An AI-based education tutoring interaction system

Abstract

This invention belongs to the field of teaching interaction technology, specifically referring to an AI-based educational tutoring interaction system. The system includes an online education platform module, an active Q&A decision-making module, a tiered Q&A prompt triggering module, a visual sensor, a pressure sensor, an infrared thermal imaging sensor, a haptic interaction module, and a dynamic correction module for the Q&A process. The haptic interaction module consists of a haptic feedback device and indicator lights. This solution mines students' historical mastery of knowledge points, initiating high-power state judgment only when the problem-solving time exceeds a threshold, achieving tiered monitoring of students' problem-solving status. This transforms passive Q&A into data-driven proactive Q&A demand prediction; dynamically optimizing the Q&A progress, updating students' historical mastery of knowledge points, and using Q&A process data as training samples to iteratively optimize the prediction model, ultimately achieving cyclical, proactive, and personalized educational tutoring on a per-question basis, significantly improving the targeting and effectiveness of online educational tutoring.

Description

Technical Field

[0001] This invention relates to the field of interactive teaching technology, specifically to an AI-based interactive educational tutoring system. Background Technology

[0002] An AI-based interactive education tutoring system is a system that uses artificial intelligence technology to automatically perceive, predict, and respond to educational tutoring needs.

[0003] Among existing similar solutions, such as CN116740998B, a remote teaching system, method, and device for information interaction, this solution addresses the technical problems of inflexible pop-up settings in existing remote teaching, the inability to help students build a logical knowledge system by simply providing answers, and the difficulty in real-time monitoring of students' learning status and providing targeted tutoring, resulting in low learning efficiency. This solution generates personalized teaching pop-up windows by combining data on students' actual learning status and teacher evaluation data, collects data on students' facial expressions and voice in real time to identify abnormal learning statuses and asks questions via pop-ups, extracts knowledge points and matches them with relevant videos for playback when answers are incorrect, pushes advanced questions and provides feedback on students' difficulties to teachers, and combines keywords, audio, and handwritten notes to generate remote teaching videos. This achieves personalized pop-up settings, accurate monitoring of learning status, helps students build a logical knowledge system, realizes efficient information interaction between teachers and students, and significantly improves the efficiency of remote learning. However, existing online education and tutoring still suffer from technical problems such as passively waiting for students to initiate question-and-answer requests, the inability to proactively and accurately predict individual question-and-answer needs, and the lack of power consumption tiering design for monitoring students' problem-solving status, resulting in serious waste of computing resources.

[0004] Furthermore, for example, CN118675381B describes an adjustable desk-based intelligent interactive teaching platform. This solution addresses the issues in existing online teaching where teachers' teaching focus and students' periods of high concentration are easily misaligned, leading to student fatigue and the lack of effective adjustment capabilities of smart desks. It also addresses the technical problems of low accuracy and difficulty in real-time adjustment of student learning status using traditional image recognition monitoring. By building a multi-module intelligent interactive teaching platform, it utilizes force sensors to detect desk pressure and assess student learning status in layers. It sets teaching focus nodes and statistically analyzes the difference rate to determine learning fatigue. Mechanical devices adjust the desk's tilt angle to alleviate fatigue. By using a prompting module to remind students of their attention and resetting the difference rate for recalculation after adjustment, this solution achieves more accurate monitoring of students' concentration and real-time assessment of learning fatigue. It effectively alleviates learning fatigue by adjusting posture, matches teaching focus with students' concentration, and improves online learning efficiency. It also enhances the technical effects of improving teaching interaction monitoring capabilities and the platform's user-friendliness. However, this solution still has technical problems such as low accuracy in judging students' Q&A needs, Q&A prompts easily interfering with students' problem-solving process, the inability to dynamically adjust the Q&A process according to students' real-time status, lack of Q&A data feedback and model optimization mechanisms, and difficulty in achieving personalized cyclical tutoring. Summary of the Invention

[0005] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an AI-based interactive education and tutoring system. Addressing the problems in current online education and tutoring, such as passively waiting for students to initiate question-and-answer requests, the inability to proactively and accurately predict individual question-and-answer needs, and the lack of power consumption tiering for monitoring student problem-solving status, resulting in significant waste of computing resources, this solution establishes a structured framework map of knowledge points for each grade level, mines students' historical mastery of knowledge points, and uses a timer for low-power dynamic monitoring. High-power status analysis is only initiated when problem-solving time exceeds a threshold, achieving tiered monitoring of students' problem-solving status and rationally allocating computing resources. This transforms passive question-and-answering into data-driven proactive prediction of question-and-answer needs, laying the foundation for accurate determination of these needs. Furthermore, it addresses the low accuracy of student question-and-answer need assessment, the tendency of question prompts to interfere with the student's problem-solving process, and the inability to dynamically adjust the question-and-answer process based on the student's real-time status. The solution addresses the technical challenges of lacking feedback mechanisms for Q&A data and model optimization, hindering personalized, cyclical tutoring. During the high-power state assessment phase, this solution utilizes visual, pressure, and infrared thermal imaging technologies to collect multi-dimensional data on students' problem-solving operations, emotions, and postures. This data is input into a predictive model trained on students' historical error sets to intelligently determine the type of Q&A need. Subtle tactile vibrations on the desktop and soft indicator lights provide non-intrusive, tiered Q&A prompts, along with a pop-up Q&A entry on the interface. This dynamically optimizes the Q&A progress, updates students' historical knowledge mastery, and uses Q&A process data as training samples to iteratively optimize the predictive model. This achieves accurate and intelligent determination of Q&A needs, making Q&A prompts both user-friendly and targeted. Furthermore, it enables dynamic adaptation and model self-learning during the Q&A process, ultimately achieving cyclical, proactive, and personalized educational tutoring on a per-question basis, significantly improving the relevance and effectiveness of online educational tutoring.

[0006] The technical solution adopted by the present invention is as follows: The present invention provides an AI-based educational tutoring interactive system, which includes an online education platform module, an active Q&A decision module, a graded Q&A prompt trigger module, a visual sensor, a pressure sensor, an infrared thermal imaging sensor, a somatosensory interaction module, and a Q&A process dynamic correction module. The somatosensory interaction module consists of a haptic feedback device and an indicator light.

[0007] The online education platform module uses online education analysis methods to explore students' historical mastery of various knowledge points. The online education platform module is equipped with a timer. When students perform homework tasks on the online education platform, the timer dynamically monitors students' single-question Q&A needs with low power consumption.

[0008] The proactive question-answering decision module uses a proactive question-answering decision method to perform multimodal fusion analysis on single question-answering needs, generate question-answering prompt hierarchical instructions, and transmit the question-answering prompt hierarchical instructions to the hierarchical question-answering prompt trigger module;

[0009] The tiered Q&A prompt trigger module emits a slight vibration through the haptic feedback device integrated on the desktop, guiding the indicator light to emit a soft light, thereby achieving non-intrusive proactive Q&A prompts, and simultaneously linking the online education platform interface to pop up the tiered Q&A entry.

[0010] The dynamic correction module for the Q&A process uses a dynamic correction method to optimize the Q&A progress.

[0011] Furthermore, the online education analysis method specifically includes the following steps:

[0012] Step S1: Knowledge point modeling, which is used to establish a basis for predicting the demand for answering questions in a single question. Specifically, a structured framework map of the student's grade is established. The structured framework map is composed of knowledge points. The student's current homework is scanned, and the homework reference answer is analyzed using natural language processing technology. The knowledge points tested in each question in the homework are bound to the knowledge points in the structured framework map.

[0013] Step S2: Mastery inference, specifically, collecting students' historical wrong questions, establishing a wrong question set, and using knowledge point graph matching technology to bind the knowledge points tested by each historical wrong question with the knowledge points in the structured framework graph, clarifying the core knowledge points, related knowledge points and knowledge point levels corresponding to the wrong questions in the structured framework graph, and exploring the students' historical mastery of each knowledge point;

[0014] Step S3: Low power dynamic monitoring, specifically, based on the historical mastery of the knowledge points tested by the current target question, a benchmark problem-solving time threshold is preset for the target question. When the timer of the online education platform detects that the student's problem-solving time exceeds the benchmark problem-solving time threshold, step S4 high power state judgment is triggered.

[0015] Step S4: High power consumption state assessment, specifically, activating the visual sensor to collect first-order operation anomaly data and second-order state anomaly data in real time when the student is doing the current target question. The first-order operation anomaly data includes the frequency of question selection and the touch trajectory of writing on the draft paper. The second-order state anomaly data includes facial orientation and gaze focus state. Based on the second-order state anomaly data, the student's facial expressions are analyzed through an emotion recognition algorithm to generate emotion state labels.

[0016] The pressure sensor synchronously collects the changes in the tactile pressure of the student's hands resting on the desk, and the infrared thermal imaging sensor performs real-time thermal imaging contour capture and dynamic trajectory analysis of the student to obtain student activity posture data.

[0017] The system transmits historical mastery data, first-order operational anomaly data, emotional state labels, and student activity posture data to the proactive Q&A decision-making module.

[0018] Furthermore, the proactive question-answering decision-making method specifically includes the following steps:

[0019] Step R1: Question-answering demand prediction. Specifically, feature-level fusion technology and XGBoost multi-class prediction model are used to perform structured fusion and intelligent judgment on the data transmitted by the online education platform module. The XGBoost multi-class prediction model is referred to as the prediction model. First, the prediction model is trained using the student's historical wrong question set. Second, the historical mastery, first-order operation abnormal data, emotional state labels and student activity posture data are quantified as features. Finally, they are integrated into a unified dimension of student problem-solving state feature vector. The student problem-solving state feature vector is used as the input of the prediction model. The prediction model performs intelligent reasoning and judgment on the type of student's question-answering demand and outputs the judgment result.

[0020] Step R2: The prediction model's judgment result is synchronized into a Q&A prompt grading instruction signal, which is then transmitted to the grading Q&A prompt trigger module.

[0021] Furthermore, the dynamic correction method for the question-answering process specifically includes the following steps:

[0022] Step T1: Real-time monitoring of the Q&A process. Specifically, after a student clicks the tiered Q&A entry on the screen to trigger the Q&A operation, the dynamic correction module for the Q&A process is immediately activated, the visual sensor is turned on again, and the student's attention data during listening and reading Q&A is continuously collected and analyzed in real time. At the same time, the student's facial expressions are analyzed through the emotion recognition algorithm, the emotion state label is updated, the pressure sensor, infrared thermal imaging sensor and body interaction module are turned off, and the online education platform module collects the student's interaction data on the Q&A interface simultaneously to optimize the Q&A progress.

[0023] Step T2: Q&A data feedback and model optimization. Specifically, record the Q&A process data, update the students' historical mastery of each knowledge point, use the Q&A process data as training samples, and iteratively optimize the prediction model of the active Q&A decision module to achieve model self-learning.

[0024] Step T3: Cyclic proactive Q&A interaction. Specifically, after a single question is answered, the online education platform module immediately switches to low-power monitoring mode and repeats the entire process of answering a single question for each subsequent target question for the student.

[0025] The beneficial effects achieved by the present invention using the above solution are as follows:

[0026] (1) In response to the technical problems in existing online education and tutoring, such as passively waiting for students to initiate question-answering requests, being unable to actively and accurately predict the question-answering needs of individual questions, and lacking a power consumption classification design for monitoring students' problem-solving status, resulting in serious waste of computing resources, this solution establishes a structured framework map of knowledge points for the corresponding grade, explores the historical mastery of students' knowledge points, uses a timer for low-power dynamic monitoring, and only initiates high-power status judgment when the problem-solving time exceeds the threshold, thereby realizing the classification monitoring of students' problem-solving status, rationally allocating computing resources, and transforming from passive question-answering to data-based proactive question-answering demand prediction, laying the foundation for accurate judgment of question-answering needs;

[0027] (2) In response to the technical problems of low accuracy in judging students' question-answering needs, easy interference of question-answering prompts with students' problem-solving process, inability to dynamically adjust the question-answering process according to students' real-time status, lack of question-answering data feedback and model optimization mechanism, and difficulty in realizing personalized cyclical tutoring, this solution collects multi-dimensional data of students' problem-solving operations, emotions, and postures through visual, pressure, and infrared thermal imaging technologies during the high power consumption state judgment stage. The data is input into the prediction model trained by the students' historical wrong question set to intelligently determine the type of question-answering needs. The solution achieves non-interference graded question-answering prompts through slight tactile vibration of the desktop and soft light prompts of indicator lights, and links the interface to pop up corresponding question-answering entry points, dynamically optimizes the question-answering progress, updates the students' historical mastery of knowledge points, and uses the question-answering process data as training samples to iteratively optimize the prediction model, thereby achieving accurate and intelligent judgment of question-answering needs. This makes the question-answering prompts both friendly and targeted, while realizing dynamic adaptation of the question-answering process and model self-learning. Finally, it achieves cyclical proactive personalized education and tutoring based on single questions, greatly improving the targeting and effectiveness of online education and tutoring. Attached Figure Description

[0028] Figure 1 This invention provides a module connection diagram for an AI-based interactive educational tutoring system.

[0029] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0030] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] Example 1: See Figure 1This embodiment provides an AI-based educational tutoring interaction system, which includes an online education platform module, an active Q&A decision module, a tiered Q&A prompt trigger module, a visual sensor, a pressure sensor, an infrared thermal imaging sensor, a haptic interaction module, and a dynamic correction module for the Q&A process. The haptic interaction module consists of a haptic feedback device and an indicator light.

[0032] The online education platform module uses online education analysis methods to explore students' historical mastery of various knowledge points. The online education platform module is equipped with a timer. When students perform homework tasks on the online education platform, the timer dynamically monitors students' single-question Q&A needs with low power consumption.

[0033] The proactive question-answering decision module uses a proactive question-answering decision method to perform multimodal fusion analysis on single question-answering needs, generate question-answering prompt hierarchical instructions, and transmit the question-answering prompt hierarchical instructions to the hierarchical question-answering prompt trigger module;

[0034] The tiered Q&A prompt trigger module emits a slight vibration through the haptic feedback device integrated on the desktop, guiding the indicator light to emit a soft light, thereby achieving non-intrusive proactive Q&A prompts, and simultaneously linking the online education platform interface to pop up the tiered Q&A entry.

[0035] The dynamic correction module for the Q&A process uses a dynamic correction method to optimize the Q&A progress.

[0036] Example 2: See Figure 1 This embodiment is based on the above embodiment, and the online education analysis method specifically includes the following steps:

[0037] Step S1: Knowledge point modeling, which is used to establish a basis for predicting the demand for answering questions in a single question. Specifically, a structured framework map of the student's grade is established. The structured framework map is composed of knowledge points. The student's current homework is scanned, and the homework reference answer is analyzed using natural language processing technology. The knowledge points tested in each question in the homework are bound to the knowledge points in the structured framework map.

[0038] Step S2: Mastery inference, specifically, collecting students' historical wrong questions, establishing a wrong question set, and using knowledge point graph matching technology to bind the knowledge points tested by each historical wrong question with the knowledge points in the structured framework graph, clarifying the core knowledge points, related knowledge points and knowledge point levels corresponding to the wrong questions in the structured framework graph, and exploring the students' historical mastery of each knowledge point;

[0039] Step S3: Low power dynamic monitoring, specifically, based on the historical mastery of the knowledge points tested by the current target question, a benchmark problem-solving time threshold is preset for the target question. When the timer of the online education platform detects that the student's problem-solving time exceeds the benchmark problem-solving time threshold, step S4 high power state judgment is triggered.

[0040] Step S4: High power consumption state assessment, specifically, activating the visual sensor to collect first-order operation anomaly data and second-order state anomaly data in real time when the student is doing the current target question. The first-order operation anomaly data includes the frequency of question selection and the touch trajectory of writing on the draft paper. The second-order state anomaly data includes facial orientation and gaze focus state. Based on the second-order state anomaly data, the student's facial expressions are analyzed through an emotion recognition algorithm to generate emotion state labels.

[0041] The pressure sensor synchronously collects the changes in the tactile pressure of the student's hands resting on the desk, and the infrared thermal imaging sensor performs real-time thermal imaging contour capture and dynamic trajectory analysis of the student to obtain student activity posture data.

[0042] The system transmits historical mastery data, first-order operational anomaly data, emotional state labels, and student activity posture data to the proactive Q&A decision-making module.

[0043] Example 3: See Figure 1 This embodiment is based on the above embodiment, and the proactive question-answering decision method specifically includes the following steps:

[0044] Step R1: Question-answering demand prediction. Specifically, feature-level fusion technology and XGBoost multi-class prediction model are used to perform structured fusion and intelligent judgment on the data transmitted by the online education platform module. The XGBoost multi-class prediction model is referred to as the prediction model. First, the prediction model is trained using the student's historical wrong question set. Second, the historical mastery, first-order operation abnormal data, emotional state labels and student activity posture data are quantified as features. Finally, they are integrated into a unified dimension of student problem-solving state feature vector. The student problem-solving state feature vector is used as the input of the prediction model. The prediction model performs intelligent reasoning and judgment on the type of student's question-answering demand and outputs the judgment result.

[0045] Step R2: The prediction model's judgment result is synchronized into a Q&A prompt grading instruction signal, which is then transmitted to the grading Q&A prompt trigger module.

[0046] Example 4: See Figure 1 This embodiment is based on the above embodiment, and the dynamic correction method for the question-answering process specifically includes the following steps:

[0047] Step T1: Real-time monitoring of the Q&A process. Specifically, after a student clicks the tiered Q&A entry on the screen to trigger the Q&A operation, the dynamic correction module for the Q&A process is immediately activated, the visual sensor is turned on again, and the student's attention data during listening and reading Q&A is continuously collected and analyzed in real time. At the same time, the student's facial expressions are analyzed through the emotion recognition algorithm, the emotion state label is updated, the pressure sensor, infrared thermal imaging sensor and body interaction module are turned off, and the online education platform module collects the student's interaction data on the Q&A interface simultaneously to optimize the Q&A progress.

[0048] Step T2: Q&A data feedback and model optimization. Specifically, record the Q&A process data, update the students' historical mastery of each knowledge point, use the Q&A process data as training samples, and iteratively optimize the prediction model of the active Q&A decision module to achieve model self-learning.

[0049] Step T3: Cyclic proactive Q&A interaction. Specifically, after a single question is answered, the online education platform module immediately switches to low-power monitoring mode and repeats the entire process of answering a single question for each subsequent target question for the student.

[0050] Example 5: See Figure 1 This embodiment is based on the above embodiment. The online education platform interface pops up a tiered Q&A entry. Specifically, the lightweight Q&A entry corresponds to targeted tips on problem-solving strategies, while the in-depth Q&A entry corresponds to a review of core knowledge points and step-by-step guidance on problem-solving steps. Students can respond to the Q&A prompts by lifting their hands off the desktop or using a designated area on the touch terminal.

[0051] Example 6: See Figure 1 This embodiment is based on the above embodiment. In step R1, the feature quantization refers to converting the first-order operation anomaly data into numerical feature A. Numerical feature A specifically includes the proportion of time exceeding the threshold, the decrease rate of question selection frequency, and the number of touch trajectory interruptions. The emotional state label is used as a one-hot encoding, and the student activity posture data is converted into numerical feature B. Numerical feature B specifically includes stress fluctuation value and body activity frequency. The historical mastery is converted into mastery scores of core knowledge points, related knowledge points, and knowledge point levels. Finally, feature vectors are concatenated through feature-level fusion technology.

[0052] Example 7: See Figure 1 This embodiment is based on the above embodiment. In step R1, the output determination result specifically includes the following steps:

[0053] Step R11: If the judgment result is that the mastery score of the corresponding knowledge point is not up to standard and there is no effective attempt trajectory in the problem-solving behavior, it is judged that the corresponding knowledge point is not mastered, and in-depth Q&A prompts are triggered directly.

[0054] Step R12: The second judgment result is that the knowledge point mastery score is in the fuzzy range, the problem-solving behavior is abnormal, and the emotional and physical state matches the characteristics of distraction, irritability, and operation stagnation. It is judged as a problem-solving stagnation, distraction and irritability, triggering a lightweight Q&A prompt.

[0055] Step R13: The third judgment result is that the knowledge point mastery score meets the standard, the problem-solving time exceeds the benchmark problem-solving time threshold but the emotional focus meets the standard, and the operation is normal. It is judged as slightly longer time but focused state. The Q&A prompt will not be triggered for the time being. The online education platform will continue to perform low power dynamic monitoring in step S3.

[0056] Example 8: See Figure 1 This embodiment is based on the above embodiment. In step T1, the focus data specifically includes eye movement trajectory, gaze focus and blink frequency.

[0057] Example 9: See Figure 1 This embodiment is based on the above embodiment. In step T1, the interactive data specifically includes fast forward, rewind, and pause.

[0058] Example 10: See Figure 1 This embodiment is based on the above embodiment. In step T1, the optimization of the Q&A progress specifically includes:

[0059] If the system detects that a student's gaze is off the Q&A interface or that the Q&A content is frequently fast-forwarded, it is determined that the student has understood the current Q&A content and will automatically skip the subsequent basic explanations and go directly to the key steps of solving the problem.

[0060] If a student's face shows confusion and the same Q&A session is replayed multiple times, it is determined that the student still does not understand. The video clips of basic knowledge points bound to the structured framework diagram in the knowledge point modeling process are immediately retrieved and seamlessly added to the current Q&A content. At the same time, the speaking speed of the Q&A explanation is reduced.

[0061] Example 11: See Figure 1 This embodiment is based on the above embodiment. In step T2, the data of this Q&A process specifically includes the reason for triggering the Q&A request, the classification type of the Q&A prompt, the dynamic correction node of the Q&A process, the student's emotional and concentration recovery status after the Q&A, the final correct answer rate of the question, and the change in the answering time.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0064] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An AI-based interactive educational tutoring system, characterized in that, It includes an online education platform module, an active Q&A decision module, a tiered Q&A prompt trigger module, a visual sensor, a pressure sensor, an infrared thermal imaging sensor, a haptic interaction module, and a dynamic correction module for the Q&A process. The haptic interaction module consists of a tactile feedback device and an indicator light. The online education platform module uses online education analysis methods to explore students' historical mastery of various knowledge points. The online education platform module is equipped with a timer. When students perform homework tasks on the online education platform, the timer dynamically monitors students' single-question Q&A needs with low power consumption. The proactive question-answering decision module uses a proactive question-answering decision method to perform multimodal fusion analysis on single question-answering needs, generate question-answering prompt hierarchical instructions, and transmit the question-answering prompt hierarchical instructions to the hierarchical question-answering prompt trigger module; The tiered Q&A prompt trigger module emits a slight vibration through the haptic feedback device integrated on the desktop, guiding the indicator light to emit a soft light, thereby achieving non-intrusive proactive Q&A prompts, and simultaneously linking the online education platform interface to pop up the tiered Q&A entry. The dynamic correction module for the Q&A process uses a dynamic correction method to optimize the Q&A progress.

2. The AI-based interactive educational tutoring system according to claim 1, characterized in that, The online education analysis method specifically includes the following steps: Step S1: Knowledge point modeling, which is used to establish a basis for predicting the demand for answering questions in a single question. Specifically, a structured framework map of the student's grade is established. The structured framework map is composed of knowledge points. The student's current homework is scanned, and the homework reference answer is analyzed using natural language processing technology. The knowledge points tested in each question in the homework are bound to the knowledge points in the structured framework map. Step S2: Mastery inference, specifically, collecting students' historical wrong questions, establishing a wrong question set, and using knowledge point graph matching technology to bind the knowledge points tested by each historical wrong question with the knowledge points in the structured framework graph, clarifying the core knowledge points, related knowledge points and knowledge point levels corresponding to the wrong questions in the structured framework graph, and exploring the students' historical mastery of each knowledge point; Step S3: Low power dynamic monitoring, specifically, based on the historical mastery of the knowledge points tested by the current target question, a benchmark problem-solving time threshold is preset for the target question. When the timer of the online education platform detects that the student's problem-solving time exceeds the benchmark problem-solving time threshold, step S4 high power state judgment is triggered. Step S4: High power consumption state assessment, specifically, activating the visual sensor to collect first-order operation anomaly data and second-order state anomaly data in real time when the student is doing the current target question. The first-order operation anomaly data includes the frequency of question selection and the touch trajectory of writing on the draft paper. The second-order state anomaly data includes facial orientation and gaze focus state. Based on the second-order state anomaly data, the student's facial expressions are analyzed through an emotion recognition algorithm to generate emotion state labels. The pressure sensor synchronously collects the changes in the tactile pressure of the student's hands resting on the desk, and the infrared thermal imaging sensor performs real-time thermal imaging contour capture and dynamic trajectory analysis of the student to obtain student activity posture data. The system transmits historical mastery data, first-order operational anomaly data, emotional state labels, and student activity posture data to the proactive Q&A decision-making module.

3. The AI-based interactive educational tutoring system according to claim 2, characterized in that, The proactive question-answering decision-making method specifically includes the following steps: Step R1: Question-answering demand prediction. Specifically, feature-level fusion technology and XGBoost multi-class prediction model are used to perform structured fusion and intelligent judgment on the data transmitted by the online education platform module. The XGBoost multi-class prediction model is referred to as the prediction model. First, the prediction model is trained using the student's historical wrong question set. Second, the historical mastery, first-order operation abnormal data, emotional state labels and student activity posture data are quantified as features. Finally, they are integrated into a unified dimension of student problem-solving state feature vector. The student problem-solving state feature vector is used as the input of the prediction model. The prediction model performs intelligent reasoning and judgment on the type of student's question-answering demand and outputs the judgment result. Step R2: The prediction model's judgment result is synchronized into a Q&A prompt grading instruction signal, which is then transmitted to the grading Q&A prompt trigger module.

4. The AI-based interactive educational tutoring system according to claim 3, characterized in that, The dynamic correction method for the Q&A process specifically includes the following steps: Step T1: Real-time monitoring of the Q&A process. Specifically, after a student clicks the tiered Q&A entry on the screen to trigger the Q&A operation, the dynamic correction module for the Q&A process is immediately activated, the visual sensor is turned on again, and the student's attention data during listening and reading Q&A is continuously collected and analyzed in real time. At the same time, the student's facial expressions are analyzed through the emotion recognition algorithm, the emotion state label is updated, the pressure sensor, infrared thermal imaging sensor and body interaction module are turned off, and the online education platform module collects the student's interaction data on the Q&A interface simultaneously to optimize the Q&A progress. Step T2: Q&A data feedback and model optimization, specifically, recording the Q&A process data, updating the students' historical mastery of each knowledge point, using the Q&A process data as training samples, and iteratively optimizing the prediction model of the active Q&A decision module; Step T3: Cyclic proactive Q&A interaction. Specifically, after a single question is answered, the online education platform module immediately switches to low-power monitoring mode and repeats the entire process of answering a single question for each subsequent target question for the student.

Images