Intelligent learning resource recommendation method and system based on knowledge graph

Abstract

The invention relates to the technical field of resource recommendation of a knowledge graph, in particular to an intelligent learning resource recommendation method and system based on the knowledge graph, and the method comprises the steps: collecting the multi-modal behavior data of a student in a preset simulation dangerous case scene, and carrying out the feature extraction to obtain a student behavior feature sequence; performing multi-modal time sequence association analysis on the student behavior feature sequence through a cognitive-physiological dynamic association knowledge graph to obtain an association analysis result; identifying a target cognitive strategy defect of the student and a corresponding target physiological response mode from a cognitive-physiological association knowledge graph according to the association analysis result; and according to the target cognitive strategy defect and the target physiological response mode, generating a corresponding learning resource sequence from the cognitive-physiological dynamic association knowledge graph. The cognitive process and psychological load of the student are effectively evaluated, so that the pertinence and effectiveness of learning resource recommendation are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph resource recommendation, and particularly to a knowledge graph-based learning resource intelligent recommendation method and system. BACKGROUND

[0002] In training scenarios involving high-pressure, complex dynamic simulation of dangerous situations, it is crucial to provide precise and effective learning resource recommendations to learners to improve their response capabilities and overall quality. Such training aims to enable learners to master the correct strategies and skills for dealing with various dangerous situations through repeated practice and experience in a highly realistic simulation environment, so that they can respond quickly and accurately in actual work.

[0003] However, existing training evaluation and resource recommendation methods often focus too much on the operational results of learners, such as whether the task is successfully completed, the time and accuracy of task completion, and other surface indicators. Although these indicators can reflect the learning effectiveness of learners to some extent, they cannot penetrate the surface of "operational results" and deeply understand the learners' internal and implicit "cognitive strategy defects" and "psychophysiological vulnerabilities". In a high-pressure, complex dynamic simulation of dangerous situations, learners' cognitive strategies and psychophysiological states play a decisive role in task execution. For example, learners may make mistakes due to incorrect cognitive strategies, such as relying too much on experience and ignoring new situations; or be affected by psychophysiological vulnerabilities, such as excessive nervousness and divided attention, which affect the quality of task completion.

[0004] Therefore, there is an urgent need for a knowledge graph-based learning resource intelligent recommendation method to effectively evaluate learners' cognitive processes and psychological load to improve the relevance and effectiveness of learning resource recommendations. SUMMARY

[0005] (1) Technical problem to be solved The purpose of the present application is to provide a knowledge graph-based learning resource intelligent recommendation method and system to solve the problem that in the simulation of dangerous situations, it is impossible to penetrate the surface of the operational results, effectively analyze the learners' internal and implicit cognitive strategy defects and psychophysiological vulnerabilities, and evaluate the learners' cognitive processes and psychological load, resulting in inaccurate learning resource recommendations.

[0006] (2) Technical solution To achieve the above-mentioned purpose, on the one hand, the present application provides a knowledge graph-based learning resource intelligent recommendation method, which comprises: S1, collecting multi-modal behavior data of learners in a preset simulation of dangerous situations, the multi-modal behavior data including operation behavior data, eye movement trajectory data, and physiological monitoring data.

[0007] S2. The multimodal behavioral data is subjected to feature extraction to obtain a student behavior feature sequence, which includes an operational feature subsequence, a visual scanning feature subsequence, and a physiological response feature subsequence.

[0008] S3. The trainee's behavioral feature sequence is subjected to multimodal temporal association analysis through a cognitive-physiological dynamic association knowledge graph to obtain the association analysis results. The cognitive-physiological dynamic association knowledge graph includes behavioral feature pattern nodes, cognitive strategy defect nodes, physiological response pattern nodes, and learning resource nodes.

[0009] S4. Based on the association analysis results, identify the trainees' target cognitive strategy deficiencies and corresponding target physiological response patterns from the cognitive-physiological association knowledge graph.

[0010] S5. Generate a corresponding learning resource sequence from the cognitive-physiological dynamic association knowledge graph based on the target cognitive strategy defects and target physiological response patterns.

[0011] Furthermore, the method for obtaining the association analysis results by performing multimodal temporal association analysis on the trainee behavioral feature sequence through a cognitive-physiological dynamic association knowledge graph includes: Obtain the operational sequence matching degree, visual sequence matching degree, and physiological sequence matching degree corresponding to the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence.

[0012] Based on the operation sequence matching degree, visual sequence matching degree, and physiological sequence matching degree, a set of candidate behavioral feature pattern nodes is selected from multiple behavioral feature pattern nodes in the cognitive-physiological dynamic association knowledge graph, and preliminary weights are determined.

[0013] Deep temporal correlation analysis is performed on the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence of each candidate behavioral feature pattern node in the candidate behavioral feature pattern node set to obtain the collaborative consistency verification result. The comprehensive confidence of each candidate behavioral feature pattern node is calculated based on the collaborative consistency verification result and the preliminary weight. The target behavioral feature pattern node is determined from the candidate behavioral feature pattern node set based on the comprehensive confidence of the node as the correlation analysis result.

[0014] Furthermore, the method for obtaining the operational sequence matching degree, visual sequence matching degree, and physiological sequence matching degree corresponding to the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence includes: The matching degree of the operation sequence is obtained by performing dynamic time warping calculation on the operation feature subsequence and the standard operation feature sequence corresponding to the behavioral feature pattern node in the cognitive-physiological dynamic association knowledge graph. The standard operation feature sequence includes operation response time feature sequence, operation accuracy feature sequence and operation stability feature sequence.

[0015] The visual sequence matching degree is obtained by performing dynamic time warping calculation on the visual scanning feature subsequence and the standard visual scanning feature sequence corresponding to the behavioral feature pattern node in the cognitive-physiological dynamic association knowledge graph. The standard visual scanning feature sequence includes the visual scanning frequency distribution feature sequence, the visual fixation duration distribution feature sequence, and the visual transfer path feature sequence.

[0016] The physiological response feature subsequence is dynamically time-warped with the standard physiological response feature sequence corresponding to the behavioral feature pattern node in the cognitive-physiological dynamic association knowledge graph to obtain the physiological sequence matching degree. The standard physiological response feature sequence includes heart rate variability feature sequence, skin conductance response amplitude feature, and skin conductance response duration feature sequence.

[0017] Furthermore, the method for obtaining the collaborative consistency verification result by performing deep temporal correlation analysis on the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence of each candidate behavioral feature pattern node in the candidate behavioral feature pattern node set includes: Key operational event sequences are extracted from the operational feature subsequence, visual attention event sequences are extracted from the visual scanning feature subsequence, and physiological stress event sequences are extracted from the physiological response feature subsequence.

[0018] Calculate the first causal association strength between the key operational event sequence and the visual attention event sequence, calculate the second causal association strength between the visual attention event sequence and the physiological stress event sequence, and calculate the third causal association strength between the key operational event sequence and the physiological stress event sequence.

[0019] A temporal causal network is constructed based on the first causal association strength, the second causal association strength, and the third causal association strength. The temporal causal network is matched with the typical causal paths predefined for the candidate behavioral feature pattern nodes in the cognitive-physiological dynamic association knowledge graph to obtain the path matching degree as the collaborative consistency verification result.

[0020] Furthermore, the method for extracting key operational event sequences from the operational feature subsequence, visual attention event sequences from the visual scanning feature subsequence, and physiological stress event sequences from the physiological response feature subsequence includes: Change point detection is performed on the operation feature subsequence to identify the moment when the operation behavior changes significantly. The identified moment points are then classified into key operation event sequences according to predefined key operation event types.

[0021] The visual scanning feature subsequence is subjected to gaze clustering analysis to identify the moment points of visual attention region transition. The identified moment points are labeled with events according to a predefined visual attention pattern to form a visual attention event sequence.

[0022] Peak detection and baseline drift correction are performed on the physiological response feature subsequence to identify the moment points of physiological state abrupt change. The identified moment points are then classified into physiological stress event sequences according to predefined physiological stress types.

[0023] Furthermore, the method for generating corresponding learning resource sequences from the cognitive-physiological dynamic association knowledge graph based on the target cognitive strategy defects and target physiological response patterns includes: Based on the identified target cognitive strategy deficiencies, a first set of learning resources with corrective associations is queried in the cognitive-physiological dynamic association knowledge graph; based on the target physiological response patterns, a second set of learning resources with regulatory associations is queried in the cognitive-physiological dynamic association knowledge graph.

[0024] The first and second learning resource sets are sorted according to the progressive dependency relationship between learning resource nodes to form a basic learning resource sequence; the basic resource sequence is dynamically adapted to generate an adapted resource sequence based on the real-time skill mastery status of the learners.

[0025] The adapted resource sequence is optimized by cross-arranging cognitive training resources and physiological regulation resources to obtain the learning resource sequence.

[0026] Furthermore, the method for obtaining the learning resource sequence by cross-arranging and optimizing the adapted resource sequence with cognitive training resources and physiological regulation resources includes: Cognitive load assessment indicators are extracted from operational feature subsequences and visual scanning feature subsequences, and physiological recovery capacity indicators are extracted from physiological response feature subsequences. Based on the mapping relationship between cognitive-physiological state and training load tolerance, the appropriate training intensity range for trainees in the current training phase is determined.

[0027] Based on the optimal ratio between cognitive training units and physiological regulation units, the cognitive training resources and physiological regulation resources in the adapted resource sequence are reorganized. During the reorganization process, the insertion frequency and duration of physiological regulation resources are dynamically adjusted according to the stress recovery characteristics reflected by the physiological response feature subsequence, so as to obtain a learning resource sequence that conforms to the individual characteristics of the learners.

[0028] Furthermore, the method for dynamically adjusting the insertion frequency and duration of physiological regulatory resources based on stress recovery characteristics reflected in the physiological response feature subsequence includes: Heart rate variability and skin conductance response trend features are extracted from the physiological response feature subsequence, and the current stress recovery status of the trainees is identified based on the trend features.

[0029] An initial adjustment plan is generated based on the predefined recommended insertion frequency and duration parameters according to the current stress recovery state; after fine-tuning the initial adjustment plan by combining the cognitive load change trends reflected in the operational feature subsequence and visual scan feature subsequence, the insertion frequency and duration of physiological regulatory resources are obtained.

[0030] On the other hand, based on the same inventive concept, the present invention also provides a knowledge graph-based intelligent recommendation system for learning resources, the system comprising: a multimodal behavior data acquisition module, a behavior feature sequence extraction module, a multimodal temporal correlation analysis module, a target defect and pattern recognition module, and a learning resource generation module, wherein each module is sequentially connected in communication. The multimodal behavior data acquisition module is used to collect multimodal behavior data of trainees in preset simulated dangerous scenarios. The multimodal behavior data includes operational behavior data, eye movement trajectory data, and physiological monitoring data.

[0031] The behavioral feature sequence extraction module is used to extract features from the multimodal behavioral data to obtain a trainee behavioral feature sequence, which includes an operational feature subsequence, a visual scanning feature subsequence, and a physiological response feature subsequence.

[0032] The multimodal temporal correlation analysis module is used to perform multimodal temporal correlation analysis on the trainee's behavioral feature sequence through a cognitive-physiological dynamic correlation knowledge graph to obtain the correlation analysis results. The cognitive-physiological dynamic correlation knowledge graph includes behavioral feature pattern nodes, cognitive strategy defect nodes, physiological response pattern nodes, and learning resource nodes.

[0033] The target defect and pattern recognition module is used to identify the trainee's target cognitive strategy defects and corresponding target physiological response patterns from the cognitive-physiological association knowledge graph based on the association analysis results.

[0034] The learning resource generation module is used to generate a corresponding learning resource sequence from the cognitive-physiological dynamic association knowledge graph based on the target cognitive strategy defects and the target physiological response patterns.

[0035] (3) Beneficial effects Compared with the prior art, the beneficial effects of the present invention are: 1. By collecting multimodal behavioral data of trainees in simulated dangerous scenarios and performing feature extraction and time-series correlation analysis, it is possible to accurately identify trainees' target cognitive strategy deficiencies and corresponding target physiological response patterns from both cognitive and physiological dimensions, thereby gaining a more comprehensive and accurate understanding of the trainees' situation.

[0036] 2. Based on the identified cognitive strategy deficiencies and physiological response patterns, corresponding learning resources are retrieved from the cognitive-physiological dynamic association knowledge graph. Multi-objective optimization sorting, dynamic difficulty adaptation, and cross-arrangement optimization are then performed to ensure that the generated learning resource sequence not only matches the learner's current knowledge and skill level but also reasonably combines cognitive training and physiological regulation resources, effectively improving the rationality and applicability of the recommended learning resources.

[0037] 3. By dynamically adjusting the insertion frequency and duration of physiological regulatory resources based on the stress recovery characteristics reflected in the physiological response feature subsequence, and by fine-tuning it in conjunction with the trend of cognitive load changes, the learning resource sequence can better adapt to the physical and mental state of trainees at different training stages. This effectively avoids problems such as poor training results or excessive burden on trainees caused by inappropriate training intensity, and further enhances the effectiveness of training and trainees' acceptance. Attached Figure Description

[0038] Figure 1 This is a flowchart of the intelligent recommendation method for learning resources based on knowledge graphs according to Embodiment 1 of the present invention.

[0039] Figure 2 This is a schematic diagram of the module composition of the knowledge graph-based intelligent recommendation system for learning resources according to Embodiment 2 of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] Example 1: As Figure 1 As shown, this embodiment provides an intelligent recommendation method for learning resources based on knowledge graphs. The method includes: S1. Collect multimodal behavioral data of trainees in preset simulated emergency scenarios. This multimodal behavioral data includes operational behavior data, eye-tracking data, and physiological monitoring data. When trainees enter the preset simulated emergency scenarios for training, multi-dimensional data collection is initiated simultaneously. For example, in flight simulation training, when a trainee encounters an engine malfunction in the simulator, their performance is continuously recorded through three channels. The first channel collects operational behavior data, specifically including the trainee's push-pull actions on the control stick, the magnitude and timing of throttle adjustments, the sequence of operations on various switches and buttons, and reaction times. This data is stored as a timestamped operational event stream with millisecond-level accuracy. The second channel collects eye-tracking data using an eye-tracking device, recording the trainee's gaze path across different areas of the instrument panel, the duration of fixation on key instruments, changes in pupil diameter, etc., thus reflecting the trainee's attention allocation pattern. The third channel uses a wearable physiological monitoring device to collect physiological monitoring data, acquiring real-time physiological indicators such as heart rate, heart rate variability, and skin conductance. These indicators objectively reflect the trainee's level of tension and stress state when dealing with emergencies.

[0042] S2. The multimodal behavioral data is used to extract features to obtain a student behavior feature sequence, which includes an operational feature subsequence, a visual scanning feature subsequence, and a physiological response feature subsequence. After acquiring the original multimodal behavioral data, feature extraction processing is performed to convert the original data into a structured student behavior feature sequence. During feature extraction, for operational behavioral data, feature parameters such as operational response time, operational accuracy, and operational stability are calculated to form an operational feature subsequence. For example, in an engine failure scenario, specific values ​​such as the response delay time from the occurrence of the failure to the start of response, the accuracy deviation of throttle adjustment, and the jitter amplitude of joystick control are extracted and organized into a sequence in chronological order. For eye-tracking trajectory data, features such as visual scanning frequency distribution, fixation duration distribution, and gaze transfer path are extracted to form a visual scanning feature subsequence. For example, it records which instruments the student scanned within 30 seconds, the duration of fixation on each instrument, and whether the scanning order conforms to the standard inspection procedure. Based on physiological monitoring data, features such as heart rate variability indicators, amplitude and duration of skin conductance response are extracted to form physiological response feature subsequences. For example, the curve of the trainee's heart rate rising from the normal 75 beats / minute to 110 beats / minute is recorded, as well as the time point when the skin conductance response peaks.

[0043] S3. The trainee's behavioral feature sequence is subjected to multimodal temporal association analysis through a cognitive-physiological dynamic association knowledge graph to obtain the association analysis results. The cognitive-physiological dynamic association knowledge graph includes behavioral feature pattern nodes, cognitive strategy defect nodes, physiological response pattern nodes, and learning resource nodes.

[0044] S4. Based on the correlation analysis results, identify the trainee's target cognitive strategy deficiencies and corresponding target physiological response patterns from the cognitive-physiological correlation knowledge graph. This involves more than just assessing whether the trainee successfully handled the engine malfunction; it's about diagnosing the underlying causes of operational errors or suboptimal operations. For example, analysis revealed that after the emergency occurred, the trainee's visual attention was excessively focused on the engine temperature gauge, repeatedly and for extended periods, while neglecting to scan other critical instruments such as the altimeter and speedometer. Simultaneously, the trainee's response time was significantly prolonged. This behavioral pattern was identified in the knowledge graph as the "attention tunneling effect," a typical cognitive strategy deficiency. At the same time, the trainee's physiological data showed a sharp increase in heart rate and a significant decrease in heart rate variability, with a persistently high skin conductance response. This pattern corresponds to the "high stress, low regulatory capacity" physiological response pattern in the knowledge graph. The trainee's target cognitive strategy deficiency was accurately identified as "incomplete information acquisition due to imbalanced attention allocation," and the corresponding target physiological response pattern was "decreased cognitive flexibility due to excessive tension."

[0045] S5. Generate a corresponding learning resource sequence from the cognitive-physiological dynamic association knowledge graph based on the target cognitive strategy defects and target physiological response patterns.

[0046] The method for obtaining association analysis results by performing multimodal temporal association analysis on the trainee behavioral feature sequence through a cognitive-physiological dynamic association knowledge graph includes: Obtain the operational sequence matching degree, visual sequence matching degree, and physiological sequence matching degree corresponding to the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence.

[0047] Based on the operational sequence matching degree, visual sequence matching degree, and physiological sequence matching degree, a set of candidate behavioral feature pattern nodes is selected from multiple behavioral feature pattern nodes in the cognitive-physiological dynamic association knowledge graph, and a preliminary weight is determined. The knowledge graph contains dozens of behavioral feature pattern nodes, such as calm rapid decision-making mode, tense delayed response mode, and distracted operation mode. The system sets a matching degree threshold, such as requiring at least two dimensions of operational sequence matching degree, visual sequence matching degree, and physiological sequence matching degree to exceed 0.65 in order to enter the set of candidate behavioral feature pattern nodes. Through this selection condition, the original dozens of nodes are reduced to a few candidate behavioral feature pattern nodes, forming a set of candidate behavioral feature pattern nodes. During the selection, a preliminary weight is determined for each candidate behavioral feature pattern node. The preliminary weight comprehensively considers the weighted average of the matching degree of the three dimensions and the credibility score of the behavioral feature pattern node in the knowledge graph. For example, a node with an operational matching degree of 0.78, a visual matching degree of 0.82, and a physiological matching degree of 0.71 has a preliminary weight of 0.77 after weighted calculation.

[0048] Deep temporal correlation analysis is performed on the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence of each candidate behavioral feature pattern node in the candidate behavioral feature pattern node set to obtain the synergistic consistency verification result. Based on the synergistic consistency verification result and preliminary weights, the comprehensive confidence score of each candidate behavioral feature pattern node is calculated. The target behavioral feature pattern node is then selected from the candidate behavioral feature pattern node set as the correlation analysis result based on the comprehensive confidence score. This synergistic consistency verification aims to answer the following questions: Did the trainee's visual attention guide operational decisions? Did the timing of the physiological stress response coincide with the key operational event? Do these cross-modal causal relationships conform to the expected typical patterns of behavioral pattern nodes? The comprehensive confidence score calculation formula typically uses a weighted fusion method, for example, comprehensive confidence score = preliminary weight × 0.6 + synergistic consistency verification result × 0.4. This considers both the basic score of single-dimensional matching and emphasizes the deep verification of multimodal synergy. For example, if the initial weight of the "delayed recognition stress response pattern" node is 0.77 and the collaborative consistency verification result is 0.85, then the overall confidence level = 0.77 × 0.6 + 0.85 × 0.4 = 0.802. This calculation is performed on all nodes in the candidate set, and the behavioral feature pattern node with the highest overall confidence level is finally selected as the target behavioral feature pattern node.

[0049] The method for obtaining the operational sequence matching degree, visual sequence matching degree, and physiological sequence matching degree corresponding to the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence includes: The matching degree of the operation sequence is obtained by performing dynamic time warping calculations on the operation feature subsequences and the standard operation feature sequences corresponding to the behavioral feature pattern nodes in the cognitive-physiological dynamic association knowledge graph. The standard operation feature sequences include operation response time feature sequences, operation accuracy feature sequences, and operation stability feature sequences. For matching operation feature subsequences, the trainee's actual operation sequence needs to be aligned and compared with the standard operation feature sequences corresponding to the behavioral feature pattern nodes in the knowledge graph. The standard operation feature sequences contain three key dimensions: the operation response time feature sequence records the standard time interval from receiving the danger signal to performing various operations; the operation accuracy feature sequence records the accuracy standard that each operation should achieve; and the operation stability feature sequence records the standard value of the stability index during the operation process. The dynamic time warping algorithm can handle this time scaling by finding the optimal time alignment path to calculate the similarity between the two sequences. The same principle is applied to the matching calculation of the operation accuracy feature sequences and the operation stability feature sequences.

[0050] The visual sequence matching degree is obtained by dynamically time-warping the visual scanning feature subsequence with the standard visual scanning feature sequence corresponding to the behavioral feature pattern node in the cognitive-physiological dynamic association knowledge graph. The standard visual scanning feature sequence includes a visual scanning frequency distribution feature sequence, a visual fixation duration distribution feature sequence, and a visual transfer path feature sequence. For matching the visual scanning feature subsequence, the trainee's actual visual behavior is compared with the standard visual scanning feature sequence. The standard visual scanning feature sequence also contains three dimensions: the visual scanning frequency distribution feature sequence defines which areas should be scanned and the scanning frequency at different time periods; the visual fixation duration distribution feature sequence specifies the fixation duration that should be maintained on key information areas; and the visual transfer path feature sequence describes the standard order in which the gaze moves between different information points. In flight training, when encountering an engine malfunction, the standard visual transfer path may be warning light → engine temperature gauge → altimeter → airspeed gauge → operating manual prompt → control panel, which constitutes an ordered attention allocation sequence. However, the trainee's actual path may be warning light → engine temperature gauge → engine temperature gauge (repeated fixation) → airspeed gauge → control panel, which involves repeated fixation and missed scans. When processing such sequence comparisons, the Dynamic Temporal Warping algorithm considers not only the order of elements but also handles cases of repetition, omission, and insertion, obtaining the visual sequence matching degree by calculating the minimum editing cost. For example, in the above case, due to a repeated gaze and the omission of scanning the altimeter and the instruction manual, the matching degree calculated by the algorithm might be 0.68.

[0051] The physiological response feature subsequences are dynamically time-warped with the standard physiological response feature sequences corresponding to the behavioral feature pattern nodes in the cognitive-physiological dynamic association knowledge graph to obtain the physiological sequence matching degree. The standard physiological response feature sequences include heart rate variability feature sequences, skin conductance response amplitude features, and skin conductance response duration feature sequences. For matching the physiological response feature subsequences, the trainee's physiological data is dynamically time-warped with the standard physiological response feature sequences. The standard physiological response feature sequences include heart rate variability feature sequences, skin conductance response amplitude features, and skin conductance response duration feature sequences. These sequences describe the change curves of the normal physiological stress response pattern under specific dangerous stimuli. For example, the standard heart rate variability feature sequence may show that: in the first 5 seconds of the dangerous situation, the heart rate should rise from the baseline of 75 beats / min to 95 beats / min, then remain in the 90-100 beats / min range for the next 10 seconds, and then gradually decline. The actual trainee's heart rate curve may rise rapidly from 75 to 110, then remain at a high level for a long time, and the decline rate is very slow. Although these two curves differ in shape, the dynamic time warping algorithm can find the optimal correspondence between them. For example, it aligns the rising segment of the trainee's curve with the rising segment of the standard curve, aligns the maintenance segments with each other, calculates the numerical deviation at each time point, and finally synthesizes the physiological sequence matching degree. If the trainee's heart rate rises too much and recovers too slowly, the matching degree may only be 0.62, indicating that the trainee's physiological stress response deviates significantly from the standard pattern. In actual training scenarios, different trainees have individual differences in reaction speed and decision-making rhythm. The same cognitive strategy deficiency may manifest at different time scales in different people, but the essential behavioral patterns are similar. Dynamic time warping can identify this essential similarity and avoid misjudgment due to differences in time scale.

[0052] The method for obtaining the collaborative consistency verification result by performing deep temporal correlation analysis on the operational feature subsequence, visual scanning feature subsequence, and physiological response feature subsequence of each candidate behavioral feature pattern node in the candidate behavioral feature pattern node set includes: Key operational event sequences are extracted from the operational feature subsequence, visual attention event sequences are extracted from the visual scanning feature subsequence, and physiological stress event sequences are extracted from the physiological response feature subsequence.

[0053] The system calculates the strength of a first causal association between the sequence of key operational events and the sequence of visual attention events, the strength of a second causal association between the sequence of visual attention events and the sequence of physiological stress events, and the strength of a third causal association between the sequence of key operational events and the sequence of physiological stress events. A sliding time window method is used when calculating the strength of the first causal association between the sequence of key operational events and the sequence of visual attention events. For example, if the "attention to abnormal indication event" occurs at T4=22.8 seconds and the "emergency lever pull event" occurs at T1=23.5 seconds, a difference of 0.7 seconds, and the visual event precedes the operational event, conforming to the causal chain of "visual recognition → decision → operation," the system calculates the strength of this temporal sequence. If the time interval is within a reasonable range (e.g., 0.5-3 seconds), and the object of visual attention has a semantic connection with the subsequent operation (attention to engine abnormality → execution of power reduction operation), then a strong causal association is determined, and a first causal association strength value of 0.82 is assigned. Similarly, the strength of the second causal association between the sequence of visual attention events and the sequence of physiological stress events is calculated. The difference of only 0.2 seconds between the "attention to abnormal indication event" T4=22.8 seconds and the "acute stress initiation event" T7=23.0 seconds indicates that visual recognition of threat information triggers a physiological stress response almost immediately, consistent with the physiological mechanism of stress response. Therefore, a second causal association strength value of 0.91 is assigned. For the third causal association strength between the key operational event sequence and the physiological stress event sequence, it is examined whether the physiological state affects operational performance. For example, after the "weakened regulatory ability event" T9=27.5 seconds, does the precision and stability of the operation decrease? If increased operational jitter is observed, a causal association is determined, and a third causal association strength value of 0.75 is assigned.

[0054] A temporal causal network is constructed based on the first, second, and third causal association strengths. This temporal causal network is then matched with typical causal paths predefined for candidate behavioral feature pattern nodes in the cognitive-physiological dynamic association knowledge graph to obtain the path matching degree, which serves as the result of the collaborative consistency verification. The temporal causal network is a directed graph structure where nodes represent events, edges represent causal associations, and the weight of an edge represents the causal association strength. For example, there is an edge from the "Attention Abnormal Indicator" node to the "Emergency Lever" node with a weight of 0.82; an edge from the "Attention Abnormal Indicator" node to the "Acute Stress Initiation" node with a weight of 0.91; and an edge from the "Weakened Regulatory Ability" node to the "Decreased Operational Stability" node with a weight of 0.75. The temporal causal network comprehensively depicts how cognitive and physiological processes interact and influence trainees when responding to emergencies. The constructed temporal causal network is then matched with typical causal paths predefined for candidate behavioral feature pattern nodes in the knowledge graph. Suppose a candidate behavioral feature pattern node is the "high-stress rapid response pattern." The typical causal path predefined for this pattern in the knowledge graph is: visual threat recognition → rapid physiological stress initiation → rapid action driven by stress → physiological recovery after action. The system checks whether the learner's temporal causal network contains this path, whether all key nodes in the path exist, and whether the causal association strength between nodes reaches a threshold. The system calculates the structural similarity and weight similarity of the two networks using a graph matching algorithm to obtain the path matching degree. If the learner's causal network highly matches the typical path, the path matching degree may reach 0.88, which is the collaborative consistency verification result, indicating that the learner's multimodal behavior indeed conforms to the intrinsic mechanism of the "high-stress rapid response pattern." If the matching degree is very low, such as only 0.35, it means that although some single-dimensional features are similar, the multimodal collaborative causal mechanism does not conform to this pattern, and this candidate behavioral feature pattern node needs to be excluded. Through this deep temporal association analysis, the authenticity of the behavioral pattern can be verified, ensuring that the recognition results are not only similar in surface features but also consistent in the underlying cognitive and physiological mechanisms.

[0055] The method for extracting key operational event sequences from the operational feature subsequence, visual attention event sequences from the visual scanning feature subsequence, and physiological stress event sequences from the physiological response feature subsequence includes: The algorithm detects changes in the operational feature subsequence to identify moments when significant changes occur in operational behavior. These moments are then categorized according to predefined key operational event types to form a key operational event sequence. The operational feature subsequence is a multi-dimensional time series containing changes in multiple parameters over time, such as operational force, speed, and direction. The principle of the change point detection algorithm is to find locations in the time series where statistical characteristics abruptly change. A sliding window method is used, for example, setting a 5-second window. The window slides across the sequence, calculating the mean and variance of the data within each window. When the difference in the mean between adjacent windows exceeds a certain threshold (e.g., more than 3 times the standard deviation), a change point is identified. For example, during a stable operation phase, the joystick force fluctuates slightly between 15-20 Newtons. When the situation escalates, the student suddenly pulls the joystick rapidly with a force of 65 Newtons. This sudden change in force is detected by the algorithm as a significant change point, timestamped at T=32.7 seconds. After detecting the change point, it needs to be categorized as an event. The system predefines a library of critical operation event types, including categories such as emergency braking, rapid steering, sudden power drop, and mode switching. Each category has a clear feature description. For example, emergency braking is defined as an increase in operating force of more than 50% within 0.5 seconds in the direction of deceleration. The detected change point features are matched with these predefined types. If the sudden increase in force at the change point matches the feature description of emergency braking, it is classified as an "emergency braking event". By traversing the entire operation feature subsequence, all change points are identified and classified to form an ordered sequence of critical operation events, such as [(T1=32.7 seconds, emergency braking), (T2=35.1 seconds, rapid steering), (T3=38.5 seconds, return to normal operation)].

[0056] The visual scan feature subsequence is subjected to gaze clustering analysis to identify the moment points of visual attention region transitions. Based on predefined visual attention patterns, the identified moment points are labeled to form a visual attention event sequence. The raw data collected by the eye-tracking device is a sequence of coordinate points of the trainee's gaze on the screen or in the real environment, recorded at a high sampling rate (e.g., 120 sampling points per second). These raw coordinate points need to be first identified by gaze point recognition to distinguish between gaze and saccades. Gazing refers to the state where the gaze is relatively stationary at a certain position, while saccades are the state where the gaze moves rapidly between different positions. A speed threshold method is used for differentiation: when the gaze movement speed is less than 30 degrees / second, it is determined to be gaze; when it is higher than this threshold, it is determined to be saccades. After identifying gaze points, spatially close gaze points need to be clustered into gaze regions. For example, on the instrument panel of a flight simulator, the trainee's gaze may generate 8 gaze points near the altimeter. K-means clustering or the DBSCAN algorithm is used to aggregate these points into a gaze region, "altimeter region," and the total gaze duration is calculated to be 2.1 seconds. A visual attention shift occurs when a trainee's gaze moves from one cluster area to another. For example, if a trainee shifts their gaze from the "altimeter area" to the "engine temperature gauge area" at T=28.3 seconds, this moment is identified as a critical moment. Predefined visual attention patterns include types such as active information search, abnormal signal response, operational target confirmation, and repetitive verification. Each type is associated with specific fixation duration, fixation sequence, and environmental context. For example, the "abnormal signal response" pattern is characterized by: the fixated area displaying abnormal information (such as a red warning light), a recent visual change in the area before fixation, and a short fixation duration (0.5-1.5 seconds) indicating rapid recognition. If the engine temperature gauge has just turned red and the trainee's fixation duration is 1.2 seconds, the visual shift event at T=28.3 seconds is labeled as an "abnormal signal response event." By processing the entire visual scan feature subsequence in this way, a visual attention event sequence is formed, such as [(T1=28.3 seconds, abnormal signal response), (T2=29.7 seconds, operation target confirmation), (T3=31.5 seconds, repeated verification)].

[0057] Peak detection and baseline drift correction are performed on the physiological response feature subsequence to identify the moment points of abrupt physiological state changes. Based on predefined physiological stress types, the identified moment points are categorized into events to form a physiological stress event sequence. Physiological signals such as heart rate and skin conductance typically contain multiple components: baseline, trend, noise, and the true physiological response. Baseline drift refers to the slow change of the DC component of the signal over time, which may be caused by factors such as sensor position movement and changes in ambient temperature. A high-pass filter is used to remove baseline drift; for example, a high-pass filter with a cutoff frequency of 0.05Hz is used to retain the high-frequency components reflecting true physiological changes. Simultaneously, a median filter is used to remove impulse noise caused by electromyographic interference. After preprocessing, the system performs peak detection on the clean signal. For heart rate signals, the R-wave peak point is detected, and the time interval between adjacent R waves is calculated to obtain an instantaneous heart rate sequence. Then, moments of significant increase or decrease are identified on the instantaneous heart rate sequence. For example, a trainee's heart rate remains stable at 78±5 beats / minute during the steady period, then begins to rise sharply at T=45.2 seconds, reaching 105 beats / minute within 5 seconds. This initial rise at T=45.2 seconds is identified as the physiological state abrupt change point. For the skin conductance response (SCR) signal, the rapid rise in amplitude is detected. The SCR typically appears 1-3 seconds after stimulation, characterized by a rapid increase in signal amplitude followed by a slow decline. The rising edge is detected using a first-order difference method; when the difference value continuously exceeds a threshold, it is determined as the starting point of the SCR peak. After identifying the abrupt change point, it is categorized according to predefined physiological stress types. Predefined types include heightened alertness, acute stress response, persistent tension, and stress recovery, each corresponding to a specific pattern of physiological indicator changes. For example, an "acute stress response" is defined as a heart rate increase of more than 20% within 5 seconds accompanied by a significant peak in the SCR. If a trainee's heart rate rises rapidly at T=45.2 seconds and skin conductance peaks at T=45.5 seconds, this time period is classified as an "acute stress response event". By traversing the entire physiological response characteristic subsequence, a physiological stress event sequence is formed, such as [(T1=45.2 seconds, acute stress response), (T2=52.8 seconds, persistent stress), (T3=68.3 seconds, stress recovery)].

[0058] The method for generating corresponding learning resource sequences from the cognitive-physiological dynamic association knowledge graph based on the target cognitive strategy defects and target physiological response patterns includes: Based on the identified target cognitive strategy deficiencies, a first set of learning resources with corrective associations is queried in the cognitive-physiological dynamic association knowledge graph. Based on the target physiological response patterns, a second set of learning resources with regulatory associations is queried in the same knowledge graph. For each identified target cognitive strategy deficiency, the path "cognitive strategy deficiency node → corrective association edge → learning resource node" is traversed to collect all learning resources with corrective associations, forming the first set of learning resources. For example, for the cognitive strategy deficiency of "imbalanced attention allocation," the knowledge graph may contain multiple corrective association edges, pointing to learning resource nodes such as systematic instrument scanning training, multi-task attention switching practice, information priority judgment training, and attention maintenance training under pressure. These resource nodes not only contain basic information about the resource (such as name, description, and difficulty level) but also metadata such as the corrective effect rating of the resource for the target deficiency, applicable conditions, and expected training duration. These resources are collected to form the first set of learning resources, assuming it contains 12 cognitive training resources. For the identified target physiological response patterns, a query is performed along the path "physiological response pattern node → regulatory association edge → learning resource node" to collect all learning resources with regulatory associations, forming a second learning resource set. For example, for the physiological response pattern of "high stress and low recovery ability," the regulatory association edge in the knowledge graph points to resource nodes such as progressive muscle relaxation training, rhythmic breathing regulation exercises, biofeedback-assisted training, and mindfulness attention training. These resources focus on helping learners reduce physiological arousal levels and improve self-regulation abilities, and are collected as the second learning resource set, assuming it contains 8 physiological regulation resources.

[0059] The first and second learning resource sets are sorted using a multi-objective optimization process based on the progressive dependencies between learning resource nodes to form a basic learning resource sequence. The basic resource sequence is then dynamically adapted to the difficulty level of the learners, generating a customized resource sequence, taking into account their real-time skill mastery. This sorting process is not a simple linear arrangement but considers the progressive dependencies between learning resource nodes. In the knowledge graph, there are "pre-dependency" edges between learning resource nodes, indicating that one resource needs to be learned after another. For example, "complex scene multi-task attention training" depends on "basic attention switching practice" because the former is more difficult and requires learners to master the basic skills first. The system constructs a directed acyclic graph to represent these dependencies and then uses a topological sorting algorithm to determine the learning order of the learning resources. Simultaneously, multi-objective optimization sorting is performed, comprehensively considering multiple optimization objectives: first, maximizing the correction effect by prioritizing learning resources with high scores for correcting target defects; second, rationalizing the training duration to avoid overly intensive or sparse training; and third, gradually evolving the cognitive load to ensure that the training difficulty increases gradually rather than in abruptly. A weighted scoring method is used to calculate a comprehensive score for each learning resource. Then, while satisfying dependency constraints, the resources are ranked from highest to lowest comprehensive score. For example, the ranked sequence of basic learning resources might be: [Basic Instrument Scanning Procedure Training (20 minutes) → Breathing Regulation Exercise (10 minutes) → Rapid Attention Switching Exercise (15 minutes) → Multitasking Parallel Processing Training (25 minutes) → Relaxation and Recovery Training (10 minutes) → Complex Scenario Comprehensive Drill (30 minutes)]. The difficulty of the basic resource sequence is dynamically adjusted based on the learner's real-time skill mastery status. Learner skill mastery information comes from historical training records and current ability assessment results. The system maintains a learner skill profile, recording the learner's mastery level in each skill dimension. For example, a learner's mastery level in "Instrument Scanning" is 65%, and their mastery level in "Stress Regulation" is 40%. The system adjusts the difficulty parameters of resources based on these levels of mastery. For resources corresponding to skills with lower mastery, it selects simplified versions of the learning resources or reduces the complexity of the training scenarios. For learning resources corresponding to skills with higher mastery, it skips introductory resources and directly arranges advanced resources, or adds distracting factors to the training scenarios to increase the challenge. For example, because the learner's mastery of "instrument scanning" is moderate, the system adjusts "Basic Instrument Scanning Procedure Training" to a medium difficulty mode, setting 6 key instruments in the training scenario instead of the standard 4. Because the learner's mastery of "stress regulation" is low, the system keeps "breathing regulation exercises" at an introductory difficulty level, using simple guided training. After difficulty adaptation, each resource is adjusted to match the learner's current level, forming an adapted resource sequence.

[0060] The adapted resource sequence is optimized by cross-arranging cognitive training resources and physiological regulation resources to obtain the learning resource sequence.

[0061] The method for obtaining a learning resource sequence by cross-arranging and optimizing the adapted resource sequence with cognitive training resources and physiological regulation resources includes: Cognitive load assessment indicators are extracted from operational feature subsequences and visual scanning feature subsequences, while physiological recovery ability indicators are extracted from physiological response feature subsequences. Based on the mapping relationship between cognitive-physiological state and training load tolerance, the appropriate training intensity range for trainees in the current training phase is determined. Cognitive load refers to the burden on the trainee's working memory during task execution; excessive cognitive load leads to decreased information processing efficiency and increased decision-making errors. Indicators extracted from operational feature subsequences include the degree of extension of operational reaction time, operational error rate, and degree of decline in operational coordination. For example, in the early stages of training, a trainee takes 8 seconds to complete a standard operational sequence with an operational error rate of 5%; after 30 minutes of training, it takes 11 seconds to complete the same operation, and the operational error rate rises to 12%, indicating that cognitive load is accumulating. Indicators extracted from visual scanning feature subsequences include fixation point dispersion, the degree of disorder in the scanning path, and the frequency of repeated fixations. Under normal circumstances, the visual scanning path of skilled trainees is orderly and efficient; when cognitive load is too high, the trainee's gaze becomes wandering, resulting in frequent repeated scanning. For example, trainees' average fixation dispersion was 15 regions of interest at the beginning of training, increasing to 23 regions of interest after 30 minutes of training. The number of repeated fixations on the same instrument also increased from 1.2 to 2.8 times; these are all signals of increased cognitive load. A weighted calculation method was used to derive the cognitive load assessment index, for example, a trainee's cognitive load assessment index at a certain moment was 7.8 (out of 10). Physiological recovery capacity reflects the speed and quality of the trainee's physiological system recovering to baseline levels after experiencing stress. The system focuses on the trend of heart rate variability (HRV) changes and the recovery characteristics of skin conductance response. Heart rate variability (HRV) is an important indicator for assessing the regulatory capacity of the autonomic nervous system; a higher HRV indicates good regulatory capacity, while a lower HRV indicates weakened regulatory capacity. The recovery time of skin conductance response is also an important indicator. Under normal circumstances, the skin conductance response begins to decline 3-5 seconds after the stimulation ends and basically returns to baseline within 10-15 seconds. If the recovery time is prolonged to more than 20 seconds, it indicates that the sympathetic nervous system is continuously activated and the recovery capacity is insufficient. These physiological indicators are used to calculate the physiological recovery capacity index. For example, a trainee's current physiological recovery capacity index is 4.2 (out of 10), indicating a relatively weak recovery capacity. After obtaining the cognitive load assessment index and the physiological recovery capacity index, the appropriate training intensity range for the trainee in the current training phase is determined based on the mapping relationship between cognitive-physiological state and training load tolerance. The mapping relationship is based on a knowledge base established from training science and human factors engineering research, describing the training intensity that trainees can effectively withstand under different cognitive-physiological states.For example, the mapping relationship stipulates that when the cognitive load index is between 6 and 8 and the physiological recovery ability index is between 4 and 6, the appropriate training intensity range is low to medium (intensity value 3-5), and high-intensity training (intensity value 7-9) should be avoided; only when the cognitive load index is below 5 and the physiological recovery ability index is above 7 is high-intensity training suitable. For the trainee in the aforementioned case (cognitive load 7.8, physiological recovery ability 4.2), after consulting the mapping relationship, it is determined that his appropriate training intensity range is 2-4 (low to medium level). This means that the subsequent training resources should be of lower difficulty and shorter duration, and physiological adjustment resources should be arranged as soon as possible for recovery.

[0062] Based on the optimal ratio between cognitive training units and physiological regulation units, the cognitive training resources and physiological regulation resources in the adapted resource sequence are reorganized. During the reorganization process, the insertion frequency and duration of physiological regulation resources are dynamically adjusted according to the stress recovery characteristics reflected in the physiological response feature subsequence, resulting in a learning resource sequence that conforms to the individual characteristics of the learner. The optimal ratio between cognitive training units and physiological regulation units also comes from training science research, which specifies the time ratio that cognitive training and physiological regulation should occupy in different training stages. Generally, in the early stage of training, when the learner is in good condition, the ratio of cognitive training to physiological regulation can be 7:3, that is, 70% of the time is used for cognitive training and 30% for regulation and recovery. In the middle stage of training, as fatigue accumulates, the ratio is adjusted to 6:4. In the later stage of training or when the learner is in poor condition, the ratio may need to be adjusted to 5:5 or even 4:6 to ensure sufficient recovery. For learners currently in a state of high cognitive load and weak physiological recovery ability, the optimal ratio is determined to be 5:5. Assuming the total duration of the adapted resource sequence is 60 minutes, with 50 minutes dedicated to cognitive training and 10 minutes to physiological regulation (a ratio of 8:2), this clearly does not meet the optimal ratio requirement. The system needs to be reorganized: either increase the quantity or duration of physiological regulation resources, or reduce cognitive training resources to bring the final ratio closer to 5:5. This could involve replacing some high-intensity cognitive training resources with medium- to low-intensity resources, or inserting more short-duration physiological regulation resources between cognitive training sessions.

[0063] The method for dynamically adjusting the insertion frequency and duration of physiological regulatory resources based on stress recovery characteristics reflected in physiological response feature subsequences includes: The system extracts the trend features of heart rate variability and skin conductance response from the physiological response feature subsequence, and identifies the trainee's current stress recovery state based on these trend features. The trajectory of heart rate variability throughout the training process is calculated using a moving window method, calculated every 5 minutes to obtain a time series. Statistical analysis of this time series feature allows the system to determine whether the decrease in heart rate variability is linear or accelerated, gradual or steep. For skin conductance response, the system focuses not only on its amplitude but also on its recovery time constant. After each stress stimulus, the skin conductance response rises rapidly and then decays exponentially. The system fits this decay curve and calculates the time constant τ; a larger τ value indicates slower recovery. For example, at the beginning of training, the trainee's skin conductance recovery time constant τ = 8 seconds; after 30 minutes of training, τ increases to 18 seconds, indicating a significantly slower recovery speed. The system also calculates the baseline drift trend of the skin conductance response. If the baseline continues to rise, it indicates that the sympathetic nervous system is in a state of continuous activation and has not received sufficient rest and recovery opportunities. The system predefines multiple stress recovery state types, each corresponding to a specific set of physiological indicator patterns. For example, the characteristics of "rapid recovery type" are: small decrease in HRV (total decrease <20%), stable skin conductance recovery time constant (τ change <30%), and no significant baseline drift; the characteristics of "slow recovery type" are: moderate decrease in HRV (total decrease 20-40%), prolonged skin conductance recovery time constant (τ increase 30-60%), and slight baseline drift; the characteristics of "difficult recovery type" are: large decrease in HRV (total decrease >40%), significantly prolonged skin conductance recovery time constant (τ increase >60%), and significant baseline drift.

[0064] An initial adjustment plan is generated based on predefined recommended insertion frequency and duration parameters according to the current stress recovery state. After fine-tuning the initial adjustment plan by combining the cognitive load change trends reflected in the operational feature subsequence and visual scan feature subsequence, the insertion frequency and duration of physiological regulation resources are obtained. The knowledge base stores recommended insertion frequency and duration parameters based on extensive training data and physiological research for each stress recovery state. For example, for the "difficult recovery" state, the recommended parameters are typically: an insertion frequency of once every 10-15 minutes, a duration of 10-15 minutes, and a regulation intensity of medium to high intensity (including a combination of techniques such as deep breathing and progressive muscle relaxation). An initial adjustment plan is generated based on these recommended parameters. Assuming the current training plan has a total duration of 60 minutes, the initial plan is generated according to the rule of inserting once every 12 minutes for 12 minutes each time. That is, 12 minutes of physiological regulation resources are inserted at the 12th, 24th, 36th, and 48th minutes, for a total of 48 minutes of physiological regulation time. This plan is reasonable from a physiological recovery perspective and can provide trainees with sufficient recovery opportunities. However, a purely physiologically-based approach may conflict with cognitive training needs. Mastering cognitive skills requires a certain amount of practice time and intensity. If physiological adjustments are too frequent or prolonged, it will compress the cognitive training time and affect skill improvement. Therefore, it is necessary to fine-tune the initial adjustment plan by combining the cognitive load variation trends reflected in the operational feature subsequences and visual scanning feature subsequences, to find the optimal balance between physiological recovery and cognitive training. System analysis of cognitive load variation trends revealed that cognitive load does not increase monotonically but exhibits a fluctuating pattern. Cognitive load varies depending on the type of training task performed. For example, during basic operational procedure training (minutes 0-15), cognitive load remained at a low to medium level (4.5-5.5); during multi-task attention allocation training (minutes 15-30), cognitive load rose to a high to medium level (6.5-7.5); during complex scenario emergency response training (minutes 30-45), cognitive load reached a high level (8.0-9.0); and during the summary and review period (minutes 45-60), cognitive load dropped back to a low level (3.5-4.5). The system identified two peak cognitive load periods: minutes 15-30 and minutes 30-45. These two periods represent the times when trainees' cognitive systems experience the greatest stress and are most prone to cognitive fatigue. Based on this cognitive load trend, the initial adjustment plan was fine-tuned in three aspects. First, the timing of insertion was adjusted, prioritizing the insertion of physiological regulatory resources after the peak of cognitive load. The initial plan inserted adjustment at the 12-minute mark, but at this time the cognitive load was not high (5.0) and there was no need for immediate adjustment. However, after the 15-minute mark, the cognitive load began to rise. If adjustment were inserted at the 12-minute mark, it would interrupt the ongoing training rhythm.The adjustment at 12 minutes was postponed to 15 minutes, just before the end of the low-to-medium load task and the beginning of the high-load task, serving as a transitional buffer. Similarly, the cognitive load peaked at 30 minutes, and the initial protocol's adjustment at 24 minutes was too early; it was adjusted to 28 minutes to provide preventative recovery before the peak. The initial protocol's adjustment at 36 minutes was appropriate during the high-load task phase and remained unchanged. The need for adjustment decreased at 48 minutes during the low-load summary phase; it was considered to postpone it or merge it with the end of training. Second, the duration was adjusted to match the intensity of the cognitive load. The higher the cognitive load, the longer the subsequent recovery time required. The adjustment inserted at 36 minutes occurred after nearly 6 minutes of extremely high cognitive load (8.0-9.0), which, although short in duration, was very intense; the system extended the duration to 15 minutes to ensure sufficient recovery. The adjustment inserted at 15 minutes occurred before a lower cognitive load (4.5-5.5), and its duration could be appropriately shortened to 10 minutes. In this way, by comprehensively considering the physiological recovery characteristics and cognitive load change trends of trainees, it not only ensures that trainees have sufficient physiological recovery time, but also maintains the continuity and effectiveness of cognitive training.

[0065] Example 2: Based on the same inventive concept, such as Figure 2 As shown, this embodiment also provides a knowledge graph-based intelligent recommendation system for learning resources. The system includes: a multimodal behavior data acquisition module, a behavior feature sequence extraction module, a multimodal temporal correlation analysis module, a target defect and pattern recognition module, and a learning resource generation module. The modules are connected in sequence for communication. The multimodal behavior data acquisition module is used to collect multimodal behavior data of trainees in preset simulated dangerous scenarios. The multimodal behavior data includes operational behavior data, eye movement trajectory data, and physiological monitoring data.

[0066] The behavioral feature sequence extraction module is used to extract features from the multimodal behavioral data to obtain a trainee behavioral feature sequence, which includes an operational feature subsequence, a visual scanning feature subsequence, and a physiological response feature subsequence.

[0067] The multimodal temporal correlation analysis module is used to perform multimodal temporal correlation analysis on the trainee's behavioral feature sequence through a cognitive-physiological dynamic correlation knowledge graph to obtain the correlation analysis results. The cognitive-physiological dynamic correlation knowledge graph includes behavioral feature pattern nodes, cognitive strategy defect nodes, physiological response pattern nodes, and learning resource nodes.

[0068] The target defect and pattern recognition module is used to identify the trainee's target cognitive strategy defects and corresponding target physiological response patterns from the cognitive-physiological association knowledge graph based on the association analysis results.

[0069] The learning resource generation module is used to generate a corresponding learning resource sequence from the cognitive-physiological dynamic association knowledge graph based on the target cognitive strategy defects and the target physiological response patterns.

[0070] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0071] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A knowledge graph-based learning resource intelligent recommendation method, characterized in that, The method comprises: Collecting multi-modal behavior data of trainees in a preset simulated dangerous situation, the multi-modal behavior data comprising operation behavior data, eye movement trajectory data and physiological monitoring data; Extracting features from the multi-modal behavior data to obtain a trainee behavior feature sequence, the trainee behavior feature sequence comprising an operation feature sub-sequence, a visual scanning feature sub-sequence and a physiological response feature sub-sequence; Performing multi-modal time sequence correlation analysis on the trainee behavior feature sequence by means of a cognitive-physiological dynamic correlation knowledge graph to obtain a correlation analysis result, the cognitive-physiological dynamic correlation knowledge graph comprising a behavior feature mode node, a cognitive strategy defect node, a physiological response mode node and a learning resource node; Identifying a target cognitive strategy defect and a corresponding target physiological response mode of the trainee from the cognitive-physiological correlation knowledge graph according to the correlation analysis result; Generating a corresponding learning resource sequence from the cognitive-physiological dynamic correlation knowledge graph according to the target cognitive strategy defect and the target physiological response mode. 2.The knowledge graph-based learning resource intelligent recommendation method according to claim 1, characterized in that, The method of performing multi-modal time sequence correlation analysis on the trainee behavior feature sequence by means of a cognitive-physiological dynamic correlation knowledge graph to obtain a correlation analysis result comprises: Obtaining operation sequence matching degrees, visual sequence matching degrees and physiological sequence matching degrees corresponding to the operation feature sub-sequence, the visual scanning feature sub-sequence and the physiological response feature sub-sequence; Screening a candidate behavior feature mode node set from a plurality of behavior feature mode nodes in the cognitive-physiological dynamic correlation knowledge graph according to the operation sequence matching degrees, the visual sequence matching degrees and the physiological sequence matching degrees and determining a preliminary weight; Performing deep time sequence correlation analysis on the operation feature sub-sequence, the visual scanning feature sub-sequence and the physiological response feature sub-sequence of each candidate behavior feature mode node in the candidate behavior feature mode node set to obtain a collaborative consistency verification result, calculating a comprehensive confidence of each candidate behavior feature mode node according to the collaborative consistency verification result and the preliminary weight, and determining a target behavior feature mode node from the candidate behavior feature mode node set as the correlation analysis result according to the comprehensive confidence. 3.The knowledge graph-based learning resource intelligent recommendation method according to claim 2, characterized in that, The method of obtaining operation sequence matching degrees, visual sequence matching degrees and physiological sequence matching degrees corresponding to the operation feature sub-sequence, the visual scanning feature sub-sequence and the physiological response feature sub-sequence comprises: Performing dynamic time warping calculation on the operation feature sub-sequence and a standard operation feature sequence corresponding to a behavior feature mode node in the cognitive-physiological dynamic correlation knowledge graph to obtain an operation sequence matching degree, the standard operation feature sequence comprising an operation response time feature sequence, an operation precision feature sequence and an operation stability feature sequence; Performing dynamic time warping calculation on the visual scanning feature sub-sequence and a standard visual scanning feature sequence corresponding to a behavior feature mode node in the cognitive-physiological dynamic correlation knowledge graph to obtain a visual sequence matching degree, the standard visual scanning feature sequence comprising a visual scanning frequency distribution feature sequence, a visual fixation duration distribution feature sequence and a visual transfer path feature sequence; The physiological response feature subsequence is subjected to dynamic time warping calculation with a standard physiological response feature sequence corresponding to a behavior feature mode node in the cognitive-physiological dynamic correlation knowledge graph to obtain a physiological sequence matching degree, and the standard physiological response feature sequence includes a heart rate variation feature sequence, a skin electrical response amplitude feature, and a skin electrical response duration feature sequence. 4.The knowledge graph-based learning resource intelligent recommendation method according to claim 3, characterized in that, The method for obtaining the collaborative consistency verification result by performing deep time sequence correlation analysis on the operation feature subsequence, the visual scanning feature subsequence, and the physiological response feature subsequence of each candidate behavior feature mode node in the candidate behavior feature mode node set includes: extracting a key operation event sequence from the operation feature subsequence, extracting a visual attention event sequence from the visual scanning feature subsequence, and extracting a physiological stress event sequence from the physiological response feature subsequence; calculating a first causal correlation strength between the key operation event sequence and the visual attention event sequence, calculating a second causal correlation strength between the visual attention event sequence and the physiological stress event sequence, and calculating a third causal correlation strength between the key operation event sequence and the physiological stress event sequence; constructing a time sequence causal network according to the first causal correlation strength, the second causal correlation strength, and the third causal correlation strength, and performing matching calculation on the time sequence causal network and a typical causal path predefined for the candidate behavior feature mode node in the cognitive-physiological dynamic correlation knowledge graph to obtain a path matching degree as the collaborative consistency verification result. 5.The knowledge graph-based learning resource intelligent recommendation method according to claim 4, characterized in that, The method for extracting the key operation event sequence from the operation feature subsequence, the visual attention event sequence from the visual scanning feature subsequence, and the physiological stress event sequence from the physiological response feature subsequence includes: performing change point detection on the operation feature subsequence to identify time points at which operation behavior significantly changes, performing event classification on the identified time points according to a predefined key operation event type to form a key operation event sequence; performing gaze clustering analysis on the visual scanning feature subsequence to identify time points at which visual attention regions are switched, performing event annotation on the identified time points according to a predefined visual attention mode to form a visual attention event sequence; performing peak detection and baseline drift correction on the physiological response feature subsequence to identify time points at which physiological states mutate, and performing event classification on the identified time points according to a predefined physiological stress type to form a physiological stress event sequence. 6.The knowledge graph-based learning resource intelligent recommendation method according to claim 1, characterized in that, The method for generating a corresponding learning resource sequence from the cognitive-physiological dynamic correlation knowledge graph according to the target cognitive strategy defect and the target physiological response mode includes: querying a first learning resource set having a correction association in the cognitive-physiological dynamic correlation knowledge graph according to the identified target cognitive strategy defect, and querying a second learning resource set having a regulation association in the cognitive-physiological dynamic correlation knowledge graph according to the target physiological response mode; The first learning resource set and the second learning resource set are sorted according to the progressive dependence relationship between the learning resource nodes to form a basic learning resource sequence; the basic resource sequence is dynamically difficulty-adapted in combination with the real-time skill mastering state of the trainee to generate an adapted resource sequence; The adapted resource sequence is cross-arranged and optimized for cognitive training resources and physiological regulation resources to obtain a learning resource sequence. 7.The knowledge graph-based learning resource intelligent recommendation method according to claim 6, characterized in that, The method for cross-arranging and optimizing the adapted resource sequence for cognitive training resources and physiological regulation resources to obtain a learning resource sequence comprises: Cognitive load evaluation indexes are extracted from the operation feature subsequence and the visual scanning feature subsequence, and physiological recovery capability indexes are extracted from the physiological response feature subsequence; the cognitive load evaluation indexes and the physiological recovery capability indexes are used to determine the suitable training intensity range of the trainee in the current training stage according to the mapping relationship between the cognitive-physiological state and the training load bearing capacity; The cognitive training resources and the physiological regulation resources in the adapted resource sequence are reorganized according to the optimal matching relationship between the cognitive training unit and the physiological regulation unit; in the reorganization process, the insertion frequency and the duration of the physiological regulation resources are dynamically adjusted according to the stress recovery characteristics reflected by the physiological response feature subsequence to obtain a learning resource sequence that conforms to the individual characteristics of the trainee. 8.The knowledge graph-based learning resource intelligent recommendation method according to claim 7, characterized in that, The method for dynamically adjusting the insertion frequency and the duration of the physiological regulation resources according to the stress recovery characteristics reflected by the physiological response feature subsequence comprises: Variation trend characteristics of heart rate variability and skin electric reaction are extracted from the physiological response feature subsequence, and the current stress recovery state of the trainee is identified according to the variation trend characteristics; An initial adjustment scheme is generated by querying the recommended insertion frequency and duration parameters predefined according to the current stress recovery state; the initial adjustment scheme is fine-tuned in combination with the cognitive load variation trend reflected by the operation feature subsequence and the visual scanning feature subsequence to obtain the insertion frequency and the duration of the physiological regulation resources.

9. The learning resource intelligent recommendation system based on a knowledge graph, characterized in that, The system comprises a multi-modal behavior data acquisition module, a behavior feature sequence extraction module, a multi-modal time sequence correlation analysis module, a target defect and pattern recognition module, and a learning resource generation module, which are sequentially and communicatively connected; The multi-modal behavior data acquisition module is configured to acquire multi-modal behavior data of the trainee in a preset simulated dangerous situation, wherein the multi-modal behavior data comprises operation behavior data, eye movement trajectory data, and physiological monitoring data; The behavior feature sequence extraction module is configured to extract features from the multi-modal behavior data to obtain a trainee behavior feature sequence, wherein the trainee behavior feature sequence comprises an operation feature subsequence, a visual scanning feature subsequence, and a physiological response feature subsequence; The multi-modal time sequence correlation analysis module is configured to perform multi-modal time sequence correlation analysis on the trainee behavior feature sequence through a cognitive-physiological dynamic correlation knowledge graph to obtain a correlation analysis result, wherein the cognitive-physiological dynamic correlation knowledge graph comprises a behavior feature mode node, a cognitive strategy defect node, a physiological response mode node, and a learning resource node. A target defect and pattern recognition module is configured to recognize a target cognitive strategy defect and a corresponding target physiological response pattern of the learner from the cognitive-physiological correlation knowledge graph according to the correlation analysis result. A learning resource generation module is configured to generate a corresponding learning resource sequence from the cognitive-physiological dynamic correlation knowledge graph according to the target cognitive strategy defect and the target physiological response pattern.

Images