Personalized job generation method based on artificial intelligence algorithm

Abstract

The present invention discloses a personalized homework generation method based on artificial intelligence algorithm, S1, collect learners' multi-dimensional learning behavior data; S2, build learners' knowledge point mastery model based on knowledge graph; S3, generate learners' behavior feature vector set; S4, build learners' cognitive load model according to behavior feature vector set and knowledge point mastery model, combined with knowledge point relevance; S5, generate and dynamically adjust personalized homework; S6, adjust the recommended content of personalized homework and the relevance weight of knowledge points; S7, update the generation strategy of personalized homework in real time; S8, enable learners to complete appropriate exercise content at the right time; S9, provide learners with personalized homework generation scheme according to the updated behavior feature vector set, knowledge point mastery model and cognitive load model. The present invention realizes the generation and push of homework that dynamically adapts to learners' needs, and improves the accuracy and adaptability of personalized homework generation.

Description

Technical Field

[0001] The present invention relates to the field of operation technology, and in particular to a personalized operation generation method based on an artificial intelligence algorithm. Background Art

[0002] With the rapid development of artificial intelligence and big data technologies, personalized learning path planning has received widespread attention in the field of education. The goal of personalized learning paths is to provide learners with customized learning content to improve learning efficiency and experience. However, in actual applications, most existing online education platforms or learning management systems still use static content recommendation methods, that is, content push based on a single user portrait and a fixed learning progress. This method makes it difficult to dynamically track the real-time status of learners, resulting in poor adaptability of learning path recommendations and the inability to truly achieve personalized learning, especially in homework generation and learning task design.

[0003] The personalized learning path planning system in the existing technology is usually based on static recommendation algorithms, which cannot adjust the homework content in real time according to the learner's actual learning behavior, cognitive load and knowledge point mastery. In the traditional recommendation algorithm, the generation and push of homework are mostly based on the fixed knowledge point distribution, and ignore the real-time changes in demand shown by learners in the learning process. In addition, the current personalized learning system is relatively rough in data processing and often relies only on simple behavioral data. It fails to comprehensively collect learners' multi-dimensional learning data, resulting in large deviations in the recommended homework content in matching learning needs.

[0004] The existing personalized learning systems still have the following problems: First, traditional recommendation methods have limited ability to dynamically adjust homework content and are difficult to respond to learners' cognitive load and learning preferences in real time, resulting in learners facing overly complex or overly simple homework content under high or low load conditions, affecting the learning experience. Secondly, the homework generation method based on static knowledge point associations cannot identify the correlation between interdisciplinary knowledge points, resulting in the generated homework content in actual applications being difficult to support the expansion of interdisciplinary or multi-dimensional learning paths. Third, the feedback mechanism of the existing system is relatively simple, lacking in-depth feedback and adaptive optimization of learning behavior, and unable to adjust subsequent content in time according to learners' performance in homework, making it difficult to meet dynamic learning needs. In addition, the existing system cannot predict the order of personalized homework push based on learners' long-term behavior patterns and cognitive load, and often finds it difficult to achieve a reasonable arrangement of learning content.

[0005] In summary, the existing technology has great deficiencies in the dynamic adaptability of personalized assignment generation, cross-disciplinary content association and feedback mechanism optimization, and cannot meet the learners' growing personalized needs. Therefore, a new personalized assignment generation method is urgently needed to solve the above problems. Summary of the invention

[0006] One purpose of the present invention is to propose a personalized homework generation method based on an artificial intelligence algorithm. The present invention realizes the generation and push of homework that dynamically adapts to learners' needs, thereby improving the accuracy and adaptability of personalized homework generation.

[0007] A personalized job generation method based on an artificial intelligence algorithm according to an embodiment of the present invention includes the following steps:

[0008] S1. Collect learners’ multi-dimensional learning behavior data;

[0009] S2. Build a learner's knowledge point mastery model based on the knowledge graph, generate the learner's current knowledge point mastery status, and identify the learner's mastered, unmastered, and partially mastered knowledge points;

[0010] S3, preprocessing the learner's learning behavior data to generate a learner's behavior feature vector set;

[0011] S4. Based on the behavioral feature vector set and the knowledge point mastery model, combined with the knowledge point relevance, the learner's cognitive load model is constructed, the learner's cognitive load value is calculated in real time, and the knowledge point relevance weight in the knowledge graph is dynamically adjusted;

[0012] S5. Generate and dynamically adjust personalized assignments based on the behavioral feature vector set, knowledge point mastery model, and cognitive load model;

[0013] S6. By real-time monitoring of learners’ homework completion rate, accuracy rate, problem-solving time distribution and difficulty evaluation, the learners’ behavioral feature vector set and knowledge point mastery model are updated, and the recommended content of personalized homework and the relevance weight of knowledge points are adjusted;

[0014] S7, dynamically adjust the structure of the knowledge graph according to the changes in the behavioral feature vector set, knowledge point mastery model and cognitive load model, and update the generation strategy of personalized tasks in real time;

[0015] S8. Combine learners’ long-term behavior patterns and learning preferences to optimize the order of homework generation through adaptive recommendation algorithms, and plan homework content and difficulty levels in advance;

[0016] S9. Provide learners with personalized homework generation solutions based on the updated behavioral feature vector set, knowledge point mastery model and cognitive load model, and optimize homework content and generation path in real time based on the dynamic association of different knowledge points in the knowledge graph.

[0017] Optionally, the S1 specifically includes:

[0018] S11. Define the learner's multi-dimensional learning behavior data feature matrix M. Each data item in the multi-dimensional learning behavior data feature matrix contains the following dimensions:

[0019] Study time T i Record the time learners spend on a specific knowledge point or homework task in seconds;

[0020] Learning Progress i Indicates the learner's current learning path progress, calculated by the percentage of knowledge points completed, ranging from [0,100];

[0021] Test score S i Record the learner's score in the test related to the current knowledge point, expressed as a percentage, ranging from [0,100];

[0022] Learning Preference i is the learner's preference information in terms of content, resource form, etc., expressed in the form of a numerical vector;

[0023] Knowledge points mastered K i The learner's mastery of a specific knowledge point ranges from [0,1], where 1 indicates complete mastery and 0 indicates no mastery;

[0024] S12. Construct a multi-dimensional learning behavior data feature matrix M, where the elements of the multi-dimensional learning behavior data feature matrix M are the learner's multi-dimensional learning behavior data:

[0025]

[0026] Among them, n is the number of data items, which represents the total number of learning behavior data items of learners in different learning scenarios.

[0027] Optionally, the S2 specifically includes:

[0028] S21. Constructing an interdisciplinary knowledge graph:

[0029] G={G m ,G p ,…,G k};

[0030] Each interdisciplinary knowledge graph G i Corresponding to the knowledge point association structure of a subject, including multiple nodes and edges, the node represents the knowledge point set K in the subject i ={k i1 ,k i2 ,…,k im}, the edge represents the association between knowledge points, and each association has a weight value w i,j , w i,j Represents knowledge point ki With knowledge point k j The strength of association between disciplines, the association between disciplines is represented by interdisciplinary nodes, which are defined as common knowledge points or logically related knowledge points between two interdisciplinary knowledge graphs of different disciplines;

[0031] S22, based on the learner's multi-dimensional learning behavior data feature matrix M of the knowledge point mastery data K i , generate the learner's subject knowledge point mastery state vector V i :

[0032] V i = {v i1 ,v i2 ,…,v im};

[0033] Among them, v ij represents the learner's understanding of knowledge point k in subject i ij The degree of mastery is in the range of [0,1], where 0 means not mastered and 1 means fully mastered;

[0034] S23, based on the learner's knowledge points, master the state vector V i and the interdisciplinary knowledge graph set G to construct the learner's interdisciplinary knowledge point mastery model M CK , interdisciplinary knowledge point mastery model M CK Including the mastery of knowledge points within the subject and the mastery of interdisciplinary related knowledge points:

[0035] The state vector V of the learner's knowledge points in the interdisciplinary knowledge graph i , identify the set of knowledge points that have been mastered in the subject A collection of unlearned knowledge points And a collection of partially mastered knowledge points in:

[0036] A collection of knowledge points that have been mastered θ a To master the threshold;

[0037] A collection of unlearned knowledge points θ u The threshold is not mastered;

[0038] A collection of partially mastered knowledge points Indicates the knowledge point between mastering and not mastering.

[0039] Optionally, the S3 specifically includes:

[0040] S31, standardize the data of each dimension in the learner's learning behavior data feature matrix M to generate a standardized learning behavior feature matrix M', in which each data item includes a standardized value of learning time, learning progress, test scores, learning preferences and knowledge point mastery;

[0041] S32, based on the standardized learning behavior feature matrix M', extracting the learner's learning habit features H including the learner's learning frequency, learning time period preference and learning duration;

[0042] S33, based on the mastery of knowledge points K i Generate the learner's knowledge point mastery level feature C, which is the learner's mastery distribution of each knowledge point;

[0043] S34, calculate the learner's learning efficiency feature E, which is calculated by analyzing the learner's learning time T on each knowledge point. i With test score S i The ratio of generates the learning efficiency vector E;

[0044] S35, based on learning preference L i Generate learners' problem-solving preference features P, which represent learners' preferences for different types of assignments or resources;

[0045] S36, combining the generated learning habit feature vector H, knowledge point mastery vector C, learning efficiency vector E and problem-solving preference vector P to form the learner's behavior feature vector set F:

[0046] F={H,C,E,P}.

[0047] Optionally, the S4 specifically includes:

[0048] S41, based on the behavior feature vector set F and the knowledge point mastering model M K , construct the learner's learning fatigue characteristic F d , learning fatigue feature F d By analyzing the learners' learning time distribution and knowledge mastery level:

[0049] F d =α·T+β·(1-C);

[0050] Among them, T is the learning time, C is the knowledge point mastery vector, α and β are weight coefficients;

[0051] S42. Based on the relevance weight of knowledge points in the interdisciplinary knowledge graph G, combined with the complexity of knowledge points, calculate the learner's knowledge point complexity feature D c :

[0052] D c =∑ i,j w i,j ·d j ;

[0053] Among them, w i,j is the knowledge point k in the interdisciplinary knowledge graph i With k j The relevance weight, d j is knowledge point k j The complexity value of

[0054] S43. Based on the learning efficiency feature E in the behavioral feature vector set, calculate the learner's problem-solving speed feature S r , problem-solving speed characteristic S r Defined as the inverse of the learning efficiency vector E:

[0055]

[0056] Among them, E is the learning efficiency feature, which is calculated by the problem-solving speed S r To reflect the learner's problem-solving speed;

[0057] S44, based on the test score feature S in the behavior feature vector set, calculate the learner's accuracy feature A, where the accuracy feature A represents the learner's accuracy in the test questions, and the range is [0,1];

[0058] S45. Based on the learning fatigue characteristics F d , Knowledge point complexity feature D c , Problem-solving speed characteristics S r and accuracy feature A, construct the learner's cognitive load model L, and the cognitive load value L of the cognitive load model L v Defined as:

[0059] L v =γ·F d +δ·D c +η·S r +θ·(1-A);

[0060] Among them, γ, δ, η, and η are weight coefficients, which are used to adjust the weight of each feature in the calculation of cognitive load value;

[0061] S46. Based on the calculated cognitive load value L v , the relevance weights of knowledge points in the interdisciplinary knowledge graph G are dynamically adjusted, the complexity of the relevance weights of knowledge points is reduced when the cognitive load value is high, and the complexity of the relevance weights of knowledge points is increased when the cognitive load value is low.

[0062] Optionally, the S5 specifically includes:

[0063] S51, based on the behavior feature vector set F, knowledge point mastering model M K and the cognitive load value L of the cognitive load model L v , judge the learner's current cognitive load state, if the cognitive load value L v Above the preset threshold θ h , it is considered that the learner's cognitive load is high and enters the low-difficulty assignment generation process. If the cognitive load value L v Below the preset threshold θ l , it is considered that the learner’s cognitive load is low and enters the high-difficulty assignment generation process;

[0064] S52, when the learner's cognitive load value L v ≥θ h When the learner has a learning habit feature H and a knowledge point mastery vector C, a set of low-difficulty exercises Q related to basic knowledge points is generated. b :

[0065] Q b = {q i ∈Q|c i ≤α};

[0066] Among them, Q is the question bank, q i Indicates low-difficulty exercises, c i Representation and Exercises q i The mastery degree of relevant knowledge points, α is the mastery threshold of basic knowledge;

[0067] S53, when the learner's cognitive load value L v ≤θ l When , according to the learning efficiency feature E and the knowledge point mastery vector C, a set of high-difficulty exercises Q related to advanced knowledge points is generated. a :

[0068] Q a = {q i ∈Q|c i >α};

[0069] S54, the generated low-difficulty exercise set Q b Or a collection of difficult exercises Q a Push it to the learner's personalized homework, and update the behavior feature vector set F and knowledge point mastery model M based on the learner's completion of each question, including completion time, accuracy and cognitive load value feedback data K ;

[0070] S55. Based on the learner's homework completion data and feedback information, the generation threshold α and cognitive load threshold θ of low-difficulty exercises and high-difficulty exercises are adjusted in real time h With θ l .

[0071] Optionally, the S8 specifically includes:

[0072] S81. Construct a long-term learning behavior model B based on the learner's long-term behavior data. The long-term learning behavior model B is generated by analyzing the learner's learning frequency, learning efficiency, and changes in knowledge points at different learning stages.

[0073] S82, based on the learner's learning preference feature vector P and long-term learning behavior pattern B, predict the learner's demand probability for different types of homework content and generate a learning demand prediction vector D;

[0074] S83. Master Model M by combining knowledge points K and the cognitive load change value ΔL of cognitive load model L v , predict the change trend of learners' future learning needs and generate a future learning needs prediction model F P :

[0075]

[0076] Where, ΔL v Indicates the change in cognitive load between the current stage and the previous stage, (1-L v ) is used to reflect the inverse demand relationship of the learner’s current cognitive load state;

[0077] S84. Prediction model based on future learning needs F P Plan the learner's personalized homework content and difficulty level in advance and determine the generation order vector S:

[0078]

[0079] Among them, d j is the probability of learners’ demand for knowledge point j, w i,j is the knowledge point k in the knowledge graph i With knowledge point k j The association weight is used to evaluate the association strength between knowledge points, (1-C j ) is the inverse relationship of knowledge point mastery, that is, the degree to which the learner has not mastered knowledge point j, F d is the learner’s learning fatigue, reflecting their current learning status, p is the learning preference feature, which indicates the learner's preference for specific content or question type. λ is the adjustment coefficient, which is used to control the impact of cognitive load on the generation order. vis the learner's current cognitive load value, indicating the learner's current learning pressure, μ is the learner's ability coefficient, which is used to adjust the difficulty of knowledge points, and D c is the complexity of the knowledge point, which is used to influence the generation order according to the difficulty of the knowledge point;

[0080] S85. Based on the adaptive recommendation algorithm, personalized assignments are pushed according to the generated sequence vector S, and after each push, the long-term learning behavior pattern B and the learning demand prediction vector D are updated according to the learner's feedback data, so that the assignment generation sequence is dynamically matched with the learner's current and future needs.

[0081] The beneficial effects of the present invention are:

[0082] (1) The present invention introduces a dynamic generation mechanism based on learners' cognitive load, knowledge point mastery model and behavioral feature vector set in the adaptive recommendation algorithm. By constructing a real-time updated cognitive load model and behavioral feature vector set, the current load state of the learner can be accurately calculated, and learning content and homework adapted to the current state can be generated. When the learner's cognitive load is high, the system will automatically generate low-difficulty homework questions related to basic knowledge points to help learners gradually improve their mastery of knowledge points. When the learner's cognitive load is low, the system will dynamically generate high-difficulty questions to enhance the challenge of learning and avoid wasting learning efficiency. Traditional systems often lack response to learners' real-time status, while the dynamic generation mechanism of the present invention effectively solves the problem of low matching of homework content.

[0083] (2) The present invention introduces an interdisciplinary knowledge graph model that can identify the correlation between interdisciplinary knowledge points and generate homework content with a high degree of correlation to meet the cross-disciplinary knowledge needs of different disciplines. By constructing knowledge graphs of multiple disciplines and establishing correlations between knowledge points in the graphs, the recommendation algorithm of the present invention can generate interdisciplinary homework content to help learners achieve a wider range of knowledge transfer and understanding. The interdisciplinary association mechanism significantly improves the diversity of homework content and knowledge coverage, enabling learners to integrate different disciplines and improve the breadth and depth of learning. The system can adjust the structure of the knowledge graph in real time according to the mastery model of knowledge points to ensure that the homework content always meets the learners' interdisciplinary needs.

[0084] (3) The present invention optimizes the order of homework generation by analyzing learners' long-term behavior patterns and learning preferences. By using the learning demand prediction model and the adaptive update mechanism of learning behavior patterns, the system can plan the order of pushing homework content in advance, so that learners can obtain the most suitable learning resources and homework questions at the best time node. The push sequence optimization effectively solves the problem of inaccurate timing of homework content push in traditional recommendation systems, ensuring that the push of learning content matches the learner's long-term learning pattern, thereby improving learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] 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:

[0086] Figure 1 A flowchart of a personalized job generation method based on an artificial intelligence algorithm proposed by the present invention;

[0087] Figure 2 This is a structural diagram of the knowledge graph construction and dynamic update in the personalized job generation method based on artificial intelligence algorithm proposed in the present invention. DETAILED DESCRIPTION

[0088] 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.

[0089] refer to Figure 1-Figure 2 , a personalized job generation method based on artificial intelligence algorithm, comprising the following steps:

[0090] S1. Collect learners’ multi-dimensional learning behavior data;

[0091] S2. Build a learner's knowledge point mastery model based on the knowledge graph, generate the learner's current knowledge point mastery status, and identify the learner's mastered, unmastered, and partially mastered knowledge points;

[0092] S3, preprocessing the learner's learning behavior data to generate a learner's behavior feature vector set;

[0093] S4. Based on the behavioral feature vector set and the knowledge point mastery model, combined with the knowledge point relevance, the learner's cognitive load model is constructed, the learner's cognitive load value is calculated in real time, and the knowledge point relevance weight in the knowledge graph is dynamically adjusted;

[0094] S5. Generate and dynamically adjust personalized assignments based on the behavioral feature vector set, knowledge point mastery model, and cognitive load model;

[0095] S6. By real-time monitoring of learners’ homework completion rate, accuracy rate, problem-solving time distribution and difficulty evaluation, the learners’ behavioral feature vector set and knowledge point mastery model are updated, and the recommended content of personalized homework and the relevance weight of knowledge points are adjusted;

[0096] S7, dynamically adjust the structure of the knowledge graph according to the changes in the behavioral feature vector set, knowledge point mastery model and cognitive load model, and update the generation strategy of personalized tasks in real time;

[0097] S8. Combine learners’ long-term behavior patterns and learning preferences to optimize the order of homework generation through adaptive recommendation algorithms, and plan homework content and difficulty levels in advance;

[0098] S9. Provide learners with personalized homework generation solutions based on the updated behavioral feature vector set, knowledge point mastery model and cognitive load model, and optimize homework content and generation path in real time based on the dynamic association of different knowledge points in the knowledge graph.

[0099] In this implementation, S1 specifically includes:

[0100] S11. Define the learner's multi-dimensional learning behavior data feature matrix M. Each data item in the multi-dimensional learning behavior data feature matrix contains the following dimensions:

[0101] Study time T i Record the time learners spend on a specific knowledge point or homework task in seconds;

[0102] Learning Progress i Indicates the learner's current learning path progress, calculated by the percentage of knowledge points completed, ranging from [0,100];

[0103] Test score S i Record the learner's score in the test related to the current knowledge point, expressed as a percentage, ranging from [0,100];

[0104] Learning Preference i is the learner's preference information in terms of content, resource form, etc., expressed in the form of a numerical vector;

[0105] Knowledge points mastered K i The learner's mastery of a specific knowledge point ranges from [0,1], where 1 indicates complete mastery and 0 indicates no mastery;

[0106] S12. Construct a multi-dimensional learning behavior data feature matrix M, where the elements of the multi-dimensional learning behavior data feature matrix M are the learner's multi-dimensional learning behavior data:

[0107]

[0108] Among them, n is the number of data items, which represents the total number of learning behavior data items of learners in different learning scenarios.

[0109] In this implementation, S2 specifically includes:

[0110] S21. Constructing an interdisciplinary knowledge graph:

[0111] G={Gm ,G p ,…,G k};

[0112] Each interdisciplinary knowledge graph G i Corresponding to the knowledge point association structure of a subject, including multiple nodes and edges, the node represents the knowledge point set K in the subject i ={k i1 ,k i2 ,…,k im}, the edge represents the association between knowledge points, and each association has a weight value w i,j , w i,j Represents knowledge point k i With knowledge point k j The strength of association between disciplines, the association between disciplines is represented by interdisciplinary nodes, which are defined as common knowledge points or logically related knowledge points between two interdisciplinary knowledge graphs of different disciplines;

[0113] S22, based on the learner's multi-dimensional learning behavior data feature matrix M of the knowledge point mastery data K i , generate the learner's subject knowledge point mastery state vector V i :

[0114] V i = {v i1 ,v i2 ,…,v im};

[0115] Among them, v ij represents the learner's understanding of knowledge point k in subject i ij The degree of mastery is in the range of [0,1], where 0 means not mastered and 1 means fully mastered;

[0116] S23, based on the learner's knowledge points, master the state vector V i and the interdisciplinary knowledge graph set G to construct the learner's interdisciplinary knowledge point mastery model M CK , interdisciplinary knowledge point mastery model M CK Including the mastery of knowledge points within the subject and the mastery of interdisciplinary related knowledge points:

[0117] The state vector V of the learner's knowledge points in the interdisciplinary knowledge graph i , identify the set of knowledge points that have been mastered in the subject A collection of unlearned knowledge points And a collection of partially mastered knowledge points in:

[0118] A collection of knowledge points that have been mastered θa To master the threshold;

[0119] A collection of unlearned knowledge points θ u The threshold is not mastered;

[0120] A collection of partially mastered knowledge points Indicates the knowledge point between mastering and not mastering.

[0121] In this implementation, S3 specifically includes:

[0122] S31, standardize the data of each dimension in the learner's learning behavior data feature matrix M to generate a standardized learning behavior feature matrix M', in which each data item includes a standardized value of learning time, learning progress, test scores, learning preferences and knowledge point mastery;

[0123] S32, based on the standardized learning behavior feature matrix M', extracting the learner's learning habit features H including the learner's learning frequency, learning time period preference and learning duration;

[0124] S33, based on the mastery of knowledge points K i Generate the learner's knowledge point mastery level feature C, which is the learner's mastery distribution of each knowledge point;

[0125] S34, calculate the learner's learning efficiency feature E, which is calculated by analyzing the learner's learning time T on each knowledge point. i With test score S i The ratio of generates the learning efficiency vector E;

[0126] S35, based on learning preference L i Generate learners' problem-solving preference features P, which represent learners' preferences for different types of assignments or resources;

[0127] S36, combining the generated learning habit feature vector H, knowledge point mastery vector C, learning efficiency vector E and problem-solving preference vector P to form the learner's behavior feature vector set F:

[0128] F={H,C,E,P}.

[0129] In this implementation, S4 specifically includes:

[0130] S41, based on the behavior feature vector set F and the knowledge point mastering model M K , construct the learner's learning fatigue characteristic F d , learning fatigue feature F dBy analyzing the learners' learning time distribution and knowledge mastery level:

[0131] F d =α·T+β·(1-C);

[0132] Among them, T is the learning time, C is the knowledge point mastery vector, α and β are weight coefficients;

[0133] S42. Based on the relevance weight of knowledge points in the interdisciplinary knowledge graph G, combined with the complexity of knowledge points, calculate the learner's knowledge point complexity feature D c :

[0134] D c =∑ i,j w i,j ·d j ;

[0135] Among them, w i,j is the knowledge point k in the interdisciplinary knowledge graph i With k j The relevance weight, d j is knowledge point k j The complexity value of

[0136] S43. Based on the learning efficiency feature E in the behavioral feature vector set, calculate the learner's problem-solving speed feature S r , problem-solving speed characteristic S r Defined as the inverse of the learning efficiency vector E:

[0137]

[0138] Among them, E is the learning efficiency feature, which is calculated by the problem-solving speed S r To reflect the learner's problem-solving speed;

[0139] S44, based on the test score feature S in the behavior feature vector set, calculate the learner's accuracy feature A, where the accuracy feature A represents the learner's accuracy in the test questions, and the range is [0,1];

[0140] S45. Based on the learning fatigue characteristics F d , Knowledge point complexity feature D c , Problem-solving speed characteristics S r and accuracy feature A, construct the learner's cognitive load model L, and the cognitive load value L of the cognitive load model L v Defined as:

[0141] L v =γ·F d +δ·D c +η·S r +θ·(1-A);

[0142] Among them, γ, δ, η, and η are weight coefficients, which are used to adjust the weight of each feature in the calculation of cognitive load value;

[0143] S46. Based on the calculated cognitive load value L v , the relevance weights of knowledge points in the interdisciplinary knowledge graph G are dynamically adjusted, the complexity of the relevance weights of knowledge points is reduced when the cognitive load value is high, and the complexity of the relevance weights of knowledge points is increased when the cognitive load value is low.

[0144] In this implementation, S5 specifically includes:

[0145] S51, based on the behavior feature vector set F, knowledge point mastering model M K and the cognitive load value L of the cognitive load model L v , judge the learner's current cognitive load state, if the cognitive load value L v Above the preset threshold θ h , it is considered that the learner's cognitive load is high and enters the low-difficulty assignment generation process. If the cognitive load value L v Below the preset threshold θ l , it is considered that the learner’s cognitive load is low and enters the high-difficulty assignment generation process;

[0146] S52, when the learner's cognitive load value L v ≥θ h When the learner has a learning habit feature H and a knowledge point mastery vector C, a set of low-difficulty exercises Q related to basic knowledge points is generated. b :

[0147] Q b = {q i ∈Q|c i ≤α};

[0148] Among them, Q is the question bank, q i Indicates low-difficulty exercises, c i Representation and Exercises q i The mastery degree of relevant knowledge points, α is the mastery threshold of basic knowledge;

[0149] S53, when the learner's cognitive load value L v ≤θ l When , according to the learning efficiency feature E and the knowledge point mastery vector C, a set of high-difficulty exercises Q related to advanced knowledge points is generated. a :

[0150] Q a = {q i ∈Q|c i >α};

[0151] S54, the generated low-difficulty exercise set Q b Or a collection of difficult exercises Q a Push it to the learner's personalized homework, and update the behavior feature vector set F and knowledge point mastery model M based on the learner's completion of each question, including completion time, accuracy and cognitive load value feedback data K ;

[0152] S55. Based on the learner's homework completion data and feedback information, the generation threshold α and cognitive load threshold θ of low-difficulty exercises and high-difficulty exercises are adjusted in real time h With θ l .

[0153] In this implementation, S8 specifically includes:

[0154] S81. Construct a long-term learning behavior model B based on the learner's long-term behavior data. The long-term learning behavior model B is generated by analyzing the learner's learning frequency, learning efficiency, and changes in knowledge points at different learning stages.

[0155] S82, based on the learner's learning preference feature vector P and long-term learning behavior pattern B, predict the learner's demand probability for different types of homework content and generate a learning demand prediction vector D;

[0156] S83. Master Model M by combining knowledge points K and the cognitive load change value ΔL of cognitive load model L v , predict the change trend of learners' future learning needs and generate a future learning needs prediction model F P :

[0157]

[0158] Where, ΔL v Indicates the change in cognitive load between the current stage and the previous stage, (1-L v ) is used to reflect the inverse demand relationship of the learner’s current cognitive load state;

[0159] S84. Prediction model based on future learning needs F P Plan the learner's personalized homework content and difficulty level in advance and determine the generation order vector S:

[0160]

[0161] Among them, d j is the probability of learners’ demand for knowledge point j, w i,j is the knowledge point k in the knowledge graph i With knowledge point kj The association weight is used to evaluate the association strength between knowledge points, (1-C j ) is the inverse relationship of knowledge point mastery, that is, the degree to which the learner has not mastered knowledge point j, F d is the learner’s learning fatigue, reflecting their current learning status, p is the learning preference feature, which indicates the learner's preference for specific content or question type. λ is the adjustment coefficient, which is used to control the impact of cognitive load on the generation order. v is the learner's current cognitive load value, indicating the learner's current learning pressure, μ is the learner's ability coefficient, which is used to adjust the difficulty of knowledge points, and D c is the complexity of the knowledge point, which is used to influence the generation order according to the difficulty of the knowledge point;

[0162] S85. Based on the adaptive recommendation algorithm, personalized assignments are pushed according to the generated sequence vector S, and after each push, the long-term learning behavior pattern B and the learning demand prediction vector D are updated according to the learner's feedback data, so that the assignment generation sequence is dynamically matched with the learner's current and future needs.

[0163] Embodiment 1:

[0164] In a middle school in City A, mathematics and physics teachers of the second grade of junior high school used the personalized homework generation method of the present invention to carry out an interdisciplinary learning experiment of "quadratic functions and physical motion". 60 students were selected for the experiment and divided into an experimental group and a control group with 30 students in each group. The experimental group used the method of the present invention to generate personalized homework, and the control group used traditional methods to assign uniform homework.

[0165] At 8:30 am on October 10, the system collected and analyzed the data of the experimental group of students. The learning behavior data of a student A in the same grade in the previous week showed that the student scored 72 points in the test of mathematical quadratic functions and 65 points in the test of physical kinematics. The analysis found that the correct rate of student A was significantly reduced when the questions involved the physical application of quadratic functions, and the time to solve the questions was 30 seconds longer than that of ordinary questions. The system generated the knowledge point mastery status of student A based on the behavioral feature vector set and the knowledge mastery model, showing that his mastery of the basic concept of quadratic functions was 85%, while his mastery of the application of quadratic functions in free fall kinematics was only 50%. The cognitive load value of student A was 0.72, which was higher than the light load threshold set by the system, indicating that the current learning pressure was relatively high. Based on this, the system assigned student A to the light load personalized homework mode and generated review questions to help him consolidate his understanding.

[0166] In the generated homework, the system recommends a question to student A: "An object falls freely from a certain height. Ignoring air resistance, describe the relationship between its displacement and time." At the same time, the system provides step-by-step instructions for A. For example, the first step is to draw a displacement-time curve, and the second step is to compare the curve with the quadratic function y=ax in mathematics. 2 Student A spent 5 minutes to answer the question, with a correct rate of 90%, which was 15 percentage points higher than before without step-by-step guidance.

[0167] At 9:00 that evening, after the system collected the problem-solving data of student A and analyzed his homework completion and accuracy, it was found that student A's comprehension of physics questions had improved, his knowledge point mastery had increased by 10%, and his cognitive load value had dropped to 0.65, entering the medium load range set by the system. So the system generated the next stage of personalized homework for student A the next day, with a slightly higher difficulty, involving the application of quadratic functions in the projectile motion of objects. The homework question was: "An object is thrown at a certain initial velocity, ignoring air resistance, try to write the relationship between displacement y and time t." A spent 8 minutes on this question and successfully completed the question through the system's step-by-step guidance, with an accuracy rate of 85%.

[0168] In the control group, student B's situation was similar to that of student A, but because the personalized homework generation method was not adopted, student B encountered more difficulties in answering similar questions. Student B spent 15 minutes on the quadratic function-related questions in the physical kinematics test, and the correct rate was only 60%. When analyzing the data, the system found that student B's problem-solving time was 50% higher than that of student A, while the correct rate was 25% lower, which shows that the traditional method has not effectively helped student B master the associated knowledge in interdisciplinary applications.

[0169] In order to verify the adaptability of the system, on October 15, the system analyzed another student C in the experimental group. C showed a higher learning preference score in the long-term behavioral data, especially in the more difficult questions. The system generated a difficult assignment for C with multi-step reasoning. The question content involved the curve change of the quadratic function under different motion conditions. The system also set an increasing difficulty mode. For example, the first question was: "An object is thrown vertically with a certain initial velocity. Find the displacement of the highest point." On this question, student C spent 6 minutes and achieved a 90% accuracy rate.

[0170] On October 18, the system summarized the learning feedback data of student C and found that his mastery of relevant knowledge points increased by 15%, and the time to solve the problem was shortened by 30%. At this time, the system generated more challenging questions, such as "Application of quadratic function relationship of object motion on inclined plane". On this question, student C's completion rate remained above 85%, and the time to solve the problem was further shortened to 5 minutes.

[0171] The following is a comparison of the specific data between the experimental group and the control group:

[0172] Table 1 Comparison of student data between the experimental group and the control group

[0173]

[0174]

[0175] Through the above specific homework generation and push process, the method of the present invention demonstrates significant personalized adaptability, effectively solving the problems of lack of personalized support and difficulty in dynamically adapting to learners' status in traditional methods. The experimental group is better than the control group in homework completion time, accuracy and knowledge mastery. At the same time, the significant decrease in cognitive load reflects the system's effective regulation of learners' learning pressure, enabling students to make continuous progress under an appropriate load level, significantly improving learning efficiency and effectiveness.

[0176] The present invention introduces a dynamic generation mechanism based on learners' cognitive load, knowledge point mastery model and behavior feature vector set in the adaptive recommendation algorithm. By constructing a real-time updated cognitive load model and behavior feature vector set, the current load state of the learner can be accurately calculated, and learning content and homework adapted to the current state can be generated. When the learner's cognitive load is high, the system will automatically generate low-difficulty homework questions related to basic knowledge points to help learners gradually improve their mastery of knowledge points. When the learner's cognitive load is low, the system will dynamically generate high-difficulty questions to enhance learning challenges and avoid waste of learning efficiency. Traditional systems often lack response to learners' real-time status, while the dynamic generation mechanism of the present invention effectively solves the problem of low matching of homework content.

[0177] The present invention introduces an interdisciplinary knowledge graph model that can identify the correlation between interdisciplinary knowledge points and generate homework content with a high degree of correlation to meet the cross-disciplinary knowledge needs of different disciplines. By constructing knowledge graphs of multiple disciplines and establishing correlations between knowledge points in the graphs, the recommendation algorithm of the present invention can generate interdisciplinary homework content to help learners achieve a wider range of knowledge transfer and understanding. The interdisciplinary association mechanism significantly improves the diversity of homework content and knowledge coverage, enabling learners to integrate different disciplines and improve the breadth and depth of learning. The system can adjust the structure of the knowledge graph in real time according to the mastery model of the knowledge points to ensure that the homework content always meets the learners' interdisciplinary needs.

[0178] The present invention optimizes the order of homework generation by analyzing learners' long-term behavior patterns and learning preferences. By utilizing the learning demand prediction model and the adaptive update mechanism of learning behavior patterns, the system can plan the order of pushing homework content in advance, so that learners can obtain the most suitable learning resources and homework questions at the best time node. The push sequence optimization effectively solves the problem of inaccurate timing of homework content push in traditional recommendation systems, ensuring that the push of learning content matches the learner's long-term learning pattern, thereby improving learning efficiency.

[0179] 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 personalized job generation method based on artificial intelligence algorithm, characterized in that: The steps include: S1. Collect learners’ multi-dimensional learning behavior data; S2. Build a learner's knowledge point mastery model based on the knowledge graph, generate the learner's current knowledge point mastery status, and identify the learner's mastered, unmastered, and partially mastered knowledge points; S3, preprocessing the learner's learning behavior data to generate a learner's behavior feature vector set; S4. Based on the behavioral feature vector set and the knowledge point mastery model, combined with the knowledge point relevance, the learner's cognitive load model is constructed, the learner's cognitive load value is calculated in real time, and the knowledge point relevance weight in the knowledge graph is dynamically adjusted; S5. Generate and dynamically adjust personalized assignments based on the behavioral feature vector set, knowledge point mastery model, and cognitive load model; S6. By real-time monitoring of learners’ homework completion rate, accuracy rate, problem-solving time distribution and difficulty evaluation, the learners’ behavioral feature vector set and knowledge point mastery model are updated, and the recommended content of personalized homework and the relevance weight of knowledge points are adjusted; S7, dynamically adjust the structure of the knowledge graph according to the changes in the behavioral feature vector set, knowledge point mastery model and cognitive load model, and update the generation strategy of personalized tasks in real time; S8. Combine learners’ long-term behavior patterns and learning preferences to optimize the order of homework generation through adaptive recommendation algorithms, and plan homework content and difficulty levels in advance; The S8 includes combining knowledge points to master the model M K and the cognitive load change value ΔL of cognitive load model L v , predict the change trend of learners' future learning needs and generate a future learning needs prediction model F P : Where, ΔL v Indicates the change in cognitive load between the current stage and the previous stage, (1-L v ) is used to reflect the inverse demand relationship of the learner’s current cognitive load state; Based on the future learning demand prediction model F P Plan the learner's personalized homework content and difficulty level in advance and determine the generation order vector S: Among them, d j is the probability of learners’ demand for knowledge point j, w i,j is the knowledge point k in the knowledge graph i With knowledge point k j The association weight is used to evaluate the association strength between knowledge points, (1-C j ) is the inverse relationship of knowledge point mastery, that is, the degree to which the learner has not mastered knowledge point j, F d is the learner’s learning fatigue, reflecting their current learning status, p is the learning preference feature, which indicates the learner's preference for specific content or question type. λ is the adjustment coefficient, which is used to control the impact of cognitive load on the generation order. v is the learner's current cognitive load value, indicating the learner's current learning pressure, μ is the learner's ability coefficient, which is used to adjust the difficulty of the knowledge point, D c is the complexity of the knowledge point, which is used to influence the generation order according to the difficulty of the knowledge point; S9. Provide learners with personalized homework generation solutions based on the updated behavioral feature vector set, knowledge point mastery model and cognitive load model, and optimize homework content and generation path in real time based on the dynamic association of different knowledge points in the knowledge graph.

2. The method for generating personalized tasks based on artificial intelligence algorithm according to claim 1, characterized in that: The S1 specifically includes: S11. Define the learner's multi-dimensional learning behavior data feature matrix M. Each data item in the multi-dimensional learning behavior data feature matrix contains the following dimensions: Study time T i Record the time learners spend on a specific knowledge point or homework task in seconds; Learning Progress i Indicates the learner's current learning path progress, calculated by the percentage of knowledge points completed, ranging from [0,100]; Test score S i Record the learner's score in the test related to the current knowledge point, expressed as a percentage, ranging from [0,100]; Learning Preference i is the learner's preference information in terms of content, resource form, etc., expressed in the form of a numerical vector; Knowledge points mastered K i The learner's mastery of a specific knowledge point ranges from [0,1], where 1 indicates complete mastery and 0 indicates no mastery; S12. Construct a multi-dimensional learning behavior data feature matrix M, where the elements of the multi-dimensional learning behavior data feature matrix M are the learner's multi-dimensional learning behavior data: Among them, n is the number of data items, which represents the total number of learning behavior data items of learners in different learning scenarios.

3. The method for generating personalized tasks based on artificial intelligence algorithm according to claim 1, characterized in that: The S2 specifically includes: S21. Constructing an interdisciplinary knowledge graph: G={G m ,G p ,…,G k }; Each interdisciplinary knowledge graph G i Corresponding to the knowledge point association structure of a subject, including multiple nodes and edges, the node represents the knowledge point set K in the subject i ={k i1 ,k i2 ,…,k im }, the edge represents the association between knowledge points, and each association has a weight value w i,j , w i,j Represents knowledge point k i With knowledge point k j The strength of association between disciplines, the association between disciplines is represented by interdisciplinary nodes, which are defined as common knowledge points or logically related knowledge points between two interdisciplinary knowledge graphs of different disciplines; S22, based on the learner's multi-dimensional learning behavior data feature matrix M of the knowledge point mastery data K i , generate the learner's subject knowledge point mastery state vector V i : V i ={v i1 ,v i2 ,…,v im }; Among them, v ij represents the learner's understanding of knowledge point k in subject i ij The degree of mastery is in the range of [0,1], where 0 means not mastered and 1 means fully mastered; S23, based on the learner's knowledge points, master the state vector V i and the interdisciplinary knowledge graph set G to construct the learner's interdisciplinary knowledge point mastery model M CK , interdisciplinary knowledge point mastery model M CK Including the mastery of knowledge points within the subject and the mastery of interdisciplinary related knowledge points: The state vector V of the learner's knowledge points in the interdisciplinary knowledge graph i , identify the set of knowledge points that have been mastered in the subject A collection of unlearned knowledge points And a collection of partially mastered knowledge points in: A collection of knowledge points that have been mastered θ a To master the threshold; A collection of unlearned knowledge points θ u The threshold is not mastered; A collection of partially mastered knowledge points Indicates the knowledge point between mastering and not mastering.

4. The method for generating personalized tasks based on artificial intelligence algorithm according to claim 1, characterized in that: The S3 specifically includes: S31, standardize the data of each dimension in the learner's learning behavior data feature matrix M to generate a standardized learning behavior feature matrix M', in which each data item includes a standardized value of learning time, learning progress, test scores, learning preferences and knowledge point mastery; S32, based on the standardized learning behavior feature matrix M', extracting the learner's learning habit features H including the learner's learning frequency, learning time period preference and learning duration; S33, based on the mastery of knowledge points K i Generate the learner's knowledge point mastery level feature C, which is the learner's mastery distribution of each knowledge point; S34, calculate the learner's learning efficiency feature E, which is calculated by analyzing the learner's learning time T on each knowledge point. i With test score S i The ratio of generates the learning efficiency vector E; S35, based on learning preference L i Generate learners' problem-solving preference features P, which represent learners' preferences for different types of assignments or resources; S36, combining the generated learning habit feature vector H, knowledge point mastery vector C, learning efficiency vector E and problem-solving preference vector P to form the learner's behavior feature vector set F: F={H,C,E,P}.

5. The method for generating personalized tasks based on artificial intelligence algorithm according to claim 1, characterized in that: The S4 specifically includes: S41, based on the behavior feature vector set F and the knowledge point mastering model M K , construct the learner's learning fatigue characteristic F d , learning fatigue feature F d By analyzing the learners' learning time distribution and knowledge mastery level: F d =α·T+β·(1-C); Among them, T is the learning time, C is the knowledge point mastery vector, α and β are weight coefficients; S42. Based on the relevance weight of knowledge points in the interdisciplinary knowledge graph G, combined with the complexity of knowledge points, calculate the learner's knowledge point complexity feature D c : D c =∑ i,j w i,j ·d j ; Among them, w i,j is the knowledge point k in the interdisciplinary knowledge graph i With k j The relevance weight, d j is knowledge point k j The complexity value of S43. Based on the learning efficiency feature E in the behavioral feature vector set, calculate the learner's problem-solving speed feature S r , problem-solving speed characteristic S r Defined as the inverse of the learning efficiency vector E: Among them, E is the learning efficiency feature, which is calculated by the problem-solving speed S r To reflect the learner's problem-solving speed; S44, based on the test score feature S in the behavior feature vector set, calculate the learner's accuracy feature A, where the accuracy feature A represents the learner's accuracy in the test questions, and the range is [0,1]; S45. Based on the learning fatigue characteristics F d , Knowledge point complexity feature D c , Problem-solving speed characteristics S r and accuracy feature A, construct the learner's cognitive load model L, and the cognitive load value L of the cognitive load model L v Defined as: L v =γ·F d +δ·D c +η·S r +θ·(1-A); Among them, γ, δ, η, and η are weight coefficients, which are used to adjust the weight of each feature in the calculation of cognitive load value; S46. Based on the calculated cognitive load value L v , the relevance weights of knowledge points in the interdisciplinary knowledge graph G are dynamically adjusted, the complexity of the relevance weights of knowledge points is reduced when the cognitive load value is high, and the complexity of the relevance weights of knowledge points is increased when the cognitive load value is low.

6. The method for generating personalized tasks based on artificial intelligence algorithm according to claim 1, characterized in that: The S5 specifically includes: S51, based on the behavior feature vector set F, knowledge point mastering model M K and the cognitive load value L of the cognitive load model L v , judge the learner's current cognitive load state, if the cognitive load value L v Above the preset threshold θ h , it is considered that the learner's cognitive load is high and enters the low-difficulty assignment generation process. If the cognitive load value L v Below the preset threshold θ l , it is considered that the learner’s cognitive load is low and enters the high-difficulty assignment generation process; S52, when the learner's cognitive load value L v ≥θ h When the learner's learning habit characteristics H and the knowledge point mastery vector C are used, a set of low-difficulty exercises Q related to basic knowledge points is generated. b : Q b ={q i ∈Q∣c i ≤α}; Among them, Q is the question bank, q i Indicates low-difficulty exercises, c i Representation and Exercises q i The mastery degree of relevant knowledge points, α is the mastery threshold of basic knowledge; S53, when the learner's cognitive load value L v ≤θ l When , according to the learning efficiency feature E and the knowledge point mastery vector C, a set of high-difficulty exercises Q related to advanced knowledge points is generated. a : Q a ={q i ∈Q∣c i >α}; S54, the generated low-difficulty exercise set Q b Or a collection of difficult exercises Q a Push it to the learner's personalized homework, and update the behavior feature vector set F and knowledge point mastery model M based on the learner's completion of each question, including completion time, accuracy and cognitive load value feedback data K ; S55. Based on the learner's homework completion data and feedback information, the generation threshold α and cognitive load threshold θ of low-difficulty exercises and high-difficulty exercises are adjusted in real time h With θ l .

7. The method for generating personalized tasks based on artificial intelligence algorithm according to claim 1, characterized in that: The S8 specifically includes: S81. Construct a long-term learning behavior model B based on the learner's long-term behavior data. The long-term learning behavior model B is generated by analyzing the learner's learning frequency, learning efficiency, and changes in knowledge points at different learning stages. S82, based on the learner's learning preference feature vector P and long-term learning behavior pattern B, predict the learner's demand probability for different types of homework content and generate a learning demand prediction vector D; S83. Master Model M by combining knowledge points K and the cognitive load change value ΔL of cognitive load model L v , predict the change trend of learners' future learning needs and generate a future learning needs prediction model F P ; S84. Prediction model based on future learning needs F P Plan the learner's personalized homework content and difficulty level in advance and determine the generation order vector S; S85. Based on the adaptive recommendation algorithm, personalized assignments are pushed according to the generated sequence vector S, and after each push, the long-term learning behavior pattern B and the learning demand prediction vector D are updated according to the learner's feedback data, so that the assignment generation sequence is dynamically matched with the learner's current and future needs.

Images