Learning interest nonlinear driving method based on Bell inequality optimization

Abstract

The invention discloses a Bell inequality optimization-based learning interest nonlinear driving method. The method comprises the following steps of S1, acquiring a learning behavior data set of a learner; s2, constructing a learning interest correlation analysis model based on the acquired learning behavior data set, and generating a learning interest correlation matrix; s3, establishing a learning interest distribution state model based on the dynamic entropy by using the learning interest incidence matrix; s4, generating an interest driving function matched with the learning interest dynamic characteristics of the learner; s5, constructing a dynamic adaptive learning path model according to the interest-driven function, and generating an optimal learning path and learning content recommendation scheme; and S6, outputting an optimal learning path and learning content recommendation scheme, continuously collecting a learning behavior data set of the learner, and repeatedly executing the steps S1-S5 to realize a loop mechanism for dynamically optimizing learning interest driving and learning path recommendation. According to the method, the driving parameters are optimized to adapt to the requirements of the learner, and the continuity and personalized expressive force of learning interest excitation are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of learning technology, and in particular to a nonlinear driving method for learning interest based on Bell inequality optimization. Background Art

[0002] With the rapid development of artificial intelligence and big data technologies, the field of education has begun to introduce intelligent technologies to improve teaching quality and learning efficiency. Learning interest, as an important factor driving students' learning motivation, is one of the core of personalized education research. Scientifically and rationally analyzing learners' interest distribution, optimizing learning paths and recommending content can not only improve learning effects, but also improve learners' initiative and learning experience. However, existing technical methods have significant deficiencies in learning interest modeling and dynamic optimization.

[0003] At present, most learning interest analysis and learning path optimization technologies rely on simple linear models or rule-based systems. They usually evaluate and recommend learning paths by statistical learning behavior data or using fixed interest weights. Static methods based on fixed rules are difficult to capture the dynamic changes of learning interests and cannot cope with the complex interest distribution characteristics of learners. Existing learning path optimization technologies usually rely on predefined knowledge point association rules and ignore the real-time changes of learners' interest in knowledge points during the learning process, resulting in the recommended learning paths and content being unable to accurately meet the learners' personalized needs, thus affecting learning efficiency.

[0004] In addition, some learning interest modeling methods based on machine learning have been applied in educational scenarios, using collaborative filtering technology or deep learning models to predict learning interests. Deep learning models often focus too much on the static characteristics of interest points and lack the ability to deeply model the dynamic changes of learning interests. When the distribution of learning interests is more complex, it is difficult to effectively quantify the learners' interest fluctuation characteristics.

[0005] Existing technologies also have deficiencies in the real-time adaptability of learning content recommendations. The current recommendation system is unable to dynamically adjust the recommended content according to learners' interest fluctuations in a timely manner, resulting in a deviation between the recommended content and learners' actual needs. This not only wastes learners' time but may also reduce their interest in learning and weaken learning outcomes.

[0006] To sum up, the existing technologies have obvious defects in the accuracy of learning interest modeling, the ability to dynamically optimize learning paths, and the real-time adaptability of recommended content. It is difficult to meet the needs of modern personalized education for intelligence and efficiency. New methods are urgently needed to solve these problems in order to achieve accurate analysis and dynamic optimization of learning interests, thereby improving learning experience and learning outcomes. Summary of the invention

[0007] One purpose of the present invention is to propose a nonlinear driving method for learning interest based on Bell's inequality optimization. The present invention significantly improves the sustainability and personalized expressiveness of learning interest motivation by optimizing driving parameters to adapt to learners' needs.

[0008] A nonlinear driving method for learning interest based on Bell inequality optimization according to an embodiment of the present invention comprises the following steps:

[0009] S1. Obtain the learner's learning behavior dataset;

[0010] S2. Building a learning interest correlation analysis model based on the acquired learning behavior data set. The learning interest correlation analysis model calculates the interest correlation between knowledge points based on Bell's inequality, performs conditional probability analysis on the learning behavior data of each pair of knowledge points, and generates a learning interest correlation matrix.

[0011] S3. Using the learning interest association matrix and combining it with the dynamic entropy analysis technology, the dynamic entropy value of the learning interest distribution is calculated, and a learning interest distribution state model is established based on the dynamic entropy value;

[0012] S4. Construct a nonlinear driving model of learning interest based on the learning interest association matrix and the learning interest distribution state model to generate an interest driving function that matches the dynamic characteristics of the learner's learning interest;

[0013] S5. Build a dynamic adaptive learning path model based on the interest-driven function to generate the optimal learning path and learning content recommendation plan;

[0014] S6. Output the optimal learning path and learning content recommendation plan, continuously collect learners' learning behavior data sets, and repeat steps S1-S5 to realize a cyclic mechanism of dynamically optimizing learning interest drive and learning path recommendation.

[0015] Optionally, the construction of the learning behavior data set D includes collecting the learner's learning time data T, learning content data C, knowledge point learning feedback data R, and relevant background information data B of the learning behavior, to establish the learner's learning behavior data set D:

[0016] D={(T i ,C i ,R i ,B i )|i∈[1,n]};

[0017] Among them, T i represents the time data of the i-th learning behavior, including learning duration and learning start time, C i represents the learning content involved in the i-th learning behavior, R irepresents the learning feedback of the knowledge point of the i-th learning behavior, specifically the learner's evaluation of the mastery of the relevant knowledge point, B i It represents the background information of the i-th learning behavior, including the learning scene, the equipment used, and other external conditions, and n is the total number of learning behaviors.

[0018] Optionally, S2 includes the following steps:

[0019] S21. Extract the knowledge point set K using the learning behavior dataset D, where K represents the set of all knowledge points involved in the learner's learning behavior:

[0020] K={K1,K2,…,K m};

[0021] Among them, m is the total number of knowledge points, K j Represents the jth knowledge point.

[0022] S22. For each pair of knowledge points K p And knowledge point K q The learning behavior data is analyzed by conditional probability to calculate the learner's probability of understanding the knowledge point K. p Is the interest affected by the knowledge point K q The interest influence is defined as the conditional probability P(K p |K q )for:

[0023]

[0024] Among them, P(K p |K q ) is used to measure the knowledge point K q With K p The relevance of interests between them;

[0025] S23, based on the conditional probability P(K p |K q ) Use Bell's inequality to verify whether there is a significant nonlinear interest association between knowledge points:

[0026] |P(K p |K q )-P(K p |K r )|-P(K q |K r )≤Δ,Δ>0;

[0027] Among them, P(K r |K q ) indicates that the learner starts from knowledge point K q Turn to knowledge point K rThe conditional probability of K is Δ, which is the maximum deviation allowed and is used to control the permissible range of interest association. When the inequality is violated, it indicates that the knowledge point K p and K q There is a significant nonlinear interest relationship between them;

[0028] S24, matrix processing is performed on the interest associations between all knowledge points to generate a learning interest association matrix A, where the matrix elements are defined as:

[0029]

[0030] Among them, a pq Represents knowledge point K p And knowledge point K q The strength of interest association between p and H q Represents knowledge point K p and K q The entropy value of the learning heat distribution is:

[0031]

[0032] Among them, β is the heat distribution weight parameter, which controls the influence of heat difference on the association strength, δ(K p ∈C i .K) indicates that the learner is in task C i Does it involve knowledge point K? p , if involved, the value is 1, otherwise 0;

[0033] Learning interest association matrix A = [a pq ] m×m It indicates the strength of learners' interest association between knowledge points and is used to comprehensively describe the interest association pattern between knowledge points.

[0034] Optionally, S3 includes the following steps:

[0035] S31, using the learning interest association matrix A to calculate the learner's interest distribution probability P(K) for each knowledge point in the knowledge point set K. p ):

[0036]

[0037] Among them, P(K p ) represents the learner's understanding of knowledge point K p The probability of interest distribution, a pq Represents knowledge point K p With knowledge point K q The interest association strength between them, m is the total number of knowledge points;

[0038] S32, based on the interest distribution probability P(K p ) Calculate the dynamic entropy value H of the knowledge point set K:

[0039]

[0040] The dynamic entropy value H is used to quantify the uniformity of learners' interest distribution in the knowledge point set K. When the H value is large, it means that the interest distribution is more uniform; when the H value is small, it means that the interest is concentrated on a few knowledge points.

[0041] S33. Take the dynamic entropy value H as the input of time series data and combine it with the learner's learning behavior data set D to build a learning interest distribution state model:

[0042] S(t)={H(t),T(t)};

[0043] Among them, S(t) represents the dynamic state of learning interest distribution, H(t) is the dynamic entropy value at time t, and T(t) is the learning behavior time data at time t;

[0044] S34. Define the learning interest change rate ΔH(t) based on the learning interest distribution state model S(t):

[0045] ΔH(t)=H(t+Δt)-H(t);

[0046] Among them, ΔH(t) is used to represent the changing trend of learners’ interest distribution between time t and time t+Δt;

[0047] If ΔH(t)>0, it indicates that the interest distribution tends to be uniform;

[0048] If ΔH(t)<0, it indicates that the interest concentration has increased.

[0049] Optionally, S4 includes the following steps:

[0050] S41. Use the learning interest association matrix A and the learning interest distribution state model S(t) to construct an initial learning interest nonlinear driving model:

[0051]

[0052] Among them, f(t) is the learning interest driving function at time t, α and β1 are weight parameters, which respectively control the influence of dynamic entropy value H(t) and interest association strength on learning interest driving, and w pq K is the knowledge point p and K q The weighting coefficient is used to reflect the weight importance of a specific association;

[0053] S42, optimize the initial learning interest driving model f(t), use the learner's learning interest change rate ΔH(t) as a constraint, and dynamically adjust the driving parameters α and β1:

[0054] α opt =α0+γ·ΔH(t);

[0055] β opt =β0-γ·ΔH(t);

[0056] Among them, α0 and β0 are the initial weight values, γ is the learning rate, which controls the influence of dynamic entropy changes on the adjustment of driving parameters; when ΔH(t)>0, the contribution of dynamic entropy to learning interest driving is increased; when ΔH(t)<0, the contribution of dynamic entropy is reduced to strengthen interest concentration;

[0057] S43, based on the optimized driving parameter α opt and β opt Generate dynamic learning interest curve f opt (t):

[0058]

[0059] Among them, f opt (t) It is used to dynamically characterize the nonlinear changing characteristics of learners’ learning interests over time;

[0060] S44. Using dynamic learning interest curves opt (t) Generate the final learning interest driving function g(K p ,t), which is used to reflect the learner’s understanding of knowledge point K at time t p Intensity of interest:

[0061] g(K p ,t)=f opt (t)·P(K p );

[0062] Among them, P(K p ) is the knowledge point K p The probability of interest distribution is combined with the dynamic learning interest driving function f opt (t) Adjust the nonlinear incentive intensity of learning interest.

[0063] Optionally, S5 includes the following steps:

[0064] S51, based on learning interest driving function g(K p ,t) and the learning interest association matrix A to construct a dynamic adaptive learning path model. Let the learning path of the dynamic adaptive learning path model be P(t). The goal of learning path construction is to maximize the learning interest driving function g(K p,t), combined with the interest association strength a between knowledge points pq Optimize path order:

[0065]

[0066] in, Represents the i-th knowledge point in the learning path;

[0067] S52. Analyze the real-time adaptability of the learning path P(t) by using the fluctuation trend of the learning interest curve, and define the learning interest fluctuation trend Δg(K p ,t) is:

[0068] Δg(K p ,t)=g(K p ,t+Δt)-g(K p ,t);

[0069] When Δg(K p ,t)>0, the current path is kept first; when Δg(K p ,t)<0, according to the interest association matrix A and the dynamic learning interest driving function f opt (t), dynamically adjust the order of knowledge points in the path or replace mismatched knowledge points;

[0070] S53. Design a path optimization strategy based on the learning interest fluctuation trend and the association matrix A. The path optimization strategy represents the matching degree of the learning path by defining the adaptation weight W(P(t)):

[0071]

[0072] Among them, W(P(t)) is used to measure the interest intensity matching and relevance of adjacent knowledge points in the learning path. Represents the knowledge points in the learning path and The interest association strength between them, n represents the total number of knowledge points in the learning path;

[0073] S54, according to the optimization strategy W(P(t)), generate the optimal learning path P in real time opt (t):

[0074]

[0075] Among them, the optimal path P opt (t) represents the learning path that maximizes the learning interest drive and the association between knowledge points at time t;

[0076] S55, based on the optimal learning path P opt (t) Combined with the interest intensity of knowledge points g(K p,t) Recommend learning content that matches the learning path, recommended content set C opt (t) is defined as:

[0077]

[0078] Among them, θ is the interest intensity threshold, which is used to filter low-interest content.

[0079] Optionally, the S6 includes the following steps:

[0080] S61. Output the optimal learning path P opt (t) and learning content recommendation scheme C opt (t), use it as the learning path and learning content of the current learning stage, and push it to learners;

[0081] S62, continuously collect learners' optimal learning path P opt (t) and learning content C opt (t) The learning behavior dataset D generated in the process new ;

[0082] S63, the newly collected learning behavior data set D new Fusion into the existing learning behavior dataset D, update the complete learning behavior dataset to D updated :

[0083] D updated =D∪D new ;

[0084] S64, based on the updated learning behavior dataset D updated Repeat steps S1 to S5;

[0085] S65, the learning path P generated each time during the repeated execution process opt (t) and content recommendation scheme C opt (t) Perform real-time comparison to ensure that the optimization direction meets the following criteria:

[0086] Learning interest driving function g(K p ,t) continues to increase with time t;

[0087] The dynamic entropy value H(t) of the learning interest distribution state model maintains reasonable fluctuations;

[0088] Learning Path opt (t) and content recommendation scheme C opt (t) Dynamically match learners’ interests and ability levels;

[0089] S66. The final optimization results are used as new learning paths and recommended solutions and are fed back to the learning system in a loop.

[0090] The beneficial effects of the present invention are:

[0091] (1) The present invention introduces Bell's inequality to perform nonlinear modeling of learners' interest associations in knowledge points, breaking through the limitations of traditional linear models. It can accurately capture the potential association characteristics between learners' different knowledge points. By calculating conditional probabilities and nonlinear association values ​​that violate Bell's inequality, the recognition of learning interest associations is significantly improved. The accuracy of association evaluation in complex interest distribution scenarios is improved by 23%, effectively solving the problem that traditional methods are difficult to capture dynamic changes in interests.

[0092] (2) The present invention adopts dynamic entropy analysis technology to quantify the distribution uniformity and change trend of learning interest, and combines time series data to establish a learning interest distribution state model. Through the real-time calculation of dynamic entropy values ​​and dynamic feedback of interest change rate, it realizes accurate description and prediction of learners' interest distribution, can sensitively capture the real-time fluctuation trend of learning interest, and adapt to learners' needs by optimizing driving parameters, which significantly improves the sustainability and personalized expression of learning interest motivation.

[0093] (3) The present invention constructs a dynamic adaptive learning path model based on the nonlinear driving function of learning interest and the learning interest association matrix. It realizes efficient dynamic optimization of the learning path by adjusting the path optimization strategy and content recommendation scheme in real time. Combined with the analysis of dynamic interest curves and interest fluctuation trends, it can accurately adjust the learning path and recommended content according to the learners' real-time changing interests, effectively avoiding the problem of disconnection between content and learning needs in traditional static recommendation systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] 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:

[0095] Figure 1 The present invention provides a flowchart of a nonlinear driving method for learning interest based on Bell inequality optimization. DETAILED DESCRIPTION

[0096] 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.

[0097] refer to Figure 1 , a nonlinear driving method for learning interest based on Bell inequality optimization, comprising the following steps:

[0098] S1. Obtain the learner's learning behavior dataset;

[0099] S2. Building a learning interest correlation analysis model based on the acquired learning behavior data set. The learning interest correlation analysis model calculates the interest correlation between knowledge points based on Bell's inequality, performs conditional probability analysis on the learning behavior data of each pair of knowledge points, and generates a learning interest correlation matrix.

[0100] S3. Using the learning interest association matrix and combining it with the dynamic entropy analysis technology, the dynamic entropy value of the learning interest distribution is calculated, and a learning interest distribution state model is established based on the dynamic entropy value;

[0101] S4. Construct a nonlinear driving model of learning interest based on the learning interest association matrix and the learning interest distribution state model to generate an interest driving function that matches the dynamic characteristics of the learner's learning interest;

[0102] S5. Build a dynamic adaptive learning path model based on the interest-driven function to generate the optimal learning path and learning content recommendation plan;

[0103] S6. Output the optimal learning path and learning content recommendation plan, continuously collect learners' learning behavior data sets, and repeat steps S1-S5 to realize a cyclic mechanism of dynamically optimizing learning interest drive and learning path recommendation.

[0104] In this embodiment, the construction of the learning behavior data set D includes collecting the learner's learning time data T, learning content data C, knowledge point learning feedback data R and relevant background information data B of the learning behavior, and establishing the learner's learning behavior data set D:

[0105] D={(T i ,C i ,R i ,B i )|i∈[1,n]};

[0106] Among them, T i represents the time data of the i-th learning behavior, including learning duration and learning start time, C i represents the learning content involved in the i-th learning behavior, R i represents the learning feedback of the knowledge point of the i-th learning behavior, specifically the learner's evaluation of the mastery of the relevant knowledge point, B i It represents the background information of the i-th learning behavior, including the learning scene, the equipment used, and other external conditions, and n is the total number of learning behaviors.

[0107] In this implementation, S2 includes the following steps:

[0108] S21. Extract the knowledge point set K using the learning behavior dataset D, where K represents the set of all knowledge points involved in the learner's learning behavior:

[0109] K={K1,K2,…,K m};

[0110] Among them, m is the total number of knowledge points, K j Represents the jth knowledge point.

[0111] S22. For each pair of knowledge points K p And knowledge point K q The learning behavior data is analyzed by conditional probability to calculate the learner's probability of understanding the knowledge point K. p Is the interest affected by the knowledge point K q The interest influence is defined as the conditional probability P(K p |K q )for:

[0112]

[0113] Among them, P(K p |K q ) is used to measure the knowledge point K q With K p The relevance of interests between them;

[0114] S23, based on the conditional probability P(K p |K q ) Use Bell's inequality to verify whether there is a significant nonlinear interest association between knowledge points:

[0115] |P(K p |K q )-P(K p |K r )|-P(K q |K r )≤Δ,Δ>0;

[0116] Among them, P(K r |K q ) indicates that the learner starts from knowledge point K q Turn to knowledge point K r The conditional probability of K is Δ, which is the maximum deviation allowed and is used to control the permissible range of interest association. When the inequality is violated, it indicates that the knowledge point K p and K q There is a significant nonlinear interest relationship between them;

[0117] S24, matrix processing is performed on the interest associations between all knowledge points to generate a learning interest association matrix A, where the matrix elements are defined as:

[0118]

[0119] Among them, apq Represents knowledge point K p And knowledge point K q The strength of interest association between p and H q Represents knowledge point K p and K q The entropy value of the learning heat distribution is:

[0120]

[0121] Among them, β is the heat distribution weight parameter, which controls the influence of heat difference on the association strength, δ(K p ∈C i .K) indicates that the learner is in task C i Does it involve knowledge point K? p , if involved, the value is 1, otherwise 0;

[0122] Learning interest association matrix A = [a pq ] m×m It indicates the strength of learners' interest association between knowledge points and is used to comprehensively describe the interest association pattern between knowledge points.

[0123] In this implementation, S3 includes the following steps:

[0124] S31, using the learning interest association matrix A to calculate the learner's interest distribution probability P(K) for each knowledge point in the knowledge point set K. p ):

[0125]

[0126] Among them, P(K p ) represents the learner's understanding of knowledge point K p The probability of interest distribution, a pq Represents knowledge point K p With knowledge point K q The interest association strength between them, m is the total number of knowledge points;

[0127] S32, based on the interest distribution probability P(K p ) Calculate the dynamic entropy value H of the knowledge point set K:

[0128]

[0129] The dynamic entropy value H is used to quantify the uniformity of learners' interest distribution in the knowledge point set K. When the H value is large, it means that the interest distribution is more uniform; when the H value is small, it means that the interest is concentrated on a few knowledge points.

[0130] S33. Take the dynamic entropy value H as the input of time series data and combine it with the learner's learning behavior data set D to build a learning interest distribution state model:

[0131] S(t)={H(t),T(t)};

[0132] Among them, S(t) represents the dynamic state of learning interest distribution, H(t) is the dynamic entropy value at time t, and T(t) is the learning behavior time data at time t;

[0133] S34. Define the learning interest change rate ΔH(t) based on the learning interest distribution state model S(t):

[0134] ΔH(t)=H(t+Δt)-H(t);

[0135] Among them, ΔH(t) is used to represent the changing trend of learners’ interest distribution between time t and time t+Δt;

[0136] If ΔH(t)>0, it indicates that the interest distribution tends to be uniform;

[0137] If ΔH(t)<0, it indicates that the interest concentration has increased.

[0138] In this implementation, S4 includes the following steps:

[0139] S41. Use the learning interest association matrix A and the learning interest distribution state model S(t) to construct an initial learning interest nonlinear driving model:

[0140]

[0141] Among them, f(t) is the learning interest driving function at time t, α and β1 are weight parameters, which respectively control the influence of dynamic entropy value H(t) and interest association strength on learning interest driving, and w pq K is the knowledge point p and K q The weighting coefficient is used to reflect the weight importance of a specific association;

[0142] S42, optimize the initial learning interest driving model f(t), use the learner's learning interest change rate ΔH(t) as a constraint, and dynamically adjust the driving parameters α and β1:

[0143] α opt =α0+γ·ΔH(t);

[0144] β opt =β0-γ·ΔH(t);

[0145] Among them, α0 and β0 are the initial weight values, γ is the learning rate, which controls the influence of dynamic entropy changes on the adjustment of driving parameters; when ΔH(t)>0, the contribution of dynamic entropy to learning interest driving is increased; when ΔH(t)<0, the contribution of dynamic entropy is reduced to strengthen interest concentration;

[0146] S43, based on the optimized driving parameter α opt and β opt Generate dynamic learning interest curve f opt (t):

[0147]

[0148] Among them, f opt (t) It is used to dynamically characterize the nonlinear changing characteristics of learners’ learning interests over time;

[0149] S44. Using dynamic learning interest curves opt (t) Generate the final learning interest driving function g(K p ,t), which is used to reflect the learner’s understanding of knowledge point K at time t p Intensity of interest:

[0150] g(K p ,t)=f opt (t)·P(K p );

[0151] Among them, P(K p ) is the knowledge point K p The probability of interest distribution is combined with the dynamic learning interest driving function f opt (t) Adjust the nonlinear incentive intensity of learning interest.

[0152] In this implementation, S5 includes the following steps:

[0153] S51, based on learning interest driving function g(K p ,t) and the learning interest association matrix A to construct a dynamic adaptive learning path model. Let the learning path of the dynamic adaptive learning path model be P(t). The goal of learning path construction is to maximize the learning interest driving function g(K p ,t), combined with the interest association strength a between knowledge points pq Optimize path order:

[0154]

[0155] in, Represents the i-th knowledge point in the learning path;

[0156] S52. Analyze the real-time adaptability of the learning path P(t) by using the fluctuation trend of the learning interest curve, and define the learning interest fluctuation trend Δg(K p ,t) is:

[0157] Δg(K p ,t)=g(K p ,t+Δt)-g(K p ,t);

[0158] When Δg(K p ,t)>0, the current path is kept first; when Δg(K p ,t)<0, according to the interest association matrix A and the dynamic learning interest driving function f opt (t), dynamically adjust the order of knowledge points in the path or replace mismatched knowledge points;

[0159] S53. Design a path optimization strategy based on the learning interest fluctuation trend and the association matrix A. The path optimization strategy represents the matching degree of the learning path by defining the adaptation weight W(P(t)):

[0160]

[0161] Among them, W(P(t)) is used to measure the interest intensity matching and relevance of adjacent knowledge points in the learning path. Represents the knowledge points in the learning path and The interest association strength between them, n represents the total number of knowledge points in the learning path;

[0162] S54, according to the optimization strategy W(P(t)), generate the optimal learning path P in real time opt (t):

[0163]

[0164] Among them, the optimal path P opt (t) represents the learning path that maximizes the learning interest drive and the association between knowledge points at time t;

[0165] S55, based on the optimal learning path P opt (t) Combined with the interest intensity of knowledge points g(K p ,t) Recommend learning content that matches the learning path, recommended content set C opt (t) is defined as:

[0166]

[0167] Among them, θ is the interest intensity threshold, which is used to filter low-interest content.

[0168] In this implementation, S6 includes the following steps:

[0169] S61. Output the optimal learning path P opt (t) and learning content recommendation scheme C opt (t), use it as the learning path and learning content of the current learning stage, and push it to learners;

[0170] S62, continuously collect learners' optimal learning path P opt (t) and learning content C opt (t) The learning behavior dataset D generated in the process new ;

[0171] S63, the newly collected learning behavior data set D new Fusion into the existing learning behavior dataset D, update the complete learning behavior dataset to D updated :

[0172] D updated =D∪D new ;

[0173] S64, based on the updated learning behavior dataset D updated Repeat steps S1 to S5;

[0174] S65, the learning path P generated each time during the repeated execution process opt (t) and content recommendation scheme C opt (t) Perform real-time comparison to ensure that the optimization direction meets the following criteria:

[0175] Learning interest driving function g(K p ,t) continues to increase with time t;

[0176] The dynamic entropy value H(t) of the learning interest distribution state model maintains reasonable fluctuations;

[0177] Learning Path opt (t) and content recommendation scheme C opt (t) Dynamically match learners’ interests and ability levels;

[0178] S66. The final optimization results are used as new learning paths and recommended solutions and are fed back to the learning system in a loop.

[0179] Embodiment 1:

[0180] In order to verify the feasibility and effectiveness of the present invention, the embodiment takes the mathematics competition tutoring scenario of a key urban high school as an example, and selects the learning behavior data of 20 students on the learning platform for test and analysis. This experiment lasts for 4 weeks and simulates the whole process of students' learning interest modeling, path optimization and content recommendation in the three knowledge modules of algebra, geometry and probability statistics.

[0181] In the first week of the experiment, a student (student number 001) systematically recorded his learning behavior data, including learning time, learning content and feedback scores, while studying the algebra module. In the learning record on November 6, 2024, student 001 studied the knowledge point "A1" for 45 minutes with a mastery feedback score of 85 / 100. He then studied the knowledge point "A2" for 20 minutes but the mastery feedback score was only 60 / 100. Through preliminary calculation of the learning interest association matrix, the system identified that the interest association strength between knowledge points "A1" and "A2" of student 001 was 0.72, which was higher than the default threshold of 0.5.

[0182] At 20:00 on the same night, the system automatically analyzed the interest distribution of student 001 and calculated its dynamic entropy value to be 1.45, indicating that his learning interest was evenly distributed in the algebra module. The system generated the initial learning path "A1→A2→A3" and pushed the task to the student. However, in the learning record at 8:15 am on November 7, the system detected that student 001's learning time dropped sharply to 10 minutes when learning "A2", and the interest-driven function value dropped to 0.35, while the interest-driven function value when learning "A3" rose to 0.82. According to the trend of dynamic entropy value changes, the system adjusted the learning path, and preferentially inserted the geometry module knowledge point "G1" with a higher correlation strength with "A3" into the path, generating a new learning path "A1→A3→G1→A2".

[0183] In the test on November 10, student 001 completed all recommended tasks in the algebra and geometry modules. The system re-evaluated his learning interest and found that his interest focus had shifted from the algebra module to the probability and statistics module. The dynamic entropy value dropped from 1.45 to 1.12, indicating that his interest was gradually concentrated. The system further optimized the learning path according to the rate of change of interest to generate the learning path "P1→P2→P4→P3", and recommended highly relevant statistics module tasks.

[0184] In the third week, the system monitored the learning path of student 001 in real time. At 17:30 on November 18, the student began to learn the probability statistics knowledge point "P1". After completion, the feedback test score was 72 / 100 and the learning time was 35 minutes. When learning "P4", the learning time was shortened to 8 minutes, and the mastery score was only 50 / 100. The system analysis found that when learning "P4", the interest-driven function value dropped to 0.22 and the correlation strength with the next knowledge point "P3" was only 0.28. Therefore, the system generated a new learning path to replace "P4" with the more relevant knowledge point "P5" and pushed a new content recommendation task.

[0185] Finally, after the fourth week of the experiment, the system summarized and analyzed the learning records of all students. The following Table 1 is the comparative data between the experimental group (using the method of the present invention) and the control group (using the traditional fixed path recommendation method):

[0186] Table 1 Comparison data between the method of the present invention and the traditional fixed path recommendation method

[0187]

[0188]

[0189] Taking student 001 as an example, the method of the present invention increases the time of interest retention from the average 95 minutes of the traditional method to 145 minutes, and the dynamic adjustment of the learning path increases the mastery rate of the algebra module to 98%. The final test score is 92 points, while the students who use the traditional method only score 84 points.

[0190] The present invention introduces Bell's inequality to perform nonlinear modeling of learners' knowledge point interest associations, breaking through the limitations of traditional linear models and being able to accurately capture learners' potential association characteristics between different knowledge points. By calculating conditional probabilities and nonlinear association values ​​that violate Bell's inequality, the recognition of learning interest associations is significantly improved, and the accuracy of association evaluation in complex interest distribution scenarios is improved by 23%, effectively solving the problem that traditional methods are difficult to capture dynamic changes in interests.

[0191] The present invention adopts dynamic entropy analysis technology to quantify the distribution uniformity and change trend of learning interest, and establishes a learning interest distribution state model in combination with time series data. Through the real-time calculation of dynamic entropy values ​​and dynamic feedback of interest change rate, it realizes accurate description and prediction of learners' interest distribution, can sensitively capture the real-time fluctuation trend of learning interest, and adapt to learners' needs by optimizing driving parameters, which significantly improves the sustainability and personalized expression of learning interest motivation.

[0192] The present invention constructs a dynamic adaptive learning path model based on the nonlinear driving function of learning interest and the learning interest association matrix, and realizes efficient dynamic optimization of the learning path by adjusting the path optimization strategy and content recommendation scheme in real time. Combined with the analysis of dynamic interest curve and interest fluctuation trend, the learning path and recommended content can be accurately adjusted according to the learner's real-time changing interests, effectively avoiding the problem of disconnection between content and learning needs in traditional static recommendation systems.

[0193] The above data and real cases show that the method of the present invention has significant advantages in dynamically adjusting learning paths, optimizing content recommendations, and improving the sustainability of learning interests and learning effects.

[0194] 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 nonlinear driving method for learning interest based on Bell inequality optimization, characterized in that: The steps include: S1. Obtain the learner's learning behavior dataset; S2. Building a learning interest correlation analysis model based on the acquired learning behavior data set. The learning interest correlation analysis model calculates the interest correlation between knowledge points based on Bell's inequality, performs conditional probability analysis on the learning behavior data of each pair of knowledge points, and generates a learning interest correlation matrix. S3, using the learning interest association matrix and combining the dynamic entropy analysis technology to calculate the dynamic entropy value of the learning interest distribution, and establishing a learning interest distribution state model based on the dynamic entropy value; S4. Construct a nonlinear driving model of learning interest based on the learning interest association matrix and the learning interest distribution state model to generate an interest driving function that matches the dynamic characteristics of the learner's learning interest; S5. Build a dynamic adaptive learning path model based on the interest-driven function to generate the optimal learning path and learning content recommendation plan; S6. Output the optimal learning path and learning content recommendation plan, continuously collect learners' learning behavior data sets, and repeat steps S1-S5 to realize a cyclic mechanism of dynamically optimizing learning interest drive and learning path recommendation.

2. The nonlinear driving method for learning interest based on Bell inequality optimization according to claim 1, characterized in that: The construction of the learning behavior data set D includes collecting the learner's learning time data T, learning content data C, knowledge point learning feedback data R and relevant background information data B of the learning behavior, and establishing the learner's learning behavior data set D: D={(T i ,C i ,R i ,B i )|i∈[1,n]}; Among them, T i represents the time data of the i-th learning behavior, including learning duration and learning start time, C i represents the learning content involved in the i-th learning behavior, R i represents the learning feedback of the knowledge point of the i-th learning behavior, specifically the learner's evaluation of the mastery of the relevant knowledge point, B i It represents the background information of the i-th learning behavior, including the learning scene, the equipment used, and other external conditions, and n is the total number of learning behaviors.

3. The nonlinear driving method for learning interest based on Bell inequality optimization according to claim 1, characterized in that: The S2 comprises the following steps: S21. Extract the knowledge point set K using the learning behavior dataset D, where K represents the set of all knowledge points involved in the learner's learning behavior: K={K1,K2,…,K m }; Among them, m is the total number of knowledge points, K j Represents the jth knowledge point. S22. For each pair of knowledge points K p And knowledge point K q The learning behavior data is analyzed by conditional probability to calculate the learner's probability of understanding the knowledge point K. p Is the interest affected by the knowledge point K q The interest influence is defined as the conditional probability P(K p |K q )for: Among them, P(K p |K q ) is used to measure the knowledge point K q With K p The relevance of interests between them; S23, based on the conditional probability P(K p |K q ) Use Bell's inequality to verify whether there is a significant nonlinear interest association between knowledge points: |P(K p |K q )-P(K p |K r )|-P(K q |K r )≤Δ,Δ>0; Among them, P(K r |K q ) indicates that the learner starts from knowledge point K q Turn to knowledge point K r The conditional probability of K is Δ, which is the maximum deviation allowed and is used to control the permissible range of interest association. When the inequality is violated, it indicates that the knowledge point K p and K q There is a significant nonlinear interest relationship between them; S24, matrix processing is performed on the interest associations between all knowledge points to generate a learning interest association matrix A, where the matrix elements are defined as: Among them, a pq Represents knowledge point K p And knowledge point K q The strength of interest association between p and H q Represents knowledge point K p and K q The entropy value of the learning heat distribution is: Among them, β is the heat distribution weight parameter, which controls the influence of heat difference on the association strength, δ(K p ∈C i .K) indicates that the learner is in task C i Does it involve knowledge point K? p , if involved, the value is 1, otherwise 0; Learning interest association matrix A = [a pq ] m×m It indicates the strength of learners' interest association between knowledge points and is used to comprehensively describe the interest association pattern between knowledge points.

4. The nonlinear driving method for learning interest based on Bell inequality optimization according to claim 1, characterized in that: The S3 comprises the following steps: S31, using the learning interest association matrix A to calculate the learner's interest distribution probability P(K) for each knowledge point in the knowledge point set K. p ): Among them, P(K p ) represents the learner's understanding of knowledge point K p The probability of interest distribution, a pq Represents knowledge point K p With knowledge point K q The interest association strength between them, m is the total number of knowledge points; S32, based on the interest distribution probability P(K p ) Calculate the dynamic entropy value H of the knowledge point set K: The dynamic entropy value H is used to quantify the uniformity of learners' interest distribution in the knowledge point set K. When the H value is large, it means that the interest distribution is more uniform; when the H value is small, it means that the interest is concentrated on a few knowledge points. S33. Take the dynamic entropy value H as the input of time series data and combine it with the learner's learning behavior data set D to build a learning interest distribution state model: S(t) = {H(t), T(t)}; Among them, S(t) represents the dynamic state of learning interest distribution, H(t) is the dynamic entropy value at time t, and T(t) is the learning behavior time data at time t; S34. Define the learning interest change rate ΔH(t) based on the learning interest distribution state model S(t): ΔH(t)=H(t+Δt)-H(t); Among them, ΔH(t) is used to represent the changing trend of learners’ interest distribution between time t and time t+Δt; If ΔH(t)>0, it indicates that the interest distribution tends to be uniform; If ΔH(t)<0, it indicates that the interest concentration increases.

5. The method for nonlinearly driving learning interest based on Bell inequality optimization according to claim 1, characterized in that: The S4 comprises the following steps: S41. Use the learning interest association matrix A and the learning interest distribution state model S(t) to construct an initial learning interest nonlinear driving model: Among them, f(t) is the learning interest driving function at time t, α and β1 are weight parameters, which respectively control the influence of dynamic entropy value H(t) and interest association strength on learning interest driving, and w pq K is the knowledge point p and K q The weighting coefficient is used to reflect the weight importance of a specific association; S42, optimize the initial learning interest driving model f(t), use the learner's learning interest change rate ΔH(t) as a constraint, and dynamically adjust the driving parameters α and β1: a opt =α0+γ·ΔH(t); b opt =β0-γ·ΔH(t); Among them, α0 and β0 are the initial weight values, γ is the learning rate, which controls the influence of dynamic entropy changes on the adjustment of driving parameters; when ΔH(t)>0, the contribution of dynamic entropy to learning interest driving is increased; when ΔH(t)<0, the contribution of dynamic entropy is reduced to strengthen interest concentration; S43, based on the optimized driving parameter α opt and β opt Generate dynamic learning interest curve f opt (t): Among them, f opt (t) It is used to dynamically characterize the nonlinear changing characteristics of learners’ learning interests over time; S44. Using dynamic learning interest curves opt (t) Generate the final learning interest driving function g(K p , t), which is used to reflect the learner’s understanding of knowledge point K at time t p Intensity of interest: g(K p ,t)=f opt (t)·P(K p ); Among them, P(K p ) is the knowledge point K p The probability of interest distribution is combined with the dynamic learning interest driving function f opt (t) Adjust the nonlinear incentive intensity of learning interest.

6. The method for nonlinearly driving learning interest based on Bell inequality optimization according to claim 1, characterized in that: The S5 comprises the following steps: S51, based on learning interest driving function g(K p , t) and the learning interest association matrix A to construct a dynamic adaptive learning path model. Let the learning path of the dynamic adaptive learning path model be P(t). The goal of learning path construction is to maximize the learning interest driving function g(K p , t), combined with the interest association strength a between knowledge points pq Optimize path order: in, Represents the i-th knowledge point in the learning path; S52. Analyze the real-time adaptability of the learning path P(t) by using the fluctuation trend of the learning interest curve, and define the learning interest fluctuation trend Δg(K p , t) is: Δg(K p ,t)=g(K p ,t+Δt)-g(K p ,t); When Δg(K p , t)>0, the current path is kept first; when Δg(K p , t)<0, according to the interest association matrix A and the dynamic learning interest driving function f opt (t), dynamically adjust the order of knowledge points in the path or replace mismatched knowledge points; S53. Design a path optimization strategy based on the learning interest fluctuation trend and the association matrix A. The path optimization strategy represents the matching degree of the learning path by defining the adaptation weight W(P(t)): Among them, W(P(t)) is used to measure the interest intensity matching and relevance of adjacent knowledge points in the learning path. Represents the knowledge points in the learning path and The interest association strength between them, n represents the total number of knowledge points in the learning path; S54, according to the optimization strategy W(P(t)), generate the optimal learning path P in real time opt (t): Among them, the optimal path P opt (t) represents the learning path that maximizes the learning interest drive and the association between knowledge points at time t; S55, based on the optimal learning path P opt (t) Combined with the interest intensity of knowledge points g(K p ,t) Recommend learning content that matches the learning path, recommended content set C opt (t) is defined as: Among them, θ is the interest intensity threshold, which is used to filter low-interest content.

7. The method for nonlinearly driving learning interest based on Bell inequality optimization according to claim 1, characterized in that: The S6 comprises the following steps: S61. Output the optimal learning path P opt (t) and learning content recommendation scheme C opt (t), use it as the learning path and learning content of the current learning stage, and push it to learners; S62, continuously collect learners' optimal learning path P opt (t) and learning content C opt (t) The learning behavior dataset D generated in the process new ; S63, the newly collected learning behavior data set D new Fusion into the existing learning behavior dataset D, update the complete learning behavior dataset to D updated : D updated =D∪D new ; S64, based on the updated learning behavior dataset D updated Repeat steps S1 to S5; S65, the learning path P generated each time during the repeated execution process opt (t) and content recommendation scheme C opt (t) Perform real-time comparison to ensure that the optimization direction meets the following criteria: Learning interest driving function g(K p ,t) continues to increase with time t; The dynamic entropy value H(t) of the learning interest distribution state model maintains reasonable fluctuations; Learning Path opt (t) and content recommendation scheme C opt (t) Dynamically match learners’ interests and ability levels; S66. The final optimization results are used as new learning paths and recommended solutions and are looped back into the learning system.

Images