Personalized learning path intelligent planning system

Abstract

The invention belongs to the technical field of intelligent education, and discloses a personalized learning path intelligent planning system, which comprises a data acquisition module used for acquiring learning activity data, user behavior data and learning result data; the data processing module is used for preprocessing the collected learning activity data, user behavior data and learning achievement data to obtain a learning activity feature data set, a user behavior feature data set and a learning achievement feature data set; the learning fitness prediction module is used for fusing the learning activity feature data set, the user behavior feature data set and the learning achievement feature data set to obtain a comprehensive feature data set; a learning fitness prediction model is trained and constructed according to the comprehensive feature data set, and the learning fitness of the user is obtained through prediction of the learning fitness prediction model; each learner provides a personalized learning path and recommendation, so that learning better meets personal requirements, and the learning effect and efficiency of the learner are improved.

Description

Technical Field

[0001] This invention relates to the field of technology, and more specifically, to a personalized learning path intelligent planning system. Background Technology

[0002] Patent publication number CN118195854A discloses a method and system for personalized learning path planning based on learner profiles. This method focuses on designing personalized learning programs by analyzing multi-dimensional data such as learners' basic attributes, learning behaviors, and learning outcomes. It utilizes advanced machine learning and swarm intelligence algorithms to deeply analyze learner data and construct a comprehensive learner profile. Based on this profile, it plans the most suitable learning path for each learner's individual needs and goals, while optimizing the matching and recommendation of learning resources. Furthermore, the method integrates a dynamic adjustment and feedback mechanism, adjusting the learning plan in real time based on the learner's progress and feedback to ensure high efficiency and a personalized learning experience. This invention is applicable to various scenarios such as online education and vocational training, significantly improving learning efficiency and outcomes while providing learners with a more precise and satisfying learning experience.

[0003] Existing intelligent learning path planning systems mainly suffer from the following problems:

[0004] The failure to integrate datasets with different features resulted in missing a comprehensive perspective on learner behavior and outcomes, thus affecting the overall understanding of the learning process. The lack of consideration for centering and whitening may lead to correlations between data features affecting model stability, reducing predictive performance, and increasing the complexity of model training. The absence of independent component analysis may prevent the identification of potential learning patterns, making it difficult to understand learner needs. The failure to optimize the non-Gaussianity of independent components may result in insufficient discriminative power between features, thus reducing the model's adaptability to complex learning behaviors. Furthermore, the failure to select the most representative independent components, leading to excessively high dimensionality, may increase computational burden, slow down training, increase the risk of overfitting, and affect the model's generalization ability.

[0005] Failure to implement a global search strategy may cause the algorithm to get stuck in local optima, thus missing better hyperparameter combinations and affecting model performance; lack of a mechanism to dynamically adjust the search strategy may cause the search process to become rigid and unable to adapt to the optimization needs at different stages, increasing the time cost of tuning; improper temperature adjustment may prevent the model from converging effectively when it is close to the optimal solution, increasing the waste of computing resources and prolonging the tuning cycle; failure to limit the perturbation amplitude may cause the randomness of the search process to get out of control, increasing the instability of the results and reducing the reliability of the model output.

[0006] Failure to comprehensively consider homework completion rate, practice question accuracy, exam pass rate, and average answering time may result in an incomplete assessment of user learning outcomes, leading to inaccurate feedback. Furthermore, the lack of a comprehensive evaluation of learning outcomes may prevent the system from effectively recommending suitable learning paths for users, resulting in the ineffective use of learning resources and activities and reduced learning effectiveness.

[0007] In view of this, the present invention proposes a personalized learning path intelligent planning system to solve the above problems. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a personalized learning path intelligent planning system, comprising:

[0009] The data acquisition module is used to collect learning activity data, user behavior data, and learning outcome data.

[0010] The data processing module is used to preprocess the collected learning activity data, user behavior data, and learning outcome data to obtain the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0011] The learning fitness prediction module is used to fuse the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset to obtain a comprehensive feature dataset; a learning fitness prediction model is trained and constructed based on the comprehensive feature dataset, and the user's learning fitness is predicted by the learning fitness prediction model;

[0012] The learning adaptability assessment module is used to compare the predicted learning adaptability of users with the preset learning adaptability threshold of users to determine whether users need to change their learning path.

[0013] The learning path changing module allows users to continue learning on the current learning path if they do not need to change it; if they do need to change it, the intelligent learning management terminal will issue a change command to remind them to change the learning path in a timely manner.

[0014] The personalized learning recommendation module is used to build a model for evaluating the effect of changing learning paths, assess the learning effect of users after changing learning paths, and recommend personalized learning paths to users based on the learning effect.

[0015] Furthermore, the learning activity data includes the user's single learning duration, average answering time, total daily learning duration, number of learning sessions per day, and number of completed learning tasks; user behavior data includes the user's course click frequency, page dwell time, keyword search frequency, and number of learning exchanges; learning outcome data includes the user's homework completion rate, practice question accuracy rate, exam pass rate, and exam score.

[0016] Furthermore, the method for preprocessing the collected learning activity data, user behavior data, and learning outcome data to obtain the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset includes:

[0017] Density clustering algorithm is used to identify and remove outliers in learning activity data, user behavior data and learning outcome data to obtain learning activity feature dataset, user behavior feature dataset and learning outcome feature dataset.

[0018] The learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset are dimensionality reduced. The standard deviation of the dimensionality-reduced learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset is then normalized to convert them into a standard normal distribution, resulting in the normalized learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0019] Furthermore, the method for dimensionality reduction of the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset includes:

[0020] S41. Integrate the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset into a single dataset X, dataset X = [X...]. a ,X b ,X c ]∈R m×n ; where X a For the learning activity feature dataset; X b For user behavior feature datasets; X c Let R be the learning outcome feature dataset; R is the set of real numbers, indicating that the feature values ​​of the dataset belong to the range of real numbers; m is the number of the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset; n is the number of features in the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0021] S42. For each feature dataset, center it using a centering formula; the centering formula is: X′=X-μ; where X′ is the dataset after centering; μ is the average value of each feature;

[0022] The formula for calculating the average value of each feature is as follows: Among them, X i Let be the feature vector of the i-th sample;

[0023] S43. Calculate the covariance matrix of each centered dataset; the formula for calculating the covariance matrix is: Where C is the covariance matrix, representing the covariance among the features in the centered dataset; (X′) T This is the transpose of the centralized dataset X′;

[0024] S44. The covariance matrix is ​​whitened using a whitening formula to make the data features uncorrelated and the variance equal to 1; the whitening formula is: Z = DV T X′; where Z is the whitened covariance matrix; D is the whitening matrix, whose diagonal elements are the reciprocal square roots of the eigenvalues ​​of the covariance matrix; V is the eigenvector matrix, where the columns of V are the eigenvectors of the covariance matrix, and each eigenvector corresponds to an eigenvalue of the covariance matrix; V T This is the transpose of the eigenvector matrix;

[0025] S45. Use improved independent component analysis to further process the whitened covariance matrix Z to extract independent components; find a separation matrix such that the independent components are represented as: S = WZ; where S is the independent component matrix; W is the separation matrix used to transform Z into the independent component matrix S;

[0026] The independent component matrix is ​​defined as S = [s1, s2, ..., s...]. r ,...,s u ]; where s r Let s be the r-th independent component, where r is the independent component s. r The dimension index, r = 1,...,u; u is the total number of independent components, i.e., the number of columns in the independent component matrix;

[0027] Define an objective function that quantifies the non-Gaussianity of independent components and maximizes the objective function during the optimization process; the objective function is: Where J(W) is the objective optimization function; ney(s) r ) is the r-th independent component s r Negative entropy;

[0028] The r-th independent component s r Negative entropy ney(s) r ) is defined as: ney(s r )=H(s g )-H(s r ); where H(s) r ) is an independent component s r The entropy of H(s); g ) is an independent component s r The entropy of the Gaussian approximation; s g It is the g-th independent component;

[0029] The independent component sr The formula for calculating entropy is: H(s) r )=-∫p(s r )log p(s r )ds r ; where p(s) r ) is the r-th independent component s r Each value s in r The probability of ds r For independent component s r The increment of continuous values;

[0030] The independent component s is determined by the entropy range constraint formula. r The entropy is limited so that the entropy of each independent component does not exceed the entropy of the limited independent component; the formula for limiting the entropy range is: Among them, H lit (s r ) is the restricted independent component s r The entropy of dm(s) r ) is an independent component s r The dimension; A is a constant representing the degree of influence on entropy;

[0031] S46. Use an iterative method to update the separation matrix W using the separation matrix update formula until the preset number of iterations is reached, and then stop to obtain the final separation matrix.

[0032] The formula for updating the separation matrix is: Among them, W M+1 Let W be the updated separation matrix, representing the new separation matrix obtained after M+1 iterations; M η is the current separation matrix; η is the learning rate, used to control the magnitude of each update; For the separation matrix W M The objective function J(W) M The gradient of the separation matrix W; The gradient operator with respect to the separation matrix W;

[0033] S47. Select the first c independent components using the independent component selection formula, wherein the independent component selection formula is: Where, f(s) r ) is an independent component s r Selection priority; b W ξ is the number of times the separation matrix W is updated; ξ is the factor controlling the influence of the number of updates on the selection priority; τ is the factor controlling the influence of the dimension on the selection priority.

[0034] S48, calculate each independent component s r Selection priority f(s) rSort all independent components by selection priority to obtain a priority list; preset selection priority threshold, select independent components with priority greater than the selection priority threshold;

[0035] The first c selected independent components are combined into a new feature matrix S′, where the columns of S′ correspond to the selected independent components. The dataset X is mapped to the dimensionality reduction space using the separation matrix W to obtain the dimensionality reduction datasets of learning activity features, user behavior features, and learning outcome features.

[0036] Furthermore, the method for fusing the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset to obtain a comprehensive feature dataset includes:

[0037] The learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset are fused together using a weighted formula to form a comprehensive feature dataset; the learning activity feature dataset is denoted as E1, the user behavior feature dataset as E2, and the learning outcome feature dataset as E3;

[0038] The weighting formula is: UW=E1·δ1+E2·δ2+E3·δ3; where δ1 is the weight coefficient of the learning activity feature dataset; δ2 is the weight coefficient of the user behavior feature dataset; and δ3 is the weight coefficient of the learning outcome feature dataset.

[0039] Furthermore, the training method for the learning fitness prediction model includes:

[0040] The dataset is divided into training, validation, and test sets for training and evaluating model performance. A learning fitness prediction model is constructed, with each sample set being a subset of the dataset. Each sample set includes historical comprehensive feature datasets and the corresponding user's learning fitness. The learning fitness prediction model includes an input layer, hidden layers, and an output layer. The hidden layer uses the ReLU activation function.

[0041] The input layer of the learning fitness prediction model is used to input the historical comprehensive feature dataset, and the number of neurons in the input layer should match the number of features in the historical comprehensive feature dataset; the output layer of the model is used to output the user's learning fitness, and the number of neurons in the output layer corresponds to the number of prediction targets, with a single neuron outputting the predicted value; the learning fitness prediction model is a fully connected neural network model;

[0042] Mean squared error is used as the loss function to measure the error between the model's predicted values ​​and the actual values; the mean squared error loss function is: Where n′ is the number of datasets; Y i The actual value of data point i in the dataset; The predicted value for data point i in the dataset;

[0043] The model is trained using the training set data, and the loss function is minimized using the Adam optimizer. The model's performance is evaluated using the validation set, and the model's hyperparameters are tuned until the model's performance no longer improves or reaches the preset stopping condition.

[0044] The model's performance in the prediction task is evaluated using a test set. The current comprehensive feature dataset is input into the trained fitness prediction model to obtain the user's learning fitness.

[0045] Furthermore, the method for tuning the hyperparameters of the model includes:

[0046] S71. Define the initial temperature as T0 to control the search range in the initial stage; define the initial solution as z0, representing the current model's hyperparameter combination; define the standard cooling rate as α; define the objective function value as the model's accuracy on the validation set; preset the number of iterations at each temperature as L; and the termination temperature as T. f ;

[0047] S72. For each temperature T, perform the following steps:

[0048] The current solution is set to z, and the current iteration number is k. Based on the current solution z, a new solution is generated using the new solution generation formula: z′=z+β k ·rd(-1,1); where z′ is the newly generated solution; rd(-1,1) is a random number generated in the interval (-1,1); β k The perturbation magnitude is related to the current iteration number k and the standard cooling rate α;

[0049] The perturbation amplitude β, related to the current iteration number k and the standard cooling rate α, is constrained by the perturbation amplitude limit formula. k The disturbance amplitude is limited; the formula for limiting the disturbance amplitude is: Where, β k ′ represents the amplitude of the disturbance after limitation;

[0050] S73. Calculate the objective function value E′(z′) of the new solution z′. The objective function is defined as the accuracy of the model on the validation set. Calculate the energy difference ΔE′=E′(z′)-E′(z) between the objective function value E′(z′) of the new solution z′ and the objective function value E′(z) of the current solution z.

[0051] S74. If ΔE′<0, the new solution is judged as superior and the new solution z′ is accepted; if ΔE′≥0, then the probability is used to determine the superior solution. Accept the new solution; if rd(-1,1)≥P, then keep the current solution unchanged; if rd(-1,1)<P, then accept the new solution.

[0052] S75. Update the temperature T using the temperature update formula, which is: Where T′ is the updated temperature; α is the standard cooling rate; and θ is the accelerated cooling rate.

[0053] The accelerated cooling rate θ is dynamically adjusted using an accelerated cooling rate adjustment formula to achieve the optimal accelerated cooling rate; the accelerated cooling rate adjustment formula is as follows: Where θ′ is the adjusted accelerated cooling rate; K is the total number of iterations; and m is the rate factor that controls the decrease of θ with the number of iterations.

[0054] S76. When the updated temperature T′ is less than or equal to the termination temperature T f The iteration stops when the preset number of iterations L is reached; otherwise, it returns to S72 to continue the next round of iteration. After the iteration stops, the objective function value corresponding to the final solution is recorded. The hyperparameter combination at this time is the optimal hyperparameter combination.

[0055] Furthermore, the method of comparing the predicted user's learning adaptability with a preset user's learning adaptability threshold to determine whether the user needs to change their learning path includes:

[0056] If the predicted learning adaptability of a user is greater than or equal to the preset learning adaptability threshold for a user, then it is determined that the user does not need to change the learning path.

[0057] If the predicted learning fitness level of a user is less than the preset learning fitness threshold, then it is determined that the user needs to change their learning path.

[0058] Furthermore, the method for constructing the replacement effect evaluation model includes:

[0059] The learning effectiveness of users after changing their learning paths is evaluated by constructing a change effectiveness evaluation model. The change effectiveness evaluation model is as follows: Where F is the user's overall learning performance score; b is a bias term used to adjust the user's base score; A is the user's homework completion rate; B is the user's practice question accuracy rate; P is the user's exam pass rate; t is the user's average answering time; ω A To control the impact of users' assignment completion rate on their overall learning performance score; ω B To control the impact of users' practice question accuracy on their overall learning performance score; ω P To control the impact of users' exam pass rates on their overall learning performance scores; ω t To control the impact of users' average answering time on users' overall learning performance score; x is an integral variable.

[0060] Furthermore, the method for recommending personalized learning paths to users based on learning outcomes includes:

[0061] Personalized learning path recommendations are made for users based on their overall learning performance score. A threshold for the overall learning performance score is preset, and this threshold is further divided into a low overall learning performance score threshold (CV). S And the high score threshold of comprehensive learning effect CV L ;

[0062] The user's overall learning performance score F is compared with the low score threshold CV of the overall learning performance. S And the high score threshold of comprehensive learning effect CV L In comparison, if F < CV S If the user's overall learning performance score is low, then the overall learning performance score is determined to be low; if CV S ≤F<CV L If F > CV, then the user's overall learning performance score is determined to be moderate; L If so, the user's overall learning performance score is determined to be a high score;

[0063] For users with low scores, the intelligent learning management terminal automatically provides them with course resources including video tutorials, online practice question banks, and interactive learning games;

[0064] For users with average scores, the intelligent learning management terminal automatically provides them with opportunities including case studies, mock exams, and interaction with users with high scores.

[0065] For users with high scores, the intelligent learning management terminal automatically provides them with innovative courses and professional training, including expert lectures, cutting-edge technical documents, and learning seminars.

[0066] The technical effects and advantages of the personalized learning path intelligent planning system of the present invention are as follows:

[0067] This invention integrates different feature datasets into a single dataset, enabling a more comprehensive analysis of learners' behavior and outcomes, thus gaining deeper insights. Centering and whitening processes eliminate correlations between data features, improving model stability and performance. Extracting independent components through ICA identifies potential learning patterns, aiding in understanding the intrinsic relationships between different features. Optimizing the non-Gaussianity of independent components improves feature discriminative power and enhances the model's adaptability to complex learning behaviors. Selecting the most representative independent components reduces data dimensionality, which helps improve the training speed and accuracy of subsequent models and reduces the risk of overfitting.

[0068] Using simulated annealing, a global search can be performed in a wide hyperparameter space, avoiding getting trapped in local optima. Dynamically adjusting the search strategy (such as temperature and perturbation amplitude) based on the quality of the current solution makes the search process more flexible and adaptable to optimization needs at different stages. By accepting inferior solutions with probability, new solutions can be explored, increasing model diversity. The gradual decrease in temperature and the dynamic adjustment of the cooling rate make model convergence more efficient when approaching the optimal solution, reducing unnecessary computation. Introducing a constrained perturbation amplitude effectively controls randomness in the search process, avoiding disordered searching in the hyperparameter space and increasing the stability and reliability of the results.

[0069] By comprehensively considering homework completion rate, practice question accuracy, exam pass rate, and average answering time, the model can fully reflect the user's learning effectiveness and provide more accurate feedback. By evaluating learning effectiveness, the system can more effectively recommend learning paths suitable for users, ensuring the relevance and effectiveness of learning resources and activities, and helping to improve learners' learning outcomes and efficiency. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the structure of a personalized learning path intelligent planning system according to the present invention;

[0071] Figure 2 This is a schematic diagram of a personalized learning path intelligent planning method according to the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example 1

[0074] Please see Figure 1 As shown in the figure, this embodiment of a personalized learning path intelligent planning system includes:

[0075] The data acquisition module is used to collect learning activity data, user behavior data, and learning outcome data.

[0076] The data processing module is used to preprocess the collected learning activity data, user behavior data, and learning outcome data to obtain the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0077] The learning fitness prediction module is used to fuse the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset to obtain a comprehensive feature dataset; a learning fitness prediction model is trained and constructed based on the comprehensive feature dataset, and the user's learning fitness is predicted by the learning fitness prediction model;

[0078] The learning adaptability assessment module is used to compare the predicted learning adaptability of users with the preset learning adaptability threshold of users to determine whether users need to change their learning path.

[0079] The learning path changing module allows users to continue learning on the current learning path if they do not need to change it; if they do need to change it, the intelligent learning management terminal will issue a change command to remind them to change the learning path in a timely manner.

[0080] The personalized learning recommendation module is used to build a model for evaluating the effect of changing learning paths, assess the learning effect of users after changing learning paths, and recommend personalized learning paths to users based on the learning effect.

[0081] Learning activity data includes a user's single learning session duration, average answering time, total daily learning time, number of learning sessions per day, and number of completed learning tasks; user behavior data includes a user's course click frequency, page dwell time, keyword search frequency, and number of learning interactions; learning outcome data includes a user's homework completion rate, practice question accuracy rate, exam pass rate, and exam score.

[0082] Methods for preprocessing collected learning activity data, user behavior data, and learning outcome data to obtain learning activity feature datasets, user behavior feature datasets, and learning outcome feature datasets include:

[0083] Density clustering algorithm is used to identify and remove outliers in learning activity data, user behavior data and learning outcome data to obtain learning activity feature dataset, user behavior feature dataset and learning outcome feature dataset.

[0084] The learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset are dimensionality reduced. The standard deviation of the dimensionality-reduced learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset is then normalized to convert them into a standard normal distribution, resulting in the normalized learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0085] Methods for dimensionality reduction of learning activity feature datasets, user behavior feature datasets, and learning outcome feature datasets include:

[0086] S41. Integrate the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset into a single dataset X, dataset X = [X...]. a ,X b ,X c ]∈R m×n ; where X a For the learning activity feature dataset; X b For user behavior feature datasets; X c Let R be the learning outcome feature dataset; R is the set of real numbers, indicating that the feature values ​​of the dataset belong to the range of real numbers; m is the number of the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset; n is the number of features in the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0087] S42. For each feature dataset, center it using the centering formula: X′=X-μ; where X′ is the dataset after centering; μ is the average value of each feature.

[0088] The formula for calculating the average value of each feature is: Among them, X i Let be the feature vector of the i-th sample;

[0089] S43. Calculate the covariance matrix of each centered dataset; the formula for calculating the covariance matrix is: Where C is the covariance matrix, representing the covariance among the features in the centered dataset; (X′) T This is the transpose of the centralized dataset X′;

[0090] S44. Whiten the covariance matrix using the whitening formula to make the data features uncorrelated and with a variance of 1; the whitening formula is: Z = DV T X′; where Z is the whitened covariance matrix; D is the whitening matrix, whose diagonal elements are the reciprocal square roots of the eigenvalues ​​of the covariance matrix; V is the eigenvector matrix, where the columns of V are the eigenvectors of the covariance matrix, and each eigenvector corresponds to an eigenvalue of the covariance matrix; V T This is the transpose of the eigenvector matrix;

[0091] S45. Use improved independent component analysis to further process the whitened covariance matrix Z to extract independent components; find a separation matrix such that the independent components are represented as: S = WZ; where S is the independent component matrix; W is the separation matrix used to transform Z into the independent component matrix S;

[0092] The independent component matrix is ​​defined as S = [s1, s2, ..., s...].r ,...,s u ]; where s r Let s be the r-th independent component, where r is the independent component s. r The dimension index, r = 1,...,u; u is the total number of independent components, i.e., the number of columns in the independent component matrix;

[0093] Define an objective function that quantifies the non-Gaussianity of independent components and maximizes the objective function during the optimization process; the objective function is: Where J(W) is the objective optimization function; ney(s) r ) is the r-th independent component s r Negative entropy;

[0094] The r-th independent component s r Negative entropy ney(s) r ) is defined as: ney(s r )=H(s g )-H(s r ); where H(s) r ) is an independent component s r The entropy of H(s); g ) is an independent component s r The entropy of the Gaussian approximation; s g It is the g-th independent component;

[0095] The formula for calculating the entropy of the independent component sr is: H(s) r )=-∫p(s r )log p(s r )ds r ; where p(s) r ) is the r-th independent component s r Each value s in r The probability of ds r For independent component s r The increment of continuous values;

[0096] The independent component s is determined by the entropy range constraint formula. r The entropy is limited so that the entropy of each independent component does not exceed the entropy of the limited independent component; the formula for limiting the entropy range is: Among them, H lit (s r ) is the restricted independent component s r The entropy of dm(s) r ) is an independent component s r The dimension; A is a constant representing the degree of influence on entropy;

[0097] For example, the influence constant A on entropy is 0.5, the total number of independent components u is 4, and the number of independent components sr If the dimension is 5, then the restricted independent components s r entropy

[0098] S46. Use an iterative method to update the separation matrix W using the separation matrix update formula until the preset number of iterations is reached, and then stop to obtain the final separation matrix.

[0099] The formula for updating the separation matrix is: Among them, W M+1 Let W be the updated separation matrix, representing the new separation matrix obtained after M+1 iterations; M η is the current separation matrix; η is the learning rate, used to control the magnitude of each update; For the separation matrix W M The objective function J(W) M The gradient of the separation matrix W; The gradient operator with respect to the separation matrix W;

[0100] S47. Select the first c independent components using the independent component selection formula. The independent component selection formula is: Where, f(s) r ) is an independent component s r Selection priority; b W ξ is the number of times the separation matrix W is updated; ξ is the factor controlling the influence of the number of updates on the selection priority; τ is the factor controlling the influence of the dimension on the selection priority.

[0101] S48, calculate each independent component s r Selection priority f(s) r Sort all independent components by selection priority to obtain a priority list; preset selection priority threshold, select independent components with priority greater than the selection priority threshold;

[0102] The first c selected independent components are combined into a new feature matrix S′, where the columns of S′ correspond to the selected independent components. The dataset X is mapped to the dimensionality reduction space using the separation matrix W to obtain the dimensionality reduction datasets of learning activity features, user behavior features, and learning outcome features.

[0103] For example, suppose there are three independent components s1, s2, and s3. The negative entropy ney(s1) of the first independent component s1 is 0.8, the negative entropy ney(s2) of the second independent component s2 is 0.6, and the negative entropy ney(s3) of the third independent component s3 is 0.9. dm(s1) is 5, dm(s2) is 4, and dm(s3) is 6. Then calculate the selection priority for each independent component:

[0104] If the preset priority threshold is 0.6, then the independent components s1 and s3 will be retained.

[0105] Methods for fusing learning activity feature datasets, user behavior feature datasets, and learning outcome feature datasets to obtain a comprehensive feature dataset include:

[0106] The learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset are fused together using a weighted formula to form a comprehensive feature dataset; the learning activity feature dataset is denoted as E1, the user behavior feature dataset as E2, and the learning outcome feature dataset as E3;

[0107] The weighting formula is: UW=E1·δ1+E2·δ2+E3·δ3; where δ1 is the weight coefficient of the learning activity feature dataset; δ2 is the weight coefficient of the user behavior feature dataset; and δ3 is the weight coefficient of the learning outcome feature dataset.

[0108] Training methods for fitness prediction models include:

[0109] The dataset is divided into training, validation, and test sets for training and evaluating model performance. A learning fitness prediction model is constructed, with each sample set being a subset of the dataset. Each sample set includes historical comprehensive feature datasets and the corresponding user's learning fitness. The learning fitness prediction model includes an input layer, hidden layers, and an output layer. The hidden layer uses the ReLU activation function.

[0110] The input layer of the learning fitness prediction model is used to input the historical comprehensive feature dataset. The number of neurons in the input layer should match the number of features in the historical comprehensive feature dataset. The output layer of the model is used to output the user's learning fitness. The number of neurons in the output layer corresponds to the number of prediction targets. The predicted value is output through one neuron. The learning fitness prediction model is a fully connected neural network model.

[0111] Mean squared error is used as the loss function to measure the error between the model's predicted values ​​and the actual values; the mean squared error loss function is: Where n′ is the number of datasets; Y i The actual value of data point i in the dataset; The predicted value for data point i in the dataset;

[0112] The model is trained using the training set data, and the loss function is minimized using the Adam optimizer. The model's performance is evaluated using the validation set, and the model's hyperparameters are tuned until the model's performance no longer improves or reaches the preset stopping condition.

[0113] The model's performance in the prediction task is evaluated using a test set. The current comprehensive feature dataset is input into the trained fitness prediction model to obtain the user's learning fitness.

[0114] Methods for tuning the hyperparameters of a model include:

[0115] S71. Define the initial temperature as T0 to control the search range in the initial stage; define the initial solution as z0, representing the current model's hyperparameter combination; define the standard cooling rate as α; define the objective function value as the model's accuracy on the validation set; preset the number of iterations at each temperature as L; and the termination temperature as T. f ;

[0116] S72. For each temperature T, perform the following steps:

[0117] Let the current solution be z, and the current iteration number be k. Based on the current solution z, generate a new solution using the new solution generation formula: z′=z+β k ·rd(-1,1); where z′ is the newly generated solution; rd(-1,1) is a random number generated in the interval (-1,1); β k The perturbation magnitude is related to the current iteration number k and the standard cooling rate α;

[0118] The perturbation amplitude β, related to the current iteration number k and the standard cooling rate α, is constrained by the perturbation amplitude limit formula. k Limitations are imposed; the formula for limiting the disturbance amplitude is: Where, β k ′ represents the amplitude of the disturbance after limitation;

[0119] S73. Calculate the objective function value E′(z′) of the new solution z′. The objective function is defined as the accuracy of the model on the validation set. Calculate the energy difference ΔE′ = E′(z′) - E′(z′) between the objective function value E′(z′) of the new solution z′ and the objective function value E′(z) of the current solution z. )；

[0120] S74. If ΔE′<0, the new solution is judged as superior and the new solution z′ is accepted; if ΔE′≥0, then the probability is used to determine the superior solution. Accept the new solution; if rd(-1,1)≥P, then keep the current solution unchanged; if rd(-1,1)<P, then accept the new solution.

[0121] S75. Update the temperature T using the temperature update formula, which is: Where T′ is the updated temperature; α is the standard cooling rate; and θ is the accelerated cooling rate.

[0122] The accelerated cooling rate θ is dynamically adjusted using an accelerated cooling rate adjustment formula to achieve the optimal accelerated cooling rate; the accelerated cooling rate adjustment formula is as follows: Where θ′ is the adjusted accelerated cooling rate; K is the total number of iterations; and m is the rate factor that controls the decrease of θ with the number of iterations.

[0123] S76. When the updated temperature T′ is less than or equal to the termination temperature T f The iteration stops when the preset number of iterations L is reached; otherwise, it returns to S72 to continue the next round of iteration. After the iteration stops, the objective function value corresponding to the final solution is recorded. The hyperparameter combination at this time is the optimal hyperparameter combination.

[0124] For example, suppose the goal is to minimize the following quadratic function: E′(z)=(z-3) 2 The initial temperature T0 is 100, the standard cooling rate α is 0.95, the initial solution z0 is 0, the number of iterations L at each temperature is 100, and the termination temperature T f It is 0.1;

[0125] In the first iteration, when k = 0, the current solution z = 0, and the perturbation amplitude is limited in the formula. The random number rd(-1,1) = 0.5, the new solution z′ = 0.5, and the new objective function value E′(z′) = (0.5-3) 2 =7.75;

[0126] The original objective function value is E′(z) = (0-3). 2 =9; Objective function difference ΔE′ = 7.75 - 9 = -1.25; Since ΔE′ < 0, accept the new solution and update z = 0.5; Update temperature T′ = α·T = 0.95·100 = 95;

[0127] In the second iteration, when k=1, the current solution z=0.5, and the perturbation amplitude is limited in the formula. The random number rd(-1,1) = -0.3, the new solution z′ = 0.3375, and the new objective function value E′(z′) = (0.3375-3) 2 ≈7.72;

[0128] The original objective function value E′(z) = 7.75; the objective function difference ΔE′ = 7.72 - 7.75 = -0.03; since ΔE′ < 0, we accept the new solution and update z = 0.3375; the updated temperature T′ = α·T = 0.95·95 ≈ 90.25.

[0129] Methods for determining whether a user needs to change their learning path by comparing the predicted user's learning adaptability with the preset user learning adaptability threshold include:

[0130] If the predicted learning adaptability of a user is greater than or equal to the preset learning adaptability threshold for a user, then it is determined that the user does not need to change the learning path.

[0131] If the predicted learning fitness level of a user is less than the preset learning fitness threshold, then it is determined that the user needs to change their learning path.

[0132] The methods for constructing a new performance evaluation model include:

[0133] The learning effectiveness of users after changing their learning paths is evaluated by constructing a change effectiveness evaluation model. The change effectiveness evaluation model is as follows: Where F is the user's overall learning performance score; b is a bias term used to adjust the user's base score; A is the user's homework completion rate; B is the user's practice question accuracy rate; P is the user's exam pass rate; t is the user's average answering time; ω A To control the impact of users' assignment completion rate on their overall learning performance score; ω B To control the impact of users' practice question accuracy on their overall learning performance score; ω P To control the impact of users' exam pass rates on their overall learning performance scores; ω t To control the impact of users' average answering time on users' overall learning performance score; x is an integral variable used to represent the cumulative impact of assignment completion rate on overall learning performance score; dx is a differential symbol representing the change in integral variable x;

[0134] The numerator of the improved performance evaluation model now incorporates the nonlinear effects of each variable, making the evaluation of learning outcomes more accurate and reflecting the true learning process. The denominator controls the impact of the user's average answering time t on the overall score. As the average answering time increases, the value of the denominator also increases, thus reducing the overall score improvement and emphasizing the importance of quick answering.

[0135] For example, a user's learning data is as follows:

[0136] The user's homework completion rate is 80%, the user's practice question accuracy rate is 90%, the user's exam pass rate is 85%, the user's average test-taking time is 30 minutes, the bias term is 2.5, and the factor ω controlling for the user's homework completion rate on the user's overall learning performance score is determined. A The factor ω, which controls the impact of the user's accuracy rate on the user's overall learning outcome score, is set to 0.4. B The factor ω, which controls the impact of the user's exam pass rate on the user's overall learning outcome score, is set to 0.3. PThe factor ω, which controls the average time taken by users to answer questions, is set to 0.2 to influence their overall learning performance score. t If the score is 0.1, then the user's overall learning performance score is...

[0137] Methods for recommending personalized learning paths based on learning outcomes include:

[0138] Personalized learning path recommendations are made for users based on their overall learning performance score. A threshold for the overall learning performance score is preset, and this threshold is further divided into a low overall learning performance score threshold (CV). S And the high score threshold of comprehensive learning effect CV L ;

[0139] The user's overall learning performance score F is compared with the low score threshold CV of the overall learning performance. S And the high score threshold of comprehensive learning effect CV L In comparison, if F < CV S If the user's overall learning performance score is low, then the overall learning performance score is determined to be low; if CV S ≤F<CV L If F > CV, then the user's overall learning performance score is determined to be moderate; L If so, the user's overall learning performance score is determined to be a high score;

[0140] For users with low scores, the intelligent learning management terminal automatically provides them with course resources including video tutorials, online practice question banks, and interactive learning games;

[0141] For users with average scores, the intelligent learning management terminal automatically provides them with opportunities including case studies, mock exams, and interaction with users with high scores.

[0142] For users with high scores, the intelligent learning management terminal automatically provides them with innovative courses and professional training, including expert lectures, cutting-edge technical documents, and learning seminars.

[0143] The preset learning adaptation threshold for users is set by staff. The learning adaptation of different users is collected through the intelligent learning management terminal, and the average of the learning adaptation of multiple users is taken as the preset learning adaptation threshold for users; similarly, preset priority thresholds are set.

[0144] In this embodiment, by integrating different feature datasets into one dataset, learner behavior and outcomes can be analyzed more comprehensively, leading to deeper insights. Centering and whitening processes can eliminate correlations between data features, improving model stability and performance. Extracting independent components through ICA can identify potential learning patterns, helping to understand the intrinsic relationships between different features. Optimizing the non-Gaussianity of independent components can improve feature discriminativeness and enhance the model's adaptability to complex learning behaviors. By selecting the most representative independent components, data dimensionality is reduced, which helps improve the training speed and accuracy of subsequent models and reduces the risk of overfitting.

[0145] Using simulated annealing, a global search can be performed in a wide hyperparameter space, avoiding getting trapped in local optima. Dynamically adjusting the search strategy (such as temperature and perturbation amplitude) based on the quality of the current solution makes the search process more flexible and adaptable to optimization needs at different stages. By accepting inferior solutions with probability, new solutions can be explored, increasing model diversity. The gradual decrease in temperature and the dynamic adjustment of the cooling rate make model convergence more efficient when approaching the optimal solution, reducing unnecessary computation. Introducing a constrained perturbation amplitude effectively controls randomness in the search process, avoiding disordered searching in the hyperparameter space and increasing the stability and reliability of the results.

[0146] By comprehensively considering homework completion rate, practice question accuracy, exam pass rate, and average answering time, the model can fully reflect the user's learning effectiveness and provide more accurate feedback. By evaluating learning effectiveness, the system can more effectively recommend learning paths suitable for users, ensuring the relevance and effectiveness of learning resources and activities, and helping to improve learners' learning outcomes and efficiency.

[0147] Example 2

[0148] Please see Figure 2 As shown, parts not described in detail in this embodiment are described in Embodiment 1. A personalized learning path intelligent planning method is provided, including:

[0149] S1. Collect learning activity data, user behavior data, and learning outcome data;

[0150] S2. Preprocess the collected learning activity data, user behavior data, and learning outcome data to obtain the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

[0151] S3. Merge the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset to obtain a comprehensive feature dataset; train and construct a learning fitness prediction model based on the comprehensive feature dataset, and predict the user's learning fitness through the learning fitness prediction model;

[0152] S4. Compare the predicted learning adaptability of the user with the preset learning adaptability threshold of the user to determine whether the user needs to change the learning path.

[0153] S5. If the user does not need to change the learning path, the current learning path will continue; if the user needs to change the learning path, the intelligent learning management terminal will issue a change instruction to remind the user to change the learning path in a timely manner.

[0154] S6. Construct a learning path change effect evaluation model to evaluate the learning effect of users after changing their learning path, and recommend personalized learning paths to users based on the learning effect.

[0155] Since the electronic device described in this embodiment is the electronic device used to implement the personalized learning path intelligent planning system in this application embodiment, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the personalized learning path intelligent planning system described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. As long as those skilled in the art implement the personalized learning path intelligent planning system in this application embodiment, the electronic device used falls within the protection scope of this application.

[0156] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0157] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A personalized learning path intelligent planning system, characterized in that, include: The data acquisition module is used to collect learning activity data, user behavior data, and learning outcome data. The data processing module is used to preprocess the collected learning activity data, user behavior data, and learning outcome data to obtain the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset. The learning fitness prediction module is used to fuse the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset to obtain a comprehensive feature dataset; a learning fitness prediction model is trained and constructed based on the comprehensive feature dataset, and the user's learning fitness is predicted by the learning fitness prediction model; The learning adaptability assessment module is used to compare the predicted learning adaptability of users with the preset learning adaptability threshold of users to determine whether users need to change their learning path. The learning path changing module allows users to continue learning on the current learning path if they do not need to change their learning path. If a user needs to change their learning path, the intelligent learning management terminal will issue a change command to remind the user to change their learning path in a timely manner. The personalized learning recommendation module is used to build a model for evaluating the effect of changing learning paths, assess the learning effect of users after changing learning paths, and recommend personalized learning paths to users based on the learning effect.

2. The personalized learning path intelligent planning system according to claim 1, characterized in that, The learning activity data includes the user's single learning time, average answering time, total daily learning time, number of learning sessions per day, and number of completed learning tasks; user behavior data includes the user's course click frequency, page dwell time, keyword search frequency, and number of learning exchanges; learning outcome data includes the user's homework completion rate, practice question accuracy rate, exam pass rate, and exam score.

3. The personalized learning path intelligent planning system according to claim 2, characterized in that, The method for preprocessing the collected learning activity data, user behavior data, and learning outcome data to obtain the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset includes: Density clustering algorithm is used to identify and remove outliers in learning activity data, user behavior data and learning outcome data to obtain learning activity feature dataset, user behavior feature dataset and learning outcome feature dataset. The learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset are dimensionality reduced. The standard deviation of the dimensionality-reduced learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset is then normalized to convert them into a standard normal distribution, resulting in the normalized learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset.

4. The personalized learning path intelligent planning system according to claim 3, characterized in that, The method for dimensionality reduction processing of the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset includes: S41. Integrate the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset into a single dataset X, dataset X = [X...]. a ,X b ,X c ]∈R m×n ; where X a For the learning activity feature dataset; X b For user behavior feature datasets; X c Let R be the learning outcome feature dataset; R is the set of real numbers, indicating that the feature values ​​of the dataset belong to the range of real numbers; m is the number of the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset; n is the number of features in the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset. S42. For each feature dataset, center it using a centering formula; the centering formula is: X′=X-μ; where X′ is the dataset after centering; μ is the average value of each feature; The formula for calculating the average value of each feature is as follows: Among them, X i Let be the feature vector of the i-th sample; S43. Calculate the covariance matrix of each centered dataset; the formula for calculating the covariance matrix is: Where C is the covariance matrix, representing the covariance among the features in the centered dataset; (X′) T This is the transpose of the centralized dataset X′; S44. The covariance matrix is ​​whitened using a whitening formula to make the data features uncorrelated and the variance equal to 1; the whitening formula is: Z = DV T X′; where Z is the whitened covariance matrix; D is the whitening matrix, whose diagonal elements are the reciprocal square roots of the eigenvalues ​​of the covariance matrix; V is the eigenvector matrix, where the columns of V are the eigenvectors of the covariance matrix, and each eigenvector corresponds to an eigenvalue of the covariance matrix; V T This is the transpose of the eigenvector matrix; S45. Use improved independent component analysis to further process the whitened covariance matrix Z to extract independent components; find a separation matrix such that the independent components are represented as: S = WZ; where S is the independent component matrix; W is the separation matrix used to transform Z into the independent component matrix S; The independent component matrix is ​​defined as S = [s1, s2, ..., s...]. r ,...,s u ]; where s r Let s be the r-th independent component, where r is the independent component s. r The dimension index, r = 1,...,u; u is the total number of independent components, i.e., the number of columns in the independent component matrix; Define an objective function that quantifies the non-Gaussianity of independent components and maximizes the objective function during the optimization process; the objective function is: Where J(W) is the objective optimization function; ney(s) r ) is the r-th independent component s r Negative entropy; The r-th independent component s r Negative entropy ney(s) r ) is defined as: ney(s r )=H(s g )-H(s r ); where H(s) r ) is an independent component s r The entropy of H(s); g ) is an independent component s r The entropy of the Gaussian approximation; s g It is the g-th independent component; The entropy of the independent component sr is calculated using the formula: H(s) r )=-∫p(s r logp(s) r )ds r ; where p(s) r ) is the r-th independent component s r Each value s in r The probability of ds r For independent component s r The increment of continuous values; The independent component s is determined by the entropy range constraint formula. r The entropy is limited so that the entropy of each independent component does not exceed the entropy of the limited independent component; the formula for limiting the entropy range is: Among them, H lit (s r ) is the restricted independent component s r The entropy of dm(s) r ) is an independent component s r The dimension; A is a constant representing the degree of influence on entropy; S46. Use an iterative method to update the separation matrix W using the separation matrix update formula until the preset number of iterations is reached, and then stop to obtain the final separation matrix. The formula for updating the separation matrix is: W M+1 =W M +η▽ W J(W M ); where W M+1 Let W be the updated separation matrix, representing the new separation matrix obtained after M+1 iterations; M Here is the current separation matrix; η is the learning rate, used to control the magnitude of each update; η▽ W J(W M W is the separation matrix. M The objective function J(W) M The gradient of the separation matrix W; ▽ W The gradient operator with respect to the separation matrix W; S47. Select the first c independent components using the independent component selection formula, wherein the independent component selection formula is: Where, f(s) r ) is an independent component s r Selection priority; b W ξ is the number of times the separation matrix W is updated; ξ is the factor controlling the influence of the number of updates on the selection priority; τ is the factor controlling the influence of the dimension on the selection priority. S48, calculate each independent component s r Selection priority f(s) r Sort all independent components by selection priority to obtain a priority list; preset selection priority threshold, select independent components with priority greater than the selection priority threshold; The first c selected independent components are combined into a new feature matrix S′, where the columns of S′ correspond to the selected independent components. The dataset X is mapped to the dimensionality reduction space using the separation matrix W to obtain the dimensionality reduction datasets of learning activity features, user behavior features, and learning outcome features.

5. The personalized learning path intelligent planning system according to claim 4, characterized in that, The method for fusing the learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset to obtain a comprehensive feature dataset includes: The learning activity feature dataset, user behavior feature dataset, and learning outcome feature dataset are fused together using a weighted formula to form a comprehensive feature dataset; the learning activity feature dataset is denoted as E1, the user behavior feature dataset as E2, and the learning outcome feature dataset as E3; The weighting formula is: UW=E1·δ1+E2·δ2+E3·δ3; where δ1 is the weight coefficient of the learning activity feature dataset; δ2 is the weight coefficient of the user behavior feature dataset; and δ3 is the weight coefficient of the learning outcome feature dataset.

6. The personalized learning path intelligent planning system according to claim 5, characterized in that, The training method for the learning fitness prediction model includes: The dataset is divided into training, validation, and test sets for training and evaluating model performance. A learning fitness prediction model is constructed, with each sample set being a subset of the dataset. Each sample set includes historical comprehensive feature datasets and the corresponding user's learning fitness. The learning fitness prediction model includes an input layer, hidden layers, and an output layer. The hidden layer uses the ReLU activation function. The input layer of the learning fitness prediction model is used to input the historical comprehensive feature dataset, and the number of neurons in the input layer should match the number of features in the historical comprehensive feature dataset; the output layer of the model is used to output the user's learning fitness, and the number of neurons in the output layer corresponds to the number of prediction targets, with a single neuron outputting the predicted value; the learning fitness prediction model is a fully connected neural network model; The mean squared error is used as the loss function to measure the error between the model's predicted values ​​and the actual values; the mean squared error loss function is: Where n′ is the number of datasets; Y i The actual value of data point i in the dataset; The predicted value for data point i in the dataset; The model is trained using the training set data, and the loss function is minimized using the Adam optimizer. The model's performance is evaluated using the validation set, and the model's hyperparameters are tuned until the model's performance no longer improves or reaches the preset stopping condition. The model's performance in the prediction task is evaluated using a test set. The current comprehensive feature dataset is input into the trained fitness prediction model to obtain the user's learning fitness.

7. The personalized learning path intelligent planning system according to claim 6, characterized in that, The methods for tuning the hyperparameters of the model include: S71. Define the initial temperature as T0 to control the search range in the initial stage; define the initial solution as z0, representing the current model's hyperparameter combination; define the standard cooling rate as α; define the objective function value as the model's accuracy on the validation set; preset the number of iterations at each temperature as L; and the termination temperature as T. f ; S72. For each temperature T, perform the following steps: The current solution is set to z, and the current iteration number is k. Based on the current solution z, a new solution is generated using the new solution generation formula: z′=z+β k ·rd(-1,1); where z′ is the newly generated solution; rd(-1,1) is a random number generated in the interval (-1,1); β k The perturbation magnitude is related to the current iteration number k and the standard cooling rate α; The perturbation amplitude β, related to the current iteration number k and the standard cooling rate α, is constrained by the perturbation amplitude limit formula. k The disturbance amplitude is limited; the formula for limiting the disturbance amplitude is: Where, β k ′ represents the amplitude of the disturbance after limitation; S73. Calculate the objective function value E′(z′) of the new solution z′. The objective function is defined as the accuracy of the model on the validation set. Calculate the energy difference ΔE′=E′(z′)-E′(z) between the objective function value E′(z′) of the new solution z′ and the objective function value E′(z) of the current solution z. S74. If ΔE′<0, the new solution is judged as superior and the new solution z′ is accepted; if ΔE′≥0, then the probability is used to determine the superior solution. Accept the new solution; if rd(-1,1)≥P, then keep the current solution unchanged; if rd(-1,1)<P, then accept the new solution. S75. Update the temperature T using the temperature update formula, which is: Where T′ is the updated temperature; α is the standard cooling rate; and θ is the accelerated cooling rate. The accelerated cooling rate θ is dynamically adjusted using an accelerated cooling rate adjustment formula to achieve the optimal accelerated cooling rate; the accelerated cooling rate adjustment formula is as follows: Where θ′ is the adjusted accelerated cooling rate; K is the total number of iterations; and m is the rate factor that controls the decrease of θ with the number of iterations. S76. When the updated temperature T′ is less than or equal to the termination temperature T f The iteration stops when the preset number of iterations L is reached; otherwise, it returns to S72 to continue the next round of iteration. After the iteration stops, the objective function value corresponding to the final solution is recorded. The hyperparameter combination at this time is the optimal hyperparameter combination.

8. The personalized learning path intelligent planning system according to claim 7, characterized in that, The method of comparing the predicted user's learning adaptability with the preset user's learning adaptability threshold to determine whether the user needs to change their learning path includes: If the predicted learning adaptability of a user is greater than or equal to the preset learning adaptability threshold for a user, then it is determined that the user does not need to change the learning path. If the predicted learning fitness level of a user is less than the preset learning fitness threshold, then it is determined that the user needs to change their learning path.

9. The personalized learning path intelligent planning system according to claim 8, characterized in that, The method for constructing the replacement effect evaluation model includes: The learning effectiveness of users after changing their learning paths is evaluated by constructing a change effectiveness evaluation model. The change effectiveness evaluation model is as follows: Where F is the user's overall learning performance score; b is a bias term used to adjust the user's base score; A is the user's homework completion rate; B is the user's practice question accuracy rate; P is the user's exam pass rate; t is the user's average answering time; ω A To control the impact of users' assignment completion rate on their overall learning performance score; ω B To control the impact of users' practice question accuracy on their overall learning performance score; ω P To control the impact of users' exam pass rates on their overall learning performance scores; ω t To control the impact of users' average answering time on users' overall learning performance score; x is an integral variable.

10. The personalized learning path intelligent planning system according to claim 9, characterized in that, The method for recommending personalized learning paths to users based on learning outcomes includes: Personalized learning path recommendations are made for users based on their overall learning performance score. A threshold for the overall learning performance score is preset, and this threshold is further divided into a low overall learning performance score threshold (CV). S And the high score threshold of comprehensive learning effect CV L ; The user's overall learning performance score F is compared with the low score threshold CV of the overall learning performance. S And the high score threshold of comprehensive learning effect CV L In comparison, if F < CV S If the user's overall learning performance score is low, then the overall learning performance score is determined to be low; if CV S ≤F<CV L If F > CV, then the user's overall learning performance score is determined to be moderate; L If so, the user's overall learning performance score is determined to be a high score; For users with low scores, the intelligent learning management terminal automatically provides them with course resources including video tutorials, online practice question banks, and interactive learning games; For users with average scores, the intelligent learning management terminal automatically provides them with opportunities including case studies, mock exams, and interaction with users with high scores. For users with high scores, the intelligent learning management terminal automatically provides them with innovative courses and professional training, including expert lectures, cutting-edge technical documents, and learning seminars.

Images