Artificial intelligence adaptive interactive teaching system

Abstract

The invention relates to the technical field of interactive learning platforms, and discloses an artificial intelligence self-adaptive interactive teaching system, which comprises the following parts: (1) a multi-mode sensing layer; (2) an AI analysis layer; (3) an adaptive decision layer; (4) a content generation layer; and (5) an interactive execution layer. According to the artificial intelligence adaptive interactive teaching system, an AI analysis layer performs real-time modeling according to a cognitive state, calculates a knowledge attenuation factor lambda = 0.02 / s through a KGTD model, dynamically tracks a knowledge mastery degree curve, and solves the problem of a fixed progress bar; the self-adaptive decision-making layer generates different strategies according to personal habits of different learners, teaching paradigms (abstract concept-> 3D modeling / concrete knowledge-> AR operation) are automatically switched based on learning style vectors (Kolb scale), the course completion rate is improved by 55% compared with MOOC, the content generation layer varies the difficulty coefficient (delta D = + / -0.3) of questions by applying a genetic algorithm, and the content generation layer is used for generating a learning model. And the exercise difficulty is matched with the cognitive level in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of online learning platforms, and in particular to an artificial intelligence adaptive interactive teaching system. Background Art

[0002] With the development of information technology, online education and educational software systems have become increasingly popular. Traditional teaching models are unable to meet the personalized needs of learners. Early online learning platforms (such as LMS systems) focused primarily on the digital dissemination and management of learning content, lacking effective interaction and intelligent feedback mechanisms. To address this personalized problem, adaptive learning systems (ALSs) emerged.

[0003] Traditional online education platforms (such as MOOCs and LMSs) use a fixed course structure and are unable to dynamically adjust content based on individual cognitive levels and learning pace. In addition, recorded courses and simple exercise systems lack real-time feedback mechanisms, which can easily reduce learners' participation due to one-way indoctrination. Insufficient data utilization: Although answer data can be recorded, there is a lack of in-depth exploration of behavioral patterns (such as hesitation duration and error trajectories), resulting in low accuracy of personalized recommendations. Therefore, it is necessary to propose an artificial intelligence adaptive interactive teaching system to solve the above problems. Summary of the Invention

[0004] (1) Technical problems solved

[0005] In response to the shortcomings of the existing technology, the present invention provides an artificial intelligence adaptive interactive teaching system, which has the advantages of being able to analyze the learner's status in real time and dynamically optimize the teaching strategy, solving the problem in the background technology that it is unable to analyze the learner's status in real time and dynamically optimize the teaching strategy.

[0006] (2) Technical solution

[0007] To achieve the above objectives, the present invention provides the following technical solutions: an artificial intelligence adaptive interactive teaching system, comprising the following:

[0008] (1) Multimodal perception layer: used to holographically capture the learner’s state;

[0009] The multimodal perception layer includes the following:

[0010] The multimodal perception module is used to collect learners' biological behavior data, operational behavior data, and environmental data. The physiological sensing unit includes a visual collector (camera), a voice collector (microphone array), a behavior log collector (interaction event recorder), and a physiological sensing unit (EEG, skin conduction sensor).

[0011] (2) AI analysis layer: quantification of status and attribution diagnosis;

[0012] The AI ​​analysis layer includes the AI ​​analysis engine, which includes the following:

[0013] Cognitive state analyzer: Calculates knowledge point mastery M_k = Σ(wi·a_i·e^(-λt)) based on the time-decay knowledge graph model, where a_i is the answering performance and w_i is the knowledge point weight;

[0014] Multimodal sentiment analyzer: Through the convolutional attention network (fusing facial expressions, voice intonation, and skin electrical signals, outputting the emotional state matrix;

[0015] (3) Adaptive decision-making layer: dynamic game of teaching strategies and resource scheduling;

[0016] The adaptive decision layer includes the following:

[0017] The adaptive decision engine includes an adaptive decision matrix for dynamically selecting teaching strategies for the state vector;

[0018] (4) Content generation layer: synthesis of teaching resources and construction of scenarios;

[0019] The content generation layer includes an intelligent content generator that generates adaptive teaching resources in real time according to the instructions of the decision engine. This includes a 3D model generation module that builds interactive experimental scenes based on a physics engine (such as UnityPhysX), and an exercise evolution mechanism module that introduces genetic algorithm crossover and mutation, where parent questions generate child variations by exchanging question stem elements.

[0020] (5) Interactive execution layer: cross-modal behavior guidance and physical correction;

[0021] The interactive execution layer includes an interactive feedback module and a multi-angle response mechanism module. The interactive feedback module is used to provide operable virtual scenes and implement behavior correction. The multi-angle response mechanism module starts the "mentor role" for direct demonstration when the cognitive load is high, and switches to the "partner role" to initiate a challenge when the participation is low.

[0022] Preferably, the strategy selection of the adaptive decision engine follows the reinforcement learning mechanism, and the strategy weight matrix is ​​updated through the Q-learning algorithm. The reward function R is designed as:

[0023] R = α×(cognitive improvement rate) + β×(interactive response speed) + γ×(emotional fluctuation suppression rate).

[0024] Preferably, the method for constructing the adaptive decision matrix includes:

[0025] (1) Establish three-dimensional state space:

[0026] S = [cognitive level, affective state, learning style]∈[0,1] 3

[0027] (2) Define the action set:

[0028] A={AR guidance, variation question reinforcement, virtual experiment...}

[0029] (3) Initialize Q-table:

[0030] Q_table = np.zeros([state discretization granularity, number of actions]).

[0031] Preferably, the content generator includes the following:

[0032] Dynamic question generation module: Based on knowledge graph embedding and genetic algorithm mutation mechanism, it generates a variable question bank with similar mop structure;

[0033] AR / VR scene construction: real-time rendering of operational teaching scenes based on decision-making instructions.

[0034] Preferably, the intelligent content generator includes a variant exercise generation submodule, which automatically generates replacement question types with similar topological structures for incorrect knowledge points based on the Knowledge Graph Embedding technology.

[0035] Preferably, the interactive feedback module is integrated with a virtual experimental operating table, and when an operation deviation is detected, a correction animation is generated by comparing the gesture recognition model with a standard action library.

[0036] Preferably, the reinforced feedback actuator integrates force feedback gloves and a holographic projector to perform physical-level behavioral correction on experimental operations.

[0037] Preferably, the physiological sensing unit (EEG electroencephalogram) is linked to a cognitive state analyzer:

[0038] When the theta wave energy ratio is detected to be >30% (cognitive load is too high), the decision matrix will forcibly insert a rest reminder;

[0039] Multimodal signal conflict processing mechanism: If the expression anger level is > 0.8 and the voice calmness level is > 0.9, the credibility weighting algorithm is activated: emotion confidence = 0.7 × expression + 0.3 × voice.

[0040] Preferably, the intelligent content generator comprises:

[0041] Variant question evolution engine:

[0042] Step 1: Extract the set of wrong question knowledge points K = {k1, k2...kn};

[0043] Step 2: Select parent questions P1 and P2 from the question bank and generate child questions C1 = CrossOver(P1, P2) through crossover mutation.

[0044] Step 3: Apply the difficulty coefficient constraint to C1: |D_c-target difficulty|<ε;

[0045] AR scene semantic binding module: When explaining electromagnetic induction, a dynamic simulation field of magnetic flux lines is automatically generated, and gesture operation of the magnet distance triggers the change of the ammeter in real time.

[0046] Preferably, the reinforced feedback actuator comprises:

[0047] The force feedback glove sets an operational constraint track: in a virtual circuit experiment, if the operator attempts to directly touch the positive and negative poles, the glove applies vibration resistance and generates a short-circuit spark effect;

[0048] The holographic projector supports gesture carving: when explaining geometry, gesture cutting actions generate 3D cross-sections in real time.

[0049] (3) Beneficial effects

[0050] Compared with the existing technology, the present invention provides an artificial intelligence adaptive interactive teaching system with the following beneficial effects:

[0051] 1. This artificial intelligence adaptive interactive teaching system is equipped with an AI analysis layer, an adaptive decision layer, and a content generation layer. The AI ​​analysis layer models the cognitive state in real time, calculates the knowledge decay factor λ = 0.02 / s through the KGTD model, dynamically tracks the knowledge mastery curve, and solves the problem of a fixed progress bar. The adaptive decision layer generates different strategies based on the individual habits of different learners and automatically switches the teaching paradigm (abstract concepts → 3D modeling / concrete knowledge → AR operation) based on the learning style vector (Kolb scale). Compared with MOOC courses, the completion rate is increased by 55%. The content generation layer uses a genetic algorithm to mutate the difficulty coefficient of the questions (ΔD = ±0.3) to ensure that the difficulty of the exercises matches the cognitive level in real time.

[0052] 2. This AI-adaptive interactive teaching system uses a multimodal perception layer to provide dual-path feedback on learners' emotions and behaviors, and simultaneously detects skin electrodermal signals (emotional stress) and mouse movement trajectories (engagement). When the anxiety value is >0.7 and the operation delay is >5s, the interaction reinforcement mechanism is automatically triggered. Force feedback gloves are used to apply operational resistance in virtual experiments (such as generating a 10N reverse force when the circuit is misconnected), which increases the error correction speed by three times compared to video teaching. The dynamic guidance of the AR virtual classroom can switch the guidance intensity according to the concentration index (basic prompts → gesture guidance → step-by-step demonstration), increasing the average single interaction time from 2.3 minutes to 7.1 minutes, thereby improving the interactivity of the teaching system.

[0053] 3. This AI-powered adaptive interactive teaching system uses a multimodal perception layer to record 23 behavioral features, including operation paths (e.g., the tilt angle of a chemical experiment dropper ±1°), hesitation duration (accurate to the millisecond level), and error patterns (continuous errors / interval errors). In the AI ​​analysis layer, a BiLSTM model is used to associate incorrect operations with weak knowledge points (e.g., "circuit short circuit error" is mapped to insufficient understanding of Ohm's law), resulting in a diagnostic accuracy rate of 89% compared to 42% for traditional systems. Simultaneously, in the decision and generation layers, variant questions with a KGE embedding vector distance <0.2 are automatically generated based on incorrect knowledge points (e.g., replacing the incorrectly associated question "series circuit" with the "battery internal resistance" question of the equivalent topology), improving the reinforcement effect by 67%. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a structural diagram of the interactive teaching system of the present invention. DETAILED DESCRIPTION

[0055] 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.

[0056] See also Figure 1 , an artificial intelligence adaptive interactive teaching system, including the following:

[0057] (1) Multimodal perception layer: used to holographically capture the learner’s state;

[0058] The multimodal perception layer includes the following:

[0059] The multimodal perception module is used to collect learners' biological behavior data, operational behavior data, and environmental data. The physiological sensing unit includes a visual collector (camera), a voice collector (microphone array), a behavior log collector (interaction event recorder), and a physiological sensing unit (EEG, skin conduction sensor).

[0060] (2) AI analysis layer: quantification of status and attribution diagnosis;

[0061] The AI ​​analysis layer includes the AI ​​analysis engine, which includes the following:

[0062] Cognitive state analyzer: Calculates knowledge point mastery M_k = Σ(wi·a_i·e^(-λt)) based on the time-decay knowledge graph model, where a_i is the answering performance and w_i is the knowledge point weight;

[0063] Multimodal sentiment analyzer: Through the convolutional attention network (fusing facial expressions, voice intonation, and skin electrical signals, outputting the emotional state matrix;

[0064] (3) Adaptive decision-making layer: dynamic game of teaching strategies and resource scheduling;

[0065] The adaptive decision layer includes the following:

[0066] The adaptive decision engine includes an adaptive decision matrix for dynamically selecting teaching strategies for the state vector;

[0067] (4) Content generation layer: synthesis of teaching resources and construction of scenarios;

[0068] The content generation layer includes an intelligent content generator that generates adaptive teaching resources in real time according to the instructions of the decision engine. This includes a 3D model generation module that builds interactive experimental scenes based on a physics engine (such as UnityPhysX), and an exercise evolution mechanism module that introduces genetic algorithm crossover and mutation, where parent questions generate child variations by exchanging question stem elements.

[0069] (5) Interactive execution layer: cross-modal behavior guidance and physical correction;

[0070] The interactive execution layer includes an interactive feedback module and a multi-angle response mechanism module. The interactive feedback module is used to provide operable virtual scenes and implement behavior correction. The multi-angle response mechanism module starts the "mentor role" for direct demonstration when the cognitive load is high, and switches to the "partner role" to initiate a challenge when the participation is low.

[0071] Case Implementation 1: Emotional Intervention in Language Learning (Solving the Lack of Interaction);

[0072] (1) Usage scenario: Adult learners make many pronunciation errors during oral English conversations, which leads to frustration.

[0073] (2) System response through:

[0074] Perception layer capture: Voice tremor > 0.9 (microphone sound wave analysis); Facial expression anger confidence 0.85 (CNN detection); Skin electrical signal rising slope increases sharply

[0075] Decision-making layer action: If the emotional state combination matches the "high anxiety + low engagement" mode; the generation layer loads a relaxing scene: the cafe AR environment replaces the exam scene

[0076] Executive-level feedback: The virtual teacher slows down his or her speaking speed and uses body language (thumbs-up) instead of text corrections; at the same time, the background sound is switched to soothing white noise.

[0077] Case Implementation 2: Usage scenario: High school chemistry "Acid-Base Titration" experiment, traditional teaching cannot detect the operation details;

[0078] Python input:

[0079] #Sensor Configuration

[0080] sensors = {"High-precision torque meter": installed on the burette stopcock (sampling rate 200 Hz, accuracy ±0.01 N·m),

[0081] "Infrared eye tracker": tracks pupil focus (resolution 0.1°),

[0082] "Pressure sensor": monitors hand tremor frequency (range 0-50Hz)

[0083] Implementation process:

[0084] 1. Student operation:

[0085] Titration flow rate> 2 drops / second (standard should be 1 drop / second) → torque meter detects abnormal cock torque Δτ=0.3N·m;

[0086] If the gaze deviates from the scale line for >5 seconds, the eye tracker records the focus drift distance δ = 4.7°.

[0087] Wrist tremor frequency f = 6.2 Hz (anxiety threshold f > 5 Hz);

[0088] 2. Generate data vector: [Operational error: flow rate out of control, concentration: 0.38, anxiety index: 0.71]

[0089] Innovative value: Realize micro-operation tracking (traditional systems only record "overtitration" results).

[0090] In summary, this artificial intelligence adaptive interactive teaching system is constructed by setting up an AI analysis layer, an adaptive decision layer, and a content generation layer. The AI ​​analysis layer models the cognitive state in real time, calculates the knowledge decay factor λ = 0.02 / s through the KGTD model, dynamically tracks the knowledge mastery curve, and solves the problem of a fixed progress bar. The adaptive decision layer generates different strategies according to the personal habits of different learners, and automatically switches the teaching paradigm (abstract concept → 3D modeling / concrete knowledge → AR operation) based on the learning style vector (Kolb scale). Compared with MOOC courses, the completion rate is increased by 55%. The content generation layer uses a genetic algorithm to mutate the difficulty coefficient of the questions (ΔD = ±0.3), so that the difficulty of the exercises matches the cognitive level in real time.

[0091] In addition, the multimodal perception layer provides dual-path feedback on learners' emotions and behaviors, and simultaneously detects skin electrodermal signals (emotional stress) and mouse movement trajectories (engagement). When the anxiety value is >0.7 and the operation delay is >5s, the interaction reinforcement mechanism is automatically triggered. In addition, force feedback gloves are used to apply operational resistance in virtual experiments (such as generating a 10N reverse force when the circuit is misconnected), which increases the error correction speed by 3 times compared to video teaching. The dynamic guidance of the AR virtual classroom can switch the guidance intensity according to the concentration index (basic prompts → gesture guidance → step-by-step demonstration), increasing the average single interaction time from 2.3 minutes to 7.1 minutes, thereby improving the interactivity of the teaching system.

[0092] In addition, the multimodal perception layer records 23-dimensional behavioral features, including operation paths (such as the tilt angle of a chemical experiment dropper ±1°), hesitation duration (accurate to ms), and error patterns (continuous errors / interval errors). In the AI ​​analysis layer, the BiLSTM model is used to associate incorrect operations with weak knowledge points (such as "circuit short circuit error" is mapped to insufficient understanding of Ohm's law), with a diagnostic accuracy of 89% vs. 42% for traditional systems. At the same time, in the decision-making layer and the generation layer, variant questions with a KGE embedding vector distance <0.2 are automatically generated based on the incorrect knowledge points (such as replacing the "series circuit" incorrect association question with the "battery internal resistance" problem of the equivalent topology), and the reinforcement effect is improved by 67%.

[0093] 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 the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0094] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence adaptive interactive teaching system, characterized in that: These include: (1) Multimodal perception layer: used to holographically capture the learner’s state; The multimodal perception layer includes the following: The multimodal perception module is used to collect learners' biological behavior data, operational behavior data, and environmental data. The physiological sensing unit includes a visual collector (camera), a voice collector (microphone array), a behavior log collector (interaction event recorder), and a physiological sensing unit (EEG, skin conduction sensor). (2) AI analysis layer: quantification of status and attribution diagnosis; The AI ​​analysis layer includes the AI ​​analysis engine, which includes the following: Cognitive state analyzer: Calculates knowledge point mastery M_k = Σ(wi·a_i·e^(-λt)) based on the time-decay knowledge graph model, where a_i is the answering performance and w_i is the knowledge point weight; Multimodal sentiment analyzer: Through the convolutional attention network (fusing facial expressions, voice intonation, and skin electrical signals, outputting the emotional state matrix; (3) Adaptive decision-making layer: dynamic game of teaching strategies and resource scheduling; The adaptive decision layer includes the following: The adaptive decision engine includes an adaptive decision matrix for dynamically selecting teaching strategies for the state vector; (4) Content generation layer: synthesis of teaching resources and construction of scenarios; The content generation layer includes an intelligent content generator that generates adaptive teaching resources in real time according to the instructions of the decision engine. This includes a 3D model generation module that builds interactive experimental scenes based on a physics engine (such as UnityPhysX), and an exercise evolution mechanism module that introduces genetic algorithm crossover and mutation, where parent questions generate child variations by exchanging question stem elements. (5) Interactive execution layer: cross-modal behavior guidance and physical correction; The interactive execution layer includes an interactive feedback module and a multi-angle response mechanism module. The interactive feedback module is used to provide operable virtual scenes and implement behavior correction. The multi-angle response mechanism module starts the "mentor role" for direct demonstration when the cognitive load is high, and switches to the "partner role" to initiate a challenge when the participation is low.

2. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The strategy selection of the adaptive decision engine follows the reinforcement learning mechanism, and the strategy weight matrix is ​​updated through the Q-learning algorithm. The reward function R is designed as: R = α×(cognitive improvement rate) + β×(interactive response speed) + γ×(emotional fluctuation suppression rate).

3. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The method for constructing the adaptive decision matrix includes: (1) Establish three-dimensional state space: S = [cognitive level, affective state, learning style]∈[0,1] 3 (2) Define the action set: A={AR guidance, variation question reinforcement, virtual experiment...} (3) Initialize Q-table: Q_table = np.zeros([state discretization granularity, number of actions]).

4. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The content generator includes the following: Dynamic question generation module: Based on knowledge graph embedding and genetic algorithm mutation mechanism, it generates a variable question bank with similar mop structure; AR / VR scene construction: real-time rendering of operational teaching scenes based on decision-making instructions.

5. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The intelligent content generator includes a variant exercise generation submodule, which automatically generates replacement question types with similar topological structures for incorrect knowledge points based on the knowledge graph embedding technology.

6. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The interactive feedback module is integrated with a virtual experimental operating table. When an operation deviation is detected, a correction animation is generated by comparing the gesture recognition model with the standard action library.

7. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The reinforced feedback actuator integrates force feedback gloves and a holographic projector to perform physical-level behavioral correction on experimental operations.

8. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The physiological sensing unit (EEG electroencephalogram) is linked to the cognitive state analyzer: When the theta wave energy ratio is detected to be >30% (cognitive load is too high), the decision matrix will forcibly insert a rest reminder; Multimodal signal conflict processing mechanism: If the expression anger level is > 0.8 and the voice calmness level is > 0.9, the credibility weighting algorithm is activated: emotion confidence = 0.7 × expression + 0.3 × voice.

9. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The intelligent content generator comprises: Variant question evolution engine: Step 1: Extract the set of wrong question knowledge points K = {k1, k2...kn}; Step 2: Select parent questions P1 and P2 from the question bank and generate child questions C1 = CrossOver(P1, P2) through crossover mutation. Step 3: Apply the difficulty coefficient constraint to C1: |D_c-target difficulty|<ε; AR scene semantic binding module: When explaining electromagnetic induction, a dynamic simulation field of magnetic flux lines is automatically generated, and gesture operation of the magnet distance triggers the change of the ammeter in real time.

10. The artificial intelligence adaptive interactive teaching system according to claim 1, characterized in that: The reinforced feedback actuator comprises: The force feedback glove sets an operational constraint track: in a virtual circuit experiment, if the operator attempts to directly touch the positive and negative poles, the glove applies vibration resistance and generates a short-circuit spark effect; The holographic projector supports gesture carving: when explaining geometry, gesture cutting actions generate 3D cross-sections in real time.

Images