Virtual teaching environment construction and feedback adjustment method and device based on immersion interaction

Abstract

The invention discloses a virtual teaching environment construction and feedback adjustment method and device based on immersion interaction, and relates to the technical field of virtual teaching. The virtual teaching environment construction and feedback adjustment method and device based on immersion interaction comprises the following steps: S1, obtaining class teaching behavior data, and preprocessing the class teaching behavior data; s2, constructing a content label vector, screening a candidate set, performing joint quantitative analysis on an adaptation relationship between recommended contents and students, and labeling a recommended state label of each recommended content; s3, collecting interaction feedback of the teacher to construct a recommendation interaction log, and analyzing the adaptation deviation degree of the recommendation content in the teaching scene; and S4, quantitatively analyzing recommended response fluctuation intensity, grading response fluctuation states, and triggering self-adaptive adjustment of weights according to fluctuation grading. The problem of misjudgment caused by difficulty in judging whether the recommended content is suitable for the student level by a teacher is solved.

Description

Technical Field

[0001] The present invention relates to the field of virtual teaching technology, and in particular to a method and device for constructing a virtual teaching environment and providing feedback adjustment based on immersive interaction. Background Art

[0002] The introduction of AI-optimized operating systems is gradually driving the evolution of educational technology towards greater intelligence and adaptability. Within existing educational technology systems, intelligent recommendation systems are already widely used to assist with personalized delivery of teaching content.

[0003] For example, the invention with publication number: CN115599204A provides an interactive virtual reality medical practice teaching method with operation feedback function. A medical practice simulation teaching scene unit is built through virtual reality technology, motion capture technology, mechanical feedback technology and big data analysis technology. The VR virtual reality simulation module accurately presents the three-dimensional data of the actual medical operation scene in the virtual scene; the motion capture module realizes precise interaction through precise hand motion capture technology; the mechanical feedback module writes the big data analysis results of clinical practice operations and digitizes the medical practice operation experience.

[0004] For example, the invention with announcement number: CN114327060A provides a working method of a virtual teaching system based on an AI assistant, including the construction of a teaching scene data set; the generation of an intelligent virtual avatar; virtual teaching based on an AI assistant; collecting videos of classroom environments and teaching activities, using diversified labels to construct a teaching scene data set, generating an intelligent virtual avatar, perceiving and understanding the virtual teaching environment, using visual methods to demonstrate knowledge points in collaborative teaching, building an AI test assistant, supporting training and assessment modes, and creating operation statistics, evaluation and feedback mechanisms.

[0005] However, although existing technologies have achieved initial results in improving the efficiency of teaching resource distribution, there are still problems such as the disconnection between content recommendations and real-time teaching behaviors, the inability to dynamically perceive adaptation errors, and the inability to adaptively update adjustment strategies. These problems make it difficult to meet the comprehensive requirements of high-frequency feedback and refined intervention in immersive and multimodal teaching scenarios.

[0006] Therefore, in response to the above problems, there is an urgent need for a method and device for constructing a virtual teaching environment and feedback adjustment based on immersive interaction. Summary of the Invention

[0007] Technical problems solved

[0008] In response to the shortcomings of the existing technology, the present invention provides a method and device for constructing a virtual teaching environment and feedback adjustment based on immersive interaction, which solves the problem that teachers have difficulty in judging whether the recommended content is suitable for the students' level and may make misjudgments.

[0009] Technical Solution

[0010] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for constructing a virtual teaching environment based on immersive interaction and feedback adjustment, comprising the following steps: S1, acquiring class teaching behavior data through real-time data collection and synchronous processing of multiple types of behavior recognition and interaction perception devices in an immersive teaching system, and preprocessing the class teaching behavior data to obtain preprocessed class teaching behavior data; S2, constructing a content label vector, screening a candidate set based on the preprocessed class teaching behavior data and the content label vector, performing a joint quantitative analysis of the adaptation relationship between recommended content and students, and annotating each recommended content with a recommendation status label; S3, pushing the recommendation status label and adaptation evaluation value to a display interface, collecting teacher interaction feedback to construct a recommendation interaction log, analyzing the degree of adaptation deviation of the recommended content in the teaching scenario based on the recommendation interaction log, extracting intervention features, and calculating the difference between the recommendation adaptation evaluation values ​​before and after adjustment to form a single adjustment residual set; S4, quantitatively analyzing the recommendation response fluctuation intensity and grading the response fluctuation state based on the adjustment residual set and the recommendation interaction log, triggering adaptive adjustment of the weight according to the fluctuation classification, and archiving the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop.

[0011] Furthermore, by collecting and synchronously processing real-time data from various types of behavior recognition and interactive perception devices in the immersive teaching system, the specific steps for obtaining class teaching behavior data are as follows: by collecting and synchronously processing real-time data from various types of behavior recognition and interactive perception devices in the immersive teaching system, class teaching behavior data is obtained. Class teaching data include the duration of answering a single teaching question, the accuracy of answering questions, the frequency of mouse operation, the average reaction time, the mean of gaze point offset, the duration of gaze, the duration of classroom questions, the number of words in recommended content, the number of recommended formulas, the number of recommended charts, and the average grade of the class; among them, the duration of answering a single teaching question is obtained by reading the timestamp difference between the question loading time and the submission time recorded in the teaching system; by comparing the students' answer results with the standard answers, the number of correct questions on the knowledge points is counted and the proportion is calculated to obtain the accuracy of the answer; by analyzing the number of mouse clicks, movements and scrolling events of students per minute in the client log The method comprises the following steps: first, calculating the frequency of mouse operation by counting the number of times the mouse clicks on the screen; second, calculating the average reaction time of students by the time interval between the appearance of prompt elements on the teaching interface and the first input operation of students, and obtaining the average reaction time; third, obtaining the mean value of gaze point offset by reading the spatial offset between the gaze coordinates collected by the eye tracking device and the center point of the teaching content; fourth, obtaining the gaze duration by analyzing the duration window of continuous gaze events; fifth, obtaining the duration of classroom questions by recording the start and end time of the audio during the period when students actively ask questions through the voice recognition module; fifth, obtaining the number of words in the recommended content by counting the words in the text area in the recommended teaching content; sixth, obtaining the number of recommended formulas by parsing the mathematical formula markup structure embedded in the content and counting the number of formulas; sixth, obtaining the number of recommended charts by identifying image tags and table elements in the content structure; and fifth, obtaining the average score of the class by reading the score data of each student in the specified assessment task in the teaching system and calculating the average value.

[0012] Furthermore, the class teaching behavior data are preprocessed, and the specific steps to obtain the preprocessed class teaching behavior data are as follows: through the collaborative detection of the median residual detection method based on the sliding window and the local outlier factor algorithm, the mutation values ​​and behavioral abnormal data in the average single-question answering time, gaze duration and average reaction time are identified and eliminated; by utilizing the combined completion mechanism of the time window linear interpolation method and the autoregressive prediction model, the missing fields in the mouse operation frequency, classroom question duration and recommended content word count are completed in time series; through the joint modeling strategy of fusing Gaussian kernel weighted sliding average and local polynomial regression algorithm, the class teaching data are smoothed and fitted to suppress high-frequency fluctuations; by adopting the cascade transformation of logarithmic transformation and maximum and minimum normalization algorithm, the class teaching data are rescaled and normalized.

[0013] Furthermore, the content label vector is constructed, and the specific steps of screening the candidate set based on the preprocessed class teaching behavior data and the content label vector are as follows: perform text parsing on the recommended content, extract keywords and map them to corresponding knowledge points and topics in the course knowledge graph; generate word frequency vectors based on knowledge points, generate embedding representations based on topics, and normalize them after splicing to form a content label vector; form a class knowledge portrait vector based on the preprocessed answer accuracy and class average scores, calculate the cosine similarity between the class knowledge portrait vector and each content label vector, and screen recommended content with a cosine similarity not lower than the knowledge fit threshold as the candidate set; if the number of candidates is lower than the lower limit of the candidate number threshold, then supplement them in descending order according to the cosine similarity of the recommended content until the set size requirement is met; if the number of candidates exceeds the upper limit of the candidate number threshold, then extract a specified number of recommended contents with the same cosine similarity value at fixed intervals to meet the set number constraint.

[0014] Furthermore, the adaptation relationship between the recommended content and students is jointly quantitatively analyzed, and the specific steps of marking the recommendation status label of each recommended content are as follows: extract the teaching single question answering time, answering accuracy, gaze duration, class average score, recommended content word count and number of recommended charts of each recommended content in the candidate set, and conduct a joint quantitative analysis of the adaptation relationship between the recommended content and students: calculate the ratio of the product of the class average score and the answering accuracy divided by the teaching single question answering time plus 1, calculate the logarithm of the gaze duration after adding 1, add this ratio to the logarithm and multiply it by the class learning performance weight, and use the product as the learning performance item; calculate the square root of the recommended content word count, calculate the square value of the number of recommended charts, add this square root to the square value and multiply it by the complexity of the recommended content The weight is multiplied as the content complexity item; the absolute value of the difference between the learning performance item and the content complexity item is calculated to obtain the recommendation adaptation evaluation value; the recommendation adaptation evaluation value of each recommended content in the candidate set is compared with the recommendation evaluation threshold, and the recommendation evaluation threshold includes the first-level recommendation evaluation threshold and the second-level recommendation evaluation threshold; if the recommendation adaptation evaluation value is less than or equal to the first-level recommendation evaluation threshold, the recommended content is marked as a low adaptation state; if the recommendation adaptation evaluation value is greater than the first-level recommendation evaluation threshold and less than the second-level recommendation evaluation threshold, the recommended content is marked as a medium adaptation state; if the evaluation value is greater than or equal to the second-level recommendation evaluation threshold, the recommended content is marked as a high adaptation state; the recommendation status label and the recommendation adaptation evaluation value are attached to the candidate set as the basis for subsequent manual interaction and selection strategy.

[0015] Furthermore, the recommendation status label and adaptation evaluation value are pushed to the display interface, and the specific steps for collecting teachers' interactive feedback to construct a recommendation interaction log are as follows: push the recommendation status label to the candidate content display interface, attach the recommendation adaptation evaluation value, learning performance item, and content complexity item of each recommended content, and list the teaching single question answering time, answering accuracy, gaze duration, recommended content word count, and number of recommended charts corresponding to the recommended content, and display them in a structured manner according to field numbers; provide three adjustment options of accept, replace, and fine-tune in the interface, and teachers can provide interactive feedback on specific recommended content based on actual teaching needs; if fine-tuning is selected, teachers can further set the adjustment ratio of the learning performance item and the content complexity item for temporary weight modification of the current recommended content; count the cumulative number of adjustment operations of the replacement and fine-tuning type in the current interaction cycle to obtain the recommendation adjustment frequency; calculate the total number of times a field appears in all recommended content in the current cycle and divide it by the number of recommended content including the field to obtain the average number of recommendations for the field; generate a corresponding recommendation interaction log for each teacher's operation, which includes the recommendation adjustment frequency, the average number of recommendations for the field, the recommendation status label, and class teaching data.

[0016] Furthermore, based on the recommendation interaction log, the degree of adaptation deviation of the recommended content in the teaching scenario is analyzed, the intervention features are extracted, and the difference between the recommendation adaptation evaluation values ​​before and after adjustment is calculated to form a single adjustment residual set. The specific steps are as follows: After the interaction cycle ends, the degree of deviation of the recommended content with the adjustment operation type of replacement and fine-tuning in the actual teaching scenario is comprehensively analyzed based on the recommendation interaction log: Calculate the product of the class average score and the correct answer rate, calculate the square value of the time taken to answer a single teaching question plus 1, and divide this product by the square to obtain the learning efficiency ratio term; calculate the logarithm of the natural logarithm after adding 1 to the gaze duration, calculate the square root of the number of words in the recommended content plus 1, and divide it by the square. This logarithmic value is divided by the square root to obtain the visual input adjustment item; the square value of the number of recommended charts is calculated, and this square value is divided by the mouse operation frequency plus 1 to obtain the chart load index item; the square value of the adjustment frequency is calculated, and this square value is divided by the average number of recommendations plus 1 to obtain the manual intervention item; the learning efficiency ratio item is added to the visual input adjustment item and then subtracted from the chart load index item, and the difference is multiplied by the manual intervention item and the absolute value is taken to obtain the recommendation deviation evaluation value; based on all recommended content that has been fine-tuned, the recommendation deviation evaluation values ​​before and after adjustment are extracted and the difference between the two is calculated as the single adjustment residual of each recommended content, and the adjustment residual set of the current interaction cycle is summarized.

[0017] Furthermore, based on the adjustment residual set and the recommendation interaction log, the response intensity of the recommended content in the current teaching cycle to the interaction between the class learning status and the teacher's adjustment behavior is measured; the product of the class average score and the correct answer rate is calculated, and the square value of the teaching single question answering time plus 1 is calculated, and this product is divided by the square value to obtain the learning load factor; the product of the number of words in the recommended content and the number of recommended charts is calculated and added with 1, and then the natural logarithm is taken to obtain the structural complexity factor; the mouse operation frequency of the recommended content is added with 1 and the natural logarithm is taken, and this logarithm is multiplied with the single adjustment residual to obtain the behavioral response factor; the learning load factor and the structural complexity factor are added, and then multiplied by the behavioral response factor to obtain the recommendation response fluctuation value; after the recommendation response fluctuation value is calculated, the recommendation response fluctuation is compared in real time The response fluctuation threshold includes the first-level response fluctuation threshold and the second-level response fluctuation threshold; when the response fluctuation threshold is less than or equal to the first-level response fluctuation threshold, the current recommended content is marked as low-response fluctuation content, and the regular adjustment frequency is maintained; when the response fluctuation threshold is greater than the first-level response fluctuation threshold and less than the second-level response fluctuation threshold, the current recommended content is marked as medium-response fluctuation status and automatically included in the short-term key observation queue; when the response fluctuation threshold is greater than or equal to the second-level response fluctuation threshold, the current recommended content is marked as high-response fluctuation status, and automatically included in the high-sensitivity review list of the next round of recommended content, and its manual intervention and review priority is increased; the fluctuation status mark is attached to the recommended content metadata to form a dynamic adjustment index for the teaching adaptation feedback chain.

[0018] Furthermore, adaptive weight adjustment is triggered based on fluctuation classification. The updated results are archived to generate a strategy snapshot to drive the next recommendation cycle in a closed loop. The specific steps are as follows: Based on the numerical distribution of recommendation response fluctuation values, samples with continuously increasing fluctuation amplitudes are identified in medium- and high-response recommended content. The recommendation response fluctuation value sequence for two or more consecutive interaction cycles is extracted to determine whether a monotonically increasing trend is met. Simultaneously, the adjustment frequency sequence of the corresponding recommended content within each cycle is extracted to calculate the change amplitude of the adjustment density. If medium- and high-response recommended content shows an increasing trend in both recommendation response fluctuation value and adjustment frequency, an adjustment sensitivity factor is constructed as the weighted average of the two change amplitudes. When the adjustment sensitivity factor exceeds the weight adjustment threshold, the weight update mechanism is automatically triggered, and the weights of class learning performance and recommendation content complexity are steadily updated using a sliding weighted average method. At the same time, the calculation structure parameters of the recommendation adaptation evaluation value in the next interaction cycle are reconstructed. The recommendation response fluctuation value, response classification label, single adjustment residual, adjustment sensitivity factor, and weight update results in the current cycle are archived and saved to generate a recommendation strategy snapshot, which serves as the input for recommendation content adaptation and strategy correction in subsequent cycles, thus achieving a complete adjustment feedback loop.

[0019] The second aspect of the present invention provides a virtual teaching environment construction and feedback adjustment device based on immersive interaction, including: a class teaching data acquisition and preprocessing module, a teaching content intelligent recommendation and adaptation evaluation module, a recommended content interaction adjustment and deviation analysis module and a response fluctuation analysis and weight adaptive update module, wherein: the class teaching data acquisition and preprocessing module is used to acquire class teaching behavior data through real-time data acquisition and synchronous processing of multiple types of behavior recognition and interactive perception devices in the immersive teaching system, and preprocess the class teaching behavior data to obtain preprocessed class teaching behavior data; the teaching content intelligent recommendation and adaptation evaluation module is used to construct a content label vector, screen a candidate set based on the preprocessed class teaching behavior data and the content label vector, and perform preprocessing on the class teaching behavior data. The adaptation relationship between recommended content and students is jointly quantitatively analyzed, and the recommendation status label of each recommended content is marked; the recommendation content interaction adjustment and deviation analysis module is used to push the recommendation status label and adaptation evaluation value to the display interface, collect teachers' interactive feedback to build a recommendation interaction log, analyze the degree of adaptation deviation of recommended content in the teaching scenario based on the recommendation interaction log, extract intervention features and calculate the difference between the recommendation adaptation evaluation value before and after adjustment to form a single adjustment residual set; the response fluctuation analysis and weight adaptive update module is used to quantitatively analyze the recommendation response fluctuation intensity and grade the response fluctuation status based on the adjustment residual set and recommendation interaction log, trigger adaptive adjustment of the weight according to the fluctuation grade, archive the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop.

[0020] Beneficial effects

[0021] The present invention has the following beneficial effects:

[0022] (1) The method and device for constructing a virtual teaching environment and adjusting feedback based on immersive interaction realizes quantitative matching analysis of the degree of knowledge fit between recommended content and class learning status through the content label vector constructed based on the course knowledge graph and the class knowledge portrait vector formed by the correct answer rate, class average score, etc., and selects candidate recommended content with highly consistent semantic structure and learning status, thereby strengthening the semantic association matching ability between recommended content and class learning performance.

[0023] (2) The method and device for constructing a virtual teaching environment and adjusting feedback based on immersive interaction jointly quantifies the degree of adaptation between the recommended content and the current learning load and structural complexity of the class based on multiple indicators such as the duration of answering a single teaching question, the duration of gaze, and the number of words in the recommended content, and assigns a recommendation status label, so that teachers can intuitively understand the adaptation status of the recommended content and conduct targeted intervention, thereby enhancing the interpretability of the adaptability of the recommended content and the ability to adjust feedback.

[0024] (3) The method and device for constructing a virtual teaching environment and adjusting feedback based on immersive interaction quantifies the degree of fluctuation in the feedback of recommended content on learning status and teacher operations during the current teaching cycle by integrating multi-source data such as teaching performance and structural characteristics, and classifies the recommended content into low, medium, and high response fluctuation states based on the grading results, thereby assisting in the dynamic adjustment of priority allocation, thereby improving the system's recognition sensitivity and timely response to unstable content.

[0025] (4) The method and device for constructing a virtual teaching environment and feedback adjustment based on immersive interaction adjusts the residual set and the fluctuation trend of the recommendation response, analyzes the changes in the adjustment density and response amplitude, and extracts the adjustment sensitive factor, which drives the smooth update mechanism of the class learning performance weight and the recommendation content complexity weight, and finally archives it as a strategy snapshot to achieve the cyclic iteration and adaptive closed loop of the recommendation mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flowchart of the construction and feedback adjustment method of a virtual teaching environment based on immersive interaction;

[0027] Figure 2 The structure diagram of the feedback adjustment device for building a virtual teaching environment based on immersive interaction;

[0028] Figure 3 Adapt feature item comparison charts for recommended content;

[0029] Figure 4 Recommended content bias evaluation value distribution chart. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] See also Figure 1-Figure 4An embodiment of the present invention provides a technical solution: a method for constructing a virtual teaching environment based on immersive interaction and feedback adjustment, comprising the following steps: S1, acquiring class teaching behavior data by collecting and synchronously processing real-time data from multiple types of behavior recognition and interaction perception devices in an immersive teaching system, and preprocessing the class teaching behavior data to obtain preprocessed class teaching behavior data; S2, constructing a content label vector, screening a candidate set based on the preprocessed class teaching behavior data and the content label vector, performing a joint quantitative analysis of the adaptation relationship between recommended content and students, and annotating a recommendation status label for each recommended content; S3, pushing the recommendation status label and adaptation evaluation value to a display interface, collecting teacher interaction feedback to construct a recommendation interaction log, analyzing the degree of adaptation deviation of the recommended content in the teaching scenario based on the recommendation interaction log, extracting intervention features, and calculating the difference between the recommendation adaptation evaluation values ​​before and after adjustment to form a single adjustment residual set; S4, quantitatively analyzing the recommendation response fluctuation intensity and grading the response fluctuation state based on the adjustment residual set and the recommendation interaction log, triggering adaptive adjustment of the weight according to the fluctuation classification, and archiving the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop.

[0032] Specifically, by collecting and synchronously processing real-time data of various types of behavior recognition and interactive perception devices in the immersive teaching system, the specific steps for obtaining class teaching behavior data are as follows: by collecting and synchronously processing real-time data of various types of behavior recognition and interactive perception devices in the immersive teaching system, class teaching behavior data are obtained. Class teaching data include the duration of answering a single teaching question, the accuracy of answering questions, the frequency of mouse operation, the average reaction time, the mean of gaze point offset, the duration of gaze, the duration of classroom questions, the number of words in recommended content, the number of recommended formulas, the number of recommended charts, and the average grade of the class; among them, the duration of answering a single teaching question is obtained by reading the timestamp difference between the question loading time and the submission time recorded in the teaching system; by comparing the students' answer results with the standard answers, the number of correct questions on the knowledge points is counted and the proportion is calculated to obtain the accuracy of the answer; by analyzing the students' mouse click, movement and scrolling events per minute in the client log The frequency of mouse operation is obtained; the average reaction time of students is calculated through the time interval between the appearance of prompt elements in the teaching interface and the first input operation of students to obtain the average reaction time; the spatial offset between the gaze coordinates collected by the eye tracking device and the center point of the teaching content is read to obtain the mean value of the gaze point offset; the gaze duration is obtained by analyzing the duration window of continuous gaze events; the audio start and end time of the period when students actively ask questions is recorded through the voice recognition module to obtain the duration of classroom questions; the number of words in the text area of ​​the recommended teaching content is obtained by counting the words; the number of formulas is counted by parsing the mathematical formula markup structure embedded in the content to obtain the number of recommended formulas; the number of recommended charts is obtained by identifying image tags and table elements in the content structure; the average score of the class is obtained by reading the score data of each student in the specified assessment task in the teaching system and calculating the average value.

[0033] In this implementation plan, through clear data collection and processing methods, we can accurately obtain key teaching behavior data such as the duration of answering single questions, answering accuracy, mouse operation frequency, average reaction time, mean gaze point offset, gaze duration, classroom questioning duration, number of words in recommended content, number of recommended formulas, number of recommended charts and class average grades. This provides complete and structured basic data support for the subsequent intelligent recommendation, adaptive evaluation and adjustment strategies based on class teaching status, and ensures the consistency and traceability of data input links in each calculation module in the system.

[0034] Specifically, the specific steps for preprocessing the class teaching behavior data to obtain the preprocessed class teaching behavior data are as follows: through the collaborative detection of the median residual detection method based on the sliding window and the local outlier factor algorithm, the mutation values ​​and behavioral abnormal data in the average single question answering time, gaze duration and average reaction time are jointly identified and accurately eliminated to ensure the data stability of key behavioral variables in subsequent calculations; by utilizing the combined completion mechanism of the time window linear interpolation method and the autoregressive prediction model, the field missing situations in the mouse operation frequency, classroom question duration and recommended content word count are time-series completed to improve data Continuity and time series integrity; through the joint modeling strategy of fusing Gaussian kernel weighted sliding average and local polynomial regression algorithm, class teaching data such as gaze point offset mean, answering accuracy and class average score are smoothed and fitted to suppress high-frequency fluctuations and reduce the interference of short-term abnormal fluctuations on overall trend analysis; through the cascade transformation of logarithmic transformation and maximum and minimum normalization algorithm, fields with large differences in numerical scales such as single-question answering time, gaze duration, and number of recommended charts are rescaled and normalized to unify the dimensions and distribution characteristics of different fields, providing a standardized input basis for subsequent calculation modules.

[0035] In this implementation plan, by preprocessing the class teaching behavior data, not only the mutation values ​​and behavioral abnormalities in the average single-question answering time, gaze duration and average reaction time are effectively identified and eliminated, but also the missing fields in the mouse operation frequency, classroom questioning time and recommended content word count are completed, thereby improving the data integrity and continuity; at the same time, by smoothing the trends of key fields such as the mean gaze point offset, answering accuracy and class average score, high-frequency fluctuation interference is reduced and the overall data stability is enhanced; finally, by unifying the numerical scales of variables such as the single-question answering time, gaze duration and number of recommended charts, the comparability and universality of the data in the subsequent recommendation adaptation analysis and deviation evaluation links are achieved, laying a reliable data foundation for the generation of recommendation status labels and adjustment behavior modeling.

[0036] Specifically, the specific steps of constructing a content label vector and screening a candidate set based on the pre-processed class teaching behavior data and the content label vector are as follows: perform text parsing on the recommended content, extract keywords from the recommended content, and map the keywords to corresponding knowledge points and topic nodes in the course knowledge graph to ensure that the semantic structure is consistent with the course system; the course knowledge graph is constructed based on the national curriculum standards and textbook content, combined with the teaching syllabus and concept system of each subject area, and covers the graph structure of the main knowledge points and teaching topics; construct a word frequency vector based on the frequency of words appearing in the knowledge point extraction content, generate an embedding representation vector based on the topic, and further capture the implicit semantic associations between topics; after splicing the word frequency vector and the embedding representation vector, use the maximum and minimum normalization method to complete the unified scaling process to form a normalized content label vector; according to the pre-processed answer accuracy and the class average score, aggregate them by knowledge point number to generate the class score. The comprehensive mastery of each knowledge point dimension by the class is used to construct a class knowledge portrait vector to express the current learning performance status of the class; the cosine similarity between each content label vector and the class knowledge portrait vector is calculated as a quantitative indicator of the degree of match between the recommended content and the current learning performance of the class; the recommended content with a cosine similarity not lower than the knowledge fit threshold is screened and included in the candidate set to ensure that the content fit meets the basic teaching adaptation requirements; if the number of recommended content in the candidate set is lower than the lower limit of the candidate number threshold, the system will automatically supplement the recommended content from the unselected content in descending order of cosine similarity until the set size requirement is met to ensure sufficient alternative content; if the number of candidate sets exceeds the upper limit of the candidate number threshold, a specified number of content samples will be extracted at fixed intervals from the recommended content with the same cosine similarity to compress the set size while retaining the similarity distribution structure to meet the set number constraint and control subsequent processing overhead.

[0037] This implementation effectively improves the alignment between recommended content and class learning performance by constructing normalized content label vectors and class knowledge profile vectors, and performing precise screening based on cosine similarity. Supported by an AI-optimized operating system, knowledge fit thresholds and set size constraints ensure that candidate sets are both targeted and comprehensive, reducing interference from irrelevant or redundant content and providing a high-quality content foundation for subsequent recommendation adaptation evaluation and adjustment decisions.

[0038] Specifically, the specific steps of conducting a joint quantitative analysis on the adaptation relationship between the recommended content and students and marking the recommendation status label of each recommended content are as follows: extract the teaching single question answering time, answering accuracy, gaze duration, class average score, recommended content word count and recommended chart number of each recommended content in the candidate set, and conduct a joint quantitative analysis on the adaptation relationship between the recommended content and students: calculate the product of the class average score and the answering accuracy rate divided by the teaching single question answering time plus 1, calculate the logarithm of the gaze duration after adding 1, add this ratio to the logarithm and multiply it by the class learning performance weight, and use the product as the learning performance item; calculate the square root of the recommended content word count, calculate the square value of the number of recommended charts, add this square root to the square value and multiply it by the recommended content complexity weight. The product is taken as the content complexity term; the absolute value of the difference between the learning performance term and the content complexity term is calculated to obtain the recommendation adaptation evaluation value; the recommendation adaptation evaluation value of each recommended content in the candidate set is compared with the recommendation evaluation threshold, and the recommendation evaluation threshold includes the first-level recommendation evaluation threshold and the second-level recommendation evaluation threshold; if the recommendation adaptation evaluation value is less than or equal to the first-level recommendation evaluation threshold, the recommended content is marked as a low adaptation state; if the recommendation adaptation evaluation value is greater than the first-level recommendation evaluation threshold and less than the second-level recommendation evaluation threshold, the recommended content is marked as a medium adaptation state; if the evaluation value is greater than or equal to the second-level recommendation evaluation threshold, the recommended content is marked as a high adaptation state; the recommendation status label and the recommendation adaptation evaluation value are attached to the candidate set as the basis for subsequent manual interaction and selection strategy.

[0039] The specific calculation formula for the recommended adaptation evaluation value is:

[0040]

[0041] In the formula, R represents the recommendation adaptation evaluation value, α represents the class learning performance weight, β represents the weight of the complexity of the recommended content, S represents the class average score, Y represents the correct answer rate, T represents the time it takes to answer a single teaching question, A represents the duration of fixation, and N represents the average score of the class. w Indicates the number of words in the recommended content, N s Indicates the number of recommended charts.

[0042] In this embodiment, Table 1 is a recommended adaptation evaluation value data table, which records in detail the key indicator variables and recommended adaptation evaluation value calculation results of the 5 recommended contents in the adaptation analysis process, and is used to quantify the degree of match between the recommended content and the class learning performance. Among them: in recommended content 1, the class average score is 85, the answer accuracy rate is 0.92, the teaching single question answering time is 60, the gaze duration is 4.5, the recommended content word count is 350, the number of recommended charts is 2, and the recommended adaptation evaluation value is 7.99; in recommended content 2, the class average score is 78, the answer accuracy rate is 0.85, the teaching single question answering time is 75, the gaze duration is 5.2, the recommended content word count is 420, the number of recommended charts is 3, and the recommended adaptation evaluation value is 10.91; in recommended content 3, the class average score is 92, the answer accuracy rate is 0.95, the teaching single question answering time is 50, The gaze duration is 3.8, the number of words in the recommended content is 310, the number of recommended charts is 1, and the recommended adaptation evaluation value is 6.14; in recommended content 4, the class average score is 88, the answer accuracy rate is 0.90, the teaching time for answering a single question is 68, the gaze duration is 4.0, the number of words in the recommended content is 390, the number of recommended charts is 2, and the recommended adaptation evaluation value is 8.53; in recommended content 5, the class average score is 81, the answer accuracy rate is 0.88, the teaching time for answering a single question is 80, the gaze duration is 5.5, the number of words in the recommended content is 450, the number of recommended charts is 3, and the recommended adaptation evaluation value is 11.18.

[0043] Table 1 Recommended adaptation evaluation value data table

[0044] Recommended content S Y T A <![CDATA[N w ]]> <![CDATA[N s ]]> R Recommended content 1 85 0.92 60 4.5 350 2 7.99 Recommended content 2 78 0.85 75 5.2 420 3 10.91 Recommended content 3 92 0.95 50 3.8 310 1 6.14 Recommended content 4 88 0.90 68 4.0 390 2 8.53 Recommended content 5 81 0.88 80 5.5 450 3 11.18

[0045] like Figure 3 As shown in the figure, it is a comparison chart of the adaptation features of recommended contents, which shows the distribution of evaluation values ​​of the 5 recommended contents in the recommendation adaptation evaluation analysis. Through the juxtaposition of the learning performance item, content complexity item and recommendation adaptation evaluation value of the recommended contents in the figure, it can be seen that there are obvious differences in the three indicators of each recommended content; among them, the content complexity item of recommended content 2 and recommended content 5 is relatively high, but because the learning performance item is also strong, the final recommendation adaptation evaluation value remains high; and the learning performance item of recommended content 3 is the lowest, resulting in its recommendation adaptation evaluation value being significantly low; this shows that the recommendation adaptation evaluation value has better integrated the dynamic relationship between class learning performance and content complexity under the setting of weight parameters.

[0046] This implementation extracts key metrics such as the duration of a single teaching question, accuracy rate, fixation duration, class average, word count in recommended content, and number of recommended graphics. This is combined with the weight of class performance and the weight of the complexity of the recommended content to quantify the compatibility between the recommended content and the student, and this is used to calculate a recommendation compatibility assessment. By optimizing the operating system through artificial intelligence, the evaluation calculation process is constructed, and primary and secondary recommendation assessment thresholds are set. Recommended content in the candidate set is labeled with recommendation status, achieving a detailed classification of the degree of recommendation compatibility, providing an accurate and quantifiable basis for subsequent human interaction and selection strategies.

[0047] Specifically, the recommendation status label and adaptation evaluation value are pushed to the display interface, and the specific steps of collecting teachers' interactive feedback to build a recommendation interaction log are as follows: push the recommendation status label to the candidate content display interface, attach the recommendation adaptation evaluation value, learning performance item and content complexity item of each recommended content, and list the teaching single question answering time, answering accuracy, gaze duration, recommended content word count and recommended chart number corresponding to the recommended content, and display them in a structured manner according to field numbers to support teachers to intuitively identify the key features of the recommended content; provide three types of adjustment options: accept, replace and fine-tune in the interface, and teachers can make personalized interactive feedback on specific recommended content based on the current teaching objectives and class status; if fine-tuning is selected, teachers can further Set the adjustment ratio of learning performance items and content complexity items for temporary weight modification of current recommended content, and enhance the flexibility and pertinence of recommendation adjustment; count the cumulative number of replacement and fine-tuning adjustment operation types in the current interaction cycle, and obtain the recommendation adjustment frequency, which is used to measure the activeness of the teacher's adjustment behavior; calculate the total number of times a field appears in all recommended content in the current cycle and divide it by the number of recommended content that includes the field to obtain the average number of field recommendations, which assists in analyzing the attention density of different elements in the recommended content; generate a corresponding recommendation interaction log for each teacher's operation, which includes the recommendation adjustment frequency, the average number of field recommendations, the recommendation status label and class teaching data, and provides a data basis for subsequent deviation analysis and weight update.

[0048] In this implementation plan, by intuitively pushing the recommendation status label and the recommendation adaptation evaluation value to the display interface, and displaying key variables such as the duration of answering a single teaching question, the accuracy of the answer, the duration of gaze, the number of words in the recommended content, and the number of recommended charts, teachers can fully perceive the composition characteristics of the recommended content; combined with interaction methods such as acceptance, replacement, and fine-tuning, teachers are supported to provide differentiated feedback on recommended content based on teaching needs; by collecting the recommendation adjustment frequency and the average number of field recommendations, the adjustment behavior characteristics of the recommended content in the actual teaching process are recorded; finally, the above information is summarized to form a recommendation interaction log, which provides data support for subsequent analysis of the adaptation deviation of the recommended content, updating the weights of learning performance items and content complexity items, and improving the degree of fit between recommended content and teaching scenarios under the artificial intelligence optimization operating system.

[0049] Specifically, based on the recommendation interaction log, the degree of adaptation deviation of the recommended content in the teaching scenario is analyzed, the intervention characteristics are extracted, and the difference in the recommendation adaptation evaluation value before and after adjustment is calculated to form a single adjustment residual set. The specific steps are as follows: After the interaction cycle ends, the degree of deviation of the recommended content with the adjustment operation type of replacement and fine-tuning in the actual teaching scenario is comprehensively analyzed based on the recommendation interaction log: calculate the product of the class average score and the correct answer rate, calculate the square value of the answering time of a single teaching question plus 1, and divide this product by the square to obtain the learning efficiency ratio item; calculate the logarithm of the natural logarithm after calculating the duration of gaze plus 1, calculate the square root of the number of words in the recommended content plus 1, and divide this logarithm by the square root to obtain the visual input adjustment item; calculate the square value of the number of recommended charts, and divide this square value by the mouse operation frequency plus 1 to obtain the chart load index item; calculate the adjustment frequency The square value is divided by the average number of recommendations plus 1 to obtain the manual intervention item; the learning effectiveness ratio item is added to the visual input adjustment item and then the chart load index item is subtracted. This difference is multiplied by the manual intervention item and the absolute value is taken to obtain the recommendation deviation evaluation value; among them, by minimizing the residual sum of squares and adopting the grid search method, the class learning performance weight and the recommendation content complexity weight corresponding to the historical recommended content are fitted by using the joint change relationship between the duration of answering single questions, the accuracy of answering questions, the duration of gaze and the average score of the class in multiple dimensions. The value range of the class learning performance weight and the recommendation content complexity weight are both [1,0]; based on all the recommended content that has been fine-tuned, the recommendation deviation evaluation values ​​before and after adjustment are extracted and the difference between the two is calculated as the single adjustment residual of each recommended content, and the adjustment residual set of the current interaction cycle is summarized.

[0050] The specific calculation formula for the recommended deviation assessment value is:

[0051]

[0052] Where D represents the recommended deviation evaluation value, S represents the average score of the class; Y represents the correct answer rate, T represents the time it takes to answer a single question, A represents the duration of fixation, and N represents the correct answer rate. w Indicates the number of words in the recommended content, N s Indicates the number of recommended charts, M indicates the frequency of mouse operation, and f indicates the frequency of recommended adjustments. Indicates the average number of recommendations for a field.

[0053] In this embodiment, Table 2 is a recommendation deviation evaluation value data table, which records in detail the key input variables of the five recommended contents in the recommendation interaction feedback cycle and the calculation results of the recommendation deviation evaluation values, which is used to analyze the degree of response difference between the recommended content in the actual teaching scenario and the class learning status. Among them: in the recommended content C1, the class average score is 85, the answer accuracy rate is 0.92, the teaching single question answering time is 60, the gaze duration is 4.5, the recommended content word count is 350, the number of recommended charts is 2, the mouse operation frequency is 25, the recommended adjustment frequency is 3, the average number of field recommendations is 2, and the recommendation deviation evaluation value is 0.13; in the recommended content C2, the class average score is 78, the answer accuracy rate is 0.85, the teaching single question answering time is 75, the gaze duration is 4.2, the recommended content word count is 390, the number of recommended charts is 2, the mouse operation frequency is 25, the recommended adjustment frequency is 3, the average number of field recommendations is 2, and the recommendation deviation evaluation value is 0.18; in the recommended content C3, the class average score is 92, the answer accuracy rate is 0.95, the teaching single question answering time is 50, the gaze duration is 3.8, and the recommended content The number of words is 310, the number of recommended charts is 1, the mouse operation frequency is 20, the recommended adjustment frequency is 2, the average number of field recommendations is 2, and the recommendation deviation evaluation value is 0.10; in the recommended content C4, the class average score is 88, the answer accuracy rate is 0.90, the teaching single question answering time is 68, the gaze duration is 4.0, the recommended content word count is 390, the number of recommended charts is 2, the mouse operation frequency is 22, the recommended adjustment frequency is 4, the average number of field recommendations is 3, and the recommendation deviation evaluation value is 0.30; in the recommended content C5, the class average score is 81, the answer accuracy rate is 0.88, the teaching single question answering time is 80, the gaze duration is 4.0, the recommended content word count is 410, the number of recommended charts is 2, the mouse operation frequency is 24, the recommended adjustment frequency is 3, the average number of field recommendations is 2, and the recommendation deviation evaluation value is 0.21.

[0054] Table 2 Recommended deviation evaluation value data table

[0055]

[0056] like Figure 4As shown, it is a distribution diagram of the recommendation deviation evaluation value of recommended content. The horizontal axis is the recommended content number, which is C1 to C5 in sequence, and the vertical axis is the recommendation deviation evaluation value, showing the difference in the recommendation deviation evaluation value of each recommended content in the current interaction cycle. Among them, the recommendation deviation evaluation value of recommended content C4 is the highest at 0.30, and the recommendation deviation evaluation value of recommended content C3 is the lowest at 0.10. The rest are C5, C2 and C1 in sequence. By visually comparing the recommendation deviation evaluation values ​​corresponding to different content numbers, teachers can quickly identify recommended content with higher deviation values ​​as key review and priority adjustment objects, and at the same time provide a basis for the system to subsequently perform recommendation strategy optimization and weight adaptive adjustment, thus realizing explainable analysis and closed-loop control of the recommendation process.

[0057] In this implementation plan, through in-depth analysis of the recommended content involving replacement and fine-tuning operations in the recommendation interaction log, and comprehensive use of key variables such as class average grades, answer accuracy, teaching single question answering time, gaze duration, recommended content word count, recommended chart number and mouse operation frequency, we construct learning effectiveness ratio items, visual input adjustment items, chart load index items and manual intervention items, quantify the degree of adaptation deviation of recommended content in actual teaching scenarios, and calculate the recommendation deviation evaluation value; further extract the difference in recommendation deviation evaluation values ​​before and after adjustment under all fine-tuning operations to form a single adjustment residual set, thereby capturing the adaptation change trajectory of recommended content before and after manual intervention, and providing core intervention feature basis for subsequent recommendation strategy feedback and adaptive updates.

[0058] Specifically, based on the adjustment residual set and the recommendation interaction log, the response intensity of the recommended content in the current teaching cycle to the interaction between the class learning status and the teacher's adjustment behavior is measured; the product of the class average score and the answer accuracy rate is calculated, and the square value of the teaching time for answering a single question plus 1 is calculated, and this product is divided by the square value to obtain the learning load factor; the product of the number of words in the recommended content and the number of recommended charts is calculated, and then 1 is added, and then the natural logarithm is taken to obtain the structural complexity factor; the mouse operation frequency of the recommended content is added by 1 and the natural logarithm is taken, and this logarithm is multiplied by the single adjustment residual to obtain the behavioral response factor; the learning load factor is added to the structural complexity factor, and then multiplied by the behavioral response factor to obtain the recommended response fluctuation value; after the recommended response fluctuation value is calculated, the recommended response fluctuation value and the response fluctuation threshold are compared in real time. The response fluctuation threshold includes the first-level response fluctuation threshold and the second-level response fluctuation threshold; when the response When the response fluctuation threshold is less than or equal to the first-level response fluctuation threshold, the current recommended content will be marked as low-response fluctuation content, and the regular adjustment frequency will be maintained; when the response fluctuation threshold is greater than the first-level response fluctuation threshold and less than the second-level response fluctuation threshold, the current recommended content will be marked as medium-response fluctuation status and automatically included in the short-term key observation queue; when the response fluctuation threshold is greater than or equal to the second-level response fluctuation threshold, the current recommended content will be marked as high-response fluctuation status and automatically included in the high-sensitivity review list for the next round of recommended content, and its manual intervention and review priority will be increased; among them, the high-sensitivity review list is constructed based on the high-response fluctuation status sample library of historical recommended content, covering typical recommended content items that trigger high-fluctuation tags in multiple recommendation cycles, and is the focus of subsequent recommendation quality control; the fluctuation status tag is attached to the recommended content metadata to form a dynamic adjustment index for the teaching adaptation feedback chain.

[0059] The specific calculation formula for the recommended response fluctuation value is:

[0060]

[0061] In the formula, E represents the recommended response fluctuation value, S represents the average score of the class; Y represents the correct answer rate, T represents the time it takes to answer a single question, and N represents the correct answer rate. w Indicates the number of words in the recommended content, N s represents the number of recommended charts, M represents the frequency of mouse operation, and ε represents the residual error of a single adjustment.

[0062] In this implementation plan, by integrating the adjustment residual set and the recommendation interaction log, and comprehensively using variables such as class average score, answer accuracy rate, teaching single question answering time, recommended content word count, recommended chart number, mouse operation frequency and single adjustment residual, we construct the learning load factor, structural complexity factor and behavioral response factor, and quantify the recommendation response fluctuation value based on this, so as to effectively measure the response intensity of the recommended content to the interaction between the class learning status and the teacher's adjustment behavior during the teaching cycle; further, based on the comparison results of the recommendation response fluctuation value and the response fluctuation threshold, the response status label of the recommended content is dynamically marked, and the metadata structure of the recommended content is improved to achieve accurate management of the recommendation content adjustment priority and dynamic tracking of the teaching adaptation feedback chain.

[0063] Specifically, based on the adaptive adjustment of the fluctuation classification trigger weight, the specific steps of archiving the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop are as follows: Based on the numerical distribution of the recommendation response fluctuation value, combined with the pre-processed recommendation interaction log and the recommendation content metadata, identify samples with continuously increasing fluctuation amplitudes in the medium-response and high-response recommendation content, and judge whether the monotonically increasing trend is met by extracting the recommendation response fluctuation value sequence of two or more consecutive interaction cycles; at the same time, combined with the recommendation adjustment frequency and the average number of field recommendations, extract the adjustment frequency sequence of the corresponding recommendation content in each cycle, calculate the adjustment density change amplitude, and ensure the dynamic traceability of the behavioral reaction intensity; if the medium- and high-response recommendation content has an increasing trend in both the recommendation response fluctuation value and the adjustment frequency, the two change amplitudes are used to calculate the dynamic traceability of the behavioral reaction intensity; The weighted average of the degrees of recommendation is used to construct an adjustment sensitivity factor, which is used to quantify the degree to which the current recommended content triggers the adjustment of strategy parameters; when the adjustment sensitivity factor is higher than the weight adjustment threshold, the weight update mechanism is automatically triggered, and the sliding weighted average method is used to smoothly update the class learning performance weight and the recommended content complexity weight to prevent excessive fluctuations from causing structural deviations in the recommendation logic. At the same time, the calculation structure parameters of the recommended adaptation evaluation value in the next interaction cycle are reconstructed to enhance the dynamic adaptability of the evaluation system; finally, the recommended response fluctuation value, response classification label, single adjustment residual, adjustment sensitivity factor and weight update result in the current cycle are uniformly archived and saved to generate a recommendation strategy snapshot, which serves as the input basis for the recommended content adaptation and strategy correction in subsequent cycles, realizing a full-process, full-variable closed-loop adjustment feedback linkage.

[0064] In this implementation plan, by extracting the dynamic changes of recommendation response fluctuation values ​​and adjustment frequencies in multiple interaction cycles, an adjustment sensitivity factor is constructed, and combined with the weight adjustment threshold, adaptive adjustment of the class learning performance weight and the recommendation content complexity weight is achieved, thereby enhancing the adaptability of the recommendation adaptation evaluation value calculation structure in different teaching stages; at the same time, by archiving the recommendation response fluctuation value, response classification label, single adjustment residual, adjustment sensitivity factor and weight update results, a traceable recommendation strategy snapshot is formed, which provides a basis for subsequent recommendation content optimization, thereby realizing closed-loop drive and continuous optimization of the recommendation adjustment chain.

[0065] like Figure 2 As shown, the second aspect of the present invention provides a virtual teaching environment construction and feedback adjustment device based on immersive interaction, including: a class teaching data acquisition and preprocessing module, a teaching content intelligent recommendation and adaptation evaluation module, a recommended content interaction adjustment and deviation analysis module and a response fluctuation analysis and weight adaptive update module, wherein: the class teaching data acquisition and preprocessing module is used to obtain class teaching behavior data through real-time data acquisition and synchronous processing of multiple types of behavior recognition and interactive perception devices in the immersive teaching system, and preprocess the class teaching behavior data to obtain preprocessed class teaching behavior data; the teaching content intelligent recommendation and adaptation evaluation module is used to construct a content label vector and screen a candidate set based on the preprocessed class teaching behavior data and content label vector , conduct a joint quantitative analysis of the adaptation relationship between the recommended content and students, and mark the recommendation status label of each recommended content; the recommended content interaction adjustment and deviation analysis module is used to push the recommendation status label and adaptation evaluation value to the display interface, collect the teacher's interactive feedback to build a recommendation interaction log, and analyze the degree of adaptation deviation of the recommended content in the teaching scenario based on the recommendation interaction log, extract the intervention characteristics and calculate the difference between the recommended adaptation evaluation value before and after adjustment to form a single adjustment residual set; the response fluctuation analysis and weight adaptive update module is used to quantitatively analyze the recommendation response fluctuation intensity and grade the response fluctuation status based on the adjustment residual set and recommendation interaction log, trigger the adaptive adjustment of the weight according to the fluctuation grade, archive the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop.

[0066] In this implementation plan, by constructing a class teaching data collection and preprocessing module, a teaching content intelligent recommendation and adaptation evaluation module, a recommended content interaction adjustment and deviation analysis module, and a response fluctuation analysis and weight adaptive update module, it is possible to efficiently obtain and preprocess class teaching behavior data, combine content label vectors to accurately screen recommended content and perform adaptation analysis, collect teacher adjustment feedback to form a recommendation interaction log, and then evaluate the adaptation deviation of the recommended content and extract adjustment features; at the same time, through quantitative analysis of the recommendation response fluctuation intensity and response grading, the adaptive update of the class learning performance weight and the recommended content complexity weight is achieved, the strategy snapshot is archived, and a complete recommendation adjustment feedback closed loop is constructed, providing strong support for content optimization in the virtual teaching environment.

[0067] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0068] The preferred embodiments of the present invention disclosed above are intended only to help illustrate the present invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the present invention to the specific embodiments described. Obviously, many modifications and variations are possible based on the content of this specification. These embodiments are selected and described in detail in this specification to better explain the principles and practical applications of the present invention, thereby enabling those skilled in the art to better understand and utilize the present invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction, characterized in that: The following steps are involved: S1, through the real-time data collection and synchronous processing of multiple types of behavior recognition and interactive perception devices in the immersive teaching system, obtain class teaching behavior data, and pre-process the class teaching behavior data to obtain pre-processed class teaching behavior data; S2: Construct a content label vector, screen the candidate set based on the pre-processed class teaching behavior data and the content label vector, conduct a joint quantitative analysis of the adaptation relationship between the recommended content and the students, and annotate the recommendation status label of each recommended content; S3: Push the recommendation status label and adaptation evaluation value to the display interface, collect teachers' interactive feedback to build a recommendation interaction log, analyze the degree of adaptation deviation of the recommended content in the teaching scenario based on the recommendation interaction log, extract intervention features, and calculate the difference between the recommendation adaptation evaluation values ​​before and after adjustment to form a single adjustment residual set; S4, based on the adjustment residual set and recommendation interaction log, quantitatively analyzes the intensity of recommendation response fluctuations and grades the response fluctuation status. According to the fluctuation grade, the adaptive adjustment of the weight is triggered. The archived update results generate a strategy snapshot to drive the next recommendation cycle in a closed loop.

2. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps for obtaining class teaching behavior data by real-time data collection and synchronous processing of multiple types of behavior recognition and interaction perception devices in the immersive teaching system are as follows: Class teaching behavior data is obtained through real-time collection and synchronous processing of multiple types of behavior recognition and interactive perception devices in the immersive teaching system; among them, the time difference between the loading and submission time of the question is calculated to obtain the duration of answering a single teaching question; the accuracy rate of the answer is obtained by comparing the student's answer results with the standard answer; the frequency of mouse operation is obtained by analyzing the frequency of mouse clicks, movements and scrolling events in the client log; the average reaction time is obtained by identifying the interval between the appearance of the prompt element and the first input; the mean value of the gaze point offset and the gaze duration are obtained by extracting the gaze coordinate offset and gaze duration collected by the eye tracking device; the duration of classroom questions is obtained by analyzing the start and end time of the audio marked by the speech recognition module; the number of words in the recommended content and the number of recommended charts are obtained by counting the number of text words and image table elements in the recommended content; the class average score is obtained by summarizing the scores of each student in the assessment task.

3. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of preprocessing the class teaching behavior data to obtain the preprocessed class teaching behavior data are as follows: Through the collaborative detection of the median residual detection method based on the sliding window and the local outlier factor algorithm, the mutation values ​​and behavioral abnormal data in the average single-question answering time, gaze duration and average reaction time are identified and eliminated; by utilizing the combined completion mechanism of the time window linear interpolation method and the autoregressive prediction model, the time series completion of the missing fields in the mouse operation frequency, classroom question duration and recommended content word count is performed; through the joint modeling strategy of fusing the Gaussian kernel weighted sliding average and the local polynomial regression algorithm, the class teaching data is smoothed and fitted to suppress high-frequency fluctuations; by adopting the cascade transformation of logarithmic transformation and maximum and minimum normalization algorithm, the class teaching data is rescaled and normalized.

4. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of constructing the content label vector and screening the candidate set based on the pre-processed class teaching behavior data and the content label vector are as follows: Perform text parsing on the recommended content, extract keywords, and map them to corresponding knowledge points and topics in the course knowledge graph. Generate word frequency vectors based on knowledge points and embedding representations based on topics. These two are then concatenated and normalized to form a content label vector. Based on the pre-processed correct answer rate and the class average score, the class knowledge profile vector is aggregated by knowledge point. The cosine similarity between the class knowledge profile vector and each content label vector is calculated. Recommended content with a cosine similarity not less than the knowledge fit threshold is selected as the candidate set. If the number of candidates is lower than the lower limit of the candidate number threshold, the recommended content is supplemented in descending order of cosine similarity until the set size requirement is met. If the number of candidates exceeds the upper threshold of the candidate number, a specified number of recommended contents with the same cosine similarity value will be extracted at fixed intervals to meet the set quantity constraint.

5. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of performing a joint quantitative analysis of the adaptation relationship between the recommended content and the students and labeling the recommended status label of each recommended content are as follows: From the candidate set, we extracted the duration of each recommended content, the accuracy rate, the fixation duration, the class average score, the number of words in the recommended content, and the number of recommended graphics. We then conducted a joint quantitative analysis of the compatibility between the recommended content and the students: we calculated the ratio of the class average score and the accuracy rate divided by the duration of each teaching question plus 1. We also calculated the logarithm of the fixation duration plus 1, added this ratio to the logarithm, and multiplied it by the class learning performance weight. The product was used as the learning performance item. Calculate the square root of the number of words in the recommended content, calculate the square value of the number of recommended graphics, add the square root to the square value, and multiply it by the complexity weight of the recommended content. The product is the content complexity item; Calculate the absolute value of the difference between the learning performance item and the content complexity item to obtain the recommended adaptation evaluation value; Compare the recommendation adaptation evaluation value of each recommendation content in the candidate set with the recommendation evaluation threshold, which includes the first-level recommendation evaluation threshold and the second-level recommendation evaluation threshold; If the recommended adaptation evaluation value is less than or equal to the first-level recommendation evaluation threshold, the recommended content is marked as a low adaptation state; if the recommended adaptation evaluation value is greater than the first-level recommendation evaluation threshold and less than the second-level recommendation evaluation threshold, the recommended content is marked as a medium adaptation state; if the evaluation value is greater than or equal to the second-level recommendation evaluation threshold, the recommended content is marked as a high adaptation state; the recommendation status label and the recommended adaptation evaluation value are appended to the candidate set as the basis for subsequent manual interaction and selection strategy.

6. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of pushing the recommendation status label and the adaptation evaluation value to the display interface and collecting the teacher's interactive feedback to build the recommendation interaction log are as follows: Push the recommendation status label to the candidate content display interface, attach the recommendation adaptation evaluation value, learning performance item and content complexity item of each recommended content, and list the corresponding teaching question answering time, answer accuracy rate, fixation duration, recommended content word count and recommended chart number of the recommended content, and display them in a structured manner according to the field number; The interface provides three adjustment options: accept, replace, and fine-tune. Teachers can provide interactive feedback on specific recommended content based on actual teaching needs. If fine-tuning is selected, teachers can further set the adjustment ratio of learning performance and content complexity to temporarily modify the weight of the current recommended content. Count the cumulative number of replacement and fine-tuning adjustment operations in the current interaction cycle to obtain the recommended adjustment frequency; Calculate the total number of times a field appears in all recommended content in the current cycle and divide it by the number of recommended content that includes the field to obtain the average number of times the field is recommended. A corresponding recommendation interaction log is generated for each teacher's operation, which includes the frequency of recommendation adjustment, the average number of field recommendations, the recommendation status label and class teaching data.

7. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of analyzing the degree of adaptation deviation of the recommended content in the teaching scenario based on the recommendation interaction log, extracting the intervention features, and calculating the difference between the recommendation adaptation evaluation values ​​before and after adjustment to form a single adjustment residual set are as follows: After the interaction cycle ends, a comprehensive analysis is conducted based on the recommendation interaction log to determine the degree of deviation of the recommended content with replacement and fine-tuning adjustment operations in actual teaching scenarios: the product of the class average score and the correct answer rate is calculated, the square of the time taken to answer a single question in the teaching session plus 1 is calculated, and this product is divided by the square to obtain the learning effectiveness ratio item; the logarithm of the natural logarithm of the gaze duration plus 1 is calculated, the square root of the number of words in the recommended content plus 1 is calculated, and this logarithm is divided by the square root to obtain the visual input adjustment item; the square of the number of recommended charts is calculated, and this square is divided by the mouse operation frequency plus 1 to obtain the chart load index item; the square of the adjustment frequency is calculated, and this square is divided by the average number of recommendations plus 1 to obtain the manual intervention item; the learning effectiveness ratio item is added to the visual input adjustment item, and the chart load index item is subtracted. The absolute value of this difference is multiplied by the manual intervention item to obtain the recommendation deviation assessment value; Based on all the recommended content that has been fine-tuned, the recommendation deviation evaluation values ​​before and after adjustment are extracted and the difference between the two is calculated as the single adjustment residual of each recommended content, and the adjustment residual set of the current interaction cycle is summarized.

8. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of quantitatively analyzing the intensity of recommendation response fluctuations and grading the response fluctuation states based on the adjusted residual set and recommendation interaction log are as follows: Based on the adjustment residual set and recommendation interaction log, the response intensity of the recommended content in the current teaching cycle to the interaction between the class learning status and the teacher's adjustment behavior is measured; the product of the class average score and the correct answer rate is calculated, and the square of the teaching time for answering a single question plus 1 is calculated, and this product is divided by the square value to obtain the learning load factor; the product of the number of words in the recommended content and the number of recommended charts is calculated, plus 1, and then the natural logarithm is taken to obtain the structural complexity factor; the mouse operation frequency of the recommended content is added by 1 and the natural logarithm is taken, and this logarithm is multiplied by the single adjustment residual to obtain the behavioral response factor; the learning load factor is added to the structural complexity factor, and then multiplied by the behavioral response factor to obtain the recommendation response fluctuation value; After the recommended response fluctuation value is calculated, the recommended response fluctuation value is compared with the response fluctuation threshold in real time. The response fluctuation threshold includes the first-level response fluctuation threshold and the second-level response fluctuation threshold; When the response fluctuation threshold is less than or equal to the first-level response fluctuation threshold, the current recommended content is marked as low-response fluctuation content and the normal adjustment frequency is maintained; When the response fluctuation threshold is greater than the first-level response fluctuation threshold and less than the second-level response fluctuation threshold, the current recommended content is marked as a medium response fluctuation state and automatically included in the short-term key observation queue; When the response fluctuation threshold is greater than or equal to the secondary response fluctuation threshold, the current recommended content is marked as a high response fluctuation state and automatically included in the high-sensitivity review list for the next round of recommended content, and its manual intervention and review priority is increased; The fluctuation state marker is attached to the metadata of the recommended content to form a dynamic adjustment index for the teaching adaptation feedback chain.

9. The method for constructing a virtual teaching environment and adjusting feedback based on immersive interaction according to claim 1, characterized in that: The specific steps of adaptively adjusting the weights based on the volatility classification trigger and archiving the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop are as follows: Based on the numerical distribution of recommendation response fluctuation values, we identify samples with continuously increasing fluctuations in medium- and high-response recommendations. We extract the recommendation response fluctuation value sequence for two or more consecutive interaction cycles to determine whether it meets the monotonically increasing trend. At the same time, we extract the adjustment frequency sequence of the corresponding recommended content in each cycle and calculate the change amplitude of the adjustment density. If the medium-to-high-response recommendation content has an upward trend in both the recommendation response fluctuation value and the adjustment frequency, the adjustment sensitivity factor is constructed by the weighted average of the two change amplitudes; When the adjustment sensitivity factor is higher than the weight adjustment threshold, the weight update mechanism is automatically triggered, and the weight of class learning performance and the weight of recommended content complexity are smoothly updated using a sliding weighted average method. At the same time, the calculation structure parameters of the recommended adaptation evaluation value in the next interaction cycle are reconstructed; The recommendation response fluctuation value, response classification label, single adjustment residual, adjustment sensitivity factor and weight update results in the current cycle are archived and saved to generate a recommendation strategy snapshot, which serves as the input basis for the subsequent cycle recommendation content adaptation and strategy correction, realizing a complete adjustment feedback loop.

10. A virtual teaching environment construction and feedback adjustment device based on immersive interaction, characterized by: It includes a class teaching data collection and preprocessing module, a teaching content intelligent recommendation and adaptation evaluation module, a recommended content interactive adjustment and deviation analysis module, and a response fluctuation analysis and weight adaptive update module, among which: The class teaching data collection and preprocessing module is used to obtain class teaching behavior data through real-time data collection and synchronous processing of multiple types of behavior recognition and interactive perception devices in the immersive teaching system, and preprocess the class teaching behavior data to obtain preprocessed class teaching behavior data; The teaching content intelligent recommendation and adaptation evaluation module is used to construct a content label vector, screen a candidate set based on the pre-processed class teaching behavior data and the content label vector, conduct a joint quantitative analysis of the adaptation relationship between the recommended content and the students, and annotate the recommendation status label of each recommended content; The recommended content interaction adjustment and deviation analysis module is used to push the recommendation status label and adaptation evaluation value to the display interface, collect teachers' interactive feedback to build a recommendation interaction log, analyze the degree of adaptation deviation of the recommended content in the teaching scenario based on the recommendation interaction log, extract intervention features and calculate the difference between the recommendation adaptation evaluation values ​​before and after adjustment to form a single adjustment residual set; The response fluctuation analysis and weight adaptive update module is used to quantitatively analyze the recommendation response fluctuation intensity and classify the response fluctuation status based on the adjustment residual set and recommendation interaction log, trigger the adaptive adjustment of the weight according to the fluctuation classification, and archive the update results to generate a strategy snapshot to drive the next recommendation cycle in a closed loop.

Images