Real-time learning intervention method and system based on dynamic cognitive load evaluation

Abstract

The invention discloses a real-time learning intervention method and system based on dynamic cognitive load evaluation, and relates to the technical field of learning intervention. Comprising the following steps: extracting a feature vector sequence representing a cognitive state; dividing the cognitive state into multi-level interval cognitive states, and establishing a cognitive state coding sequence; constructing a cognitive state transition trend matrix according to the cognitive state switching frequency; extracting a state jump amplitude between two continuous cognitive states, taking the state jump amplitude as a dynamic adjustment factor to adjust an opening threshold, and marking an initial intervention node; constructing an intervention record chain table, and performing linkage comparison with the cognitive state transition trend matrix; and when the continuous intervention failure accumulatively exceeds a set tolerance frequency, calculating an amplitude difference of the cognitive state before and after intervention, and adaptively adjusting a cognitive state transfer damping factor according to the amplitude difference. The real-time performance, the stability and the decision robustness of the intelligent intervention system in the learning process are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of learning intervention, and in particular to a real-time learning intervention method and system based on dynamic cognitive load assessment. Background Art

[0002] In recent years, with the development of artificial intelligence, multimodal sensing, and educational big data, intelligent cognitive assessment and personalized intervention based on learning process data have gradually become important research areas in intelligent education. Traditional learning interventions often rely on static behavioral rules or teaching experience, making it difficult to respond in real time to the dynamic fluctuations of learners' cognitive states. To this end, relevant technologies have begun to introduce real-time sensing, multimodal data fusion, and deep learning algorithms to model behavioral performance during the learning process, identify potential variables such as attention level, learning cognitive state, and cognitive load, and thus implement intelligent feedback and auxiliary intervention. Such methods have played a positive role in educational evaluation, personalized teaching recommendations, and learning behavior correction, promoting profound changes in educational informatization. However, in existing technical solutions, cognitive state modeling remains relatively crude, resulting in delayed intervention responses or a single adjustment mechanism.

[0003] For example, CN109859078A describes a student learning behavior analysis intervention method, device, and system. This method collects video data and environmental parameters in the classroom, constructs feature fusion data, and uses a neural network model to analyze the correlation between air composition and student behavioral cognitive state. On this basis, a normal value model is established to achieve judgment of student behavioral cognitive state and environmental intervention. This method has certain real-time and systematic properties in feature extraction and fusion analysis, but there are two prominent problems: First, this method focuses on modeling the static correlation between the external environment and macroscopic behavior, ignoring the continuous evolution of cognitive state over time and internal jump characteristics, making it difficult to characterize the microscopic changes in individual cognitive load; second, the intervention method is mainly based on environmental adjustment, lacking a closed-loop adaptive adjustment mechanism based on cognitive state feedback, which is prone to problems of ineffective intervention or excessive intervention. Therefore, this method has obvious deficiencies in the modeling granularity of cognitive state and the dynamic response capability of intervention.

[0004] CN118968555A proposes a method and system for intervening in the learning cognitive state of AI study rooms. This method uses infrared and depth cameras to capture head posture and learning material area data, determine learning cognitive state, and implement appropriate interventions. This method demonstrates certain technological advancements in image acquisition, enabling relatively accurate acquisition of posture data and inference of attention levels. However, this method still faces the following limitations: First, cognitive state determination relies primarily on static features and single-frame information, lacking modeling of cognitive state evolution paths and transition trends, making it unsuitable for complex cognitive fluctuation scenarios. Second, its intervention mechanism is relatively simple, outputting control signals based solely on the current cognitive state. It fails to form a feedback regulation path after multiple interventions, lacks a posteriori modeling and dynamic optimization of intervention effectiveness, and is difficult to adapt to individual learner differences. Therefore, while this solution improves the accuracy of self-study monitoring, it still leaves room for improvement in terms of responsiveness and feedback in cognitive load regulation. Summary of the Invention

[0005] In view of the problems existing in existing learning intervention technologies based on behavior analysis or image monitoring, the present invention is proposed.

[0006] Therefore, the present invention aims to solve the technical bottlenecks of the above-mentioned existing technologies, such as the inability to accurately capture cognitive state transitions and the lack of intervention feedback linkage and regulation mechanisms.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a real-time learning intervention method based on dynamic cognitive load assessment, which includes extracting a sequence of feature vectors representing cognitive states based on historical behavioral data and current operating parameters in the learning process; dividing the cognitive state into multi-level interval cognitive states in combination with the cognitive state gradient change rate, and establishing a cognitive state coding sequence using the interval cognitive states; constructing a cognitive state transfer trend matrix based on the cognitive state coding sequence and the cognitive state switching frequency; extracting the state jump amplitude between two consecutive cognitive states from the cognitive state coding sequence, and constructing a sliding trigger window, using the state jump amplitude as a dynamic adjustment factor to adjust the opening threshold, and marking the initial intervention node; when the initial intervention node is marked, constructing an intervention record linked list, and performing a linkage comparison with the cognitive state transfer trend matrix; when the cumulative number of consecutive intervention failures exceeds the set tolerance number, calculating the amplitude difference of the cognitive state before and after the intervention, and when it exceeds the upper limit of the mean standard deviation of the historical state jump amplitude, it is used as a candidate factor to participate in the damping factor adjustment, so as to reduce the influence of short-term cognitive state fluctuations on subsequent intervention judgments.

[0009] As a preferred embodiment of the real-time learning intervention method based on dynamic cognitive load assessment of the present invention, the generation of the cognitive state feature vector includes: collecting the learner's behavioral modal data at the current time frame t; taking the current time frame t as the center, taking t0 frames forward and backward, and selecting a total of 2t frames before and after 0+1 The modal context tensor is constructed for each frame, with t0 being a constant. First-order differences are performed on the modal context tensor in the time dimension to obtain the change sequence of each modality. The nonlinear change rate of each modality is calculated as the modal weight and concatenated with the mean of the modalities to form the cognitive state feature vector of the current time frame. The nonlinear change rate is the weighted sum of the logarithmic differences of the change rates of each modality.

[0010] As a preferred embodiment of the real-time learning intervention method based on dynamic cognitive load assessment described in the present invention, the establishment of the cognitive state coding sequence includes: performing sliding window statistical analysis based on the nonlinear change rate on the continuous time series, extracting the median and skewness, and constructing dynamic interval boundaries; mapping all cognitive state feature vectors to multi-level cognitive state levels based on the dynamic interval boundaries; constructing a cognitive state coding sequence based on the divided cognitive state level sequence, and recording the direction and frequency of each pair of adjacent cognitive state transitions.

[0011] As a preferred solution of the real-time learning intervention method based on dynamic cognitive load assessment described in the present invention, the construction of the cognitive state transition trend matrix includes: extracting each pair of adjacent cognitive state transition pairs in chronological order to construct a cognitive state transition pair sequence set; within a sliding interval of a fixed window length, counting the local occurrence frequency of each pair of adjacent cognitive state transition pairs in the cognitive state transition pair sequence set, and accumulating them into a cognitive state transition count matrix; calculating the corresponding directional entropy for the cognitive state transition distribution of each row in the cognitive state transition count matrix to measure the directional dispersion degree of the current cognitive state to different cognitive states; and weightedly fusing the cognitive state transition count matrix and the corresponding directional entropy to construct a cognitive state transition trend matrix.

[0012] As a preferred solution of the real-time learning intervention method based on dynamic cognitive load assessment described in the present invention, wherein: the state jump amplitude is used as a dynamic adjustment factor to adjust the start threshold and mark the initial intervention node, including: setting a sliding window, backtracking from the current time frame t to form a sliding window sequence, and calculating the cumulative value of the sliding window state jump amplitude; setting a basic start threshold θ0, with the average state jump amplitude in the sliding window is the adjustment factor, and the threshold value θ(t) is adaptively updated:

[0013]

[0014] Where γ is the jump adjustment coefficient; if the cumulative value of the state jump amplitude of the current time frame exceeds the opening threshold, the corresponding time frame will be marked as the initial intervention node.

[0015] As a preferred solution of the real-time learning intervention method based on dynamic cognitive load assessment described in the present invention, the linkage comparison with the cognitive state transfer trend matrix includes: the intervention record linked list contains the pre-intervention cognitive state, intervention parameters, post-intervention cognitive state and delay time difference; the pre-intervention cognitive state and post-intervention cognitive state of each record in the intervention record linked list are mapped to the transfer path elements corresponding to the cognitive state transfer trend matrix, and based on the historical trend cumulative value in the cognitive state transfer trend matrix, it is judged whether the corresponding cognitive state transfer has trend support; if the historical trend cumulative value is lower than the preset reference level, it is deemed that there is no trend support and it is recorded as an intervention failure.

[0016] As a preferred embodiment of the real-time learning intervention method based on dynamic cognitive load assessment described in the present invention, the adaptive adjustment of the cognitive state transfer damping factor based on the amplitude difference includes: after calculating the amplitude difference, normalizing all amplitude differences to generate an amplitude difference vector sequence; setting a basic damping factor, and constructing a damping adjustment factor that is positively correlated with the corresponding amplitude difference based on the normalized amplitude difference between the cognitive states before and after the intervention.

[0017] In a second aspect, the present invention provides a real-time learning intervention system based on dynamic cognitive load assessment, comprising: a cognitive encoding module for extracting a sequence of feature vectors representing cognitive states based on historical behavioral data and current operating parameters during the learning process; dividing the cognitive state into multiple interval cognitive states based on the gradient change rate of the cognitive state, and establishing a cognitive state encoding sequence using the interval cognitive states;

[0018] A trend matrix module is used to construct a cognitive state transition trend matrix based on the cognitive state coding sequence and the cognitive state switching frequency;

[0019] The threshold adjustment module is used to extract the state jump amplitude between two consecutive cognitive states from the cognitive state coding sequence, construct a sliding trigger window, use the state jump amplitude as a dynamic adjustment factor to adjust the opening threshold, and mark the initial intervention node;

[0020] The intervention linkage module is used to construct an intervention record list after the initial intervention node is marked, and perform linkage comparison with the cognitive state transfer trend matrix;

[0021] The damping adjustment module is used to calculate the amplitude difference of the cognitive state before and after the intervention when the cumulative number of consecutive intervention failures exceeds the set tolerance. When it exceeds the upper limit of the mean standard deviation of the historical state jump amplitude, it participates in the damping factor adjustment as a candidate factor to reduce the impact of short-term cognitive state fluctuations on subsequent intervention judgments.

[0022] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the real-time learning intervention method based on dynamic cognitive load assessment as described in the first aspect of the present invention are implemented.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the real-time learning intervention method based on dynamic cognitive load assessment as described in the first aspect of the present invention are implemented.

[0024] The beneficial effects of the present invention are as follows: through cognitive state encoding, cognitive state transfer trend analysis and state jump amplitude adjustment mechanism, the present invention effectively improves the capture sensitivity of cognitive state changes, making the judgment of intervention nodes more accurate; compared with traditional strategies that rely on static thresholds or empirical rules, the present invention realizes dynamic adjustment of intervention trigger conditions, and has stronger adaptability and individual characteristics; at the same time, by constructing an intervention record list and comparing it with the trend matrix, the present invention can track the effectiveness of intervention and optimize the intervention strategy in time, further improving the efficiency of intervention; when the continuous intervention effect is not ideal, the cognitive state transfer damping factor can also be adjusted by amplitude difference to reduce the interference of local fluctuations on the overall evaluation and judgment. In summary, the present invention improves the real-time, stability and decision-making robustness of the intelligent intervention system during the learning process. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 A flowchart of the real-time learning intervention method based on dynamic cognitive load assessment;

[0027] Figure 2 This is a structural diagram of the real-time learning intervention system based on dynamic cognitive load assessment. DETAILED DESCRIPTION

[0028] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0029] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0030] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0031] As mentioned in the above background technology, traditional learning interventions mostly rely on static behavioral rules or teaching experience, and it is difficult to respond to the dynamic fluctuations of learners' cognitive states in real time. To this end, relevant technologies have begun to try to introduce real-time sensing, multimodal data fusion and deep learning algorithms to model behavioral performance in the learning process, identify potential variables such as attention level, learning cognitive state and cognitive load, and then achieve intelligent feedback and auxiliary intervention. This type of method has played a positive role in educational evaluation, personalized teaching recommendations, learning behavior correction, etc., and has promoted deep changes in educational informatization. However, in existing technical solutions, the modeling of cognitive states is still relatively extensive, resulting in delayed response of intervention measures or a single adjustment mechanism.

[0032] Figure 1 FIG is a flow chart of a real-time learning intervention method based on dynamic cognitive load assessment according to an embodiment of the present invention. Figure 1 As shown in the real-time learning intervention method based on dynamic cognitive load assessment, it includes:

[0033] S1: Based on the historical behavioral data and current operating parameters during the learning process, a feature vector sequence representing the cognitive state is extracted; combined with the gradient change rate of the cognitive state, the cognitive state is divided into multi-level interval cognitive states, and the interval cognitive states are used to establish a cognitive state coding sequence.

[0034] In the specific implementation, we first obtain a multi-dimensional modal behavior data set from the learner's interactive operations based on the behavioral data collection mechanism of the current time frame t. The corresponding data set includes multiple modal dimensions such as mouse click density, operation rhythm parameters, learning task dwell time, and interface switching frequency. The data of each dimension are uniformly stored in a modal vector structure to represent the comprehensive behavioral cognitive state at that time point.

[0035] In order to enhance the responsiveness of the current frame to the changes in the previous and next behaviors, we take the current time frame t as the center, take the t0 frame forward and backward, and select a total of 2t frames before and after. 0+1 The modal context tensor is constructed by the frame. If it is located near the first or last frame of the time series, the tensor is supplemented at both ends by mirroring the critical frame to ensure that the time dimension length is constant at 2t. 0+1 , t0 is a constant.

[0036] In order to reveal the direction and amplitude of changes in behavioral modes in continuous time frames, the modal context tensor is subjected to first-order differences in the time dimension to obtain the change sequence of each mode. That is, the change trajectory of each mode in the local time series can be obtained through the difference sequence.

[0037] Furthermore, the nonlinear change rate of each modality is calculated as a modal weight and concatenated with the modal mean to form the cognitive state feature vector for the current time frame. The nonlinear change rate is the weighted sum of the logarithmic differences of the change rates of each modality. This concatenated feature vector not only captures the behavioral change trend of the current frame within the local temporal window, but also retains the ability to express static behavioral levels.

[0038] It should be noted that in traditional methods, the mean of modal changes is often calculated directly, but the asymmetry and explosiveness of the change amplitude are ignored. The present invention constructs a weighted change rate of modal change by applying a nonlinear mapping to the modal differential value. The rate of change of each modality is converted in the form of a logarithmic transformation, and combined with the modal weight, the nonlinear change rate index corresponding to a single frame is calculated. The above-mentioned nonlinear change rate can not only suppress the interference of outliers on the judgment results, but also has the characteristics of being sensitive to sudden behavioral changes, thereby improving the accuracy and stability of feature expression.

[0039] Furthermore, the establishment of the cognitive state coding sequence includes the following steps:

[0040] A sliding window statistical analysis is performed based on the nonlinear rate of change on the continuous time series to extract the median and skewness, and construct dynamic interval boundaries. Based on the dynamic interval boundaries, all cognitive state feature vectors are mapped to a multi-level cognitive state scale (in the embodiment of the present invention, they are mapped to a five-level cognitive state scale, which is only an example and must be based on actual conditions), corresponding to five categories of cognitive load cognitive states: very low, low, medium, high, and very high (the division rule can be: interval = median ± skewness coefficient × standard deviation, which can be set according to actual conditions). A cognitive state coding sequence is constructed based on the divided cognitive state level sequence, and the direction and frequency of each pair of adjacent cognitive state transitions are recorded (for example, a positive transition from a low level to a high level and a negative transition from a high level to a low level). A reasonable transition interval is defined as a state transition direction of a single-step adjacent transition.

[0041] It's important to note that the median measures the central tendency of the rate of change, while the skewness describes the direction and degree of fluctuation. By using both as input variables for dynamic boundaries, we generate time-varying boundaries for cognitive state level intervals. This boundary demarcation mechanism outperforms static threshold demarcation methods, enabling adaptive adjustment of level intervals to suit the learner's current cognitive state distribution, thereby improving the effectiveness and accuracy of cognitive state mapping.

[0042] It can be seen that the present invention is different from traditional methods that only rely on single-point behavioral indicators. Through multimodal behavioral collaborative analysis, nonlinear change rate modeling and sliding window statistical mechanism, the present invention achieves high-precision identification of the dynamic evolution of cognitive state, providing a data basis and cognitive state judgment basis for subsequent real-time intervention decisions. It has strong scalability and can adapt to multi-platform data access scenarios and personalized learning behavior differences.

[0043] S2: Based on the cognitive state coding sequence, a cognitive state transition trend matrix is ​​constructed according to the cognitive state switching frequency.

[0044] The construction of the cognitive state transition trend matrix includes the following steps:

[0045] a. Extract each pair of adjacent cognitive state transitions in chronological order to construct a cognitive state transition pair sequence set.

[0046] b. To enhance the local sensitivity of cognitive state trend modeling and suppress the influence of global outliers, the local occurrence frequency of each pair of adjacent cognitive state transition pairs in the cognitive state transition pair sequence set is counted within a sliding interval of a fixed window length and accumulated into a cognitive state transition count matrix.

[0047] Specifically, the sliding window length is the same as the window length of step S1, 2t 0+1 , in each sliding window 2t 0+1 The number of occurrences of all cognitive state transitions is counted internally, and the number is accumulated into a two-dimensional structure to form a cognitive state transition count matrix, where each row of the matrix corresponds to the cognitive state as the starting cognitive state, and each column corresponds to the cognitive state as the target cognitive state. The value of the element in the matrix represents the cumulative number of occurrences of the transition from the starting cognitive state to the target cognitive state.

[0048] By covering the entire cognitive state sequence through window sliding, we can capture long-term trends while retaining short-term transition characteristics, thereby improving the time response capability of cognitive state modeling.

[0049] c. Calculate the corresponding directional entropy for each row of the cognitive state transition distribution in the cognitive state transition count matrix to measure the degree of directional dispersion of the current cognitive state to different cognitive states.

[0050] It should be noted that within each row, there are multiple transition paths to the target cognitive state, and the distribution of these paths often directly affects the adaptability of the cognitive state regulation strategy. To measure whether the transition direction of the current starting cognitive state is concentrated or discrete, the relative frequency of each transition target in the row is calculated and the directional entropy value is calculated accordingly.

[0051] Among them, the directional entropy reflects the degree of dispersion of the corresponding cognitive state to each target cognitive state. The higher the entropy value, the more random or divergent the cognitive state diffusion path is. The lower the entropy value, the more significant the dominant path is, which has trend characteristics. The present invention can effectively avoid the misjudgment of transition fluctuations caused by accidental noise and improve the structural robustness of subsequent trend judgment.

[0052] d. Perform weighted fusion of the cognitive state transition count matrix and the corresponding directional entropy to construct the cognitive state transition trend matrix.

[0053] Specifically, the transition frequency value at each position in the cognitive state transition count matrix is ​​weighted and fused with the directional entropy of the corresponding starting state to generate the final cognitive state transition trend matrix. The weighting method can set different fusion coefficients according to application requirements to adjust the weight distribution of frequency and directional entropy in the overall trend. For example, in some task scenarios, the cognitive state shows a typical concentrated diffusion path, so the weight of directional entropy can be increased to highlight the stability of the path; in exploratory tasks, frequency can be more critical and used to indicate the main state transition hotspots.

[0054] Through the above construction process, the cognitive state transition trend matrix finally formed not only retains the directionality of state level transitions, but also embeds the stability characteristics of the transition path, which can be used as the core basis for subsequent dynamic judgment of intervention trigger thresholds, state anomaly identification, and path confidence backtracking. In conventional models, only the frequency of cognitive state transitions is often counted, without considering the biased characteristics of the distribution of migration directions between cognitive states, resulting in an inability to accurately judge whether the state change has a trend structure. The present invention effectively compensates for the above shortcomings by adding directional entropy and performing trend fusion, significantly improving the interpretability and judgment accuracy of cognitive state transition paths.

[0055] Furthermore, the sliding window mechanism employed by this invention possesses local sensitivity and temporal progression, enabling dynamic perception of the temporal evolutionary trends of learner state fluctuations, resulting in high timeliness. In real-world learning scenarios, learners' cognitive state fluctuations are highly nonlinear and abrupt. By employing windowed trend statistics and information entropy constraints, this invention effectively distinguishes normal state transitions from potentially abnormal ones, enhancing the foresight and accuracy of subsequent intervention node assessments.

[0056] S3: Extract the state jump amplitude between two consecutive cognitive states from the cognitive state encoding sequence, construct a sliding trigger window, use the state jump amplitude as a dynamic adjustment factor to adjust the opening threshold, and mark the initial intervention node.

[0057] S3.1: First, all consecutive adjacent state levels in the cognitive state coding sequence are compared one by one, and based on the numerical setting of the state level, the amplitude difference between the two previous and next cognitive states is calculated.

[0058] Among them, the amplitude difference represents the mutation intensity of the learner's cognitive state between two adjacent time frames, thus constituting a state jump amplitude sequence of continuous cognitive states.

[0059] S3.2: After the state jump amplitude sequence is generated, a trend weight indicator is further added to the state jump amplitude component at each time point. The trend weight is derived from the directional entropy of the previous stage and can reflect whether the cognitive state transition path is regular or deviant in the historical state distribution.

[0060] It should be noted that the trend weight is extracted by directly reading the value corresponding to the cognitive state transfer trend matrix as the trend weight; if the corresponding path has a sparse historical frequency or a high directional entropy, the trend weight is low, indicating an unconventional transition.

[0061] S3.3: Use the state transition amplitude as a dynamic adjustment factor to adjust the start threshold and mark the initial intervention nodes, including:

[0062] S3.3.1: Set another sliding window κ, trace back from the current time frame t to form a sliding window sequence, and calculate the cumulative value of the sliding window state jump amplitude; among them, the cumulative value of the state jump amplitude is used to measure the overall level of state change intensity within the sliding window κ. If this value is high, it means that the learner's cognitive state shows continuous instability or high volatility during this period.

[0063] Specifically, when calculating the jump cumulative value, the current implementation method adopts the absolute value summation method of the jump amplitude within the window to enhance the cumulative recognition ability of frequent but weak transition states. The jump amplitude corresponding to each time point is calculated by the state level difference. After taking the absolute value, it is summed one by one within the sliding window to form a jump cumulative value that represents the total intensity of the state change in this section.

[0064] S3.3.2: Set the basic opening threshold θ0 to the average state jump amplitude within the sliding window κ is the adjustment factor, and the threshold value θ(t) is adaptively updated:

[0065]

[0066] Among them, γ is the jump adjustment coefficient.

[0067] S3.3.3: If the cumulative value of the state jump amplitude of the current time frame exceeds the opening threshold, the time frame is determined to be a fluctuation node where cognitive abnormalities may exist, and the corresponding time frame is marked as the initial intervention node to form the intervention candidate cognitive state set.

[0068] S4: After the initial intervention node is marked, a linked list of intervention records is constructed and compared with the cognitive state transition trend matrix.

[0069] First, based on the set of candidate cognitive states identified in step S3, the key state change information corresponding to each initial intervention node is extracted one by one. This includes the cognitive state level before the intervention, the applied intervention parameter configuration, the target state level reported after the intervention behavior is completed, and the response delay difference before and after the intervention in the time dimension. These four key fields are combined into a structured quadruple, corresponding to the pre-intervention state, intervention parameters, post-intervention state, and delay time difference, respectively. This quadruple data is then appended to the intervention record linked list.

[0070] Secondly, the pre-intervention cognitive state and post-intervention cognitive state of each record in the intervention record list are mapped to the corresponding transfer path elements in the cognitive state transfer trend matrix. Based on the cumulative value of the historical trend in the cognitive state transfer trend matrix, it is judged whether the corresponding cognitive state transfer has trend support. If the cumulative value of the historical trend is lower than the preset reference level, it is considered to have no trend support and is recorded as an intervention failure.

[0071] Furthermore, if some prognostic cognitive states do not fall within the expected transfer interval, the intervention is determined to be ineffective, and the intervention parameters are updated by going back to the cognitive state node before the intervention.

[0072] The expected transition interval is determined by the hierarchical mapping logic established during the previous cognitive state classification, representing the range of target states that can be reasonably transitioned from a given cognitive level. If the post-intervention state is in a state of leapfrogging or reverse degradation based on the hierarchical logic, it indicates that the current intervention parameters have failed to effectively guide the cognitive state toward the desired direction, and the intervention should be considered ineffective.

[0073] For cases where interventions have been determined to be invalid, the initial node of the current intervention record is traced back and the parameter correction mechanism is entered:

[0074] First, the most recent successful state trajectory of the pre-intervention state in the intervention record list is taken out as the basic reference for the correction operation; combined with the low-weight direction of the intervention failure state path in the cognitive state transfer trend matrix, the candidate state path with the closest directionality to the corresponding failure path but with a higher trend weight is selected; then, by adjusting the adjustable factors in the intervention parameters (such as intervention intensity, intervention rhythm, or intervention duration frames, etc.), a backup configuration set for the next round of intervention actions is constructed.

[0075] In order to prevent the system from falling into infinite correction and repeated intervention on the same node, the present invention sets an intervention node retry counter. Whenever an intervention node fails and triggers a parameter update, its counter increments once. If the number of retries exceeds the maximum tolerance threshold, the corresponding node will be marked as "frozen" in the intervention record list and the intervention response will be stopped; otherwise, the corrected intervention configuration will be re-judged in the next round.

[0076] Through the above approach, the present invention can provide a flexible parameter retry mechanism and a strongly constrained response termination mechanism in intervention failure scenarios, taking into account both system intervention robustness and resource investment efficiency.

[0077] S5: When the cumulative number of consecutive intervention failures exceeds the set tolerance, the amplitude difference of the cognitive state before and after the intervention is calculated. When it exceeds the upper limit of the mean standard deviation of the historical state jump amplitude, it is used as a candidate factor to participate in the damping factor adjustment to reduce the impact of short-term cognitive state fluctuations on subsequent intervention judgments.

[0078] First, based on the marked intervention failure records, the pre-intervention state and post-intervention state of each intervention operation in the record are extracted, and the corresponding cognitive state feature vectors are taken out respectively.

[0079] For the above two groups of cognitive state feature vectors, the amplitude difference is calculated dimension by dimension according to the unified modal feature sorting method, and the absolute value of the difference is taken to represent the total magnitude change of the state response caused by the intervention.

[0080] Furthermore, adaptively adjusting the cognitive state transfer damping factor according to the amplitude difference includes:

[0081] After calculating the amplitude differences, all amplitude differences are normalized to generate an amplitude difference vector sequence. A basic damping factor is set, and based on the normalized amplitude differences between the cognitive states before and after the intervention, a damping adjustment factor is constructed that is positively correlated with the corresponding amplitude difference. The damping adjustment factor is designed as a monotonically increasing linear function, ensuring that the larger the amplitude difference, the higher the damping strength. The damping adjustment factor is directly applied to the trend response value of the corresponding path in the cognitive state transition trend matrix, resulting in a trend value attenuation adjustment. By suppressing the values ​​of the cognitive state transition trend matrix, the selectivity weight of the current path in the trend comparison is reduced, thereby reducing the probability of triggering the path again in subsequent decisions.

[0082] Further, such as Figure 2 As shown, this embodiment also provides a real-time learning intervention system based on dynamic cognitive load assessment, including:

[0083] The cognitive encoding module 100 is used to extract a sequence of feature vectors representing cognitive states based on historical behavioral data and current operating parameters during the learning process; divide the cognitive state into multiple interval cognitive states based on the cognitive state gradient change rate, and use the interval cognitive states to establish a cognitive state encoding sequence;

[0084] A trend matrix module 200 is used to construct a cognitive state transition trend matrix based on the cognitive state coding sequence and the cognitive state switching frequency;

[0085] The threshold adjustment module 300 is used to extract the state transition amplitude between two consecutive cognitive states from the cognitive state coding sequence, construct a sliding trigger window, use the state transition amplitude as a dynamic adjustment factor to adjust the activation threshold, and mark the initial intervention node;

[0086] The intervention linkage module 400 is used to construct an intervention record linked list after the initial intervention node is marked, and perform linkage comparison with the cognitive state transition trend matrix;

[0087] The damping adjustment module 500 is used to calculate the amplitude difference of the cognitive state before and after the intervention when the cumulative number of consecutive intervention failures exceeds the set tolerance. When it exceeds the upper limit of the mean standard deviation of the historical state jump amplitude, it is used as a candidate factor to participate in the damping factor adjustment to reduce the impact of short-term cognitive state fluctuations on subsequent intervention judgments.

[0088] This embodiment also provides a computer device suitable for a real-time learning intervention method based on dynamic cognitive load assessment, comprising a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the real-time learning intervention method based on dynamic cognitive load assessment proposed in the above embodiment.

[0089] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0090] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the real-time learning intervention method based on dynamic cognitive load assessment as proposed in the above embodiment.

[0091] In summary, through cognitive state encoding, cognitive state transfer trend analysis and state jump amplitude adjustment mechanism, the present invention effectively improves the capture sensitivity of cognitive state changes, making the judgment of intervention nodes more accurate; compared with traditional strategies that rely on static thresholds or empirical rules, the present invention realizes dynamic adjustment of intervention trigger conditions, and has stronger adaptability and individual characteristics; at the same time, by constructing an intervention record list and linking and comparing it with the cognitive state transfer trend matrix, the present invention can track the effectiveness of intervention and timely optimize the intervention strategy, further improving the efficiency of intervention; when the continuous intervention effect is not ideal, the cognitive state transfer damping factor can also be adjusted by amplitude difference to reduce the interference of local fluctuations on the overall evaluation and judgment. In summary, the present invention improves the real-time, stability and decision-making robustness of the intelligent intervention system during the learning process.

[0092] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A real-time learning intervention method based on dynamic cognitive load assessment, characterized by: include: Extracting a sequence of feature vectors representing cognitive states based on historical behavioral data and current operating parameters during the learning process; Combined with the cognitive state gradient change rate, the cognitive state is divided into multi-level interval cognitive states, and the interval cognitive states are used to establish the cognitive state coding sequence; Based on the cognitive state coding sequence, constructing a cognitive state transition trend matrix according to the cognitive state switching frequency; Extracting the state transition amplitude between two consecutive cognitive states from the cognitive state coding sequence, constructing a sliding trigger window, using the state transition amplitude as a dynamic adjustment factor to adjust the activation threshold, and marking the initial intervention node; After the initial intervention node is marked, an intervention record list is constructed and linked and compared with the cognitive state transition trend matrix; When the cumulative number of consecutive intervention failures exceeds the set tolerance, the amplitude difference of the cognitive state before and after the intervention is calculated. When it exceeds the upper limit of the mean standard deviation of the historical state jump amplitude, it is used as a candidate factor to participate in the damping factor adjustment to reduce the impact of short-term cognitive state fluctuations on subsequent intervention judgments.

2. The real-time learning intervention method based on dynamic cognitive load assessment according to claim 1, characterized in that: The generation of the cognitive state feature vector includes: At the current time frame t, collect the learner's behavioral modal data; take the current time frame t as the center, take the t0 frame forward and backward, and select a total of 2t 0+1 The frame constructs the modality context tensor, t0 is a constant; Perform first-order difference on the modal context tensor in the time dimension to obtain the change sequence of each modality; The nonlinear change rate of each mode is calculated as the mode weight and concatenated with the mean of the mode to form the cognitive state feature vector of the current time frame; The nonlinear change rate is the weighted sum of the logarithmic differences of the change rates of each mode.

3. The real-time learning intervention method based on dynamic cognitive load assessment according to claim 2, characterized in that: The establishment of the cognitive state coding sequence includes: Perform sliding window statistical analysis based on the nonlinear change rate of continuous time series, extract the median and skewness, and construct dynamic interval boundaries; Mapping all cognitive state feature vectors to a multi-level cognitive state hierarchy according to dynamic interval boundaries; A cognitive state coding sequence is constructed based on the divided cognitive state level sequence, and the direction and frequency of each pair of adjacent cognitive state transitions are recorded.

4. The real-time learning intervention method based on dynamic cognitive load assessment according to claim 1, characterized in that: The construction of the cognitive state transfer trend matrix includes: Extract each pair of adjacent cognitive state transitions in time order to construct a cognitive state transition pair sequence set; Within a sliding interval of a fixed window length, the local occurrence frequency of each adjacent cognitive state transfer pair in the cognitive state transfer pair sequence set is counted and accumulated into a cognitive state transfer count matrix; For each row of the cognitive state transition distribution in the cognitive state transition count matrix, the corresponding directional entropy is calculated to measure the directional dispersion of the current cognitive state to different cognitive states; The cognitive state transition count matrix and the corresponding directional entropy are weightedly fused to construct the cognitive state transition trend matrix.

5. The real-time learning intervention method based on dynamic cognitive load assessment according to claim 1, characterized in that: The method of adjusting the start threshold using the state transition amplitude as a dynamic adjustment factor and marking the initial intervention node includes: Set the sliding window, trace back from the current time frame t to form a sliding window sequence, and calculate the cumulative value of the sliding window state jump amplitude; Set the basic opening threshold θ0 to the average state jump amplitude within the sliding window is the adjustment factor, and the threshold value θ(t) is adaptively updated: Among them, γ is the jump adjustment coefficient; If the cumulative value of the state jump amplitude of the current time frame exceeds the opening threshold, the corresponding time frame will be marked as the initial intervention node.

6. The real-time learning intervention method based on dynamic cognitive load assessment according to claim 1, characterized in that: The linkage comparison with the cognitive state transfer trend matrix includes: The intervention record linked list includes the cognitive state before intervention, intervention parameters, cognitive state after intervention and delay time difference; The pre-intervention cognitive state and post-intervention cognitive state of each record in the intervention record list are mapped to the corresponding transfer path element of the cognitive state transfer trend matrix. Based on the historical trend cumulative value in the cognitive state transfer trend matrix, it is judged whether the corresponding cognitive state transfer has trend support; if the historical trend cumulative value is lower than the preset reference level, it is considered to have no trend support and is recorded as an intervention failure.

7. The real-time learning intervention method based on dynamic cognitive load assessment according to claim 1, characterized in that: The adaptive adjustment of the cognitive state transfer damping factor according to the amplitude difference includes: After calculating the amplitude difference, all amplitude differences are normalized to generate an amplitude difference vector sequence; A basic damping factor was set, and based on the normalized amplitude difference between cognitive states before and after intervention, a damping adjustment factor was constructed that was positively correlated with the corresponding amplitude difference.

8. A real-time learning intervention system based on dynamic cognitive load assessment, based on the real-time learning intervention method based on dynamic cognitive load assessment according to any one of claims 1 to 7, characterized in that: Also includes: The cognitive encoding module is used to extract a sequence of feature vectors representing cognitive states based on historical behavioral data and current operating parameters during the learning process; Combined with the cognitive state gradient change rate, the cognitive state is divided into multi-level interval cognitive states, and the interval cognitive states are used to establish the cognitive state coding sequence; A trend matrix module is used to construct a cognitive state transition trend matrix based on the cognitive state coding sequence and the cognitive state switching frequency; The threshold adjustment module is used to extract the state jump amplitude between two consecutive cognitive states from the cognitive state coding sequence, construct a sliding trigger window, use the state jump amplitude as a dynamic adjustment factor to adjust the opening threshold, and mark the initial intervention node; The intervention linkage module is used to construct an intervention record list after the initial intervention node is marked, and perform linkage comparison with the cognitive state transfer trend matrix; The damping adjustment module is used to calculate the amplitude difference of the cognitive state before and after the intervention when the cumulative number of consecutive intervention failures exceeds the set tolerance. When it exceeds the upper limit of the mean standard deviation of the historical state jump amplitude, it participates in the damping factor adjustment as a candidate factor to reduce the impact of short-term cognitive state fluctuations on subsequent intervention judgments.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the real-time learning intervention method based on dynamic cognitive load assessment according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the real-time learning intervention method based on dynamic cognitive load assessment according to any one of claims 1 to 7 are implemented.

Images