Intelligent course recommendation system and method combined with cognitive level recognition

Abstract

The invention relates to the field of intelligent course recommendation, and discloses an intelligent course recommendation system and method combined with cognitive level recognition, and the method comprises the steps: constructing a multi-dimensional operation task scene through an interactive game task, and generating an initial cognitive state sequence; mapping the initial cognitive state sequence, and guiding an adaptive feature screening mechanism to highlight key behavior features in a specific emotional state so as to construct a cognitive state map; based on the cognitive state map, capturing short-term cognitive fluctuation and extracting core cognitive indexes in different time periods; fusing the cognitive level evolution trajectory with the historical learning trajectory of the student, the knowledge point mastering condition, the individual preference and the emotional state to generate a personalized cognitive level judgment result which reflects the real ability of the student in the current task and the learning demand of emotion adaptation; and dynamically generating a course recommendation strategy according to a personalized cognitive level judgment result. The method has the advantage of improving the judgment precision of the current cognitive level of the student.

Description

Technical Field

[0001] This invention relates to the field of intelligent course recommendation, specifically to an intelligent course recommendation system and method that combines cognitive level recognition. Background Technology

[0002] Existing intelligent course recommendation systems largely rely on students' historical learning data, including indicators such as exam scores, homework completion, study time, and classroom interaction frequency, to indirectly infer students' learning abilities and cognitive levels. While this approach can reflect overall learning trends with large samples, it has significant limitations in personalized recommendations. On one hand, historical data is updated infrequently, making it difficult to capture short-term cognitive fluctuations. On the other hand, grades and homework results are often influenced by test anxiety, differences in question types, and environmental factors, leading to biases in the cognitive levels obtained by the system. In practical applications, this recommendation method, which relies on lagging data, is prone to specific problems: for example, when a student's performance on a mid-term test is low due to temporary distraction in class, the system may incorrectly classify their cognitive level as declining, recommending courses that are too easy and providing insufficient learning challenge. Conversely, when a student performs well on a test due to short-term cramming, the system may misjudge their cognitive level as too high, recommending course content beyond their actual ability and increasing learning frustration. Such mismatches not only reduce the accuracy of course recommendations but may also weaken students' learning motivation and sustained engagement. Therefore, it is necessary to design an intelligent course recommendation system and method that combines cognitive level recognition to improve the accuracy of judging students' current cognitive level. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an intelligent course recommendation system and method that combines cognitive level recognition, which has the advantage of improving the accuracy of judging students' current cognitive level and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving the accuracy of judging students' current cognitive level, this invention provides the following technical solution: an intelligent course recommendation method combining cognitive level recognition, comprising the following steps: Multi-dimensional operational task scenarios are constructed through interactive game tasks, and behavioral indicator collection points are set in each task scenario to infer students' learning ability and emotional state, thereby generating an initial cognitive state sequence. The initial cognitive state sequence is mapped, and the student's real-time emotional state is used as a weight adjustment factor to guide the adaptive feature selection mechanism to highlight the key behavioral features under specific emotional states. At the same time, noise interference is removed and emotion-related cognitive indicators are strengthened, thereby constructing a cognitive state map. Based on the cognitive state map, an adaptive evolutionary mapping model combining multi-scale time series convolutional networks and reinforcement learning is used to predict students' cognitive levels. The influence coefficient of emotional state is introduced in real time to adjust the evolutionary trajectory of cognitive level in a weighted manner, capture short-term cognitive fluctuations and extract core cognitive indicators for different time periods. By integrating the evolutionary trajectory of cognitive level with students' historical learning trajectory, mastery of knowledge points, individual preferences and emotional state, personalized cognitive level assessment results are generated, reflecting students' true abilities in the current task and their emotionally appropriate learning needs. Based on the results of personalized cognitive level assessment, the parameters of interactive tasks are adjusted through periodic real-time feedback to dynamically generate course recommendation strategies.

[0005] Preferably, the process of generating the initial cognitive state sequence is as follows: By setting up multiple behavioral indicator collection points in each interactive task scenario, students' operational behavior, task completion time, error rate, operation sequence, and emotional state information are recorded in real time. The collected multi-source behavioral data is time-series encoded and supplemented with task scenario identifiers, environmental parameters, and task difficulty level information to form an initial cognitive state sequence with timestamps.

[0006] Preferably, the process of mapping the initial cognitive state sequence is as follows: The initial cognitive state sequence is constructed as a multi-dimensional feature vector, with the vector dimensions including behavioral indicator dimension, task scenario dimension, and real-time emotion weight dimension. Multidimensional feature vectors are input into a mapping network for standardization and high-dimensional mapping. By iteratively optimizing the mapping parameters, a balance between feature consistency and discriminativeness is achieved under different task scenarios. An emotional state weighting factor is introduced during the mapping process to adjust the cognitive behaviors that are sensitive to emotions, highlight key behavioral features, and suppress low-relevance noise.

[0007] Preferably, the process of guiding the adaptive feature selection mechanism to highlight key behavioral features under specific emotional states is as follows: An adaptive feature selection algorithm is used to iteratively analyze the mapped multidimensional feature vectors to calculate the contribution and correlation of each behavioral indicator in the current emotional state. High-contribution behaviors are assigned enhancement weights, while low-contribution and noisy features are removed to form a set of emotion-related enhancement features; Dynamically update feature importance scores to identify and highlight core cognitive behaviors in real time across different task scenarios.

[0008] Preferably, the process of constructing a cognitive state map is as follows: The key behavioral indicators and emotional weights after adaptive screening are mapped as graph nodes according to time series and task scenarios, and the nodes are connected by the behavioral dependency and the intensity of emotional influence as edge weights. By embedding nodes and normalizing the graph structure, a cognitive state graph is formed, which reflects the behavioral associations of students under different task scenarios and emotional conditions. By analyzing node centrality and behavioral dependency, key nodes and core behavioral paths in the graph are identified.

[0009] Preferably, the process of predicting students' cognitive levels using a multi-scale time-series convolutional network combined with an adaptive evolutionary mapping model of reinforcement learning is as follows: The cognitive state map is input into a multi-scale time series convolutional network to extract short-term and long-term cognitive features and capture the dynamic patterns of key behaviors as emotional states change. Reinforcement learning algorithms are embedded in convolutional networks to adaptively optimize network weights and emotion sensitivity parameters. The predicted continuous cognitive states are coded and time-labeled to identify core cognitive indicators and behavioral patterns in different time periods, and output the evolution trajectory of cognitive level.

[0010] Preferably, the process of weighted adjustment of the cognitive level evolution trajectory is as follows: The evolutionary trajectory of cognitive level is processed in a multidimensional weighted manner by combining real-time emotional state, task difficulty and individual student preferences; By analyzing short-term fluctuations and instantaneous capability changes in the trajectory using a sliding window mechanism, core cognitive indicators for each time period are extracted, and outliers or atypical fluctuations in the trajectory are removed and corrected.

[0011] Preferably, the process of integrating the evolutionary trajectory of cognitive level with students' historical learning trajectory, mastery of knowledge points, individual preferences, and emotional state is as follows: The core cognitive indicators are multidimensionally weighted and integrated with historical learning performance, knowledge mastery probability and individual preference data, and real-time emotional state regulation factors are embedded. The fused feature vector is standardized and aligned with time series, and the fused feature vector reflects the students’ abilities and emotional state in different task scenarios. The output comprehensive feature vector serves as the input for personalized cognitive level assessment, and the assessment result includes students' historical performance, real-time task execution, and emotional adaptation information.

[0012] The preferred process for dynamically generating course recommendation strategies is as follows: Based on comprehensive feature vectors, a multi-dimensional weighted reasoning method is used to generate personalized cognitive level assessment results, while considering the impact of students' emotional state on cognitive performance in real time. During the assessment process, core behavioral indicators for different knowledge points and task stages are dynamically analyzed to reflect students' ability performance and emotional adaptation in the current task. Mark and track abnormal or mutated behaviors in the judgment results, and output personalized cognitive level judgment results; Based on the results of individualized cognitive level assessment, the difficulty, content order, task allocation, and learning pace of interactive tasks are dynamically adjusted, while parameters are configured in conjunction with students' real-time cognitive load and emotional state. Through a periodic feedback mechanism, students' cognitive level and emotional fluctuations during tasks can be monitored in real time. The task parameters in the recommendation strategy are adjusted in multiple dimensions, including the weight of behavioral indicators, the allocation of task duration, and the order of knowledge connection, and the course recommendation strategy is output.

[0013] An intelligent course recommendation system that combines cognitive level recognition includes: Task perception module: Constructs multi-dimensional operation task scenarios and collects students' behavioral indicators and emotional states in real time to generate an initial cognitive state sequence; Feature mapping module: Maps the initial cognitive state sequence into a multi-dimensional feature vector, introduces emotional weights and highlights key behavioral features to construct a cognitive state map; Cognitive prediction module: Based on the cognitive state map, it uses a multi-scale time series convolutional network and reinforcement learning model to predict students' cognitive level, generate evolutionary trajectory and capture short-term fluctuations; Horizontal Fusion Module: This module integrates the evolutionary trajectory of cognitive level with historical learning trajectory, knowledge mastery, individual preferences, and emotional state to generate personalized cognitive level assessment results. The strategy generation module periodically adjusts task parameters and dynamically generates course recommendation strategies based on the results of personalized cognitive level assessment.

[0014] Compared with existing technologies, the present invention provides an intelligent course recommendation system and method that combines cognitive level recognition, which has the following beneficial effects: This invention constructs multi-dimensional operational task scenarios and collects student behavioral indicators and emotional states in real time, enabling a comprehensive and detailed reflection of students' immediate cognitive abilities and emotional adaptation in different learning tasks. Through mapping and adaptive feature filtering of the initial cognitive state sequence, key behavioral features are highlighted and noise interference is eliminated, achieving high-precision identification of students' core cognitive behaviors. An adaptive evolutionary mapping model combining a multi-scale time-series convolutional network based on cognitive state maps and reinforcement learning can effectively predict short-term fluctuations and long-term trends in students' cognitive levels. Weighted adjustments capture instantaneous ability changes, providing a dynamic and quantifiable representation of the cognitive level evolution trajectory. By integrating the cognitive level evolution trajectory with students' historical learning records, knowledge mastery, individual preferences, and emotional states in a multi-dimensional manner, personalized cognitive level assessment results accurately reflect students' true abilities and learning needs in the current task, balancing ability development and emotional adaptation. Based on the periodic real-time feedback and dynamic course recommendation strategy of the judgment results, the difficulty of tasks, the order of content, task allocation and learning pace can be optimized in real time. This allows the recommended content to continuously adapt to changes in students' cognitive and emotional states, thereby improving learning efficiency, enhancing learning initiative and participation, and effectively supporting personalized learning path planning. It also provides precise, dynamic and context-aware decision-making basis for intelligent education systems. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0016] 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.

[0017] Example 1: Please refer to Figure 1 As shown in the figure, an intelligent course recommendation method combining cognitive level recognition according to an embodiment of the present invention includes the following steps: S1: Construct multi-dimensional operational task scenarios through interactive game tasks, and set behavioral indicator collection points in each task scenario to infer students' learning abilities and emotional states, thereby generating an initial cognitive state sequence.

[0018] The process of generating the initial cognitive state sequence in S1 is as follows: By setting multiple behavioral indicator collection points in each interactive task scenario, student operation behavior, task completion time, error rate, operation sequence, and emotional state information are recorded in real time. When designing interactive tasks, each task is broken down into quantifiable operation units, such as clicking, dragging, selecting answers, or completing operation steps. Collection nodes are deployed at the corresponding positions of each operation unit. These collection nodes can be event listening modules on the software side or hardware sensors (such as touchpads, keyboards, mice, or haptic feedback devices). Each student's operation is timestamped to form an operation event stream. At the same time, the operation time is measured, and the task completion time is calculated by the time difference between the start and end. The operation results are judged in real time, and correct, incorrect, or time-out operations are recorded as error rate indicators. Error statistics for each task or task segment are accumulated and generated. Operation sequences, that is, the order of actions and execution paths of students to complete tasks, are collected. This information is used to analyze students' strategy choices and cognitive preferences. The collected multi-source behavioral data is time-series encoded and supplemented with task scenario identifiers, environmental parameters, and task difficulty level information to form an initial cognitive state sequence with timestamps. Each operation event is sorted according to its occurrence time to form a continuous time series. Behavioral indicators, operation duration, error information, and operation order are encoded into structured data formats (such as vectors or tables) to ensure that different types of data can be processed uniformly in the same sequence. A task scenario identifier is added to each record to distinguish different task types, enabling subsequent models to identify the meaning and dependencies of operations in different contexts. Environmental parameter information is added, such as the interface complexity of the current task, the number of interfering elements, the screen size of the learning device, or the lighting conditions, to ensure that each data point in the sequence reflects the external environment in which the operation takes place. A task difficulty level (such as easy, medium, or complex) is added to each data point, calculated based on a predefined difficulty standard or based on historical task completion, to facilitate subsequent analysis of students' behavioral performance at different difficulty levels. After the above processing, the generated initial cognitive state sequence not only includes the time, action, and result of each operation, but also the task scenario, environmental factors, difficulty level, and real-time emotional information.

[0019] S2: Map the initial cognitive state sequence, use the student's real-time emotional state as a weight adjustment factor, guide the adaptive feature selection mechanism to highlight the key behavioral features under specific emotional states, remove noise interference and strengthen emotionally related cognitive indicators, thereby constructing a cognitive state map.

[0020] The process of mapping the initial cognitive state sequence in S2 is as follows: The initial cognitive state sequence is constructed as a multi-dimensional feature vector, with dimensions including behavioral indicator dimension, task scenario dimension, and real-time emotion weight dimension. The data of each operation event in the initial cognitive state sequence is decomposed into different types of feature dimensions, including operation behavior type (such as click, drag, select), operation duration, error rate, and operation order. These are quantified or categorized as behavioral indicator dimension. The task scenario to which each operation belongs is encoded as an independent dimension, for example, by using one-hot encoding to represent different task modules or task types, so that the model can distinguish operation behaviors in different contexts. The real-time emotion state associated with each operation event is quantified, for example, by mapping emotion categories to numerical weights (positive, negative, neutral) or emotion intensity indicators to form the emotion weight dimension, so that the influence of emotion on cognitive behavior can be reflected in the subsequent feature mapping. The behavioral indicator dimension, task scenario dimension, and emotion weight dimension are concatenated to form a complete multi-dimensional feature vector for each operation event, ensuring that the feature vector can simultaneously reflect the comprehensive information of student operation behavior, task environment, and emotion state. Multidimensional feature vectors are input into a mapping network for standardization and high-dimensional mapping. By iteratively optimizing the mapping parameters, a balance between feature consistency and discriminability across different task scenarios is achieved. Multidimensional feature vectors are input into a deep mapping network (such as a fully connected network or a convolutional network) for processing and standardization. The values ​​of each feature dimension are adjusted to a uniform scale to eliminate the impact of differences in feature scales on model training. After standardization, the features are mapped to a high-dimensional feature space. The network's non-linear activation function enables the combination and interaction between different features, enhancing the expressive power of the features. Through iterative training and parameter optimization, the network gradually learns which features should maintain consistency to reflect common cognitive behaviors and which features should maintain discriminability to reflect individual differences or task characteristics in different task scenarios. This ensures that the mapping results are both comparable and discriminative across multiple task scenarios. During training, loss function constraints and gradient update methods are used to continuously adjust the network weights and bias parameters, so that the high-dimensional representation after mapping retains key behavioral information while reducing the influence of irrelevant or noisy features. In the mapping process, an emotional state weighting factor is introduced to adjust the weight of emotionally sensitive cognitive behaviors, highlighting key behavioral features while suppressing low-relevance noise. For the emotional dimension in each feature vector, the influence weight of emotional state on behavioral indicators is calculated, and highly relevant behavioral features are given higher weight coefficients, so that these features are further amplified in the high-dimensional mapping space. For behavioral features with low correlation to emotions or little relation to task objectives, their weights are reduced or their contribution to the overall feature representation is suppressed through attention mechanisms, thereby reducing noise interference. By introducing an emotional weighting layer in the mapping network, the emotional state of each operation event is dynamically integrated with the behavioral indicators, so that the final high-dimensional mapping features can not only highlight key cognitive behaviors, but also reflect the moderating effect of students' real-time emotions on behavioral patterns. The final output mapping feature vector is a high-dimensional representation that comprehensively reflects the cognitive information after weighting behavioral indicators, task scenarios, and emotional states. The process by which the adaptive feature selection mechanism in S2 highlights key behavioral features under specific emotional states is as follows: An adaptive feature selection algorithm is used to iteratively analyze the mapped multidimensional feature vectors to calculate the contribution and correlation of each behavioral indicator in the current emotional state. High-contribution behaviors are assigned enhancement weights, while low-contribution and noisy features are removed to form a set of emotion-related enhancement features; Dynamically update feature importance scores to identify and highlight core cognitive behaviors in real time across different task scenarios.

[0021] The process of constructing the cognitive state map in S2 is as follows: The adaptively filtered key behavioral indicators and emotional weights are mapped as graph nodes according to time series and task scenarios. Connections between nodes are established based on behavioral dependencies and the intensity of emotional influence. The mapped high-dimensional feature vectors are decomposed according to the behavioral indicator dimensions, making the representation of each behavioral indicator independent and evaluable. For each behavioral indicator, its contribution to the prediction of cognitive level is quantitatively evaluated in conjunction with the current student's real-time emotional state. For example, the sensitivity of the indicator to cognitive judgment under positive, high-stress, or anxious emotional states is analyzed. The correlation between behavioral indicators and other feature dimensions is calculated to determine whether a certain behavior is highly correlated with the task scenario or other key behaviors, so as to retain important linkage information in the screening process. Through an iterative analysis mechanism, the contribution and correlation of each behavioral indicator under different emotional states are continuously updated to ensure that the screening process can adapt to the student's emotional fluctuations and accurately capture behavioral features that have a significant impact on cognitive level. The graph structure is processed by node embedding and graph normalization to form a cognitive state graph, which reflects the behavioral correlations of students under different task scenarios and emotional conditions. Based on the calculated contribution, behavioral indicators that have a significant impact on the prediction of cognitive level are given higher weights, making them central in subsequent processing and graph construction. For features with low contribution and no significant correlation with task objectives or cognitive assessment, the system removes them through screening rules to reduce the interference of noise on the overall feature representation. Combined with emotional state information, highly emotionally sensitive indicators are included in the reinforcement feature set, so that the feature set not only reflects the importance of the behavior itself, but also reflects the moderating effect of emotions on behavioral performance. The output feature set is weighted and screened to form a set of reinforcement cognitive behavioral features for specific emotional states. By analyzing node centrality and behavioral dependency, key nodes and core behavioral paths in the cognitive state graph are identified. During task execution, the system continuously monitors students' operations and emotional states, dynamically evaluates the actual contribution of each feature, and adjusts the importance scores of the features. For behavioral features that still show high contribution in new task scenarios, their weights are maintained or further enhanced. For behaviors with declining contribution, their weights are reduced or temporarily removed to ensure that the feature set always focuses on core cognitive behaviors. Through a real-time update mechanism, the enhanced feature set can adapt to different task scenarios and emotional changes, achieving continuous identification and prominent expression of key cognitive behaviors. The dynamic adjustment process ensures that the cognitive state graph can stably reflect students' core cognitive characteristics in multi-task and multi-emotional environments, while eliminating the interference of noise or low-value features on graph construction.

[0022] S3: Based on the cognitive state map, an adaptive evolutionary mapping model combining multi-scale time series convolutional networks and reinforcement learning is used to predict students' cognitive levels. The influence coefficient of emotional state is introduced in real time to adjust the evolutionary trajectory of cognitive level in a weighted manner, capture short-term cognitive fluctuations and extract core cognitive indicators for different time periods.

[0023] The process of predicting students' cognitive levels using a multi-scale time-series convolutional network combined with an adaptive evolutionary mapping model of reinforcement learning in S3 is as follows: The cognitive state map is input into a multi-scale time-series convolutional network to extract short-term and long-term cognitive features and capture the dynamic patterns of key behaviors changing with emotional states. The constructed cognitive state map is flattened or embedded into a multi-dimensional vector space in time-series order, so that the behavioral indicators, emotional weights, and task scenario information of each node can be used as input features of the convolutional network. The convolutional network uses temporal convolutional kernels of different lengths (short kernels and long kernels) to process the input data simultaneously to capture short-term instantaneous fluctuations and long-term trend changes in students' cognitive states, ensuring sensitivity to rapid behavioral reactions and cross-task patterns. In the convolutional operation, each convolutional layer extracts local temporal patterns of node features while retaining cross-node task dependency information, ensuring that the changing patterns of key behaviors under different emotional states can be identified. Through multi-scale convolutional operations, the network can form multi-level cognitive feature representations, so that short-term operational errors, attention fluctuations, and long-term cognitive development trends are fully encoded. Reinforcement learning algorithms are embedded in convolutional networks to adaptively optimize network weights and emotion sensitivity parameters. During training, the reinforcement learning algorithm uses the cognitive level prediction accuracy and emotional state adaptability as reward functions to dynamically adjust the weights of the convolutional network and the emotion sensitivity coefficient of each behavioral indicator. When updating the gradient of specific convolutional kernels and connection weights in the network, the moderating effect of emotional state on feature response is considered, enabling the network to adaptively enhance or suppress the influence of certain behavioral features in different emotional scenarios. Reinforcement learning also continuously optimizes the prediction strategy of the network in continuous task sequences through a policy iteration mechanism, ensuring that the network can accurately track the cognitive evolution trend of students, rather than just making single-step predictions. This adaptive optimization ensures that the network can effectively extract the most distinctive cognitive features when facing different task scenarios, different emotional fluctuations, and different student behavior patterns. The predicted continuous cognitive states are coded and time-labeled to identify core cognitive indicators and behavioral patterns in different time periods, and output the cognitive level evolution trajectory. The continuous cognitive states output by the convolutional network are organized into a sequence in chronological order, and each time step is labeled to record behavioral indicator values, emotional weights, and task scenario information. The continuous state sequence is coded to ensure that the cognitive states output by the network not only retain numerical changes but also reflect the evolution of behavioral patterns over time, including the frequency, duration, and emotional sensitivity of core behaviors. Time labeling is used to mark key time nodes in the sequence to identify short-term fluctuation stages, task transition points, and stages with significant emotional influence, thereby highlighting the performance of core cognitive indicators in different time periods. The generated cognitive level evolution trajectory is a continuous time series that can reflect the changes in students' cognitive levels throughout the interactive task.

[0024] The process of weighted adjustment of the cognitive level evolution trajectory in S3 is as follows: By combining real-time emotional state, task difficulty, and individual student preferences, the cognitive level evolution trajectory is processed with multidimensional weighting. Emotional intensity and type information are extracted from real-time emotional data collected from students during interactive tasks to quantify the immediate impact of emotions on cognitive performance, giving personalized weights to cognitive level adjustments under different emotional states. The task difficulty coefficient is introduced into the trajectory weighting mechanism to give greater attention to cognitive performance during high-difficulty task phases, ensuring that the trajectory reflects students' true ability changes under different task loads. Individual student preference data, including learning style, attention concentration patterns, and operating habits, are integrated to assign personalized weights to cognitive indicators at each time step in the trajectory, making the weighted trajectory more consistent with students' individual characteristics. Finally, emotional weights, task difficulty weights, and individual preference weights are multidimensionally integrated to form a composite weighting coefficient, which is applied to each time point in the cognitive level evolution trajectory to achieve global adjustment. By analyzing short-term fluctuations and instantaneous ability changes in the trajectory using a sliding window mechanism, core cognitive indicators for each time period are extracted, and outliers or atypical fluctuations in the trajectory are removed and corrected. A sliding window analysis is applied to the weighted cognitive level evolution trajectory, dividing the trajectory into continuous short time segments to facilitate the capture of students' short-term cognitive fluctuations and instantaneous ability changes. Within each sliding window, the mean, variance, and trend of behavioral indicators are statistically analyzed to identify the most representative core cognitive indicators for that time period, such as operation speed, accuracy, and concentration, highlighting key performance indicators. Outliers or atypical fluctuations in the trajectory, such as unusually low or high scores within a short period, are assessed in conjunction with trends and emotional states in preceding and following time segments to remove noisy fluctuations. These are then corrected using interpolation or smoothing strategies to ensure the continuity and reliability of the trajectory. The generated multidimensional weighted trajectory not only reflects the evolution of students' cognitive levels throughout the task process but also highlights the core cognitive indicators for each time period while eliminating abnormal interference, forming a refined cognitive trajectory that can be used for personalized assessment and course recommendation.

[0025] S4: Integrate the cognitive level evolution trajectory with students' historical learning trajectory, knowledge mastery, individual preferences and emotional state to generate personalized cognitive level assessment results, reflecting students' true abilities in the current task and their emotionally appropriate learning needs.

[0026] The process in S4 that integrates the evolutionary trajectory of cognitive level with students' historical learning trajectory, mastery of knowledge points, individual preferences, and emotional state is as follows: This approach integrates core cognitive indicators with historical learning performance, knowledge mastery probability, and individual preference data through multidimensional weighting, and embeds a real-time emotional state adjustment factor. Core cognitive indicators from the resulting cognitive level evolution trajectory, such as operation speed, accuracy, and attention concentration, are matched with student performance data at each stage of their historical learning trajectory. Historical performance serves as a benchmark to reflect long-term learning trends. Knowledge mastery probability information is integrated into the feature vector. By analyzing students' mastery of different knowledge points, the cognitive weight of knowledge points with high mastery probabilities is appropriately reduced for the current task, while the weight of knowledge points with low mastery probabilities is increased to highlight learning difficulties. Students' individual preference information, including learning style, interest bias, attention cycle, and operation habits, is mapped as weighting factors to adjust the contribution of different cognitive indicators in the integrated feature vector, thereby achieving personalized feature expression. A real-time emotional state adjustment factor is introduced to dynamically adjust the weights of cognitive indicators. For example, when students are highly focused or under mild stress, the weights of core cognitive indicators are appropriately increased to reflect the immediate impact of emotions on learning performance. The fused comprehensive feature vector is standardized and aligned with time series. The comprehensive feature vector reflects students' abilities and emotional states in different task scenarios. The comprehensive feature vector obtained by fusing multi-source data is standardized to unify indicators of different dimensions to the same numerical scale, so as to ensure the comparability and fairness of data in each dimension in subsequent judgment. The time series features are aligned to ensure that historical learning data, real-time task performance and emotional state data are synchronized in the time dimension, so as to avoid data offset or misalignment from interfering with the judgment results. Through time series alignment, the comprehensive feature vector can accurately reflect students' cognitive abilities and emotional states at each task time point, so that the model can understand the changes and dynamic trends of students' states in the continuous learning process. The output comprehensive feature vector serves as input for personalized cognitive level assessment. The assessment result includes students' historical performance, real-time task execution, and emotional adaptation information. The standardized and time-aligned comprehensive feature vector is input into the personalized cognitive level assessment model. This model comprehensively considers the influence of long-term learning trends, current task performance, and emotional state to evaluate students' real-time cognitive abilities. The assessment result not only includes students' operational abilities and knowledge mastery in the current task but also reflects their adaptability to different task difficulties and emotional states, making personalized assessment more comprehensive and accurate. The output assessment result provides a direct basis for subsequent course recommendation strategies, ensuring that recommended content can dynamically adapt to students' cognitive levels, learning history, and emotional states.

[0027] The process in S4 that reflects students' true abilities and emotionally appropriate learning needs in the current task is as follows: Based on a comprehensive feature vector, a multidimensional weighted inference method is used to generate personalized cognitive level assessment results, while considering the impact of students' emotional state on cognitive performance in real time. The generated comprehensive feature vector is input into the multidimensional weighted inference model, which integrates historical learning data, current task performance, knowledge point mastery, and individual student preferences into a unified feature space. During the inference process, weights are assigned to each dimension, with core cognitive indicators and important knowledge points receiving higher weights based on students' past learning performance and task priority, while non-core indicators and low-relevance data have reduced weights to ensure the accuracy of the assessment results. Real-time emotional state is introduced to dynamically adjust the indicators of each dimension. For example, when students are in a state of high concentration or slight tension, the model automatically adjusts the contribution of behavioral indicators so that the assessment results can reflect the immediate impact of emotions on cognitive performance, ensuring the personalization and emotional adaptability of cognitive level assessment. During the assessment process, core behavioral indicators for different knowledge points and task stages are dynamically analyzed to reflect students' ability performance and emotional adaptation in the current task. The cognitive level assessment results are matched with the core behavioral indicators corresponding to each knowledge point after task decomposition. The operation accuracy, completion time, error rate, and behavior sequence of each task stage are analyzed independently to quantify students' ability level for each knowledge point in the current task. The core behavioral indicators for each stage are weighted and adjusted in conjunction with real-time collected emotional data to assess students' adaptability and learning efficiency under different emotional states, thereby revealing their true cognitive performance and emotional adaptation in the current task. The dynamic analysis process tracks indicator changes through sliding window or rolling statistical methods, which can identify short-term fluctuations, progress trends, or attention declines, so that the ability assessment can reflect the overall level and capture instantaneous changes. Abnormal or mutated behaviors in the judgment results are marked and tracked, and personalized cognitive level judgment results are output. During the judgment process, the system identifies abnormal behaviors that deviate from the normal cognitive trajectory, such as a sudden increase in operational errors, a sharp drop in completion time, or abnormal emotional fluctuations, and marks these behaviors to distinguish between typical manifestations and abnormal events. The marked abnormal or mutated behaviors are tracked in real time, and their changing trends are analyzed in conjunction with the overall cognitive level to assess the impact of these abnormal behaviors on students' task ability and emotional adaptation, and corrections or weighting are made to prevent occasional abnormalities from interfering with the judgment results. The output personalized cognitive level judgment results not only include the student's ability scores for each knowledge point in the current task, but also reflect emotional adaptation information and key behavioral dynamics.

[0028] S5: Based on the results of personalized cognitive level assessment, the interactive task parameters are adjusted through periodic real-time feedback to dynamically generate course recommendation strategies.

[0029] The process of dynamically generating course recommendation strategies in S5 is as follows: Based on the personalized cognitive level assessment results, the difficulty, content order, task allocation, and learning pace of interactive tasks are dynamically adjusted. Parameters are configured in conjunction with students' real-time cognitive load and emotional state. The generated personalized cognitive level assessment results are input into the course recommendation system to analyze students' ability levels at different knowledge points and task stages, as well as their emotional adaptation in the current task. The task difficulty is dynamically adjusted according to students' cognitive abilities and emotional states. For example, the task complexity is increased for students with strong abilities and high concentration, while the difficulty is reduced or supplementary prompts are added for students with lower abilities or decreased attention, to match students' immediate learning needs. Regarding content order and task allocation, the order of knowledge points is rearranged based on students' historical mastery and individual preferences, prioritizing practice of weaker knowledge points. The workload and learning pace are flexibly allocated to ensure that students' cognitive load is moderate in each time period, avoiding excessive pressure or ineffective learning. Through a periodic feedback mechanism, students' cognitive level and emotional fluctuations during tasks are monitored in real time. During the interactive tasks, behavioral data, completion time, error rate, and operation sequence are continuously collected, while emotional state, including attention concentration, anxiety or interest level, is monitored in real time. The periodic feedback mechanism compares the collected data with the expected cognitive level and emotional state to assess whether the current task exceeds the student's immediate capacity or whether there are cognitive bottlenecks or low mood. Based on the feedback results, a real-time report is generated, including short-term learning performance, task adaptability, and emotional fluctuation trends, providing data support for the dynamic adjustment of course recommendation strategies and ensuring that the strategies are highly matched with the student's current state. The task parameters in the recommendation strategy are adjusted in multiple dimensions, including the weight of behavioral indicators, the allocation of task duration, and the order of knowledge point connections, to output a course recommendation strategy. Based on periodic feedback, the task parameters are finely adjusted, such as adjusting the weight of different core behavioral indicators to give key cognitive behaviors a higher weight in task evaluation, while reducing the impact of low-relevance or noisy behaviors. The task duration and learning pace are dynamically planned, high-load tasks are executed in segments, and short breaks or intensive practice sessions are added to ensure that students can learn efficiently at an appropriate pace while preventing cognitive overload. Based on the mastery of knowledge points and the prediction results of cognitive level, the task content and the order of knowledge point connections are reordered, prioritizing unmastered or easily confused knowledge points to optimize the continuity and coherence of the learning path. The output course recommendation strategy includes adjusted task difficulty, content order, behavioral weight, and duration, which can adapt to students' cognitive abilities and emotional states in real time, providing precise and dynamic guidance for personalized learning.

[0030] Example 2: Figure 2 As shown, an intelligent course recommendation system that combines cognitive level recognition includes: Task perception module: Constructs multi-dimensional operation task scenarios and collects students' behavioral indicators and emotional states in real time to generate an initial cognitive state sequence; Feature mapping module: Maps the initial cognitive state sequence into a multi-dimensional feature vector, introduces emotional weights and highlights key behavioral features to construct a cognitive state map; Cognitive prediction module: Based on the cognitive state map, it uses a multi-scale time series convolutional network and reinforcement learning model to predict students' cognitive level, generate evolutionary trajectory and capture short-term fluctuations; Horizontal Fusion Module: This module integrates the evolutionary trajectory of cognitive level with historical learning trajectory, knowledge mastery, individual preferences, and emotional state to generate personalized cognitive level assessment results. The strategy generation module periodically adjusts task parameters and dynamically generates course recommendation strategies based on the results of personalized cognitive level assessment.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent course recommendation method combined with cognitive level recognition, characterized in that, The method comprises the following steps: A multi-dimensional operation task scene is constructed through an interactive game task, and a behavior index collection point for inferring learning ability and emotional state of a student is set in each task scene to generate an initial cognitive state sequence; The initial cognitive state sequence is mapped, a real-time emotional state of the student is taken as a weight adjustment factor, a self-adaptive feature screening mechanism is guided to highlight key behavior features in a specific emotional state, noise interference is removed, and emotional related cognitive indicators are strengthened to construct a cognitive state atlas; Based on the cognitive state atlas, a multi-scale time sequence convolution network is used in combination with a self-adaptive evolution mapping model of reinforcement learning to predict the cognitive level of the student, an emotional state influence coefficient is introduced in real time to weight and adjust the cognitive level evolution track, short-term cognitive fluctuations are captured, and core cognitive indicators in different time periods are extracted; The cognitive level evolution track is fused with a historical learning track of the student, a knowledge point mastering situation, individual preferences and emotional states to generate a personalized cognitive level determination result, and the real ability of the student in the current task and the learning demand of the student in the emotional adaptation are reflected; According to the personalized cognitive level determination result, interactive task parameters are adjusted through periodic real-time feedback, and a course recommendation strategy is dynamically generated. 2.The intelligent course recommendation method in combination with cognitive level recognition of claim 1, wherein, The process of generating the initial cognitive state sequence is: A plurality of behavior index collection points are set in each interactive task scene, and operation behavior, task completion time, error rate, operation sequence and emotional state information of the student are recorded in real time; The collected multi-source behavior data are time sequence coded, and task scene identification, environmental parameters and task difficulty level information are added to form a time-stamped initial cognitive state sequence. 3.The intelligent course recommendation method in combination with cognitive level identification according to claim 2, characterized in that, The process of mapping the initial cognitive state sequence is: The initial cognitive state sequence is constructed into a multi-dimensional feature vector, and the vector dimension includes a behavior index dimension, a task scene dimension and a real-time emotional weight dimension; The multi-dimensional feature vector is input into a mapping network for standardization processing and high-dimensional mapping, and iteration optimization of mapping parameters is performed to balance the consistency and distinguishability of features in different task scenes; An emotional state weight factor is introduced in the mapping process to weight and adjust the emotional sensitive cognitive behavior, highlight the key behavior features, and suppress the low correlation noise. 4.The intelligent course recommendation method in combination with cognitive level identification of claim 3, wherein, The process of guiding the self-adaptive feature screening mechanism to highlight the key behavior features in a specific emotional state is: An adaptive feature screening algorithm is used to iteratively analyze the mapped multi-dimensional feature vector, and the contribution degree and correlation degree of each behavior index in the current emotional state are calculated; High-contribution-degree behaviors are given enhanced weights, and low-contribution-degree and noise features are removed to form an emotional related reinforced feature set; The feature importance score is dynamically updated to identify and highlight core cognitive behaviors in real time in different task scenes. 5.The intelligent course recommendation method in combination with cognitive level identification of claim 4, wherein, The process of constructing the cognitive state atlas is: The key behavior indicators and emotional weights screened adaptively are mapped into graph nodes according to time sequence and task scene, and the nodes are connected with behavior dependence relationship and emotional influence strength as edge weight; Node embedding and graph normalization processing are performed on the graph structure to form a cognitive state atlas, and the behavior correlation of the student in different task scenes and emotional conditions is embodied; The key nodes and core behavior paths in the cognitive state graph are identified through node centrality and behavior dependency analysis. 6.The intelligent course recommendation method in combination with cognitive level identification according to claim 5, characterized in that, The process of predicting the student's cognitive level using a multi-scale time series convolutional network combined with a reinforcement learning adaptive evolution mapping model is as follows: Input the cognitive state graph into the multi-scale time series convolutional network to extract short-term and long-term cognitive features and capture the dynamic patterns of key behaviors changing with emotional states; Embed the reinforcement learning algorithm in the convolutional network to adaptively optimize the network weights and emotional sensitivity parameters; Encode and time-tag the predicted continuous cognitive states to identify core cognitive indicators and behavior patterns in different time periods, and output the cognitive level evolution trajectory. 7.The intelligent course recommendation method in combination with cognitive level identification of claim 6, wherein, The process of weighting and adjusting the cognitive level evolution trajectory is as follows: Combine real-time emotional states, task difficulty, and individual preferences to perform multi-dimensional weighting on the cognitive level evolution trajectory; Analyze short-term fluctuations and instantaneous ability changes in the trajectory through a sliding window mechanism, extract core cognitive indicators for each time period, and remove or correct abnormal values or atypical fluctuations in the trajectory. 8.The intelligent course recommendation method in combination with cognitive level identification of claim 7, wherein, The process of fusing the cognitive level evolution trajectory with the student's historical learning trajectory, knowledge mastery, individual preferences, and emotional state is as follows: Multi-dimensionally weight and fuse the core cognitive indicators with historical learning performance, knowledge mastery probability, and individual preference data, and embed real-time emotional state adjustment factors; Standardize and time-align the fused comprehensive feature vector, which reflects the student's ability and emotional state in different task scenarios; Output the comprehensive feature vector as the input for personalized cognitive level determination, and the determination result includes the student's historical performance, real-time task execution, and emotional adaptation information. 9.The intelligent course recommendation method in combination with cognitive level identification of claim 8, wherein, The process of dynamically generating course recommendation strategies is as follows: Based on the comprehensive feature vector, use multi-dimensional weighted reasoning to generate personalized cognitive level determination results, while considering the impact of the student's emotional state on cognitive performance in real time; During the determination process, dynamically analyze the core behavior indicators for different knowledge points and task stages to reflect the student's ability performance and emotional adaptation in the current task; Mark and track abnormal or sudden behaviors in the determination result, and output the personalized cognitive level determination result; Based on the personalized cognitive level determination result, dynamically adjust the difficulty, content sequence, task allocation, and learning pace of interactive tasks, and configure parameters based on the student's real-time cognitive load and emotional state; Through a periodic feedback mechanism, monitor the student's cognitive level and emotional fluctuations in real time during the task; Adjust the task parameters in the recommendation strategy, including behavior indicator weights, task duration allocation, and knowledge point connection sequence, and output the course recommendation strategy.

10. An intelligent course recommendation system in combination with cognitive level recognition, characterized in that, The system includes: Task perception module: Construct multi-dimensional operation task scenarios and collect student behavior indicators and emotional states in real time to generate initial cognitive state sequences; Feature mapping module: Map the initial cognitive state sequences to multi-dimensional feature vectors, introduce emotional weights, highlight key behavior features, and construct a cognitive state graph. Cognitive prediction module: Based on the cognitive state atlas, the multi-scale time series convolution network and the reinforcement learning model are used to predict the student's cognitive level, generate the evolution trajectory and capture the short-term fluctuations; Horizontal fusion module: The cognitive level evolution trajectory is fused with the historical learning trajectory, knowledge mastery, individual preference and emotional state to generate the personalized cognitive level judgment result; Generation strategy module: According to the personalized cognitive level judgment result, the task parameters are periodically adjusted and the course recommendation strategy is dynamically generated.

Images