Intelligent coaching adjustment system and methods

Abstract

This invention belongs to the field of artificial intelligence technology and discloses an intelligent coaching adjustment system and method. It solves the problem that existing technologies rely solely on preset scripts and simulated training data, resulting in a single data collection dimension and difficulty in comprehensively assessing the user's cognitive load. The method includes: collecting multi-dimensional data on users, behavior, and physiology; comprehensively extracting structured features and cognitive load sequences; processing to obtain cross-dimensional features, difficulty constraints, and ability vectors; modifying the ability vectors using an improved model; using the TOPS IS algorithm to analyze and generate a reinforcement task queue and then modifying it; dynamically adjusting the learning task weights; and finally obtaining a composite learning task, achieving a comprehensive assessment and task optimization of the user's cognitive load. This invention generates a reinforcement task queue using the TOPS IS algorithm and dynamically adjusts task weights based on cross-dimensional features and difficulty constraints to form a composite learning task, solving the problem of insufficient single-ability assessment.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to an intelligent coaching adjustment system and method. Background Technology

[0002] As artificial intelligence technology penetrates the education field, the limitations of traditional standardized tutoring models are becoming increasingly apparent.

[0003] Chinese patent application CN113627801A discloses an intelligent tutoring method, device, electronic device, and storage medium. The method includes: a tutor building simulated practice scripts for different service scenarios on their end; a training center training the simulated practice scripts with corresponding intentions and quality control rules to obtain an intention model and a quality control rule model, which are then sent to the student's end; the student simulating and training on the simulated practice scripts for different service scenarios on their end to obtain simulated training information; and the student's end scoring the simulated training information in real time based on the intention model and the quality control rule model to obtain a first score. The quality control rules include at least: emotion recognition rules, interruption monitoring rules, speech rate monitoring rules, and service-related prohibited words. This invention saves on training manpower costs.

[0004] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0005] Relying solely on pre-defined scripts and simulated training data results in a single data collection dimension, making it difficult to comprehensively assess the user's cognitive load. Scoring simulated training results based solely on intent models and quality control rules lacks dynamic task adjustment logic based on real-time data. Relying solely on predefined simulated scripts and fixed scoring rules leads to insufficient refinement of personalized measurement.

[0006] In view of this, the present invention proposes an intelligent coaching adjustment system and method to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution:

[0008] The intelligent coaching adjustment method includes the following steps:

[0009] Collect user data, behavioral data, and physiological data;

[0010] Comprehensive feature extraction is performed on user data, behavioral data, and physiological data to obtain structured features; comprehensive feature extraction is performed on behavioral data and physiological data to obtain cognitive load sequences;

[0011] Structured features are processed to obtain cross-dimensional features; cognitive load sequences are analyzed to obtain difficulty constraints.

[0012] Based on a predefined knowledge graph, the structured features are processed to obtain a capability vector; the capability vector is then corrected based on an improved Ebbinghaus model to obtain a corrected capability vector.

[0013] Based on the TOPSIS multi-criteria decision-making algorithm, the structured features are analyzed according to the modified capability vector to obtain the reinforcement task queue; the reinforcement task queue is processed according to the cross-dimensional features to obtain the modified task queue.

[0014] The learning task weights are dynamically adjusted based on the cognitive load index sequence to obtain the task weight matrix;

[0015] By analyzing the difficulty constraints, the modified task queue, and the task weight matrix, a composite learning task is obtained.

[0016] Furthermore, the user data includes spoken recordings and answer records; the answer records include learning task type tags, question IDs, user answers, correctness tags, and response times. The learning task type tags in the answer records include grammar, vocabulary, and listening comprehension. A correctness tag of 1 indicates a correct answer, and a correctness tag of 0 indicates an incorrect answer. The behavioral data includes learning period timestamps, and the physiological data is RR interval data.

[0017] Furthermore, the method for obtaining the structured features includes:

[0018] The spoken recording is converted into spoken text, compared with the preset answer text, and the total number of words and the number of incorrect words in the spoken text are counted. The ratio of the number of incorrect words to the total number of words is calculated to obtain the spoken error density.

[0019] The speech rate fluctuation rate is obtained by statistically analyzing the average and standard deviation of the number of words per second in spoken text, and calculating the ratio of the standard deviation of the number of words per second to the average number of words per second. The coherence index is obtained by statistically analyzing the number of occurrences of preset filler words and adjacent repeated phrases per unit time, and calculating the ratio of the occurrences to the total number of words in the spoken text. The phoneme error rate is obtained by using the spoken text as input to a pre-trained speech model.

[0020] Obtain the number of questions in the most recent R learning task with the category label "vocabulary" and the accuracy label "1". Calculate the ratio of the number of questions with the accuracy label "1" to R to obtain the vocabulary accuracy rate. Count the total number of times users reviewed vocabulary in the most recent H weeks. Calculate the ratio of the total number of reviews to H to obtain the average number of reviews per week.

[0021] Obtain the number of questions in the most recent E-learning task with the type label "grammar" and the correctness label "1". Calculate the ratio of the number of questions with the correctness label "1" to E-learning to obtain the grammar accuracy.

[0022] Get all question IDs with a correctness label of 0. Based on the predefined knowledge graph, get the knowledge point labels and corresponding basic coefficients of all question IDs with a correctness label of 0. Count the number of errors corresponding to the same knowledge point label. Calculate the product of the number of errors and the basic coefficient of the knowledge point to obtain the knowledge point weight. Count all knowledge point weights to obtain the defect matrix.

[0023] Obtain the response time for the first preset number of responses, calculate the standard deviation and mean of the response time for the first preset number of responses, and calculate the ratio of the standard deviation to the mean of the response time for the first preset number of responses to obtain the response time dispersion; obtain the physiological data for the second preset number of responses, calculate the standard deviation and mean of the physiological data for the second preset number of responses, and calculate the ratio of the standard deviation to the mean of the physiological data for the second preset number of responses to obtain the physiological characteristics; weight the response time dispersion and the physiological characteristics to obtain the cognitive load index;

[0024] Structured features are obtained by concatenating spoken error density, speech rate fluctuation rate, coherence index, phoneme error rate, vocabulary accuracy, average weekly review frequency, grammatical accuracy, defect matrix, cognitive load index, and corresponding timestamps.

[0025] Furthermore, the method for obtaining the cognitive load sequence includes:

[0026] The cognitive load index, data source labels, and corresponding timestamps from the structured features within a preset time period are obtained and then concatenated to obtain the cognitive load sequence.

[0027] The methods for obtaining cross-dimensional features include:

[0028] The structured features are normalized to obtain standardized features, and the standardized features are used as input to the feature extraction model to obtain cross-dimensional features.

[0029] Methods for obtaining difficulty constraints include:

[0030] Obtain the real-time cognitive load index. When the real-time cognitive load index is greater than the second preset load upper limit, adjust the maximum proportion of new knowledge to the preset first threshold. The first difficulty constraint range is [dX, d), where d is the user's current ability and X is the first adjustment value. When the real-time cognitive load index is less than the second preset load lower limit, adjust the maximum proportion of new knowledge to the preset second threshold. The second difficulty constraint range is [d, d+Y), where Y is the second adjustment value.

[0031] Furthermore, the method for obtaining the capability vector includes:

[0032] Spoken language proficiency is obtained by weighting spoken error density, speech rate fluctuation rate, coherence index, and phoneme error rate.

[0033] Vocabulary mastery rate is obtained by weighting vocabulary accuracy and average number of weekly reviews;

[0034] Based on the knowledge graph, obtain the knowledge point weights corresponding to questions with the learning task type label of grammar in the structured features, as well as the basic knowledge point coefficients of the corresponding knowledge points in the knowledge graph. The grammar mastery is obtained by weighting the knowledge point weights and basic knowledge point coefficients of each knowledge point.

[0035] The system retrieves the user's historical listening practice records, calculates the accuracy rate at different speaking speeds, obtains the highest speaking speed with an accuracy rate not lower than the preset highest threshold as the upper limit of recognition, and obtains the lowest speaking speed with an accuracy rate not lower than the preset lowest threshold as the lower limit of recognition; the difference between the upper and lower limits of recognition is calculated as the range of speaking speeds to be recognized.

[0036] The system compares the listening text with the list of words the user has already learned, counts the number of unlearned words, and calculates the ratio of the number of unlearned words to the total number of words in the listening text as the word frequency.

[0037] The listening comprehension mastery rate is obtained by weighting the range of speech speed and the frequency of new words.

[0038] The ability vector is obtained by concatenating oral proficiency, vocabulary mastery, grammar mastery, and listening comprehension mastery.

[0039] Furthermore, the method for obtaining the modified capability vector includes:

[0040] The memory retention rate is calculated based on the cumulative knowledge base coefficient and the accuracy rate of the most recent C questions. The modified ability vector is obtained by multiplying the memory retention rate and the ability vector. The cumulative knowledge base coefficient is obtained by weighting the knowledge point base coefficients of the knowledge points that the user has mastered in the knowledge graph.

[0041] Furthermore, the method for obtaining the enhanced task queue includes:

[0042] Extract the corrected oral proficiency, corrected vocabulary proficiency, corrected grammar proficiency, and corrected listening comprehension threshold from the corrected ability vector, and calculate the grammar ability deviation and listening ability deviation corresponding to the corrected oral proficiency, corrected vocabulary proficiency, corrected grammar proficiency, and corrected listening comprehension threshold;

[0043] Calculate cognitive load adaptability based on the cognitive load index;

[0044] Standardized deviation is obtained by standardizing the deviations of grammatical and listening abilities and taking their reciprocals. Standardized knowledge coefficients are obtained by standardizing the basic coefficients of knowledge points. Standardized cognitive load is obtained by standardizing cognitive load adaptability. Standardized deviation, standard knowledge coefficients, and standard cognitive load are weighted to obtain a standard matrix. The maximum standardized deviation, the maximum standard knowledge coefficient, and the maximum standard cognitive load are obtained and weighted to obtain the ideal solution. The minimum standardized deviation, the minimum standard knowledge coefficient, and the minimum standard cognitive load are obtained and weighted to obtain the negative ideal solution. The distances from the standard matrix to the ideal solution and the negative ideal solution are calculated separately. The proximity score is calculated based on the distances from the standard matrix to the ideal solution and the distances from the standard matrix to the negative ideal solution for each knowledge point. The knowledge points are sorted in descending order of proximity score to obtain the reinforcement task queue.

[0045] Furthermore, the method for obtaining the correction task queue includes:

[0046] The knowledge points in the reinforcement task queue are encoded in ascending order. The contribution weight of each knowledge point to the cross-dimensional features is calculated based on the attention mechanism. The contribution weight is multiplied by the corresponding code of the knowledge point in the reinforcement task queue to obtain the corrected code. The knowledge points in the reinforcement task queue are reordered in ascending order according to the corrected code. The result of the reordering is used as the corrected task queue.

[0047] Furthermore, the method for obtaining the task weight matrix includes:

[0048] The accuracy rate of learning tasks in different dimensions is obtained by statistically analyzing the ratio of the number of questions with a correctness label of 1 to the number of questions in the corresponding dimension. The actual task load coefficient is then calculated by combining the calculated load coefficient with the preset task load coefficient. The different dimensions include speaking, vocabulary, grammar, and listening.

[0049] The updated load coefficient is calculated based on the preset task load coefficient, the cognitive load index at the current moment, and the smoothing factor.

[0050] The task weights are calculated by combining the sensitivity coefficient, the cognitive load index at the current moment, the preset task load coefficient, and the update load coefficient. The task weight matrix is ​​obtained by concatenating the task weights of different dimensions.

[0051] Furthermore, the method for obtaining the complex learning task includes:

[0052] Obtain the first N knowledge points in the correction task queue, generate a task sequence, and combine the task weights to design the task combination ratio in the task sequence to obtain composite tasks.

[0053] Query related knowledge points in the knowledge graph, and design vertical progression tasks and horizontal linkage tasks respectively. Vertical progression tasks represent learning N knowledge points sequentially, and horizontal linkage tasks represent learning N knowledge points simultaneously. Obtain the user's current cognitive load index, design the time allocation ratio of learning tasks in different dimensions according to the knapsack algorithm, classify the user's cognitive load level according to the preset cognitive load level, and set the number of questions in each dimension for different cognitive load levels according to the preset ratio.

[0054] Furthermore, the methods for obtaining the time allocation ratios for the learning tasks of different dimensions include:

[0055] Obtain the weight value of each learning task in each dimension, the difficulty constraint range of each dimension, and the duration corresponding to the user's learning period. Treat each learning task in each dimension as an item, the value of the learning task as the task weight, and the weight of the learning task as the estimated time to complete the learning task. When selecting tasks, ensure that the difficulty parameters of the selected learning tasks meet the preset difficulty constraint range. Under the premise that the total time does not exceed the estimated total time, select learning task combinations to maximize the total weight. Calculate the time allocation percentage for each task type based on the total weight of the task combination, and statistically analyze the time allocation percentage for learning tasks in different dimensions. Use the time allocation percentage as the time distribution ratio for learning tasks in different dimensions.

[0056] Methods for obtaining the difficulty parameters of a learning task include:

[0057] Obtain the user's most recent Z questions' answer data, count the number of correct answers, calculate the ratio of the number of correct answers to Z, and obtain the user's answer accuracy rate;

[0058] Based on maximum likelihood estimation, a model is built to represent user answer accuracy, user ability, and question difficulty. User ability is calculated based on answer accuracy by fixing the question difficulty. Prediction accuracy is calculated based on user ability and the preset question difficulty. The actual accuracy of the user's most recent M questions is statistically analyzed. A difficulty adjustment parameter is calculated based on the predicted and actual accuracy. If the predicted accuracy is greater than the actual accuracy, the difficulty parameter is updated by adding the difficulty adjustment parameter; if the predicted accuracy is less than the actual accuracy, the difficulty parameter is updated by subtracting the difficulty adjustment parameter. Furthermore, user ability is updated based on updated answer records.

[0059] The intelligent coaching adjustment system implements the aforementioned intelligent coaching adjustment method, including:

[0060] Data acquisition module: collects user data, behavioral data, and physiological data;

[0061] The first analysis module performs comprehensive feature extraction on user data, behavioral data, and physiological data to obtain structured features; it also performs comprehensive feature extraction on behavioral data and physiological data to obtain a cognitive load sequence.

[0062] The second analysis module processes structured features to obtain cross-dimensional features and analyzes cognitive load sequences to obtain difficulty constraints.

[0063] The third analysis module processes the structured features based on a predefined knowledge graph to obtain capability vectors; it then modifies the capability vectors based on an improved Ebbinghaus model to obtain modified capability vectors.

[0064] The fourth analysis module: Based on the TOPSIS multi-criteria decision-making algorithm, it analyzes the structured features according to the modified capability vector to obtain the reinforcement task queue; and processes the reinforcement task queue according to the cross-dimensional features to obtain the modified task queue.

[0065] The fifth analysis module dynamically adjusts the weights of the learning tasks based on the cognitive load index sequence to obtain the task weight matrix;

[0066] Task Adjustment Module: Based on difficulty constraints, the task queue is adjusted, and the task weight matrix is ​​analyzed to obtain composite learning tasks.

[0067] The technical effects and advantages of the intelligent coaching adjustment system and method of this invention are as follows:

[0068] This invention overcomes the limitations of fixed-script training by integrating physiological data to monitor cognitive load in real time and correcting ability vectors using the Ebbinghaus model to accurately match learning progress and memory patterns. It leverages knowledge graphs to pinpoint ability weaknesses and dynamically adjusts question difficulty and task pace based on load constraints, avoiding the poor adaptability issues caused by standardized training. By generating a reinforcement task queue using the TOPSIS algorithm and dynamically adjusting task weights based on cross-dimensional features and difficulty constraints, it forms composite learning tasks, solving the problem of insufficient single-ability assessment. Simultaneously, it iteratively updates model parameters using real-time load sequences and learning effect data, ensuring the effectiveness of the long-term learning process and compensating for the lack of a proper model update mechanism. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the intelligent coaching adjustment method according to Embodiment 1 of the present invention;

[0070] Figure 2 This is a schematic diagram of the intelligent coaching adjustment method according to Embodiment 2 of the present invention;

[0071] Figure 3 This is a block diagram of the intelligent coaching adjustment system of Embodiment 3 of the present invention. Detailed Implementation

[0072] 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.

[0073] Example 1

[0074] Please see Figure 1 As shown, this embodiment provides an intelligent coaching adjustment method, including the following steps:

[0075] The system collects user data, behavioral data, and physiological data. User data includes spoken recordings and answer logs. Spoken recordings are audio recordings of spoken practice taken through a microphone, along with pre-set answer texts. By analyzing these audio recordings and texts, pronunciation and content completeness can be accurately assessed, optimizing the feedback mechanism and practice content for spoken practice, and specifically improving expression skills. Answer logs include learning task type tags (including grammar, vocabulary, and listening), question IDs, user answers, correctness tags (1 for correct, 0 for incorrect), and response time. Answer log collection can be based on task type, question type, and answer format. The system tracks correct answers and response times to pinpoint weaknesses in vocabulary, grammar, and listening comprehension, intelligently recommending relevant reinforcement exercises for precise learning. Behavioral data includes learning time stamps, with start and end timestamps, enabling analysis of learning pace and focus based on the learning period, dynamically adjusting exercise duration and task intervals to optimize the continuity of the learning process. Physiological data, collected through wearable devices such as smartwatches, provides real-time monitoring of cognitive load, automatically adjusting the difficulty of practice questions to avoid excessive or insufficient load and maintain an efficient learning state.

[0076] By comprehensively extracting features from user data, behavioral data, and physiological data, structured features are obtained. Furthermore, the cognitive load sequence extracted from behavioral and physiological data comprehensively depicts a user's knowledge mastery and learning behavior patterns. Simultaneously, the cognitive load sequence extracted from behavioral and physiological data accurately reflects the user's real-time workload. Based on this data, intelligent tutoring dynamically adjusts the difficulty and pace of practice content, achieving personalized adaptation of practice plans and improving learning efficiency and experience.

[0077] Methods for obtaining structured features include:

[0078] The spoken recording is converted into spoken text, compared with the preset answer text, and the total number of words and the number of incorrect words in the spoken text are counted. The ratio of the number of incorrect words to the total number of words is calculated to obtain the spoken error density.

[0079] The average and standard deviation of the number of words per second in spoken text are statistically analyzed, and the ratio of the standard deviation of the number of words per second to the average number of words per second is calculated to obtain the speech rate fluctuation rate.

[0080] The frequency of preset filler words and adjacent repeated phrases within a unit of time is counted, and the ratio of the frequency of occurrence to the total number of words in the spoken text is calculated to obtain a coherence index.

[0081] The phoneme error rate is obtained by using spoken text as input to a pre-trained speech model (such as Wav2Vec 2.0 or a localized model);

[0082] Get the number of questions in the most recent R learning task with the type label "vocabulary" and the correctness label "1". Calculate the ratio of the number of questions with the correctness label "1" to R to obtain the vocabulary accuracy.

[0083] Calculate the total number of times users reviewed vocabulary in the past H weeks, and then calculate the ratio of the total number of reviews to H to obtain the average number of reviews per week.

[0084] Obtain the number of questions in the most recent E-learning task with the type label "grammar" and the correctness label "1". Calculate the ratio of the number of questions with the correctness label "1" to E-learning to obtain the grammar accuracy.

[0085] Get all question IDs with a correctness label of 0. Based on the predefined knowledge graph, get the knowledge point labels and corresponding basic coefficients of all question IDs with a correctness label of 0. Count the number of errors corresponding to the same knowledge point label. Calculate the product of the number of errors and the basic coefficient of the knowledge point to obtain the knowledge point weight. Count all knowledge point weights to obtain the defect matrix.

[0086] Obtain the response time of the preset first number of times, calculate the standard deviation and mean of the response time of the preset first number of times, calculate the ratio of the standard deviation to the mean of the response time of the preset first number of times, and obtain the answer time dispersion.

[0087] Obtain a preset second set of physiological data, calculate the standard deviation and mean of the preset second set of physiological data, calculate the ratio of the standard deviation to the mean of the preset second set of physiological data, and obtain physiological characteristics;

[0088] The cognitive load index is obtained by weighting the time dispersion of answering questions and physiological characteristics; the weighting coefficients can be obtained by optimizing a natural heuristic optimization algorithm.

[0089] Structured features are obtained by concatenating spoken error density, speech rate fluctuation rate, coherence index, phoneme error rate, vocabulary accuracy, average weekly review frequency, grammatical accuracy, defect matrix, cognitive load index, and corresponding timestamps.

[0090] Methods for obtaining cognitive load sequences include:

[0091] The cognitive load index, data source labels, and corresponding timestamps from the structured features within a preset time period are obtained and then concatenated to obtain the cognitive load sequence.

[0092] By processing structured features, cross-dimensional features are obtained; by analyzing cognitive load sequences, difficulty constraints are obtained. The cross-dimensional features obtained from processing structured features can integrate multi-dimensional information such as user knowledge mastery and behavioral patterns to deeply characterize user learning characteristics and needs. The difficulty constraints derived from analyzing cognitive load sequences define the range of question difficulty that users can currently accept. Intelligent tutoring relies on cross-dimensional features to accurately pinpoint user skill gaps and dynamically matches question difficulty with difficulty constraints. This allows for targeted reinforcement training for knowledge weaknesses while avoiding excessive difficulty that leads to inefficient learning. It achieves scientific adjustment of practice content, difficulty, and pace, ensuring that the learning process is both targeted and aligned with cognitive capacity, optimizing the learning experience and effectiveness.

[0093] Methods for obtaining cross-dimensional features include:

[0094] The structured features are normalized to obtain standardized features, and the standardized features are used as input to the feature extraction model to obtain cross-dimensional features.

[0095] Training methods for feature extraction models include:

[0096] P sets of training data were collected in advance, including standardized features and corresponding cross-dimensional features;

[0097] Standardized features are used as input to the feature extraction model, and cross-dimensional features are used as output. With the goal of minimizing the error between the output cross-dimensional features and the actual cross-dimensional features, the network parameters of the feature extraction model are optimized using a natural heuristic optimization algorithm. The network parameters that minimize the error between the cross-dimensional features output by the feature extraction model and the actual cross-dimensional features are obtained. The feature extraction model constructed with the corresponding network parameters is used as the trained feature extraction model.

[0098] Methods for obtaining difficulty constraints include:

[0099] The system acquires a real-time cognitive load index. When the real-time cognitive load index exceeds a second preset load limit, the maximum proportion of new knowledge is adjusted to a preset first threshold, with a first difficulty constraint range of [dX, d), where d represents the user's current ability and X is a first adjustment value, such as 0.1. When the real-time cognitive load index exceeds the second preset load limit, it indicates that the user is currently experiencing high cognitive pressure. To prevent the user from experiencing learning fatigue or resistance due to excessive load, the difficulty of the learning content needs to be reduced. Adjusting the maximum proportion of new knowledge to the preset first threshold controls the proportion of new knowledge in the learning content, thereby alleviating cognitive load. Setting the first difficulty constraint range to [dX, d) means that the difficulty of the learning content will be limited to the user's current ability d, with a maximum reduction of X (e.g., 0.1). This is because under high cognitive load, appropriately reducing the difficulty helps users better understand and absorb knowledge, making the learning process smoother. With this setting, a negative correlation is established between the cognitive load index and the difficulty constraint range. That is, as the cognitive load index exceeds the upper limit, the difficulty constraint range is adjusted to a lower difficulty level to adapt to the user's current cognitive state and ensure learning effectiveness. When the real-time cognitive load index is less than the second preset lower limit, the maximum proportion of new knowledge is adjusted to the preset second threshold. The second difficulty constraint range is [d, d+Y), where Y is the second adjustment value, such as 0.2.

[0100] Based on a predefined knowledge graph, structured features are processed to obtain a capability vector. This capability vector is then modified using an improved Ebbinghaus forgetting curve model to obtain a revised capability vector. The capability vector generated by processing structured features based on the knowledge graph quantifies a user's knowledge mastery across four dimensions: speaking, vocabulary, grammar, and listening. Furthermore, the capability vector is corrected using memory retention rate, incorporating the forgetting curve to accurately assess knowledge retention and weaknesses. During intelligent tutoring, the system can tailor practice directions based on the user's actual ability level, such as prioritizing knowledge points with rapid memory decay, ensuring that tutoring content is both relevant to current abilities and meets memory optimization needs.

[0101] Methods for obtaining capability vectors include:

[0102] Spoken language proficiency is obtained by weighting spoken error density, speech rate fluctuation rate, coherence index, and phoneme error rate.

[0103] Vocabulary mastery rate is obtained by weighting vocabulary accuracy and average number of weekly reviews;

[0104] Based on the knowledge graph, obtain the knowledge point weights corresponding to questions with the learning task type label of grammar in the structured features, as well as the basic knowledge point coefficients of the corresponding knowledge in the knowledge graph. The grammar mastery is obtained by weighting the knowledge point weights and basic knowledge point coefficients of each knowledge point.

[0105] The system retrieves the user's historical listening practice records, calculates the accuracy rate at different speaking speeds, obtains the highest speaking speed with an accuracy rate not lower than the preset highest threshold as the upper limit of recognition, and obtains the lowest speaking speed with an accuracy rate not lower than the preset lowest threshold as the lower limit of recognition; the difference between the upper and lower limits of recognition is calculated as the range of speaking speeds to be recognized.

[0106] The system compares the listening text with the list of words the user has already learned, counts the number of unlearned words, and calculates the ratio of the number of unlearned words to the total number of words in the listening text as the word frequency.

[0107] The listening comprehension mastery rate is obtained by weighting the range of speech speed and the frequency of new words.

[0108] The ability vector is obtained by concatenating oral proficiency, vocabulary mastery, grammar mastery, and listening comprehension mastery.

[0109] Methods for obtaining the modified capability vector include:

[0110] The memory retention rate is calculated based on the cumulative knowledge base coefficient and the accuracy rate of the most recent C questions. The revised ability vector is then obtained by multiplying the memory retention rate by the ability vector. (e.g., memory retention rate...) Wherein, τ is the cumulative knowledge base coefficient, which is obtained by weighting the knowledge point base coefficients of the knowledge points that the user has mastered in the knowledge graph. The knowledge point base coefficient is determined based on the inherent attributes of the knowledge point, including difficulty (scored by education experts based on the course syllabus, 1-5 points) and importance (determined by the proportion of the knowledge point in the exam syllabus, 1-5 points). The formula for calculating the cumulative knowledge base coefficient τ is τ = (0.6 × difficulty score + 0.4 × importance score - 1) / (5 - 1) (normalized to the range of 0-1). The cumulative knowledge base coefficient can also be adjusted based on the user's historical learning performance. For example, if the user's accuracy rate in learning this knowledge point in the last 3 times is ≥80%, then the cumulative knowledge base coefficient τ is adjusted. ' =τ×0.9, if the accuracy rate is <50%, then adjust the cumulative knowledge base coefficient τ. '=τ×1.1; γ is the accuracy rate of the most recent C questions; Δt is the time interval between the collection of the most recent C questions, exp(·) represents the exponential function; parameter 1.5 is an empirical correction factor, derived from the fitting of experimental data on memory retention rate of 1000 learners. By comparing the deviation between the actual memory retention rate and the theoretical value (without correction) (which can be calculated using the improved Ebbinghaus model), it was found that when the correction factor is 1.5, the mean square error between the model prediction value and the actual value is the smallest (error ≤3%), so this parameter was selected; Corrected ability vector GS' = GS×MR; where GS is the ability vector; The memory retention rate formula combines the cumulative knowledge base coefficient τ, the recent accuracy rate γ, and the time interval Δt, and simulates the Ebbinghaus forgetting curve through an exponential function to quantify the degree of knowledge forgetting over time. For example, if the user's cumulative knowledge base coefficient is high and the recent answer accuracy rate is high, the formula will calculate a higher memory retention rate, and vice versa, reflecting the risk of knowledge forgetting; By correcting the ability vector through memory retention rate, the ability assessment can not only reflect the current knowledge level, but also incorporate memory factors. For example, if a user's original ability vector shows that they have a good grasp of a certain dimension, but their memory retention rate is low (e.g., they haven't reviewed it for a long time), the revised ability vector will reduce the weight of that dimension, more accurately reflecting the "retrieval status" of knowledge, and providing a more accurate basis for planning subsequent learning tasks.

[0111] Based on the TOPSIS multi-criteria decision-making algorithm, structured features are analyzed according to the modified ability vector to obtain a reinforcement task queue. This reinforcement task queue is then processed based on cross-dimensional features to obtain a modified task queue. The TOPSIS multi-criteria decision-making algorithm is used to analyze the modified ability vector, selecting high-priority reinforcement task queues to clarify the focus of knowledge consolidation. Further processing using cross-dimensional features eliminates mismatched tasks and adds suitable tasks, forming the modified task queue. This makes the task recommendation for intelligent tutoring more scientific, focusing on core weaknesses while also ensuring a balance across learning dimensions.

[0112] Methods to obtain an enhanced task queue include:

[0113] Extract the corrected oral proficiency, corrected vocabulary proficiency, corrected grammar proficiency, and corrected listening comprehension threshold from the corrected ability vector, and calculate the grammar ability deviation and listening ability deviation corresponding to the corrected oral proficiency, corrected vocabulary proficiency, corrected grammar proficiency, and corrected listening comprehension threshold;

[0114] Cognitive load adaptability is calculated based on the cognitive load index, such as Here, CLIM represents cognitive load adaptability; parameter 5 is a normalization coefficient used to adjust the sensitivity of the cognitive load index CLI to the influence of cognitive load adaptability CLIM; when parameter = 5, the actual distribution of cognitive load adaptability is 0.076-0.924 points (covering 90% of user scenarios). This normalization parameter ensures that the impact of CLI changes on CLIM is neither too gradual (if the normalization parameter is too small, CLIM will lack discrimination) nor too extreme (if the normalization parameter is too large, CLIM will tend to concentrate at both ends), thus preserving the discrimination of the score and meeting the actual needs of "adaptability" assessment; parameter 0.5 is used to calibrate the center position of the function, so that when CLI = 0.5, CLIM is in an intermediate state, balancing the impact of high and low CLI values ​​on CLIM, ensuring that the assessment logic conforms to the setting of "cognitive load and adaptability are negatively correlated" (i.e., the higher the CLI, the lower the CLIM); CLI is the cognitive load index, CLI = ω CLD ×CLD+ω CLP ×CLP, where ω CLD The CLD weights can be obtained through a natural heuristic optimization algorithm; CLD is the time dispersion of responses (normalized to the range of 0-1); ω CLP The CLP weights can be obtained through natural heuristic optimization algorithms. CLP is a comprehensive value of physiological characteristics (standardized to the range of 0-1). The reason for choosing the RR interval (heart rate variability index) as the physiological characteristic is that, as verified by medical research, the correlation between the RR interval and cognitive load (e.g., Pearson correlation coefficient 0.82) is significantly higher than that of other physiological indicators (e.g., skin conductance response correlation coefficient 0.75). For a very small number of users whose RR intervals do not change significantly, the system automatically switches to skin conductance response data and adjusts the calculation method of CLP to CLP = 0.5 × skin conductance response + 0.5 × pupil diameter change rate (the correlation between the two indicators is ≥0.7, or other physiological indicators with a correlation higher than 0.7 can be selected based on the actual calculation results). By converting the cognitive load index into cognitive load adaptability, a quantitative assessment of the user's current load status can be achieved. By mapping the abstract cognitive load index to an adaptability score in the [0,1] interval, the degree of matching between task difficulty and user load can be intuitively reflected. Moreover, the higher the cognitive load index, the lower the cognitive load adaptability score, and the system can dynamically adjust the task difficulty accordingly. For example, when CLI scores are higher than the preset CLI threshold, CLIM scores are low. In this case, high-difficulty tasks are reduced to avoid learning fatigue caused by user overload. Conversely, if the load is low, challenging tasks are increased to make full use of cognitive resources and balance learning efficiency and experience.

[0115] Standardized deviation is obtained by standardizing the deviations of grammatical and listening abilities and taking their reciprocals. Standardized knowledge coefficients are obtained by standardizing the basic coefficients of knowledge points. Standardized cognitive load is obtained by standardizing cognitive load adaptability. Standardized deviation, standard knowledge coefficients, and standard cognitive load are weighted to obtain a standard matrix. The maximum standardized deviation, the maximum standard knowledge coefficient, and the maximum standard cognitive load are obtained and weighted to obtain the ideal solution. The minimum standardized deviation, the minimum standard knowledge coefficient, and the minimum standard cognitive load are obtained and weighted to obtain the negative ideal solution. The distances from the standard matrix to the ideal solution and the negative ideal solution are calculated separately. The proximity score is calculated based on the distances from the standard matrix to the ideal solution and the distances from the standard matrix to the negative ideal solution for each knowledge point. The knowledge points are sorted in descending order of proximity score to obtain the reinforcement task queue.

[0116] Methods for obtaining the correction task queue include:

[0117] The knowledge points in the reinforcement task queue are encoded in ascending order. The contribution weight of each knowledge point to the cross-dimensional features is calculated based on the attention mechanism. The contribution weight is multiplied by the corresponding code of the knowledge point in the reinforcement task queue to obtain the corrected code. The knowledge points in the reinforcement task queue are reordered in ascending order according to the corrected code. The result of the reordering is used as the corrected task queue.

[0118] The learning task weights are dynamically adjusted based on the cognitive load index sequence to obtain a task weight matrix. This dynamic adjustment matches the user's current cognitive load status in real time. When the load is high, the weight of complex tasks is reduced; when the load is low, the weight of challenging tasks is increased. This ensures that the task allocation in intelligent tutoring better matches the user's physiological and psychological endurance, preventing learning fatigue or inefficiency caused by unreasonable task weights and maintaining comfort and focus during the learning process.

[0119] Methods for obtaining the task weight matrix include:

[0120] The accuracy rate of learning tasks in different dimensions is obtained by statistically analyzing the ratio of the number of questions with a correctness label of 1 to the number of questions in the corresponding dimension. The actual task load coefficient is then calculated by combining the calculated load coefficient with the preset task load coefficient. The different dimensions include speaking, vocabulary, grammar, and listening.

[0121] The updated load coefficient is calculated based on the preset task load coefficient, the current cognitive load index, and the smoothing factor. The smoothing factor is used to adjust the current cognitive load index and the preset task load coefficient, making the calculation results more stable and reasonable, avoiding drastic changes in the updated load coefficient due to instantaneous fluctuations in the cognitive load index, making the entire calculation process smoother, and reducing the adverse effects caused by data mutations.

[0122] Task weights are calculated by combining the sensitivity coefficient, the current cognitive load index, the preset task load coefficient, and the updated load coefficient. These task weights from different dimensions are then concatenated to obtain a task weight matrix. The sensitivity coefficient measures how sensitive the task weights are to changes in the current cognitive load index and the preset task load coefficient. It determines the extent to which changes in the cognitive load index and the preset task load coefficient affect the final task weight result during the calculation of task weights. A higher sensitivity coefficient indicates that the task weights are more sensitive to changes in these factors, and vice versa.

[0123] By analyzing the modified task queue, task weight matrix, and difficulty constraints, a composite learning task is obtained. The difficulty constraints (defining the manageable range of question difficulty), the modified task queue (clarifying the core practice content), and the task weight matrix (matching task priorities to the workload) are integrated to ultimately generate the composite learning task. This enables the intelligent tutor to output personalized task combinations that are "reasonably difficult, accurately content-driven, and scientifically weighted," ensuring both knowledge reinforcement and adaptation to cognitive load. This achieves a deep fit between learning content, pace, and user state, maximizing learning efficiency.

[0124] Methods for obtaining complex learning tasks include:

[0125] Obtain the first N knowledge points in the correction task queue, generate a task sequence, and design the task combination ratio in the task sequence based on the task weights to obtain composite tasks.

[0126] Query related knowledge points in the knowledge graph, and design vertical progression tasks and horizontal linkage tasks respectively. Vertical progression tasks represent learning N knowledge points sequentially, while horizontal linkage tasks represent learning N knowledge points simultaneously. Obtain the user's current cognitive load index, design the time allocation ratio of learning tasks in different dimensions according to the knapsack algorithm, classify the user's cognitive load level according to the preset cognitive load level, and set the number of questions in each dimension for different cognitive load levels according to the preset ratio. The design ratio of each dimension can be optimized based on the natural heuristic optimization algorithm.

[0127] Methods for obtaining the time allocation ratio for learning tasks of different dimensions include:

[0128] Obtain the weight value of each learning task in each dimension, the difficulty constraint range of each dimension, and the duration corresponding to the user's learning period. Treat each learning task in each dimension as an item, the value of the learning task as the task weight, and the weight of the learning task as the time required to complete the learning task. When selecting tasks, ensure that the difficulty parameters of the selected learning tasks meet the preset difficulty constraint range. Under the premise that the total time does not exceed the available time, select learning task combinations to maximize the total weight. Calculate the time allocation ratio of each task type based on the total weight of the task combination, and statistically analyze the time allocation ratio of learning tasks in different dimensions. Use the time allocation ratio as the time distribution ratio of learning tasks in different dimensions.

[0129] Methods for obtaining the difficulty parameters of a learning task include:

[0130] Obtain the user's most recent Z questions' answer data, count the number of correct answers, calculate the ratio of the number of correct answers to Z, and obtain the user's answer accuracy rate;

[0131] Based on maximum likelihood estimation, a model is built to represent user answer accuracy, user ability, and question difficulty. By fixing the question difficulty, user ability is calculated based on user answer accuracy.

[0132] Based on the user's ability and the preset question difficulty, the prediction accuracy is calculated, and the actual accuracy of the user's most recent M questions is statistically analyzed. The difficulty adjustment parameter is calculated based on the prediction accuracy and the actual accuracy. If the prediction accuracy is greater than the actual accuracy, the difficulty parameter is updated by adding the difficulty adjustment parameter to obtain the difficulty parameter of the learning task. If the prediction accuracy is less than the actual accuracy, the difficulty parameter is updated by subtracting the difficulty adjustment parameter from the difficulty parameter. In addition, the user's ability is updated based on the updated answer records.

[0133] Example 2

[0134] Reference Figure 2 This embodiment provides a method for optimizing complex learning tasks, including the following steps:

[0135] Collect interrupt events, which consist of a timestamp sequence of pause and resume events;

[0136] The structured features are optimized based on the interruption event to obtain optimized structured features;

[0137] Methods for obtaining optimized structured features include:

[0138] Obtain the timestamps and interruption events of the learning period, aggregate them through a preset time window, count the learning duration within the preset time window, calculate the ratio of the learning duration to the preset time of the preset time window, and obtain the effective learning duration percentage.

[0139] The interrupt frequency is obtained by counting the number of interrupt events within a preset time window and calculating the ratio of the number of occurrences to the preset time window.

[0140] The effective learning time percentage and interruption frequency are concatenated with the structured features to obtain optimized structured features.

[0141] The recommended learning period is obtained by processing the effective learning time ratio and historical learning periods in the optimized structured features; the interruption frequency and task type history records in the optimized structured features are processed to obtain the interruption risk score of the learning task.

[0142] Methods for obtaining recommended study time slots include:

[0143] Divide the day into time periods, obtain the time period to which the timestamp of the learning period belongs, and count the learning period with the highest percentage of effective learning time among all the time periods of the day. Select the learning period with the highest percentage of effective learning time as the recommended learning period. Concatenate the start time, end time and percentage of effective time of the recommended learning period to obtain the learning endurance feature.

[0144] Methods for obtaining an interruption risk score include:

[0145] Obtain the interruption frequency from the optimized structured features and the corresponding learning task type label. Use user logistic regression to predict the interruption probability for different task types, and use the interruption probability as an interruption risk score; for example... Wherein, P(n) i |i) represents the interruption probability of the i-th task type; e is a mathematical constant; α is an adjustment coefficient, obtained through natural heuristic optimization, serving as the baseline offset for the model and adjusting the basic level of the interruption probability; β is a frequency coefficient, obtained through natural heuristic optimization, used to measure the weight of the impact of interruption frequency on the interruption probability; if the absolute value of β is large, it indicates that the interruption frequency has a significant impact on the interruption probability, that is, the higher the interruption frequency, the more obvious the increase in the interruption probability; conversely, if β approaches 0, the impact of interruption frequency on the interruption probability is negligible. Both work together to enable the model to more accurately capture the correlation between task type and interruption behavior, improving the reliability of interruption risk prediction; n i The interrupt frequency for the i-th task type; n avg The average number of interruptions in the user's history (the average of the last 10 learning sessions); ∈ is a very small positive number used to avoid n avg=0. Converting interruption frequency into a measurable interruption risk score can intuitively reflect the likelihood of different task types triggering user interruptions in learning. For example, if the interruption probability calculation result for a certain type of task (such as difficult grammar questions) is high, it means that users are prone to interruption when performing this task. This can provide a basis for decision-making for the intelligent tutoring system, allowing adjustments to task difficulty, duration, or presentation methods (such as breaking down complex tasks) for task types with high interruption risk, thereby reducing the probability of learning interruption and improving learning continuity.

[0146] Optimize complex learning tasks based on recommended learning periods and interruption risk scores;

[0147] Methods for optimizing complex learning tasks include:

[0148] The available time for complex learning tasks is obtained based on recommended learning periods. Interruption monitoring is implemented based on the interruption risk score. If the current interruption risk score is greater than the preset interruption probability threshold, a rest reminder is automatically inserted or the learning task type is switched.

[0149] Example 3

[0150] Please see Figure 3 As shown, this embodiment provides an intelligent coaching adjustment system, including:

[0151] Data acquisition module: collects user data, behavioral data, and physiological data;

[0152] The first analysis module performs comprehensive feature extraction on user data, behavioral data, and physiological data to obtain structured features; it also performs comprehensive feature extraction on behavioral data and physiological data to obtain a cognitive load sequence.

[0153] The second analysis module processes structured features to obtain cross-dimensional features and analyzes cognitive load sequences to obtain difficulty constraints.

[0154] The third analysis module processes the structured features based on a predefined knowledge graph to obtain capability vectors; it then modifies the capability vectors based on an improved Ebbinghaus model to obtain modified capability vectors.

[0155] The fourth analysis module: Based on the TOPSIS multi-criteria decision-making algorithm, it analyzes the structured features according to the modified capability vector to obtain the reinforcement task queue; and processes the reinforcement task queue according to the cross-dimensional features to obtain the modified task queue.

[0156] The fifth analysis module dynamically adjusts the weights of the learning tasks based on the cognitive load index sequence to obtain the task weight matrix;

[0157] Task Adjustment Module: Based on difficulty constraints, the task queue is adjusted, and the task weight matrix is ​​analyzed to obtain composite learning tasks.

[0158] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0159] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent coaching adjustment method, characterized in that, The method comprises the following steps: Collecting user data, behavior data and physiological data; Comprehensive feature extraction is performed on the user data, behavior data and physiological data to obtain structured features; Comprehensive feature extraction is performed on the behavior data and physiological data to obtain a cognitive load sequence; the cognitive load index, data source label and corresponding timestamp in the structured features within a preset time period are obtained, and then spliced to obtain the cognitive load sequence; A preset first number of response times is obtained, the standard deviation and mean of the preset first number of response times are calculated, the ratio of the standard deviation to the mean of the preset first number of response times is calculated, and the answer time dispersion is obtained; A preset second number of physiological data is obtained, the standard deviation and mean of the preset second number of physiological data are calculated, the ratio of the standard deviation to the mean of the preset second number of physiological data is calculated, and the physiological features are obtained; The answer time dispersion and the physiological features are weighted to obtain the cognitive load index; The structured features are processed to obtain cross-dimension features; The cognitive load sequence is analyzed to obtain a difficulty constraint; the structured features are normalized to obtain standardized features, and the standardized features are used as input of a feature extraction model to obtain the cross-dimension features; According to a predefined knowledge graph, the oral error density, speech speed fluctuation rate, coherence index, phoneme error rate, vocabulary accuracy rate, weekly review frequency, grammar accuracy rate and defect matrix in the structured features are processed to obtain an ability vector spliced from oral mastery, vocabulary mastery rate, grammar mastery and listening mastery rate; the ability vector is corrected based on an improved Ebbinghaus model to obtain a corrected ability vector; The memory retention rate is calculated based on the cumulative knowledge base coefficient and the correct rate of the last C questions, and the corrected ability vector is obtained based on the product of the memory retention rate and the ability vector; Based on the TOPSIS multi-criteria decision algorithm, the structured features are analyzed according to the corrected ability vector to obtain a reinforcement task queue; The reinforcement task queue is processed according to the cross-dimension features to obtain a corrected task queue; The learning task weight is dynamically adjusted according to the cognitive load index sequence to obtain a task weight matrix; The composite learning task is obtained by analyzing the difficulty constraint, the corrected task queue and the task weight matrix: the first N knowledge points in the corrected task queue are obtained, a task sequence is generated, the task combination proportion in the task sequence is designed in combination with the task weight, and the composite task is obtained; The associated knowledge points in the knowledge graph are queried, and vertical advancement tasks and horizontal linkage tasks are respectively designed; the vertical advancement task represents sequentially learning N knowledge points, and the horizontal linkage task represents simultaneously learning N knowledge points; The current cognitive load index of the user is obtained, the time allocation proportion of learning tasks in different dimensions is designed according to the knapsack algorithm, the cognitive load level to which the user belongs is divided according to a preset cognitive load level, and the number of questions in each dimension for different cognitive load levels is set according to a preset proportion.

2. The intelligent coaching adjustment method of claim 1, wherein, The method for obtaining the structured features comprises: The oral audio is converted into oral text, compared with a preset answer text, the total number of words and the number of error words of the oral text are counted, the ratio of the number of error words to the total number of words is calculated, and the oral error density is obtained. The average value and the standard deviation of the number of words per second in the spoken language text are counted, the ratio of the standard deviation of the number of words per second to the average value of the number of words per second is calculated, and the speech speed fluctuation rate is obtained; the number of occurrences of a preset filler and the number of occurrences of adjacent repeated phrases in a unit time are counted, the ratio of the number of occurrences to the total number of words in the spoken language text is calculated, and the coherence index is obtained; the spoken language text is taken as the input of a pre-trained speech model, and the phoneme error rate is obtained; The number of questions with the learning task type label of the Rth column as vocabulary and the correctness label as 1 is obtained, the ratio of the number of questions with the correctness label as 1 to R is calculated, and the vocabulary accuracy rate is obtained; the total number of reviews of vocabulary by the user in the last H weeks is counted, and the ratio of the total number of reviews to H is calculated to obtain the average number of reviews per week; The number of questions with the learning task type label of the Eth column as grammar and the correctness label as 1 is obtained, the ratio of the number of questions with the correctness label as 1 to E is calculated, and the grammar accuracy rate is obtained; The ID of all questions with the correctness label as 0 is obtained, the knowledge point label corresponding to the question ID with the correctness label as 0 and the corresponding knowledge point basic coefficient are obtained according to the pre-defined knowledge graph, the number of errors corresponding to the same knowledge point label is counted, and the product of the number of errors and the knowledge point basic coefficient is obtained to obtain the knowledge point weight; the defect matrix is obtained by counting all knowledge point weights; The spoken language error density, the speech speed fluctuation rate, the coherence index, the phoneme error rate, the vocabulary accuracy rate, the average number of reviews per week, the grammar accuracy rate, the defect matrix and the cognitive load index, and the corresponding time stamp are spliced to obtain the structured feature.

3. The intelligent coaching adjustment method of claim 2, wherein, The method for obtaining the difficulty constraint comprises: The real-time cognitive load index is obtained, when the real-time cognitive load index is greater than the second preset upper limit of the load, the maximum new knowledge proportion is adjusted to a preset first threshold value, and the first difficulty constraint interval is [d-X, d), wherein d is the current ability of the user, and X is a first adjustment value; when the real-time cognitive load index is less than the second preset lower limit of the load, the maximum new knowledge proportion is adjusted to a preset second threshold value, and the second difficulty constraint interval is [d, d+Y), Y is a second adjustment value.

4. The intelligent coaching adjustment method of claim 3, wherein, The method for obtaining the ability vector comprises: The spoken language error density, the speech speed fluctuation rate, the coherence index and the phoneme error rate are weighted to obtain the spoken language mastery degree; The vocabulary accuracy rate and the average number of reviews per week are weighted to obtain the vocabulary mastery rate; According to the knowledge graph, the knowledge point weight corresponding to the question with the learning task type label as grammar in the structured feature and the knowledge point basic coefficient of the corresponding knowledge point in the knowledge graph are obtained, and the knowledge point weight and the knowledge point basic coefficient of each knowledge point are weighted to obtain the grammar mastery degree; The user's historical listening practice records are obtained, the correct rates at different speech speeds are counted, the highest speech speed with a correct rate not lower than a preset highest threshold value is obtained as an upper limit of recognition, and the lowest speech speed with a correct rate not lower than a preset lowest threshold value is obtained as a lower limit of recognition; the difference between the upper limit of recognition and the lower limit of recognition is calculated as the recognition speech speed range; The listening text and the list of words learned by the user are counted, the number of unlearned new words is counted, and the ratio of the number of new words to the total number of words in the listening text is calculated as the new word frequency. The listening mastery rate is obtained by weighting the recognition speech speed range and the frequency of new words; The ability vector is obtained by splicing the oral mastery degree, the vocabulary mastery rate, the grammar mastery degree and the listening mastery rate.

5. The intelligent coaching adjustment method of claim 4, wherein, The cumulative knowledge base coefficient is obtained by weighting the knowledge point base coefficient of the knowledge points mastered by the user in the knowledge graph.

6. The intelligent coaching adjustment method of claim 5, wherein, The method for obtaining the reinforcement task queue comprises: extracting the modified oral mastery degree, the modified vocabulary mastery rate, the modified grammar mastery degree and the modified listening comprehension threshold in the modified ability vector, and calculating the grammar ability deviation and the listening ability deviation corresponding to the modified oral mastery degree, the modified vocabulary mastery rate, the modified grammar mastery degree and the modified listening comprehension threshold; calculating the cognitive load adaptability based on the cognitive load index; standardizing the grammar ability deviation and the listening ability deviation to obtain the standardized deviation, standardizing the knowledge point base coefficient to obtain the standard knowledge coefficient, standardizing the cognitive load adaptability to obtain the standard cognitive load, weighting the standardized deviation, the standard knowledge coefficient and the standard cognitive load to obtain the standard matrix, obtaining the maximum value of the standardized deviation, the maximum value of the standard knowledge coefficient and the maximum value of the standard cognitive load, and weighting them to obtain the ideal solution, obtaining the minimum value of the standardized deviation, the minimum value of the standard knowledge coefficient and the minimum value of the standard cognitive load, and weighting them to obtain the negative ideal solution, calculating the distance from the standard matrix to the ideal solution and the negative ideal solution respectively, calculating the closeness degree based on the distance from the standard matrix to the ideal solution and the distance from the standard matrix to the negative ideal solution of each knowledge point, sorting the knowledge points in descending order based on the closeness degree, and obtaining the reinforcement task queue.

7. The intelligent coaching adjustment method of claim 6, wherein, The method for obtaining the modified task queue comprises: encoding the knowledge points in the reinforcement task queue in ascending order according to the sorting, calculating the contribution weight of each knowledge point to the cross-dimension feature based on the attention mechanism, multiplying the contribution weight and the encoding of the corresponding knowledge point in the reinforcement task queue to obtain the modified encoding, re-sorting the knowledge points in the reinforcement task queue in ascending order according to the modified encoding, and taking the re-sorting result as the modified task queue.

8. The intelligent coaching adjustment method of claim 7, wherein, The method for obtaining the task weight matrix comprises: statistically obtaining the accuracy rate of the learning tasks in different dimensions by calculating the ratio of the number of questions with a correctness label of 1 to the number of questions in the corresponding dimension, and obtaining the actual task load coefficient by combining the preset task load coefficient; wherein the different dimensions include oral, vocabulary, grammar and listening; obtaining the updated load coefficient by calculating based on the preset task load coefficient, the cognitive load index at the current moment and the smoothing factor; obtaining the task weight by combining the sensitivity coefficient, the cognitive load index at the current moment, the preset task load coefficient and the updated load coefficient, and obtaining the task weight matrix by splicing the task weights of different dimensions.

9. A smart coaching adjustment system implementing the smart coaching adjustment method of any one of claims 1-8, characterized by, It comprises: a data collection module for collecting user data, behavior data and physiological data; a first analysis module for comprehensively extracting features from the user data, behavior data and physiological data to obtain structured features; comprehensively extracting features from the behavior data and physiological data to obtain a cognitive load sequence; The second analysis module processes the structured features to obtain cross-dimension features; The cognitive load sequence is analyzed to obtain difficulty constraints; The third analysis module processes the structured features according to a predefined knowledge graph to obtain an ability vector, and corrects the ability vector based on an improved Ebbinghaus model to obtain a corrected ability vector; The fourth analysis module analyzes the structured features based on the corrected ability vector according to a TOPSIS multi-criteria decision algorithm to obtain a reinforcement task queue; The reinforcement task queue is processed according to the cross-dimension features to obtain a corrected task queue; The fifth analysis module dynamically adjusts learning task weights according to the cognitive load index sequence to obtain a task weight matrix; The task adjustment module analyzes the difficulty constraints, the corrected task queue and the task weight matrix to obtain a composite learning task.

Images