Deep learning-based artistic course personalized learning path method and system

Abstract

The invention provides an artistic course personalized learning path method and system based on deep learning. Firstly, an artistic course learning basic information set of a learner and a preset artistic course module library are acquired; calling a pre-trained deep learning course dynamic association model to generate association strength description of the learning basis of the learner and each course module; screening and sorting based on association strength description to form an initial learning module sequence; acquiring a real-time learning behavior data set of the first course module learned by the learner, and generating a module adjustment signal; and adjusting the initial learning module sequence according to the module adjustment signal to obtain a personalized learning path adaptive to the real-time learning state of the learner, and generating a personalized learning path document comprising a course module learning sequence and module connection guidance, thereby realizing accurate customization and dynamic optimization of the learning path, and improving the learning efficiency. And the learning effect of artistic courses is improved.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and more specifically, to a method and system for personalized learning paths for art courses based on deep learning. Background Technology

[0002] In the field of arts education, developing suitable learning paths for learners is crucial for improving learning outcomes and stimulating interest. However, existing methods for planning learning paths in arts courses have many limitations.

[0003] On the one hand, traditional methods often arrange learning content based on fixed syllabi and teacher experience, lacking sufficient consideration for individual differences among learners. Different learners have different art learning backgrounds; some may have already been exposed to some art knowledge, some have unique performance in art practice, and others have a strong desire to learn in a specific art direction. However, traditional methods struggle to plan precise learning paths based on these individual characteristics, resulting in a mismatch between learning content and learners' actual needs, which affects learning motivation and effectiveness.

[0004] On the other hand, existing methods lack a dynamic adjustment mechanism during the learning process. Learners' learning states are constantly changing. In the initial stages of learning, they may show different levels of acceptance and comprehension of certain course modules. However, traditional methods cannot adjust the learning path in a timely manner based on learners' real-time learning behavior. This may cause learners to encounter difficulties or feel that the learning content is not challenging enough, thus failing to fully realize their potential. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for personalized learning paths in art courses based on deep learning, the method comprising:

[0006] The system acquires a set of basic information about the learner's art courses and a pre-set library of art course modules. The set of basic information about the learner's art courses includes the art course knowledge items that the learner has already encountered, the learner's practical performance records in art courses, and the learner's expected direction of art course learning. The pre-set library of art course modules includes multiple art course modules with knowledge-related relationships.

[0007] The pre-trained deep learning course dynamic association model is invoked, and the learner's set of basic learning information for art courses is input into the deep learning course dynamic association model to generate a description of the association strength between the learner's learning foundation and each course module in the preset art course module library.

[0008] Based on the association strength description, an initial course module subset is selected from the preset art course module library, and the initial course module subset is sorted according to the association strength to form an initial learning module sequence;

[0009] Collect a set of real-time learning behavior data generated by learners when learning the first course module in the initial learning module sequence, and input the set of real-time learning behavior data into the deep learning course dynamic association model to generate a module adjustment signal;

[0010] Based on the module adjustment signal, the initial learning module sequence is replaced or its order is adjusted to obtain a personalized learning path that adapts to the learner's real-time learning status. Based on the personalized learning path, a personalized learning path document containing the learning order of course modules and module connection guidance is generated.

[0011] In another aspect, embodiments of the present invention also provide a personalized learning path system for art courses based on deep learning, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the machine-readable storage medium to implement the above-described method.

[0012] Based on the above, this embodiment of the invention integrates the learner's individual characteristics and art course knowledge system by acquiring a set of basic learning information for art courses and a pre-set art course module library. It then invokes a pre-trained deep learning dynamic association model to accurately generate a description of the correlation strength between the learner's learning foundation and each course module, achieving quantitative analysis of the relationship between the learner and the course modules. Based on the correlation strength description, an initial subset of course modules is selected and sorted to form an initial learning module sequence, initially planning a learning path suitable for the learner's foundation. Real-time learning behavior data of the learner during the first course module is collected and input again into the deep learning dynamic association model to generate module adjustment signals, enabling timely capture of changes in the learner's real-time learning state. The initial learning module sequence is dynamically adjusted according to the module adjustment signals to obtain a personalized learning path adapted to the learner's real-time learning state, generating a personalized learning path document with detailed guidance. This achieves precise customization and dynamic optimization of the learning path, effectively improving the relevance and effectiveness of art course learning and stimulating learners' learning interest and potential. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the execution flow of the personalized learning path method for art courses based on deep learning provided in an embodiment of the present invention.

[0014] Figure 2This is a schematic diagram of exemplary hardware and software components of the personalized learning path system for art courses based on deep learning provided in an embodiment of the present invention. Detailed Implementation

[0015] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for personalized learning paths in art courses based on deep learning, provided in one embodiment of the present invention. The following is a detailed description of this method for personalized learning paths in art courses based on deep learning.

[0016] Step S110: Obtain the learner's basic information set on art courses and the preset art course module library. The learner's basic information set on art courses includes the art course knowledge items that the learner has been exposed to, the learner's practical performance records in art courses, and the learner's expected direction of art course learning. The preset art course module library contains multiple art course modules with knowledge-related relationships.

[0017] This embodiment uses painting-related art courses as a unified application scenario, covering sub-fields such as basic sketching, still life watercolor, landscape oil painting, and portraiture. Specific content includes sketching perspective training, watercolor water control exercises, oil painting color mixing, and Impressionist landscape creation. Two types of core data are synchronized through the user center and course resource library interface of the painting course learning management platform.

[0018] The learner's basic information on art courses is organized by dimension: the knowledge items already encountered cover "single-point perspective in sketching", "dry painting techniques in watercolor", "the application of impasto brushstrokes in oil painting", and "the theory of the three primary colors of color", etc. Each knowledge item is marked with the level of mastery (understanding, familiar, mastery) and associated with the corresponding learning time record; the practical performance record includes works such as sketching geometric shapes, watercolor still life studies, and oil paintings submitted in the past six months, with comments (such as "the vanishing point of perspective is accurate but the line expression is insufficient" and "the color transition is natural but the brushstroke layering is monotonous"), suggestions for improvement, and records of the completion scene (such as home practice and offline training); the expected learning direction is clearly "Impressionist landscape oil painting creation", with sub-preferences including "Monet-style light and shadow expression" and "Renoir's use of color", and the expected goal is "to independently complete an outdoor landscape oil painting work that conforms to the Impressionist style".

[0019] The pre-designed art course module library is constructed according to the logic of knowledge progression and ability development, containing 15 core modules: M01 - Basic Sketching (Perspective and Lines), M02 - Advanced Sketching (Light and Shadow and Texture), M03 - Basic Watercolor (Tools and Water Control), M04 - Still Life Watercolor (Light and Shadow Shaping), M05 - Landscape Watercolor (Spatial Layers), M06 - Basic Oil Painting (Materials and Substrate Treatment), M07 - Oil Painting Color (Harmony and Contrast), M08 - Oil Painting Brushstrokes (Wet and Dry, Thick and Thin), M09 - Still Life Oil Painting (Detail Depiction), M10 - Introduction to Landscape Oil Painting (Composition and Tone), M11 - Advanced Landscape Oil Painting (Light and Shadow and Atmosphere), M12 - Basic Impressionist Oil Painting (Color Juxtaposition), M13 - Impressionist Landscape Techniques (Outdoor Sketching), M14 - Oil Painting Creation Guidance (Theme and Style), M15 - Comprehensive Creative Practice (Artwork Refinement and Display). Each module includes a knowledge graph (marking prerequisite dependencies and subsequent extensions with other modules), competency development goals (e.g., M12 focuses on "color perception ability" and "rapid capture of light and shadow"), a list of teaching resources (video tutorials, illustrated manuals, and sample works), and suggested learning scenarios (online self-study and offline practical training).

[0020] To address the privacy concerns of learners' phone numbers and home addresses in the collected data, field anonymization was implemented, replacing them with random characters. Learning records and project data were transmitted via an encrypted tunnel and stored using distributed encrypted storage. Access permissions were set up with tiered controls, allowing only teachers and administrators to view the complete data, thus preventing privacy leaks.

[0021] Step S120: Call the pre-trained deep learning course dynamic association model, input the learner's art course learning foundation information set into the deep learning course dynamic association model, and generate a description of the association strength between the learner's learning foundation and each course module in the preset art course module library.

[0022] The system invokes a deep learning course dynamic association model deployed on a cloud server. This model is built on the Transformer-XL architecture, incorporating multi-head attention and residual connections. It comprises four layers: an input layer, an association computation layer, a weight fusion layer, and an output layer. The model has been trained and validated using data from over 500,000 learners. Learning information is converted into feature vectors and input into the model. Multi-dimensional association analysis is then used to generate a description of the association strength. The specific process is as follows.

[0023] Step S121: Activate the input layer processing unit of the deep learning course dynamic association model, and convert the knowledge items, practice performance records and learning directions in the learner's art course learning basic information set into structured feature vectors respectively. The dimension of each structured feature vector is consistent with the preset dimension of the input layer of the deep learning course dynamic association model.

[0024] The input layer's feature encoding unit is activated to perform structured processing on three types of information: knowledge items are encoded using a bag-of-words model combined with pre-trained art domain word vectors. Each knowledge item is mapped to a fixed-dimensional vector, with vector components reflecting the correlation between the knowledge and other art concepts. The level of mastery is adjusted by the vector scaling factor. Practice performance records are extracted using a convolutional neural network to extract visual features of the works (such as line density, color distribution, and brushstroke direction). These features are then combined with sentiment analysis and keyword extraction results from the comment text to form a practice feature vector, which includes dimensions such as skill gaps, strengths, and progress trends. Learning directions are extracted using a topic model to extract core demand keywords (such as "Impressionism," "landscape," and "outdoor sketching"), which are mapped to directional feature vectors. The vector components correspond to the matching degree of different learning direction categories.

[0025] The lengths of the three feature vectors are adjusted by the dimension unification module to match the number of neurons in the model's input layer, ensuring that the data format meets the model's input requirements and forming structured feature vectors.

[0026] Step S122: Input the transformed multiple structured feature vectors into the association calculation layer of the deep learning course dynamic association model. The association calculation layer includes a knowledge association sublayer, an ability association sublayer, and a direction association sublayer.

[0027] The knowledge feature vector, practice feature vector, and direction feature vector are respectively input into the three parallel sub-layers of the association calculation layer. Each sub-layer uses a dedicated network structure for association analysis. The specific process is as follows.

[0028] Step S1221: After receiving the structured feature vector containing the feature components of knowledge items through the knowledge association sublayer, the built-in knowledge graph matching unit is called. The knowledge graph matching unit loads the art course knowledge graph, which contains the hierarchical and association relationships between each knowledge item.

[0029] After receiving the knowledge feature vector, the knowledge association sublayer activates the knowledge graph matching unit and loads a dedicated knowledge graph for painting courses. This dedicated knowledge graph for painting courses uses "Fundamentals of Painting Art" as its root node, with four first-level branches: "Fundamentals of Form," "Fundamentals of Color," "Materials and Techniques," and "Styles and Schools." Each first-level branch has second-level nodes (e.g., "Fundamentals of Form - Sketching - Perspective" and "Fundamentals of Color - Theory - Contrast"); second-level nodes have tertiary nodes (e.g., "Perspective - Single-Point Perspective - Vanishing Point Positioning" and "Contrast - Hue Contrast - Complementary Colors"). Nodes are connected through four relationships: "Prerequisite Learning," "Skill Support," "Style Derivation," and "Cross-Media Application." For example, "Single-Point Perspective in Sketching" and "Landscape Oil Painting Composition" have a "Skill Support" relationship, and "Oil Painting Color Harmony" and "Impressionist Color Juxtaposition" have a "Style Derivation" relationship. The graph also links typical teaching cases and explanations of common misconceptions for each node.

[0030] Step S1222: Through the knowledge graph matching unit, the knowledge item feature components in the structured feature vector are mapped to the knowledge coverage features of each course module in the preset art course module library and then to the art course knowledge graph to determine the node positions and path distances between the nodes in the art course knowledge graph.

[0031] The entries such as "single-point perspective in sketching" and "impasto brushstrokes in oil painting" in the knowledge feature vector are mapped to the third-level nodes of the knowledge graph; at the same time, the knowledge coverage features of each course module are extracted (such as the knowledge coverage features of M12-Impressionist Oil Painting Basics being "color juxtaposition techniques", "quick color mixing techniques", and "outdoor light and shadow capture"), and are also mapped to the corresponding nodes in the graph.

[0032] The shortest path algorithm is used to calculate the path between learner knowledge nodes and module knowledge nodes. The path is based on the direct relationship between nodes. For example, the path between the learner's "color theory" node and the M12 "color juxtaposition technique" node is "color theory - color contrast - color juxtaposition". The relationship path and the number of nodes between the two are clearly defined.

[0033] Step S1223: Calculate the knowledge association coefficient based on the path distance between nodes. Use the knowledge association coefficient as the core calculation basis for the knowledge dimension association value. The shorter the path distance, the larger the knowledge association coefficient.

[0034] The knowledge association coefficient is calculated based on the negative correlation between path distance and knowledge relevance. The shorter the path distance, the stronger the connection between the learner's existing knowledge and the knowledge required by the module, and the larger the association coefficient. When the path distance exceeds a set range, the association coefficient is zero, indicating that the connection between the two is extremely low. For each course module, the association coefficients of all its knowledge coverage nodes and learner knowledge nodes are combined, and the average value is taken as the knowledge dimension association value of the module, reflecting the degree of matching between the learner's knowledge base and the module's knowledge requirements.

[0035] Step S1224: After receiving the structured feature vector containing practical performance feature components through the ability association sublayer, start the ability dimension analysis unit. The ability dimension analysis unit contains multiple ability assessment dimensions, and each ability assessment dimension corresponds to a core ability in art course learning.

[0036] After receiving the practical feature vector, the competency-related sublayer activates the competency dimension analysis unit. This unit sets up eight core competency assessment dimensions for painting courses: modeling ability (proportion, structural accuracy), line expression ability (smoothness, expressiveness), light and shadow shaping ability (light and shadow levels, three-dimensionality), color application ability (harmony, contrast, emotional expression), brushstroke control ability (diversity, targeted approach), composition ability (balance, highlighting key elements), scene capture ability (rapid observation, summarization), and creative expression ability (style integration, theme presentation). Each dimension is associated with corresponding assessment indicators (e.g., modeling ability is associated with "vanishing point accuracy" and "human proportion error rate").

[0037] Step S1225: Decompose the practical performance feature components in the structured feature vector into each ability assessment dimension, calculate the performance score under each ability assessment dimension, and extract the training intensity value of the ability training features of each course module under the corresponding ability assessment dimension.

[0038] The work features and comments keywords in the practice feature vector are decomposed into 8 ability dimensions according to their correspondence: for example, "insufficient line expression" corresponds to a low score in line expression ability, and "natural color transition" corresponds to a high score in color application ability; combined with the improvement directions in the modification suggestions (such as "strengthening brushstroke layering practice"), the development potential of each dimension is determined.

[0039] The competency development characteristics of each course module were extracted. Each module has a development intensity setting in 8 dimensions, reflecting the emphasis of the module on specific competencies (e.g., the scene capture ability development intensity of M13-Impressionist landscape technique is the highest, followed by the ability to use color). The development intensity is determined based on the module's teaching objectives and historical teaching effects.

[0040] Step S1226: By comparing the matching degree between performance scores and cultivation intensity values, generate the capability matching coefficient for each capability assessment dimension, and calculate the average of the capability matching coefficients for all capability assessment dimensions to obtain the capability dimension correlation value.

[0041] For each competency dimension, the degree of fit between the learner's performance score and the module's training intensity is compared: if a learner performs poorly in a certain dimension and the module's training intensity for that dimension is high, the match is high and the competency matching coefficient is large; if a learner has reached proficiency in a certain dimension and the module's training intensity for that dimension is low, the match is low and the competency matching coefficient is small. The average of the competency matching coefficients across all dimensions is taken as the competency dimension correlation value for that module, reflecting the degree of fit between the learner's current competency and the module's training objectives.

[0042] Step S1227: After receiving the structured feature vector containing the learning direction feature components through the direction association sublayer, activate the direction matching analysis unit. The direction matching analysis unit loads a preset learning direction classification system. The learning direction classification system includes multiple subdivided learning direction categories and the core learning objective corresponding to each category.

[0043] After receiving the directional feature vector, the directional association sublayer activates the directional matching analysis unit and loads the painting course learning direction classification system. This painting course learning direction classification system is divided into four primary categories: "Traditional Realism," "Impressionism," "Abstract Expressionism," and "Contemporary Syntheticism." Each primary category is further divided into secondary sub-directions (such as "Impressionism-Landscape," "Impressionism-Portrait," and "Impressionism-Still Life"). Each secondary direction is associated with core learning objectives (such as "Impressionism-Landscape," whose core objectives are "mastering the techniques of quickly capturing outdoor light and shadow" and "using color juxtaposition to express atmosphere"), representative artists, and typical work style analysis.

[0044] Step S1228: Classify the learning direction feature components in the structured feature vector into the corresponding categories in the learning direction classification system, determine the learner's core learning objectives, and analyze the correlation between the goal-oriented features of each course module and the core learning objectives.

[0045] Learners’ directional characteristics of “Impressionist landscape oil painting creation” are categorized into the secondary category of “Impressionism-Landscape”. Based on their preferences, the core learning objectives are determined as “mastering the expression of light, shadow and color in the style of Monet” and “improving the ability to quickly compose in outdoor sketching”.

[0046] Extract the goal-oriented characteristics of each course module and analyze their correlation with the core learning objectives: the higher the overlap between the module's teaching content, case demonstrations, practical tasks and the core objectives, the stronger the correlation; for example, the correlation between M12-Impressionist Oil Painting Fundamentals and M13-Impressionist Landscape Techniques and the core objectives is significantly higher than that of M06-Oil Painting Fundamentals.

[0047] Step S1229: Generate a directional fit coefficient based on the degree of correlation. Use the directional fit coefficient as a component of the directional dimension correlation value. The higher the degree of correlation, the larger the directional fit coefficient.

[0048] The degree of correlation between a module and the core learning objective is directly converted into a directional fit coefficient. The higher the correlation, the larger the coefficient, indicating that the module better matches the learner's learning direction needs. This directional fit coefficient is the directional dimension correlation value, reflecting the degree of matching between the module and the learner's desired direction.

[0049] Step S123: Input the knowledge dimension association value, ability dimension association value and direction dimension association value into the weight fusion layer of the deep learning course dynamic association model. The weight fusion layer performs weighted summation of the association values ​​of the three dimensions according to the weight coefficients of each dimension determined during the model training phase.

[0050] The weighted fusion layer loads weight coefficients determined through cross-validation and grid search during the model training phase. These weight coefficients are optimized and adjusted based on historical learner learning performance data, reflecting the degree of influence of each dimension on the learning path's suitability. The correlation values ​​of the three dimensions—knowledge, ability, and direction—are weighted and summed according to the weight coefficients to obtain the comprehensive correlation value for each course module, comprehensively reflecting the overall suitability between the learner and the module.

[0051] Step S124: Through the output layer of the deep learning course dynamic association model, the weighted summation result is converted into an association index with a uniform range. Each course module corresponds to an association index, and the association indices of all course modules together constitute a description of the association strength between the learner's learning foundation and each course module.

[0052] The output layer generates correlation indices through normalization and anomaly correction processes, as follows.

[0053] For example, in step S1241: the output layer of the deep learning course dynamic association model receives the weighted summation result from the weight fusion layer, and the normalization processing unit is started. The normalization processing unit loads a preset association index range.

[0054] After receiving the comprehensive correlation values ​​from all modules, the output layer starts the normalization processing unit and loads the preset correlation index range. This correlation index range is determined based on the distribution characteristics of a large amount of historical data to ensure that the correlation indices of different learners are comparable.

[0055] Step S1242: Input the weighted summation result corresponding to each course module into the normalization processing unit, and map the weighted summation result to the preset correlation index range through linear transformation to obtain the preliminary correlation index.

[0056] By using a linear normalization method, the comprehensive correlation values ​​of each course module are mapped to a preset range, eliminating the impact of differences in dimensions and making the initial correlation index uniform, which facilitates horizontal comparison.

[0057] Step S1243: Activate the outlier correction unit to perform a range check on the preliminary correlation index. If the preliminary correlation index exceeds the preset correlation index range, it is corrected to the range boundary value; if the preliminary correlation index is within the range, it remains unchanged, and the final correlation index is obtained.

[0058] The outlier correction unit verifies the preliminary correlation index. If the preliminary correlation index of a module exceeds the upper or lower limit of the preset range, it indicates that the module is too well-suited to the learner or too poorly suited, and it is corrected to the range boundary value. The index within the range remains unchanged, forming the final correlation index.

[0059] Step S1244: Establish a mapping relationship between each course module identifier and its corresponding final correlation index to form a correlation index mapping table.

[0060] Construct a table showing the correspondence between module identifiers, module names, and final correlation indices, and clarify the correlation index of each module. For example, M13 - Impressionist Landscape Techniques - High Correlation Index, M12 - Impressionist Oil Painting Basics - High Correlation Index, M06 - Oil Painting Basics - Medium Correlation Index, M09 - Still Life Oil Painting - Low Correlation Index, etc.

[0061] Step S1245: Sort the final correlation indices in the correlation index mapping table in descending order, while retaining the course module identifier and module attribute information corresponding to each correlation index.

[0062] The association index mapping table is sorted in descending order of the final association index. During the sorting process, the complete attribute information of the module is preserved, including knowledge graph, ability training objectives, teaching resource types, and suggested learning scenarios, to ensure that the characteristics of the module can be comprehensively considered during subsequent screening.

[0063] Step S1246: Integrate the sorted association index mapping table with the attribute information of each course module to form a structured document containing course module identifier, module attributes, association index and sorting position. This structured document describes the association strength between learners' learning foundation and each course module.

[0064] The sorted mapping table is integrated with module attribute information to generate a structured document containing fields such as sorting position, module identifier, module name, core knowledge points, key skills development areas, correlation index, and a list of teaching resources. For example, position 1: M13 - Impressionist Landscape Techniques - Core Knowledge: "Outdoor Light and Shadow Capture," "Color Juxtaposition," "Rapid Composition" - Skills Development: "Scene Capture," "Color Application" - Correlation Index: High - Resources: "Outdoor Sketching Videos," "Demonstration Works Collection." This structured document describes the correlation strength, comprehensively reflecting the learner's compatibility with each module.

[0065] Step S130: Based on the association strength description, select an initial course module subset from the preset art course module library, and sort the initial course module subset according to the association strength to form an initial learning module sequence.

[0066] Based on the correlation strength description, a correlation index threshold is set, and modules with correlation indices higher than the threshold are selected as a subset of initial course modules. In this embodiment, five modules are selected: M13 - Impressionist Landscape Techniques, M12 - Impressionist Oil Painting Basics, M11 - Advanced Landscape Oil Painting, M10 - Introduction to Landscape Oil Painting, and M08 - Oil Painting Brushstrokes. These modules are arranged in descending order of correlation index, forming the initial learning module sequence: M13—M12—M11—M10—M08. This initial learning module sequence is sorted according to the learner's directional preferences, knowledge base, and current abilities, prioritizing modules with high suitability.

[0067] Step S140: Collect a set of real-time learning behavior data generated by the learner when learning the first course module in the initial learning module sequence, and input the set of real-time learning behavior data into the deep learning course dynamic association model to generate a module adjustment signal.

[0068] A behavior collection and analysis system is deployed on the learning management platform to capture learners' behavior data in real time during the learning process of the first module M13 - Impressionist Landscape Techniques. The data is then analyzed by the model to generate adjustment signals. The specific process is as follows.

[0069] Step S141: Deploy the behavior data collection unit to continuously capture the learner's operational behavior during the learning of the first course module. The captured operational behavior includes module content browsing behavior, practical task submission behavior, and interactive Q&A participation behavior.

[0070] Behavioral data collection units are deployed on both the web and mobile app versions of the learning management platform. Through front-end event listening, back-end log recording, and API calls, three core operational behaviors are continuously captured: module content browsing behavior (including playback, pause, fast forward, and rewind operations for video tutorials; page turning, pausing, and marking operations for illustrated manuals; and zooming in, zooming out, and viewing details of demonstration works); practical task submission behavior (including uploading periodic sketching assignments, recording the usage trajectory of online drawing tools, and saving records of revised assignments); and interactive Q&A participation behavior (including posting questions in the course forum, replying to others' questions, liking and commenting, and recording posts in the online Q&A room). The collection units capture these behaviors in real time, recording basic information such as the timestamp of the operation, operation type, operation object, and operation duration for each type of behavior.

[0071] Step S142: Extract data from the captured module content browsing behavior, record the duration distribution of learners browsing each module sub-content, the identifiers of repeatedly browsed sub-content and skipped sub-content, and form a subset of browsing behavior data.

[0072] Key information was extracted from the collected module content browsing behavior data: Module M13 - Impressionist Landscape Techniques contains 5 sub-contents (C1 - Outdoor Sketching Tool Preparation, C2 - Light and Shadow Observation Methods, C3 - Quick Composition Techniques, C4 - Color Juxtaposition Practice, C5 - Key Points for Artwork Correction). The time spent by learners on each sub-content was recorded to form a time distribution; the number of times learners browsed each sub-content was counted, and sub-contents with more than a set number of repeated browsings were marked (e.g., C2 - Light and Shadow Observation Methods was repeatedly browsed multiple times); sub-contents that learners fast-forwarded or skipped directly were identified (e.g., C1 - Outdoor Sketching Tool Preparation was directly skipped). The time distribution, repeated browsing markers, and skip markers were categorized and organized by sub-content to form a subset of browsing behavior data.

[0073] Step S143: Extract data from the captured practice task submission behavior, record the number of times learners submit practice tasks, the completion progress percentage of each submission, and the presentation of key elements in the task results, forming a subset of practice behavior data.

[0074] The core data of practical task submission behavior is extracted: Module M13 contains three practical tasks (T1 - Outdoor Light and Shadow Sketch, T2 - Small-Scale Color Juxtaposition Sketch, T3 - Complete Landscape Sketch). It records the number of times learners submit each task (e.g., multiple revised versions were submitted for T2). Based on the task's completion standards, the percentage of completion for each submission is calculated (e.g., the initial submission of T1 was partially completed, and after revisions, it was fully completed). The presentation of key elements in the task results is analyzed (e.g., the density of color juxtaposition, the number of light and shadow layers, and the balance of composition in T2). Feature parameters of these elements are extracted using image recognition technology. The number of submissions, percentage of completion progress, and key element feature parameters are associated with task identifiers to form a subset of practical behavior data.

[0075] Step S144: Extract data from the captured interactive question-and-answer participation behaviors, record the number of times learners ask questions, the knowledge domain corresponding to the question content, the number of times they answer other people's questions, and the recognition of the answer content, forming a subset of interactive behavior data.

[0076] Relevant information was extracted from the interactive Q&A participation data: the total number of questions asked by learners in the course forum and online Q&A room was recorded; each question was categorized by topic to determine its corresponding knowledge domain (e.g., "choosing the angle of light and shadow observation" belongs to the knowledge domain of light and shadow observation, and "controlling the proportion of color juxtaposition" belongs to the knowledge domain of color application); the number of times learners answered other learners' questions was counted, and the number of likes, adoptions, and other recognition received for each answer was recorded. The number of questions, knowledge domain categories, number of answers, and recognition were integrated to form a subset of interactive behavior data.

[0077] Step S145: The browsing behavior data subset, the practice behavior data subset, and the interaction behavior data subset are structurally integrated and unified in data format to form a real-time learning behavior data set.

[0078] The three data subsets were structured and integrated using JSON format, with unified field naming rules and data types: the browsing behavior data subset was organized according to the field structure of "sub-content identifier - duration distribution - number of repeated views - whether skipped"; the practice behavior data subset was organized according to the field structure of "task identifier - number of submissions - completion progress - key element characteristics"; and the interaction behavior data subset was organized according to the field structure of "number of questions asked - knowledge domain - number of answers received - recognition indicators". The integrated data underwent format validation to ensure that no fields were missing or formatted incorrectly, forming a complete real-time learning behavior data set.

[0079] Step S146: Input the real-time learning behavior data set into the behavior analysis layer of the deep learning course dynamic association model for feature extraction and generate behavior feature vectors.

[0080] Behavioral feature vectors are generated through a multi-dimensional feature extraction mechanism in the behavioral analysis layer. The specific process is as follows.

[0081] Step S1461: Receive the browsing behavior data subset through the browsing feature extraction unit in the behavior analysis layer, perform statistical analysis on the browsing duration distribution, calculate the browsing duration ratio and average browsing duration of each module's sub-content, and convert them into duration feature components.

[0082] After receiving a subset of browsing behavior data, the browsing feature extraction unit calculates the proportion of browsing time for each sub-content to the total browsing time, as well as the average browsing time for all sub-contents. These proportions and duration values ​​are then standardized and converted into a duration feature component, which reflects the learner's level of attention to different knowledge content.

[0083] Step S1462: Count the frequency of repeatedly viewed sub-content identifiers, determine the sub-content type with the highest frequency of repeated viewing, generate repeated viewing feature components, and classify the skipped sub-content identifiers by type to generate skipped content feature components.

[0084] The frequency of repeatedly viewed sub-content identifiers is counted to determine the sub-content type with the highest frequency (such as light and shadow observation content). This type and its corresponding frequency are converted into repeated viewing feature components. Skipped sub-content identifiers are classified according to knowledge type (such as tool preparation), and the classification results are converted into skipped content feature components. These two components reflect the learner's knowledge weaknesses and existing knowledge, respectively.

[0085] Step S1463: Receive the subset of practice behavior data through the practice feature extraction unit in the behavior analysis layer, count the number of times the practice task is submitted, calculate the ratio of the number of submissions to a preset submission threshold, and generate a submission frequency feature component.

[0086] After receiving a subset of practice behavior data, the practice feature extraction unit counts the number of submissions for each practice task, compares the number of submissions for each task with the preset submission threshold for that task, obtains a ratio, and integrates the ratios of all tasks into a submission frequency feature component, reflecting the learner's practice participation and perceived difficulty in completing the task.

[0087] Step S1464: Perform trend analysis on the completion progress percentage of each submitted task, extract the slope parameter of the progress change trend, generate progress trend feature components, and at the same time, perform feature encoding on the presentation of key elements in the task results to generate result element feature components.

[0088] The multiple submission records for each practical task are arranged in chronological order. The changing trend of the completion progress percentage is analyzed, the slope parameter of the trend line is extracted, and it is converted into a progress trend feature component to reflect the learner's progress speed. The presentation of key elements in the task results (such as color juxtaposition and light and shadow levels) is converted into numerical result element feature components through feature encoding algorithm to reflect the degree of mastery of practical skills.

[0089] Step S1465: Receive the subset of interactive behavior data through the interactive feature extraction unit in the behavior analysis layer, count the number of questions asked, and generate question frequency and domain distribution feature components by combining the knowledge domain corresponding to the question content.

[0090] After receiving a subset of interactive behavior data, the interactive feature extraction unit counts the total number of questions, classifies and counts the question content according to knowledge domain, calculates the proportion of questions in each domain, and integrates the total number of questions and the domain proportion into question frequency and domain distribution feature components, reflecting the distribution of learners' knowledge confusion.

[0091] Step S1466: Statistically count the number of times you answer other people's questions, and calculate the answer acceptance coefficient based on the acceptance of the answer content to generate answer quality feature components.

[0092] The total number of times a learner answers others' questions is counted. The number of likes and adoptions for each answer is compared with the total number of answers to calculate the answer recognition coefficient. The coefficient is combined with the number of answers to generate answer quality feature components, which reflect the learner's knowledge output ability and the solidity of their mastery.

[0093] Step S1467: Standardize and arrange the duration feature components, repeated browsing feature components, skipped content feature components, submission frequency feature components, progress trend feature components, result element feature components, question frequency and domain distribution feature components, and answer quality feature components in a preset order to form a behavior feature vector with fixed dimensions.

[0094] The eight feature components are arranged in a preset order, and the differences in the dimensions of different components are eliminated through standardization, so that each component is in the same data range, and finally a fixed-dimensional behavioral feature vector is formed. This behavioral feature vector fully represents the learner's real-time learning behavior state.

[0095] Step S147: Compare the behavioral feature vector with the preset standard behavioral feature library in the deep learning course dynamic association model to identify the deviation type between the learner's current learning state and the standard learning state.

[0096] The system retrieves a pre-defined standard behavioral feature library from the deep learning course dynamic association model. This library contains feature vectors for various standard learning states (such as "highly efficient mastery state," "moderate adaptation state," and "weak foundation state") for module M13 - Impressionist Landscape Techniques. The generated behavioral feature vectors are compared with the feature vectors in the standard behavioral feature library to calculate the distance between them. Based on the state corresponding to the standard feature vector with the smallest distance, and combined with the specific components of the behavioral feature vector, the deviation type is identified. For example, if the learner's repeated browsing feature component shows frequent repetition of light and shadow observation content, and the achievement element feature component shows weak light and shadow performance, and the progress trend feature component shows slow progress, then the deviation type is "insufficient ability in light and shadow observation and performance."

[0097] Step S148: According to the deviation type, call the corresponding adjustment rule from the adjustment strategy library of the deep learning course dynamic association model, and generate a module adjustment signal containing module replacement suggestions or order adjustment suggestions based on the adjustment rules. The module replacement suggestion specifies the course module identifier to be replaced and the course module identifier after replacement, and the order adjustment suggestion specifies the arrangement order of the adjusted course modules.

[0098] The module adjustment signal is generated through the following process:

[0099] Step S1481: The adjustment strategy library of the deep learning course dynamic association model stores adjustment rules corresponding to various deviation types. Each deviation type is associated with at least one adjustment rule. The adjustment rule includes the rule triggering condition, adjustment operation type, and operation parameters.

[0100] The adjustment strategy library has preset adjustment rules for common deviation types in painting courses. For example, the deviation type of "insufficient ability to observe and express light and shadow" is associated with two adjustment rules: Rule 1 is a sequential adjustment rule (trigger condition: minor deviation; operation type: sequential adjustment; operation parameter: advance the basic light and shadow module); Rule 2 is a module replacement rule (trigger condition: severe deviation; operation type: module replacement; operation parameter: identifier of the module to be replaced, filter conditions for the replacement module identifier).

[0101] Step S1482: Based on the identified deviation type between the learner's current learning state and the standard learning state, perform rule matching in the adjustment strategy library to find an adjustment rule that perfectly matches the deviation type.

[0102] Based on the identified "insufficient ability to observe and express light and shadow" deviation type, the corresponding rule 1 and rule 2 are matched in the adjustment strategy library. Then, the degree of deviation (such as the speed of progress and the degree of weakness) is analyzed according to the behavioral feature vector to determine the trigger rule 2 (the degree of deviation is more serious).

[0103] Step S1483: Extract the adjustment operation type from the matched adjustment rules and determine whether the adjustment operation type is module replacement or sequence adjustment.

[0104] The adjustment operation type in rule 2 is module replacement.

[0105] Step S1484: If the adjustment operation type is module replacement, then extract the operation parameters in the adjustment rules. The operation parameters include the filter conditions for the course module identifier to be replaced and the filter conditions for the course module identifier after replacement.

[0106] Extract the operation parameters for rule 2: The filter condition for the course module identifier to be replaced is "higher difficulty modules following the current learning module"; the filter condition for the course module identifier after replacement is "modules containing basic training in light and shadow".

[0107] Step S1485: Determine the course module identifier to be replaced from the initial learning module sequence according to the filtering conditions for the course module identifier to be replaced; determine the course module identifier to be replaced from the preset art course module library according to the filtering conditions for the course module identifier to be replaced.

[0108] Based on the selection criteria, the module to be replaced from the initial learning module sequence (M13—M12—M11—M10—M08) is M12-Impressionist Oil Painting Basics (which belongs to the later advanced modules); the module to be replaced from the course module library is M11-Landscape Oil Painting Advanced (which includes basic training content on light and shadow).

[0109] Step S1486: Integrate the course module identifier to be replaced, the course module identifier after replacement, and the adjustment operation type to form a module replacement suggestion.

[0110] The module identifier M12 to be replaced, the module identifier M11 to be replaced, and the operation type "module replacement" are combined to form the module replacement suggestion: "Replace M12 in the initial sequence with M11".

[0111] Step S1487: If the adjustment operation type is sequential adjustment, then extract the operation parameters in the adjustment rules. The operation parameters include the priority sorting basis for the sequential adjustment of modules.

[0112] If rule 1 is triggered, the priority sorting criterion for extracting operation parameters is "basic capability modules take precedence over advanced modules".

[0113] Step S1488: Based on the priority sorting criteria, reorder the course modules in the initial learning module sequence to determine the adjusted course module arrangement order.

[0114] If rule 1 is triggered, the initial sequence will be adjusted to "M11—M13—M12—M10—M08" based on the principle of "priority of basic capability modules".

[0115] Step S1489: Add adjustment signal identifiers and generation time information to the module replacement suggestion or sequence adjustment suggestion, and encapsulate them into a module adjustment signal that conforms to the model output format.

[0116] Add an adjustment signal identifier (unique code) and generate a timestamp to the module replacement suggestion, and encapsulate it in the model's preset JSON format to form a module adjustment signal.

[0117] Step S150: Based on the module adjustment signal, the initial learning module sequence is replaced or the order is adjusted to obtain a personalized learning path that adapts to the learner's real-time learning status. Based on the personalized learning path, a personalized learning path document containing the course module learning order and module connection guidance is generated.

[0118] The initial sequence is adjusted based on the module adjustment signal, and a path document is generated. The specific process is as follows.

[0119] Step S151: Parse the adjustment type identifier in the module adjustment signal to determine whether the adjustment type is module replacement or sequential adjustment. If the adjustment type is module replacement, extract the identifier of the course module to be replaced and the identifier of the course module after replacement from the module adjustment signal.

[0120] The module adjustment signal is analyzed to determine that the adjustment type is module replacement, and the identifier of the module to be replaced, M12, and the identifier of the replacement module, M11, are extracted.

[0121] Step S152: Locate the course module that matches the identifier of the course module to be replaced from the initial learning module sequence, remove the course module from the initial learning module sequence, and insert the replacement course module into the position of the original module to be replaced to form a temporary module sequence.

[0122] Remove M12 from the initial sequence and insert M11 at the original position of M12 to form a temporary module sequence: M13—M11—M12—M10—M08.

[0123] Step S153: If the adjustment type is sequential adjustment, extract the adjusted course module arrangement order from the module adjustment signal, and reorder the course modules in the initial learning module sequence according to this arrangement order to form a temporary module sequence.

[0124] If the adjustment type is sequential adjustment, the temporary sequence is generated by directly reordering the order of the adjustment signals.

[0125] Step S154: Input the temporary module sequence into the path verification layer of the deep learning course dynamic association model. The path verification layer loads the course module association rule library, which contains the pre-learning relationships and connection logic between each course module.

[0126] The validity of the temporary sequence is verified through a path verification layer. The specific process is as follows:

[0127] Step S1541: Start the rule loading unit through the path verification layer, and retrieve the course module association rule library from the built-in rule storage area. Each rule in the course module association rule library includes the identifier of the preceding course module, the identifier of the following course module, and the association strength parameter.

[0128] The rule loading unit retrieves the course module association rule library, which contains rules such as "M11-Advanced Landscape Oil Painting is the prerequisite module for M10-Introduction to Landscape Oil Painting" and "M12-Basic Impressionist Oil Painting is the prerequisite module for M11-Advanced Landscape Oil Painting". Each rule is marked with an association strength parameter (such as strong association, medium association).

[0129] Step S1542: Classify and organize the rules in the course module association rule base, group them according to the identifier of the preceding course module, and form a list of subsequent course modules corresponding to each preceding course module. The list of subsequent course modules contains all subsequent course modules that are based on the preceding course module.

[0130] Grouped by the front module identifier, forming a list such as "M10's rear module: M11", "M11's rear module: M12", "M12's rear module: M13", etc.

[0131] Step S1543: Number each course module in the temporary module sequence in sequence to form a module number sequence, and select each module except the first module from the module number sequence as the current module to be verified.

[0132] The temporary module sequence numbers are 1-M13, 2-M11, 3-M12, 4-M10, and 5-M08. 2-M11, 3-M12, 4-M10, and 5-M08 are selected as the current modules to be verified.

[0133] Step S1544: For each current module to be verified, extract its module identifier and search for the corresponding preceding course module identifier in the course module association rule base.

[0134] Extract the identifiers of 2-M11 and find that its front-end module identifier is M10; extract the identifiers of 3-M12 and find that its front-end module identifier is M11; extract the identifiers of 4-M10 and find that its front-end module identifier is none; extract the identifiers of 5-M08 and find that its front-end module identifier is none.

[0135] Step S1545: Determine the position of the prerequisite course module identifier in the temporary module sequence, and check whether the prerequisite course module is located before the current module to be verified. If the prerequisite course module is located before the current module to be verified, it is determined that the position of the current module to be verified conforms to the prerequisite learning relationship.

[0136] Checking 2-M11, the predecessor module M10 is located at position 4 in the temporary sequence, after position 2, so it is determined that the predecessor relationship does not meet the requirements; checking 3-M12, the predecessor module M11 is located at position 2, before position 3, so it is determined that the predecessor relationship meets the requirements; 4-M10 and 5-M08 have no predecessor modules, so they are determined to meet the requirements.

[0137] Step S1546: If the prerequisite course module is not located before the current module to be verified, extract the association strength parameter between the prerequisite course module and the current module to be verified from the course module association rule base. The association strength parameter is used to represent the degree of necessity of the prerequisite learning relationship.

[0138] The correlation strength parameter between M10 and M11 is extracted as "strong correlation", indicating that M10 is a necessary prerequisite module for M11.

[0139] Step S1547: Simultaneously, the connection logic in the course module association rule base is extracted through the connection logic analysis unit of the path verification layer. The connection logic includes the smoothness requirements of knowledge transition between modules and the progressive requirements of ability development.

[0140] The connection logic analysis unit extracts the connection logic: the knowledge transition requires that "basic composition knowledge should precede the learning of light and shadow expression knowledge"; the ability development requires that "composition ability development should precede color application ability development".

[0141] Step S1548: Analyze the knowledge transition between the current module to be verified and the previous module based on the connection logic to see if it is smooth and whether the ability development is progressive. If the knowledge transition is not smooth or the ability development is not progressive, it is marked as a connection anomaly.

[0142] Analysis of 2-M11 (the previous module was 1-M13): M13 is Impressionist landscape techniques (advanced), and M11 is advanced landscape oil painting (intermediate difficulty). The transition of knowledge from advanced to intermediate difficulty is not smooth, and the development of abilities from comprehensive performance to specialized training is not progressive. This is marked as an abnormal connection.

[0143] Step S1549: For modules with abnormal connection marks and those that do not conform to the prior learning relationship, adjust the temporary module sequence according to the association strength parameters and connection logic in the course module association rule base.

[0144] Since the preceding module M10 of M11 is located after it and the two are strongly related, and there is an abnormal connection between M13 and M11 with an uneven knowledge transition, the temporary module sequence is adjusted: M10 is moved before M11, and M13 is adjusted after M12, forming the adjusted temporary sequence: M10—M11—M12—M13—M08.

[0145] Step S155: Through the path verification layer, check whether the arrangement of each course module in the temporary module sequence conforms to the prior learning relationship in the course module association rule base. If the check passes, the temporary module sequence is determined as a personalized learning path that adapts to the learner's real-time learning status. If the check fails, the temporary module sequence is fine-tuned according to the connection logic in the course module association rule base, adjusting the position of modules that do not conform to the prior relationship until all association rules are met, and then it is determined as a personalized learning path.

[0146] The adjusted temporary sequence M10—M11—M12—M13—M08 underwent a further check of its prerequisite relationships: M11's prerequisite module M10 precedes it, M12's prerequisite module M11 precedes it, and M13's prerequisite module M12 precedes it, all conforming to the prerequisite learning relationship. Simultaneously, the knowledge transition and ability development progression were analyzed: M10 (composition and tone)—M11 (light and shadow and atmosphere) meets the knowledge transition requirement of "basic composition precedes light and shadow expression"; M11—M12 (color juxtaposition) meets the ability development requirement of "composition ability precedes color application"; and M12—M13 (outdoor sketching) achieves a progression from basic techniques to practical application. The knowledge transition is smooth, and the ability development is progressive, with no abnormal connections. Therefore, this sequence was determined as the personalized learning path.

[0147] Step S156: Generate a personalized learning path document containing the learning order of course modules and module connection guidelines based on the personalized learning path. The specific process is as follows.

[0148] Step S1561: Extract all course module identifiers in the personalized learning path, arrange them in the learning order to form a course module learning order list, and add the corresponding module name and a brief introduction of the core content of the module to each course module identifier in the course module learning order list.

[0149] Extract the module identifiers and related information from the personalized learning path to form a learning sequence list: 1. M10 - Introduction to Landscape Oil Painting (Core content: principles of landscape composition, basic color matching, simple landscape sketching steps); 2. M11 - Advanced Landscape Oil Painting (Core content: expression of light and shadow layers, creation of spatial atmosphere, practice of sketching medium and complex landscapes); 3. M12 - Impressionist Oil Painting Basics (Core content: Impressionist color juxtaposition techniques, quick color mixing techniques, analysis of Impressionist style characteristics); 4. M13 - Impressionist Landscape Techniques (Core content: outdoor light and shadow observation methods, quick on-site composition, practical Impressionist landscape sketching); 5. M08 - Oil Painting Brushstrokes (Core content: application of wet and dry brushstrokes, brushstrokes and emotional expression, comparison of brushstroke characteristics of different styles).

[0150] Step S1562: For two adjacent course modules, analyze the connection points between the core knowledge of the previous course module and the core knowledge of the next course module, determine the key content for knowledge connection, and generate knowledge connection guidelines.

[0151] Analyze the knowledge connections between adjacent modules: The connection between M10 and M11 is "composition principles - light and shadow layout", and the knowledge connection guide is "based on the composition framework mastered in M10, in M11, focus on learning how to strengthen the composition focus through light and shadow distribution, and understand the relationship between light and shadow direction and compositional balance"; The connection between M11 and M12 is "light and shadow layers - color juxtaposition", and the knowledge connection guide is "transfer the light and shadow contrast logic in M11 to color juxtaposition, express the light and shadow effect through the juxtaposition of warm and cool colors, and achieve the integration of light and shadow and color"; The connection between M12 and M13 is "Color Juxtaposition Techniques - Outdoor Practical Application," and the knowledge connection guide is "Based on the color juxtaposition techniques mastered in M12, combine them with the outdoor light and shadow observation methods in M13 to flexibly adjust the proportion and brightness of the juxtaposed colors to adapt to changes in outdoor light." The connection between M13 and M08 is "Outdoor Sketching Brushstrokes - Specialized Brushstroke Training," and the knowledge connection guide is "Summarize the commonly used brushstroke types in outdoor sketching in M13, and conduct specialized training in M08 for weak brushstrokes (such as pointillist brushstrokes) to improve brushstroke expressiveness."

[0152] Step S1563: Analyze the progressive relationship between the competency development objectives of the previous course module and the competency development objectives of the next course module, determine the focus of competency transition, and generate competency transition guidelines.

[0153] Analyze the progressive relationship of abilities between adjacent modules: M10 (composition ability, basic color tone control ability) — M11 (light and shadow shaping ability, spatial expression ability), the ability transition guide is "based on the composition ability formed in M10, focus on improving the light and shadow shaping ability, and transform planar composition into three-dimensional spatial expression through multiple sets of light and shadow contrast exercises"; M11 (light and shadow shaping ability, spatial expression ability) — M12 (color application ability, style imitation ability), the ability transition guide is "based on the spatial expression foundation of M11, focus on the flexibility of color application, and master the combination of color juxtaposition and style expression techniques through imitating Impressionist works"; The transition from M12 (color application ability, style imitation ability) to M13 (scene capture ability, practical application ability) is guided by the principle of "transferring the color and style abilities of M12 to outdoor scenes, focusing on the ability to quickly capture scenes and make real-time adjustments, and improving the practical application level of techniques through multiple outdoor sketching practices." The transition from M13 (scene capture ability, practical application ability) to M08 (brushstroke control ability, personalized expression ability) is guided by the principle of "based on the brushstroke problems exposed in M13 practice, focusing on the accuracy and diversity of brushstroke control, and combining personal style preferences, forming a distinctive brushstroke expression through specialized training."

[0154] Step S1564: Combine the knowledge connection guide with the ability transition guide to form a module connection guide for adjacent modules. Each combination of adjacent modules corresponds to one module connection guide.

[0155] The knowledge and skills guidelines are combined to form module transition guidelines: M10-M11 transition guidelines are "Knowledge transition: Combining composition framework with light and shadow layout; Skill transition: Improving composition skills from light and shadow shaping skills, enhancing spatial expression through light and shadow"; M11-M12 transition guidelines are "Knowledge transition: Juxtaposing and integrating light and shadow layers with color; Skill transition: Expanding spatial expression skills from color application skills, combining color juxtaposition to imitate Impressionist style"; M12-M13 transition guidelines are "Knowledge transition: Combining color juxtaposition techniques with outdoor light and shadow observation; Skill transition: Transforming color and style skills into scene capture and practical application skills, improving practical adaptability"; M13-M08 transition guidelines are "Knowledge transition: Corresponding outdoor sketching brushstrokes with specialized brushstroke types; Skill transition: Deepening practical application skills from brushstroke control and personalized expression skills, forming a personal brushstroke style".

[0156] Step S1565: Integrate the course module learning order list and all module connection guides according to the preset document structure, which includes a path overview, module details and connection guide.

[0157] Step S1566: In the path overview section, describe the overall learning objectives and expected learning outcomes of the personalized learning path. In the module details section, present the identifier, name, core content summary and suggested learning period of each course module in sequence. In the connection guide section, present the corresponding module connection guide in the order of adjacent modules to form a personalized learning path document.

[0158] The content is integrated according to a pre-set structure: The path overview section explains the overall goal of the personalized learning path ("Through module learning from basic to advanced, from theory to practice, master the core abilities of Impressionist landscape oil painting creation and form a personalized creative style"), expected learning outcomes ("Independently complete 2-3 outdoor landscape oil paintings in the Impressionist style, possessing solid abilities in composition, light and shadow, color, and brushstroke application"), and suggested total learning period; The module details section presents the identifier, name, core content summary, teaching resource list (e.g., M10 includes "composition principle video tutorial, color matching illustrated manual, 10 demonstration works"), and suggested learning period for each module in sequence; The bridging guidance section presents the corresponding module bridging guidance in the order of adjacent modules, including key points of knowledge bridging and methods of ability transition.

[0159] The overall learning objectives and expected learning outcomes of the personalized learning path are explained in the path overview section. The module details section presents the identifier, name, core content summary and suggested learning period of each course module in sequence. The connection guide section presents the corresponding module connection guide in the order of adjacent modules, forming a personalized learning path document.

[0160] The generated personalized learning path document is available in both PDF and online document formats. The online document allows learners to mark the completion status of modules and add learning notes. Each module name in the document has a hyperlink, which can be clicked to directly jump to the corresponding module page on the learning management platform to access teaching resources such as videos, images, and sample works. The bridging guide section includes an interactive Q&A entry point, where learners can ask questions about problems encountered during the bridging process, and teachers or the system will provide intelligent answers.

[0161] Figure 2 The illustration shows exemplary hardware and software components of a deep learning-based personalized learning path system 100 for art courses, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the deep learning-based personalized learning path system 100 for art courses and to perform the functions in this application.

[0162] The deep learning-based personalized learning path system for art courses 100 can be a general-purpose server or a special-purpose server; both can be used to implement the deep learning-based personalized learning path method for art courses described in this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0163] For example, a deep learning-based personalized learning path system 100 for art courses may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the deep learning-based personalized learning path system 100 for art courses may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The deep learning-based personalized learning path system 100 for art courses also includes an I / O interface 150 between the computer and other input / output devices.

[0164] For ease of explanation, only one processor is described in the deep learning-based personalized learning path system for art courses 100. However, it should be noted that the deep learning-based personalized learning path system for art courses 100 in this application may also include multiple processors. Therefore, the steps executed by one processor as described in this application may also be executed jointly or individually by multiple processors. For example, if the processor of the deep learning-based personalized learning path system for art courses 100 executes steps A and B, it should be understood that steps A and B may also be executed jointly by two different processors or individually by one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor jointly execute steps A and B.

[0165] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-mentioned method for personalized learning paths for art courses based on deep learning is implemented.

[0166] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for personalized learning paths in art courses based on deep learning, characterized in that, The method includes: The system acquires a set of basic information about the learner's art courses and a pre-set library of art course modules. The set of basic information about the learner's art courses includes the art course knowledge items that the learner has already encountered, the learner's practical performance records in art courses, and the learner's expected direction of art course learning. The pre-set library of art course modules includes multiple art course modules with knowledge-related relationships. The pre-trained deep learning course dynamic association model is invoked, and the learner's set of basic learning information for art courses is input into the deep learning course dynamic association model to generate a description of the association strength between the learner's learning foundation and each course module in the preset art course module library. Based on the association strength description, an initial course module subset is selected from the preset art course module library, and the initial course module subset is sorted according to the association strength to form an initial learning module sequence; Collect a set of real-time learning behavior data generated by learners when learning the first course module in the initial learning module sequence, and input the set of real-time learning behavior data into the deep learning course dynamic association model to generate a module adjustment signal; Based on the module adjustment signal, the initial learning module sequence is replaced or its order is adjusted to obtain a personalized learning path that adapts to the learner's real-time learning status. Based on the personalized learning path, a personalized learning path document containing the learning order of course modules and module connection guidance is generated.

2. The method for personalized learning paths for art courses based on deep learning according to claim 1, characterized in that, The process involves calling a pre-trained deep learning course dynamic association model, inputting the learner's basic information on art courses into the model, and generating a description of the association strength between the learner's basic information and each course module in the preset art course module library, including: The input layer processing unit of the deep learning course dynamic association model is activated, and the knowledge items, practice performance records and learning directions in the learner's art course learning basic information set are converted into structured feature vectors respectively. The dimension of each structured feature vector is consistent with the preset dimension of the input layer of the deep learning course dynamic association model. The transformed multiple structured feature vectors are input into the association calculation layer of the deep learning course dynamic association model, which includes a knowledge association sublayer, an ability association sublayer, and a direction association sublayer. In the knowledge association sub-layer, the knowledge-related feature components in each structured feature vector are extracted and similarity analysis is performed with the knowledge coverage features of each course module in the preset art course module library to generate knowledge dimension association values; In the ability association sublayer, feature components related to practical performance are extracted from each structured feature vector and matched with the ability cultivation features of each course module in the preset art course module library to generate ability dimension association values. In the direction association sublayer, feature components related to the learning direction are extracted from each structured feature vector, and a fit analysis is performed with the target orientation features of each course module in the preset art course module library to generate a direction dimension association value. The knowledge dimension association value, ability dimension association value, and direction dimension association value are input into the weight fusion layer of the deep learning course dynamic association model. The weight fusion layer performs weighted summation of the association values ​​of the three dimensions according to the weight coefficients of each dimension determined during the model training phase. The output layer of the deep learning course dynamic association model converts the weighted summation result into an association index with a uniform range. Each course module corresponds to an association index, and the association indices of all course modules together constitute a description of the association strength between the learner's learning foundation and each course module.

3. The method for personalized learning paths for art courses based on deep learning according to claim 2, characterized in that, The process involves inputting the transformed multiple structured feature vectors into the association computation layer of the deep learning course dynamic association model. This association computation layer includes a knowledge association sublayer, a capability association sublayer, and a direction association sublayer, comprising: After receiving the structured feature vector containing the feature components of knowledge items through the knowledge association sublayer, the built-in knowledge graph matching unit is invoked. The knowledge graph matching unit loads the art course knowledge graph, which contains the hierarchical and association relationships between each knowledge item. The knowledge entry feature components in the structured feature vector are mapped to the knowledge coverage features of each course module in the preset art course module library through the knowledge graph matching unit, thereby determining the node positions and path distances between the two in the art course knowledge graph. The knowledge association coefficient is calculated based on the path distance between nodes. The knowledge association coefficient is used as the core calculation basis for the knowledge dimension association value. The shorter the path distance, the larger the knowledge association coefficient. After receiving the structured feature vector containing practical performance feature components through the capability association sublayer, the capability dimension analysis unit is activated. The capability dimension analysis unit contains multiple capability assessment dimensions, and each capability assessment dimension corresponds to a core capability in art course learning. The practical performance feature components in the structured feature vector are decomposed into each ability assessment dimension, the performance score under each ability assessment dimension is calculated, and the training intensity value of the ability training features of each course module under the corresponding ability assessment dimension is extracted. By comparing the degree of matching between performance scores and training intensity values, a capability matching coefficient is generated for each capability assessment dimension. The mean of the capability matching coefficients of all capability assessment dimensions is calculated to obtain the capability dimension correlation value. After receiving the structured feature vector containing the learning direction feature components through the direction association sublayer, the direction matching analysis unit is activated. The direction matching analysis unit loads a preset learning direction classification system, which includes multiple subdivided learning direction categories and the core learning objective corresponding to each category. The learning direction feature components in the structured feature vector are classified into the corresponding categories in the learning direction classification system to determine the learner's core learning objectives. At the same time, the correlation between the goal-oriented features of each course module and the core learning objectives is analyzed. A directional fit coefficient is generated based on the degree of correlation. The directional fit coefficient is used as a component of the directional dimension correlation value. The higher the degree of correlation, the larger the directional fit coefficient.

4. The method for personalized learning paths for art courses based on deep learning according to claim 1, characterized in that, The data collection includes real-time learning behavior data generated by learners when learning the first course module in the initial learning module sequence. This real-time learning behavior data collection is then input into the deep learning course dynamic association model to generate module adjustment signals, including: Deploy behavioral data collection units to continuously capture learners' operational behaviors during the learning of the first course module. The captured operational behaviors include module content browsing behavior, practical task submission behavior, and interactive Q&A participation behavior. Data is extracted from the captured module content browsing behavior, and the duration distribution of learners browsing each module sub-content, the identifiers of repeatedly browsed sub-content, and the identifiers of skipped sub-content are recorded to form a subset of browsing behavior data; Data is extracted from the captured practical task submission behavior, and the number of times learners submit practical tasks, the completion progress percentage of each submission, and the presentation of key elements in the task results are recorded to form a subset of practical behavior data. Data is extracted from the captured interactive Q&A participation behaviors, recording the number of times learners ask questions, the knowledge domain corresponding to the questions, the number of times they answer other people's questions, and the recognition of their answers, forming a subset of interactive behavior data; The browsing behavior data subset, practice behavior data subset, and interaction behavior data subset are structured and integrated, and a unified data format is used to form a real-time learning behavior data set. The real-time learning behavior data set is input into the behavior analysis layer of the deep learning course dynamic association model for feature extraction, generating a behavior feature vector. The behavioral feature vector is compared with the preset standard behavioral feature library in the deep learning course dynamic association model to identify the type of deviation between the learner's current learning state and the standard learning state. According to the deviation type, the corresponding adjustment rule is called from the adjustment strategy library of the deep learning course dynamic association model. Based on the adjustment rule, a module adjustment signal containing module replacement suggestions or order adjustment suggestions is generated. The module replacement suggestion specifies the course module identifier to be replaced and the course module identifier after replacement. The order adjustment suggestion specifies the arrangement order of the adjusted course modules.

5. The method for personalized learning paths for art courses based on deep learning according to claim 4, characterized in that, The real-time learning behavior data set is input into the behavior analysis layer of the deep learning course dynamic association model for feature extraction, generating a behavior feature vector, including: The browsing behavior data subset is received by the browsing feature extraction unit in the behavior analysis layer, the browsing time distribution is statistically analyzed, the browsing time ratio and average browsing time of each module sub-content are calculated, and converted into time feature components. Frequency statistics are performed on the sub-content identifiers that are repeatedly viewed to determine the sub-content type with the highest frequency of repeated viewing and generate repeated viewing feature components. At the same time, the sub-content identifiers that are skipped are classified by type and skipped content feature components are generated. The practice behavior data subset is received by the practice feature extraction unit in the behavior analysis layer, the number of times the practice task is submitted is counted, the ratio of the number of submissions to the preset submission threshold is calculated, and the submission frequency feature component is generated. The progress percentage of each submitted task is analyzed for trend, the slope parameter of the progress change trend is extracted, and the progress trend feature component is generated. At the same time, the presentation of key elements in the task results is feature-coded to generate result element feature components. The interaction behavior data subset is received by the interaction feature extraction unit in the behavior analysis layer, the number of questions asked is counted, and the frequency and domain distribution feature components of the questions are generated by combining the knowledge domain corresponding to the question content. The number of times someone answers another person's question is counted, and the degree of acceptance of the answer is combined to calculate the answer acceptance coefficient and generate answer quality feature components. The duration feature component, repeated browsing feature component, skipped content feature component, submission frequency feature component, progress trend feature component, result element feature component, question frequency and domain distribution feature component, and answer quality feature component are standardized and arranged in a preset order to form a behavioral feature vector with fixed dimensions.

6. The method for personalized learning paths for art courses based on deep learning according to claim 4, characterized in that, The step of calling the corresponding adjustment rule from the adjustment strategy library of the deep learning course dynamic association model according to the deviation type, and generating a module adjustment signal containing module replacement suggestions or sequence adjustment suggestions based on the adjustment rule, includes: The adjustment strategy library of the deep learning course dynamic association model stores adjustment rules corresponding to various deviation types. Each deviation type is associated with at least one adjustment rule. The adjustment rule includes the rule triggering condition, adjustment operation type, and operation parameters. Based on the identified deviation type between the learner's current learning state and the standard learning state, rule matching is performed in the adjustment strategy library to find an adjustment rule that perfectly matches the deviation type. Extract the adjustment operation type from the matched adjustment rules and determine whether the adjustment operation type is module replacement or sequence adjustment; If the adjustment operation type is module replacement, then extract the operation parameters from the adjustment rules. The operation parameters include the filter conditions for the course module identifier to be replaced and the filter conditions for the course module identifier after replacement. Based on the selection criteria for the course module identifier to be replaced, the course module identifier to be replaced is determined from the initial learning module sequence; based on the selection criteria for the course module identifier to be replaced, the course module identifier to be replaced is determined from the preset art course module library. The module identifier to be replaced, the module identifier after replacement, and the adjustment operation type are integrated to form a module replacement suggestion; If the adjustment operation type is sequential adjustment, then the operation parameters in the adjustment rules are extracted. The operation parameters include the priority sorting basis for the sequential adjustment of modules. Based on priority ranking criteria, the course modules in the initial learning module sequence are reordered to determine the adjusted course module arrangement order; The revised course module order and adjustment operation types are integrated to form a suggested order adjustment; Add adjustment signal identifiers and generation time information to module replacement suggestions or sequence adjustment suggestions, and encapsulate them into module adjustment signals that conform to the model output format.

7. The method for personalized learning paths for art courses based on deep learning according to claim 1, characterized in that, The step of replacing or adjusting the order of the initial learning module sequence according to the module adjustment signal to obtain a personalized learning path adapted to the learner's real-time learning state includes: The adjustment type identifier in the module adjustment signal is parsed to determine whether the adjustment type is module replacement or sequential adjustment. If the adjustment type is module replacement, the identifier of the course module to be replaced and the identifier of the course module after replacement are extracted from the module adjustment signal. Locate the course module that matches the identifier of the course module to be replaced from the initial learning module sequence, remove the course module from the initial learning module sequence, and insert the replacement course module into the position of the original module to be replaced to form a temporary module sequence; If the adjustment type is sequential adjustment, the adjusted course module arrangement order is extracted from the module adjustment signal, and the course modules in the initial learning module sequence are reordered according to this arrangement order to form a temporary module sequence. The temporary module sequence is input into the path verification layer of the deep learning course dynamic association model. The path verification layer loads the course module association rule library, which contains the pre-learning relationships and connection logic between each course module. The path verification layer checks whether the arrangement of each course module in the temporary module sequence conforms to the prior learning relationships in the course module association rule base. If the check passes, the temporary module sequence is determined as a personalized learning path that adapts to the learner's real-time learning status. If the check fails, the temporary module sequence is fine-tuned according to the connection logic in the course module association rule base, adjusting the positions of modules that do not conform to the prior relationships until all association rules are met, and then it is determined as a personalized learning path.

8. The method for personalized learning paths for art courses based on deep learning according to claim 7, characterized in that, The step of inputting the temporary module sequence into the path verification layer of the deep learning course dynamic association model, wherein the path verification layer loads the course module association rule base, includes: The rule loading unit is started through the path verification layer, and the course module association rule library is retrieved from the built-in rule storage area. Each rule in the course module association rule library includes the identifier of the preceding course module, the identifier of the following course module, and the association strength parameter. The rules in the course module association rule base are classified and organized, and grouped according to the identifier of the preceding course module to form a list of subsequent course modules corresponding to each preceding course module. The list of subsequent course modules contains all subsequent course modules that are based on the preceding course module. Each course module in the temporary module sequence is numbered sequentially to form a module number sequence. Each module except the first module is selected from the module number sequence as the current module to be verified. For each module to be verified, extract its module identifier and search for the corresponding preceding course module identifier in the course module association rule base. Determine the position of the prerequisite course module identifier in the temporary module sequence, check whether the prerequisite course module is located before the current module to be verified, and if the prerequisite course module is located before the current module to be verified, determine that the position of the current module to be verified conforms to the prerequisite learning relationship; If the prerequisite course module is not located before the current module to be verified, the association strength parameter between the prerequisite course module and the current module to be verified is extracted from the course module association rule base. The association strength parameter is used to represent the degree of necessity of the prerequisite learning relationship. Meanwhile, the connection logic in the course module association rule base is extracted through the connection logic analysis unit of the path verification layer. The connection logic includes the smoothness requirements of knowledge transition between modules and the progressive requirements of ability development. Based on the connection logic analysis, it is determined whether the knowledge transition between the current module to be verified and the previous module is smooth and whether the ability development is progressive. If the knowledge transition is not smooth or the ability development is not progressive, it is marked as a connection anomaly. For modules with abnormal connection marks or that do not conform to prior learning relationships, the temporary module sequence is adjusted according to the association strength parameters and connection logic in the course module association rule base.

9. The method for personalized learning paths for art courses based on deep learning according to claim 1, characterized in that, The process of generating a personalized learning path document based on the personalized learning path, which includes the learning order of course modules and guidance on module connections, includes: Extract all course module identifiers from the personalized learning path, arrange them in the learning order to form a course module learning order list, and add the corresponding module name and a brief description of the core content of the module to each course module identifier in the course module learning order list. For two adjacent course modules, analyze the connections between the core knowledge of the previous course module and the core knowledge of the next course module, identify the key content for knowledge connection, and generate knowledge connection guidelines; Analyze the progressive relationship between the competency development objectives of the previous course module and the competency development objectives of the next course module, determine the focus of competency transition, and generate competency transition guidelines; The knowledge connection guide and the ability transition guide are combined to form a module connection guide for adjacent modules, with each combination of adjacent modules corresponding to one module connection guide. The course module learning order list and all module connection guides are integrated according to a preset document structure, which includes a path overview section, a module details section, and a connection guide section. The overall learning objectives and expected learning outcomes of the personalized learning path are explained in the path overview section. The module details section presents the identifier, name, core content summary and suggested learning period of each course module in sequence. The connection guide section presents the corresponding module connection guide in the order of adjacent modules, forming a personalized learning path document.

10. A personalized learning path system for art courses based on deep learning, characterized in that, The deep learning-based personalized learning path system for art courses includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the deep learning-based personalized learning path method for art courses as described in any one of claims 1-9.

Images