Dynamic course planning method based on adaptive learning algorithm

Abstract

This invention discloses a dynamic curriculum planning method based on an adaptive learning algorithm. The method comprises the following steps: S1. Constructing a multi-dimensional data model of the learner's knowledge space; S2. Forming a knowledge graph with fractal characteristics; S3. Identifying weak areas and potential knowledge blind spots in the learner's knowledge space; S4. Recursively decomposing the knowledge graph using a fractal recursive algorithm based on the learner's real-time knowledge status analysis and learning objectives; S5. Optimizing recommended content and ranking dynamically based on the learner's knowledge mastery, weak knowledge points, and learning behavior preferences; S6. Based on the learning resources recommended in S5, dynamically adjusting the learning path and recommended resource content by monitoring the learner's performance data in real time and combining the performance data analysis results, optimizing learning suggestions, and updating the learner's knowledge status in real time until the curriculum planning is completed. This invention can better meet personalized learning needs, improve learning effectiveness, and enhance the system's cross-scenario adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of course planning, and in particular to a dynamic course planning method based on an adaptive learning algorithm. Background Art

[0002] In the existing technology, the demand for personalized learning in the field of education is growing. The traditional course planning method has many shortcomings in meeting the real-time learning needs of learners. Usually, the education platform provides unified teaching content for different learners through a predefined course structure, lacks the ability to dynamically respond to individual differences of learners, and cannot make targeted adjustments according to the real-time performance of learners during the learning process. The weak links of learners in certain knowledge points cannot be identified and strengthened in real time, which easily leads to low learning efficiency and uneven mastery of knowledge points.

[0003] In order to improve this problem, some existing technologies attempt to introduce adaptive learning algorithms to analyze learners' learning status by collecting their learning data. However, adaptive learning algorithms still have some obvious defects in practice: first, most existing systems are limited to simple linear analysis and cannot fully capture the complexity of learners' knowledge structure; second, existing dynamic course planning systems only rely on static rules or simple matching algorithms when recommending learning resources, which makes it difficult to achieve personalization and dynamic adjustment, resulting in the allocation of learning resources not meeting the actual needs of learners.

[0004] In addition, some existing technologies attempt to optimize curriculum planning through data mining and learning path analysis, but existing methods lack the ability to track and dynamically respond to learners' real-time performance in a fine-grained manner. When identifying learners' knowledge blind spots or weak areas, existing technologies mainly rely on predefined thresholds or static rules, which make it difficult to flexibly adapt to learners' changing learning states. Furthermore, existing systems have technical bottlenecks in multi-platform compatibility and resource integration. Curriculum planning systems are difficult to efficiently migrate and apply between different educational platforms, limiting their scalability and adaptability.

[0005] The above problems have greatly limited the effectiveness of personalized learning algorithms in practice. A new technical means is urgently needed to overcome the above defects and improve the dynamic response capability of the course planning system and the accuracy of resource optimization recommendations. Summary of the invention

[0006] One object of the present invention is to propose a dynamic course planning method based on an adaptive learning algorithm, which can better meet personalized learning needs, improve learning effects and the cross-scenario adaptability of the system.

[0007] A dynamic course planning method based on an adaptive learning algorithm according to an embodiment of the present invention comprises the following steps:

[0008] S1. Collect learners’ past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data, and build a multidimensional data model of the learner’s knowledge space based on the collected data;

[0009] S2. Based on the multidimensional data model, use fractal geometry to analyze the learner's knowledge space, calculate the fractal dimension of the learner's knowledge space, quantify the complexity of the learner's knowledge structure, and form a knowledge graph characterized by fractal characteristics;

[0010] S3. Based on the knowledge graph, combined with the adaptive learning algorithm, the learner's mastery of each knowledge point in the knowledge space is evaluated to identify the weak areas and potential knowledge blind spots in the learner's knowledge space;

[0011] S4. Based on the real-time knowledge status analysis results of learners and their learning goals, the knowledge graph is recursively decomposed using the fractal recursive algorithm, and the knowledge points are divided into the main knowledge module, the secondary extension module and the detailed knowledge module. The dynamic learning path is achieved by adjusting the arrangement order of the main modules, dynamically expanding the secondary modules and loading the detailed modules.

[0012] S5. Based on the learner's real-time knowledge status and dynamic learning path, the fractal characteristics are used to optimize the recommendation of multi-level learning resources. The optimization of recommended content and ranking are dynamically adjusted according to the learner's knowledge mastery, weak knowledge points and learning behavior preferences;

[0013] S6. Based on the learning resources recommended by S5, by real-time monitoring of learners’ performance data during the learning process, dynamically adjust the learning path and resource recommendation content based on the performance data analysis results, optimize learning suggestions and update learners’ knowledge status in real time until the course planning is completed.

[0014] Optionally, the S1 includes the following steps:

[0015] S11. Collect learners’ past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data:

[0016] Past course grades data i represents the learner's score in the i-th course, and its value range is [0,1];

[0017] Online learning time data T i represents the learner's cumulative online learning time related to the i-th knowledge point;

[0018] Knowledge point mastery data K i Indicates the learner's mastery of the i-th knowledge point, with a value range of [0,1], where 0 indicates no mastery and 1 indicates complete mastery;

[0019] Learning behavior preference data P i Indicates the learner's preference for using learning resources related to the i-th knowledge point;

[0020] S12. Normalize the collected past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data so that different types of data can be compared on a unified scale;

[0021] S13. Construct a multidimensional data model of the learner knowledge space based on the normalized learner data:

[0022]

[0023] Among them, S is the multidimensional data model matrix of the knowledge space.

[0024] Optionally, S2 includes the following steps:

[0025] S21. Use the knowledge space multidimensional data model S to represent each knowledge point as a node v i , the relationship between nodes is represented by edge e ij , construct the learner's knowledge space network diagram:

[0026] G=(V,E,W):

[0027] Among them, V is the knowledge point node set, E is the associated edge set between nodes, W = {w ij} is the edge weight set, edge weight w ij According to knowledge point v i And knowledge points v j Calculation of association strength:

[0028]

[0029] Among them, K i ,K j For knowledge point v i And knowledge points v j The mastery level, λ is the correlation adjustment coefficient, ∥K i ∥,∥K j ∥ is the norm of the mastery degree of the corresponding knowledge point;

[0030] S22. Use the box dimension method to calculate the fractal dimension of the knowledge space network graph G, embed the knowledge space network into the metric space, and select the box size ε of different scales. l , covering the entire knowledge space, the number of boxes N(ε l ) represents the scale ε l The minimum number of boxes required to cover the knowledge space is established by the box size εl With the number of boxes N(ε l ) to form a data set {(logε l ,logN(ε l ))};

[0031] Use linear regression method to fit the data set and calculate the fractal dimension D:

[0032]

[0033] Among them, k is the slope of the fitted line, and D is the fractal dimension of the knowledge space network, which quantifies the complexity of the learner's knowledge structure;

[0034] S23. Generate a knowledge graph T with fractal characteristics based on the calculated fractal dimension D and the knowledge space network G, and adjust the node v according to the fractal dimension D and the topological characteristics of the node i Weight

[0035]

[0036] Among them, K i For knowledge point v i The degree of mastery, deg(v i ) is the node v i The degree, is the weight adjustment index, D max is the maximum possible fractal dimension of the knowledge space;

[0037] According to the node weight and fractal characteristics, adjust the edge e ij The weight W eij :

[0038]

[0039] in, is the adjusted node weight, φ is the edge weight adjustment index;

[0040] Construct the weighted adjacency matrix T of the knowledge graph = [t ij ]:

[0041] t ij ={W0 e,ij , if node v i No and then v j Directly connected;

[0042] S24. Store the generated knowledge graph T with fractal characteristics in the database of the learning management system.

[0043] Optionally, S3 includes the following steps:

[0044] S31. Based on the knowledge graph T, combined with the degree of knowledge mastery K i and node weights in the knowledge graph Assess learners' comprehensive mastery of knowledge points:

[0045]

[0046] Among them, M i Represents knowledge point v i The comprehensive mastery level of K i For learners to understand the knowledge point v i The degree of mastery, deg(v i ) is the knowledge point v i The degree of represents the number of knowledge points directly connected to it. is the sum of all node degrees in the knowledge graph;

[0047] S32. Edge weights through knowledge graph T and node mastery level M i ,M j Calculation knowledge point v i and v j The correlation strength R ij and the weakness L ij :

[0048]

[0049] L ij =1-R ij ;

[0050] Among them, R ij Represents knowledge point v i and v j The greater the value, the stronger the association between the two knowledge points. ij Indicates the degree of weakness between two knowledge points. A larger value indicates that the relationship requires more learning attention.

[0051] S33. Based on the comprehensive mastery of all knowledge points M i The strength of association between the knowledge point and ij , generate the learner's real-time knowledge state matrix S t =[s ij ]:

[0052]

[0053] Among them, S t is the learner's real-time knowledge state matrix, s ij Indicates that the learner is at knowledge point v i and v jThe comprehensive learning status between

[0054] S34. Through the real-time knowledge state matrix S t Weakness analysis identifies weak areas and potential knowledge blind spots in learners’ knowledge space:

[0055] Weak area analysis: For all non-zero comprehensive learning states s ij The values ​​are sorted and the comprehensive learning state s is selected ij The knowledge points with values ​​lower than the preset values ​​are regarded as weak areas to focus on;

[0056] Identification of potential knowledge blind spots: through real-time knowledge status matrix S t Sparsity analysis of , locate the corresponding s that have not been learned for a long time ij = 0 and marked as potential knowledge blind spots.

[0057] Optionally, S4 includes the following steps:

[0058] S41. Based on real-time knowledge state matrix S t and the learner's learning goal O, use the fractal recursive algorithm to recursively decompose the knowledge graph T and divide the knowledge points into the main knowledge modules M core , Secondary expansion module M sub and refinement knowledge module M fine , define knowledge point v i The module to which it belongs is determined as follows:

[0059]

[0060] Among them, deg(v i ) is the knowledge point v i The degree of represents the number of knowledge points directly connected to it, μ1, μ2 and τ1, τ2 are the hierarchical thresholds of mastery degree and degree, which are dynamically adjusted according to the learning objectives;

[0061] S42. Based on the main knowledge module M core The correlation strength R of the knowledge point ij Calculate the learning order P within the main knowledge module i , give priority to learning nodes with high correlation strength between knowledge points:

[0062]

[0063] Among them, P i For knowledge point v i Priority learning ranking value;

[0064] S43. According to the secondary expansion module M sub The weakness of the knowledge point in ijReal-time status of learners i Dynamically adjust the module content and determine the recursive depth d of the secondary module sub :

[0065]

[0066] Among them, d sub is the recursion depth of the secondary module. A larger value indicates that a deeper content expansion is required. sub | is the number of knowledge points in the secondary module;

[0067] S44. For the refined knowledge module M fine , dynamically load relevant learning resources based on learners’ potential knowledge blind spots and content priorities:

[0068]

[0069] Among them, L fine represents the total amount of detailed module resources that need to be loaded, α1, β1 are the learning resource loading weight coefficients, and N(i) is the number of resources related to the knowledge point v i A set of connected adjacent knowledge points;

[0070] S45. Dynamically generate learners' personalized learning paths P according to the priority learning order of the main modules, the recursive expansion depth of the secondary modules, and the loading content of the detailed modules t :

[0071] P t =Sort({M core ,M sub ,M fine},Priority);

[0072] Among them, Sort means arranging the order of modules by priority, Priority means the priority of the module, the trunk module has the highest priority, the secondary module is the second, and the detailed module has the lowest priority.

[0073] Optionally, S5 includes the following steps:

[0074] S51. Based on personalized learning path P t and the learner's real-time knowledge state matrix S t , combined with the weakness of knowledge points and learning behavior preferences P t Calculate the priority P of each learning resource based on the strength of association with the knowledge point k :

[0075]

[0076] S52. Combined with the learners' comprehensive mastery of each knowledge point Mi The specific coverage of the learning resources that need to be recommended is quantified by the weak area analysis results C k :

[0077]

[0078] Among them, C k represents the content coverage of learning resources, γ and δ are the adjustment coefficients of learning resource coverage;

[0079] S53. According to the learning resource priority P k and content coverage C k , using fractal characteristics to optimize the multi-level ranking of learning resources, and calculating the ranking value R of each learning resource in the final recommendation sequence rank,k :

[0080] R rank,k =α2P k +β2C k ;

[0081] Among them, R rank,k is the ranking value of learning resources. The higher the value, the higher the priority. α2 and β2 are ranking weight coefficients.

[0082] S54. Generate a learning resource recommendation matrix based on the learning resource ranking optimization results. Each element of the learning resource recommendation matrix represents the matching relationship between the resource type and the knowledge point:

[0083]

[0084] Among them, R rec,ij For learning resources to knowledge points v i The recommended weight, R rank,k Resources for learning R k The sort value of

[0085] S55. Dynamically generate a personalized learning resource recommendation list based on the learning resource recommendation matrix, and adjust the recommended content according to the learner's real-time performance data during the learning process to update the learning resource recommendation matrix:

[0086]

[0087] in, is the updated recommendation weight, η1 is the feedback adjustment coefficient, Score actual For learners in learning resources R k The actual learning performance on expected The expected learning performance of the system.

[0088] The beneficial effects of the present invention are:

[0089] (1) The present invention introduces fractal geometry methods to perform multi-scale analysis on the learner's knowledge space, quantifies the complexity of the learner's knowledge structure by calculating the fractal dimension of the knowledge space, and dynamically generates a knowledge graph characterized by fractal characteristics. Compared with traditional linear analysis methods, the present invention can capture the nonlinear associations between multiple knowledge points and the distribution of knowledge blind spots, accurately identify weak areas and uncovered points in the learner's knowledge structure, and use a fractal recursive algorithm to modularly decompose the knowledge graph to divide the knowledge points into main, secondary and detailed modules, thereby realizing hierarchical management and targeted optimization of knowledge points, and significantly improving the adaptability of course planning to learners' real-time needs.

[0090] (2) The present invention combines the real-time knowledge state matrix and fractal characteristics to dynamically generate personalized learning paths. By optimizing the priority learning order of the main module, the recursive expansion depth of the secondary module, and the resource loading amount of the detailed module, the learning path can adapt to the learner's knowledge mastery status, weak knowledge points and learning goals. Compared with the traditional fixed path planning method, the dynamic path generation mechanism of the present invention can flexibly respond to the learner's real-time performance, avoid repeated learning of the mastered content, and at the same time strengthen the intensive learning of weak knowledge points, effectively improving the learning efficiency and knowledge point coverage.

[0091] (3) The present invention introduces fractal characteristic optimization in learning resource recommendation. By combining the comprehensive mastery level, weakness level, learning behavior preference and correlation strength between knowledge points, the content and order of learning resources are dynamically adjusted. The recommendation matrix and feedback mechanism are used to update the recommendation results according to the real-time performance of learners, ensuring a high degree of matching between learning resources and learning needs. Compared with the existing recommendation technology based on simple matching rules, the resource optimization recommendation mechanism of the present invention significantly improves the accuracy and dynamics of resource recommendation, providing learners with more efficient learning support. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0093] Figure 1 The present invention provides a flow chart of a dynamic course planning method based on an adaptive learning algorithm. DETAILED DESCRIPTION

[0094] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0095] refer to Figure 1, a dynamic course planning method based on an adaptive learning algorithm, comprising the following steps:

[0096] S1. Collect learners’ past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data, and build a multidimensional data model of the learner’s knowledge space based on the collected data;

[0097] S2. Based on the multidimensional data model, use fractal geometry to analyze the learner's knowledge space, calculate the fractal dimension of the learner's knowledge space, quantify the complexity of the learner's knowledge structure, and form a knowledge graph characterized by fractal characteristics;

[0098] S3. Based on the knowledge graph, combined with the adaptive learning algorithm, the learner's mastery of each knowledge point in the knowledge space is evaluated to identify the weak areas and potential knowledge blind spots in the learner's knowledge space;

[0099] S4. Based on the real-time knowledge status analysis results of learners and their learning goals, the knowledge graph is recursively decomposed using the fractal recursive algorithm, and the knowledge points are divided into the main knowledge module, the secondary extension module and the detailed knowledge module. The dynamic learning path is achieved by adjusting the arrangement order of the main modules, dynamically expanding the secondary modules and loading the detailed modules.

[0100] S5. Based on the learner's real-time knowledge status and dynamic learning path, the fractal characteristics are used to optimize the recommendation of multi-level learning resources. The optimization of recommended content and ranking are dynamically adjusted according to the learner's knowledge mastery, weak knowledge points and learning behavior preferences;

[0101] S6. Based on the learning resources recommended by S5, by real-time monitoring of learners’ performance data during the learning process, dynamically adjust the learning path and resource recommendation content based on the performance data analysis results, optimize learning suggestions and update learners’ knowledge status in real time until the course planning is completed.

[0102] In this implementation, S1 includes the following steps:

[0103] S11. Collect learners’ past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data:

[0104] Past course grades data i represents the learner's score in the i-th course, and its value range is [0,1];

[0105] Online learning time data T i represents the learner's cumulative online learning time related to the i-th knowledge point;

[0106] Knowledge point mastery data K iIndicates the learner's mastery of the i-th knowledge point, with a value range of [0,1], where 0 indicates no mastery and 1 indicates complete mastery;

[0107] Learning behavior preference data P i Indicates the learner's preference for using learning resources related to the i-th knowledge point;

[0108] S12. Normalize the collected past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data so that different types of data can be compared on a unified scale;

[0109] S13. Construct a multidimensional data model of the learner knowledge space based on the normalized learner data:

[0110]

[0111] Among them, S is the multidimensional data model matrix of the knowledge space.

[0112] In this implementation, S2 includes the following steps:

[0113] S21. Use the knowledge space multidimensional data model S to represent each knowledge point as a node v i , the relationship between nodes is represented by edge e ij , construct the learner's knowledge space network diagram:

[0114] G=(V,E,W):

[0115] Among them, V is the knowledge point node set, E is the associated edge set between nodes, W = {w ij} is the edge weight set, edge weight w ij According to knowledge point v i And knowledge points v j Calculation of association strength:

[0116]

[0117] Among them, K i ,K j For knowledge point v i And knowledge points v j The mastery level, λ is the correlation adjustment coefficient, ∥K i ∥,∥K j ∥ is the norm of the mastery degree of the corresponding knowledge point;

[0118] S22. Use the box dimension method to calculate the fractal dimension of the knowledge space network graph G, embed the knowledge space network into the metric space, and select the box size ε of different scales. l , covering the entire knowledge space, the number of boxes N(εl ) represents the scale ε l The minimum number of boxes required to cover the knowledge space is established by the box size ε l With the number of boxes N(ε l ) to form a data set {(logε l ,logN(ε l ))};

[0119] Use linear regression method to fit the data set and calculate the fractal dimension D:

[0120]

[0121] Among them, k is the slope of the fitted line, and D is the fractal dimension of the knowledge space network, which quantifies the complexity of the learner's knowledge structure;

[0122] S23. Generate a knowledge graph T with fractal characteristics based on the calculated fractal dimension D and the knowledge space network G, and adjust the node v according to the fractal dimension D and the topological characteristics of the node i The weight W vi :

[0123]

[0124] Among them, K i For knowledge point v i The degree of mastery, deg(v i ) is the node v i The degree, is the weight adjustment index, D max is the maximum possible fractal dimension of the knowledge space;

[0125] According to the node weight and fractal characteristics, adjust the edge e ij Weight

[0126]

[0127] in, is the adjusted node weight, φ is the edge weight adjustment index;

[0128] Construct the weighted adjacency matrix T of the knowledge graph = [t ij ]:

[0129]

[0130] S24. Store the generated knowledge graph T with fractal characteristics in the database of the learning management system.

[0131] In this implementation, S3 includes the following steps:

[0132] S31. Based on the knowledge graph T, combined with the degree of knowledge mastery K i and node weights in the knowledge graph Assess learners' comprehensive mastery of knowledge points:

[0133]

[0134] Among them, M i Represents knowledge point v i The comprehensive mastery level of K i For learners to understand the knowledge point v i The degree of mastery, deg(v i ) is the knowledge point v i The degree of represents the number of knowledge points directly connected to it. is the sum of all node degrees in the knowledge graph;

[0135] S32. Edge weights through knowledge graph T and node mastery level M i ,M j Calculation knowledge point v i and v j The correlation strength R ij and the weakness L ij :

[0136]

[0137] L ij =1-R ij ;

[0138] Among them, R ij Represents knowledge point v i and v j The greater the value, the stronger the association between the two knowledge points. ij Indicates the degree of weakness between two knowledge points. A larger value indicates that the relationship requires more learning attention.

[0139] S33. Based on the comprehensive mastery of all knowledge points M i The strength of association between the knowledge point and ij , generate the learner's real-time knowledge state matrix S t =[s ij ]:

[0140]

[0141] Among them, S t is the learner's real-time knowledge state matrix, s ij Indicates that the learner is at knowledge point v i and v jThe comprehensive learning status between

[0142] S34. Through the real-time knowledge state matrix S t Weakness analysis identifies weak areas and potential knowledge blind spots in learners’ knowledge space:

[0143] Weak area analysis: For all non-zero comprehensive learning states s ij The values ​​are sorted and the comprehensive learning state s is selected ij The knowledge points with values ​​lower than the preset values ​​are regarded as weak areas to focus on;

[0144] Identification of potential knowledge blind spots: through real-time knowledge status matrix S t Sparsity analysis of , locate the corresponding s that have not been learned for a long time ij = 0 and marked as potential knowledge blind spots.

[0145] In this implementation, S4 includes the following steps:

[0146] S41. Based on real-time knowledge state matrix S t and the learner's learning goal O, use the fractal recursive algorithm to recursively decompose the knowledge graph T and divide the knowledge points into the main knowledge modules M core , Secondary expansion module M sub and refinement knowledge module M fine , define knowledge point v i The module to which it belongs is determined as follows:

[0147]

[0148] Among them, deg(v i ) is the knowledge point v i The degree of represents the number of knowledge points directly connected to it, μ1, μ2 and τ1, τ2 are the hierarchical thresholds of mastery degree and degree, which are dynamically adjusted according to the learning objectives;

[0149] S42. Based on the main knowledge module M core The correlation strength R of the knowledge point ij Calculate the learning order P within the main knowledge module i , give priority to learning nodes with high correlation strength between knowledge points:

[0150]

[0151] Among them, P i For knowledge point v i Priority learning ranking value;

[0152] S43. According to the secondary expansion module M sub The weakness of the knowledge point in ijReal-time status of learners i Dynamically adjust the module content and determine the recursive depth d of the secondary module sub :

[0153]

[0154] Among them, d sub is the recursion depth of the secondary module. A larger value indicates that a deeper content expansion is required. sub | is the number of knowledge points in the secondary module;

[0155] S44. For the refined knowledge module M fine , dynamically load relevant learning resources based on learners’ potential knowledge blind spots and content priorities:

[0156]

[0157] Among them, L fine represents the total amount of detailed module resources that need to be loaded, α1, β1 are the learning resource loading weight coefficients, and N(i) is the number of resources related to the knowledge point v i A set of connected adjacent knowledge points;

[0158] S45. Dynamically generate learners' personalized learning paths P according to the priority learning order of the main modules, the recursive expansion depth of the secondary modules, and the loading content of the detailed modules t :

[0159] P t =Sort({M core ,M sub ,M fine},Priority);

[0160] Among them, Sort means arranging the order of modules by priority, Priority means the priority of the module, the trunk module has the highest priority, the secondary module is the second, and the detailed module has the lowest priority.

[0161] In this implementation, S5 includes the following steps:

[0162] S51. Based on personalized learning path P t and the learner's real-time knowledge state matrix S t , combined with the weakness of knowledge points and learning behavior preferences P t Calculate the priority P of each learning resource based on the strength of association with the knowledge point k :

[0163]

[0164] S52. Combined with the learners' comprehensive mastery of each knowledge point Mi The specific coverage of the learning resources that need to be recommended is quantified by the weak area analysis results C k :

[0165]

[0166] Among them, C k represents the content coverage of learning resources, γ and δ are the adjustment coefficients of learning resource coverage;

[0167] S53. According to the learning resource priority P k and content coverage C k , using fractal characteristics to optimize the multi-level ranking of learning resources, and calculating the ranking value R of each learning resource in the final recommendation sequence rank,k :

[0168] R rank,k =α2P k +β2C k ;

[0169] Among them, R rank,k is the ranking value of learning resources. The higher the value, the higher the priority. α2 and β2 are ranking weight coefficients.

[0170] S54. Generate a learning resource recommendation matrix based on the learning resource ranking optimization results. Each element of the learning resource recommendation matrix represents the matching relationship between the resource type and the knowledge point:

[0171]

[0172] Among them, R rec,ij For learning resources to knowledge points v i The recommended weight, R rank,k Resources for learning R k The sort value of

[0173] S55. Dynamically generate a personalized learning resource recommendation list based on the learning resource recommendation matrix, and adjust the recommended content according to the learner's real-time performance data during the learning process to update the learning resource recommendation matrix:

[0174]

[0175] in, is the updated recommendation weight, η1 is the feedback adjustment coefficient, Score actual For learners in learning resources R k The actual learning performance on expected The expected learning performance of the system.

[0176] Embodiment 1:

[0177] Example 1 In February 2024, an educational platform carried out a pilot project based on the dynamic curriculum planning method of the present invention in Shanghai. The participants were fifth-grade students of Shanghai Hongqiao No. 2 Primary School. Student Xiao Ming was one of the pilot students. His math scores were at an average level in the class, and he performed relatively weakly in the tests of fractions and geometry. The school teachers hoped to use the method of the present invention to help Xiao Ming significantly improve his mastery of fraction-related knowledge within one month and strengthen his geometric reasoning ability.

[0178] On February 5, Xiao Ming entered the learning platform and completed the initial assessment test provided by the system. The test covered all mathematical knowledge points, including fraction addition and subtraction, fraction multiplication and division, geometric reasoning, and figure area calculation. The test had a total of 45 questions, divided into four modules. The completion time and accuracy of each module were recorded in the system. After the test was completed, the system generated Xiao Ming's initial knowledge graph and marked the weak knowledge points.

[0179] The analysis shows that Xiao Ming's mastery of the "Fraction Addition and Subtraction" part is relatively low, at only 56%, and his mastery of the "Fraction Multiplication and Division" part is 68%, while his accuracy rate in the "Geometric Reasoning" part is only 47%. In addition, Xiao Ming's average time to complete the test is 20% higher than the class average, indicating that he has a problem of slow calculation speed in some modules.

[0180] The system constructed an initial knowledge graph based on Xiao Ming's test data. The fractal dimension of the graph calculated by fractal geometry method was 2.3, indicating that Xiao Ming's knowledge structure is relatively sparse and there are obvious weak areas. The analysis also found that the correlation strength between knowledge points related to scores is low, while the geometric reasoning part has almost no correlation.

[0181] The system identified "fraction addition and subtraction" as the main knowledge module, and prioritized the learning path according to Xiao Ming's learning status and generated personalized tasks. On February 6, Xiao Ming started the first round of "fraction addition and subtraction" learning. The system recommended the following content:

[0182] Micro-lesson video "Basic concepts of fractions and addition and subtraction operations" (15 minutes);

[0183] Sample question set "Denominator Unification Exercise" (10 questions in total).

[0184] After completing the task, Xiao Ming's test accuracy rate increased to 74%. The system recorded that Xiao Ming still made mistakes in the "denominator unification" related questions and was not proficient in calculating the lowest common multiple of different denominators. Based on this performance, the system automatically adjusted the learning task and added the following resources:

[0185] Experimental topic: "Intensified Practice on Unifying Denominators" (20 questions, with different levels of difficulty);

[0186] Animated explanation of "Quick calculation techniques for common multiples of denominators" (10 minutes).

[0187] On February 7, Xiao Ming completed the reinforcement test again and his accuracy rate increased to 89%. The system marked the module as "basic mastery" and dynamically adjusted the learning path to the secondary module "Fraction Multiplication and Division".

[0188] In the "Fractional Multiplication and Division" module, the system detected that Xiao Ming performed poorly in the "Fractional Multiplication Reduction" part, with an accuracy rate of only 63%. The system used a recursive decomposition algorithm to expand the module tasks and dynamically adjust the content depth. On February 9, Xiao Ming completed the following tasks:

[0189] Video course "Tips for Reducing Fractional Multiplication" (8 minutes);

[0190] In-class test (10 questions in total, 78% accuracy).

[0191] Based on the fractal characteristics analysis, the system found that while Xiao Ming had mastered the reduction of fractions, he had missed the understanding of "multiplication and denominator transformation". On February 10, the system added the following tasks:

[0192] In-class interactive questions (gradually increasing in difficulty, involving denominator transformation problems).

[0193] After Xiao Ming completed the task, the system recorded that his accuracy rate increased to 92%. At this point, the learning task of the score module was completed.

[0194] On February 12, the system switched to the geometry reasoning module. Preliminary testing showed that Xiao Ming had a high error rate in the "properties of parallel lines" section and took a long time to solve geometry problems. The system recommended the following resources:

[0195] Animated explanation of "Application of the properties of parallel lines" (5 minutes);

[0196] In-class test "Properties of parallel lines and geometric reasoning" (20 questions).

[0197] After completing the task, Xiao Ming's accuracy rate increased to 75%. Based on his error records, the system added the following detailed module content:

[0198] Problem set "Parallel Line Reasoning" (15 questions, from easy to difficult);

[0199] Video course "Quick Problem Solving Techniques for Geometric Reasoning" (12 minutes).

[0200] During the learning process of this module, Xiao Ming showed a gradual improvement in his problem-solving speed and accuracy. By the end of February, Xiao Ming's geometry module test accuracy reached 87%, and the problem-solving time was shortened by 18%.

[0201] After the experiment, the school evaluated Xiao Ming's performance and compared it with students who used traditional methods:

[0202] Table 1 Comparison data between dynamic curriculum planning method and traditional method

[0203]

[0204] In the data in Table 1 above, Xiao Ming achieved significant improvement in learning results through the method of the present invention, especially showing obvious advantages in strengthening weak knowledge points and problem-solving efficiency. Due to the lack of targeted dynamic adjustment in traditional methods, there are duplications and redundancies in the allocation of learning resources, resulting in low learning efficiency. The dynamic course planning method of the present invention provides a more efficient and accurate solution for personalized education.

[0205] The present invention introduces fractal geometry methods to perform multi-scale analysis on the learner's knowledge space, quantifies the complexity of the learner's knowledge structure by calculating the fractal dimension of the knowledge space, and dynamically generates a knowledge graph represented by fractal characteristics. Compared with traditional linear analysis methods, the present invention can capture the nonlinear correlation between multiple knowledge points and the distribution of knowledge blind spots, accurately identify weak areas and uncovered points in the learner's knowledge structure, and use a fractal recursive algorithm to perform modular decomposition of the knowledge graph to divide the knowledge points into main, secondary and detailed modules, thereby realizing hierarchical management and targeted optimization of knowledge points, and significantly improving the adaptability of course planning to learners' real-time needs.

[0206] The present invention combines the real-time knowledge state matrix and fractal characteristics to dynamically generate personalized learning paths. Through the priority learning sorting of the main modules, the recursive expansion depth of the secondary modules and the optimization of the resource loading amount of the detailed modules, the learning path can adapt to the learner's knowledge mastery status, weak knowledge points and learning goals. Compared with the traditional fixed path planning method, the dynamic path generation mechanism of the present invention can flexibly respond to the learner's real-time performance, avoid repeated learning of the mastered content, and at the same time strengthen the intensive learning of weak knowledge points, effectively improving the learning efficiency and knowledge point coverage.

[0207] The present invention introduces fractal characteristic optimization in learning resource recommendation, and dynamically adjusts the content and sorting of learning resources by combining the comprehensive mastery level, weakness level, learning behavior preference and correlation strength between knowledge points. The recommendation matrix and feedback mechanism can update the recommendation results according to the real-time performance of learners, ensuring a high degree of matching between learning resources and learning needs. Compared with the existing recommendation technology based on simple matching rules, the resource optimization recommendation mechanism of the present invention significantly improves the accuracy and dynamics of resource recommendation, and provides learners with more efficient learning support.

[0208] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A dynamic course planning method based on an adaptive learning algorithm, characterized in that: The steps include: S1. Collect learners’ past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data, and build a multidimensional data model of the learner’s knowledge space based on the collected data; S2. Based on the multidimensional data model, use fractal geometry to analyze the learner's knowledge space, calculate the fractal dimension of the learner's knowledge space, quantify the complexity of the learner's knowledge structure, and form a knowledge graph characterized by fractal characteristics; S3. Based on the knowledge graph, combined with the adaptive learning algorithm, the learner's mastery of each knowledge point in the knowledge space is evaluated to identify the weak areas and potential knowledge blind spots in the learner's knowledge space; S4. Based on the real-time knowledge status analysis results of learners and their learning goals, the knowledge graph is recursively decomposed using the fractal recursive algorithm, and the knowledge points are divided into the main knowledge module, the secondary extension module and the detailed knowledge module. The dynamic learning path is achieved by adjusting the arrangement order of the main modules, dynamically expanding the secondary modules and loading the detailed modules. S5. Based on the learner's real-time knowledge status and dynamic learning path, the fractal characteristics are used to optimize the recommendation of multi-level learning resources. The optimization of recommended content and ranking are dynamically adjusted according to the learner's knowledge mastery, weak knowledge points and learning behavior preferences; S6. Based on the learning resources recommended in S5, by real-time monitoring of learners’ performance data during the learning process, dynamically adjust the learning path and resource recommendation content based on the performance data analysis results, optimize learning suggestions and update learners’ knowledge status in real time until the course planning is completed; The S2 comprises the following steps: S21. Use the knowledge space multidimensional data model S to represent each knowledge point as a node v i , the relationship between nodes is represented by edge e ij , construct the learner's knowledge space network diagram: G=(V,E,W): Among them, V is the knowledge point node set, E is the associated edge set between nodes, W = {w ij } is the edge weight set, edge weight w ij According to knowledge point v i And knowledge points v j Calculation of association strength: Among them, K i ,K j For knowledge point v i And knowledge points v j The mastery level, λ is the correlation adjustment coefficient, ∥K i ∥,∥K j ∥ is the norm of the mastery degree of the corresponding knowledge point; S22. Use the box dimension method to calculate the fractal dimension of the knowledge space network graph G, embed the knowledge space network into the metric space, and select the box size ε of different scales. l , covers the entire knowledge space, the number of boxes N(ε l ) represents the scale ε l The minimum number of boxes required to cover the knowledge space is established by the box size ε l With the number of boxes N(ε l ) to form the data set {(logε l ,logN(ε l ))}; Use linear regression method to fit the data set and calculate the fractal dimension D: Among them, k is the slope of the fitted line, and D is the fractal dimension of the knowledge space network, which quantifies the complexity of the learner's knowledge structure; S23. Generate a knowledge graph T with fractal characteristics based on the calculated fractal dimension D and the knowledge space network G, and adjust the node v according to the fractal dimension D and the topological characteristics of the node i Weight Among them, K i For knowledge point v i The degree of mastery, deg(v i ) is the node v i The degree, is the weight adjustment index, D max is the maximum possible fractal dimension of the knowledge space; According to the node weight and fractal characteristics, adjust the edge e ij Weight in, is the adjusted node weight, φ is the edge weight adjustment index; Construct the weighted adjacency matrix T of the knowledge graph = [t ij ]: S24. Store the generated knowledge graph T with fractal characteristics in the database of the learning management system.

2. A dynamic course planning method based on an adaptive learning algorithm according to claim 1, characterized in that: The S1 comprises the following steps: S11. Collect learners’ past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data: Past course grades data i represents the learner's score in the i-th course, and its value range is [0,1]; Online learning time data T i represents the learner's cumulative online learning time related to the i-th knowledge point; Knowledge point mastery data K i Indicates the learner's mastery of the i-th knowledge point, with a value range of [0,1], where 0 indicates no mastery and 1 indicates complete mastery; Learning behavior preference data P i Indicates the learner's preference for using learning resources related to the i-th knowledge point; S12. Normalize the collected past course performance data, online learning time data, knowledge point mastery data, and learning behavior preference data so that different types of data can be compared on a unified scale; S13. Construct a multidimensional data model of the learner knowledge space based on the normalized learner data: Among them, S is the multidimensional data model matrix of the knowledge space.

3. A dynamic course planning method based on an adaptive learning algorithm according to claim 1, characterized in that: The S3 comprises the following steps: S31. Based on the knowledge graph T, combined with the degree of knowledge mastery K i and node weights in the knowledge graph Assess learners' comprehensive mastery of knowledge points: Among them, M i Represents knowledge point v i The comprehensive mastery level of K i For learners to understand the knowledge point v i The degree of mastery, deg(v i ) is the knowledge point v i The degree of represents the number of knowledge points directly connected to it. is the sum of all node degrees in the knowledge graph; S32. Edge weights through knowledge graph T and node mastery level M i ,M j Calculation knowledge point v i and v j The correlation strength R ij and the weakness L ij : Among them, R ij Represents knowledge point v i and v j The greater the value, the stronger the association between the two knowledge points. ij Indicates the degree of weakness between two knowledge points. A larger value indicates that the relationship requires more learning attention. S33. Based on the comprehensive mastery of all knowledge points M i The strength of association between the knowledge point and ij , generate the learner's real-time knowledge state matrix S t =[s ij ]: Among them, S t is the learner's real-time knowledge state matrix, s ij Indicates that the learner is at knowledge point v i and v j The comprehensive learning status between S34. Through the real-time knowledge state matrix S t Weakness analysis identifies weak areas and potential knowledge blind spots in learners’ knowledge space: Weak area analysis: For all non-zero comprehensive learning states s ij The values ​​are sorted and the comprehensive learning state s is selected ij The knowledge points with values ​​lower than the preset values ​​are regarded as weak areas to focus on; Identification of potential knowledge blind spots: through real-time knowledge status matrix S t Sparsity analysis of , locate the corresponding s that have not been learned for a long time ij = 0 and marked as potential knowledge blind spots.

4. A dynamic course planning method based on an adaptive learning algorithm according to claim 1, characterized in that: The S4 comprises the following steps: S41. Based on real-time knowledge state matrix S t and the learner's learning goal O, use the fractal recursive algorithm to recursively decompose the knowledge graph T and divide the knowledge points into the main knowledge modules M core , Secondary expansion module M sub and refinement knowledge module M fine , define knowledge point v i The module to which it belongs is determined as follows: Among them, deg(v i ) is the knowledge point v i The degree of represents the number of knowledge points directly connected to it, μ1, μ2 and τ1, τ2 are the hierarchical thresholds of mastery degree and degree, which are dynamically adjusted according to the learning objectives; S42. Based on the main knowledge module M core The correlation strength R of the knowledge point ij Calculate the learning order P within the main knowledge module i , give priority to learning nodes with high correlation strength between knowledge points: Among them, P i For knowledge point v i Priority learning ranking value; S43. According to the secondary expansion module M sub The weakness of the knowledge point in ij Real-time status of learners i Dynamically adjust the module content and determine the recursive depth d of the secondary module sub : Among them, d sub is the recursion depth of the secondary module. A larger value indicates that a deeper content expansion is required. sub | is the number of knowledge points in the secondary module; S44. For the refined knowledge module M fine , dynamically load relevant learning resources based on learners’ potential knowledge blind spots and content priorities: Among them, L fine represents the total amount of detailed module resources that need to be loaded, α1, β1 are the learning resource loading weight coefficients, and N(i) is the number of resources related to the knowledge point v i A set of connected adjacent knowledge points; S45. Dynamically generate learners' personalized learning paths P according to the priority learning order of the main modules, the recursive expansion depth of the secondary modules, and the loading content of the detailed modules t : P t =Sort({M core ,M sub ,M fine },Priority); Among them, Sort means arranging the order of modules by priority, Priority means the priority of the module, the trunk module has the highest priority, the secondary module is the second, and the detailed module has the lowest priority.

5. A dynamic course planning method based on an adaptive learning algorithm according to claim 1, characterized in that: The S5 comprises the following steps: S51. Based on personalized learning path P t and the learner's real-time knowledge state matrix S t , combined with the weakness of knowledge points and learning behavior preferences P t Calculate the priority P of each learning resource based on the strength of association with the knowledge point k : S52. Combined with the learners' comprehensive mastery of each knowledge point M i The specific coverage of the learning resources that need to be recommended is quantified by the weak area analysis results C k : Among them, C k represents the content coverage of learning resources, γ and δ are the adjustment coefficients of learning resource coverage; S53. According to the learning resource priority P k and content coverage C k , using fractal characteristics to optimize the multi-level ranking of learning resources, and calculating the ranking value R of each learning resource in the final recommendation sequence rank,k : R rank,k =α2P k +β2C k ; Among them, R rank,k is the ranking value of learning resources. The higher the value, the higher the priority. α2 and β2 are ranking weight coefficients. S54. Generate a learning resource recommendation matrix based on the learning resource ranking optimization results. Each element of the learning resource recommendation matrix represents the matching relationship between the resource type and the knowledge point: Among them, R rec,ij For learning resources to knowledge points v i The recommended weight, R rank,k Resources for learning R k The sort value of S55. Dynamically generate a personalized learning resource recommendation list based on the learning resource recommendation matrix, and adjust the recommended content according to the learner's real-time performance data during the learning process to update the learning resource recommendation matrix: in, is the updated recommendation weight, η1 is the feedback adjustment coefficient, Score actual For learners in learning resources R k The actual learning performance on expected The expected learning performance of the system.

Images