Adaptive interdisciplinary learning path recommendation method based on deep learning

Abstract

The invention discloses a self-adaptive interdisciplinary learning path recommendation method based on deep learning. The method comprises the following steps: S1, constructing a historical learning data set; s2, constructing a knowledge point difficulty scoring model based on a deep learning algorithm; s3, generating a student learning ability portrait; s4, generating a preliminary interdisciplinary learning path according to the learning ability portrait of the student and the output of the knowledge point difficulty scoring model; s5, dynamically adjusting a knowledge point sequence and a difficulty level in the interdisciplinary learning path; s6, the system evaluates the performance of the students on each knowledge point in the learning process in real time through a knowledge point difficulty scoring model, and dynamically adjusts the difficulty level of the knowledge points; and S7, after the student completes the learning task, the system optimizes the recommendation strategy of the interdisciplinary learning path according to the learning performance and the learning feedback. According to the method, the actual learning ability of the student can be reflected, and personalized learning path recommendation which better conforms to the current state of the student is provided for the student through the portrait, so that the learning efficiency and effect are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of interdisciplinary learning paths, and in particular to an adaptive interdisciplinary learning path recommendation method based on deep learning. Background Art

[0002] With the rapid development of information technology and educational technology, personalized learning path recommendation systems have gradually become one of the important tools of modern education. Many learning path recommendation systems in existing technologies rely on static rules or recommendation models based on a single subject, and usually construct learning paths through students' test scores and class schedule data. However, such systems are usually limited to recommending knowledge points in a single subject and it is difficult to accurately respond to students' interdisciplinary learning needs. For example, in the field of science and engineering, there is a close knowledge connection between the disciplines of mathematics, physics, and computer science. When students study between these disciplines, the existing single-subject recommendation system is difficult to effectively capture the knowledge connection between disciplines and provide students with interdisciplinary learning paths.

[0003] In addition, the learning path recommendation system in the prior art usually relies on students' fixed learning progress or preset learning abilities for recommendation, and cannot be dynamically adjusted according to students' real-time learning progress and behavioral feedback. When students' mastery of knowledge points at a certain stage exceeds expectations or encounters learning bottlenecks, the system cannot respond in time and adjust the learning path. The static recommendation mechanism can easily lead to students encountering learning content that is too simple or too difficult during the learning process, which reduces learning efficiency and interest.

[0004] The existing technology also has another significant flaw, that is, the assessment of students' learning ability is relatively rough. It usually relies on only a few indicators such as test scores or homework completion, ignoring more details in learning behavior, such as answering speed, error rate, and multi-dimensional data of learning feedback. It is unable to comprehensively evaluate students' actual learning ability, resulting in inaccurate recommendation of learning paths. In addition, the existing technology fails to fully utilize deep learning technology to conduct detailed analysis of students' learning data, resulting in difficulty for learning paths to dynamically adapt to changes in students' individual needs.

[0005] To sum up, the existing technology has the following main shortcomings: first, it is unable to effectively provide students with cross-disciplinary learning path recommendations and cannot capture the knowledge associations between disciplines; second, the existing learning path recommendation system lacks the ability to dynamically adjust and cannot flexibly adjust the learning path based on students' real-time feedback and learning progress; finally, the existing system has fewer evaluation dimensions for students' learning ability and it is difficult to provide personalized learning path recommendations. Summary of the invention

[0006] One purpose of the present invention is to propose an adaptive interdisciplinary learning path recommendation method based on deep learning, which can more accurately reflect the students' actual learning ability and provide students with personalized learning path recommendations that are more in line with their current status through the portrait, thereby improving learning efficiency and effectiveness.

[0007] According to an embodiment of the present invention, a method for recommending an adaptive interdisciplinary learning path based on deep learning includes the following steps:

[0008] S1. Collect students' learning behavior data, learning history data, knowledge point mastery data, and cross-disciplinary knowledge point correlation data to construct a historical learning data set. Learning behavior data includes learning time, learning frequency, and answer accuracy. Learning history data includes students' completed learning tasks, mastered knowledge points, and their difficulty levels. Knowledge point mastery data includes students' mastery of knowledge points in various subjects and the correlation between knowledge points.

[0009] S2. Construct a knowledge point difficulty scoring model based on a deep learning algorithm. The knowledge point difficulty scoring model is trained according to the preset subject knowledge difficulty dimension and historical learning data set, and outputs the difficulty score of each knowledge point. The subject knowledge difficulty dimension includes the complexity of the knowledge point, the learner's past learning performance, and the subject relevance of the knowledge point;

[0010] S3. Establish a student learning ability assessment model by analyzing the collected historical learning data set. The student learning ability assessment model is based on the multi-dimensional data of students' mastery of different subject knowledge points, answering speed, error rate and learning feedback, and evaluates students' current learning ability status in real time to generate a student learning ability portrait;

[0011] S4. Generate a preliminary interdisciplinary learning path based on the student's learning ability profile and the output of the knowledge point difficulty scoring model. The interdisciplinary learning path is sorted according to the difficulty score of the knowledge point and the student's current mastery, so that the difficulty of the knowledge points in the interdisciplinary learning path gradually increases, and the subject correlation between the interdisciplinary learning paths maintains reasonable continuity and interdisciplinary connectivity;

[0012] S5. The system monitors students’ learning behavior data in real time during the learning process, and dynamically adjusts the order and difficulty level of knowledge points in the interdisciplinary learning path based on students’ learning progress and learning feedback information. When the system detects that students have a learning bottleneck in a certain knowledge point or have mastered a knowledge point beyond expectations, the system adjusts the difficulty of the subsequent knowledge points in the interdisciplinary learning path in real time;

[0013] S6. The system uses a knowledge point difficulty scoring model to evaluate students' performance on each knowledge point in real time during the learning process and dynamically adjust the difficulty level of the knowledge point. According to students' learning ability, learning progress and real-time feedback data, the system increases or decreases the difficulty of each knowledge point in the interdisciplinary learning path and maintains a reasonable degree of subject relevance between knowledge points.

[0014] S7. After students complete their learning tasks, the system optimizes the recommended strategy for interdisciplinary learning paths based on their learning performance and learning feedback, and quantitatively evaluates the learning effect through the knowledge point difficulty scoring model, including the test results of students' knowledge point mastery, knowledge point learning time and learning feedback, and optimizes subsequent interdisciplinary learning paths.

[0015] Optionally, S1 includes the following steps:

[0016] S11. Collect students’ learning behavior data D B :

[0017] D B ={T s ,F s ,C s};

[0018] Among them, T s Indicates the average learning time of students on each knowledge point, in hours; F s Indicates the student's daily learning frequency, in times; C s Indicates the correct rate of students' answers to this knowledge point, ranging from 0 to 1.

[0019] S12. Collect students’ learning history data H :

[0020] D H ={L c ,K m ,D k};

[0021] Among them, L c ={l1,l2,...,l n} indicates the learning tasks that students have completed, K m ={k1,k2,...,k m} represents the set of knowledge points that students have mastered, D k The difficulty of the corresponding knowledge point ranges from 1 to 10, with 1 representing the simplest and 10 representing the most complex;

[0022] S13. Collect students’ knowledge mastery data D P , including students' mastery of knowledge points in various subjects k, where P k Indicates the student's mastery of knowledge point k, a floating point number ranging from 0 to 1, where 0 indicates no mastery at all and 1 indicates complete mastery;

[0023] S14. Collect cross-disciplinary knowledge point related data D R , where the interdisciplinary knowledge point correlation matrix R is a two-dimensional matrix used to analyze the relationship between knowledge points in different disciplines. R(i,j) represents the relationship between knowledge points k i With knowledge point k j The degree of association between them is a floating point number ranging from 0 to 1, where 0 indicates no association and 1 indicates complete association;

[0024] S15. Based on learning behavior data D B , learning historical data D H , Knowledge point mastery data D P and knowledge point related data D R Constructing historical learning dataset D S :

[0025] D S ={D B ,D H ,D P ,D R}.

[0026] Optionally, S2 includes the following steps:

[0027] S21. Constructing the Difficulty Dimension Set D of Subject Knowledge D :

[0028] D D ={C k ,P s ,A k};

[0029] Among them, C k represents the complexity of knowledge point k, an integer ranging from 1 to 10, 1 represents the simplest and 10 represents the most complex; P s A represents the past learning performance of student s, including average score, correct answer rate and learning speed index; k Indicates the subject relevance of knowledge point k, a floating point number ranging from 0 to 1, where 0 indicates no relevance and 1 indicates complete relevance;

[0030] S22. Treat each knowledge point k as a quantum field φ k , the students’ learning behavior is used as the excitation of the quantum field to construct the Lagrangian of the field corresponding to the knowledge point k

[0031]

[0032] Among them, g μν is a metric tensor used to describe the space-time geometric structure of the knowledge point field. represents the partial derivative with respect to the space-time coordinate μ, φ k is the quantum field representing the knowledge point k, V(φ k ) is the potential function of knowledge point k, ξ is the coupling constant, describing the coupling between the field and the space curvature R, γ∈ ijk is the three-dimensional rotational symmetry coefficient of the knowledge point, which represents the relationship between the relevance of the knowledge point and its rotational invariance;

[0033] S23, define the potential function V(φ k )for:

[0034]

[0035] Among them, m k For the field φ k The effective quality of , represents the complexity C of knowledge point k k :

[0036] m k =α·C k ;

[0037] λ k For the field φ k The self-interaction strength represents the subject association degree A of knowledge point k k :

[0038] λ k =β·A k ;

[0039] Among them, α and β are parameters that control the complexity and relevance of knowledge points;

[0040] S24. Using deep learning algorithms to solve the Lagrangian The derived field equation of motion gives the field excited state ψ for each knowledge point k k :

[0041]

[0042] in, is the D'Alembert operator, which represents the propagation characteristics of the knowledge point field in space and time. The nonlinear term on the right reflects the coupling effect between knowledge points in different disciplines. ijk The term represents the rotational invariance and interaction between interdisciplinary knowledge points;

[0043] S25, the field excited state ψ k P with student s’ past learning performances Mapped to knowledge point difficulty score S k :

[0044]

[0045] Among them, μ k and σ k are the average mastery level and standard deviation of knowledge point k, respectively. The integral region Ω is the student’s historical learning data set. The integral term reflects the matching degree between the student’s learning behavior and the field excitation state, which determines the final knowledge point difficulty score S. k ;

[0046] S26. Using historical learning data D S and the subject knowledge difficulty dimension set D D Training knowledge point difficulty scoring model M D , optimize the knowledge point difficulty scoring model parameters by minimizing the loss function L, and the loss function is defined as:

[0047]

[0048] in, is the difficulty score of knowledge point k predicted by the knowledge point difficulty scoring model, and λ is the regularization coefficient.

[0049] Optionally, S3 includes the following steps:

[0050] S31. By analyzing the collected historical learning data set D S Constructing the dimension set D of students’ learning ability A :

[0051] D A = {G k ,V k ,E k ,F k};

[0052] Among them, G k V is the degree of students’ mastery of knowledge point k, a floating point number ranging from 0 to 1; k is the answering speed, which means the time it takes for students to complete questions related to knowledge point k, in seconds; E k is the error rate of knowledge point k, a floating point number ranging from 0 to 1; F k The learning feedback data of students on knowledge point k, including students’ subjective learning feedback and test results;

[0053] S32. Constructing a model for evaluating students’ learning ability A , the student learning ability evaluation model inputs the student’s historical learning dataset D S and learning ability dimension set DA , the student learning ability evaluation model outputs the student's learning ability status A in different subject knowledge points s ;

[0054] S33, by minimizing the loss function L A , the student learning ability assessment model M A For training, the loss function is defined as:

[0055]

[0056] Where N is the number of training samples, is the actual learning ability status of students, is the learning ability state predicted by the student learning ability evaluation model, λ is the regularization coefficient, and the integration area Ω is the student's historical learning data set;

[0057] S34. Using the student learning ability evaluation model M after training A Conduct real-time learning ability assessment on students and generate a learning ability profile of them A The learning ability portrait includes multi-dimensional data on students’ mastery of knowledge points in different subjects, answering speed, error rate, and learning feedback:

[0058] P A =M A (D S ,D A ).

[0059] Optionally, S4 includes the following steps:

[0060] S41. Based on the students' learning ability profile P A And the knowledge point difficulty scoring model M D The output results of the above method are used to obtain the students’ current mastery of knowledge points in different subjects and the difficulty score of knowledge points S k , and construct a preliminary set of interdisciplinary learning paths P L ;

[0061] S42. For the interdisciplinary learning path set P L Sort the knowledge points in the text according to the students' mastery of G k Perform preliminary sorting so that students can learn the knowledge points with lower difficulty first, and the difficulty score is S k Higher knowledge points are presented later in the path;

[0062] S43. Difficulty rating of knowledge points in interdisciplinary learning paths k Gradually increase to form a learning process from easy to difficult, by defining the path progressive function R(P L)Measure the rationality of interdisciplinary learning paths:

[0063]

[0064] Among them, when R(P L )≥0, it means that the difficulty of the knowledge points in the path increases reasonably;

[0065] S44, by analyzing the interdisciplinary knowledge point correlation matrix D R To maintain interdisciplinary connectivity between knowledge points in the interdisciplinary learning path, for adjacent knowledge points k in the path i and k i+1 , according to the correlation matrix R(k i ,k i+1 ) make them coherent in knowledge structure:

[0066] R(k i ,k i+1 ) ≥ τ;

[0067] Among them, τ is the minimum correlation threshold between interdisciplinary knowledge points;

[0068] S45. Generate optimized interdisciplinary learning paths

[0069] Optionally, S5 includes the following steps:

[0070] S51. During the learning process, the system monitors the students’ learning behavior data in real time:

[0071] S52. Based on students' learning progress, real-time evaluation of students' knowledge mastery A , by comparing the students' current knowledge mastery level G k and the system preset mastering threshold θ k , to determine whether students have a learning bottleneck or exceed their expected learning performance during the learning process. The rules are as follows:

[0072] ΔG k =G k -θ k ;

[0073] When ΔG k ≥δ1, it means that the student's mastery of knowledge point k exceeds expectations; when ΔG k When ≤δ2, it means that the student encounters a learning bottleneck in this knowledge point, where δ1 and δ2 are the preset over-expectation and bottleneck thresholds respectively;

[0074] S53. When the system detects that students have a learning bottleneck at a certain knowledge point, the system dynamically adjusts the difficulty of the knowledge points in the interdisciplinary learning path and reduces the difficulty score of subsequent knowledge points according to the degree of the learning bottleneck. k ;

[0075] S54. When the system detects that the student's mastery of a certain knowledge point exceeds expectations, the system increases the difficulty score of the knowledge point in the subsequent interdisciplinary learning path;

[0076] S55. Based on the real-time monitored learning data, the system continuously and dynamically adjusts the order and difficulty level of knowledge points in the interdisciplinary learning path, so that the student's interdisciplinary learning path always matches his or her current learning status, and the adjusted interdisciplinary learning path P′ L Available to students.

[0077] Optionally, S6 includes the following steps:

[0078] S61. The system evaluates the student's performance on each knowledge point in real time during the learning process based on the knowledge point difficulty scoring model, and constructs the student's current knowledge point performance set, which includes the student's mastery of the knowledge point, answering speed, error rate and learning feedback data;

[0079] S62, the system adjusts the difficulty of each knowledge point in the interdisciplinary learning path in real time according to the student's learning ability and learning progress. The system adjusts the difficulty of the knowledge point by analyzing the difference between the student's current performance set and the preset target performance set;

[0080] S63. The system determines whether it is necessary to increase or decrease the difficulty of knowledge points in the interdisciplinary learning path based on the students' real-time feedback data. If the student's performance on the current knowledge point exceeds expectations, the difficulty of the subsequent knowledge points will be increased; if the student encounters difficulties on the current knowledge point, the difficulty of the subsequent knowledge points will be reduced.

[0081] S64. When adjusting the difficulty of each knowledge point, the system analyzes the interdisciplinary knowledge point correlation matrix to make the correlation between adjacent knowledge points meet the minimum threshold;

[0082] S65. The system continuously and dynamically adjusts the difficulty of knowledge points based on students' learning performance and feedback, and updates the order and difficulty level of knowledge points in the interdisciplinary learning path in real time, so that the interdisciplinary learning path is always consistent with the students' learning status.

[0083] The beneficial effects of the present invention are:

[0084] (1) The present invention effectively solves the limitations of single-discipline path recommendation in the prior art by constructing a knowledge point difficulty scoring model based on deep learning. The system can dynamically generate interdisciplinary learning paths by analyzing the correlation between knowledge points in different disciplines, ensuring that students can establish reasonable knowledge point connections in the learning process between multiple disciplines. By introducing an interdisciplinary correlation matrix to ensure the coherence between knowledge points in different disciplines, and dynamically adjusting the order of knowledge point recommendations according to students' real-time learning situation, the information fragmentation problem of traditional methods in single-discipline learning recommendations is avoided, significantly improving the integrity and correlation of interdisciplinary learning.

[0085] (2) The present invention introduces a dynamic adjustment mechanism based on deep learning. The system can monitor students' learning behavior data in real time during the learning process, and automatically adjust the difficulty level and knowledge point sequence of the learning path through the knowledge point difficulty scoring model. When the system detects that the student performs well in a certain knowledge point or encounters a learning bottleneck, it can instantly adjust the difficulty and sequence of subsequent knowledge points, making the learning path more flexible and personalized. This effectively solves the problem of rigid learning paths and inability to respond to students' needs in real time in the prior art, and improves the adaptability and flexibility of students' learning paths.

[0086] (3) The present invention constructs a student learning ability assessment model based on multidimensional data. Combining the students' mastery level, learning speed, error rate and learning feedback multidimensional data system, the present invention can accurately assess the students' learning ability and generate a personalized learning ability portrait. This overcomes the defect of the prior art that relies on a single indicator to recommend learning paths. It can more accurately reflect the students' actual learning ability and provide students with personalized learning path recommendations that are more in line with their current status through the portrait, thereby improving learning efficiency and effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] 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:

[0088] Figure 1 A flowchart of a method for recommending an adaptive interdisciplinary learning path based on deep learning proposed by the present invention;

[0089] Figure 2 This is a schematic diagram of dynamic learning path adjustment in an adaptive interdisciplinary learning path recommendation method based on deep learning proposed by the present invention. DETAILED DESCRIPTION

[0090] 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.

[0091] refer to Figure 1-2 , an adaptive interdisciplinary learning path recommendation method based on deep learning, comprising the following steps:

[0092] S1. Collect students' learning behavior data, learning history data, knowledge point mastery data, and cross-disciplinary knowledge point correlation data to construct a historical learning data set. Learning behavior data includes learning time, learning frequency, and answer accuracy. Learning history data includes students' completed learning tasks, mastered knowledge points, and their difficulty levels. Knowledge point mastery data includes students' mastery of knowledge points in various subjects and the correlation between knowledge points.

[0093] S2. Construct a knowledge point difficulty scoring model based on a deep learning algorithm. The knowledge point difficulty scoring model is trained according to the preset subject knowledge difficulty dimension and historical learning data set, and outputs the difficulty score of each knowledge point. The subject knowledge difficulty dimension includes the complexity of the knowledge point, the learner's past learning performance, and the subject relevance of the knowledge point;

[0094] S3. Establish a student learning ability assessment model by analyzing the collected historical learning data set. The student learning ability assessment model is based on the multi-dimensional data of students' mastery of different subject knowledge points, answering speed, error rate and learning feedback, and evaluates students' current learning ability status in real time to generate a student learning ability portrait;

[0095] S4. Generate a preliminary interdisciplinary learning path based on the student's learning ability profile and the output of the knowledge point difficulty scoring model. The interdisciplinary learning path is sorted according to the difficulty score of the knowledge point and the student's current mastery, so that the difficulty of the knowledge points in the interdisciplinary learning path gradually increases, and the subject correlation between the interdisciplinary learning paths maintains reasonable continuity and interdisciplinary connectivity;

[0096] S5. The system monitors students’ learning behavior data in real time during the learning process, and dynamically adjusts the order and difficulty level of knowledge points in the interdisciplinary learning path based on students’ learning progress and learning feedback information. When the system detects that students have a learning bottleneck in a certain knowledge point or have mastered a knowledge point beyond expectations, the system adjusts the difficulty of the subsequent knowledge points in the interdisciplinary learning path in real time;

[0097] S6. The system uses a knowledge point difficulty scoring model to evaluate students' performance on each knowledge point in real time during the learning process and dynamically adjust the difficulty level of the knowledge point. According to students' learning ability, learning progress and real-time feedback data, the system increases or decreases the difficulty of each knowledge point in the interdisciplinary learning path and maintains a reasonable degree of subject relevance between knowledge points.

[0098] S7. After students complete their learning tasks, the system optimizes the recommended strategy for interdisciplinary learning paths based on their learning performance and learning feedback, and quantitatively evaluates the learning effect through the knowledge point difficulty scoring model, including the test results of students' knowledge point mastery, knowledge point learning time and learning feedback, and optimizes subsequent interdisciplinary learning paths.

[0099] In this implementation, S1 includes the following steps:

[0100] S11. Collect students’ learning behavior data D B :

[0101] D B ={T s ,F s ,C s};

[0102] Among them, T s Indicates the average learning time of students on each knowledge point, in hours; F s Indicates the student's daily learning frequency, in times; C s Indicates the correct rate of students' answers to this knowledge point, ranging from 0 to 1.

[0103] S12. Collect students’ learning history data H :

[0104] D H ={L c ,K m ,D k};

[0105] Among them, L c ={l1,l2,...,l n} indicates the learning tasks that students have completed, K m ={k1,k2,...,k m} represents the set of knowledge points that students have mastered, D k The difficulty of the corresponding knowledge point ranges from 1 to 10, with 1 representing the simplest and 10 representing the most complex;

[0106] S13. Collect students’ knowledge mastery data D P , including students' mastery of knowledge points in various subjects k , where P k Indicates the student's mastery of knowledge point k, a floating point number ranging from 0 to 1, where 0 indicates no mastery at all and 1 indicates complete mastery;

[0107] S14. Collect cross-disciplinary knowledge point related data D R, where the interdisciplinary knowledge point correlation matrix R is a two-dimensional matrix used to analyze the relationship between knowledge points in different disciplines. R(i,j) represents the relationship between knowledge points k i With knowledge point k j The degree of association between them is a floating point number ranging from 0 to 1, where 0 indicates no association and 1 indicates complete association;

[0108] S15. Based on learning behavior data D B , learning historical data D H , Knowledge point mastery data D P and knowledge point related data D R Constructing historical learning dataset D S :

[0109] D S ={D B ,D H ,D P ,D R}.

[0110] In this implementation, S2 includes the following steps:

[0111] S21. Constructing the Difficulty Dimension Set D of Subject Knowledge D :

[0112] D D ={C k ,P s ,A k};

[0113] Among them, C k represents the complexity of knowledge point k, an integer ranging from 1 to 10, 1 represents the simplest and 10 represents the most complex; P s A represents the past learning performance of student s, including average score, correct answer rate and learning speed index; k Indicates the subject relevance of knowledge point k, a floating point number ranging from 0 to 1, where 0 indicates no relevance and 1 indicates complete relevance;

[0114] S22. Treat each knowledge point k as a quantum field φ k , the students’ learning behavior is used as the excitation of the quantum field to construct the Lagrangian of the field corresponding to the knowledge point k

[0115]

[0116] Among them, g μν is a metric tensor used to describe the space-time geometric structure of the knowledge point field. represents the partial derivative with respect to the space-time coordinate μ, φ kis the quantum field representing the knowledge point k, V(φ k ) is the potential function of knowledge point k, ξ is the coupling constant, describing the coupling between the field and the space curvature R, γ∈ ijk is the three-dimensional rotational symmetry coefficient of the knowledge point, which represents the relationship between the relevance of the knowledge point and its rotational invariance;

[0117] S23, define the potential function V(φ k )for:

[0118]

[0119] Among them, m k For the field φ k The effective quality of , represents the complexity C of knowledge point k k :

[0120] m k =α·C k ;

[0121] λ k For the field φ k The self-interaction strength represents the subject association degree A of knowledge point k k :

[0122] λ k =β·A k ;

[0123] Among them, α and β are parameters that control the complexity and relevance of knowledge points;

[0124] S24. Using deep learning algorithms to solve the Lagrangian The derived field equation of motion gives the field excited state ψ for each knowledge point k k :

[0125]

[0126] in, is the D'Alembert operator, which represents the propagation characteristics of the knowledge point field in space and time. The nonlinear term on the right reflects the coupling effect between knowledge points in different disciplines. ijk The term represents the rotational invariance and interaction between interdisciplinary knowledge points;

[0127] S25, the field excited state ψ k P with student s’ past learning performance s Mapped to knowledge point difficulty score S k :

[0128]

[0129] Among them, μ kand σ k are the average mastery level and standard deviation of knowledge point k, respectively. The integral region Ω is the student’s historical learning data set. The integral term reflects the matching degree between the student’s learning behavior and the field excitation state, which determines the final knowledge point difficulty score S. k ;

[0130] S26. Using historical learning data D S and the subject knowledge difficulty dimension set D D Training knowledge point difficulty scoring model M D , optimize the knowledge point difficulty scoring model parameters by minimizing the loss function L, and the loss function is defined as:

[0131]

[0132] in, is the difficulty score of knowledge point k predicted by the knowledge point difficulty scoring model, and λ is the regularization coefficient.

[0133] In this implementation, S3 includes the following steps:

[0134] S31. By analyzing the collected historical learning data set D S Constructing the dimension set D of students’ learning ability A :

[0135] D A = {G k ,V k ,E k ,F k};

[0136] Among them, G k V is the degree of students’ mastery of knowledge point k, a floating point number ranging from 0 to 1; k is the answering speed, which means the time it takes for students to complete questions related to knowledge point k, in seconds; E k is the error rate of knowledge point k, a floating point number ranging from 0 to 1; F k The learning feedback data of students on knowledge point k, including students’ subjective learning feedback and test results;

[0137] S32. Constructing a model for evaluating students’ learning ability A , the student learning ability evaluation model inputs the student’s historical learning dataset D S and learning ability dimension set D A , the student learning ability evaluation model outputs the student's learning ability status A in different subject knowledge points s ;

[0138] S33, by minimizing the loss function L A, the student learning ability assessment model M A For training, the loss function is defined as:

[0139]

[0140] Where N is the number of training samples, is the actual learning ability status of students, is the learning ability state predicted by the student learning ability evaluation model, λ is the regularization coefficient, and the integration area Ω is the student's historical learning data set;

[0141] S34. Using the student learning ability evaluation model M after training A Conduct real-time learning ability assessment on students and generate a learning ability profile of them A The learning ability portrait includes multi-dimensional data on students’ mastery of knowledge points in different subjects, answering speed, error rate, and learning feedback:

[0142] P A =M A (D S ,D A ).

[0143] In this implementation, S4 includes the following steps:

[0144] S41, based on the student's learning ability profile P A And the knowledge point difficulty scoring model M D The output results of the above method are used to obtain the students’ current mastery of knowledge points in different subjects and the difficulty score of knowledge points S k , and construct a preliminary set of interdisciplinary learning paths P L ;

[0145] S42. For the interdisciplinary learning path set P L Sort the knowledge points in the text according to the students' mastery of G k Perform preliminary sorting so that students can learn the knowledge points with lower difficulty first, and the difficulty score is S k Higher knowledge points are presented later in the path;

[0146] S43. Difficulty rating of knowledge points in interdisciplinary learning paths k Gradually increase to form a learning process from easy to difficult, by defining the path progressive function R(P L )Measure the rationality of interdisciplinary learning paths:

[0147]

[0148] Among them, when R(P L)≥0, it means that the difficulty of the knowledge points in the path increases reasonably;

[0149] S44, by analyzing the interdisciplinary knowledge point correlation matrix D R To maintain interdisciplinary connectivity between knowledge points in the interdisciplinary learning path, for adjacent knowledge points k in the path i and k i+1 , according to the correlation matrix R(k i ,k i+1 ) make them coherent in knowledge structure:

[0150] R(k i ,k i+1 ) ≥ τ;

[0151] Among them, τ is the minimum correlation threshold between interdisciplinary knowledge points;

[0152] S45. Generate optimized interdisciplinary learning paths

[0153] In this implementation, S5 includes the following steps:

[0154] S51. During the learning process, the system monitors the students’ learning behavior data in real time:

[0155] S52. Based on students' learning progress, real-time evaluation of students' knowledge mastery A , by comparing the students' current knowledge mastery level G k and the system preset mastering threshold θ k , to determine whether students have a learning bottleneck or exceed their expected learning performance during the learning process. The rules are as follows:

[0156] ΔG k =G k -θ k ;

[0157] When ΔG k ≥δ1, it means that the student's mastery of knowledge point k exceeds expectations; when ΔG k When ≤δ2, it means that the student encounters a learning bottleneck in this knowledge point, where δ1 and δ2 are the preset over-expectation and bottleneck thresholds respectively;

[0158] S53. When the system detects that students have a learning bottleneck at a certain knowledge point, the system dynamically adjusts the difficulty of the knowledge points in the interdisciplinary learning path and reduces the difficulty score of subsequent knowledge points according to the degree of the learning bottleneck. k ;

[0159] S54. When the system detects that the student's mastery of a certain knowledge point exceeds expectations, the system increases the difficulty score of the knowledge point in the subsequent interdisciplinary learning path;

[0160] S55. Based on the real-time monitored learning data, the system continuously and dynamically adjusts the order and difficulty level of knowledge points in the interdisciplinary learning path, so that the student's interdisciplinary learning path always matches his or her current learning status, and the adjusted interdisciplinary learning path P′ L Available to students.

[0161] In this implementation, S6 includes the following steps:

[0162] S61. The system evaluates the student's performance on each knowledge point in real time during the learning process based on the knowledge point difficulty scoring model, and constructs the student's current knowledge point performance set, which includes the student's mastery of the knowledge point, answering speed, error rate and learning feedback data;

[0163] S62, the system adjusts the difficulty of each knowledge point in the interdisciplinary learning path in real time according to the student's learning ability and learning progress. The system adjusts the difficulty of the knowledge point by analyzing the difference between the student's current performance set and the preset target performance set;

[0164] S63. The system determines whether it is necessary to increase or decrease the difficulty of knowledge points in the interdisciplinary learning path based on the students' real-time feedback data. If the student's performance on the current knowledge point exceeds expectations, the difficulty of the subsequent knowledge points will be increased; if the student encounters difficulties on the current knowledge point, the difficulty of the subsequent knowledge points will be reduced.

[0165] S64. When adjusting the difficulty of each knowledge point, the system analyzes the interdisciplinary knowledge point correlation matrix to make the correlation between adjacent knowledge points meet the minimum threshold;

[0166] S65. The system continuously and dynamically adjusts the difficulty of knowledge points based on students' learning performance and feedback, and updates the order and difficulty level of knowledge points in the interdisciplinary learning path in real time, so that the interdisciplinary learning path is always consistent with the students' learning status.

[0167] Embodiment 1:

[0168] In this embodiment 1, the scene is set as an international school in City A. An international school in City A adopts the interdisciplinary adaptive learning path recommendation system based on deep learning of the present invention and applies it to the science students in the third year of high school. The student group is facing the sprint stage of the college entrance examination and needs to flexibly switch between mathematics, physics and chemistry, and efficiently master the core knowledge points of multiple subjects in a short time to solve complex interdisciplinary problems. The existing learning path recommendation system is inefficient in learning and has limited improvement in student performance due to the inability to dynamically adjust and integrate across disciplines. Therefore, the school decided to apply the method of the present invention to improve it.

[0169] On May 1, 2024, the system monitored and analyzed the learning behavior data of student A during use. Student A has a good foundation in physics and chemistry but a relatively weak foundation in mathematics. He encountered difficulties in the study of advanced mathematics calculus. At 3:30 pm on the same day, the system monitored that student A had a correct answer rate of only 20% in the five calculus-related questions he completed, which was significantly lower than the system's preset accuracy threshold of 60%. At the same time, the system recorded that student A's answering speed when calculating derivatives and integrals dropped significantly, and the average time increased to 3.2 minutes per question, while the normal speed should be within 1.5 minutes per question.

[0170] According to the data, the system generated a real-time report at 3:45 am on the same day, pointing out that Student A encountered an obvious learning bottleneck in calculus learning, especially in the calculation of multiple integrals and definite integrals. The system suggested that the system help him master the application background and break through the learning bottleneck by recommending the related content of kinematics in physics and calculus in mathematics.

[0171] On May 1, 2024, at 4 p.m., the system adjusted the mathematics content in Student A's learning path to calculus application content related to motion equations in physics based on Student A's real-time learning data. The system dynamically adjusted the interdisciplinary path to push mathematical knowledge points related to speed and acceleration calculation problems in physics to Student A. The adjusted mathematical questions are mainly based on basic derivative and integral calculations, aiming to help Student A understand the application background of calculus from the perspective of physical applications.

[0172] During the learning process, the system continuously monitored Student A's answers. At 4:30 p.m., the system detected that Student A's accuracy rate had increased to 60%, and the average time taken was reduced to 1.8 minutes per question. His mastery of knowledge points had improved significantly. At the same time, the system found that Student A had achieved better-than-expected learning results on kinematic problems in physics, with an accuracy rate of 85%.

[0173] In the subsequent learning process on May 2, the system adjusted Student A's learning path again to slightly increase the difficulty and recommended questions related to calculus in wave mechanics. Through these interdisciplinary knowledge point recommendations, Student A gradually mastered the concept of calculus and its application in physics.

[0174] To verify the effectiveness of the present invention, we compared the learning performance of student A before and after adopting the adaptive learning path recommendation system of the present invention, as well as the learning effect under the traditional system. The following Table 1 shows the specific learning data from May 1 to May 7:

[0175] Table 1 Comparison data of learning path recommendation systems

[0176]

[0177] In the above data, after adopting the learning path recommendation system of the present invention, student A's learning time was reduced by about 20%, and the accuracy of answering questions was significantly improved. The accuracy of mathematics subjects increased from 20% to 65%, while the improvement under the traditional system was only 20%. In addition, the system effectively helped student A improve his mastery of knowledge points in various subjects by dynamically adjusting the difficulty of knowledge points and interdisciplinary paths. The average time for answering questions was also shortened from the original 2.5 minutes / question to 1.8 minutes / question, and the learning efficiency was greatly improved.

[0178] By real-time monitoring of student A's learning behavior data and dynamically evaluating his learning performance based on a deep learning model, the system can quickly discover student A's bottleneck in mathematics learning and organically combine his mathematics knowledge with physical knowledge through the adjustment of interdisciplinary paths, thereby gradually improving the learning effect. The system can not only adjust the difficulty of knowledge points in real time, but also maintain a reasonable degree of subject relevance in interdisciplinary scenarios, making the learning path more flexible and adaptable. During a week of study, the system dynamically adjusted student A's learning path a total of 10 times, significantly improving the learning efficiency. Compared with traditional learning path recommendation systems, the system of the present invention shows higher learning effects and flexibility.

[0179] The present invention effectively solves the limitations of single subject path recommendation in the prior art by constructing a knowledge point difficulty scoring model based on deep learning. The system can dynamically generate interdisciplinary learning paths by analyzing the correlation between knowledge points in different subjects, ensuring that students can establish reasonable knowledge point connections in the learning process between multiple subjects. By introducing an interdisciplinary correlation matrix to ensure the coherence between knowledge points in different subjects and dynamically adjusting the order of knowledge point recommendations according to students' real-time learning situation, the information fragmentation problem of traditional methods in single subject learning recommendations is avoided, which significantly improves the integrity and correlation of interdisciplinary learning.

[0180] The present invention introduces a dynamic adjustment mechanism based on deep learning. The system can monitor students' learning behavior data in real time during the learning process, and automatically adjust the difficulty level and knowledge point sequence of the learning path through the knowledge point difficulty scoring model. When the system detects that the student performs well in a certain knowledge point or encounters a learning bottleneck, it can instantly adjust the difficulty and sequence of subsequent knowledge points, making the learning path more flexible and personalized, effectively solving the problems of rigid learning paths and inability to respond to students' needs in real time in the prior art, and improving the adaptability and flexibility of students' learning paths.

[0181] The present invention constructs a student learning ability assessment model based on multidimensional data. Combined with the students' mastery level, learning speed, error rate and learning feedback multidimensional data system, the system can accurately assess the students' learning ability and generate personalized learning ability portraits. It overcomes the defect of relying on a single indicator for learning path recommendation in the prior art, and can more accurately reflect the students' actual learning ability and provide students with personalized learning path recommendations that are more in line with their current status through the portrait, thereby improving learning efficiency and effectiveness.

[0182] 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. An adaptive interdisciplinary learning path recommendation method based on deep learning, characterized in that: The steps include: S1. Collect students’ learning behavior data, learning history data, knowledge point mastery data, and cross-disciplinary knowledge point association data to construct a historical learning data set; S2. Construct a knowledge point difficulty scoring model based on a deep learning algorithm. The knowledge point difficulty scoring model is trained according to the preset subject knowledge difficulty dimension and historical learning data set, and outputs the difficulty score of each knowledge point; S3. Establish a student learning ability assessment model by analyzing the collected historical learning data set, evaluate the student's current learning ability status in real time, and generate a student learning ability portrait; S4. Generate a preliminary interdisciplinary learning path based on the student's learning ability profile and the output of the knowledge point difficulty scoring model. The interdisciplinary learning path is sorted according to the difficulty score of the knowledge point and the student's current mastery; S5. The system monitors students’ learning behavior data in real time during the learning process, and dynamically adjusts the order and difficulty level of knowledge points in the interdisciplinary learning path based on students’ learning progress and learning feedback information. When the system detects that students have a learning bottleneck in a certain knowledge point or have mastered a knowledge point beyond expectations, the system adjusts the difficulty of the subsequent knowledge points in the interdisciplinary learning path in real time; S6. The system uses a knowledge point difficulty scoring model to evaluate students' performance on each knowledge point in real time during the learning process and dynamically adjust the difficulty level of the knowledge point. According to students' learning ability, learning progress and real-time feedback data, the system increases or decreases the difficulty of each knowledge point in the interdisciplinary learning path and maintains a reasonable degree of subject relevance between knowledge points. S7. After students complete their learning tasks, the system optimizes the recommended strategy for interdisciplinary learning paths based on their learning performance and learning feedback, quantitatively evaluates the learning effect through the knowledge point difficulty scoring model, and optimizes subsequent interdisciplinary learning paths.

2. The method for recommending an adaptive interdisciplinary learning path based on deep learning according to claim 1, characterized in that: The S1 comprises the following steps: S11. Collect students’ learning behavior data D B : D B ={T s ,F s ,C s }; Among them, T s Indicates the average learning time of students on each knowledge point, in hours; F s Indicates the student's daily learning frequency, in times; C s Indicates the correct rate of students' answers to this knowledge point, ranging from 0 to 1. S12. Collect students’ learning history data H : D H ={L c ,K m ,D k }; Among them, L c ={l1,l2,...,l n } indicates the learning tasks that students have completed, K m ={k1,k2,...,k m } represents the set of knowledge points that students have mastered, D k The difficulty of the corresponding knowledge point ranges from 1 to 10, with 1 representing the simplest and 10 representing the most complex; S13. Collect students’ knowledge mastery data D P , including students' mastery of knowledge points in various subjects k , where P k Indicates the student's mastery of knowledge point k, a floating point number ranging from 0 to 1, where 0 indicates no mastery at all and 1 indicates complete mastery; S14. Collect cross-disciplinary knowledge point related data D R , where the interdisciplinary knowledge point correlation matrix R is a two-dimensional matrix used to analyze the relationship between knowledge points in different disciplines. R(i,j) represents the relationship between knowledge points k i With knowledge point k j The degree of association between them is a floating point number ranging from 0 to 1, where 0 indicates no association and 1 indicates complete association; S15. Based on learning behavior data D B , learning historical data D H , Knowledge point mastery data D P and knowledge point related data D R Constructing historical learning dataset D S : D S ={D B ,D H ,D P ,D R }。 3. The method for recommending an adaptive interdisciplinary learning path based on deep learning according to claim 1, characterized in that: The S2 comprises the following steps: S21. Constructing the Difficulty Dimension Set D of Subject Knowledge D : D D ={C k ,P s ,A k }; Among them, C k represents the complexity of knowledge point k, an integer ranging from 1 to 10, 1 represents the simplest and 10 represents the most complex; P s A represents the past learning performance of student s, including average score, correct answer rate and learning speed index; k Indicates the subject relevance of knowledge point k, a floating point number ranging from 0 to 1, where 0 indicates no relevance and 1 indicates complete relevance; S22. Treat each knowledge point k as a quantum field φ k , the students’ learning behavior is used as the excitation of the quantum field to construct the Lagrangian of the field corresponding to the knowledge point k Among them, g μν is a metric tensor used to describe the space-time geometric structure of the knowledge point field. represents the partial derivative with respect to the space-time coordinate μ, φ k is the quantum field representing the knowledge point k, V(φ k ) is the potential function of knowledge point k, ξ is the coupling constant, describing the coupling between the field and the space curvature R, γ∈ ijk is the three-dimensional rotational symmetry coefficient of the knowledge point, which represents the relationship between the relevance of the knowledge point and its rotational invariance; S23. Define the potential function V(φ k )for: Among them, m k For the field φ k The effective quality of , represents the complexity C of knowledge point k k : m k =α·C k ; λ k For the field φ k The self-interaction strength represents the subject relevance A of knowledge point k k : l k =β·A k ; Among them, α and β are parameters that control the complexity and relevance of knowledge points; S24. Using deep learning algorithms to solve the Lagrangian The derived field equation of motion gives the field excited state ψ for each knowledge point k k : in, is the D'Alembert operator, which represents the propagation characteristics of the knowledge point field in space and time. The nonlinear term on the right reflects the coupling effect between knowledge points in different disciplines. ijk The term represents the rotational invariance and interaction between interdisciplinary knowledge points; S25, the field excited state ψ k P with student s’ past learning performance s Mapped to knowledge point difficulty score S k : Among them, μ k and σ k are the average mastery level and standard deviation of knowledge point k, respectively. The integral region Ω is the student’s historical learning data set. The integral term reflects the matching degree between the student’s learning behavior and the field excitation state, which determines the final knowledge point difficulty score S. k ; S26. Using historical learning data D S and the subject knowledge difficulty dimension set D D Training knowledge point difficulty scoring model M D , optimize the knowledge point difficulty scoring model parameters by minimizing the loss function L, and the loss function is defined as: in, is the difficulty score of knowledge point k predicted by the knowledge point difficulty scoring model, and λ is the regularization coefficient.

4. The method for recommending an adaptive interdisciplinary learning path based on deep learning according to claim 1, characterized in that: The S3 comprises the following steps: S31. By analyzing the collected historical learning data set D S Constructing the dimension set D of students’ learning ability A : D A ={G k ,V k ,E k ,F k }; Among them, G k V is the degree of students’ mastery of knowledge point k, a floating point number ranging from 0 to 1; k is the answering speed, which means the time it takes for students to complete questions related to knowledge point k, in seconds; E k is the error rate of knowledge point k, a floating point number ranging from 0 to 1; F k The learning feedback data of students on knowledge point k, including students’ subjective learning feedback and test results; S32. Constructing a model for evaluating students’ learning ability A , the student learning ability evaluation model inputs the student’s historical learning dataset D S and learning ability dimension set D A , the student learning ability evaluation model outputs the student's learning ability status A in different subject knowledge points s ; S33, by minimizing the loss function L A , the student learning ability assessment model M A For training, the loss function is defined as: Where N is the number of training samples, is the actual learning ability status of students, is the learning ability state predicted by the student learning ability evaluation model, λ is the regularization coefficient, and the integration area Ω is the student's historical learning data set; S34. Using the student learning ability evaluation model M after training A Conduct real-time learning ability assessment on students and generate a learning ability profile of them A The learning ability portrait includes multi-dimensional data on students’ mastery of knowledge points in different subjects, answering speed, error rate, and learning feedback: P A =M A (D S ,D A )。 5. The method for recommending an adaptive interdisciplinary learning path based on deep learning according to claim 1, characterized in that: The S4 comprises the following steps: S41. Based on the students' learning ability profile P A And the knowledge point difficulty scoring model M D The output results of the above method are used to obtain the students’ current mastery of knowledge points in different subjects and the difficulty score of knowledge points S k , and construct a preliminary set of interdisciplinary learning paths P L ; S42. For the interdisciplinary learning path set P L Sort the knowledge points in the text according to the students' mastery of G k Perform preliminary sorting so that students can learn the knowledge points with lower difficulty first, and the difficulty score is S k Higher knowledge points are presented later in the path; S43. Difficulty rating of knowledge points in interdisciplinary learning paths k Gradually increase to form a learning process from easy to difficult, by defining the path progressive function R(P L )Measure the rationality of interdisciplinary learning paths: Among them, when R(P L )≥0, it means that the difficulty of the knowledge points in the path increases reasonably; S44, by analyzing the interdisciplinary knowledge point correlation matrix D R To maintain interdisciplinary connectivity between knowledge points in the interdisciplinary learning path, for adjacent knowledge points k in the path i and k i+1 , according to the correlation matrix R(k i ,k i+1 ) make them coherent in knowledge structure: R(k i ,k i+1 )≥τ; Among them, τ is the minimum correlation threshold between interdisciplinary knowledge points; S45. Generate optimized interdisciplinary learning paths 6. The method for recommending an adaptive interdisciplinary learning path based on deep learning according to claim 1, characterized in that: The S5 comprises the following steps: S51. During the learning process, the system monitors the students’ learning behavior data in real time: S52. Based on students' learning progress, real-time evaluation of students' knowledge mastery A , by comparing the students' current knowledge mastery level G k and the system preset mastering threshold θ k , to determine whether students have a learning bottleneck or exceed their expected learning performance during the learning process. The rules are as follows: ΔG k =G k -θ k ; When ΔG k ≥δ1, it means that the student's mastery of knowledge point k exceeds expectations; when ΔG k When ≤δ2, it means that the student encounters a learning bottleneck in this knowledge point, where δ1 and δ2 are the preset over-expectation and bottleneck thresholds respectively; S53. When the system detects that students have a learning bottleneck at a certain knowledge point, the system dynamically adjusts the difficulty of the knowledge points in the interdisciplinary learning path and reduces the difficulty score of subsequent knowledge points according to the degree of the learning bottleneck. k ; S54. When the system detects that the student's mastery of a certain knowledge point exceeds expectations, the system increases the difficulty score of the knowledge point in the subsequent interdisciplinary learning path; S55. Based on the real-time monitored learning data, the system continuously and dynamically adjusts the order and difficulty level of knowledge points in the interdisciplinary learning path, so that the student's interdisciplinary learning path always matches his or her current learning status, and the adjusted interdisciplinary learning path P′ L Available to students.

7. The method for recommending an adaptive interdisciplinary learning path based on deep learning according to claim 1, characterized in that: The S6 comprises the following steps: S61. The system evaluates the student's performance on each knowledge point in real time during the learning process based on the knowledge point difficulty scoring model, and constructs the student's current knowledge point performance set, which includes the student's mastery of the knowledge point, answering speed, error rate and learning feedback data; S62, the system adjusts the difficulty of each knowledge point in the interdisciplinary learning path in real time according to the student's learning ability and learning progress. The system adjusts the difficulty of the knowledge point by analyzing the difference between the student's current performance set and the preset target performance set; S63. The system determines whether it is necessary to increase or decrease the difficulty of knowledge points in the interdisciplinary learning path based on the students' real-time feedback data. If the student's performance on the current knowledge point exceeds expectations, the difficulty of the subsequent knowledge points will be increased; if the student encounters difficulties on the current knowledge point, the difficulty of the subsequent knowledge points will be reduced. S64. When adjusting the difficulty of each knowledge point, the system analyzes the interdisciplinary knowledge point relevance matrix to make the relevance between adjacent knowledge points meet the minimum threshold; S65. The system continuously and dynamically adjusts the difficulty of knowledge points based on students' learning performance and feedback, and updates the order and difficulty level of knowledge points in the interdisciplinary learning path in real time, so that the interdisciplinary learning path is always consistent with the students' learning status.

Images