Intelligent personalized learning path recommendation system

Abstract

The invention discloses an intelligent personalized learning path recommendation system, and relates to the technical field of learning systems, and the system comprises a multi-source data collection module which is used for synchronizing behavior data of students in a cross-platform learning scene in real time; the cognitive feature analysis engine comprises a style recognition sub-engine and a demand prediction sub-engine to predict the potential learning demand intensity of students for unmastered knowledge points and generate a priority list comprising knowledge gaps; the personalized path generator generates a three-dimensional path plan; the proportion of guided questioning, example demonstration and autonomous exploration is automatically configured according to knowledge difficulty; the dynamic self-adaptive adjustment unit is used for designing a reward function including short-term progress speed and long-term ability growth potential by taking real-time performance of students as a state space and taking path adjustment action as a decision space based on a reinforcement learning framework; and a double-loop feedback mechanism is realized. According to the method, the dimension limitation of traditional learning analysis is broken through, the difficulty of stiffness of a static course template is broken through, and meanwhile, the semantic gap of subject cognition is broken through.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of learning systems, and specifically to an intelligent personalized learning path recommendation system. Background Art

[0002] Since the 1990s, rule-based adaptive learning systems (such as intelligent tutoring systems) have begun to attempt to adjust the content difficulty according to the learner's performance. In recent years, machine learning, deep learning, and natural language processing technologies have enabled the system to analyze learning behavior data in real time (such as answering speed, error type), and construct a more accurate user profile.

[0003] Currently, personalized learning systems in the education field face the following technical challenges: traditional systems rely on questionnaires or simple behavior logs, and cannot capture multi-modal learning features such as cognitive load and emotional state (such as EEG brain waves, eye movement trajectories), resulting in incomplete learner profiles; most systems adopt preset course templates, lacking a dynamic adjustment mechanism based on real-time cognitive states (such as knowledge defect location, metacognitive strategy use level), and it is difficult to adapt to the non-linear changes of individual learning rhythms; the knowledge representation methods of different disciplines vary significantly (such as the symbol system of mathematics and the field theory model of physics), and existing systems lack semantic mapping and transfer strategies for cross-course knowledge paths, resulting in low utilization rate of prerequisite course experience; the paths generated by deep learning models lack interpretability, and it is difficult for teachers and parents to verify the rationality of the recommendation strategy, and there is also a risk of algorithmic bias (such as systematic neglect of specific cognitive style groups);

[0004] In summary, the full-modal cognitive profile, adaptive path evolution, and cross-domain knowledge weaving in learning systems are technical problems that need to be solved urgently by those in this technical field. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide an intelligent personalized learning path recommendation system to solve the technical problems raised in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] An intelligent personalized learning path recommendation system, including a multi-source data collection module: used to synchronize the behavior data of students in cross-platform learning scenarios in real time;

[0008] A cognitive feature analysis engine: based on a hybrid machine learning model, including a style recognition sub-engine: using a dynamic Bayesian network combined with a learning behavior sequence to construct a five-dimensional learning style vector including visual, auditory, hands-on practice preferences, linear reasoning, and jumping thinking modes; a demand prediction sub-engine: processing time series data through an LSTM neural network to predict the potential learning demand intensity of students for unmastered knowledge points, and generating a list of knowledge gap priorities;

[0009] Personalized Path Generator: Adopting a multi-objective genetic algorithm, with knowledge point coverage, learning load balance, and interest matching degree as optimization objectives, it generates a three-dimensional path planning including a learning object sequence, a media type combination, and a time allocation plan; and embeds a cognitive scaffolding strategy, automatically configuring the proportion of guided questions, example demonstrations, and independent explorations according to the knowledge difficulty.

[0010] Dynamic Adaptive Adjustment Unit: Based on a reinforcement learning framework, with the student's real-time performance as the state space and the path adjustment action as the decision space, a reward function is designed to include the short-term progress speed and the long-term ability growth potential; a dual-loop feedback mechanism is implemented. Specifically, the inner loop fine-tunes the presentation method of the current learning unit every 15 minutes, and the outer loop reconstructs the overall path topology structure every 48 hours.

[0011] Preferably, the behavior data includes:

[0012] Learning behavior stream data: knowledge point residence duration, answer path selection, resource switching frequency, video viewing progress;

[0013] Explicit feedback data: scoring, collection, tag annotation, emotion expression input;

[0014] Implicit feedback data: pupil movement trajectory, keyboard keystroke interval, microphone voice intonation analysis;

[0015] Effectiveness evaluation data: knowledge point test correct rate, homework completion speed, long-term memory decay rate.

[0016] Preferably, the hybrid machine learning model further includes:

[0017] The federated learning framework is used to aggregate multi-user data while protecting privacy and construct a population learning style distribution map;

[0018] The meta-learning optimizer is used to achieve fast model adaptation when introducing new knowledge points through the MAML algorithm;

[0019] The interpretability module is used to decompose the prediction result using SHAP values and generate a path recommendation explanation report including feature contribution degrees.

[0020] Preferably, the cognitive feature analysis engine further includes:

[0021] The cognitive load assessment sub-module is used for the coupled analysis of physiological signals and learning behaviors, and dynamically calculates the current cognitive resource occupancy rate;

[0022] The flow state detector identifies and maintains the optimal learning engagement state by comprehensively analyzing the answer fluency, the number of repeated operations, and the emotional feedback.

[0023] Preferably, the personalized path generator has the following optimization constraints:

[0024] Prerequisite knowledge relationship constraint: Construct a prerequisite knowledge network through association rule mining to ensure that the path conforms to the subject logical structure;

[0025] Cognitive flexibility constraint: Set interval repetition nodes between similar knowledge points to suppress the proactive interference effect;

[0026] Attention economy constraint: Use a survival analysis model to predict the critical point of learning fatigue and automatically insert micro-break nodes.

[0027] Preferably, the dynamic adaptive adjustment unit includes:

[0028] Group wisdom enhancement mechanism: When the individual learning path falls into a local optimum, introduce the optimal path segment of a group of similar learners for genetic algorithm crossover operation;

[0029] Cross-course knowledge transfer module: By constructing an inter-disciplinary knowledge mapping table, transfer the metacognitive strategies of mastered courses to new learning fields.

[0030] Preferably, the recommendation system further includes:

[0031] The educational digital twin construction module is used to create a virtual mapping of the student's cognitive state through multi-agent modeling technology, supporting the preview and verification of path adjustment strategies;

[0032] The ethics review sub-unit is used to detect algorithmic biases in path recommendations using a generative adversarial network to ensure educational fairness.

[0033] Preferably, the personalized path generator further includes:

[0034] The three-dimensional knowledge map visualization module is used to map the learning path to a three-dimensional cognitive space, and uses a force-directed graph algorithm to display the association strength of knowledge points, supporting students to interactively explore the learning trajectory in a virtual reality manner;

[0035] The path evolution interpreter is used to display the adaptive process of the learning path with the change of cognitive state through a dynamic topology evolution animation, and provide an evolution report including cognitive inflection point identification and strategy adjustment basis.

[0036] Preferably, the multi-source data acquisition module further includes:

[0037] The multi-modal interaction interface adaptation unit is used to support speech instruction parsing, gesture control recognition and eye movement tracking input, and automatically record cross-modal interaction data;

[0038] The context awareness sub-module is used to analyze the learning scene characteristics through device sensor fusion, and dynamically adjust the data acquisition frequency and path recommendation strategy.

[0039] In summary, the present invention mainly has the following beneficial effects:

[0040] In the present invention, the dimensional limitation of traditional learning analysis is broken through. By integrating 12 types of heterogeneous data such as EEG brain waves, eye movement trajectories, and handwriting time series, a three-dimensional cognitive portrait is constructed to identify the implicit cognitive barriers of learners, establish a complete cognitive chain from neuron activities to explicit behaviors, and support the design of personalized intervention strategies.

[0041] The rigid situation of the static course template is cracked. Based on the reservoir computing framework, real-time modeling of the cognitive state is realized. When a sudden change in attention is detected, a path switch is triggered within 3 seconds to avoid cognitive breaks, and the density of the cognitive scaffold is dynamically adjusted (such as from a prompt every 5 minutes to a prompt every 2 minutes), increasing the learning fluency by 58%.

[0042] The semantic gap in disciplinary cognition is bridged. By constructing a cross-domain ontology mapping table through a graph neural network, the coverage rate of strategy migration is increased, and a composite learning path including multidisciplinary intersections is automatically generated. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the overall processing logic framework of the system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0044] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0045] As Figure 1 shown, an intelligent personalized learning path recommendation system includes a multi-source data collection module: used to synchronize the behavior data of students in cross-platform learning scenarios in real time;

[0046] A cognitive feature analysis engine: based on a hybrid machine learning model, including a style recognition sub-engine: using a dynamic Bayesian network combined with learning behavior sequences to construct a five-dimensional learning style vector including visual, auditory, hands-on practice preferences, linear reasoning, and jumping thinking modes; a demand prediction sub-engine: processing time series data through an LSTM neural network to predict the potential learning demand intensity of students for unmastered knowledge points and generating a knowledge gap priority list.

[0047] A personalized path generator: using a multi-objective genetic algorithm, with knowledge point coverage, learning load balance, and interest matching degree as optimization objectives, generating a three-dimensional path plan including a learning object sequence, a media type combination, and a time allocation plan; and embedding a cognitive scaffold strategy to automatically configure the proportion of guided questions, example demonstrations, and independent explorations according to the knowledge difficulty.

[0048] Dynamic Adaptive Adjustment Unit: Based on the reinforcement learning framework, with the real-time performance of students as the state space and path adjustment actions as the decision space, a reward function is designed to include short-term progress speed and long-term ability growth potential; a dual-loop feedback mechanism is implemented. Specifically, the inner loop fine-tunes the presentation method of the current learning unit every 15 minutes, and the outer loop reconstructs the overall path topology every 48 hours.

[0049] Behavior data includes:

[0050] Learning behavior flow data: knowledge point residence duration, answering path selection, resource switching frequency, video viewing progress;

[0051] Explicit feedback data: scoring, collection, tag annotation, emotional expression input;

[0052] Implicit feedback data: pupil movement trajectory, keyboard keystroke interval, microphone voice intonation analysis;

[0053] Effectiveness evaluation data: knowledge point test correct rate, homework completion speed, long-term memory decay rate.

[0054] The hybrid machine learning model further includes:

[0055] The federated learning framework is used to aggregate multi-user data while protecting privacy and construct a population learning style distribution map;

[0056] The meta-learning optimizer is used to achieve fast model adaptation when introducing new knowledge points through the MAML algorithm;

[0057] The interpretability module is used to decompose the prediction results using SHAP values and generate a path recommendation explanation report including feature contribution degrees.

[0058] The cognitive feature analysis engine also includes:

[0059] The cognitive load assessment sub-module is used for the coupled analysis of physiological signals and learning behaviors to dynamically calculate the current cognitive resource occupancy rate;

[0060] The flow state detector identifies and maintains the optimal learning engagement state by comprehensively analyzing answering fluency, number of repeated operations, and emotional feedback.

[0061] The personalized path generator has the following optimization constraints:

[0062] Knowledge point prerequisite relationship constraint: Construct a prerequisite knowledge network through association rule mining to ensure that the path conforms to the subject logical structure;

[0063] Cognitive flexibility constraint: Set spaced repetition nodes between similar knowledge points to inhibit the proactive interference effect;

[0064] Attention Economy Constraint: Use a survival analysis model to predict the critical point of learning fatigue and automatically insert micro-rest nodes.

[0065] The dynamic adaptive adjustment unit includes:

[0066] Group Intelligence Enhancement Mechanism: When an individual's learning path falls into a local optimum, introduce the optimal path segments of a group of similar learners for genetic algorithm crossover operations;

[0067] Cross-Curricular Knowledge Transfer Module: By constructing an inter-disciplinary knowledge mapping table, transfer the metacognitive strategies of mastered courses to new learning areas.

[0068] The recommendation system also includes:

[0069] The Educational Digital Twin Construction Module is used to create a virtual mapping of the student's cognitive state through multi-agent modeling technology, supporting the pre-play verification of path adjustment strategies;

[0070] The Ethical Review Subunit is used to detect algorithmic biases in path recommendations using a generative adversarial network to ensure educational fairness.

[0071] The personalized path generator further includes:

[0072] The 3D Knowledge Map Visualization Module is used to map the learning path to a 3D cognitive space, use a force-directed graph algorithm to display the association strength of knowledge points, and support students to interactively explore the learning trajectory in a virtual reality manner;

[0073] The Path Evolution Interpreter is used to display the adaptive process of the learning path as the cognitive state changes through dynamic topology evolution animations, and provide an evolution report containing cognitive inflection point identification and strategy adjustment basis.

[0074] The multi-source data acquisition module also includes:

[0075] The multi-modal interaction interface adaptation unit is used to support speech instruction parsing, gesture control recognition, and eye movement tracking input, and automatically record cross-modal interaction data;

[0076] The context awareness sub-module is used to analyze the learning scene characteristics through device sensor fusion, and dynamically adjust the data acquisition frequency and path recommendation strategy.

[0077] Example 1: It should be noted that in this example, it is used to illustrate the generation of personalized learning paths based on multi-modal data.

[0078] Scenario: Generate a learning path for the geometry chapter for junior high school mathematics learners

[0079] 1. Data acquisition:

[0080] Collect the sequential data of the problem-solving handwriting through an intelligent pen, specifically the pen pressure, writing speed, and graphic closure degree;

[0081] Capture the line-of-sight stay trajectory of the learner in front of the solid geometry model through a desktop camera

[0082] Record the number of repeated attempts for the problem of adding auxiliary lines in the online test;

[0083] Collect the prefrontal theta wave power ratio output by the EEG device.

[0084] 2. Cognitive feature analysis:

[0085] Style recognition: Bayesian network analysis shows that the student has outstanding visual-spatial processing ability, and the fixation duration on the 3D model > 90th percentile; however, the symbol reasoning speed is slow, and the average time-consuming for adding auxiliary lines > 2σ of the group mean;

[0086] Demand prediction: The LSTM model predicts that the demand intensity for knowledge points related to spatial imagination, such as the conversion of three-view drawings, is 0.87, while the demand intensity for knowledge points related to theorem proof is 0.42;

[0087] 3. Path generation:

[0088] The path generated by the genetic algorithm includes:

[0089] Initial stage: 70% three-dimensional model interactive demonstration + 30% dynamic geometry demonstration;

[0090] Mid-term stage: Introduce an AR auxiliary line generation tool to automatically prompt key construction points;

[0091] Late stage: Configure a comparative exercise set including error answer analysis;

[0092] Cognitive scaffolding strategy: Insert a spatial imagination micro-test every 15 minutes. When the correct rate < 60%, automatically roll back to the model demonstration stage;

[0093] 4. Dynamic adjustment:

[0094] When it is detected that the learner has an operation stagnation during the use of the AR tool, and there is no effective construction for more than 30 seconds, trigger path adjustment:

[0095] (1) Insert step-by-step construction animations;

[0096] (2) Adjust the difficulty coefficient of the subsequent exercises and reduce it by 0.2σ;

[0097] After the daily study ends, the system automatically increases the proportion of knowledge points related to theorem proof to 25% the next day according to the 12% increase in the correct rate of the spatial imagination test on the same day.

[0098] Example 2: It should be noted that in this example, it is used to illustrate cross-course knowledge transfer and group wisdom enhancement.

[0099] Scenario: High school physics learners encounter cognitive obstacles in the electromagnetism chapter

[0100] 1. Intervention of group wisdom

[0101] The system identifies that the similarity between the current path of this student and the trajectory of the "weak in mastering the concept of electric field strength" sub-cluster (n = 327) in the group reaches 83%;

[0102] Extract the optimal adjustment strategy of this sub-cluster: Introduce the analogy learning method and compare the distribution of electric field lines with the fluid pressure field.

[0103] 2. Cross-course transfer

[0104] It is detected that the learner has effectively used the vector field visualization tool in the fluid mechanics course;

[0105] Automatically transfer the interaction mode of this tool to the electromagnetic field line simulation experiment and retain its familiar color mapping scheme.

[0106] 3. Path reconstruction

[0107] Adopt the crossover operation of the genetic algorithm to integrate the transfer strategy with the current path:

[0108] Add 4 analogy learning modules (accounting for 15% of the class hours);

[0109] Adjust the experiment order: First complete the fluid pressure field experiment, and then enter the electrostatic field simulation;

[0110] The correct rate of the electric field strength concept test after the path adjustment is increased by 18% (from 52% to 70%).

[0111] Example 3: It should be noted that in this example, it is used to illustrate blockchain evidence storage and ethical review

[0112] Scenario: Generate the digital academic certificate of the learner

[0113] 1. Upload the learning process to the blockchain

[0114] Write the key path adjustment records, including timestamps, adjustment bases, and effect data, into Hyperledger Fabric;

[0115] The smart contract automatically verifies the hash continuity of each record to ensure the data cannot be tampered with

[0116] 2. Detection of algorithmic bias

[0117] Conduct an adversarial test on the path recommendation algorithm:

[0118] Generate virtual learner groups, including different genders, regions, and learning styles;

[0119] Detect the differences in the quality of recommended resources and the proportion of authoritative literature among different subgroups;

[0120] It is found that the original algorithm lacks sufficient experimental resources recommended for learners with practical ability. Through SHAP value analysis, it is found that the "operation accuracy" feature is over-weighted.

[0121] 3. Ethical compliance adjustment

[0122] Adjust the feature weight matrix, and replace the single accuracy index with the composite index of "operation speed" and "number of innovative solutions";

[0123] After retraining the model, the quality of recommended resources for the practical ability group has increased by 23%, and the difference from other groups is <5%.

[0124] Example 4: It should be noted that in this example, it is used to illustrate the three-dimensional knowledge map and multimodal interaction

[0125] Scenario: High school biology learners explore the genetics knowledge network

[0126] 1. Three-dimensional map construction

[0127] Use the Force-Directed algorithm to generate a three-dimensional knowledge graph:

[0128] X-axis: Knowledge difficulty (basic concepts → frontier research)

[0129] Y-axis: Degree of interdisciplinary (biological purity → cross with chemistry / physics)

[0130] Z-axis: Time dimension (historical discoveries → modern technologies)

[0131] 2. Multimodal interaction

[0132] Voice command: "Show experiments related to Mendel's laws" → Highlight the pea experiment node in the 19th century;

[0133] Gesture control: Zoom in and out with both hands to focus on the gene editing technology cluster;

[0134] Eye movement tracking: When staring at a node for more than 3 seconds, an animation of the historical development of this knowledge point will automatically pop up.

[0135] 3. Path visualization

[0136] The historical path of the learner is displayed as a dynamic streamer trajectory, and the current position is marked as a pulsating light spot;

[0137] The unexplored area uses a semi-transparent gradient effect to attract attention and guide;

[0138] Support the VR mode to enter the knowledge node, and display the DNA replication process with molecular-level animation.

[0139] 4. Technical effect verification

[0140] (1) Improvement in learning efficiency:

[0141] The time required for the experimental group (n = 1200) to complete the same curriculum standard was shortened by 35% compared with the traditional group (p < 0.01);

[0142] Knowledge retention rate test (30 days later): 72% in the experimental group vs 51% in the traditional group;

[0143] (2) Effectiveness of path adjustment:

[0144] The number of times of dynamic adjustment triggering is positively correlated with the learning effect (r = 0.68), and the key adjustment points are mainly distributed in the cognitive inflection point area.

[0145] (3) System robustness:

[0146] When 30% of the simulated data is missing, the accuracy of path generation only drops by 4.2%, and it increases by 7.8% after recovery through federated learning.

[0147] The above embodiments are only used to illustrate the technical idea of the present invention, and the protection scope of the present invention cannot be limited thereby. Any modification made on the basis of the technical solution according to the technical idea proposed by the present invention falls within the protection scope of the present invention.

Claims

1. An intelligent personalized learning path recommendation system, characterized in that, It includes a multi-source data acquisition module: which is used to synchronize the behavior data of students in cross-platform learning scenarios in real time; A cognitive feature analysis engine: Based on a hybrid machine learning model, it includes a style recognition sub-engine: which uses a dynamic Bayesian network combined with learning behavior sequences to construct a five-dimensional learning style vector including visual, auditory, hands-on practice preferences, linear reasoning, and jump thinking patterns; a demand prediction sub-engine: which processes time series data through an LSTM neural network to predict the potential learning demand intensity of students for unmastered knowledge points and generates a knowledge gap priority list; A personalized path generator: which uses a multi-objective genetic algorithm with knowledge point coverage, learning load balance, and interest matching degree as optimization objectives to generate a three-dimensional path plan including a learning object sequence, media type combination, and time allocation plan; and embeds a cognitive scaffolding strategy to automatically configure the proportions of guided questions, example demonstrations, and independent explorations according to the knowledge difficulty; A dynamic adaptive adjustment unit: Based on a reinforcement learning framework, with the real-time performance of students as the state space and path adjustment actions as the decision space, a reward function is designed to include short-term progress speed and long-term ability growth potential; a dual-loop feedback mechanism is implemented, specifically, the inner loop fine-tunes the presentation method of the current learning unit every 15 minutes, and the outer loop reconstructs the overall path topology structure every 48 hours.

2. The intelligent personalized learning path recommendation system according to claim 1, wherein The behavior data includes: Learning behavior flow data: knowledge point residence duration, answer path selection, resource switching frequency, video viewing progress; Explicit feedback data: scoring, collection, tag annotation, emotional expression input; Implicit feedback data: pupil movement trajectory, keyboard keystroke interval, microphone speech intonation analysis; Effectiveness evaluation data: knowledge point test correct rate, homework completion speed, long-term memory decay rate.

3. An intelligent personalized learning path recommendation system according to claim 1, wherein, The hybrid machine learning model further includes: A federated learning framework for aggregating multi-user data while protecting privacy to construct a population learning style distribution map; A meta-learning optimizer for quickly adapting the model when introducing new knowledge points through the MAML algorithm; An interpretability module for decomposing the prediction results using SHAP values to generate a path recommendation explanation report including feature contribution degrees.

4. An intelligent personalized learning path recommendation system according to claim 1, characterized in that, The cognitive feature analysis engine also includes: A cognitive load assessment sub-module for coupling analysis of physiological signals and learning behaviors to dynamically calculate the current cognitive resource occupancy rate; A flow state detector that identifies and maintains the optimal learning engagement state by comprehensively analyzing answer fluency, number of repeated operations, and emotional feedback.

5. An intelligent personalized learning path recommendation system according to claim 1, characterized in that, The personalized path generator has the following optimization constraint conditions: Knowledge point precedence relationship constraint: constructing a prerequisite knowledge network through association rule mining to ensure that the path conforms to the subject logical structure; Cognitive flexibility constraint: setting interval repetition nodes between similar knowledge points to suppress the proactive inhibition effect; Attention economy constraint: using a survival analysis model to predict the learning fatigue critical point and automatically inserting micro-break nodes.

6. An intelligent personalized learning path recommendation system according to claim 1, characterized in that, The dynamic adaptive adjustment unit includes: A group wisdom enhancement mechanism: when an individual's learning path falls into a local optimum, introducing the optimal path segment of a similar learner group for genetic algorithm crossover operations; Cross - curriculum knowledge transfer module: By constructing an inter - disciplinary knowledge mapping table, transfer the metacognitive strategies of the mastered courses to the new learning fields.

7. An intelligent personalized learning path recommendation system according to claim 1, characterized in that, The recommendation system also includes: The educational digital twin construction module is used to create a virtual mapping of the student's cognitive state through multi - agent modeling technology, supporting the pre - rehearsal verification of the path adjustment strategy; The ethics review sub - unit is used to detect algorithmic biases in path recommendations using generative adversarial networks to ensure educational fairness.

8. An intelligent personalized learning path recommendation system according to claim 5, characterized in that, The personalized path generator further includes: The 3D knowledge map visualization module is used to map the learning path to a 3D cognitive space, using a force - directed graph algorithm to display the association strength of knowledge points, supporting students to interactively explore the learning trajectory in a virtual reality manner; The path evolution interpreter is used to show the adaptive process of the learning path with the change of the cognitive state through dynamic topology evolution animation, and provide an evolution report including the identification of cognitive inflection points and the basis for strategy adjustment.

9. An intelligent personalized learning path recommendation system according to claim 1, characterized in that, The multi - source data acquisition module also includes: The multi - modal interaction interface adaptation unit is used to support voice command parsing, gesture control recognition and eye - movement tracking input, and automatically record cross - modal interaction data; The context - awareness sub - module is used to analyze the learning scenario characteristics through device sensor fusion, and dynamically adjust the data acquisition frequency and path recommendation strategy.

Images