An online course recommendation method based on a double-tower graph convolutional neural network

Abstract

The present disclosure relates to a method for online course recommendation based on a double-tower graph convolutional neural network. The method comprises: obtaining learning platform data, and performing data preprocessing to generate training sample data; using the training sample data containing a meta path as model input data, and using a course recommendation list as model output data to establish a basic model; performing feature extraction and optimizing network feature representation capability of the basic model based on a preset double-tower network structure, optimizing model output based on a self-consistency learning algorithm, and optimizing recommendation accuracy based on a weight cold start strategy to generate an online course intelligent recommendation system model; and completing online course recommendation based on the online course intelligent recommendation system model. The present disclosure combines a double-tower network structure and a self-consistency learning algorithm to optimize the performance of the online course recommendation system from multiple aspects of model structure and model training.

Description

Technical Field

[0001] This disclosure relates to the fields of big data and deep learning in the field of artificial intelligence technology, and more specifically, to an online course recommendation method, apparatus, electronic device, and computer-readable storage medium based on a dual-tower graph convolutional neural network. Background Technology

[0002] Intelligent course recommendation technology is a deep learning algorithm-based technique used to recommend suitable online learning courses based on users' interests and needs. This technology utilizes deep learning algorithms to analyze users' historical learning behavior, personal preferences, learning goals, and other information to better understand their needs and interests, thereby recommending the most relevant and useful online courses. This technology can improve users' learning efficiency and experience, while also helping online education institutions better meet user needs and increase user retention and conversion rates. Improving the accuracy of course recommendations is a pressing issue that needs to be addressed.

[0003] In existing technologies, graph convolutional neural networks (GCNNs) are commonly used to implement intelligent course recommendation systems. GCNNs utilize graph structures to effectively extract features such as user behavior and provide reasonable recommendations. While existing course recommendation systems can meet certain user needs, traditional networks cannot effectively extract user features, and the models are prone to overfitting.

[0004] Therefore, one or more methods are needed to solve the above problems.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this disclosure is to provide an online course recommendation method, apparatus, electronic device, and computer-readable storage medium based on a dual-tower graph convolutional neural network, thereby overcoming at least to some extent one or more problems caused by the limitations and defects of related technologies.

[0007] According to one aspect of this disclosure, an online course recommendation method based on a dual-tower graph convolutional neural network is provided, comprising:

[0008] Acquire learning platform data and perform data preprocessing on the learning platform data to generate training sample data;

[0009] Using the training sample data containing metapaths as model input data and the course recommendation list as model output data, a basic model for an intelligent online course recommendation system based on graph convolutional neural networks is established.

[0010] The basic model of the online course intelligent recommendation system is generated by extracting features and optimizing network feature representation capabilities based on a preset dual-tower network structure, optimizing model output based on a self-consistent learning algorithm, and optimizing recommendation accuracy based on a weighted cold start strategy.

[0011] Based on the aforementioned online course intelligent recommendation system model, online course recommendations are performed using a dual-tower graph convolutional neural network.

[0012] In one exemplary embodiment of this disclosure, the data preprocessing in the method includes data desensitization processing of learning platform data, construction of heterogeneous graphs, extraction of meta-paths, and generation of training sample data.

[0013] In one exemplary embodiment of this disclosure, the method further includes:

[0014] The data anonymization process includes deleting personal information from the learning platform data to protect the personal information of students and teachers.

[0015] The construction of the heterogeneous graph includes identifying student, teacher, course, and knowledge point entities in the learning platform data to complete the construction of the heterogeneous graph;

[0016] The meta-path extraction process includes extracting the meta-paths of students and courses from the learning platform data based on a random walk algorithm, and generating training sample data.

[0017] In one exemplary embodiment of this disclosure, the method for establishing the basic model of an intelligent online course recommendation system based on graph convolutional neural networks further includes:

[0018] Using the training sample data containing metapaths as model input data and the course recommendation list as model output data, the training sample data containing metapaths is the set of metapaths. Convert to the corresponding adjacency matrix , where n represents the number of extracted meta-paths;

[0019] A multi-layer graph convolutional neural network model is used to learn representations of students and courses;

[0020] An adaptive mechanism is adopted, adding a mapping vector s that can be automatically learned by the model to assign weights to different graph convolutional layers, thereby adaptively adjusting the proportion of each graph convolutional layer in the output representation.

[0021] An attention mechanism is adopted to learn the importance of different meta-paths by utilizing the correlation between meta-paths, and the importance is used as the weight of different meta-paths to achieve the fusion of representations generated by different meta-paths.

[0022] Multiply the representation matrices of the students and courses to obtain a matrix containing the student ratings for each course. After removing the preset learning courses, select the top K courses as the recommendation list.

[0023] In one exemplary embodiment of this disclosure, the method extracts features of students and courses based on a preset dual-tower network structure and optimizes the network feature representation capability. The optimization of the basic model of the online course intelligent recommendation system further includes:

[0024] The basic model of the online course intelligent recommendation system is replicated in the form of a dual-tower system, and a unified optimization function is used to adjust the training loss of the model.

[0025] The dual-tower structure utilizes the Dropout algorithm to randomly zero out the feature representations of the training samples. The training loss of the model is:

[0026] .

[0027] In one exemplary embodiment of this disclosure, the method outputs a model based on a self-consistent learning algorithm, and the optimization of the basic model of the online course intelligent recommendation system further includes:

[0028] The similarity value between the outputs p1 and p2 of the dual-tower network is calculated based on the cosine similarity function, and the self-consistency optimization algorithm is as follows:

[0029] ,

[0030] .

[0031] In one exemplary embodiment of this disclosure, the basic model of the online course intelligent recommendation system is optimized based on a weighted cold-start strategy to improve recommendation accuracy. This optimization also includes automatically adjusted weights that gradually increase with the number of training steps, based on the weighted cold-start strategy. :

[0032] ;

[0033] in, Indicates the current training weight values; This represents the initial weight values ​​during training; t represents the final weight value; n represents the current training step number; n represents the total number of training steps of the model.

[0034] In one aspect of this disclosure, an online course recommendation device based on a dual-tower graph convolutional neural network is provided, comprising:

[0035] The training sample generation module is used to acquire learning platform data and perform data preprocessing on the learning platform data to generate training sample data.

[0036] The basic model building module is used to build a basic model for an online course intelligent recommendation system based on graph convolutional neural networks, using the training sample data containing meta-paths as model input data and the course recommendation list as model output data.

[0037] The basic model optimization module is used to optimize the feature extraction and feature representation capabilities of the basic model of the online course intelligent recommendation system based on a preset dual-tower network structure, optimize the model output based on a self-consistent learning algorithm, and optimize the recommendation accuracy based on a weighted cold start strategy, thereby generating an online course intelligent recommendation system model.

[0038] The online course recommendation module is used to recommend online courses based on the dual-tower graph convolutional neural network, according to the online course intelligent recommendation system model.

[0039] In one aspect of this disclosure, an electronic device is provided, comprising:

[0040] Processor; and

[0041] A memory storing computer-readable instructions that, when executed by the processor, implement the method according to any one of the preceding claims.

[0042] In one aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the method according to any one of the preceding claims.

[0043] An exemplary embodiment of this disclosure discloses an online course recommendation method based on a dual-tower graph convolutional neural network. The method includes: acquiring learning platform data and performing data preprocessing to generate training sample data; establishing a basic model using the training sample data containing meta-paths as model input data and a course recommendation list as model output data; extracting features from the basic model based on a preset dual-tower network structure and optimizing the network feature representation capability; optimizing the model output based on a self-consistent learning algorithm; optimizing the recommendation accuracy based on a weighted cold-start strategy; and generating an intelligent online course recommendation system model; and completing online course recommendation based on the intelligent online course recommendation system model. This disclosure combines a dual-tower network structure and a self-consistent learning algorithm to optimize the performance of the online course recommendation system from multiple levels, including model structure and model training.

[0044] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0045] The above and other features and advantages of this disclosure will become more apparent from the detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0046] Figure 1 A flowchart illustrating an online course recommendation method based on a dual-tower graph convolutional neural network according to an exemplary embodiment of the present disclosure is shown.

[0047] Figure 2 The illustration shows a heterogeneous graph formed after preprocessing learning platform data, according to an exemplary embodiment of the present disclosure, for an online course recommendation method based on a dual-tower graph convolutional neural network.

[0048] Figure 3 The illustration shows a schematic diagram of the meta-path extracted from a heterogeneous graph using a random walk algorithm in an online course recommendation method based on a dual-tower graph convolutional neural network according to an exemplary embodiment of the present disclosure;

[0049] Figure 4 The illustration shows a flowchart of an online course recommendation method based on a dual-tower graph convolutional neural network according to an exemplary embodiment of the present disclosure, which uses a graph convolutional neural network to generate student and course representations.

[0050] Figure 5 A schematic diagram of an intelligent course recommendation system model based on a dual-tower structure, according to an exemplary embodiment of the present disclosure, is shown.

[0051] Figure 6 The illustration shows a schematic diagram of an intelligent course recommendation system model based on a self-consistency learning algorithm, which is an online course recommendation method based on a dual-tower graph convolutional neural network according to an exemplary embodiment of the present disclosure.

[0052] Figure 7 The diagram illustrates a cold-start-based weight change in an online course recommendation method based on a dual-tower graph convolutional neural network according to an exemplary embodiment of the present disclosure.

[0053] Figure 8 A structural block diagram of an online course recommendation device based on a dual-tower graph convolutional neural network according to an exemplary embodiment of the present disclosure is shown.

[0054] Figure 9 A block diagram of an electronic device according to an exemplary embodiment of the present disclosure is schematically shown; and

[0055] Figure 10 The illustration shows a schematic diagram of a computer-readable storage medium according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0056] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0057] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced without one or more of the specific details described, or other methods, components, materials, apparatuses, steps, etc., can be employed. In other instances, well-known structures, methods, apparatuses, implementations, materials, or operations are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0058] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, or in one or more software-hardened modules, or in different network and / or processor devices and / or microcontroller devices.

[0059] In this example embodiment, an online course recommendation method based on a dual-tower graph convolutional neural network is first provided; see reference Figure 1 As shown, this online course recommendation method based on a dual-tower graph convolutional neural network may include the following steps:

[0060] Step S110: Obtain learning platform data and perform data preprocessing on the learning platform data to generate training sample data;

[0061] Step S120: Using the training sample data containing metapaths as model input data and the course recommendation list as model output data, a basic model for an online course intelligent recommendation system based on graph convolutional neural networks is established.

[0062] Step S130: Based on the preset dual-tower network structure, feature extraction and network feature representation capability are performed on the basic model of the online course intelligent recommendation system. The model output is optimized based on the self-consistent learning algorithm. The recommendation accuracy is optimized based on the weighted cold start strategy to generate the online course intelligent recommendation system model.

[0063] Step S140: Based on the online course intelligent recommendation system model, complete the online course recommendation based on the dual-tower graph convolutional neural network.

[0064] An exemplary embodiment of this disclosure discloses an online course recommendation method based on a dual-tower graph convolutional neural network. The method includes: acquiring learning platform data and performing data preprocessing to generate training sample data; establishing a basic model using the training sample data containing meta-paths as model input data and a course recommendation list as model output data; extracting features from the basic model based on a preset dual-tower network structure and optimizing the network feature representation capability; optimizing the model output based on a self-consistent learning algorithm; optimizing the recommendation accuracy based on a weighted cold-start strategy; and generating an intelligent online course recommendation system model; and completing online course recommendation based on the intelligent online course recommendation system model. This disclosure combines a dual-tower network structure and a self-consistent learning algorithm to optimize the performance of the online course recommendation system from multiple levels, including model structure and model training.

[0065] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0066] Heterogeneous graphs: Heterogeneous graphs are a general graph structure that can be used to represent various complex systems and relationships. A heterogeneous graph is a graph composed of nodes and edges of different types, where nodes and edges can have different properties and semantics. Given a heterogeneous graph... ,in It is a collection of entities. It is a collection of relationships. Furthermore, heterogeneous graphs also include entity type mapping functions. mapping functions for relation types ,in and Represents the set of types for defined entities and relations, and This invention uses heterogeneous graphs to represent four entities in a learning platform: students (S), teachers (T), courses (C), and knowledge points (K), as well as five relationships between these entities: learning, teaching, inclusion, following, and viewing. By using heterogeneous graphs to represent the data in the learning platform, richer semantic and attribute information can be preserved for better recommendation tasks.

[0067] Meta-path: A meta-path is a path type in a heterogeneous graph, used to describe the relationships and combinations between different types of nodes. It is defined as follows: It can also be abbreviated. Metapaths describe the semantic relationships between nodes by abstracting and combining the relationships between different types of nodes. In this invention, metapaths between students, teachers, courses, and knowledge points are extracted from heterogeneous graphs to better capture the semantic relationships between different entities. The metapath between two students can be defined as follows: This indicates that two different students became connected because they viewed the same knowledge point. This indicates that two students became connected because they studied different courses taught by the same teacher.

[0068] Graph Neural Network (Graph Neural Network) models are machine learning models that learn and reason based on graph-structured data. Unlike traditional neural networks that process vector or matrix data, Graph Neural Networks can handle graph data, such as e-commerce networks, social networks, and biological networks. This type of model can capture the complex relationships and structural information within graph data and transform it into vector representations through node embedding and graph embedding. These vector representations can then be used for downstream tasks such as node classification, link prediction, and graph classification.

[0069] Dropout: Dropout is a regularization technique widely used in deep neural networks. It randomly sets the output of a subset of neurons in the network to 0 during training, thereby preventing overfitting. Specifically, in each training iteration, Dropout randomly selects a subset of neurons with a certain probability (usually 0.2-0.5) and sets their output to 0. These randomly selected neurons will not participate in the forward and backward propagation of that iteration.

[0070] Dual-Tower Structure: The dual-tower structure is a neural network architecture whose core idea is to extract features from the input by sharing weights, and then compare or calculate the similarity of the extracted features. The model's name comes from the shape of its neural network diagram, which resembles two parallel towers, each tower being an independent neural network with shared weights.

[0071] The following will further explain an online course recommendation method based on a dual-tower graph convolutional neural network in this example embodiment.

[0072] Example 1:

[0073] In step S110, learning platform data can be acquired, and the learning platform data can be preprocessed to generate training sample data.

[0074] In this example embodiment, the data preprocessing in the method includes data anonymization of the learning platform data, construction of heterogeneous graphs, extraction of meta-paths, and generation of training sample data.

[0075] In this example embodiment, the method further includes:

[0076] The data anonymization process includes deleting personal information from the learning platform data to protect the personal information of students and teachers.

[0077] The construction of the heterogeneous graph includes identifying student, teacher, course, and knowledge point entities in the learning platform data to complete the construction of the heterogeneous graph;

[0078] The meta-path extraction process includes extracting the meta-paths of students and courses from the learning platform data based on a random walk algorithm, and generating training sample data.

[0079] In step S120, the training sample data containing metapaths can be used as model input data, and the course recommendation list can be used as model output data to establish a basic model for an online course intelligent recommendation system based on graph convolutional neural networks.

[0080] In this example embodiment, the method for establishing the basic model of the online course intelligent recommendation system based on graph convolutional neural networks further includes:

[0081] Using the training sample data containing metapaths as model input data and the course recommendation list as model output data, the training sample data containing metapaths is the set of metapaths. Convert to the corresponding adjacency matrix , where n represents the number of extracted meta-paths;

[0082] A multi-layer graph convolutional neural network model is used to learn representations of students and courses;

[0083] An adaptive mechanism is adopted, adding a mapping vector s that can be automatically learned by the model to assign weights to different graph convolutional layers, thereby adaptively adjusting the proportion of each graph convolutional layer in the output representation.

[0084] An attention mechanism is adopted to learn the importance of different meta-paths by utilizing the correlation between meta-paths, and the importance is used as the weight of different meta-paths to achieve the fusion of representations generated by different meta-paths.

[0085] Multiply the student and course representation matrices to obtain a matrix containing student ratings for each course. After removing the preset learning courses, select the top course as the recommendation list.

[0086] In step S130, the basic model of the online course intelligent recommendation system can be subjected to feature extraction and optimization of network feature representation capabilities based on a preset dual-tower network structure, model output optimization based on a self-consistent learning algorithm, and recommendation accuracy optimization based on a weighted cold start strategy to generate an online course intelligent recommendation system model.

[0087] In this example embodiment, the method extracts features of students and courses based on a preset dual-tower network structure and optimizes the network feature representation capability. The optimization of the basic model of the online course intelligent recommendation system further includes:

[0088] The basic model of the online course intelligent recommendation system is replicated in the form of a dual-tower system, and a unified optimization function is used to adjust the training loss of the model.

[0089] The dual-tower structure utilizes the Dropout algorithm to randomly zero out the feature representations of the training samples. The training loss of the model is:

[0090] .

[0091] In this example embodiment, the method outputs the model based on a self-consistent learning algorithm, and the optimization of the basic model of the online course intelligent recommendation system further includes:

[0092] The similarity value between the outputs p1 and p2 of the dual-tower network is calculated based on the cosine similarity function, and the self-consistency optimization algorithm is as follows:

[0093] ,

[0094] .

[0095] In this example embodiment, the basic model of the online course intelligent recommendation system is optimized for recommendation accuracy based on a weighted cold start strategy. This optimization also includes automatically adjusted weights that gradually increase with the number of training steps, based on the weighted cold start strategy. :

[0096] ;

[0097] in, Indicates the current training weight values; This represents the initial weight values ​​during training; t represents the final weight value; n represents the current training step number; n represents the total number of training steps of the model.

[0098] In step S140, online course recommendation based on a dual-tower graph convolutional neural network can be completed based on the online course intelligent recommendation system model.

[0099] Example 2:

[0100] In this example embodiment, an online course recommendation technology based on graph convolutional neural networks is proposed. A heterogeneous graph is used to represent the data of the online learning platform, and a random walk algorithm is used to extract meta-paths as training data. Subsequently, a graph convolutional neural network is used to learn student and course representations, and relevant course recommendations are made based on the learned representations. This improves the accuracy of the course recommendation technology in multiple dimensions at both the data and model levels. The technology includes the following steps S1-S3:

[0101] S1 acquires data from the learning platform and performs data preprocessing on the learning platform data;

[0102] Specifically, data preprocessing includes data anonymization, construction of heterogeneous graphs, and extraction of meta-paths.

[0103] S101 data anonymization includes: removing information directly related to the personal identity of students or teachers from the learning record data, such as names, student IDs, employee IDs, email addresses, and phone numbers. This ensures that only data related to learning behavior is retained. For each learning record, the student and teacher's identity is replaced with a unique identifier. This prevents the data from being directly associated with a specific individual. Subsequently, histograms, box plots, or kernel density estimation methods are used to create a data distribution map, and the graphs before and after anonymization are compared to ensure that the general shape and characteristics of the data distribution remain consistent to meet subsequent business needs.

[0104] S102 Constructing a heterogeneous graph includes: identifying student, teacher, and course entities in the online learning platform data, and establishing knowledge point entities. Then, relationships of five types are established for the four entities: learning, teaching, containing, following, and viewing. For example... Figure 2 As shown, a heterogeneous graph containing four entity types and five relationship types is constructed.

[0105] S103 Extracting meta-paths includes: extracting meta-paths on the constructed heterogeneous graph using a random walk algorithm to generate sample data for model training. For example... Figure 3 As shown, meta-paths about students and courses are extracted and used as training sample data.

[0106] In the embodiments of this example, as Figure 4 As shown, S2 uses the extracted meta-paths as input data to the graph convolutional neural network and the output recommendation list as output data to establish an intelligent recommendation system model for online courses based on graph convolutional neural networks.

[0107] S201 will extract the metapath set Convert to the corresponding adjacency matrix , where n represents the number of metapaths.

[0108] S202 We employ a multi-layer graph convolutional neural network model to learn representations of students and courses under different meta-paths. The layered graph convolution model is defined as follows:

[0109] ;

[0110] in, express Entity representation of layers, A is the adjacency matrix corresponding to a specific metapath, and self-connections are added by adding it to the identity matrix I. It is a trainable weight matrix. It is a non-linear activation function.

[0111] S203 employs an adaptive mechanism to balance the proportion of representations from different graph convolutional layers, avoiding oversmoothing. Specifically, it adds a mapping vector that can be automatically learned by the model to assign weights to different layers, defined as follows:

[0112] ;

[0113] in, is a trainable mapping vector, and d is the dimension of the representation vector. By adding mapping vectors, the model can automatically adjust the outputs from different layers, thereby avoiding the oversmoothing phenomenon in deep graph convolutional neural networks.

[0114] S204 employs an attention mechanism, utilizing the correlation between meta-paths to learn attention weights, in order to better integrate the representations generated by different meta-paths.

[0115] ;

[0116] in, Graph convolutional neural networks based on metapaths The generated entity representation, Let H represent the attention weights, and H represent the final entity representation that incorporates the attention weights from different meta-paths. In the course recommendation task, we mainly focus on student and course entities, so we primarily generate representations for these two entities. Typically, given each corresponding meta-path... entity representation We define the attention weight calculation method as follows:

[0117] ;

[0118] in, For entity representations generated based on the target meta-path, This refers to the entity representation generated from other meta-paths. x represents a trainable attention vector. This represents a non-linear activation function. Learning attention weights based on the correlation of the meta-paths themselves allows for a better inference of the importance of different meta-paths. Using this importance as attention weights enables better learning of entity representations.

[0119] After obtaining the representations of students and courses, S205 uses a rating matrix to implement the course recommendation task for students. We consider the number of times a student clicks on a course as the rating matrix. The student's rating for a course can be defined as follows:

[0120] ;

[0121] in, For the vector representation of students, Let h be the vector representation of the course. h is the vector dimension, and n and m are the number of students and courses, respectively. The row vectors in the rating matrix represent each user's rating of the course. After removing previously accessed courses, the top K courses are selected as the final recommendation list.

[0122] In this example embodiment, S3, building upon the previous step, the present invention will design an online course recommendation system model using a dual-tower graph convolutional neural network. For example... Figure 5 We replicate the single recommender system network structure as a dual-tower design, using a unified optimization function to adjust the model's loss value. The dual-tower structure, by processing the input separately, can better extract sample features, thereby improving the overall model's feature extraction capability. The dual-tower structure is very useful for tasks requiring multi-faceted information, such as semantic matching and recommender systems. Furthermore, the dual-tower model can utilize more information during training to optimize the model, thus improving its generalization ability. In this example, the dual-tower structure will utilize the Dropout algorithm to randomly zero out the feature representations of the training samples. This algorithm randomly alters the feature space of the training samples, increasing uncertainty in a finite number of samples and thus improving feature diversity. The training loss of the model at this point is:

[0123] .

[0124] S301 designs a novel self-consistent learning algorithm to optimize the performance of online course recommendation systems using graph convolutional neural networks.

[0125] Traditional graph convolutional neural networks often use Dropout in their layers to avoid overfitting. Dropout is a regularization method commonly used in neural network models. It reduces dependencies between neurons by randomly setting their outputs to zero, thus mitigating the risk of overfitting. With Dropout, each neuron learns to generate useful features even when other neurons are absent, increasing the model's robustness. While Dropout effectively avoids overfitting, its use only during training can lead to inconsistencies between training and inference processes. These inconsistencies can negatively impact model performance. Figure 6As shown, in this example, a novel self-consistent learning algorithm will be designed by combining the S3 dual-tower network structure to optimize the Dropout problem and thus improve the model's performance. The model optimization function is as follows:

[0126] ,

[0127] ;

[0128] Where p1 and p2 represent the training outputs of the dual-tower model, respectively. The cosine similarity weight is used to calculate the similarity value between the outputs p1 and p2 of the dual-tower network. As shown in the formula above, the more similar the model outputs p1 and p2 are, the smaller their spatial distance. This effectively constrains output inconsistency. In the self-consistency learning algorithm, this invention uses the cosine similarity function to calculate the similarity value between the outputs p1 and p2 of the dual-tower network, and optimizes this similarity value through model training, gradually increasing it. In other words, the goal of self-consistency learning is to make the feature outputs p1 and p2 more similar, ensuring consistency between the model training and inference phases, thereby eliminating the impact of the Dropout inconsistency problem.

[0129] In S302, to optimize consistency between the model training and inference phases in S301, we introduced a self-consistent learning algorithm. In the self-consistent algorithm, the weight of cosine similarity... It requires manual settings, although appropriate ones Values ​​can significantly improve model performance, but exploring suitable values ​​is crucial. Repeated experiments would undoubtedly increase the training cost of the model. Therefore, in this invention, in order to set more efficiently... To address this, we designed a weighted cold-start strategy that automatically optimizes the weights based on the number of training steps. This effectively reduced the weight. Trial and error cost:

[0130] ;

[0131] in, Indicates the current training weight values; This represents the initial weight values ​​during training; The final weight value is represented by t; the current training step number is represented by t; and the total number of training steps for the model is represented by n. The weights designed in this invention... The cold start strategy uses a small weight value in the initial stage of training, and the weight increases as the number of training steps increases. As the number of weights increases, the constraints on the model also increase. This cold start strategy is used because the model is unstable in the early stages of training, and higher weights... While it can constrain inconsistencies, it can affect the model's feature learning. The cold start strategy, however, dynamically adjusts the weights. The value can effectively constrain the model. Furthermore, such as... Figure 7 As shown, compared to traditional fixed weights, the cold start strategy is more flexible and does not require manual weight setting. It also has the characteristics of automation.

[0132] In this example embodiment, the present invention employs a dual-tower graph convolutional neural network structure and a self-consistent learning algorithm to optimize the loss function of an online course recommendation system model. Starting from the practical problems of course recommendation systems, the present invention utilizes a dual-tower network structure and a self-consistent learning algorithm to optimize the model's loss function. Furthermore, addressing the weight setting problem of the self-consistent learning algorithm, the present invention designs a cold-start strategy. This strategy not only improves model performance but also avoids the problem of repeated experiments, and is more intelligent and automated. The advantages of the present invention are: First, it constructs a heterogeneous graph containing four entities—learning users, instructors, learning courses, and related knowledge concepts—based on learning platform data, and uses a graph neural network to extract user representations, thereby improving the accuracy of the course recommendation system to a certain extent. Second, to improve the feature extraction capability of the graph convolutional neural network, the present invention designs a dual-tower network structure and innovatively proposes a self-consistent learning algorithm based on cosine similarity to optimize the inconsistency problem during model training and inference. Finally, the present invention designs a cold-start strategy to optimize the setting of loss function weights, making it more intelligent; it solves the problems of traditional networks being unable to effectively extract user features and the model being prone to overfitting.

[0133] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0134] Furthermore, in this example embodiment, an online course recommendation device based on a dual-tower graph convolutional neural network is also provided. (See reference...) Figure 8 As shown, the online course recommendation device 200 based on a dual-tower graph convolutional neural network may include: a training sample generation module 210, a basic model building module 220, a basic model optimization module 230, and an online course recommendation module 240. Wherein:

[0135] The training sample generation module 210 is used to acquire learning platform data and perform data preprocessing on the learning platform data to generate training sample data.

[0136] The basic model building module 220 is used to build a basic model of an online course intelligent recommendation system based on graph convolutional neural networks, using the training sample data containing meta-paths as model input data and the course recommendation list as model output data.

[0137] The basic model optimization module 230 is used to optimize the feature extraction and feature representation capabilities of the basic model of the online course intelligent recommendation system based on a preset dual-tower network structure, optimize the model output based on a self-consistent learning algorithm, and optimize the recommendation accuracy based on a weighted cold start strategy, thereby generating an online course intelligent recommendation system model.

[0138] The online course recommendation module 240 is used to complete online course recommendation based on the dual-tower graph convolutional neural network based on the online course intelligent recommendation system model.

[0139] The specific details of each of the above-mentioned online course recommendation device modules based on dual-tower graph convolutional neural networks have been described in detail in the corresponding online course recommendation method based on dual-tower graph convolutional neural networks, so they will not be repeated here.

[0140] It should be noted that although several modules or units of the online course recommendation device 200 based on a dual-tower graph convolutional neural network have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0141] Furthermore, in an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0142] Those skilled in the art will understand that various aspects of the present invention can be implemented as systems, methods, or program products. Therefore, various aspects of the present invention can be specifically implemented as entirely hardware embodiments, entirely software embodiments (including firmware, microcode, etc.), or embodiments combining hardware and software aspects, collectively referred to herein as “circuit,” “module,” or “system.”

[0143] The following reference Figure 9 To describe an electronic device 300 according to such an embodiment of the present invention. Figure 9 The electronic device 300 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0144] like Figure 9As shown, the electronic device 300 is presented in the form of a general-purpose computing device. The components of the electronic device 300 may include, but are not limited to: at least one processing unit 310, at least one storage unit 320, a bus 330 connecting different system components (including storage unit 320 and processing unit 310), and a display unit 340.

[0145] The storage unit stores program code that can be executed by the processing unit 310, causing the processing unit 310 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 310 can perform actions such as... Figure 1 Steps S110 to S140 are shown in the diagram.

[0146] Storage unit 320 may include readable media in the form of volatile storage units, such as random access memory (RAM) 3201 and / or cache memory 3202, and may further include read-only memory (ROM) 3203.

[0147] Storage unit 320 may also include a program / utility 3204 having a set (at least one) program module 3205, such program module 3205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0148] Bus 330 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0149] Electronic device 300 can also communicate with one or more external devices 370 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 300, and / or with any device that enables electronic device 300 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 350. Furthermore, electronic device 300 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 360. As shown, network adapter 360 communicates with other modules of electronic device 300 via bus 330. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 300, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0150] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0151] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of the invention may also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the invention described in the "Exemplary Methods" section above.

[0152] refer to Figure 10 As shown, a program product 400 for implementing the above-described method according to an embodiment of the present invention is described. This product may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0153] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0154] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0155] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0156] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0157] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0158] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0159] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. An online course recommendation method based on a dual-tower graph convolutional neural network, characterized in that, The method includes: Acquire learning platform data and perform data preprocessing on the learning platform data to generate training sample data; Using the training sample data containing meta-paths as model input data and the course recommendation list as model output data, a basic model for an intelligent online course recommendation system based on graph convolutional neural networks is established. This model also includes: Using the training sample data containing metapaths as model input data and the course recommendation list as model output data, the training sample data containing metapaths is the set of metapaths. Convert to the corresponding adjacency matrix Where n represents the number of extracted meta-paths; A multi-layer graph convolutional neural network model is used to learn representations of students and courses; An adaptive mechanism is adopted, adding a mapping vector s that can be automatically learned by the model to assign weights to different graph convolutional layers, thereby adaptively adjusting the proportion of each graph convolutional layer in the output representation. An attention mechanism is adopted to learn the importance of different meta-paths by utilizing the correlation between meta-paths, and the importance is used as the weight of different meta-paths to achieve the fusion of representations generated by different meta-paths. Multiply the representation matrices of the students and courses to obtain a matrix containing the student's rating for each course. After removing the preset learning courses, select the top K courses as the recommendation list. The basic model of the online course intelligent recommendation system, based on a preset dual-tower network structure, further includes feature extraction and optimization of network feature representation capabilities, and also includes: The basic model of the online course intelligent recommendation system is replicated in the form of a dual-tower system, and a unified optimization function is used to adjust the training loss of the model. In the dual-tower network structure, the Dropout algorithm will be used to randomly zero out the feature representations of the training samples. The training loss of the model is: Where, loss model The overall loss for the dual-tower model is... model1 loss model2 Let p represent the losses of the two models in the twin-tower structure. i y is the model training output in the dual-tower network structure, where y is the true data label and M is the number of training samples. Model output optimization based on self-consistency learning algorithms also includes: The similarity value between the outputs p1 and p2 of the dual-tower network is calculated based on the cosine similarity function, and the self-consistency optimization algorithm for the similarity value is optimized through model training as follows: Wherein, function E is used to calculate the similarity value between the outputs of two models in the double tower structure using the cosine similarity function, p1 and p2 are the training outputs of the two models in the double tower structure, Loss is the loss obtained by the self-consistency algorithm, and β is the weight of the similarity value. The recommendation accuracy is optimized based on a weighted cold start strategy, generating an intelligent recommendation system model for online courses; Based on the aforementioned online course intelligent recommendation system model, online course recommendations are performed using a dual-tower graph convolutional neural network.

2. The method as described in claim 1, characterized in that, The data preprocessing in the method includes data anonymization of the learning platform data, construction of heterogeneous graphs, extraction of meta-paths, and generation of training sample data.

3. The method as described in claim 2, characterized in that, The method further includes: The data anonymization process includes deleting personal information from the learning platform data to protect the personal information of students and teachers. The construction of the heterogeneous graph includes identifying student, teacher, course, and knowledge point entities in the learning platform data to complete the construction of the heterogeneous graph; The meta-path extraction process includes extracting the meta-paths of students and courses from the learning platform data based on a random walk algorithm, and generating training sample data.

4. The method as described in claim 1, characterized in that, The online course intelligent recommendation system's basic model is optimized based on a weighted cold start strategy to improve recommendation accuracy. This optimization also includes a weight β that gradually increases with the number of training steps, automatically optimized based on the weighted cold start strategy. Where, β t β represents the current training weight values; β0 represents the initial training weight values; β e t represents the final weight value; n represents the current training step number; n represents the total number of training steps of the model.

5. An online course recommendation device based on a dual-tower graph convolutional neural network, characterized in that, The device includes: The training sample generation module is used to acquire learning platform data and perform data preprocessing on the learning platform data to generate training sample data. The basic model building module is used to build a basic model for an intelligent online course recommendation system based on graph convolutional neural networks, using the training sample data containing meta-paths as model input data and the course recommendation list as model output data. It also includes: Using the training sample data containing metapaths as model input data and the course recommendation list as model output data, the training sample data containing metapaths is the set of metapaths. Convert to the corresponding adjacency matrix Where n represents the number of extracted meta-paths; A multi-layer graph convolutional neural network model is used to learn representations of students and courses; An adaptive mechanism is adopted, adding a mapping vector s that can be automatically learned by the model to assign weights to different graph convolutional layers, thereby adaptively adjusting the proportion of each graph convolutional layer in the output representation. An attention mechanism is adopted to learn the importance of different meta-paths by utilizing the correlation between meta-paths, and the importance is used as the weight of different meta-paths to achieve the fusion of representations generated by different meta-paths. Multiply the representation matrices of the students and courses to obtain a matrix containing the student's rating for each course. After removing the preset learning courses, select the top K courses as the recommendation list. The basic model optimization module is used to optimize the feature extraction and feature representation capabilities of the basic model of the online course intelligent recommendation system based on a preset dual-tower network structure, and also includes: The basic model of the online course intelligent recommendation system is replicated in the form of a dual-tower system, and a unified optimization function is used to adjust the training loss of the model. In the dual-tower network structure, the Dropout algorithm will be used to randomly zero out the feature representations of the training samples. The training loss of the model is: Where, loss model The overall loss for the dual-tower model is... model1 loss model2 Let p represent the losses of the two models in the twin-tower structure. i y is the model training output in the dual-tower network structure, where y is the true data label and M is the number of training samples. Model output optimization based on self-consistency learning algorithms also includes: The similarity value between the outputs p1 and p2 of the dual-tower network is calculated based on the cosine similarity function, and the self-consistency optimization algorithm for the similarity value is optimized through model training as follows: Wherein, function E is used to calculate the similarity value between the outputs of two models in the double tower structure using the cosine similarity function, p1 and p2 are the training outputs of the two models in the double tower structure, Loss is the loss obtained by the self-consistency algorithm, and β is the weight of the similarity value. The recommendation accuracy is optimized based on a weighted cold start strategy, generating an intelligent recommendation system model for online courses; The online course recommendation module is used to recommend online courses based on the dual-tower graph convolutional neural network, according to the online course intelligent recommendation system model.

6. An electronic device, characterized in that, The method includes a processor; and a memory storing computer-readable instructions that, when executed by the processor, implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 4.

Images