Digital art design AI personalized teaching method and device based on multi-modal data, equipment and medium

Abstract

The invention relates to a digital art AI personalized teaching method, device and equipment based on multi-modal data and a medium, and the method comprises the steps: collecting multi-source heterogeneous data in a student creation process in real time, and constructing a behavior feature data set through space-time alignment and feature fusion; generating a dynamic teaching strategy containing stage adaptive content and weak point special training based on the identification creation stage labels and knowledge weak points; a personalized instruction which can be executed by the terminal is generated through protocol compression and priority coding, and interactive teaching contents such as AR auxiliary prompt and three-dimensional material preview are driven; and finally, according to the teaching feedback data, iteratively optimizing the weight of the rule base and the trigger threshold, and realizing the self-evolution of the teaching strategy. According to the method, the problems of evaluation misaccuracy caused by a single data source, strategy stiffness caused by a static rule base and effect attenuation caused by lack of closed-loop optimization in the prior art are solved, and the accuracy and real-time performance of personalized teaching are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of AI personalized teaching technology, and in particular to a method, device, equipment and medium for AI personalized teaching of digital art design based on multimodal data. Background Art

[0002] In recent years, artificial intelligence (AI) technology has been gradually introduced into digital art and design education, aiming to improve teaching efficiency and personalization through technological means. This intelligent teaching method analyzes students' behavioral data during the creative process to provide real-time guidance and feedback, addressing issues such as insufficient teaching staff and delayed feedback in traditional teaching. This is particularly true in professional art schools and online education platforms, where systems must simultaneously process large amounts of student creative process data and dynamically adjust teaching strategies based on individual differences. This places higher demands on the real-time and adaptable nature of teaching systems.

[0003] However, existing technical solutions still have significant bottlenecks in achieving personalized teaching. First, most systems rely on a single data source such as canvas operation records or questionnaire feedback to evaluate the learning status, resulting in inaccurate judgment of the students' true creative intentions and cognitive status, and an inability to accurately locate weak links in knowledge. Secondly, the teaching strategy generation mechanism mostly uses a static rule base, which makes it difficult to dynamically adjust the teaching content and interaction methods according to the students' real-time cognitive changes, resulting in delayed teaching interventions or deviations from actual needs. Finally, the existing solutions lack closed-loop optimization capabilities, and the teaching rule base cannot be continuously iterated based on students' feedback data. After long-term use, the teaching effect gradually decays. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a digital art design AI personalized teaching method, device, equipment and medium based on multimodal data, which can capture multi-dimensional creative behavior in real time, dynamically generate personalized teaching strategies and achieve closed-loop self-optimization.

[0005] The purpose of the present invention is achieved by the following scheme:

[0006] In a first aspect, the present invention provides an AI personalized teaching method for digital art design based on multimodal data, comprising the following steps:

[0007] S1: Acquire and preprocess multimodal data from the student's learning terminal to generate a spatiotemporally aligned behavioral feature dataset. The multimodal data includes pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information.

[0008] S2: Perform learning state recognition processing on the behavioral feature dataset. Based on the pressure fluctuation data, trajectory motion characteristics, and gaze focus distribution, the student's creative stage is determined and the knowledge weaknesses in the current state are identified. The learning state label containing the creative stage label and the weak point identifier is generated.

[0009] S3: Generate personalized strategies based on learning status tags. Combined with the preset teaching rule library, the student's creation stage tags and weakness indicators are matched to generate dynamic teaching strategy information. The dynamic teaching strategy information is used to indicate targeted teaching interactions for students.

[0010] S4: Encode the dynamic teaching strategy information, generate personalized teaching instructions and send them to the student’s learning terminal;

[0011] S5: Obtain teaching feedback data from the student's learning terminal, adjust the teaching rule library based on the teaching feedback data, and generate an iterated teaching rule library.

[0012] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data, step S1 specifically comprising the following steps:

[0013] S11: Acquire multimodal data from the student's learning terminal, including pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information;

[0014] S12: aligning the timestamps of the pressure sensing data and the hand trajectory data of the multimodal data, eliminating the sampling rate difference, and smoothing the hand motion trajectory to generate tablet data;

[0015] S13: Perform spatial sampling correction on the multimodal eye tracking data, use B-spline surface fitting to correct the student's gaze coordinate offset, and generate spatially calibrated student eye tracking data;

[0016] S14: Extract features from the real-time canvas state information of multimodal data, analyze the layer structure and brush history created by the students, and generate canvas operation feature vectors;

[0017] S15: Data fusion is performed on the digital tablet data, the student's eye movement data, and the canvas operation feature vector, and time-space alignment is performed using timestamps to generate a behavioral feature dataset. The behavioral feature dataset is used to indicate the student's pressure fluctuation data, trajectory movement characteristics, and line of sight focus distribution during the creative process.

[0018] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data S2 specifically comprising the following steps:

[0019] S21: Perform stability analysis on the pressure fluctuation data in the behavioral feature dataset, calculate the variance index and mutation frequency of the pressure value within the time window, and generate a technique stability assessment result. The technique stability assessment result is used to indicate the student's control over the painting tool;

[0020] S22: Perform motion feature analysis on the hand trajectory data in the behavioral feature dataset, extract the velocity change rate and trajectory curvature features of the trajectory point sequence, and generate an operation proficiency index. The operation proficiency index is used to reflect the student's pen-handling fluency.

[0021] S23: Cognitive load modeling is performed based on the eye tracking data and real-time canvas status information in the behavioral feature dataset to generate a knowledge mastery deficit map. The knowledge mastery deficit map is used to identify the degree to which students neglect key knowledge points.

[0022] S24: Make a comprehensive decision based on the technical stability assessment results, operation proficiency indicators and knowledge mastery defect map, call the decision tree classifier to divide the basic ability level of the students' creation, determine the type of technical defects and locate the weak points, and generate a learning status label containing the creation stage label and the weak point identifier.

[0023] In one embodiment, S23 of a digital art design AI personalized teaching method based on multimodal data provided by the present invention specifically includes the following steps:

[0024] S231: Extract key teaching areas from the real-time canvas state information in the behavioral feature dataset, identify key knowledge areas of students' creations based on composition rules and color theory, and generate a coordinate set of interest areas;

[0025] S232 generates a heat map of the eye tracking data in the behavioral feature dataset, calculates the distribution density of the gaze points within a preset time window, and generates a heat map of the gaze points;

[0026] S233: Compare and analyze the gaze point heat map with the interest region coordinate set, calculate the gaze loss index of each region, and generate a knowledge mastery defect map.

[0027] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data, step S3 specifically comprising the following steps:

[0028] S31: Matching teaching resources with the creation stage tag of the learning status tag, querying the preset teaching rule library to obtain the adaptation resources of the corresponding stage, and generating a basic teaching content package;

[0029] S32: Targetedly strengthen the weak point identification of the learning status label, query the preset teaching rule library according to the defect type, extract special training content, and generate a weak point intervention plan;

[0030] S33: Perform strategic integration processing on the basic teaching content package and the weak point intervention plan, adjust the basic content weights and weak point intervention intensity of the students' teaching through the weight distribution algorithm, and generate a personalized teaching strategy framework;

[0031] S34: Obtain the device information of the student's learning terminal, adapt the interaction mode in combination with the personalized teaching strategy framework, add interactive instructions about AR auxiliary prompts and tactile feedback based on the learning terminal device type, and generate dynamic teaching strategy information.

[0032] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data. The calculation formula for the basic content weight and weak point intervention intensity is:

[0033]

[0034] W weak =1-W base

[0035] Among them, W base is the basic content weight, W weak The intervention intensity of the weak point is is a Sigmoid type attenuation function, S c is the stage confidence, 0 c <1, k is the adjustment factor.

[0036] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data S5 specifically includes the following steps:

[0037] S51: Obtain teaching feedback data from students' learning terminals, quantify the effectiveness of the teaching feedback data, extract indicators such as the time required to correct weaknesses and the extent of skill improvement, and generate a teaching intervention effect report;

[0038] S52: Conduct rule effectiveness analysis on the teaching intervention effect report, calculate the effectiveness score of each weak point intervention rule, and generate rule optimization suggestions;

[0039] S53: Dynamically adjust the weight of the rule optimization suggestions, update the matching weight of the teaching rule library, and generate an iterated teaching rule library.

[0040] In a second aspect, the present invention provides an AI personalized teaching device for digital art design based on multimodal data, which is configured with the following modules: ​

[0041] The multimodal data preprocessing module is used to obtain and preprocess the multimodal data from the student's learning terminal to generate a spatiotemporally aligned behavioral feature dataset. The multimodal data includes pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information.

[0042] The learning state recognition module is used to perform learning state recognition processing on the behavioral feature dataset. Based on the pressure fluctuation data, trajectory motion characteristics, and gaze focus distribution, it determines the student's creative stage and identifies the knowledge weaknesses in the current state, generating a learning state label containing the creative stage label and the weak point identifier.

[0043] The personalized strategy generation module is used to generate personalized strategies based on learning status tags. It matches students' creation stage tags and weak point identifiers with the preset teaching rule library to generate dynamic teaching strategy information. The dynamic teaching strategy information is used to instruct students to take targeted teaching interactions.

[0044] The teaching instruction encoding and sending module is used to encode dynamic teaching strategy information, generate personalized teaching instructions and send them to the student learning terminal;

[0045] The teaching rule library iteration module is used to obtain teaching feedback data from students' learning terminals, adjust the teaching rule library based on the teaching feedback data, and generate an iterated teaching rule library.

[0046] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements any of the above-mentioned AI personalized teaching methods for digital art design based on multimodal data.

[0047] In a fourth aspect, the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements any of the above-mentioned AI personalized teaching methods for digital art design based on multimodal data.

[0048] In summary, the present application provides a digital art design AI personalized teaching method based on multimodal data. Through the collaborative analysis of multimodal data and closed-loop optimization mechanism, it can achieve accurate mapping of creative behavior and cognitive state, and effectively solve the teaching inaccuracy problem caused by data singleness and rigid rules in the existing technology; the use of spatiotemporal alignment technology to fuse multi-source heterogeneous data such as tablet pressure, hand trajectory, eye tracking and canvas state can overcome the behavioral feature distortion caused by device differences, and construct a behavioral feature data set that reflects the true creative intention; through the coupled analysis of pressure fluctuations, trajectory dynamics and line of sight distribution, it can achieve the matching of creative stage and knowledge gap. Collaborative diagnosis of weaknesses can break through the limitations of traditional single-dimensional assessment and be used to generate learning status labels that are both stage-adaptive and defect-specific; teaching strategy information generated based on a dynamic rule matching mechanism can achieve real-time adaptation of teaching content and interaction methods, and can automatically adjust interactive elements such as AR auxiliary prompts and three-dimensional material demonstrations according to changes in students' cognitive states, in order to achieve personalized guidance of the "teaching-learning-practice" closed loop; ultimately, through iterative optimization of the rule library driven by teaching feedback, the self-evolution capability of teaching strategies can be established, which can continuously improve the accuracy and timeliness of intervention in weaknesses, in order to solve the industry pain point of attenuated effects due to long-term use.

[0049] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a personalized AI teaching method for digital art design based on multimodal data provided in an embodiment of the present application;

[0051] Figure 2 A schematic diagram of a process for generating a learning status label including a creation stage label and a weakness identifier provided in an embodiment of the present application;

[0052] Figure 3 A schematic diagram of a process for generating dynamic teaching strategy information provided in an embodiment of the present application;

[0053] Figure 4 A schematic structural diagram of a digital art design AI personalized teaching device based on multimodal data provided in another embodiment of the present application. DETAILED DESCRIPTION

[0054] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0056] In one embodiment, Figure 1 As shown, a digital art design AI personalized teaching method based on multimodal data is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0057] S1: Acquire and preprocess the multimodal data of the student's learning terminal to generate a spatiotemporally aligned behavioral feature dataset. The multimodal data includes pressure data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information.

[0058] Specifically, the system collects pressure data through the pressure sensor built into the digital tablet. The data includes the pen tip pressure value, tilt angle and azimuth, and the sampling frequency is set to 200Hz; the hand trajectory data is collected through the electromagnetic induction positioning system of the digital tablet. The data includes the pen tip on the X / Y axis coordinates and movement speed, and the curvature radius, acceleration change rate and other derived parameters of the trajectory are generated synchronously.

[0059] At the same time, the system uses an infrared eye tracker to collect eye tracking data, including pupil center coordinates, blink frequency, and gaze dwell duration. Corneal reflection is used to calculate the mapping position of the gaze point in the canvas coordinate system, with a sampling frequency of 120Hz. The system captures canvas status information in real time through the design software SDK interface. This information includes layer structure, pixel modification history, and tool call sequence. Layer structure includes layer name, transparency, and blending mode; pixel modification history includes coordinate range and color value RGB channel changes; and tool call sequence includes parameters such as brush type and eraser radius.

[0060] The system preprocesses the collected data, including data cleaning, time synchronization, and feature extraction. The data cleaning step removes outliers, noise, and erroneous records to ensure data accuracy and reliability. Time synchronization establishes a unified timestamp coordinate system for data from different modalities, achieving spatiotemporal alignment. Feature extraction extracts key behavioral features from the preprocessed data, such as the average pressure and standard deviation of pressure data, the length and smoothness of hand trajectories, the number of fixations and the proportion of fixation durations in eye movement data, and the color diversity index and line complexity of the canvas state. This creates a spatiotemporally aligned behavioral feature dataset, which includes pressure feature vectors, trajectory kinematic parameters, eye movement heatmaps, and canvas state summaries.

[0061] S2: Perform learning status recognition processing on the behavioral feature dataset, determine the student's creative stage based on pressure fluctuation data, trajectory motion characteristics, and gaze focus distribution, identify the knowledge weaknesses in the current state, and generate a learning status label containing the creative stage label and weakness identifier.

[0062] Specifically, the system performs learning status recognition processing on the behavioral feature data set, determines the student's creative stage and identifies knowledge weaknesses, and generates a learning status label containing a creative stage label and a weakness identifier. The system comprehensively analyzes multi-dimensional behavioral characteristics such as pressure fluctuation data, trajectory movement characteristics, and line of sight focus distribution, and matches them with the preset creative stage feature pattern to determine the student's current creative stage. For example, in the early stages of creation, eye movement trajectories are scattered, pressure is light and fluctuates little, and hand movements are hesitant and slow; in the refinement stage, eye movement trajectories are concentrated in a local area, gaze duration increases, pressure is stable and the force is moderate, and hand movements are fine and regular.

[0063] The system generates a label for the student's current creative stage based on the creative stage classification criteria. It also compares the student's behavioral characteristic data with the normal behavioral characteristic range for that creative stage, identifying characteristic indicators that deviate from the normal range. Combining the teaching knowledge system and the FAQ database, the system identifies the student's knowledge weaknesses and generates a label. For example, during the coloring stage, if color selection is inconsistent, color changes are frequent, and color transitions between canvas areas are unnatural, the system identifies the student's weaknesses in color matching knowledge and generates a corresponding label.

[0064] S3: Generate personalized strategies based on learning status labels, match students’ creation stage labels and weakness identifiers with the preset teaching rule library, and generate dynamic teaching strategy information. The dynamic teaching strategy information is used to indicate targeted teaching interactions for students.

[0065] Specifically, the system generates personalized strategies based on learning status labels, matches students' creative stage labels and weak point identifiers with a preset teaching rule library, and generates dynamic teaching strategy information to instruct students to take targeted teaching interactions. It should be noted that the teaching rule library constructed in this application contains a variety of teaching strategy templates, each of which corresponds to a specific combination of creative stage labels and weak point identifiers, and is established based on the experience summary of teaching experts, teaching theory, and statistical analysis of a large amount of student data.

[0066] The system searches for matching teaching strategy templates in the teaching rule library based on the student's learning status label, and combines the student's learning history and individual characteristics, such as learning style and learning progress, to personalize and combine the teaching strategy, and generate dynamic teaching strategy information. Based on the dynamic teaching strategy information, the system designs specific teaching interaction methods, such as pop-up teaching prompt windows, voice guidance, video demonstrations, virtual teacher role guidance, and targeted practice task push, etc., to clarify the triggering conditions, presentation timing and content display form of each interaction method, to ensure that the teaching strategy can be applied to the student's learning terminal in an intuitive and effective way.

[0067] S4: Encode the dynamic teaching strategy information, generate personalized teaching instructions and send them to the student's learning terminal.

[0068] Specifically, the system encodes dynamic teaching strategy information, including video URLs, annotation coordinates, and prompt text. The system converts this information into a UTF-8-encoded string and encrypts it using Base64. The system defines eight interactive command types, including 0x01 for pop-up prompts, 0x03 for canvas annotations, and 0x05 for operation guidance. Binary encoding is used to represent command types and parameters, including the RGB value of the annotation color and the prompt display duration. The system adds a CRC32 checksum and a timestamp during the encoding process, ultimately encapsulating the data into a command packet no longer than 1024 bytes.

[0069] The system establishes a persistent connection between the learning terminal and the server via the WebSocket protocol, with a heartbeat interval set to 30 seconds to ensure command transmission delays are kept within 200ms. The server-side command queue uses priority scheduling, with weakness correction instructions taking precedence over stage guidance instructions. After receiving the command, the learning terminal decodes and verifies it, executes the corresponding operation through the local rendering engine, calls the system pop-up API, draws guide lines on the canvas layer, and sends a confirmation of command receipt to the server.

[0070] S5: Obtain teaching feedback data from the student's learning terminal, adjust the teaching rule library based on the teaching feedback data, and generate an iterated teaching rule library.

[0071] Specifically, the system collects information on students' implementation of personalized teaching instructions and feedback through the feedback interface of the students' learning terminals, including their understanding of the teaching content, their acceptance of teaching methods, their response to teaching interactions, improvements in their work creation, and their subjective evaluations. Specifically, the system continuously monitors changes in students' creative behavior data after receiving teaching instructions, compares the data differences before and after teaching, and analyzes the actual impact of teaching interventions on students' creations. The system uses data mining and machine learning algorithms to analyze the collected teaching feedback data, evaluate the effectiveness, applicability, and pertinence of each rule template in the existing teaching rule library, identify ineffective or inapplicable rules, as well as students' new knowledge needs and problem types, and explore potential connections between different creative stages and weaknesses.

[0072] Based on the feedback data analysis results, the system makes targeted adjustments and optimizations to the teaching rule library, updates the teaching strategy templates, expands the rules for identifying knowledge weaknesses, improves the teaching interaction design, and perfects the standards for dividing the creation stages, generating an iterative teaching rule library so that the teaching system can continuously adapt to changes in students' learning needs and continuously improve the quality of personalized teaching.

[0073] In summary, the present application provides a digital art design AI personalized teaching method based on multimodal data. Through the collaborative analysis of multimodal data and closed-loop optimization mechanism, it can achieve accurate mapping of creative behavior and cognitive state, and effectively solve the teaching inaccuracy problem caused by data singleness and rigid rules in the existing technology; the use of spatiotemporal alignment technology to fuse multi-source heterogeneous data such as tablet pressure, hand trajectory, eye tracking and canvas state can overcome the behavioral feature distortion caused by device differences, and construct a behavioral feature data set that reflects the true creative intention; through the coupled analysis of pressure fluctuations, trajectory dynamics and line of sight distribution, it can achieve the matching of creative stage and knowledge gap. Collaborative diagnosis of weaknesses can break through the limitations of traditional single-dimensional assessment and be used to generate learning status labels that are both stage-adaptive and defect-specific; teaching strategy information generated based on a dynamic rule matching mechanism can achieve real-time adaptation of teaching content and interaction methods, and can automatically adjust interactive elements such as AR auxiliary prompts and three-dimensional material demonstrations according to changes in students' cognitive states, in order to achieve personalized guidance of the "teaching-learning-practice" closed loop; ultimately, through iterative optimization of the rule library driven by teaching feedback, the self-evolution capability of teaching strategies can be established, which can continuously improve the accuracy and timeliness of intervention in weaknesses, in order to solve the industry pain point of attenuated effects due to long-term use.

[0074] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data, step S1 specifically comprising the following steps:

[0075] S11: Acquire multimodal data from the student's learning terminal, where the multimodal data includes pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information.

[0076] Specifically, the system acquires multimodal data from the student's learning terminal, including the tablet's pressure data, hand trajectory data, the student's eye tracking data, and real-time canvas status information. The system collects pressure data through the pressure sensor integrated in the tablet. The sensor uses an electromagnetic induction pressure detection module. The collected parameters include the pen tip pressure value, X / Y axis tilt angle, and azimuth angle. The data is transmitted to the terminal processing unit in real time through the USB-C interface. The hand trajectory data is collected through the tablet's infrared positioning module, which records the pen tip's real-time coordinates in the two-dimensional plane, the movement speed (unit: mm / s), and the movement acceleration (unit: mm / s). 2 ) and is transmitted synchronously with the pressure-sensing data through the same interface; eye tracking data is collected by an infrared eye tracker integrated above the terminal display screen. The device uses corneal reflection and pupil center dual positioning technology. The collection parameters include the three-dimensional coordinates of the left / right eye pupil center, the line of sight direction vector, the number of blinks and the duration of each blink. The data is transmitted to the system processing module through the HDMI interface; real-time canvas status information is obtained through the SDK interface of the digital art creation software. The interface supports real-time monitoring of the canvas. The content obtained includes the currently activated layer ID, the pixel matrix of each layer, the type parameters of the brush tool, and historical operation records. The information update frequency is synchronized with the canvas operation trigger.

[0077] S12: aligning the timestamps of the pressure sensing data and the hand trajectory data of the multimodal data, eliminating the sampling rate difference, smoothing the hand motion trajectory, and generating tablet data.

[0078] Specifically, the system extracts the original timestamps of the pressure data and hand trajectory data, and establishes a timeline based on the system's unified clock, with a minimum scale of milliseconds. To account for the sampling rate differences between the pressure data and hand trajectory data, the system can use interpolation to upsample the lower-sampling-rate data or a sliding window method to downsample the higher-sampling-rate data, maintaining consistent sampling intervals between the two types of data.

[0079] At the same time, the system smoothes the hand trajectory data after timestamp alignment. It can use a smoothing algorithm to eliminate jump points in the trajectory and calculate the first-order derivatives and second-order derivatives of adjacent trajectory points. When the absolute value of the derivative exceeds the preset threshold, the abnormal point is corrected by fitting the neighborhood points. The tablet data generated after processing contains a timestamp sequence, the pressure value at the corresponding moment, the pressure change rate, coordinate information and movement direction parameters. The data is stored in an array, and the array length corresponds to the total length of the time axis.

[0080] S13: Perform spatial sampling correction on the eye tracking data of the multimodal data, use B-spline surface fitting to correct the student's line of sight coordinate offset, and generate spatially calibrated student eye movement data.

[0081] Specifically, before students begin their work, the system guides them through a multi-point calibration process, collecting the original eye coordinates and actual coordinates of the students as they gaze at pre-set calibration points on the screen. The calibration points are distributed around the edges and center of the screen. The system uses the original coordinates of the calibration points as input and the actual coordinates as output, constructing a surface fitting model with a specific order and a grid of control points. The system then feeds the real-time eye-tracking data into the fitting model, calculating the corrected gaze coordinates through surface interpolation to eliminate coordinate offsets caused by head movement and device installation errors.

[0082] Preferably, the system performs outlier filtering on the corrected eye movement data, and can use statistical criteria to eliminate data points that are beyond the normal gaze range. The generated spatially calibrated eye movement data contains the corrected gaze point coordinates, gaze duration, and line of sight movement vector. The data is arranged in ascending order by timestamp, and the gaze duration is calculated by the time difference between adjacent gaze points.

[0083] S14: Extract features from the real-time canvas status information of multimodal data, analyze the layer structure and brush history created by the students, and generate canvas operation feature vectors.

[0084] Specifically, the system performs feature extraction on real-time canvas status information. The system parses the layer structure data in the canvas status information, extracts the number of layers, the stacking order of each layer, transparency parameters, blending mode parameters and pixel coverage. The pixel coverage is the ratio of the effective pixels of the layer to the total pixels of the canvas, and converts these parameters into layer feature vectors. The system parses the brush history, counts the frequency of brush types used in a specific time period, calculates the mean and discreteness of the brush size, records the distribution range of the brush hardness, and converts these statistics into brush feature vectors. The system extracts the time characteristics of canvas operations, including the layer creation interval, brush switching frequency and the number of undo / redo operations, and converts them into time feature vectors. The system concatenates the layer feature vector, brush feature vector and time feature vector by column to generate a canvas operation feature vector. Each element of the vector is normalized, and the feature vector is refreshed at fixed time intervals as the canvas status is updated.

[0085] S15: Data fusion is performed on the digital tablet data, the student's eye movement data, and the canvas operation feature vector, and time-space alignment is performed using timestamps to generate a behavioral feature dataset. The behavioral feature dataset is used to indicate the student's pressure fluctuation data, trajectory movement characteristics, and line of sight focus distribution during the creative process.

[0086] Specifically, the system uses timestamps as indexes and maps the three types of data onto the same timeline. This timeline is then divided into fixed-interval time slices, each corresponding to a data fusion unit. Within each fusion unit, the system extracts the pressure value dispersion parameters and trajectory point curvature parameters from the tablet data, the gaze point coordinate distribution parameters from the student's eye movement data, and the mean of the canvas operation feature vector within that time slice.

[0087] The system combines the above features in a specific order to form a time slice feature vector. The system concatenates the feature vectors of multiple consecutive time slices in chronological order to form behavioral feature dataset segments. These segments are linked by timestamps to form a complete behavioral feature dataset. The dataset includes a time range marker and a feature vector matrix for each time slice, stored in a feature table in a time series database. The table structure includes the dataset identifier, student identifier, creative task identifier, feature matrix data, and generation time.

[0088] In one embodiment, Figure 2 As shown, the present invention provides a digital art design AI personalized teaching method based on multimodal data S2 specifically includes the following steps:

[0089] S21: Perform stability analysis on the pressure fluctuation data in the behavioral feature dataset, calculate the variance index and mutation frequency of the pressure value within the time window, and generate a technical stability evaluation result. The technical stability evaluation result is used to indicate the student's control over the painting tools.

[0090] Specifically, the system extracts a sequence of pressure values ​​within each time window from the behavioral feature data set, and the sequence is derived from the pressure value field in the tablet data. Preferably, when calculating the variance index, the system can use the sample variance formula, that is, the sum of the squares of the deviations of all pressure values ​​in each time window from the mean pressure value of the window is divided by the number of data points minus one. The resulting variance value reflects the degree of discreteness of the pressure fluctuation, and the size of the variance value is inversely proportional to the stability of the pressure control. When calculating the mutation frequency, the system determines the pressure mutation threshold based on the pressure standard deviation of the previous time window. When the difference in the pressure values ​​of two consecutive data points exceeds the threshold, it is recorded as a pressure mutation. The total number of mutations in each time window is the mutation frequency.

[0091] The system normalizes the variance index and mutation frequency, and converts them to the [0,1] interval through Min-Max standardization. The variance index and mutation frequency are both positively mapped, that is, the larger the value, the worse the pressure control stability. The technical stability evaluation results are represented by a two-dimensional vector, which includes the normalized variance value and the normalized mutation frequency value, and are associated with the corresponding time window number to form the student's technical stability curve in different creative periods.

[0092] S22: Perform motion feature analysis on the hand trajectory data in the behavioral feature dataset, extract the velocity change rate and trajectory curvature features of the trajectory point sequence, and generate an operation proficiency index. The operation proficiency index is used to reflect the student's pen-handling fluency.

[0093] Specifically, the system obtains a sequence of hand trajectory points in each time window from the behavioral feature dataset. The sequence contains the X / Y coordinates of the trajectory points and the corresponding timestamps. Based on this, the instantaneous speed of each trajectory point is calculated. The instantaneous speed is obtained by the ratio of the distance between two adjacent points to the time interval. At the same time, the speed change rate is calculated by the ratio of the instantaneous speed difference between two adjacent trajectory points to the time interval. The speed change rate sequence in each time window is averaged after taking the absolute value to obtain the average speed change rate of the window. The value is inversely proportional to the degree of change in the pen speed. The trajectory curvature feature is calculated by the curvature value of each trajectory point. The curvature value is estimated using the three-point method, that is, the current point and one adjacent point before and after form a triangle. The curvature is determined by the inverse of the radius of the triangle's circumscribed circle. The average of the curvature sequence in each time window is used to obtain the average trajectory curvature. The value is inversely proportional to the smoothness of the pen movement. The system normalizes the average speed change rate and average trajectory curvature, converts them into standard scores using Z-score standardization, and then obtains the operation proficiency index through weighted summation. The index value is proportional to the level of operation proficiency.

[0094] S23: Cognitive load modeling is performed on the eye tracking data and real-time canvas status information in the behavioral feature dataset to generate a knowledge mastery deficit map. The knowledge mastery deficit map is used to identify the degree to which students neglect key knowledge points.

[0095] Preferably, the knowledge mastery deficiency map is generated by the following steps:

[0096] S231: Extract key teaching areas from the real-time canvas state information in the behavioral feature dataset, identify key knowledge areas created by students based on composition rules and color theory, and generate a coordinate set of interest areas.

[0097] Specifically, the system parses the layer structure, element distribution and color parameters in the real-time canvas status information. The layer structure includes the element type and position coordinates of each layer, the element distribution includes the spatial arrangement of graphics, lines, and color blocks, and the color parameters include hue, brightness, and saturation values. Preferably, the system can call a pre-stored composition rule library, which includes balance principles, proportional relationships, rhythm rules, etc., and analyze the spatial layout of elements through a graphic recognition algorithm to locate areas that meet key composition nodes. At the same time, the system calls a color theory library, which includes contrast rules, harmony principles, cold and warm properties, etc., and uses a color analysis algorithm to identify color block combination areas that have exemplary color theory significance.

[0098] The system marks the areas that conform to the rules of composition and color theory as key knowledge areas, extracts the boundary coordinates of each area, and determines the boundary coordinates by the maximum and minimum X-axis and Y-axis coordinates of the elements in the area. These coordinates are grouped by area to generate a coordinate set of the area of ​​interest. The coordinate set is stored in the form of an array, and each array element corresponds to the boundary coordinate value of a key knowledge area.

[0099] S232 generates a heat map of the eye tracking data in the behavioral feature dataset, counts the distribution density of gaze points within a preset time window, and generates a gaze point heat map.

[0100] Specifically, the system generates a heat map of the eye tracking data in the behavioral feature dataset, and extracts the eye tracking data from the behavioral feature dataset. The data contains the canvas coordinates of each gaze point and the corresponding timestamp. Preferably, the system divides the preset time windows into fixed time windows, and the length of the time window is set according to the average duration of the creation process. Each time window is continuous and non-overlapping. The system counts the distribution of all gaze points in each time window, divides the canvas into uniform grid units, calculates the number of gaze points in each grid unit, and determines the distribution density based on the number of gaze points in the grid unit. The distribution density is calculated by the ratio of the number of gaze points in the unit to the unit area.

[0101] The system maps the distribution density to visual presentation parameters, and generates a gaze point heat map covering the entire canvas range according to the density corresponding to different visual identifiers. Each pixel point of the heat map corresponds to the distribution density of a grid unit. The heat map is stored in image data format and is consistent with the canvas coordinate system.

[0102] S233: Compare and analyze the gaze point heat map with the interest region coordinate set, calculate the gaze loss index of each region, and generate a knowledge mastery defect map.

[0103] Specifically, the system establishes a coordinate mapping relationship, aligning the canvas coordinate system of the gaze heat map with the coordinate system of the ROI coordinate set to ensure precise spatial correspondence between the two. For each key knowledge area in the ROI coordinate set, the system extracts the density value data corresponding to that area in the gaze heat map, calculates the average gaze density of that area, and compares it with a preset baseline density value.

[0104] The system normalizes the difference between the average gaze density and the baseline density value to obtain a gaze loss index for each area. A negative index value indicates that the actual gaze density is lower than the baseline value. The larger the absolute value of the negative value, the lower the degree of attention paid to the area. Finally, the system summarizes the gaze loss index within each time window by knowledge category label to form a knowledge mastery deficiency map. The map uses time as the horizontal axis and knowledge category as the vertical axis to present the gaze loss situation of different knowledge categories within each time window in matrix form, intuitively identifying which key knowledge points students lack attention to during the creative process.

[0105] S24: Make a comprehensive decision based on the technical stability assessment results, operation proficiency indicators and knowledge mastery defect map, call the decision tree classifier to divide the basic ability level of the students' creation, determine the type of technical defects and locate the weak points, and generate a learning status label containing the creation stage label and the weak point identifier.

[0106] Specifically, the decision tree classifier is constructed using the C4.5 algorithm, and the training samples are derived from historical data of students of different levels. Each sample contains the variance index and mutation frequency of the technical stability assessment results, the operation proficiency index, the deviation rate of each region of the knowledge mastery defect map, and the corresponding manually labeled level and defect type. Among them, the first-level nodes of the classifier use the operation proficiency index as the split attribute to divide the basic ability level into elementary, intermediate, and advanced levels; the second-level nodes are based on the variance index of the technical stability assessment results, and subdivide the technical defect types at each level. For example, a higher variance corresponds to "unstable pressure control" and a higher mutation frequency corresponds to "sudden change in pen strength"; the third-level nodes locate weak points based on the areas with lower deviation rates in the knowledge mastery defect map. For example, a lower deviation rate in the perspective area indicates "weak perspective knowledge."

[0107] The system calls the trained decision tree classifier to infer the real-time input feature data, outputs the basic ability level, technical defect type and weak point location, and determines the student's creative stage and generates a creative stage label based on the time progress of the creative process and the proportion of completed creative content. The basic ability level, technical defect type, weak point information and creative stage label are integrated to form a structured learning status label, which is stored in the form of a data field and contains a stage identification code and a weak point identification code.

[0108] In one embodiment, Figure 3 As shown, S3 of the AI ​​personalized teaching method for digital art design based on multimodal data provided by the present invention specifically includes the following steps:

[0109] S31: Match the teaching resources with the creation stage label of the learning status label, query the preset teaching rule library to obtain the adaptation resources of the corresponding stage, and generate a basic teaching content package.

[0110] Specifically, the system queries the preset teaching rule library, which stores the adaptation resources corresponding to different creation stages. The system uses the creation stage label as the query key to retrieve the corresponding resource records in the rule library. The system extracts the matched resource types, resource paths and resource parameters. Resource types include tutorial videos, graphic explanations, sample materials, etc. Resource paths point to storage locations, and resource parameters define playback duration, display order, etc. The system integrates these resource information to generate a basic teaching content package. The content package is stored in the form of structured data and contains a resource list and presentation logic. In order to measure the accuracy of resource matching, the system uses the following formula to calculate the matching degree:

[0111]

[0112] Among them, M represents resource matching degree, R i represents the relevance score of the i-th resource, W i Represents the weight of the i-th resource. Relevance score R i Calculated based on the relevance of resources to the creation stage, the weight W i Adjust based on resource usage frequency and historical performance.

[0113] S32: Targetedly strengthen the weak point identification of the learning status label, query the preset teaching rule library according to the defect type, extract special training content, and generate a weak point intervention plan.

[0114] Specifically, the system analyzes the defect type corresponding to the weakness identification and queries the preset teaching rule library again. The system extracts special training content, which includes practice tasks, simulation cases, guidance prompts, and other types. The system adjusts the difficulty level and quantity of training content based on the severity of the weakness and the trainee's historical data. The system integrates the special training content and generates a weakness intervention plan, which includes a list of training tasks and the execution order. To determine the training intensity, the system uses the following formula to calculate the training intensity coefficient:

[0115]

[0116] Among them, I represents the training intensity coefficient, S j Indicates the severity of the jth weak point, D j Indicates the difficulty coefficient of the jth weak point; severity S j The difficulty coefficient D is calculated based on the frequency and impact range of students' errors in their weak points. j The difficulty level of the weak point is set according to the syllabus.

[0117] S33: Carry out strategic fusion processing on the basic teaching content package and the weak point intervention plan, adjust the basic content weight of student teaching and the weak point intervention intensity through the weight distribution algorithm, and generate a personalized teaching strategy framework.

[0118] Specifically, the system can use a weight distribution algorithm to comprehensively consider factors such as the student's historical learning data, current creation progress, and the priority of weak points to calculate the basic content weight and weak point intervention strength. Preferably, the calculation formula for the basic content weight and weak point intervention strength is:

[0119]

[0120] W weak =1-W base

[0121] Among them, W base is the weight of basic content, which indicates the proportion of basic teaching content in the overall teaching strategy; W weak The strength of intervention for weak points indicates the proportion of intervention plans for weak points in the overall teaching strategy; is a Sigmoid type attenuation function, which is used to achieve a smooth transition of weights; S c is the stage confidence, the value range is 0 c <1, reflecting the reliability of the creation stage label; k is the adjustment factor, which is used to adjust the sensitivity of the weight to the change of stage confidence. The system extracts the stage confidence S from the learning state label c The specific value of W is substituted into the formula to calculate base and W weak .

[0122] Based on the calculation results, the system adjusts the content ratios of the foundational teaching content package and the weak point intervention plan. The foundational content weight determines the presentation duration and frequency of the foundational content in the teaching sequence, while the weak point intervention intensity determines the number of repetitions of the specialized training content and the level of detailed interactive feedback. The system integrates the adjusted foundational teaching content and weak point intervention content in chronological order and logical relationships, generating a personalized teaching strategy framework that includes the content presentation sequence, duration allocation for each section, and interactive feedback nodes.

[0123] S34: Obtain the device information of the student's learning terminal, adapt the interaction mode in combination with the personalized teaching strategy framework, add interactive instructions about AR auxiliary prompts and tactile feedback based on the learning terminal device type, and generate dynamic teaching strategy information.

[0124] ​Specifically, the system obtains the device information of the student's learning terminal, which includes parameters such as terminal type, screen resolution, and supported interactive functions. The system adapts the interactive mode in conjunction with the personalized teaching strategy framework. The system determines whether the terminal supports AR functions and tactile feedback. If so, it adds interactive instructions for AR auxiliary prompts and tactile feedback to the strategy framework. AR auxiliary prompts are achieved by superimposing virtual guidance elements on the canvas, and tactile feedback is achieved by calling the vibration function of the terminal.

[0125] The system generates dynamic teaching strategy information, which includes teaching content, resource paths, interactive instructions, and execution logic. The system encodes the dynamic teaching strategy information into a data format that can be parsed by the terminal, preparing for subsequent instruction transmission and execution. Preferably, in order to evaluate the effect of interactive adaptation, the system uses the following formula to calculate the adaptation degree:

[0126]

[0127] Among them, A represents the interactive fitness, C k represents the compatibility score of the kth interaction function, V k Indicates the usage value of the kth interactive function. Compatibility score C k Calculate the value V based on the support level of the terminal device for interactive functions. k The degree of improvement of teaching effect is set according to the interactive function.

[0128] In one embodiment, the present invention provides a digital art design AI personalized teaching method based on multimodal data S5 specifically includes the following steps:

[0129] S51: Obtain teaching feedback data from students’ learning terminals, quantify the effectiveness of the teaching feedback data, extract indicators of time to correct weaknesses and degree of skill improvement, and generate a teaching intervention effect report.

[0130] Specifically, the system collects teaching feedback data sent by the student's learning terminal through the terminal data interface. The feedback data includes the student's viewing record of the teaching content, the operation sequence of modification according to prompts, the comparison of the canvas status before and after modification, and the subjective evaluation information submitted through the terminal interactive interface.

[0131] Specifically, the system pre-processes the feedback data, filters out invalid records such as empty data generated by erroneous operations, and sorts the valid data by timestamp. At the same time, it calculates the time it takes to correct the weak points, that is, the time interval from the sending of the teaching intervention instruction to the student's operation to make the weak point mark disappear. The time node when the weak point disappears is determined by comparing the behavioral feature data sets before and after the intervention.

[0132] The system calculates the skill improvement index by extracting the difference between the skill stability assessment results and the operational proficiency index within the same time window before and after the intervention. A positive difference indicates skill improvement, and a larger difference indicates a greater improvement. The system associates the weakness correction time and skill improvement index with the corresponding teaching strategy identifier, and integrates them chronologically into a teaching intervention effectiveness report. The report includes the strategy type, execution duration, correction time, improvement magnitude, and associated weakness type for each intervention.

[0133] S52: Conduct rule utility analysis on the teaching intervention effect report, calculate the effectiveness score of each weak point intervention rule, and generate rule optimization suggestions.

[0134] Specifically, the system conducts a rule-effectiveness analysis on teaching intervention effectiveness reports and extracts intervention rules corresponding to each weakness from the teaching rule library. These rules include rule identifiers, applicable conditions, teaching content generation logic, and interaction method definitions. The system associates the weakness correction time and skill improvement indicators in the teaching intervention effectiveness reports with the corresponding intervention rules and calculates the effectiveness score for each intervention rule. The effectiveness score is calculated by weighting the ratio of the skill improvement indicator to the weakness correction time, combined with the instruction response rate. The score reflects the rule's effectiveness in improving the weakness and the learner's acceptance of the rule.

[0135] The system sorts the effectiveness scores, identifies intervention rules with scores below the preset threshold, analyzes potential problems in these rules in terms of matching of applicable conditions, targeted teaching content, and adaptability of interactive methods, and generates rule optimization suggestions, including the direction of rule parameter adjustment, content replacement plans, and interaction method improvement measures.

[0136] S53: Dynamically adjust the weight of the rule optimization suggestions, update the matching weight of the teaching rule library, and generate an iterated teaching rule library.

[0137] Specifically, the system dynamically adjusts the weights of rule optimization suggestions. The system analyzes the rule optimization suggestions and determines the specific rule entries in the teaching rule library that need to be adjusted and the adjustment direction. Preferably, the system can call a preset weight adjustment algorithm, which determines the weight adjustment range based on the degree of deviation between the rule effectiveness score and the preset benchmark value. The higher the score, the greater the increase in the matching weight of the rule. The matching weight of the rule with a score below the threshold is reduced or remains unchanged.

[0138] The system updates the matching weights for the corresponding rules in the teaching rule library and records the timestamp, basis, and original weight values ​​of the weight adjustment, creating a weight adjustment log. The system performs a consistency check on the adjusted teaching rule library, checking for conflicts between rules and resolving them by prioritizing them. Once the check passes, an iterated teaching rule library is generated. This new rule library retains valid rules while integrating optimization suggestions, improving the accuracy and adaptability of subsequent teaching strategy generation.

[0139] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0140] Based on the same inventive concept, the embodiments of the present application also provide a multimodal data-based AI personalized teaching device for digital art design, which is used to implement the multimodal data-based AI personalized teaching method for digital art design. The implementation solution provided by this device is similar to the implementation solution described in the above-mentioned method. Therefore, the specific limitations of one or more embodiments of the multimodal data-based AI personalized teaching device for digital art design provided below can be found in the above-mentioned limitations of the multimodal data-based AI personalized teaching method for digital art design, and will not be repeated here.

[0141] In a second aspect, the present invention provides a digital art design AI personalized teaching device 600 based on multimodal data, which is configured with the following modules:

[0142] Multimodal data preprocessing module 610 is used to obtain and preprocess multimodal data from the student's learning terminal to generate a spatiotemporally aligned behavioral feature dataset. The multimodal data includes tablet pressure data and hand trajectory data, the student's eye tracking data, and real-time canvas status information.

[0143] The learning state identification module 620 is used to perform learning state identification processing on the behavioral feature dataset, determine the student's creative stage based on the pressure fluctuation data, trajectory motion characteristics, and gaze focus distribution, identify the knowledge weaknesses in the current state, and generate a learning state label containing the creative stage label and weakness identifier;

[0144] The personalized strategy generation module 630 is used to generate personalized strategies based on learning status tags, match students' creative stage tags and weak point identifiers in combination with a preset teaching rule library, and generate dynamic teaching strategy information. The dynamic teaching strategy information is used to indicate targeted teaching interactions for students.

[0145] The teaching instruction encoding and sending module 640 is used to encode the dynamic teaching strategy information, generate personalized teaching instructions and send them to the student learning terminal;

[0146] The teaching rule library iteration module 650 is used to obtain teaching feedback data from the student's learning terminal, adjust the teaching rule library based on the teaching feedback data, and generate an iterated teaching rule library.

[0147] Preferably, the multimodal data preprocessing module 610 provided in this application is configured with the following units:

[0148] A multimodal data acquisition unit, configured to acquire multimodal data from a student's learning terminal, including tablet pressure data and hand trajectory data, the student's eye tracking data, and real-time canvas status information;

[0149] A digital tablet data processing unit is used to align the timestamps of the pressure-sensing data and the hand trajectory data in the multimodal data, eliminate sampling rate differences, smooth the hand motion trajectory, and generate digital tablet data;

[0150] The eye movement data calibration unit is used to perform spatial sampling correction on the eye tracking data in the multimodal data, and use B-spline surface fitting to correct the student's line of sight coordinate offset to generate spatially calibrated student eye movement data;

[0151] The canvas feature extraction unit is used to extract features from the real-time canvas state information in the multimodal data, analyze the layer structure and brush history of the students' creation, and generate canvas operation feature vectors;

[0152] The spatiotemporal data fusion unit is used to fuse the tablet data, the student's eye movement data and the canvas operation feature vector, and use the timestamp to perform spatiotemporal alignment to generate a behavioral feature data set. The behavioral feature data set is used to indicate the student's pressure fluctuation data, trajectory movement characteristics and line of sight focus distribution during the creative process.

[0153] Preferably, the learning state identification module 620 provided in this application is configured with the following units:

[0154] The technique stability assessment unit is used to perform stability analysis on the pressure fluctuation data in the behavioral feature dataset, calculate the variance index and mutation frequency of the pressure value within the time window, and generate a technique stability assessment result. The technique stability assessment result is used to indicate the student's control over the painting tool;

[0155] The operation proficiency analysis unit is used to analyze the motion characteristics of the hand trajectory data in the behavior feature dataset, extract the velocity change rate and trajectory curvature characteristics of the trajectory point sequence, and generate the operation proficiency index, which is used to reflect the student's pen-handling fluency;

[0156] The cognitive load modeling unit is used to perform cognitive load modeling on the eye tracking data and real-time canvas status information in the behavioral feature dataset, generating a knowledge mastery deficit map. The knowledge mastery deficit map is used to identify the degree to which students neglect key knowledge points.

[0157] The comprehensive learning status judgment unit is used to make comprehensive decisions based on the technical stability assessment results, operation proficiency indicators and knowledge mastery defect maps. It calls the decision tree classifier to divide the basic ability level of the students' creation, determine the type of technical defects and locate the weak points, and generate a learning status label containing the creation stage label and the weak point identifier.

[0158] Preferably, the cognitive load modeling unit includes a teaching key area extraction subunit, an eye movement heat map generation subunit, and a gaze loss analysis subunit. The teaching key area extraction subunit is used to extract teaching key areas from the real-time canvas state information in the behavioral feature data set, identify key knowledge areas created by students based on composition rules and color theory, and generate a set of interest area coordinates; the eye movement heat map generation subunit is used to generate a heat map from the eye tracking data in the behavioral feature data set, calculate the distribution density of gaze points within a preset time window, and generate a gaze point heat map; the gaze loss analysis subunit is used to compare and analyze the gaze point heat map with the interest area coordinate set, calculate the gaze loss index of each area, and generate a knowledge mastery defect map.

[0159] Preferably, the personalized strategy generation module 630 provided in this application is configured with the following units:

[0160] The stage resource matching unit is used to match the teaching resources with the creation stage tag of the learning status tag, query the preset teaching rule library to obtain the adaptation resources of the corresponding stage, and generate the basic teaching content package;

[0161] The weak point reinforcement plan unit is used to carry out targeted reinforcement of the weak point identification of the learning status label, query the preset teaching rule library according to the defect type, extract special training content, and generate a weak point intervention plan;

[0162] The teaching strategy fusion unit is used to integrate the basic teaching content package and the weak point intervention plan. It adjusts the basic content weight and weak point intervention intensity of the students' teaching through the weight distribution algorithm to generate a personalized teaching strategy framework.

[0163] The interactive mode adaptation unit is used to obtain the device information of the student's learning terminal, adapt the interactive mode in combination with the personalized teaching strategy framework, add interactive instructions about AR auxiliary prompts and tactile feedback based on the learning terminal device type, and generate dynamic teaching strategy information.

[0164] Preferably, the teaching rule library iteration module 650 provided in this application is configured with the following units:

[0165] The teaching feedback effectiveness quantification unit is used to obtain teaching feedback data from students' learning terminals, quantify the effectiveness of the teaching feedback data, extract indicators such as the time to correct weaknesses and the degree of skill improvement, and generate a teaching intervention effect report;

[0166] The rule effectiveness analysis unit is used to perform rule effectiveness analysis on the teaching intervention effect report, calculate the effectiveness score of each weak point intervention rule, and generate rule optimization suggestions;

[0167] The rule weight updating unit is used to dynamically adjust the weight of the rule optimization suggestions, update the matching weight of the teaching rule library, and generate an iterated teaching rule library.

[0168] In one embodiment, the present application also provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the above-mentioned AI personalized teaching method for digital art design based on multimodal data is implemented.

[0169] In one embodiment, the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned AI personalized teaching method for digital art design based on multimodal data.

[0170] In the description of this specification, the reference terms "one embodiment," "some embodiments," "example," "specific example," or "some examples" mean that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described may be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art may combine and integrate different embodiments or examples described in this specification, as well as features of different embodiments or examples, unless they are mutually inconsistent.

[0171] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0172] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A personalized AI teaching method for digital art design based on multimodal data, characterized by: The following steps are involved: S1: Acquire and preprocess multimodal data from the student's learning terminal to generate a spatiotemporally aligned behavioral feature dataset. The multimodal data includes pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information. S2: Performing learning state recognition processing on the behavioral feature dataset, judging the student's creative stage based on the pressure fluctuation data, trajectory motion characteristics, and gaze focus distribution, and identifying the knowledge weaknesses in the current state, and generating a learning state label including a creative stage label and a weakness identifier; S3: Generate personalized strategies based on the learning status labels, match the student's creative stage labels and weak point identifiers in combination with a preset teaching rule library, and generate dynamic teaching strategy information. The dynamic teaching strategy information is used to instruct targeted teaching interactions for the student; S4: Encoding the dynamic teaching strategy information, generating personalized teaching instructions and sending them to the student's learning terminal; S5: Acquire teaching feedback data from the student's learning terminal, adjust the teaching rule library based on the teaching feedback data, and generate an iterated teaching rule library.

2. The method according to claim 1, characterized in that Said S1 comprises: S11: Acquire multimodal data from the student's learning terminal, wherein the multimodal data includes pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information; S12: performing time stamp alignment on the pressure sensing data and the hand trajectory data of the multimodal data, eliminating sampling rate differences and smoothing the hand motion trajectory to generate tablet data; S13: performing spatial sampling correction on the eye tracking data of the multimodal data, correcting the student's line of sight coordinate offset using B-spline surface fitting, and generating spatially calibrated student eye tracking data; S14: extracting features from the real-time canvas state information of the multimodal data, analyzing the layer structure and brush history records created by the student, and generating a canvas operation feature vector; S15: Fusing the tablet data, the student's eye movement data, and the canvas operation feature vector, and using timestamps for spatiotemporal alignment, to generate a behavioral feature dataset, where the behavioral feature dataset is used to indicate the student's pressure fluctuation data, trajectory motion characteristics, and line of sight focus distribution during the creative process.

3. The method according to claim 1, characterized in that The S2 includes: S21: performing stability analysis on the pressure fluctuation data in the behavioral feature dataset, calculating the variance index and mutation frequency of the pressure value within the time window, and generating a technique stability evaluation result, wherein the technique stability evaluation result is used to indicate the student's control over the painting tool; S22: performing motion feature analysis on the hand trajectory data in the behavioral feature dataset, extracting the velocity change rate and trajectory curvature features of the trajectory point sequence, and generating an operation proficiency index, wherein the operation proficiency index is used to reflect the student's pen-handling fluency; S23: performing cognitive load modeling on the eye tracking data and real-time canvas state information in the behavioral feature dataset to generate a knowledge mastery deficit map, wherein the knowledge mastery deficit map is used to identify the degree to which the learner neglects key knowledge points; S24: Make a comprehensive decision based on the technical stability assessment results, the operational proficiency index and the knowledge mastery defect map, call the decision tree classifier to divide the basic ability level of the students' creation, determine the type of technical defects and locate the weak points, and generate a learning status label including the creation stage label and the weak point identifier.

4. The method according to claim 3, characterized in that The S23 includes: S231: extracting key teaching areas from the real-time canvas state information in the behavioral feature dataset, identifying key knowledge areas of the student's creation based on composition rules and color theory, and generating a coordinate set of interest areas; S232 generates a heat map of the eye tracking data in the behavioral feature dataset, calculates the distribution density of gaze points within a preset time window, and generates a gaze point heat map; S233: Compare and analyze the gaze point heat map with the interest region coordinate set, calculate the gaze loss index of each region, and generate a knowledge mastery deficiency map.

5. The method according to claim 1, wherein The S3 includes: S31: Matching teaching resources with the creation stage tag of the learning status tag, querying a preset teaching rule library to obtain adapted resources for the corresponding stage, and generating a basic teaching content package; S32: Targetedly strengthen the weak point identification of the learning status label, query the preset teaching rule library according to the defect type, extract special training content, and generate a weak point intervention plan; S33: Performing strategy fusion processing on the basic teaching content package and the weak point intervention plan, adjusting the basic content weights and weak point intervention intensity of the student's teaching through a weight distribution algorithm, and generating a personalized teaching strategy framework; S34: Obtain device information of the student's learning terminal, perform interactive mode adaptation processing in combination with the personalized teaching strategy framework, add interactive instructions about AR auxiliary prompts and tactile feedback based on the learning terminal device type, and generate dynamic teaching strategy information.

6. The method according to claim 5, characterized in that The calculation formula for the basic content weight and weak point intervention intensity is: IN weak =1-W base Among them, W base is the basic content weight, W weak The intervention intensity of the weak point is is a Sigmoid type attenuation function, S c is the stage confidence, 0 c <1, k is the adjustment factor.​ 7. The method according to any one of claims 1 to 6, characterized in that The S5 includes: S51: Obtain teaching feedback data from students' learning terminals, quantify the effectiveness of the teaching feedback data, extract indicators such as the time required to correct weaknesses and the extent of skill improvement, and generate a teaching intervention effect report; S52: performing rule utility analysis on the teaching intervention effect report, calculating the effectiveness score of each weak point intervention rule, and generating rule optimization suggestions; S53: Dynamically adjust the weight of the rule optimization suggestion, update the matching weight of the teaching rule library, and generate an iterated teaching rule library.

8. A digital art design AI personalized teaching device based on multimodal data, characterized by: The device comprises: A multimodal data preprocessing module is used to obtain and preprocess multimodal data from the student's learning terminal to generate a spatiotemporally aligned behavioral feature dataset. The multimodal data includes pressure sensing data and hand trajectory data on the tablet, the student's eye tracking data, and real-time canvas status information. A learning state recognition module is used to perform learning state recognition processing on the behavioral feature dataset, determine the student's creative stage based on the pressure fluctuation data, trajectory motion characteristics, and gaze focus distribution, identify the knowledge weaknesses in the current state, and generate a learning state label containing a creative stage label and a weakness identifier; A personalized strategy generation module is used to generate personalized strategies based on the learning status tags, match the students' creation stage tags and weakness identifiers in combination with a preset teaching rule library, and generate dynamic teaching strategy information. The dynamic teaching strategy information is used to instruct targeted teaching interactions for students; A teaching instruction encoding and sending module is used to encode the dynamic teaching strategy information, generate personalized teaching instructions and send them to the student learning terminal; The teaching rule library iteration module is used to obtain teaching feedback data from the student learning terminal, adjust the teaching rule library based on the teaching feedback data, and generate an iterated teaching rule library.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images