A personalized learning route generation method and system based on graph convolutional network

Abstract

The present invention discloses a method and system for generating personalized learning routes based on a graph convolutional network, and belongs to the technical field of big data; the method comprises: acquiring learning data and user data; preprocessing the learning data and user data to obtain preprocessed learning data and preprocessed user data; constructing a learning resource graph and a learner behavior graph according to the preprocessed learning data and the preprocessed user data; and constructing a graph convolutional network model according to the learning resource graph; the present invention improves the accuracy and effect of personalized learning, introduces a graph convolutional network, and effectively captures the graph structure information in the learning data by constructing a learning resource graph and a learner behavior graph, and uses the graph convolutional network to aggregate and propagate information, so as to more accurately capture the complex relationship between learning resources and the personalized needs of learners.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a method for generating a personalized learning route based on a graph convolutional network. Background Art

[0002] With the rapid development of artificial intelligence and big data technologies, and the booming digital education today, personalized learning has become an important research direction in the field of education. It aims to tailor learning content and paths for students based on their interests, abilities, learning progress and other personalized factors, thereby improving learning efficiency and learning experience.

[0003] In the existing technology, the methods for generating personalized learning routes mostly rely on traditional recommendation algorithms, such as collaborative filtering and content recommendation. These methods often have limitations such as low accuracy and poor scalability when it comes to large-scale, high-dimensional learning resources and learner behaviors, and cannot effectively capture the deep connections between knowledge and the complex patterns of user learning behaviors.

[0004] In recent years, as a product of the combination of graph theory and deep learning, graph convolutional networks (GCNs) have provided powerful graph structure data processing capabilities and tools. They can effectively process complex relationships in graph structure data and have been widely used in social network analysis, recommendation systems and other fields. Therefore, we can use the powerful graph processing capabilities of graph convolutional networks to deeply explore the intrinsic connections between learning resources and the potential relationships between learning resources and users, provide innovative methods and ideas for the generation of personalized learning routes, solve the problem of insufficient processing of the complexity and correlation of user learning data, and improve the effect and efficiency of personalized learning.

[0005] The existing technologies have the following shortcomings in generating personalized learning routes:

[0006] First, there is a lack of in-depth understanding and utilization of the graph structure of learning data, which makes it difficult to accurately capture the complex relationships between learning resources. The internal connections between learning resources are insufficiently explored, making it difficult to form an effective learning path.

[0007] Second, it is difficult to fully utilize learners' historical learning behavior data, and the consideration of user personalization factors is not comprehensive enough, making it impossible to provide users with accurate and personalized learning routes.

[0008] Third, the algorithm is highly complex and has poor scalability, making it difficult to adapt to the rapid growth of learning resources and the number of learners, and unable to provide efficient personalized learning route recommendations in a real-time environment. Summary of the invention

[0009] The purpose of the present invention is to provide a method and system for generating personalized learning routes based on graph convolutional networks. By using the graph structure information in the learning data, a learning resource graph and a learner behavior graph are constructed. By combining the user's learning behavior and preferences, and using graph convolutional networks for information aggregation and dissemination, more accurate and efficient generation of learning routes for personalized needs can be achieved.

[0010] In order to solve the above technical problems, the present invention provides a method for generating a personalized learning route based on a graph convolutional network, comprising the following steps:

[0011] Obtain learning data and user data;

[0012] Preprocessing the learning data and the user data to obtain preprocessed learning data and preprocessed user data;

[0013] Construct a learning resource map and a learner behavior map based on the preprocessed learning data and the preprocessed user data;

[0014] Build a graph convolutional network model based on the learning resource graph;

[0015] Generate personalized learning routes based on the learner behavior graph and graph convolutional network model.

[0016] Preferably, the learning data includes learning resources and the associations between learning resources, and the user data includes learner behaviors and the associations between learner behaviors. According to the preprocessed learning data and the preprocessed user data, a learning resource map and a learner behavior map are constructed, which specifically includes the following steps:

[0017] Based on the graph theory algorithm, for each learning resource in the preprocessed learning data, a node is created; for each association relationship between learning resources in the preprocessed learning data, an edge is created between the corresponding nodes and the corresponding weight is set;

[0018] Based on the graph theory algorithm, for each learner behavior in the preprocessed user data, a node is created; for each association relationship between learner behaviors in the preprocessed user data, an edge is created between the corresponding nodes and the corresponding weights are set.

[0019] Preferably, the learning resource graph and the learner behavior graph are both undirected graphs G=(V, E); wherein: V is a set of nodes, each node vi represents a learning resource; E is a set of edges, each edge e_ij represents the association relationship between resources v_i and v_j; the edge weight w_ij represents the strength of the association between resources.

[0020] Preferably, a graph convolutional network model is constructed according to the learning resource graph, specifically comprising the following steps:

[0021] Extract the resource feature vector h_i of the node in the learning resource graph;

[0022] Aggregate the information of neighboring nodes to update the representation graph convolution layer of the current node. For the node v_i in the lth layer, its output representation h_i^(l+1) is calculated by the following formula:

[0023] h_i^(l+1) =σ(∑_{j∈N(i)} (W^(l)h_j^(l) + b^(l)))

[0024] Where N(i) represents the set of neighbor nodes of node v_i; W^(l) and b^(l) are the weight matrix and bias term of the lth layer respectively; σ is the activation function; j is an index variable that traverses the neighbor set N(i) of node i, and h_j^(l) represents the feature vector or hidden state of the node in the jth layer (or iteration / time step);

[0025] The formula of graph convolution operator is:

[0026] ;

[0027] Among them: Let the center point be ; is a normalization factor, such as the inverse of the node degree; For Node Neighbors, including itself; For Node Type; for Type node in The transformation weight parameters of the layer;

[0028] Stack multiple graph convolution layers to build a multi-layer graph convolution model;

[0029] During the training process, the output of the model is calculated through forward propagation, and then the loss function is used to calculate the difference between the predicted path and the actual path. The gradient is then calculated using the backpropagation algorithm, and the parameters of the model are updated using the optimization algorithm. This process is repeated until the model converges or the preset number of iterations is reached.

[0030] Preferably, the graph convolutional network model includes multiple graph convolutional layers, each of which learns the representation of the current node by aggregating information of its neighboring nodes.

[0031] Preferably, generating a personalized learning route according to the learner behavior graph specifically includes the following steps:

[0032] Extract the user feature vector u of the node in the learner behavior graph;

[0033] Match the user feature vector u with the resource feature vector to obtain the most relevant resource set;

[0034] Based on the graph traversal algorithm, multiple candidate learning paths are generated according to the most relevant resource sets;

[0035] The nodes in the learner behavior graph are evaluated through the graph convolutional network model to obtain the evaluation results;

[0036] According to the evaluation results, the optimal path is selected from the candidate learning paths as the personalized learning route.

[0037] Preferably, evaluating the nodes in the learner behavior graph by using a graph convolutional network model to obtain an evaluation result specifically includes the following steps:

[0038] The nodes corresponding to all candidate learning paths are input into the graph convolutional network model, and the representation of the output nodes is used as the evaluation result.

[0039] Preferably, according to the evaluation results, the optimal path is selected from the candidate learning paths as the personalized learning route, which specifically includes the following steps:

[0040] Aggregate the representations of the nodes to obtain the candidate learning path representations;

[0041] Scoring the candidate learning path representations based on the scoring function, and selecting the candidate learning path with the highest score as the optimal path;

[0042] Use the optimal path as a personalized learning route.

[0043] Preferably, the pretreatment includes cleaning, integration and standardization.

[0044] The present invention also provides a personalized learning route generation system based on a graph convolutional network, comprising:

[0045] Acquisition module, used to acquire learning data and user data;

[0046] A preprocessing module, used to preprocess the learning data and the user data to obtain preprocessed learning data and preprocessed user data;

[0047] A graph construction module, used for constructing a learning resource graph and a learner behavior graph according to the preprocessed learning data and the preprocessed user data;

[0048] The model building module is used to build a graph convolutional network model based on the learning resource graph;

[0049] The route generation module is used to generate personalized learning routes based on the learner behavior graph and graph convolutional network model.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. Improved the accuracy and effectiveness of personalized learning. The graph convolutional network is introduced to effectively capture the graph structure information in the learning data by constructing a learning resource graph and a learner behavior graph. The graph convolutional network is used for information aggregation and dissemination, which can more accurately capture the complex relationship between learning resources and the personalized needs of learners.

[0052] 2. The scalability and adaptability of the system are enhanced. By constructing graph-structured data to represent learning resources and learner behaviors, it is easy to add new resources or learners and retrain the model to update resource embeddings.

[0053] 3. Reduces the complexity and cost of the system. By introducing the graph convolutional network, an efficient graph structure data processing tool, it is possible to avoid the use of complex machine learning algorithms or deep learning models, thereby reducing the complexity and cost of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.

[0055] Figure 1 It is a flow chart of a method for generating a personalized learning route based on a graph convolutional network according to the present invention;

[0056] Figure 2 It is a schematic diagram of an example of building a learning resource map;

[0057] Figure 3 This is a sample diagram of the learner behavior map construction;

[0058] Figure 4 It is a schematic diagram of the graph convolutional network model. DETAILED DESCRIPTION

[0059] Many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of the present invention, so the present invention is not limited to the specific implementation disclosed below.

[0060] The terms used in one or more embodiments of this specification are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of this specification. The singular forms of "a", "said" and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0061] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0062] The present invention is further described in detail below in conjunction with the accompanying drawings:

[0063] The present invention provides a method for generating a personalized learning route based on a graph convolutional network, comprising the following steps:

[0064] Obtain learning data and user data;

[0065] Preprocessing the learning data and the user data to obtain preprocessed learning data and preprocessed user data;

[0066] Construct a learning resource map and a learner behavior map based on the preprocessed learning data and the preprocessed user data;

[0067] Build a graph convolutional network model based on the learning resource graph;

[0068] Generate personalized learning routes based on the learner behavior graph and graph convolutional network model.

[0069] Preferably, the learning data includes learning resources and the associations between learning resources, and the user data includes learner behaviors and the associations between learner behaviors. According to the preprocessed learning data and the preprocessed user data, a learning resource map and a learner behavior map are constructed, which specifically includes the following steps:

[0070] Based on the graph theory algorithm, for each learning resource in the preprocessed learning data, a node is created; for each association relationship between learning resources in the preprocessed learning data, an edge is created between the corresponding nodes and the corresponding weight is set;

[0071] Based on the graph theory algorithm, for each learner behavior in the preprocessed user data, a node is created; for each association relationship between learner behaviors in the preprocessed user data, an edge is created between the corresponding nodes and the corresponding weights are set.

[0072] Preferably, the learning resource graph and the learner behavior graph are both undirected graphs G=(V, E); wherein: V is a set of nodes, each node vi represents a learning resource; E is a set of edges, each edge e_ij represents the association relationship between resources v_i and v_j; the edge weight w_ij represents the strength of the association between resources.

[0073] Preferably, a graph convolutional network model is constructed according to the learning resource graph, specifically comprising the following steps:

[0074] Extract the resource feature vector h_i of the node in the learning resource graph;

[0075] Aggregate the information of neighboring nodes to update the representation graph convolution layer of the current node. For the node v_i in the lth layer, its output representation h_i^(l+1) is calculated by the following formula:

[0076] h_i^(l+1) =σ(∑_{j∈N(i)} (W^(l)h_j^(l) + b^(l)))

[0077] Where N(i) represents the set of neighbor nodes of node v_i; W^(l) and b^(l) are the weight matrix and bias term of the lth layer respectively; σ is the activation function; j is an index variable that traverses the neighbor set N(i) of node i, and h_j^(l) represents the feature vector or hidden state of the node in the jth layer;

[0078] The formula of graph convolution operator is:

[0079] ;

[0080] Among them: Let the center point be ; is a normalization factor, such as the inverse of the node degree; For Node Neighbors, including itself; For Node Type; for Type node in The transformation weight parameters of the layer;

[0081] Stack multiple graph convolution layers to build a multi-layer graph convolution model;

[0082] During the training process, the output of the model is calculated through forward propagation, and then the loss function is used to calculate the difference between the predicted path and the actual path. The gradient is then calculated using the backpropagation algorithm, and the parameters of the model are updated using the optimization algorithm. This process is repeated until the model converges or the preset number of iterations is reached.

[0083] Preferably, the graph convolutional network model includes multiple graph convolutional layers, each of which learns the representation of the current node by aggregating information of its neighboring nodes.

[0084] Preferably, generating a personalized learning route according to the learner behavior graph specifically includes the following steps:

[0085] Extract the user feature vector u of the node in the learner behavior graph;

[0086] Match the user feature vector u with the resource feature vector to obtain the most relevant resource set;

[0087] Based on the graph traversal algorithm, multiple candidate learning paths are generated according to the most relevant resource sets;

[0088] The nodes in the learner behavior graph are evaluated through the graph convolutional network model to obtain the evaluation results;

[0089] According to the evaluation results, the optimal path is selected from the candidate learning paths as the personalized learning route.

[0090] Preferably, evaluating the nodes in the learner behavior graph by using a graph convolutional network model to obtain an evaluation result specifically includes the following steps:

[0091] The nodes corresponding to all candidate learning paths are input into the graph convolutional network model, and the representation of the output nodes is used as the evaluation result.

[0092] Preferably, according to the evaluation results, the optimal path is selected from the candidate learning paths as the personalized learning route, which specifically includes the following steps:

[0093] Aggregate the representations of the nodes to obtain the candidate learning path representations;

[0094] Scoring the candidate learning path representations based on the scoring function, and selecting the candidate learning path with the highest score as the optimal path;

[0095] Use the optimal path as a personalized learning route.

[0096] Preferably, the pretreatment includes cleaning, integration and standardization.

[0097] The present invention also provides a personalized learning route generation system based on a graph convolutional network, comprising:

[0098] Acquisition module, used to acquire learning data and user data;

[0099] A preprocessing module, used to preprocess the learning data and the user data to obtain preprocessed learning data and preprocessed user data;

[0100] A graph construction module, used for constructing a learning resource graph and a learner behavior graph according to the preprocessed learning data and the preprocessed user data;

[0101] The model building module is used to build a graph convolutional network model based on the learning resource graph;

[0102] The route generation module is used to generate personalized learning routes based on the learner behavior graph and graph convolutional network model.

[0103] In order to better illustrate the technical effect of the present invention, the present invention provides the following specific embodiments to illustrate the above technical process:

[0104] Embodiment 1: A personalized learning route generation method based on graph convolutional network, which adopts graph convolutional network (GCN) as the core technology, studies and aggregates key technologies such as data preprocessing, graph structure modeling, feature extraction, model training and tuning, and learning route generation, and proposes a personalized learning route generation method based on graph convolutional network, such as Figure 1 As shown, including the following:

[0105] 1. Data preprocessing: Clean, integrate and standardize the user's learning data to form a unified data format, providing a basis for subsequent graph structure modeling and feature extraction.

[0106] 2. Graph structure modeling: Based on the relevance and hierarchy of learning data, a graph structure model of learning knowledge points is constructed. The nodes in the graph represent knowledge points, and the edges represent the relationships between knowledge points (such as predecessor, successor, similarity, etc.). Use the NetworkX graph theory library to construct a graph structure model of learning knowledge points.

[0107] 3. Feature extraction: Use graph convolutional networks to extract features from graph structure models to capture the deep connections between knowledge points and the complex patterns of user learning behaviors. Through multi-layer graph convolution operations, node-level feature vectors are extracted for subsequent learning route generation. Use graph convolutional network libraries for feature extraction.

[0108] 4. Graph convolutional network model training and tuning: Based on feature extraction, the graph convolutional network model is trained using labeled learning path data. By minimizing the difference between the predicted path and the actual path, the model parameters are continuously optimized so that it can accurately generate personalized learning routes based on the user's learning preferences and historical learning data. After training, the model is applied to new learners or learning resources, a learner behavior graph is constructed based on the user's learning behavior data, and the trained model is used to generate a learning route that meets the user's personalized needs.

[0109] 5. Learning route generation: Based on the extracted feature vectors and the user's learning preferences, a deep learning model is used to generate the learning route. By training the model, it can predict the learning route that best meets the user's personalized needs based on the user's historical learning data and current learning status.

[0110] The implementation process of the personalized learning route generation method based on the graph convolutional network in this embodiment mainly includes the following steps: construction of a learning resource graph (data preprocessing, graph structure modeling, feature extraction), construction of a learner behavior graph (data preprocessing, graph structure modeling, feature extraction), training and application of a graph convolutional network model, and generation of a personalized learning route, specifically including the following steps:

[0111] 1. Construction of learning resource map:

[0112] Based on the associations between learning resources (such as topic similarity, knowledge point dependency, etc.), a learning resource graph is constructed to represent the complex relationships between learning resources in an intuitive graph structure. The learning resource graph is an undirected graph in which nodes represent learning resources (including knowledge points, courses, etc.) and edges represent the associations between resources (including predecessor and successor relationships, similarity relationships, etc.). By introducing edge weights to represent the strength of the association, and using graph theory algorithms to process and integrate the relationships, the complex relationships between learning resources can be more accurately portrayed.

[0113] 1. Data preparation:

[0114] Collect learning resource data, including knowledge points, courses, teaching materials, etc., as well as the relationships between them, such as predecessor and successor relationships, similarity relationships, etc.

[0115] 2. Graph structure definition:

[0116] The learning resource graph is an undirected graph G=(V, E), where V is a set of nodes, each node v_i represents a learning resource; E is a set of edges, each edge e_ij represents the association between resources v_i and v_j. The edge weight w_ij can represent the strength of the association between resources.

[0117] 3. Graph construction algorithm:

[0118] Based on the collected data, we use graph theory algorithms (adjacency matrix method, adjacency list method) to build a learning resource graph. For each learning resource, we create a node; for the relationship between resources, we create an edge between the corresponding nodes and set the corresponding weight, such as Figure 2 shown.

[0119] 2. Construction of learner behavior map:

[0120] Based on the learner's historical learning behavior data (such as learning time, learning sequence, learning evaluation, etc.), a learner behavior graph is constructed. The nodes in the learner behavior graph represent learning resources, and the edges represent the learner's access relationship to resources. Similarly, by introducing edge weights to represent the learner's preference for resources and using graph theory algorithms to process and integrate relationships, the learner's personalized needs can be captured more accurately.

[0121] 1. Data preparation:

[0122] Collect user data, including basic information, learning preferences, historical learning records, etc.

[0123] 2. Graph structure definition:

[0124] The learner behavior graph is an undirected graph G=(V, E), where V is a set of nodes, each node v_i represents a learning behavior; E is a set of edges, each edge e_ij represents the association between resources v_i and v_j. The weight w_ij of the edge can represent the strength of the association between resources.

[0125] 3. Graph construction algorithm:

[0126] Based on the collected data, we use graph theory algorithms (adjacency matrix method, adjacency list method) to build a learner behavior graph. For each learner behavior, we create a node; for the relationship between resources, we create an edge between the corresponding nodes and set the corresponding weight, such as Figure 3 shown.

[0127] 3. Training and application of graph convolutional network model:

[0128] Based on the constructed learning resource graph and learner behavior graph, a graph convolutional network model is designed and implemented. The core of this model consists of multiple graph convolutional layers, each of which learns the representation of the current node by aggregating the information of its neighboring nodes. With the stacking of graph convolutional layers, the model can gradually learn higher-level resource representations, thereby capturing the complex relationships between learning resources in a deeper way.

[0129] The main goal of the model is to learn and optimize the node representation in the learning resource graph, namely, resource embedding. We expect that similar or related resources can have similar representations in the embedding space. At the same time, the model also needs to fully consider the influence of the learner behavior graph to ensure that resource embedding can accurately reflect the personalized learning needs of different learners.

[0130] During the model training phase, we adopted a supervised learning approach. Using existing user learning path data, we built a labeled dataset to guide the model training process. By minimizing the difference between the learning path predicted by the model and the actual learning path, we continuously optimized the model's parameters so that it could generate more accurate and personalized learning routes.

[0131] After training, the model can be flexibly applied to new learning scenarios. For new learners, we can build a behavior graph based on their historical learning behavior data and use the trained model to generate a personalized learning path for them. Similarly, when new learning resources are added to the learning resource graph, we can choose to retrain the model to update the resource embedding, thereby ensuring that the model can always reflect the latest learning resource information.

[0132] 1. Node feature extraction:

[0133] Each learning resource node v_i has a series of features, such as resource type, difficulty, duration, keywords, etc. We represent these features as a resource feature vector h_i. For text type features (such as keywords), we use word embedding technology (Word2Vec) to convert them into numerical vectors.

[0134] 2. Graph Convolutional Layer:

[0135] The graph convolution layer is the core part of the GCN model, which updates the representation of the current node by aggregating the information of neighboring nodes. For the node v_i in the lth layer, its output representation h_i^(l+1) is calculated by the following formula:

[0136] h_i^(l+1) =σ(∑_{j∈N(i)} (W^(l)h_j^(l) + b^(l)))

[0137] Among them, N(i) represents the set of neighbor nodes of node v_i; W^(l) and b^(l) are the weight matrix and bias term of the lth layer respectively; σ is the activation function (ReLU, Sigmoid, etc.); j is an index variable that traverses the neighbor set N(i) of node i, and h_j^(l) represents the feature vector or hidden state of the node in the jth layer (or iteration / time step).

[0138] The formula of graph convolution operator is:

[0139] ;

[0140] Among them: Let the center point be ; is a normalization factor, such as the inverse of the node degree; For Node Neighbors, including itself; For Node Type; for Type node in The transformation weight parameters of the layer;

[0141] 3. Multi-layer graph convolutional network:

[0142] In order to capture the complex relationships between learning resources, we stack multiple graph convolutional layers to build a multi-layer graph convolutional network. By passing and aggregating information layer by layer, we can learn higher-level resource representations.

[0143] 4. Loss function and optimization:

[0144] During the training process, the output of the model is calculated through forward propagation, and then the difference between the predicted path and the actual path is calculated using the loss function. Then the gradient is calculated using the backpropagation algorithm, and the parameters of the model are updated using the optimization algorithm (RMSprop). This process is repeated until the model converges or reaches the preset number of iterations, such as Figure 4 shown.

[0145] 4. Generation of personalized learning routes:

[0146] Generate a personalized learning route based on the trained model and the new learner behavior graph. Specifically, the most relevant learning resources can be selected as recommendation results based on the learner's personalized factors (learning status and interest preferences, etc.). The recommendation results can be arranged in a certain order to form a personalized learning route.

[0147] The specific process is as follows:

[0148] 1. User feature extraction:

[0149] Extract the user's personalized factors and represent them in vector form. For each user, we can extract personalized features from their historical learning records, such as learning preferences, learning progress, etc. These features are represented as a user feature vector u.

[0150] 2. User and resource matching:

[0151] Match the user vector with the resource representation and calculate the similarity between the user and each resource. By calculating the similarity between the user feature vector u and the resource representation h_i (such as cosine similarity, inner product, etc.), we can find the resource set most relevant to the user.

[0152] 3. Candidate path generation:

[0153] Based on the resource set that is most relevant to the user, we can generate multiple candidate learning paths. The generation of candidate paths can be based on graph traversal algorithms (such as depth-first search, breadth-first search, etc.) or heuristic search algorithms (such as A* algorithm, genetic algorithm, etc.).

[0154] 4. Path evaluation and selection:

[0155] The graph convolutional network model is used to evaluate the candidate paths, and the path with the best evaluation result is selected as the final personalized learning route. Specifically, we can input the resource nodes in the path into the model one by one to obtain the representation of each node; then, we can aggregate these representations (taking the average strategy and taking the maximum strategy) to obtain the representation of the entire path;

[0156] The averaging strategy is to take the arithmetic average of the representation vectors of all nodes in the path to obtain a single vector to represent the entire path. Specifically, the representation of the entire path is h The pathmean can be h pathmean= n 1∑ i =1 nhi Formula calculation, here ∑ i =1 nhi It means to sum the elements of all node representation vectors and then divide by the number of nodes. n to achieve average.

[0157] The maximum strategy is to take the maximum value along each dimension of the representation vector, so as to obtain a vector of the same dimension to represent the entire path. Specifically, the representation of the entire path is h Pathmax can be h pathmax=maxpool( h 1, h 2,…, hn ) formula. Here maxpool means taking the maximum value in each dimension. If hi is a d -dimensional vector, then h pathmax will also be a d -dimensional vector, where each element is the sum of all hi The maximum value in .

[0158] Finally, we use a scoring function (neural network scoring function) to score the path representation and select the path with the highest score as the final personalized learning route.

[0159] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules, modules or units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units, modules or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0160] The units may or may not be physically separated, and the components displayed as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0161] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0162] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present invention are executed. It should be noted that the above-mentioned computer-readable medium of the present invention can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above.

[0163] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0164] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A personalized learning route generation method based on graph convolutional network, It is characterized in that The following steps are involved: Acquire learning data and user data; the learning data includes learning resources and associations between learning resources, and the user data includes learner behaviors and associations between learner behaviors; Preprocessing the learning data and the user data to obtain preprocessed learning data and preprocessed user data; According to the preprocessed learning data and the preprocessed user data, a learning resource graph and a learner behavior graph are constructed, which includes: based on a graph theory algorithm, for each learning resource in the preprocessed learning data, a node is created; for each association relationship between the learning resources in the preprocessed learning data, an edge is created between the corresponding nodes, and corresponding weights are set; based on a graph theory algorithm, for each learner behavior in the preprocessed user data, a node is created; for each association relationship between the learner behaviors in the preprocessed user data, an edge is created between the corresponding nodes, and corresponding weights are set; According to the learning resource graph, build a graph convolutional network model: Extract resource feature vectors of nodes in the learning resource graph ; Aggregate the information of neighboring nodes to update the representation graph convolution layer of the current node. For the node in the lth layer , whose output indicates , calculated by the following formula: ； in, Representation Node The set of neighbor nodes of and They are The weight matrix and bias term of the layer; is the activation function; is an index variable that traverses the nodes The neighbor set of , Indicated in The feature vectors or hidden states of the nodes in the layer; The formula of graph convolution operator is: ； Among them: Let the center point be ; is the normalization factor, which represents the inverse of the node degree; For Node Neighbors, including itself; For Node Type; for Type node in The transformation weight parameters of the layer; Stack multiple graph convolution layers to build a multi-layer graph convolution model; During the training process, the output of the model is calculated through forward propagation, and then the difference between the predicted path and the actual path is calculated using the loss function. Then the gradient is calculated using the backpropagation algorithm, and the parameters of the model are updated using the optimization algorithm. This process is repeated until the model converges or the preset number of iterations is reached. Extract the user feature vector u of the node in the learner behavior graph; Match the user feature vector u with the resource feature vector to obtain the most relevant resource set; Based on the graph traversal algorithm, multiple candidate learning paths are generated according to the most relevant resource sets; The nodes in the learner behavior graph are evaluated through the graph convolutional network model to obtain the evaluation results; According to the evaluation results, the optimal path is selected from the candidate learning paths as the personalized learning route.

2. According to the method for generating personalized learning routes based on graph convolutional networks according to claim 1, Features: The graph convolutional network model includes multiple graph convolutional layers, each of which learns the representation of the current node by aggregating information of its neighboring nodes.

3. According to the method for generating personalized learning routes based on graph convolutional networks according to claim 2, It is characterized in that According to the evaluation results, the optimal path is selected from the candidate learning paths as the personalized learning route, which specifically includes the following steps: Aggregate the representations of the nodes to obtain the candidate learning path representations; Scoring the candidate learning path representations based on the scoring function, and selecting the candidate learning path with the highest score as the optimal path; Use the optimal path as a personalized learning route.

4. According to claim 3, the method for generating a personalized learning route based on a graph convolutional network, Features: The pretreatment includes cleaning, integration and standardization.

5. A personalized learning route generation system based on a graph convolutional network, used to implement the personalized learning route generation method based on a graph convolutional network as described in any one of claims 1 to 4, It is characterized in that include: Acquisition module, used to acquire learning data and user data; A preprocessing module, used to preprocess the learning data and the user data to obtain preprocessed learning data and preprocessed user data; A graph construction module, used for constructing a learning resource graph and a learner behavior graph according to the preprocessed learning data and the preprocessed user data; The model building module is used to build a graph convolutional network model based on the learning resource graph; The route generation module is used to generate personalized learning routes based on the learner behavior graph and graph convolutional network model.

Images