Intelligent interdisciplinary learning path planning system

Abstract

The invention belongs to the field of learning planning, and particularly discloses an intelligent interdisciplinary learning path planning system which comprises a user information management module, a learning resource management module, a learning path planning module and a learning progress tracking module. According to the scheme, a convolutional neural network is combined with a distribution estimation algorithm, convolutional layer parameters are optimized through stochastic gradient descent, convolutional layer weights are dynamically adjusted, automatic extraction of text local features is achieved, the distribution estimation algorithm is introduced into a full connection layer, a high-fitness weight matrix is dynamically generated, and a population diversity dynamic attenuation standard deviation is combined, so that automatic extraction of text local features is achieved. The fitting ability of the model to complex text features is improved; a knowledge graph and a collaborative filtering algorithm are combined, a multi-dimensional knowledge graph is constructed, the limitation of a traditional single-disciplinary path is broken, collaborative filtering is carried out based on course content, explicit behavior data and implicit semantic features are combined, and an interdisciplinary learning path is dynamically generated.

Description

Technical Field

[0001] The present invention relates to the field of learning planning, and specifically refers to an intelligent interdisciplinary learning path planning system. Background Art

[0002] Intelligent interdisciplinary learning path planning uses artificial intelligence technology to customize a learning route that involves the integration of knowledge from multiple disciplines for learners based on various factors such as their knowledge level, learning goals, and interests. When planning an interdisciplinary learning path, a large amount of time is required for manual annotation of learning resources. In terms of subject classification, due to the fact that human judgment is easily affected by subjective factors and lacks accurate criteria and sufficient knowledge coverage, the accuracy of subject classification is significantly insufficient; traditional learning methods often center around a single discipline in teaching, lacking guidance on the relationships between interdisciplinary knowledge, which affects people's overall understanding of complex interdisciplinary knowledge, and the learning sequence lacks scientificity. This single thinking training mode makes it difficult for people to break through disciplinary boundaries and understand and solve interdisciplinary problems with different thinking modes when learning interdisciplinary knowledge. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an intelligent interdisciplinary learning path planning system. Regarding the technical problem that a large amount of time is required for manual annotation of learning resources when planning an interdisciplinary learning path, and in terms of subject classification, due to the fact that human judgment is easily affected by subjective factors, resulting in a significant lack of accuracy in subject classification, this solution combines a convolutional neural network with an estimation of distribution algorithm, optimizes the parameters of the convolutional layer through stochastic gradient descent, uses the cross-entropy loss function to measure the error between the predicted label and the true label, and dynamically adjusts the weights of the convolutional layer through backpropagation to achieve automatic extraction of local features of the text. In the fully connected layer, an estimation of distribution algorithm is introduced, and by calculating the mean, variance, and covariance matrix of the candidate set parameters, a high-fitness weight matrix is dynamically generated, and the standard deviation is dynamically attenuated in combination with population diversity to improve the model's fitting ability for complex text features; regarding the technical problem that traditional learning methods often center around a single discipline in teaching, affecting people's overall understanding of complex interdisciplinary knowledge, and the learning sequence lacks scientificity, and this single thinking training mode makes it difficult for people to break through disciplinary boundaries when learning interdisciplinary knowledge, this solution combines a knowledge graph and a collaborative filtering algorithm, extracts entities and relationships from course metadata through knowledge extraction technology to construct a multi-dimensional knowledge graph, breaks the limitations of the traditional single-discipline path, conducts collaborative filtering based on course content, combines explicit behavior data and implicit semantic features, can dynamically generate an interdisciplinary learning path, and improves the matching degree of the learning path.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent interdisciplinary learning path planning system, including a user information management module, a learning resource management module, a learning path planning module, and a learning progress tracking module;

[0005] The user information management module collects and stores the basic information of users, and the basic information includes age, educational background, learning goals, interest areas, and current knowledge level;

[0006] The learning resource management module integrates and annotates various learning resources. The learning resources include online courses, e-books, and academic papers. The annotations include subject classification, difficulty level, learning duration, and prerequisite knowledge requirements. When performing subject classification, a convolutional neural network method is used to analyze the description text of the learning resources and classify them into corresponding subject fields;

[0007] The learning path planning module constructs a knowledge graph through knowledge extraction and generates an interdisciplinary learning path for users based on user information, learning resources, and the knowledge graph;

[0008] The learning progress tracking module tracks the learning progress of users, evaluates the learning effect according to the learning situation of users, and feeds back the learning effect to the learning path planning module to continuously adjust the learning path.

[0009] The learning resource management module uses a convolutional neural network method to analyze the description text of learning resources, which specifically includes the following steps:

[0010] Step S1: Construct and initialize a convolutional neural network as a text classification model. The embedding layer of the convolutional neural network uses a pre-trained Glove model to encode the text into a fixed-length sequence. The fully connected layer uses the ReLU function as the output, and the output layer uses the softmax function as the output;

[0011] Step S2: Use the gradient method to train the convolutional layer of the convolutional neural network to achieve feature extraction;

[0012] Step S3: Optimize the weight matrix of the fully connected layer based on the estimation of distribution algorithm;

[0013] Step S4: Input the description text of the learning resources into the text classification model for classification.

[0014] Further, step S2 specifically includes the following steps:

[0015] Step S21: Parameter initialization, randomly initialize the weight parameters of the convolutional layer;

[0016] Step S22: Forward propagation. Collect the training text set, where the training text set is descriptive text with true classification labels. Input the training text set into the convolutional neural network. After passing through the convolutional layer, activation function, and pooling layer in sequence, generate high-level features and obtain the predicted classification labels.

[0017] Step S23: Loss calculation. Use the cross-entropy loss function to calculate the error between the predicted classification label and the true classification label. The formula used is as follows: ;

[0018] In the formula, is the cross-entropy loss function, is the weight parameter vector of the convolutional layer of the convolutional neural network, is the training text set, is the total number of descriptive texts in the training text set, is the traversal of , is the number of categories of the true classification label, is the traversal of , is the indicator function, is the th descriptive text. When 's category is , the indicator function is 1; otherwise, the indicator function is 0. is the probability that the convolutional neural network predicts 's category is ;

[0019] Step S24: Backward propagation. Calculate the gradient of the cross-entropy loss with respect to the convolutional layer parameters, and use the stochastic gradient descent method to update the convolutional layer weight parameters.

[0020] Furthermore, step S3 specifically includes the following steps:

[0021] Step S31: Parameter initialization. Randomly initialize the weight matrix of the fully connected layer to generate the initial population. Each individual in the initial population is a weight matrix.

[0022] Step S32: Fitness evaluation. For each individual in the initial population, use the cross-entropy loss function in step S23 to calculate the cross-entropy loss value of the individual on the training text set, and use the reciprocal of the cross-entropy loss value as the fitness value of the individual.

[0023] Step S33: Selection and distribution estimation. Sort the individuals in descending order of fitness value, select the top 50% of the individuals to construct a candidate set, and use the univariate marginal distribution algorithm to calculate the mean and variance of each parameter in the candidate set to generate new individuals. The formula used is as follows: ;

[0024] In the formula, is the weight matrix of the fully connected layer of the th iteration of the th individual calculated by the univariate marginal distribution algorithm, is the standard deviation of the th iteration, used to control the search range, is a random number sampled from the standard normal distribution, is the mean of the weight matrices of the excellent individuals in the candidate set at the th iteration;

[0025] Through the multivariate normal distribution algorithm, calculate the mean vector and covariance matrix of the candidate set, and generate a new individual. The formula used is as follows: ;

[0026] In the formula, is the weight matrix of the fully connected layer of the th iteration of the th individual calculated by the multivariate normal distribution algorithm, is the global center point of the th iteration, that is, the historical optimal parameter position;

[0027] Step S34: Dynamic adjustment, dynamically narrow the search range according to the population diversity, and decay the variance. The formula used is as follows: ;

[0028] In the formula, is the standard deviation, is the number of individuals in the candidate set, is the traversal of , is the th iteration of the th individual's parameter vector in the candidate set, is the dimension of the fully connected layer parameters, is the Euclidean norm;

[0029] Step S35: Update the global center point of the initial population, record the individual with the highest fitness as the historical optimal parameter, preset the number of iterations, and repeat steps S32 to S35 until the preset number of iterations is reached.

[0030] Furthermore, the learning path planning module uses a learning path planning method to output a learning path. The learning path planning method specifically includes the following steps:

[0031] Step M1: Data collection and preprocessing. Collect user behavior data and course metadata to construct a planning dataset. The user behavior data includes user ratings, clicks, browsing durations, and course completion rates. The course metadata includes course titles, course descriptions, subject fields, difficulty levels, instructors, keyword tags, and prerequisite courses. Uniformly encode the course metadata, and divide the data in the planning dataset into a training set and a test set at a ratio of 8:2.

[0032] Step M2: Construct a knowledge graph. Perform entity recognition through knowledge extraction methods, conduct relationship extraction, construct course hierarchical relationships, user behavior relationships, and semantic association relationships, and convert entities and relationships into low-dimensional graph embedding vectors.

[0033] Step M3: Use the K-nearest neighbor algorithm to calculate the weighted cosine similarity between users and courses as a distance metric based on the user-course rating matrix. The user-course rating matrix represents the degree of interest of users in courses, and obtain the nearest K neighbors.

[0034] Step M4: Use the singular value decomposition model to decompose the user-course rating matrix into three matrices: a left singular matrix, a diagonal matrix, and a right singular matrix, which respectively represent the relationship between users and latent factors, the strength of latent factors, and the relationship between courses and latent factors. The latent factors are users' learning preferences and course metadata. Retain the first k singular values of the three matrices to reduce the matrix dimension and obtain the low-dimensional latent features of users and courses.

[0035] Step M5: Calculate the cosine similarity between users, find the course learning paths of similar users based on latent factors, calculate the dot product of the low-dimensional latent features of the courses of users and similar users to obtain a course matching score, and output the courses according to the course matching score to obtain an interdisciplinary learning path.

[0036] The beneficial effects achieved by the present invention using the above solution are as follows:

[0037] (1) Aiming at the technical problems that when planning interdisciplinary learning paths, it takes a lot of time to manually label learning resources, and in terms of subject classification, due to the easy influence of subjective factors on human judgment, the accuracy of subject classification is significantly insufficient. This solution combines a convolutional neural network and an estimation of distribution algorithm, optimizes the parameters of the convolutional layer through stochastic gradient descent, uses the cross-entropy loss function to measure the error between the predicted label and the true label, and dynamically adjusts the weights of the convolutional layer through backpropagation to achieve automatic extraction of local text features. And introduce the estimation of distribution algorithm in the fully connected layer. By calculating the mean, variance, and covariance matrix of the candidate set parameters, dynamically generate a high-fitness weight matrix, and combine the dynamic decay of the standard deviation of population diversity to improve the model's fitting ability for complex text features.

[0038] (2)Regarding the technical problem that traditional learning methods often teach centered around a single discipline, which affects people's overall grasp of complex interdisciplinary knowledge, lacks scientific learning order, and this single thinking training mode makes it difficult for people to break through the disciplinary boundaries when learning interdisciplinary knowledge, this solution combines knowledge graphs and collaborative filtering algorithms, extracts entities and relationships from curriculum metadata through knowledge extraction technology, constructs a multi-dimensional knowledge graph, breaks the limitations of traditional single-disciplinary paths, conducts collaborative filtering based on curriculum content, combines explicit behavioral data and implicit semantic features, can dynamically generate interdisciplinary learning paths, and improves the matching degree of learning paths. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a module connection diagram of an intelligent interdisciplinary learning path planning system provided by the present invention.

[0040] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0042] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present invention.

[0043] Embodiment 1: Refer to Figure 1 An intelligent interdisciplinary learning path planning system provided in this embodiment includes a user information management module, a learning resource management module, a learning path planning module, and a learning progress tracking module;

[0044] The user information management module collects and stores the basic information of users, and the basic information includes age, educational background, learning goals, interest fields, and current knowledge levels;

[0045] The learning resource management module integrates and annotates various learning resources, including online courses, e-books, and academic papers. The annotations include subject classification, difficulty level, learning duration, and prerequisite knowledge requirements. When performing subject classification, a convolutional neural network method is used to analyze the descriptive text of the learning resources and classify them into corresponding subject fields;

[0046] The learning path planning module constructs a knowledge graph through knowledge extraction and generates an interdisciplinary learning path for users based on user information, learning resources, and the knowledge graph;

[0047] The learning progress tracking module tracks the learning progress of users, evaluates the learning effect based on the learning situation of users, and feeds back the learning effect to the learning path planning module to continuously adjust the learning path.

[0048] Example 2: Refer to Figure 1 Based on the above example, the learning resource management module uses a convolutional neural network method to analyze the descriptive text of learning resources, which specifically includes the following steps:

[0049] Step S1: Construct and initialize a convolutional neural network as a text classification model. The embedding layer of the convolutional neural network uses a pre-trained Glove model to encode the text into a fixed-length sequence. The fully connected layer uses the ReLU function as the output, and the output layer uses the softmax function as the output;

[0050] Step S2: Use the gradient method to train the convolutional layer of the convolutional neural network to achieve feature extraction;

[0051] Step S3: Optimize the weight matrix of the fully connected layer based on the estimation of distribution algorithm;

[0052] Step S4: Input the descriptive text of the learning resources into the text classification model for classification.

[0053] Example 3: Refer to Figure 1 Based on the above example, step S2 specifically includes the following steps:

[0054] Step S21: Parameter initialization, randomly initialize the weight parameters of the convolutional layer;

[0055] Step S22: Forward propagation, collect the training text set, which is the descriptive text with real classification labels. Input the training text set into the convolutional neural network. After passing through the convolutional layer, activation function, and pooling layer in sequence, generate high-level features and obtain the predicted classification labels;

[0056] Step S23: Loss calculation. Use the cross-entropy loss function to calculate the error between the predicted classification label and the true classification label. The formula used is as follows: ;

[0057] In the formula, is the cross-entropy loss function, is the weight parameter vector of the convolutional layer of the convolutional neural network, is the training text set, is the total number of description texts in the training text set, is the traversal of , is the number of categories of the true classification label, is the traversal of ; is the indicator function, is the th description text. When the category of is , the indicator function is 1; otherwise, the indicator function is 0. is the probability that the convolutional neural network predicts that the category of is ;

[0058] Step S24: Backpropagation. Calculate the gradient of the cross-entropy loss with respect to the convolutional layer parameters, and use the stochastic gradient descent method to update the convolutional layer weight parameters.

[0059] Example 4: Refer to Figure 1 , which is based on the above example. Step S3 specifically includes the following steps:

[0060] Step S31: Parameter initialization. Randomly initialize the weight matrix of the fully connected layer to generate an initial population. Each individual in the initial population is a weight matrix;

[0061] Step S32: Fitness evaluation. For each individual in the initial population, use the cross-entropy loss function in Step S23 to calculate the cross-entropy loss value of the individual on the training text set, and use the reciprocal of the cross-entropy loss value as the fitness value of the individual;

[0062] Step S33: Selection and distribution estimation. Sort the individuals in descending order of fitness value, select the top 50% of the individuals to construct a candidate set, and use the univariate marginal distribution algorithm to calculate the mean and variance of each parameter in the candidate set to generate new individuals. The formula used is as follows: ;

[0063] In the formula, is the th iteration calculated by the univariate marginal distribution algorithm, and the The weight matrix of the fully connected layer of an individual is the standard deviation of the th iteration, which is used to control the search range, is a random number sampled from the standard normal distribution, and is the mean of the weight matrices of the excellent individuals in the candidate set at the

[0064] th iteration; Through the multivariate normal distribution algorithm, calculate the mean vector and covariance matrix of the candidate set, and generate a new individual. The formula used is as follows: ;

[0065] In the formula, is the weight matrix of the fully connected layer of the th individual at the th iteration calculated by the multivariate normal distribution algorithm, is the global center point of the th iteration, that is, the position of the historical optimal parameter;

[0066] Step S34: Dynamic adjustment, dynamically narrow the search range according to the population diversity, and decay the variance. The formula used is as follows: ;

[0067] In the formula, is the standard deviation, is the number of individuals in the candidate set, is the traversal of , is the th iteration, the th parameter vector of the th individual in the candidate set, is the Euclidean norm;

[0068] Step S35: Update the global center point of the initial population, record the individual with the highest fitness as the historical optimal parameter, preset the number of iterations, and repeat steps S32 to S35 until the preset number of iterations is reached.

[0069] Example Five: Refer to Figure 1 , this example is based on the above example. The learning path planning module uses a learning path planning method to output a learning path. The learning path planning method specifically includes the following steps:

[0070] Step M1: Data collection and preprocessing. Collect user behavior data and course metadata to construct a planning dataset. The user behavior data includes user ratings, clicks, browsing durations, and course completion rates. The course metadata includes course titles, course descriptions, subject areas, difficulty levels, instructors, keyword tags, and prerequisite courses. Uniformly encode the course metadata, and divide the data in the planning dataset into a training set and a test set at a ratio of 8:2.

[0071] Step M2: Construct a knowledge graph. Perform entity recognition through knowledge extraction methods, perform relationship extraction, construct course hierarchy relationships, user behavior relationships, and semantic association relationships, and convert entities and relationships into low-dimensional graph embedding vectors.

[0072] Step M3: Use the K-nearest neighbor algorithm to calculate the weighted cosine similarity between users and courses as a distance metric based on the user-course rating matrix. The user-course rating matrix represents the degree of interest of users in courses, and obtain the nearest K neighbors.

[0073] Step M4: Use the singular value decomposition model to decompose the user-course rating matrix into three matrices: a left singular matrix, a diagonal matrix, and a right singular matrix, which represent the relationships between users and latent factors, the strengths of latent factors, and the relationships between courses and latent factors respectively. The latent factors are users' learning preferences and course metadata. Retain the first k singular values of the three matrices to reduce the matrix dimension and obtain the low-dimensional latent features of users and courses.

[0074] Step M5: Calculate the cosine similarity between users. Based on the latent factors, find the course learning paths of similar users, calculate the dot product of the low-dimensional latent features of the courses of users and similar users to obtain a course matching score, and output the courses according to the course matching score to obtain an interdisciplinary learning path.

[0075] Example 6: This example is based on the above example. In Example 5, when constructing the knowledge graph, entity recognition is performed through knowledge extraction methods. The core entities in this solution are set as users, courses, subject areas, instructors, and keyword tags, and the auxiliary entities are set as users' learning goals, user ratings, and the institutions to which the instructors belong.

[0076] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0077] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

[0078] The above describes the present invention and its implementation manners. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and, without departing from the gist of the present invention, design similar structural manners and embodiments to this technical solution without creative efforts, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent interdisciplinary learning path planning system, characterized in that It includes a user information management module, a learning resource management module, a learning path planning module, and a learning progress tracking module; The user information management module collects and stores the basic information of users; The learning resource management module integrates and annotates various learning resources. The annotation includes subject classification, difficulty level, learning duration, and prerequisite knowledge requirements. When performing subject classification, a convolutional neural network method is used to analyze the description text of the learning resources and classify them into corresponding subject fields; The learning path planning module constructs a knowledge graph through knowledge extraction and generates an interdisciplinary learning path for users based on user information, learning resources, and the knowledge graph; The learning progress tracking module tracks the learning progress of users, evaluates the learning effect according to the learning situation of users, and feeds back the learning effect to the learning path planning module to continuously adjust the learning path.

2. An intelligent interdisciplinary learning path planning system according to claim 1, characterized in that, The learning resource management module uses a convolutional neural network method to analyze the description text of learning resources, which specifically includes the following steps: Step S1: Construct and initialize a convolutional neural network as a text classification model. The embedding layer of the convolutional neural network uses a pre-trained Glove model to encode the text into a fixed-length sequence. The fully connected layer uses the ReLU function as the output, and the output layer uses the softmax function as the output; Step S2: Use the gradient method to train the convolutional layer of the convolutional neural network to achieve feature extraction; Step S3: Optimize the weight matrix of the fully connected layer based on the estimation of distribution algorithm; Step S4: Input the description text of the learning resources into the text classification model for classification.

3. The intelligent interdisciplinary learning path planning system according to claim 2, characterized in that Step S2 specifically includes the following steps: Step S21: Parameter initialization, randomly initialize the weight parameters of the convolutional layer; Step S22: Forward propagation, collect the training text set, which is the description text with real classification labels. Input the training text set into the convolutional neural network to generate high-level features and obtain the predicted classification labels; Step S23: Loss calculation, use the cross-entropy loss function to calculate the error between the predicted classification label and the real classification label; Step S24: Backward propagation, calculate the gradient of the cross-entropy loss with respect to the parameters of the convolutional layer, and use the stochastic gradient descent method to update the weight parameters of the convolutional layer.

4. An intelligent interdisciplinary learning path planning system according to claim 2, characterized in that, Step S3 specifically includes the following steps: Step S31: Parameter initialization, randomly initialize the weight matrix of the fully connected layer to generate an initial population, and each individual in the initial population is a weight matrix; Step S32: Fitness evaluation, for each individual in the initial population, use the cross-entropy loss function in Step S23 to calculate the cross-entropy loss value of the individual on the training text set, and use the reciprocal of the cross-entropy loss value as the fitness value of the individual; Step S33: Selection and distribution estimation, sort the individuals in descending order of fitness value, select the top 50% of the individuals to construct a candidate set, and calculate the mean and variance of each parameter in the candidate set through the univariate marginal distribution algorithm to generate new individuals; At the same time, through the multivariate normal distribution algorithm, calculate the mean vector and covariance matrix of the candidate set to generate new individuals; Step S34: Dynamic adjustment. Dynamically narrow the search range according to the population diversity and perform variance attenuation. Step S35: Update the global center point of the initial population. Denote the individual with the highest fitness as the historical optimal parameter, preset the number of iterations, and repeat Steps S32 to S35 until the preset number of iterations is reached.

5. An intelligent interdisciplinary learning path planning system according to claim 4, characterized in that, The learning path planning module uses a learning path planning method to output a learning path. The learning path planning method specifically includes the following steps: Step M1: Data collection and preprocessing. Collect user behavior data and course metadata to construct a planning dataset, uniformly encode the course metadata, and divide the data in the planning dataset into a training set and a test set at a ratio of 8:

2. Step M2: Construct a knowledge graph. Perform entity recognition through a knowledge extraction method, perform relationship extraction, and convert the entities and relationships into low-dimensional graph embedding vectors. Step M3: Use the K-nearest neighbor algorithm to calculate the weighted cosine similarity between the user and the course as the distance metric based on the user-course rating matrix. The user-course rating matrix represents the degree of interest of the user in the course, and obtain the nearest K neighbors. Step M4: Use the singular value decomposition model to decompose the user-course rating matrix into three matrices: a left singular matrix, a diagonal matrix, and a right singular matrix, which respectively represent the relationship between the user and the latent factors, the strength of the latent factors, and the relationship between the course and the latent factors. The latent factors are the learning preferences of the user and the course metadata. Retain the first k singular values of the three matrices to reduce the matrix dimension and obtain the low-dimensional latent features of the user and the course. Step M5: Calculate the cosine similarity between users. Based on the latent factors, find the course learning paths of similar users, calculate the dot product of the low-dimensional latent features of the courses of the user and the similar users to obtain the course matching score, and output the courses according to the course matching score to obtain an interdisciplinary learning path.

Images