters are adjusted on this basis to generate three mediation schemes: Scheme 1 has a 3% decrease in model accuracy when the fairness improvement is 30%, Scheme 2 has a 10% decrease in model accuracy when the fairness improvement is 50%, and Scheme 3 has a 15% decrease in model accuracy when the fairness improvement is 60%. Each scheme corresponds to a different combination of fairness improvement and model accuracy for users to choose according to actual needs.

[0126] Step S504: Select a bias parameter according to the mediation scheme, dynamically calculate the updated constraint strength parameter, and feed the constraint strength parameter back to the fairness constraint condition for adaptive adjustment. A bias parameter of 0 represents accuracy priority, and a bias parameter of 1 represents fairness priority.

[0127] The bias parameter is a parameter set by the user based on actual needs. It is used to prioritize the model between fairness and accuracy, and its value range is [0,1].

[0128] The constraint strength parameter refers to the parameter that controls the intensity of the fairness constraint in the model. The higher the value, the stronger the constraint. The calculation formula is: λ=λa+(1-λ)(1-a), where λ represents the current constraint strength parameter and a represents the set bias parameter.

[0129] The bias parameter selected by the user and the current constraint strength parameter are weighted according to the formula, and the constraint strength parameter is dynamically adjusted. When the bias parameter is 0, it completely tends to retain the model accuracy, and when it is 1, it completely tends to improve fairness, thereby achieving adaptive adjustment of fairness constraints to balance model performance.

[0130] For example, the current constraint strength parameter λ = 0.6, the user sets the bias parameter a = 0.8 (biased towards fairness), and the updated constraint strength parameter is λ = 0.6 * 0.8 + (1-0.6) * (1-0.8) = 0.56 by substituting the formula.

[0131] By adopting the technical solutions described above, the model performance detection and multi-objective optimization mechanism are introduced, which can dynamically adjust the fairness constraint strength while ensuring the overall accuracy of the model, achieve adaptive balance between fairness and accuracy, improve the flexibility and applicability of the system in different application scenarios, ensure that the learning analysis results take into account fairness while maintaining high analysis reliability, and enhance the controllability and practical value of the model.

[0132] The embodiment of the application discloses a kind of intelligent mediation methods for teacher-student teaching conflict. Refer to Figure 6 The method comprises: Step S601: extract at least one difference characteristic parameter between teachers and students, and the difference characteristic parameter includes knowledge point coverage difference, training intensity difference or cognitive load difference.

[0133] The difference characteristic parameter is a quantitative index reflecting the understanding deviation or cognitive difference between the teacher's suggestion and the student's feedback.

[0134] The knowledge point coverage difference is obtained by comparing the knowledge point distribution of the teacher's suggestion and the attention point distribution of the student's feedback, the training intensity difference is calculated by quantifying the difference value of the training time of the suggestion of both parties, and the cognitive load difference is determined based on the difference between the evaluation scores of the task difficulty of both parties.

[0135] By natural language processing on the teacher-student exchange text, the knowledge point content, training time description and task difficulty evaluation in the teacher's suggestion and the student's feedback are extracted using text classification and keyword recognition technology, and the differences in knowledge point coverage, training intensity requirement and cognitive load of the two are compared respectively, to form characteristic parameters such as knowledge point coverage difference, training intensity difference and cognitive load difference, which are used for subsequent teaching conflict analysis and mediation scheme generation.

[0136] Step S602: calculate the comprehensive conflict index based on the difference characteristic parameter.

[0137] The comprehensive conflict index is a numerical value that measures the overall conflict degree between teachers and students in teaching content, training arrangement and cognitive understanding, and the formula is C = W1 * D k + W2 * D t + W3 * D c , wherein D k is the knowledge point coverage difference, D t is the training intensity deviation value, and D cCognition load difference coefficient, W1, W2, W3 correspond to the weight coefficients of each difference characteristic parameter, and satisfy W1+W2+W3=1.

[0138] By multiplying each difference characteristic parameter by the corresponding weight coefficient, the sum of the values obtained is used to quantify the overall conflict between teachers and students in terms of knowledge coverage, training intensity, and cognitive load. The specific calculation method is to weight and sum the knowledge coverage difference, training intensity difference, and cognitive load difference after assigning them preset weights, and obtain a comprehensive value as the conflict index to determine the severity of the teaching conflict.

[0139] For example, if D k =0.4, D t =1, and D c =2, and the weights are set as W1=0.5, W2=0.3, and W3=0.2, then the comprehensive conflict index is C=0.5×0.4+0.3×1+0.2×2=0.2+0.3+0.4=0.9.

[0140] Step S603: If the comprehensive conflict index exceeds the comprehensive threshold, based on the learning situation analysis results, predict the first expected effect of the teacher's suggestion and the second expected effect of the student's feedback scheme respectively, and generate a coordination scheme.

[0141] The first expected effect refers to the teaching effectiveness that the teaching scheme may achieve in terms of knowledge mastery improvement and learning goal achievement based on the teacher's teaching suggestion and the learning situation analysis results.

[0142] The second expected effect refers to the learning effectiveness that the learning scheme may achieve in terms of learning adaptability and task completion based on the student's feedback and the learning situation analysis results.

[0143] Through multi-objective optimization algorithm to find a compromise solution among multiple conflicting objectives, first, the teacher's suggestion of knowledge coverage, training intensity, and student's feedback scheme of cognitive load, acceptance, etc. are quantitatively modeled, then under the constraint conditions of meeting the actual learning situation and teaching requirements, the algorithm is iteratively calculated to obtain multiple feasible solutions, from which the optimal compromise scheme that takes into account the teaching effectiveness and student adaptability is selected as the coordination suggestion.

[0144] Step S604: According to the coordination scheme, dynamically configure the display content and form of the coordination scheme, and optimize the learning situation analysis results.

[0145] According to the identity of teachers and students and the type of terminal, the coordination scheme is presented in different ways to ensure the effectiveness and effectiveness of information transmission. The teaching strategy optimization suggestion containing the overall knowledge mastery improvement curve of the class is pushed to the teacher terminal, and the personalized adjustment scheme integrating individual learning characteristic data is output to the student terminal.

[0146] For example, a graph showing that the overall knowledge mastery of the class has improved from 60% to 70% is displayed to the teacher, and the personalized suggestion of "review application problems this week, train 1.5 hours a day" is pushed to the students.

[0147] By adopting the above technical solutions, the difference characteristics such as knowledge point coverage, training intensity and cognitive load between teachers and students are extracted, a coordination scheme is generated, and the display content and form are dynamically configured, which can effectively identify and alleviate the deviation in the cognition of the teaching parties, improve the pertinence and adaptability of the teaching intervention, and enhance the efficiency of the teacher-student interaction and the intelligent level of the teaching decision.

[0148] The embodiment of the application discloses a teaching mediation failure traceability intervention method. Referring to Figure 7 The method comprises the following steps: Step S701: When it is detected that the coordination scheme fails to achieve the expected effect for N consecutive times, a traceability process is triggered.

[0149] The traceability process refers to a backtracking analysis mechanism automatically started when the system detects that the coordination scheme fails to achieve the expected effect for N consecutive times.

[0150] For example, if the coordination scheme generated by the system for 3 consecutive times fails to improve the knowledge mastery of the students to the target value, the traceability process is triggered.

[0151] Step S702: Based on the traceability process, historical characteristic parameters in the coordination scheme are extracted, and a characteristic parameter of a dominant conflict is determined.

[0152] The historical characteristic parameters refer to the parameters related to the teacher-student teaching conflict extracted in the previous coordination processes, such as knowledge point coverage difference, training intensity difference, cognitive load difference and the like.

[0153] The characteristic parameter of the dominant conflict refers to the characteristic parameter that has the greatest impact on the conflict in the multiple coordination failures.

[0154] By analyzing the historical characteristic parameters in the previous coordination schemes through the attention weight distribution, the influence degree of the parameters such as the knowledge point coverage difference, the training intensity difference and the cognitive load difference in the coordination failure process is quantitatively sorted, and the characteristic parameter with a higher attention weight indicates that it plays a dominant role in the coordination conflict, so that the core conflict factor causing the coordination failure is identified.

[0155] For example, the knowledge point coverage difference, the training intensity difference and the cognitive load difference in the last 3 coordinations are extracted, and it is found through the attention weight distribution that the weight of the knowledge point coverage difference is the highest, indicating that the knowledge point coverage difference is the main reason for the current coordination failure.

[0156] Step S703: generating a conflict trace report according to the characteristic parameters, the conflict trace report including the time-varying curve of the characteristic parameters of the dominant conflict and the matching degree analysis of the related teaching resources.

[0157] The conflict trace report refers to an analysis report for recording the reason for coordination failure, showing the change trend of the characteristic parameters of the dominant conflict and the adaptation of the teaching resources.

[0158] The time-varying curve of the characteristic parameters of the dominant conflict refers to showing the change of the characteristic parameters of the dominant conflict in the previous coordination in the form of a curve, helping to analyze the evolution trend.

[0159] The matching degree analysis of the related teaching resources refers to evaluating the matching degree between the current teaching resources, such as courseware, exercises, teaching methods, and the learning needs of students.

[0160] For example, the generated conflict trace report indicates that the "knowledge point coverage difference" is continuously high, and draws its change curve in three mediations, while analyzing that the teaching resources lack coverage of related knowledge points, with a matching degree of only 60%.

[0161] Step S704: pushing an alarm signal and a conflict trace report to the administrator terminal, and locking the automatic coordination function on the student terminal and the teacher terminal.

[0162] The alarm signal refers to the prompt information sent by the system to the administrator terminal after detecting continuous coordination failure, prompting manual intervention.

[0163] The automatic coordination function refers to the function of the system automatically pushing coordination schemes for teachers and students, and locking the automatic coordination function can prevent the continuous pushing of invalid schemes.

[0164] For example, the system sends a prompt to the administrator terminal: "the number of coordination failures exceeds the standard, please check the trace report and intervene", and suspends the pushing of coordination schemes to the teacher terminal and the student terminal, preventing the continuous pushing of invalid schemes.

[0165] Step S705: in response to receiving a manual confirmation instruction, unlocking the automatic coordination function on the student terminal and the teacher terminal.

[0166] The manual confirmation instruction refers to the instruction sent by the administrator to the system after checking the trace report and confirming the processing opinion, which is used to restore the automatic sending of coordination scheme function.

[0167] For example, the administrator clicks the "confirm processing completed" button, and after the system receives the manual confirmation instruction, it restores the pushing of coordination schemes to the teacher terminal and the student terminal, and the system regains the ability to push coordination schemes.

[0168] By adopting the technical scheme, the continuous monitoring and tracing mechanism of coordination effect is introduced, which can automatically trigger tracing analysis when multiple coordination is invalid, accurately identify the key factors of dominant conflict, and generate a tracing report, combined with manual intervention to improve the reliability and safety of system decision, effectively prevent repeated execution of invalid mediation, and enhance the intelligent management level and abnormal processing capability of the teaching intervention system.

[0169] Based on the same inventive concept, the embodiments of the present application provide a learning situation analysis system based on a large language model, please refer to Figure 8 The system comprises: The acquisition module 801 is configured to acquire learning data and interaction information. The memory 802 is configured to store the program of the learning situation analysis method based on the large language model. The processor 803 can load and execute the program in the memory, and implement the learning situation analysis method based on the large language model.

[0170] By adopting the technical scheme, the acquisition module acquires student learning data in real time, the processor efficiently executes the learning situation analysis algorithm based on the large language model, and the memory continuously optimizes the analysis model and data feature library, realizing intelligent processing of the whole process from data acquisition to learning situation diagnosis to teaching intervention, improving the analysis accuracy and pertinence, significantly enhancing the teaching decision efficiency, and providing efficient and reliable technical support for precise teaching and individualized teaching in the intelligent education scene.

[0171] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional modules is taken as an example for illustration, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0172] The embodiments of the present application provide a computer readable storage medium storing a computer program capable of loading and executing the learning situation analysis method based on the large language model by the processor.

[0173] The computer storage medium includes, for example, a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0174] Based on the same inventive concept, the embodiment of the present application provides a kind of intelligent terminal, including memory and processor, computer program capable of being loaded and executed by processor based on the learning situation analysis method of large language model is stored on memory.

[0175] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above-mentioned division of functional modules is taken as an example, and in actual application, the above-mentioned functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0176] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application, any feature disclosed in the specification (including abstract and drawings) can be replaced by other equivalent or similar purpose alternative features, unless specifically described. That is, each feature is only an example of a series of equivalent or similar features.

Claims

1. A learning situation analysis method based on a large language model, characterized in that: include: Collecting students' learning data, including structured learning data and unstructured learning data. The structured learning data includes test scores and homework completion status, and the unstructured learning data includes classroom behavior data, learning process data, student self-evaluation and peer evaluation data, and teacher teaching records; Preprocessing the learning data to generate standardized data that can be used for analysis; Receive input interaction information and generate analysis requirement parameters, wherein the interaction information includes the teacher's analysis requirements or the student's feedback information; Based on the large language model, semantic understanding and in-depth analysis are performed on the standardized data and the analysis requirement parameters to obtain learning situation analysis results; The learning situation analysis results are displayed in a visual manner.

2. A learning situation analysis method based on a large language model according to claim 1, characterized in that: After the unstructured learning data is preprocessed, the following steps are further included: detecting the consistency between the classroom behavior data and the learning process data; If the consistency does not meet the preset standard, it is determined that there is a logical conflict when the classroom behavior data is higher than a first data threshold and the learning process data is lower than a second data threshold; In response to the logical conflict, analyzing effective participation features in the surveillance video data retrieved during the conflict time period to obtain an effectiveness score, wherein the effective participation features include the angle between the face and the blackboard or the teacher, the relevance of body movements to the learning task, and effective speech in the group interaction; If the effectiveness score exceeds the effectiveness threshold, increasing the first weight coefficient of the classroom behavior data and decreasing the second weight coefficient of the learning process data; If the effectiveness score does not exceed the effectiveness threshold, reducing the third weight coefficient of the classroom behavior data and increasing the fourth weight coefficient of the learning process data; The classroom behavior data and the learning process data are weighted and fused according to the first weight coefficient, the second weight coefficient, the third weight coefficient and the fourth weight coefficient to generate the standardized data.

3. A learning situation analysis method based on a large language model according to claim 2, characterized in that: The abnormal situation of detecting classroom behavior data includes: Collecting student status characteristics, including body movement amplitude characteristics, facial orientation angle change characteristics, and group interaction status characteristics; When the state characteristic exceeds a safety threshold, it is determined that an abnormal teaching event exists; If the abnormal teaching event occurs, freezing the classroom behavior data, marking the classroom behavior data as an invalid data segment and removing it; Taking the valid data before the generation time of the invalid data segment as regular data; Based on the routine data, generate and calculate missing data and confidence levels; If the confidence exceeds a threshold, adding the missing data to the classroom behavior data; If the confidence level does not exceed a threshold, the missing data is frozen.

4. The learning situation analysis method based on a large language model according to claim 1, characterized in that: Also includes: Detecting a fairness index of the learning situation analysis result, the fairness index including at least one of the following: a score distribution difference between student groups of different genders, a suggestion similarity between student groups of different academic foundations, and an attention deviation value between student groups with special needs; If any of the fairness indicators exceeds the deviation threshold, a feature importance distribution is retroactively generated based on the fairness indicators, and the learning situation analysis result is recorded as a learning situation analysis result with potential bias. The feature importance distribution includes a score of the influence of the input features of the large language model on the final analysis result. Dynamically configuring fairness constraints in the large language model according to the feature importance distribution, and regenerating a corrected learning situation analysis result based on the fairness constraints; The learning situation analysis results with potential deviations and the corrected learning situation analysis results are simultaneously displayed in a visual interface.

5. A learning situation analysis method based on a large language model according to claim 4, characterized in that: The method further comprises: Measuring the change in model accuracy when the fairness constraints are applied; When the change in the model accuracy rate drops by more than the accuracy threshold, the fairness-accuracy evaluation process is initiated, and key performance data of the large language model is output. The key performance data includes the original model accuracy, the model accuracy under fairness constraints, and the prediction difference between sensitive attribute groups; Calculating the degree of fairness improvement based on the key performance data, and generating a plurality of optional mediation schemes, wherein the mediation schemes include different combinations of fairness and accuracy; A bias parameter is selected according to the mediation scheme, an updated constraint strength parameter is dynamically calculated, and the constraint strength parameter is fed back to the fairness constraint condition for adaptive adjustment. A bias parameter of 0 represents that accuracy is prioritized, and a bias parameter of 1 represents that fairness is prioritized.

6. The learning situation analysis method based on a large language model according to claim 1, characterized in that: The method further comprises: Extracting at least one difference characteristic parameter between the teacher and the student, wherein the difference characteristic parameter includes a difference in knowledge point coverage, a difference in training intensity, or a difference in cognitive load; Calculating a comprehensive conflict index based on the difference characteristic parameters; If the comprehensive conflict index exceeds the comprehensive threshold, based on the learning situation analysis results, respectively predict the first expected effect of the teacher's suggestion and the second expected effect of the student's feedback plan, and generate a coordination plan; According to the coordination plan, the display content and form of the coordination plan are dynamically configured to optimize the learning situation analysis results.

7. A learning situation analysis method based on a large language model according to claim 6, characterized in that: When the generation of the coordination solution fails, the following steps are included: When it is detected that the coordination plan fails to achieve the expected effect for N consecutive times, the traceability process is triggered; Extracting historical characteristic parameters in the coordination plan based on the tracing process to determine characteristic parameters of the dominant conflict; Generating a conflict tracing report based on the characteristic parameters, wherein the conflict tracing report includes a time-varying curve of key conflict parameters and a matching degree analysis of relevant teaching resources; Pushing an alarm signal and the conflict tracing report to the administrator terminal, and locking the automatic coordination function on the student terminal and the teacher terminal; In response to receiving a manual confirmation instruction, unlocking the automatic coordination function on the student terminal and the teacher terminal.

8. A learning situation analysis system based on a large language model, characterized in that: The system is used to execute the learning situation analysis method based on a large language model according to any one of claims 1 to 7, comprising: Acquisition module, used to obtain learning data and interaction information; A memory, used to store a program of the learning situation analysis method based on a large language model; The program in the processor memory can be loaded and executed by the processor to implement the learning situation analysis method based on the large language model.

9. An intelligent terminal, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored which can be loaded by a processor and execute the method according to any one of claims 1 to 7.

Images