Method and system for evaluating teaching plan text based on student group state

Abstract

The application discloses a method and system for evaluating teaching plan texts based on student group state, and relates to the field of intelligent education information technology. The application obtains teaching plan texts and performs structured analysis, and simultaneously obtains a target class student portrait; a classroom process simulation is performed according to the teaching plan texts, and an interactive log is generated; feedback enhancement processing is performed according to the interactive log and the student portrait, and a double-channel detection is designed to respectively perform group layer gap detection and implicit-extrinsic behavior mapping detection, and a risk time step set is obtained, so as to calculate a risk value and convert it into a positive quality score, and meanwhile, a diagnostic report is generated by backtracking to specific teaching sentences. The application converts traditional static text evaluation into dynamic group learning effect deduction, provides objective and quantifiable teaching plan pre-prediction, reduces trial and error cost, and can perform adaptive evaluation according to different student group portraits, and supports differentiated teaching plan optimization.

Description

Technical Field

[0001] This application relates to the field of smart education information technology, and in particular to a method and system for assessing lesson plan texts based on student group status. Background Technology

[0002] Currently, lesson plan evaluation primarily relies on manual review by curriculum experts to qualitatively assess the textual standardization, completeness of teaching procedures, and rationality of content. Simultaneously, some automated tools are beginning to incorporate natural language processing technology to extract keywords, match templates, or check structural compliance in lesson plans, providing preliminary quantitative scoring references.

[0003] However, existing technologies have significant shortcomings. Human assessment relies heavily on the subjective experience of reviewers, and the evaluation criteria among different experts vary significantly, making it difficult to guarantee the consistency and repeatability of assessment results. Existing automated scoring systems only analyze surface-level text features, failing to model the execution process and potential learning outcomes of the teaching activities designed in the lesson plan in the actual classroom, nor can they identify potential logical flaws or cognitive overload risks in the teaching process. More importantly, both methods lack the ability to adapt to different student groups (such as differences in age, grade, and cognitive level), and cannot predict the applicability of the same lesson plan to different types of learners, resulting in assessment conclusions that are disconnected from actual teaching.

[0004] Therefore, there is an urgent need for a lesson plan evaluation method that can integrate teaching execution process simulation with learner characteristic analysis to overcome the above-mentioned shortcomings and achieve objective, comprehensive and differentiated evaluation of lesson plan quality. Summary of the Invention

[0005] In order to overcome at least the above-mentioned shortcomings in the prior art, the technical problem to be solved by this application is: how to improve the process granularity of lesson plan evaluation before the formal implementation of the lesson plan and when the available real process feedback is limited, so that the evaluation results can reflect the dynamic differences of different student groups in the teaching process and accurately locate the teaching stage where risks occur, thereby providing a scientific basis for lesson plan optimization. The purpose of this application is to provide a method, system, device and medium for evaluating lesson plan texts based on student group status to solve the above problems.

[0006] Firstly, this application provides a method for assessing lesson plan texts based on the state of student groups, including:

[0007] Based on the lesson plan text and the target class roster, obtain the sequence of teaching statements and the student profile matrix of the target class;

[0008] Based on the sequence of teaching statements, an interaction log is obtained through a pre-set classroom process simulation;

[0009] Based on the interaction log and the student profile matrix, a latent state time series tensor is obtained; the latent state time series tensor is configured to include at least the student, time step, and learning state.

[0010] Based on the latent state temporal tensor, a first risk time step set and a second risk time step set are obtained through preset parallel dual-channel detection.

[0011] Based on the first risk time step set and the second risk time step set, the total number of risk triggers is obtained, and the evaluation result of the lesson plan text is obtained based on the total number of risk triggers.

[0012] In one possible implementation, obtaining the sequence of teaching statements and the student profile matrix of the target class based on the lesson plan text and the target class roster includes:

[0013] Based on the lesson plan text, a structured lesson plan element set is obtained through preset semantic segmentation rules; based on the structured lesson plan element set, a sequence of teaching statements is obtained through time window segmentation and instruction annotation.

[0014] Based on the target class roster, the prior knowledge level, instruction completion rate, and attention duration parameters of each student are obtained by retrieving information from the learning database.

[0015] Based on the prior knowledge level parameters, instruction completion parameters, and attention duration parameters of each student, individual profile vectors of all students are obtained through feature vectorization to obtain the student profile matrix of the target class.

[0016] In one possible implementation, each teaching statement in the sequence of teaching statements is configured with instruction labels including: a first label indicating whether all students are given the opportunity to participate, and a second label indicating whether all students are required to raise their hands in response;

[0017] The rule for determining the first marker is configured as follows: if the teaching statement contains a word from the preset vocabulary set of the first marker, the first marker is set to 1; otherwise, it is set to 0.

[0018] The rule for determining the second marker is configured as follows: if the teaching statement contains a word from the preset second marker vocabulary set, the second marker is set to 1; otherwise, it is set to 0.

[0019] In one possible implementation, obtaining the interaction log based on the sequence of teaching statements through a preset classroom process simulation includes:

[0020] Based on the teaching statement sequence, the simulation clock is advanced according to the preset scoring grid time interval to obtain the teaching statement flow executed in time step order;

[0021] Based on the teaching statement flow, obtain the current teaching statement and cumulative practice time corresponding to each time step to obtain the interaction log;

[0022] In response to the simulation clock reaching a preset query trigger point, an explanatory response is obtained by executing an interrupt-response-recovery atomic operation chain. The explanatory response is used to update the interaction log.

[0023] In one possible implementation, the implicit state time series tensor is obtained based on the interaction log and the student profile matrix. The configuration of the implicit state time series tensor includes at least the student, time step, and learning state, including:

[0024] Based on the current teaching statement corresponding to each time step of the interaction log and the student profile matrix, the student's attention anomaly value, cognitive load value and comprehension value at each time step are obtained to obtain the original state sequence.

[0025] Based on the original state sequence, a resampled state sequence is obtained by resampling at scoring grid time intervals;

[0026] Based on the resampled state sequence, a hidden state temporal tensor with three-dimensional alignment of student × time step × learning state is obtained.

[0027] In one possible implementation, the dual-channel detection is configured with implicit-explicit behavior mapping detection;

[0028] The implicit-explicit behavior mapping detection is configured as follows:

[0029] Acquire the behavioral image sequence that is synchronously acquired with the latent state temporal tensor, and obtain the first changing temporal tensor through a graph convolutional network;

[0030] Based on the implicit state temporal tensor, a second change temporal tensor is obtained through a state-behavior association network;

[0031] Based on the implicit state temporal tensor corresponding to the lesson plan to be evaluated, the behavior change temporal tensor is obtained based on the state-behavior association network.

[0032] Based on the behavioral change temporal tensor, the hand state and facial expression state of each student at each time step are obtained through the second behavior detection network;

[0033] Based on the first and / or second markers of each teaching statement in the teaching statement sequence, obtain the label of whether all students raised their hands at each time step;

[0034] If the current time step shows a marker indicating that all students have raised their hands, but no hand-raising behavior is detected, or a frowning behavior is detected, then the student's current behavior is determined to be abnormal and they are considered an abnormal student at the current time step.

[0035] Obtain the number of abnormal students at each time step. If the number of abnormal students exceeds a preset value of the total number of students at that time step, then mark that time step as the first risk time step.

[0036] In one possible implementation, the dual-channel detection is configured with population inter-layer gap detection;

[0037] The inter-layer gap detection of the execution group is configured as follows:

[0038] Based on the latent state time series tensor, the population average state curve is obtained;

[0039] Based on the population average state curve, the corresponding time steps that exceed the first preset threshold are marked to obtain an abnormal time step set.

[0040] Based on the set of abnormal time steps, obtain the average comprehension level of each student within the set; based on the average comprehension level of each student, sort them in ascending order, and divide the students in the first preset proportion of the top-ranked group into the low comprehension group, and the students in the first preset proportion of the bottom-ranked group into the high comprehension group.

[0041] Based on the high comprehension group and the low comprehension group, the absolute difference between the mean comprehension values ​​of the two groups is obtained at all time steps; based on the absolute difference at each time step, time steps that exceed the second preset threshold are marked to obtain the second risk time step set.

[0042] Secondly, this application provides a system based on a student group status assessment lesson plan text, including a sequentially electrically connected data acquisition unit, a simulation unit, a tensor construction unit, a dual-channel detection unit, and an assessment output unit;

[0043] The data acquisition unit is configured to: obtain a sequence of teaching statements and a student profile matrix of the target class based on the lesson plan text and the target class roster;

[0044] The simulation unit is configured to obtain an interaction log by simulating a preset classroom process based on the sequence of teaching statements.

[0045] The tensor construction unit is configured to: obtain a latent state temporal tensor based on the interaction log and the student profile matrix;

[0046] The dual-channel detection unit is configured to: obtain a first risk time step set and a second risk time step set by using a preset parallel dual-channel detection based on the latent state temporal tensor.

[0047] The evaluation output unit is configured to: obtain the total number of risk triggers based on the first risk time step set and the second risk time step set, and obtain the evaluation result of the lesson plan text based on the total number of risk triggers.

[0048] Thirdly, this application provides a computer device, including: a processor, and a memory communicatively connected to the processor;

[0049] The memory stores the instructions that the computer executes;

[0050] The processor executes computer execution instructions stored in memory to implement the method of this application.

[0051] Fourthly, this application provides a computer-readable storage medium storing program code, which, when executed by a processor, is used to implement the method of this application.

[0052] In summary, the beneficial effects that this application can achieve are:

[0053] The dynamic lesson plan evaluation model proposed in this application solves the problems of the one-sidedness and lag of traditional static text evaluation by constructing a classroom simulation engine and simulating the teaching sequence and the evolution of students' status. It realizes the upgrade of lesson plan evaluation from static analysis to dynamic prediction, and effectively improves the accuracy and foresight of lesson plan evaluation.

[0054] The proposed quantifiable pre-assessment system for lesson plans constructs multi-dimensional risk quantification indicators such as cognitive load, comprehension, and attention outliers, and outputs assessment results with a 100-point quality score. This solves the problems of subjective ambiguity and lack of clear optimization basis in traditional lesson plan assessment, and enables precise empowerment for lesson plan optimization, significantly reducing the trial and error costs in actual classrooms.

[0055] The proposed adaptive and differentiated lesson plan evaluation mechanism solves the problems of traditional one-size-fits-all evaluation and inability to adapt to different learning situations by introducing multi-dimensional student profile parameters and classifying different student groups. It realizes differentiated evaluation for students with different cognitive levels, can accurately identify lesson plan adaptation risks, and support personalized and customized lesson plan optimization. Attached Figure Description

[0056] The accompanying drawings, which are included to provide a further understanding of the embodiments of this application and form part of this application, do not constitute a limitation on the embodiments of this application. In the drawings:

[0057] Figure 1 This is a schematic diagram of the method steps in an embodiment of this application;

[0058] Figure 2 This is a schematic diagram of the method steps in step S1 of an embodiment of this application;

[0059] Figure 3 This is a schematic diagram of the method steps in step S2 of an embodiment of this application;

[0060] Figure 4 This is a schematic diagram of the method steps in step S3 of an embodiment of this application;

[0061] Figure 5 This is a schematic diagram of the method steps in step S5 of an embodiment of this application;

[0062] Figure 6 This is a schematic diagram of the system structure according to an embodiment of this application;

[0063] Figure 7 This is a schematic diagram of the hardware structure of an embodiment of this application.

[0064] The diagram is labeled as follows: 610 - Data acquisition unit, 620 - Simulation unit, 630 - Tensor construction unit, 640 - Dual-channel detection unit, 650 - Evaluation output unit, 701 - Processor, 702 - Memory, 703 - Input / output interface, 704 - Communication interface, 705 - Bus. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0066] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0067] Example 1

[0068] Please refer to the following: Figure 1The above is a schematic diagram of the steps of the method for assessing the state of a student group based on a lesson plan text provided in this application embodiment. Further, the method for assessing the state of a student group based on a lesson plan text may specifically include the content described in steps S1-S5.

[0069] Step S1: Based on the lesson plan text and the target class roster, obtain the teaching statement sequence and the student profile matrix of the target class;

[0070] Step S2: Based on the sequence of teaching statements, obtain the interaction log through a pre-set classroom process simulation;

[0071] Step S3: Obtain the latent state time series tensor based on the interaction log and student profile matrix; the latent state time series tensor is configured to include at least the student, time step, and learning state.

[0072] Step S4: Based on the hidden state temporal tensor, obtain the first risk time step set and the second risk time step set through preset parallel dual-channel detection;

[0073] Step S5: Based on the first risk time step set and the second risk time step set, obtain the total number of risk triggers to obtain a positive lesson plan quality assessment value.

[0074] In implementing this embodiment, the lesson plan text to be evaluated and the target class roster are first obtained. The lesson plan text can be in the form of a plain text file, PDF document, or Word document; the target class roster can be manually entered by the teacher or imported from the school's student information management system. Subsequently, the obtained lesson plan text is processed, for example, by segmenting sentences using punctuation marks or line breaks to obtain the sequence of teaching sentences. At the same time, based on the target class roster, basic information of each student can be directly extracted from a pre-established student information database, such as age, gender, historical grades, and learning information. This information, after being encoded or statistically processed, is used to construct a student profile matrix to represent the students' basic learning characteristics.

[0075] After obtaining the sequence of instructional statements, a classroom process simulation is performed based on time steps. This simulation can be configured to proceed sequentially at fixed time intervals according to the order of the instructional statements. At each time step, the current instructional statement is recorded, and the student's response to that statement is simulated, such as understanding or not understanding. These simulated sequences of instructional statements and student responses, along with the corresponding time information, are recorded to form an interaction log. The interaction log can be viewed as a chronologically ordered list of events, recording all activities that occurred during the simulated classroom.

[0076] Based on the interaction logs and student profile matrix obtained above, a latent state time series tensor can be further obtained. Specifically, for each time step in the interaction log, combined with the student profile matrix, a preliminary estimate of each student's learning state can be made. For example, based on the learning ability parameters in the student profile matrix and the length of the current teaching statement, the student's focus level can be inferred; based on the knowledge background and the complexity of the teaching statement in the student profile matrix, the student's cognitive load can be inferred; and based on the above inferences, combined with the student's response in the simulation, a preliminary judgment of the student's comprehension level can be formed. These estimates are organized into a three-dimensional data structure, including three dimensions: student, time step, and learning state, thus forming a latent state time series tensor.

[0077] After obtaining the latent state time series tensor, a pre-defined parallel dual-channel detection mechanism is used to identify potential teaching risks. In this embodiment, the dual-channel detection includes inter-group gap detection and implicit-explicit behavior mapping detection. For implicit-explicit behavior mapping detection, some rules can be preset. For example, when a teaching statement requires a student to perform a certain behavior (such as thinking), if the student's implicit state shows low concentration, then that time step is marked as the first risk time step. For inter-group gap detection, the student group can be divided into several groups, for example, based on their historical scores or learning ability parameters in their profile vectors, and then the average learning state differences of these groups at a specific time step are compared. If the difference exceeds a certain preset standard, then that time step is marked as the second risk time step. Through these two detections, the first risk time step set and the second risk time step set are obtained respectively.

[0078] Finally, based on the obtained first and second risk time step sets, the total number of risk triggers is counted. This total number of triggers is obtained by summing the counts of all marked time steps in both sets. Based on this total number of triggers, a positive lesson plan quality assessment value can be calculated. For example, a preliminary assessment score can be obtained by subtracting the total number of risk triggers from the total number of time steps and then dividing by the total number of time steps. Simultaneously, since a correspondence between time steps and teaching statements was established during the simulation, the specific teaching statement corresponding to a given time point in the risk time step set can be traced back to that point. These teaching statements corresponding to the risk time steps, along with the types of risks they represent, are compiled into a diagnostic report, providing direct reference information for lesson plan optimization.

[0079] This method achieves dynamic simulation of the teaching process by processing lesson plan texts and combining them with a student profile matrix, overcoming the limitations of traditional static assessment. By constructing a latent state time series tensor, it can capture changes in students' internal learning states at different time steps. The introduction of a parallel dual-channel detection mechanism enables the identification of teaching risks from two dimensions: group differences and behavioral anomalies. Finally, by quantifying risks and generating diagnostic reports, this method provides an objective quality assessment of lesson plans and can accurately locate the teaching stages where risks occur, providing a scientific basis for lesson plan optimization.

[0080] Example 2

[0081] Based on Example 1, please refer to Figure 2 - Further implementations of the method for assessing the state of student groups based on lesson plan texts provided in the embodiments of this application include the following.

[0082] In this embodiment, a specific seventh-grade mathematics lesson plan on "Transforming Terms in a Linear Equation in One Variable" is used as an example to demonstrate the complete calculation process of the method proposed in this application. The lesson plan includes 5 time steps, each lasting 5 minutes, for a total duration of 25 minutes.

[0083] In one possible implementation, step S1 involves obtaining the sequence of teaching statements and the student profile matrix of the target class based on the lesson plan text and the target class roster. (See [link to relevant documentation]). Figure 2 Here is a schematic diagram of the method steps for step S1, including:

[0084] Step S101: Based on the lesson plan text, obtain a structured lesson plan element set through preset semantic segmentation rules; the structured lesson plan element set is configured to include at least: a sequence of teaching objectives, a sequence of knowledge points, a sequence of activity segments, a sequence of time allocation, and a sequence of segment connection relationships;

[0085] Step S102: Based on the structured lesson plan element set, obtain the teaching statement sequence with planned start time and duration by dividing the time window and annotating the instructions;

[0086] Step S103: Based on the target class roster, retrieve the prior knowledge level parameters, instruction completion parameters, and attention duration parameters for each student through the learning database or questionnaire input.

[0087] Step S104: Based on each student's prior knowledge level parameters, instruction completion parameters, and attention duration parameters, obtain individual profile vectors for all students through feature vectorization. These vectors are used to represent students' cognitive baselines and behavioral tendencies, thereby obtaining the student profile matrix for the target class.

[0088] In one possible implementation, step S102, each teaching statement in the teaching statement sequence is configured with instruction labels, including: a first label indicating whether to provide opportunities for all students to participate, and a second label indicating whether to require all students to raise their hands in response;

[0089] The first marker determination rule is configured as follows: if a word in the teaching statement is in the preset first marker vocabulary set, the first marker is set to 1; otherwise, it is set to 0. The first marker vocabulary set configuration includes at least: all students in the class, all students, or everyone.

[0090] The second marker determination rule is configured as follows: if a word in the preset second marker vocabulary set exists in the teaching statement, the second marker is set to 1; otherwise, it is set to 0. The second marker vocabulary set configuration includes at least the following expressions that explicitly require raising hands: please raise your hand, raise your hand to indicate, or raise your hand to answer.

[0091] In the implementation of this application embodiment, step S1 first performs structured parsing of the lesson plan text and simultaneously obtains basic profile parameters of the target class students. The original lesson plan text is obtained and structuredly decomposed using natural language processing based on preset rules. The complete lesson plan text to be evaluated is received, and through semantic segmentation and element extraction, core teaching elements such as teaching objectives, knowledge point sequences, teaching activity segments, time allocation for each segment, and the connection relationships between segments are extracted. The parsed data forms a structured lesson plan element set, providing a benchmark for subsequent timeline mapping. Based on the target student's location, basic profile data of the target class students is collected. Parameters include at least: age, grade, prior knowledge level (pre-score), attention baseline (attention duration), and group heterogeneity indicators (such as personality type, instruction completion rate, etc.), representing the individual characteristics of virtual students with different cognitive baselines and behavioral tendencies, serving as the initial state input for simulation. Specifically:

[0092] First, the original lesson plan text is imported from the lesson plan management system. This original lesson plan text typically includes a description of teaching objectives, a list of knowledge points, a schedule of teaching activities, a time allocation explanation, and a blackboard design. Then, using an existing natural language processing platform and pre-defined semantic segmentation rules, the original lesson plan text undergoes semantic analysis and extraction. These semantic segmentation rules identify the boundaries of professional terms using a pre-trained domain dictionary and combine this with teaching activity type tags to extract information about the teaching process.

[0093] Specifically, the natural language processing platform first identifies the boundaries of technical terms in the lesson plan text. For example, time indicators such as "first," "next," and "finally" are used as markers for transitions between activities; numbers combined with time units such as "5 minutes" and "10 minutes" are used as duration markers; and verb phrases such as "explain," "ask questions," "practice," and "discuss" are used as activity type markers. By combining these rules with a machine learning model, the natural language processing platform outputs a structured lesson plan element set that includes a sequence of teaching objectives, a sequence of knowledge points, a sequence of activity segments, a sequence of duration allocations, and a sequence of segment transitions.

[0094] After obtaining the structured lesson plan element set, the teaching statement segmentation and annotation operations are further performed. The lesson plan is segmented along the timeline according to the preset time window length, dividing the continuous teaching activities in the lesson plan text into several independent teaching statements. Each teaching statement includes the planned start time and duration. For example, a 25-minute math lesson plan may be segmented into 5 teaching statements, with planned start times of 0 minutes, 5 minutes, 10 minutes, 15 minutes, and 20 minutes for each statement, and a corresponding duration of 5 minutes for each statement.

[0095] After segmentation, each teaching statement is labeled with instructions. The label includes a first marker indicating whether the statement provides an opportunity for all students to participate, and a second marker indicating whether all students are explicitly required to raise their hands. The first marker is set to 1 if the teaching statement contains words such as "all students," "all students," or "everyone," otherwise it is set to 0. The second marker is set to 1 if the teaching statement contains explicit instructions such as "please raise your hand," "raise your hand to indicate," or "raise your hand to answer," otherwise it is set to 0.

[0096] Based on each teaching statement marked with a completion instruction, a sequence of teaching statements is obtained according to the time step.

[0097] While obtaining the sequence of teaching statements, based on the target class roster, parameters for each student in the target class are retrieved from the student learning database or questionnaire system, including: prior knowledge level parameters, instruction completion rate parameters, and attention span parameters.

[0098] The prior knowledge level parameter is represented by a value between 0 and 1, with a higher value indicating a better level of pre-study for the subject content;

[0099] The instruction completion parameter is represented by a value between 0 and 1. The higher the value, the better the student can complete the task according to the teacher's instructions.

[0100] Attention span parameter, measured in minutes, indicates the maximum duration a student can maintain focus.

[0101] In addition, a personality type description field for students was obtained, including three types: active questioner, silent thinker, and easily distracted. This field is used as auxiliary information for subsequent analysis but does not participate in numerical calculations.

[0102] The above parameters are converted into numerical vectors that can be used for calculation through feature vectorization, forming individual profile vectors that represent the student's cognitive baseline and behavioral tendencies. The combination of all students' individual profile vectors constitutes the student profile matrix of the target class.

[0103] In the implementation of this application embodiment, the sequence of teaching statements output in step S1 is shown in Table 1 below:

[0104]

[0105] The student profile matrix for the target class is shown in Table 2 below:

[0106]

[0107] In one possible implementation, step S2 involves obtaining an interaction log based on the sequence of teaching statements through a pre-set classroom process simulation. (See also...) Figure 3 Here is a schematic diagram of the method steps for step S2, including:

[0108] Step S201: Based on the teaching statement sequence, advance the simulation clock according to the preset scoring grid time interval to obtain the teaching statement flow executed in time step order;

[0109] Step S202: Based on the teaching statement flow, obtain the current teaching statement and cumulative practice duration corresponding to each time step to obtain the interaction log.

[0110] Step S203: In response to the simulation clock reaching the preset query trigger point, obtain an explanatory response by executing the interrupt-reply-recovery atomic operation chain to update the interaction log.

[0111] In one possible implementation, step S202 involves obtaining the interaction log based on the current teaching statement and cumulative practice time corresponding to each time step.

[0112] In one possible implementation, executing the interrupt-recovery-restore atomic operation chain does not change the time step and does not add additional scoring grid points. However, within the time step corresponding to the execution of the interrupt-recovery-restore atomic operation chain, the cumulative practice time of the current teaching statement at that time step is updated.

[0113] In one possible implementation, step S203, obtaining an interpretive response by executing an interrupt-response-recovery atomic operation chain to update the interaction log, is configured as follows:

[0114] In response to the simulation clock reaching the preset question trigger point, the timestamp of the question occurrence, the student ID of the questioner, and the question content are obtained; the current teaching statement is used as the currently interrupted teaching statement;

[0115] Based on the question and the currently interrupted teaching statement, an explanatory response text is obtained through a pre-set context-enhanced generative network.

[0116] Based on the number of characters in the explanatory response text, the duration of the explanatory response on the simulation timeline is obtained through a preset character-duration mapping rule;

[0117] Based on the duration, obtain the end timestamp after the explanatory response ends;

[0118] In response to the end of the explanatory reply, the execution of the currently interrupted instructional statements will be automatically resumed;

[0119] Based on the timestamp of the question being asked and the timestamp of the question ending, update the cumulative practice time of the current teaching statement corresponding to the currently interrupted teaching statement;

[0120] Based on the current teaching statement and cumulative practice time corresponding to each time step, as well as the student identifier, question content, currently interrupted teaching statement, and explanatory reply text within the time step corresponding to the execution of the interrupt-resume-recovery atomic operation chain, an interaction log including the entire event time is obtained.

[0121] In the implementation of this application embodiment, a dynamic simulation is constructed in step S2, mapping static lesson plan elements to a sequence of simulation events on a timeline to drive the virtual student group to perform learning activities. Based on the elements of the lesson plan text parsed in S1 and the class student profile matrix, the classroom simulation is initialized. The teaching statements are divided into nodes with planned start and end times, and the simulation clock is advanced at fixed time steps. When the teaching statements are executed sequentially on the timeline, the focus and participation status of each virtual student are calculated according to time slices, and after resampling and alignment, they are aggregated into a class group state sequence, and dual-channel risk detection is performed. During the simulation, based on the text features and cumulative duration of the teaching statements in the interaction log (including the density of professional vocabulary, sentence length, and number of abstract concepts), combined with the student profile matrix in S2, the focus anomaly value, cognitive load value, and comprehension value of each virtual student are calculated step by step to form the original state sequence. Based on the scoring grid interval, the original sequence is resampled to a fixed number of grid points using natural cubic spline interpolation, ensuring strict alignment of all students' timelines. After Savitzky-Golay smoothing filtering and interval truncation, a three-dimensional implicit state time series tensor of student × time step × learning state is constructed. The statement identifier and cumulative practice time are recorded. When a preset student question event is triggered, an "interrupt-response-resume" atomic operation chain is executed: interrupt the execution of the current statement → record the question event → call the context enhancement generation network to output an explanatory response → calculate the response time according to the "fixed time per word" rule → resume the execution of the original statement. The time spent on questioning and responding is included in the cumulative time but does not add new scoring grid points. The final output includes an interaction log with timestamps for all events. Specifically:

[0122] First, a fixed scoring grid time interval is set as the step size for advancing the simulation clock. This interval can be set according to the total duration of the lesson plan and the expected scoring accuracy. The smaller the interval, the finer the scoring granularity but the greater the computational load. It is usually set to one minute or two minutes. Then, the time interval between each pair of scoring grids is the scoring grid point, which is understood here as the moment on the time axis of the simulation time.

[0123] The simulation clock is started at time 0, and the teaching statement sequence is executed in the order of time steps to obtain the teaching statement flow that is implemented in the time interval of the scoring grid.

[0124] The teaching statements are executed sequentially at each time step of the teaching statement sequence. The simulation clock advances according to the scoring grid time interval. It checks whether the current simulation time has reached the planned start time of a certain teaching statement. If it has, the teaching statement is executed at that simulation time, and the corresponding time step also begins. The current teaching statement and the cumulative practice time from the start of the lesson plan to the current moment are recorded in real time to form the basic execution trajectory on the simulation timeline, which serves as the teaching statement flow.

[0125] When the simulation clock reaches the preset student question trigger point, the interrupt-reply-restore atomic operation chain is executed. The specific execution process is as follows:

[0126] First, interrupt the current teaching statement and save the execution status of the statement to the pause queue;

[0127] Then, record relevant information about the question event in the interaction log, including the timestamp of the question, the student ID of the student who asked the question, and the text description of the question content;

[0128] After recording is complete, a pre-set context-enhanced generative network is invoked to generate an explanatory response based on the obtained question content, the content of the currently interrupted teaching statement, and the context of the preset teaching knowledge base. This generative network typically uses a large language model based on the Transformer architecture, which has been fine-tuned and optimized for the education and teaching field, and can generate answers and explanations that conform to cognitive laws according to the teaching scenario.

[0129] The generated network output response text enters the word count stage. According to the preset rule that each character corresponds to a fixed duration, the duration of the response on the simulation timeline is calculated. Assuming that the preset character-duration mapping is 0.1 seconds per character, a 100-character response will occupy a duration of 10 seconds on the simulation timeline.

[0130] Then the duration is included in the cumulative practice duration, but no additional scoring grid points are added, maintaining the equal interval characteristic of the scoring grid;

[0131] After the response is generated, the response text content and the response end timestamp are recorded. Then, the interrupted teaching statements are automatically resumed, and the simulation continues to advance along the timeline.

[0132] The entire simulation process continues until the end time of the lesson plan, and the final output includes an interaction log containing timestamps of all events. This interaction log records in detail the start and end times of each teaching statement during the execution of the lesson plan, the occurrence time and content of student question events, and the occurrence time and content of response generation events, providing complete process data for subsequent tensor construction steps.

[0133] In the implementation of this application embodiment, the interaction log generated in the simulation step records the execution timestamp of each teaching statement. Assuming that there are student question events at time steps t3 and t5, the generation network generates 50-word and 80-word explanatory replies for each question, with corresponding time consumption of 5 seconds and 8 seconds, respectively.

[0134] In one possible implementation, step S3 involves obtaining the latent state time series tensor based on the interaction logs and the student profile matrix. The latent state time series tensor configuration includes at least the student, time step, and learning state. (See [link to relevant documentation]). Figure 4 Here is a schematic diagram of the method steps for step S3, including:

[0135] Step S301: Based on the current teaching statement corresponding to each time step of the interaction log and the student profile matrix, obtain the student's attention anomaly value, cognitive load value and comprehension value at each time step to obtain the original state sequence.

[0136] Step S302: Based on the original state sequence, obtain a resampled state sequence by resampling at scoring grid time intervals;

[0137] Step S303: Based on the resampled state sequence, obtain the implicit state temporal tensor with three-dimensional alignment of student × time step × learning state.

[0138] In this embodiment, step S304 is also included: obtaining a bidirectional backtracking mapping table between time steps and statements based on the resampled state sequence.

[0139] In one possible implementation, step S301 involves obtaining the attention anomaly value, cognitive load value, and comprehension value for each time step based on the current teaching statement corresponding to each time step in the interaction log and the student profile matrix, including:

[0140] At any given time step,

[0141] Step S3011: Based on the planned start time and the current time step, obtain the attention anomaly value corresponding to the student at the current time step;

[0142] Step S3012: Obtain the content difficulty value based on the current teaching statement;

[0143] Step S3013: Obtain the cognitive load value based on the content difficulty value and the prior knowledge level of the student profile matrix;

[0144] Step S3014: Based on the cognitive load value and the student's instruction completion rate in the student profile matrix, obtain the initial comprehension level.

[0145] Step S3015: Based on the initial comprehension level, a lower limit protection is applied to the initial comprehension level using a preset lower limit value to obtain the comprehension level value.

[0146] The original state sequence is obtained based on the attention anomaly value, cognitive load value, and comprehension value at each time step.

[0147] In one possible implementation, step S302, obtaining a resampled state sequence based on the original state sequence by resampling at scoring grid time intervals, includes:

[0148] Step S3021: Using the scoring grid time interval as the benchmark and the cumulative practice time as the horizontal axis, perform natural cubic spline interpolation on the original state sequence and resample it to a fixed number of scoring grid points so that the time axes of all students are strictly aligned.

[0149] Step S3022: Determine whether the number of grid points after resampling is greater than or equal to the preset filter window length; if so, perform Savitzky-Golay filtering on each state component; otherwise, retain the interpolation result.

[0150] Step S3023: Truncate the filtering or interpolation result to a preset standard interval to obtain the hidden state temporal tensor;

[0151] Step S3024: Save the original time step and the current teaching statement corresponding to each resampled scoring grid point to obtain a bidirectional backtracking mapping table between time steps and teaching statements.

[0152] In the implementation of this application embodiment, step S3 calculates the focus and participation status of each virtual student by time slice, and after resampling and alignment, aggregates them into a class group state sequence, and performs dual-channel risk detection. During the simulation, based on the textual features of teaching statements in the interaction log (density of professional vocabulary, sentence length, number of abstract concepts) and cumulative duration, combined with the student profile in S2, the focus anomaly value, cognitive load value, and comprehension value of each virtual student are calculated step by step to form the original state sequence. Using the scoring grid interval as a benchmark, the original sequence is resampled to a fixed number of grid points through natural cubic spline interpolation to ensure strict alignment of the time axes of all students; after Savitzky-Golay smoothing filtering and interval truncation, a three-dimensional implicit state time series tensor of student × time step × learning state is constructed. Specifically:

[0153] Tensor construction includes four sub-steps: state parameter calculation, resampling alignment, smoothing filtering, and index saving.

[0154] In the state parameter calculation sub-step S301, tensor construction calculates three core state parameters for each student at each time step based on the text features and cumulative duration of the teaching statements corresponding to the current time step in the interaction log, combined with the student profile: attention anomaly value, cognitive load value, and comprehension value.

[0155] Calculating cognitive load first requires determining the difficulty level of the current instructional statement; specifically, this involves extracting the number of specialized vocabulary words from the instructional statement. Statement length and abstract concept numbers The three text features are mapped to the interval between 0 and 1 using a fixed upper limit normalization method.

[0156] At this point, the density of professional vocabulary in the lesson plan can be directly obtained, which can be expressed as:

[0157] .

[0158] Then, a truncation function is used to ensure that the normalization result does not exceed 1, using the symbol... This represents the truncation function, which can be expressed as:

[0159] .

[0160] Similarly, statement length and the number of abstract concepts are also mapped according to their respective normalization rules, and can be represented as:

[0161] ;

[0162] ;

[0163] ;

[0164] After obtaining the three normalized eigenvalues, the difficulty value of the content can be calculated by arithmetic average, and can be expressed as:

[0165] ;

[0166] That is, to obtain the content difficulty value Then, the cognitive load value is calculated based on the difference between this value and the student's prior knowledge level parameter, which can be expressed as:

[0167] ;

[0168] in, Parameters of students' prior knowledge level 1.5 is the load amplification factor, used to amplify the gap between the difficulty of the content and the students' basic knowledge, and is a cutoff function to ensure that the result is within the standard range of 0 to 1.

[0169] For example, for a student with a prior knowledge level of 0.8, when the difficulty of the teaching content is 0.9, the calculated cognitive load is 0.1, indicating that the student can adapt to the difficulty of the teaching. However, for a student with a prior knowledge level of 0.38, the same difficulty of the teaching content will produce a cognitive load value of 0.52, indicating that the student faces greater cognitive pressure.

[0170] The comprehension score is calculated based on the cognitive load score. First, the ratio of the cognitive load score to the student instruction completion rate parameter is calculated as the initial comprehension score. Then, a preset lower limit for comprehension is used to protect this initial comprehension score. This lower limit is typically set to 0.3 and can be expressed as:

[0171] ;

[0172] in, Instruction completion parameter for individual students .

[0173] The protection mechanism ensures that even under extremely high cognitive loads, the level of understanding will not drop too low to cause anomalies in subsequent calculations.

[0174] The calculation of attention outliers uses an exponential decay model to reflect the gradual dissipation of students' attention over time. The cumulative number of minutes from the start of the lesson plan to the current time step, representing the positive attention level at this point, can be expressed as:

[0175] ;

[0176] in, For individual students' attention span parameters , This represents the cumulative number of minutes.

[0177] Therefore, focusing on outliers can be expressed as: ;

[0178] When the cumulative duration reaches approximately 4.6 times the duration of attention, the attention anomaly value will approach 1, meaning that attention is almost completely scattered.

[0179] In the resampling alignment sub-step S302, tensor construction uses the scoring grid time interval as a baseline and the cumulative practice time as the horizontal axis to perform natural cubic spline interpolation on the original state sequence, resampling the unequally spaced original event nodes to a fixed number of scoring grid points. The main purpose is to ensure strict alignment of all students in the time dimension for subsequent group aggregation analysis. The natural cubic spline interpolation method can generate a smooth resampling curve while maintaining the trend of the original data, making it particularly suitable for handling continuous changes in state parameters during the teaching process.

[0180] After resampling, each student will have the same number of state data points, forming a preliminary aligned original state sequence.

[0181] In the smoothing filtering sub-step S303, the tensor construction determines whether the number of grid points after resampling is greater than or equal to the preset filtering window length. If the condition is met, a Savitzky-Golay filtering operation is performed on each state component. This filtering method uses local polynomial regression to smooth the signal, which can remove noise while preserving the peak characteristics of the signal. The selection of the filtering window length needs to balance the noise suppression effect with the degree of signal detail preservation. In the embodiments of this application, a window length of 5 and a second-order polynomial parameter configuration are typically used.

[0182] If the number of grid points after resampling is less than the length of the filtering window, the interpolation result is retained and no further filtering is performed.

[0183] Regardless of whether filtering is performed, the final state data needs to be truncated to a standard interval to ensure that all state values ​​fall within the range of 0 to 1. After the above processing is completed, the output is a three-dimensional aligned implicit state temporal tensor of student × time step × learning state.

[0184] In the index saving sub-step S304, tensor construction establishes a mapping relationship between each time step and its corresponding original teaching statement for each resampled scoring grid point, serving as a bidirectional backtracking mapping table between time steps and statements. This bidirectional backtracking mapping table is stored in the form of a bidirectional index table, supporting bidirectional backtracking functions of querying the corresponding statement from a time step and querying the corresponding time step from a statement. This index structure provides a data foundation for locating risk statements in the subsequent evaluation output step.

[0185] In the implementation of this application embodiment, each state parameter is calculated based on the above input data.

[0186] According to the formula for cognitive load Comprehension score Each step takes 5 minutes, for a total of [time] minutes. Focus on outliers The calculation results are shown in the table below:

[0187]

[0188]

[0189]

[0190] Taking the calculation process of student S5 at time step t3 as an example, the cognitive load value is first calculated as the difference between the content difficulty and the prior knowledge level, 0.90-0.38=0.52. Multiplying it by the load amplification factor of 1.5, we get 0.78. After truncation to the standard interval, it is 0.78.

[0191] Comprehension score is calculated using student S4's calculation process at time step t3 as an example, representing cognitive load. The ratio is 0.63, the instruction completion rate is 0.65, and the original ratio is... The initial understanding level is approximately 0.0308, which is then set to 0.3 after lower bound protection.

[0192] Focusing on outliers, taking the calculation process of student S1 as an example, the calculation at time step t3 (cumulative duration 15 minutes) is 1-exp(-15 / 45)=0.2835.

[0193] After resampling alignment and smoothing filtering, a 5×5×3 hidden state time series tensor is generated, where 5 represents the number of students, 5 represents the number of time steps, and 3 represents the number of state dimensions. However, this embodiment only has five time steps, so Savitzky-Golay filtering is not performed, and the above values ​​are used directly. Dimensions .

[0194] In one possible implementation, step S4, based on the latent state temporal tensor, obtains a first risk time step set and a second risk time step set through a pre-defined parallel dual-channel detection, including:

[0195] The dual-channel detection is configured for population inter-layer gap detection;

[0196] The inter-layer gap detection for the group is configured as follows:

[0197] Step S411: Based on the latent state time series tensor, obtain the population average state curve by the arithmetic mean of all students in each state dimension at each time step.

[0198] Step S412: Calculate the absolute second difference based on the population average state curve and mark the corresponding time steps that exceed the first preset threshold to obtain the abnormal time step set.

[0199] Step S413: Based on the abnormal time step set, obtain the average comprehension level of each student in the set; based on the average comprehension level of each student, sort them in ascending order, and divide the students in the first preset proportion of the top-ranked students into the low comprehension group, and the students in the second preset proportion of the bottom-ranked students into the high comprehension group.

[0200] Step S414: Based on the high comprehension group and the low comprehension group, obtain the absolute difference between the mean comprehension values ​​of the two groups at all time steps; based on the absolute difference at each time step, mark the time steps that exceed the second preset threshold to obtain the second risk time step set.

[0201] In one possible implementation, in step S413, the inter-strata gap detection, if the abnormal time step set is empty or the total number of valid students is less than 2, then student stratification is not performed, and the second risk time step set is set to empty.

[0202] In one possible implementation, step S4, based on the latent state temporal tensor, obtains a first risk time step set and a second risk time step set through a pre-defined parallel dual-channel detection, including:

[0203] Dual-channel detection is configured with implicit-explicit behavior mapping detection;

[0204] Implicit-explicit behavior mapping detection is configured as follows:

[0205] Step S421: Obtain the behavior image sequence that is synchronously acquired with the latent state temporal tensor, process the arm key points through graph convolutional network to obtain the hand-raising related features, obtain the frowning related features through face feature extractor and expression classification head, and obtain the first change temporal tensor.

[0206] Step S422: Based on the latent state temporal tensor, a state-behavior association network is used to obtain a second changing temporal tensor through three-dimensional convolution and reconstruction; and the alignment loss is calculated using the first changing temporal tensor as a supervision signal to train the state-behavior association network.

[0207] Step S423: Based on the implicit state temporal tensor corresponding to the lesson plan to be evaluated, obtain the behavior change temporal tensor based on the state-behavior association network; based on the behavior change temporal tensor, obtain the hand state and facial expression state of each student at each time step through the second behavior detection network.

[0208] Step S424: Based on the first and / or second markers of each teaching statement in the teaching statement sequence, obtain the label of whether all students raised their hands at each time step; if the label of all students raising their hands appears at the current time step, but no hand-raising behavior is detected, or frowning behavior is detected, then the student's current behavior is determined to be abnormal and is regarded as an abnormal student at the current time step.

[0209] Step S425: Obtain the number of abnormal students at each time step. If the number of abnormal students exceeds the preset value of the total number of students at that time step, then mark that time step as the first risk time step.

[0210] In the implementation of this application embodiment, in step S4, based on this tensor, "group inter-level gap detection" (calculating the group mean, absolute second-order difference, student stratification and inter-level comprehension difference) and "implicit-explicit behavior mapping detection" (outputting the probability of raising a hand / frowning based on the state-behavior association network, and determining behavioral abnormalities by combining structured instructions) are executed respectively, and finally the first risk time step set (behavioral abnormality cluster) and the second risk time step set (group cognitive discontinuity) are output. Specifically:

[0211] The dual-channel detection is configured to include: inter-group gap detection and implicit-explicit behavior mapping detection.

[0212] The implicit-explicit behavior mapping detection channel in dual-channel detection includes two stages: pre-training and inference. The pre-training stage uses a sequence of behavior images acquired synchronously with the temporal tensor of the implicit state as training data.

[0213] The behavioral image sequence was obtained by cropping classroom videos according to the scoring grid intervals, with at least one image corresponding to a student's behavior within a time step.

[0214] A graph convolutional network is used to process the coordinate sequence of key points on the student's arm in the image and extract feature vectors related to the hand-raising behavior. At the same time, a face feature extractor is used to extract the depth feature representation of the student's face in the image and the presence of frowning features is determined by an expression classification head. The above two features are fused to form the first change temporal tensor, which records the trajectory of the student's explicit behavior changes at each time step.

[0215] The latent state temporal tensor is input into a state-behavior association network, which employs a structural design combining 3D convolutional layers and reconstruction layers. The 3D convolutional layers capture the continuous change patterns of the tensor in the temporal dimension, while the reconstruction layers map the convolutional features to the behavior space. The network is trained using supervised learning, with the first changing temporal tensor used as the supervision signal to calculate the alignment loss. Backpropagation is used to optimize the network parameters, enabling the network to learn the mapping rules from the internal learning state to the explicit behavior.

[0216] During the inference phase, the latent state temporal tensor corresponding to the lesson plan to be evaluated is input into a pre-trained state-behavior association network to obtain the predicted behavioral change temporal tensor. Subsequently, this tensor is input into a second behavior detection network that is trained independently. This network outputs the probability of each student raising their hand and frowning at each time step. These probabilities are binarized with a threshold of 0.5 to obtain a binarized representation of the hand state and facial expression state.

[0217] After obtaining the explicit behavior prediction results, abnormal behavior is determined according to the structured instructions marked in the data acquisition steps. The specific determination rules are as follows: if the current time step explicitly requires all students (i.e., marked as 1 in the first flag) to raise their hands (i.e., marked as 1 in the second flag), but a student is not detected raising their hand, that student is considered to have abnormal behavior; if a student is detected frowning, that student is directly considered to have abnormal behavior. The number of abnormal students at each time step is counted. When this number exceeds one-fifth of the total number of valid students at that time step, that time step is marked as the first risk time step. The set of first risk time steps is denoted as […]. .

[0218] In the dual-channel detection method, the inter-level gap detection channel first calculates the arithmetic mean of all students across all state dimensions at each time step, forming the group average state curve. The group's arithmetic mean represents the average comprehension level. Average focus outliers Average cognitive load , can be represented as:

[0219] ;

[0220] This represents the total number of valid students. , , , Let be the state parameters of the i-th student at time step t. The group average state curve reflects the overall state evolution trend of the class during the learning process.

[0221] After obtaining the population average state curve, the absolute second difference of the curve is calculated to detect abrupt state changes. The second difference reflects the concavity / convexity of the function curve and can effectively identify inflection points on the state curve, which can be expressed as:

[0222] ;

[0223] ;

[0224] ;

[0225] in, , , Let t be the group average of the state parameters of all students at time step t.

[0226] The absolute second difference is compared with a preset first threshold. When the absolute second difference exceeds this threshold, the corresponding time step is marked as an abnormal time step. The setting of the first threshold needs to be adjusted according to the characteristics of the teaching scenario, and is usually set to a value between 0.3 and 0.5. For example, if the thresholds are set to 0.3, 0.4, and 0.5 respectively, the set of abnormal time steps can be represented as follows: .

[0227] After obtaining the set of anomalous time steps, students in the class are stratified. Specifically, the average comprehension level of each student within the set of anomalous time steps can be calculated as follows:

[0228] ;

[0229] Then based on that average level of understanding Sort in ascending order.

[0230] After sorting, the top-ranked students (within a predetermined percentage) are grouped into the low-comprehension group, and the bottom-ranked students (within a predetermined percentage) are grouped into the high-comprehension group. The first predetermined percentage is typically set between 20% and 30%, indicating the identification of students with weak and strong comprehension abilities. If set to 20%, the top... The name is assigned to the low-comprehension group. The same number were then divided into high-comprehension groups. Then, after completing the student stratification, the absolute difference between the mean comprehension scores of the high-comprehension group and the low-comprehension group is calculated over all time steps, which can be expressed as:

[0231] ;

[0232] Here, i represents any student, and the absolute difference in the mean comprehension reflects the gap in comprehension between student groups with different ability levels, that is, the inter-level gap.

[0233] The difference is compared with a second preset threshold. When the difference exceeds the second preset threshold, the corresponding time step is marked as the second risk time step. The second preset threshold is usually set to a value between 0.1 and 0.2. If it is set to 0.1, the set of second risk time steps is:

[0234] If the abnormal time step set is empty or the total number of valid students is less than 2, then the student stratification operation will not be performed, and the second risk time step set will be set to empty. .

[0235] In the implementation of this application's embodiments, the dual-channel detection step performs group inter-layer gap detection and calculates the average understanding at each time step. Average focus outliers Average cognitive load The absolute second difference is calculated, and the results are shown in the table below:

[0236]

[0237] The population average understanding at each time step is calculated to obtain the sequence {1.000, 0.989, 0.471, 1.000, 0.701}. The absolute second difference is then used to identify the set of anomalous time steps. .

[0238] Average comprehension level of each student at abnormal time steps:

[0239] Students were ranked according to their average comprehension level at abnormal time steps, with S5 being the lowest and S4, S3, S2, and S1 being the highest.

[0240] The students were divided into a low-comprehension group and a high-comprehension group, with the top 20% (1 person) assigned to the low-comprehension group. The bottom 20% (1 person) were in the high-level group. ,

[0241] Then calculate the interlayer gap, based on the interlayer gap of all time steps. The formula and calculation results are shown in the table below:

[0242]

[0243] Identify the second risk time step set .

[0244] The dual-channel detection step simultaneously performs implicit and explicit behavior mapping detection, recording the number of aberration students at each time step as... Assume the second row is the output of the detection network as shown in the table below:

[0245]

[0246] Abnormal judgment rules: Frowning is abnormal; failing to raise one's hand when asked is abnormal.

[0247] Suppose the output of the second-line detection network shows that at time step t3, S4 and S5 frowned and did not raise their hands, the number of abnormal students is 3, which is more than one-fifth (1 person) of the total number of valid students (5). Therefore, t3 is marked as the first risk time step. At time step t5, S5 frowns and S4 and S5 do not raise their hands, indicating that there are 2 abnormal students, which also exceeds the threshold. Therefore, t5 is also marked as the first risk time step. At this time, the set of first risk time steps is... .

[0248] In one possible implementation, step S5 involves obtaining the total number of risk triggers based on the first and second risk time step sets to obtain a positive lesson plan quality assessment value. (See also...) Figure 5 This is a schematic diagram of the method steps in step S5, including:

[0249] Step S501: Obtain the total number of risk triggers based on the first risk time step set and the second risk time step set; if the same time step exists in both sets, the number of risk triggers for that time step is counted twice.

[0250] Step S502: Obtain the first valid evaluation opportunity number and the second valid evaluation opportunity number;

[0251] Step S503: Calculate the ratio based on the total number of risk triggers and the sum of the first and second effective evaluation opportunities, and convert it to a percentage to obtain the positive lesson plan quality assessment value.

[0252] Step S504: Based on the bidirectional backtracking mapping table between time steps and statements, find the original teaching statements corresponding to the first and second risk time steps, and obtain a diagnostic report including the risk time step, risk type and corresponding lesson plan segment.

[0253] In one possible implementation, step S502, obtaining the first valid evaluation opportunity count and the second valid evaluation opportunity count, includes:

[0254] Step S5021: Based on the inference results of the second behavior detection network, traverse all time steps and count the time steps in which reliable hand state and facial expression state output can be successfully obtained and the number of effective students in that time step is greater than zero as the first effective evaluation opportunity.

[0255] Step S5022: Based on the judgment results of whether the abnormal time step set is non-empty and whether the total number of valid students is greater than or equal to two in the inter-level gap detection of the group;

[0256] Step S5023: If the abnormal time step set is not empty and the total number of valid students is greater than or equal to two, then the stratification is determined to be successful, and all time steps are counted as the second valid evaluation opportunity.

[0257] Step S5024: If stratification fails, the number of second valid evaluation opportunities is set to zero.

[0258] The number of first and second valid evaluation opportunities are counted independently and do not affect each other.

[0259] The first effective evaluation opportunity number reflects the total number of available time steps in the behavior detection channel, and the second effective evaluation opportunity number reflects the total number of effective evaluation time steps after the group stratification channel is successfully established; both opportunity numbers are positive integers or zero, and do not include the process of calculating the positive lesson plan quality assessment value.

[0260] In the implementation of this application, a quantitative lesson plan score and a visual diagnostic report are generated based on the mapping between the group state sequence, the risk time step set, and the achievement of teaching objectives. According to the first and second risk time step sets output by S4, the total number of risk triggers is counted, with dual-channel hits at the same time step counted as two. Combining the number of effective evaluation opportunities obtained from the number of time steps with available behavioral outputs and the number of time steps with successfully stratified inter-stratification gaps, the risk percentage is calculated and converted into a percentage-based positive lesson plan quality assessment value. Simultaneously, using the bidirectional backtracking mapping table between time steps and original teaching statements established by S3, the risk time step is precisely located to a specific teaching statement fragment, outputting a visual diagnostic report including the risk time step coordinates, risk type (cognitive gap / behavioral abnormality), corresponding original lesson plan text, and teaching optimization suggestions. Specifically:

[0261] The evaluation output step S5 first calculates the total number of risk triggers. This number is equal to the sum of the potentials of the first risk time step set and the second risk time step set, denoted as R. When the same time step exists in both sets, the risk trigger count for that time step is counted twice.

[0262] Simultaneously, determine the first and second effective evaluation opportunities. The first effective evaluation opportunity is the total number of time steps in which behavioral output can be obtained, denoted as... The second effective evaluation opportunity number is the total number of time steps in which group stratification is successfully performed and the inter-stratum gap can be calculated, denoted as . .

[0263] After obtaining the above values, the positive quality score can be calculated using the following formula, which can be expressed as:

[0264] ;

[0265] Where Q represents the percentage score, and if the sum of the first and second valid evaluation opportunities is 0, then... If no valid evaluation opportunity is found, the system will return a message stating "No valid evaluation opportunity".

[0266] Since the total number of risk triggers R is an absolute value, it is affected by the total time steps in the lesson plan and cannot be directly used for cross-lesson plan comparisons. Therefore, this invention first counts the total number of time steps that can generate effective evaluations. As a normalized denominator, the risk density within a unit evaluation opportunity is calculated; then, this risk density is complemented by a conversion, i.e., 1 − risk density, to obtain the risk-free probability; finally, the risk-free probability is mapped to a percentage range. Through this progressive transformation, the original 'negative risk count' is successfully converted into a 'positive lesson plan quality assessment value,' allowing lesson plans of different durations to be compared on a uniform scale.

[0267] The "positive lesson plan evaluation value" means that its value is positively correlated with the quality of the lesson plan—the higher the value, the higher the quality of the lesson plan, the lower the risk, and the better the suitability for different student groups.

[0268] The word "positive" here refers to a semantic direction, used to distinguish it from all other negative indicators in this scheme:

[0269] Cognitive load is a negative indicator; the higher the value, the greater the cognitive stress on the student.

[0270] Anomalies in focus are negative indicators; the higher the value, the more severe the inattention.

[0271] The very existence of a risk time step indicates a problem.

[0272] The total number of risk triggers is a negative indicator; the higher the value, the more problems there are.

[0273] From cognitive load and the calculation of outliers to the capture of outlier time steps and the determination of risk time step sets, the entire solution revolves around "risk detection" and "outlier identification," and its essence is to collect negative evidence.

[0274] However, the final evaluation results delivered to teachers or curriculum researchers need to be intuitive, comparable, and conform to the cognitive habit of "the higher the score, the better." Therefore, the lesson plan quality evaluation value, as the only positive output in the entire document, maps all the previously accumulated negative evidence to the positive evaluation space through a defined mathematical transformation.

[0275] Finally, based on the bidirectional backtracking mapping table between time steps and teaching statements, the original teaching statements corresponding to the first and second risk time steps are found, and a diagnostic report is output, including the risk time step, risk type, and corresponding lesson plan segment. The diagnostic report is presented in a visual format, including components such as a risk timeline, a pie chart of risk type distribution, and a highlighted list of risk statements, helping teachers quickly locate problems in their lesson plans and make targeted optimizations.

[0276] In the implementation of this application's embodiments, the evaluation output step calculates the total number of risk triggers as the number of elements in the union of the first risk time step set and the second risk time step set, i.e., R = That is, 4 times. In this example, each time step has behavioral output and the layering is effective. Each channel has 5 time steps, therefore... The number of valid evaluation opportunities is 10. The quality score is... point.

[0277] The final diagnostic report showed that the lesson plan triggered risks 4 times, with a score of 60. The risk time steps were t3 and t5, corresponding to the lesson plan segments "Explaining the transposition rule and asking students to raise their hands" and "Setting up equations and asking students to raise their hands," respectively. The diagnosis means that time step t3 simultaneously exhibited abrupt changes in comprehension, cognitive load, group disparities, and clustering of abnormal behaviors, making it the highest-risk teaching segment in the lesson plan. Although time step t5 did not show a significant abrupt change in the group curve, the gap between high- and low-ability students persisted, and the number of students exhibiting abnormal behavior was relatively large, indicating that students with lower foundations faced continued learning difficulties in the comprehensive application stage. It is recommended that teachers increase individualized guidance for struggling students at this stage.

[0278] As can be seen from the above embodiments, the method of this application transforms the traditional static lesson plan text evaluation into a dynamic group learning effect deduction. By quantifying risk indicators, it identifies teaching links in the lesson plan that may have adverse effects on specific student groups, providing an objective and quantifiable decision-making basis for lesson plan optimization.

[0279] Example 3

[0280] This is the third embodiment of this application. Based on embodiments 1 and 2, please refer to the following references. Figure 2This embodiment provides a system based on a student group status assessment lesson plan text, including a sequentially electrically connected data acquisition unit 610, a simulation unit 620, a tensor construction unit 630, a dual-channel detection unit 640, and an assessment output unit 650.

[0281] The data acquisition unit 610 is configured to: obtain the teaching statement sequence and the student profile matrix of the target class based on the lesson plan text and the target class roster;

[0282] Simulation unit 620 is configured to obtain interaction logs by simulating a preset classroom process based on a sequence of teaching statements.

[0283] Tensor building unit 630 is configured to: obtain the latent state temporal tensor based on the interaction log and the student profile matrix;

[0284] The dual-channel detection unit 640 is configured to obtain a first risk time step set and a second risk time step set by using a preset parallel dual-channel detection based on the latent state temporal tensor.

[0285] The evaluation output unit 650 is configured to obtain the total number of risk triggers based on the first risk time step set and the second risk time step set, in order to obtain a positive lesson plan quality evaluation value.

[0286] When implemented, the system based on the teaching plan text for evaluating the state of a student group according to the embodiments of this application includes five core components: a data acquisition module, a simulation module, a tensor construction module, a dual-channel detection module, and an evaluation output module.

[0287] The data acquisition module is responsible for retrieving the lesson plan documents to be evaluated and the basic student information data of the target class from the teaching management system or text storage system. The simulation module dynamically simulates the classroom process based on the structured lesson plan data output by the data acquisition module, simulating the execution trajectory of teaching activities on the timeline. The tensor construction module deeply integrates the interaction logs generated during the simulation process with student profile data, calculating the learning state parameters of each student at each simulation time step, forming a three-dimensional aligned implicit state time series tensor. The dual-channel detection module executes two different risk identification strategies based on this tensor: inter-level gap detection and implicit / explicit behavior mapping detection. The evaluation output module calculates the lesson plan quality score by combining the dual-channel detection results and generates a visual diagnostic report.

[0288] Example 4

[0289] The fourth embodiment of this application differs from the previous embodiments in that:

[0290] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0291] In the several embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method for enhancing weak feedback data in education. This electronic device can be any smart terminal, including a tablet computer.

[0292] Please see Figure 7 , Figure 7 The hardware structure of an electronic device according to an embodiment of the present invention is illustrated. The electronic device includes:

[0293] The processor 701 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0294] The memory 702 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 702 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 702, and the processor 701 calls and executes the method for enhancing weak feedback data in education according to the embodiments of this application.

[0295] The input / output interface 703 is used to implement information input and output;

[0296] The communication interface 704 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0297] Bus 705 transmits information between various components of the device (e.g., processor 701, memory 702, input / output interface 703, and communication interface 704);

[0298] The processor 701, memory 702, input / output interface 703, and communication interface 704 are connected to each other within the device via bus 705.

[0299] The memory 704 stores a computer program. When the processor 702 executes the program, it implements the method described above for evaluating lesson plan texts based on student group status. The input / output interface is used to connect to external devices and receive lesson plan texts and student profile data. The communication interface 703 is used to interact with external terminals or storage systems and send evaluation results. The bus 700 is responsible for data and control signal transmission between components. The processor 702 reads instructions and data from the memory 704 through the bus 700, performs simulation deduction, tensor construction, and dual-channel risk detection, and outputs the generated diagnostic report through the output interface or communication interface 703. This embodiment uses a general-purpose computer device as a carrier to realize the hardware implementation of this method in a real teaching scenario.

[0300] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for enhancing weak educational feedback data.

[0301] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0302] The above-described specific embodiments further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method based on lesson plan texts for assessing the state of student groups, characterized in that, include: Based on the lesson plan text and the target class roster, obtain the teaching statement sequence and the student profile matrix of the target class; Based on the sequence of teaching statements, an interaction log is obtained through a pre-set classroom process simulation; Based on the interaction log and the student profile matrix, the latent state temporal tensor is obtained; Based on the latent state temporal tensor, a first risk time step set and a second risk time step set are obtained through preset parallel dual-channel detection. Based on the first risk time step set and the second risk time step set, the total number of risk triggers is obtained, and the evaluation result of the lesson plan text is obtained based on the total number of risk triggers.

2. The method for assessing lesson plan texts based on student group status according to claim 1, characterized in that, The process of obtaining the teaching statement sequence and the student profile matrix of the target class based on the lesson plan text and the target class roster includes: Based on the lesson plan text, a structured lesson plan element set is obtained through preset semantic segmentation rules; Based on the structured lesson plan element set, the sequence of teaching statements is obtained through time window segmentation and instruction annotation; Based on the target class roster, the prior knowledge level, instruction completion rate, and attention duration parameters of each student are obtained by retrieving information from the learning database. Based on the prior knowledge level parameters, instruction completion parameters, and attention duration parameters of each student, individual profile vectors of all students are obtained through feature vectorization to obtain the student profile matrix of the target class.

3. The method for assessing lesson plan texts based on student group status according to claim 2, characterized in that, Each teaching statement in the sequence of teaching statements is configured with instruction labels, the labels including: a first label indicating whether all students are given the opportunity to participate, and a second label indicating whether all students are required to raise their hands in response; The rule for determining the first marker is configured as follows: if the teaching statement contains a word from the preset vocabulary set of the first marker, the first marker is set to 1; otherwise, it is set to 0. The rule for determining the second marker is configured as follows: if the teaching statement contains a word from the preset second marker vocabulary set, the second marker is set to 1; otherwise, it is set to 0.

4. The method for assessing lesson plan texts based on student group status according to claim 1, characterized in that, The step of obtaining an interaction log based on the sequence of teaching statements through a preset classroom process simulation includes: Based on the teaching statement sequence, the simulation clock is advanced according to the preset scoring grid time interval to obtain the teaching statement flow executed in time step order; Based on the teaching statement flow, obtain the current teaching statement and cumulative practice time corresponding to each time step to obtain the interaction log; In response to the simulation clock reaching a preset query trigger point, an explanatory response is obtained by executing an interrupt-response-recovery atomic operation chain, and the explanatory response is used to update the interaction log.

5. The method for assessing lesson plan texts based on student group status according to claim 1, characterized in that, The implicit state time series tensor is obtained based on the interaction log and the student profile matrix. This implicit state time series tensor is configured to include at least the student, time step, and learning state, including: Based on the current teaching statement corresponding to each time step of the interaction log and the student profile matrix, the student's attention anomaly value, cognitive load value and comprehension value at each time step are obtained to obtain the original state sequence; Based on the original state sequence, a resampled state sequence is obtained by resampling at scoring grid time intervals; Based on the resampled state sequence, a hidden state temporal tensor with three-dimensional alignment of student × time step × learning state is obtained.

6. The method for assessing lesson plan texts based on student group status according to claim 1, characterized in that, The dual-channel detection is configured with implicit-explicit behavior mapping detection; The implicit-explicit behavior mapping detection is configured as follows: Acquire the behavioral image sequence that is synchronously acquired with the latent state temporal tensor, and obtain the first changing temporal tensor through a graph convolutional network; Based on the implicit state temporal tensor, a second change temporal tensor is obtained through a state-behavior association network; Based on the implicit state temporal tensor corresponding to the lesson plan to be evaluated, the behavior change temporal tensor is obtained based on the state-behavior association network. Based on the behavioral change temporal tensor, the hand state and facial expression state of each student at each time step are obtained through the second behavior detection network; Based on the first and / or second markers of each teaching statement in the teaching statement sequence, obtain the label of whether all students raised their hands at each time step; If the current time step shows a marker indicating that all students have raised their hands, but no hand-raising behavior is detected, or a frowning behavior is detected, then the student's current behavior is determined to be abnormal and they are considered an abnormal student at the current time step. Obtain the number of abnormal students at each time step. If the number of abnormal students exceeds a preset value of the total number of students at that time step, then mark that time step as the first risk time step.

7. The method for assessing lesson plan texts based on student group status according to claim 1, characterized in that, The dual-channel detection is configured with population inter-layer gap detection; The inter-layer gap detection of the execution group is configured as follows: Based on the latent state time series tensor, the population average state curve is obtained; Based on the population average state curve, the corresponding time steps that exceed the first preset threshold are marked to obtain an abnormal time step set. Based on the set of abnormal time steps, obtain the average comprehension level of each student within that set; Based on each student's average comprehension level, the students ranked in ascending order are divided into the low comprehension group (top 1% and bottom 1% respectively) and the high comprehension group (bottom 1% and bottom 1% respectively). Based on the high-comprehension group and the low-comprehension group, obtain the absolute difference between the mean comprehension values ​​of the two groups at all time steps; Based on the absolute difference of each time step, time steps that exceed the second preset threshold are marked to obtain the second risk time step set.

8. A system for assessing the state of student groups based on lesson plan texts, characterized in that, It includes a data acquisition unit (610) with sequential electrical connections, a simulation unit (620), a tensor construction unit (630), a dual-channel detection unit (640), and an evaluation output unit (650). The data acquisition unit (610) is configured to: obtain a sequence of teaching statements and a student profile matrix of the target class based on the lesson plan text and the target class roster; The simulation unit (620) is configured to: obtain an interaction log by simulating a preset classroom process based on the sequence of teaching statements; The tensor construction unit (630) is configured to: obtain a latent state temporal tensor based on the interaction log and the student profile matrix; The dual-channel detection unit (640) is configured to: obtain a first risk time step set and a second risk time step set by means of preset parallel dual-channel detection based on the latent state time sequence tensor; The evaluation output unit (650) is configured to: obtain the total number of risk triggers based on the first risk time step set and the second risk time step set, and obtain the evaluation result of the lesson plan text based on the total number of risk triggers.

9. A device, characterized in that, include: One or more processors; Memory; One or more programs, wherein the one or more programs are stored in memory and configured to be executed by one or more processors, the one or more programs being configured to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code that can be called by a processor to perform the method as described in any one of claims 1 to 7.

Images