A method, device, equipment, storage medium and program product for course session recommendation

Abstract

The present invention discloses a method, apparatus, device, storage medium and program product for course session recommendation. The method includes: performing channel splicing on the semantic embedding matrix of course text description data and the user behavior feature matrix to generate a multi-dimensional feature matrix; constructing an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix, and performing graph convolution operations based on the adjacency graph and the hypergraph to generate target embedding representation vectors of each course node; generating attention weights for each course node, and performing weighted aggregation on the target embedding representation vectors based on the attention weights, and performing course session recommendation based on the weighted aggregated target embedding representation vectors; because the present invention effectively improves the feature expression ability of course nodes by fusing course semantic features and user behavior features, captures local neighborhood features between courses and global features in the hypergraph structure, accurately captures the dynamic changes of user interests, and thus realizes accurate course session recommendation.

Description

Technical Field

[0001] The present invention relates to the technical field of online education, and in particular to a course conversation recommendation method, device, equipment, storage medium and program product. Background Art

[0002] With the rapid development of the online education industry, a large number of complex and diverse course resources have emerged on educational platforms. How to mine courses that users are interested in from the vast amount of course resources has become a key issue in the research of online education recommendation systems. In recent years, recommendation systems have gradually been applied to the field of online education, and various course recommendation methods for online learning environments have been proposed to generate more personalized course recommendations. As an important branch of the recommendation system, conversation recommendation is suitable for the characteristics of diverse and real-time changing user needs in online education scenarios.

[0003] However, existing conversation recommendation models still have some limitations. They mainly rely on the interaction data between users and courses, ignoring the rich semantic information contained in course resources, resulting in the lack of in-depth understanding of the characteristics of course content in the recommendation results and making it difficult to capture the potential semantic associations between courses. Moreover, when dealing with complex course relationships, existing conversation recommendation models only focus on local information and cannot capture the interaction relationships between multiple nodes. Therefore, they fail to effectively capture the complex associations between courses and cannot mine the implicit feature relationships between nodes, leading to inaccurate course recommendations. Summary of the Invention

[0004] The main purpose of the present invention is to provide a course conversation recommendation method, device, equipment, storage medium and program product, aiming to solve the technical problem that the existing technology fails to effectively utilize course semantic information, capture the complex associations and implicit feature connections between courses, resulting in inaccurate course recommendations.

[0005] To achieve the above object, the present invention provides a course conversation recommendation method, and the method includes the following steps:

[0006] Input the text description data of the course into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector, and generate a semantic embedding matrix based on the semantic feature vector;

[0007] Conduct user behavior analysis on the historical learning interaction data of the user, and construct a user behavior feature matrix according to the results of the user behavior analysis;

[0008] Perform channel splicing on the semantic embedding matrix and the user behavior feature matrix to generate a multi-dimensional feature matrix;

[0009] Construct an adjacency graph and a hypergraph among courses according to the multi-dimensional feature matrix, where the adjacency graph includes the adjacency relationships among each course node, and the hypergraph includes the course nodes and the interaction relationships between users and each course node;

[0010] Perform graph convolution operations based on the adjacency graph and the hypergraph to generate the target embedding representation vectors of each course node;

[0011] Generate the attention weights of each course node, perform weighted aggregation on the target embedding representation vectors based on the attention weights, and perform course session recommendation based on the weighted aggregated target embedding representation vectors.

[0012] Optionally, the inputting the text description data of the course into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors, and generating a semantic embedding matrix based on the semantic feature vectors includes:

[0013] Collect the text description data of the course and perform preprocessing on the text description data, where the preprocessing includes word segmentation, stop word filtering, and special character filtering;

[0014] Input the preprocessed text description data into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors;

[0015] Integrate the semantic feature vectors of each course for semantic embedding to generate an initial semantic matrix;

[0016] Perform normalization processing on the semantic feature vectors of each dimension in the initial semantic matrix to generate a semantic embedding matrix.

[0017] Optionally, the performing user behavior analysis on the historical learning interaction data of the user and constructing a user behavior feature matrix according to the user behavior analysis results includes:

[0018] Model the learning behavior of the user as a course learning session sequence in chronological order based on the historical learning interaction data of the user, where the course learning session sequence includes the set of courses learned by the user;

[0019] Obtain the chronological feature of the user's course learning according to the sequence position encoding information of the course learning session sequence;

[0020] Analyze the interaction frequency between the user and each course according to the course learning session sequence, and generate an explicit relationship weight matrix based on the interaction frequency, where the elements in the explicit relationship weight matrix represent the occurrence frequencies of different courses in the same session;

[0021] Map the course ID corresponding to each course to an initial behavior embedding vector, and construct an initial behavior feature matrix based on the initial behavior embedding vector.

[0022] Weighted aggregation is performed on the initial behavior feature matrix based on the chronological features of the course learning time and the explicit relationship weight matrix to generate enhanced behavior features:

[0023]

[0024] Wherein, represents the enhanced behavior feature, represents the set of neighbor courses that appear in the same session as course ; represents the position encoding of course ; represents the frequency of appearance of course in the same session as course ; represents the initial behavior embedding vector of course ;

[0025] Construct a user behavior feature matrix based on the enhanced behavior features of each course.

[0026] Optionally, the graph convolution operation based on the adjacency graph and the hypergraph to generate the target embedding representation vectors of each course node includes:

[0027] Perform neighborhood aggregation operation on the adjacency graph through a graph convolutional network to obtain the short-range dependency relationships between courses;

[0028] Obtain the local feature information of each course according to the short-range dependency relationships;

[0029] Perform graph convolution operation on the hypergraph through a graph convolutional network to obtain hyperedge information;

[0030] Obtain the interaction relationships between students and each course based on the hyperedge information, and obtain the potential feature information of the courses in the learning path based on the interaction relationships;

[0031] Fuse the local feature information and the potential feature information to obtain fused features;

[0032] Perform activation and normalization processing on the fused features to generate the target embedding representation vectors of each course node.

[0033] Optionally, generate the attention weights of each course node, perform weighted aggregation on the target embedding representation vectors based on the attention weights, and perform course session recommendation based on the weighted aggregated target embedding representation vectors, including:

[0034] Determine the cosine similarity between each course node according to the semantic feature vectors of each course node:

[0035]

[0036] Among them, represents the cosine similarity i between node j and node and respectively represent the semantic feature vectors i of node j and node

[0037] Determine the semantic difference information between each course node according to the cosine similarity;

[0038] Generate the attention weights of each course node based on the semantic difference information and the position encoding information of each course node:

[0039]

[0040] Among them, represents the attention weight, represents the position encoding information;

[0041] Generate the cluster center of the target embedding representation vector;

[0042] Calculate the cluster similarity between each course node and the cluster center;

[0043] Perform weighted aggregation on the target embedding representation vector based on the cluster similarity and the attention weight:

[0044]

[0045] Among them, represents the target embedding representation vector after weighted aggregation;

[0046] Generate the behavior feature vector of the user based on the user's historical learning interaction data;

[0047] Input the target embedding representation vector after weighted aggregation and the behavior feature vector into a pre-trained recommendation model for course session recommendation.

[0048] Optionally, the inputting the target embedding representation vector after weighted aggregation and the behavior feature vector into a pre-trained recommendation model for course session recommendation includes:

[0049] Input the target embedding representation vector after weighted aggregation and the behavior feature vector into a pre-trained recommendation model to calculate the predicted scores of each course node:

[0050]

[0051] Among them, represents the predicted score, represents the behavioral feature vector, represents the course embedding feature matrix constructed from the target embedding representation vector after weighted aggregation;

[0052] Determine the label difference information between the predicted score and the true label based on the cross - entropy loss function:

[0053]

[0054] where, represents the label difference information, represents the true label, represents the predicted score;

[0055] Update the model parameters of the recommendation model through the backpropagation algorithm:

[0056]

[0057] where, η represents the learning rate, represents the gradient of the loss function, represents the model parameters before update, represents the model parameters after update;

[0058] Optimize the recommendation model according to the updated model parameters, and perform course session recommendation based on the optimized recommendation model.

[0059] In addition, to achieve the above - mentioned purpose, the present invention also proposes a course session recommendation device, and the course session recommendation device includes:

[0060] A semantic analysis module, configured to input the text description data of the course into a pre - trained deep language model for semantic quantization, obtain a semantic feature vector, and generate a semantic embedding matrix based on the semantic feature vector;

[0061] A user behavior analysis module, configured to perform user behavior analysis on the user's historical learning interaction data, and construct a user behavior feature matrix according to the user behavior analysis results;

[0062] A feature splicing module, configured to perform channel splicing on the semantic embedding matrix and the user behavior feature matrix to generate a multi - dimensional feature matrix;

[0063] A graph construction module, configured to construct an adjacency graph and a hypergraph between courses according to the multi - dimensional feature matrix, where the adjacency graph includes the adjacency relationship between each course node, and the hypergraph includes the course nodes and the interaction relationship between the user and each course node;

[0064] A graph convolution processing module, configured to perform graph convolution operations based on the adjacency graph and the hypergraph to generate target embedding representation vectors for each course node;

[0065] A course recommendation module, configured to generate attention weights for each course node, perform weighted aggregation on the target embedding representation vectors based on the attention weights, and perform course session recommendation based on the weighted aggregated target embedding representation vectors.

[0066] In addition, to achieve the above object, the present application also provides a course session recommendation device, where the device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the course session recommendation method as described above.

[0067] In addition, to achieve the above object, the present application also provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the steps of the course session recommendation method as described above are implemented.

[0068] In addition, to achieve the above object, the present application also provides a computer program product, where the computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the course session recommendation method as described above are implemented.

[0069] The present invention inputs the text description data of courses into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors, and generates a semantic embedding matrix based on the semantic feature vectors; performs user behavior analysis on the historical learning interaction data of users, and constructs a user behavior feature matrix according to the results of the user behavior analysis; performs channel splicing on the semantic embedding matrix and the user behavior feature matrix to generate a multi-dimensional feature matrix; constructs an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix, where the adjacency graph includes the adjacency relationships between each course node, and the hypergraph includes the course nodes and the interaction relationships between users and each course node; performs graph convolution operations based on the adjacency graph and the hypergraph to generate target embedding representation vectors for each course node; generates attention weights for each course node, and performs weighted aggregation on the target embedding representation vectors based on the attention weights, and performs course session recommendation based on the weighted aggregated target embedding representation vectors; since the present invention fuses course semantic features and user behavior features, thereby improving the feature expression ability by fusing multi-source information, constructs an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix, thereby effectively improving the feature expression ability of course nodes in the graph structure, effectively capturing the multi-node interaction relationships in the hypergraph structure and the neighborhood local feature interaction relationships of each course node in the adjacency graph, performs graph convolution operations based on the adjacency graph and the hypergraph, thereby accurately mining the explicit feature relationships and implicit feature relationships between course nodes, realizing accurate capture of the dynamic changes of user interests, and thus accurately performing course session recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0071] Figure 1 It is a schematic structural diagram of a course session recommendation device for the hardware operating environment involved in the embodiment solution of the present invention;

[0072] Figure 2 It is a schematic flowchart of the first embodiment of the course session recommendation method of the present invention;

[0073] Figure 3 It is a schematic flowchart of the second embodiment of the course session recommendation method of the present invention;

[0074] Figure 4 It is a schematic flowchart of the third embodiment of the course session recommendation method of the present invention;

[0075] Figure 5Schematic flowchart of the fourth embodiment of the course session recommendation method of the present invention;

[0076] Figure 6 Schematic flowchart of the fifth embodiment of the course session recommendation method of the present invention;

[0077] Figure 7 Block diagram of the structure of the first embodiment of the course session recommendation device of the present invention.

[0078] The realization, functional characteristics and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0079] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0080] Refer to Figure 1 , Figure 1 Schematic diagram of the structure of the course session recommendation device in the hardware operating environment involved in the embodiment solution of the present invention.

[0081] As Figure 1 shown, the course session recommendation device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless-fidelity (WI-FI) interface). The memory 1005 may be a high-speed random access memory (RandomAccess Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0082] Those skilled in the art can understand that Figure 1 the structure shown in

[0083] does not constitute a limitation on the course session recommendation device, and may include more or fewer components than shown in the figure, or combine some components, or arrange different components. Figure 1As shown in the figure, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a course session recommendation program.

[0084] In Figure 1 In the course session recommendation device shown in the figure, the network interface 1004 is mainly used for data communication with a network server; the user interface 1003 is mainly used for data interaction with a user; the processor 1001 and the memory 1005 in the course session recommendation device of the present invention may be arranged in the course session recommendation device. The course session recommendation device calls the course session recommendation program stored in the memory 1005 through the processor 1001 and executes the course session recommendation method provided by the embodiments of the present invention.

[0085] The embodiments of the present invention provide a course session recommendation method. Referring to Figure 2 , Figure 2 is a schematic flowchart of the first embodiment of the course session recommendation method of the present invention.

[0086] In the first embodiment of the course session recommendation method of the present invention, the course session recommendation method includes the following steps:

[0087] Step S10: Input the text description data of a course into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector, and generate a semantic embedding matrix based on the semantic feature vector.

[0088] It should be understood that the execution subject of this embodiment may be a computing service device with data processing, network communication, and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or a terminal electronic device capable of implementing the above functions. Hereinafter, a course session recommendation device (abbreviated as a recommendation device) is used as an example to illustrate this embodiment and the following embodiments.

[0089] It can be understood that in this embodiment, a pre-trained deep language model (such as a BERT model) can be used to convert the text description of a course (including the title, introduction, and content) into a high-dimensional semantic vector and generate a semantic embedding matrix of the course.

[0090] In some embodiments, the recommendation device can extract semantic information from the text description of the courses through the semantic generation module, providing a high-quality semantic embedding representation for subsequent node feature generation. Specifically, this module uses a pre-trained deep language model (such as BERT) to process the text information such as the title and introduction of the courses, embeds it into a high-dimensional semantic space, and generates a semantic embedding vector for each course. By encoding the text descriptions of all courses, a semantic embedding matrix E is generated, where each row represents the semantic features of a course. To ensure the quality of the semantic embedding, the module also normalizes the embedding vectors to make them consistent in the feature space. In addition, the semantic generation module dynamically adjusts the dimension of the semantic embedding to ensure that it can adapt to course datasets of different scales. The introduction of the semantic generation module makes up for the neglect of the course content characteristics in traditional recommendation methods, providing rich semantic information support for subsequent node feature fusion and graph convolution operations.

[0091] Step S20: Perform user behavior analysis on the user's historical learning interaction data, and construct a user behavior feature matrix according to the results of the user behavior analysis.

[0092] It can be understood that in this embodiment, a user behavior sequence can be constructed based on the historical interaction records of the user's course learning, and the temporal information in the sequence and the explicit relationship between courses can be extracted to generate a user behavior feature matrix.

[0093] In some embodiments, the recommendation device can extract behavior features from the user's learning history through the student behavior sequence generation module, and construct the interaction relationship between the user and the courses. This module first collects the user's learning records and generates the user's session sequence in chronological order. Each session sequence represents the set of courses that the user has learned during a specific time period. Subsequently, the module combines the timestamp information to extract the temporal features of the user's behavior and capture the dynamic changes of the user's interests. In addition, this module also generates a user behavior matrix , which is used to represent the explicit relationship between the user and the courses. To enhance the robustness of the data, the module performs completion processing on the sparse interaction data to ensure that the generated behavior sequence can comprehensively reflect the user's learning path. The user behavior sequence generation module provides basic data support for subsequent graph construction and enhancement, and at the same time provides dynamic features of the user's interests for the course recommendation model.

[0094] Step S30: Concatenate the semantic embedding matrix and the user behavior feature matrix in channels to generate a multi-dimensional feature matrix.

[0095] It can be understood that in this embodiment, the semantic information of the course nodes can be concatenated with the user behavior features to generate a node embedding representation that fuses semantic and relationship features, which is used for the construction of the course session graph.

[0096] In some embodiments, the recommendation device can fuse semantic information and user interaction relationships through the node feature generation module to generate a high-dimensional course node feature representation. Specifically, this module first combines the course semantic embedding E output by the semantic generation module with the interaction relationship features output by the user behavior sequence generation module. Through a feature concatenation operation, the module generates a node feature representation that fuses semantics and relationships. , Subsequently, the module performs a non-linear transformation on the node features to further enhance the expressive power of the features. The design of the node feature generation module aims to make up for the problem of split modeling of course content and user behavior in traditional recommendation methods, and provides high-quality node input features for subsequent graph convolution operations.

[0097] In some embodiments, the recommendation device can fuse the semantic feature vectors in the semantic embedding matrix and the behavior feature vectors in the user behavior feature matrix by means of channel concatenation to generate a fusion feature matrix containing multiple initial fusion feature vectors. :

[0098]

[0099] Unify the feature scales through layer normalization (LayerNorm, LN), and generate a multi-dimensional feature matrix based on the fusion feature vectors after layer normalization:

[0100]

[0101] Among them, is a dimensionality reduction matrix, representing the initial fusion feature vector, represents the fusion feature vector after layer normalization.

[0102] Step S40: Construct an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix.

[0103] It should be noted that the adjacency graph includes the adjacency relationships between each course node, and the hypergraph includes the course nodes and the interaction relationships between the user and each course node.

[0104] It should be noted that the recommendation device can construct a course adjacency graph according to the neighbor co-occurrence relationship between courses, and the adjacency graph is used to capture local aggregation features. By constructing a high-order course relationship graph based on the hypergraph, it is used to model the relationships between multiple nodes.

[0105] In some embodiments, the recommendation device can extract the global clustering center in the course embedding based on the mean shift clustering mechanism to enhance the global distribution perception ability of the node features.

[0106] In some embodiments, the recommendation device can construct a relationship graph between courses through the graph construction and enhancement module, and enhance the graph structure through multiple mechanisms. First, the module generates an adjacency graph G based on the user behavior sequence to capture the local relationships between courses. Second, the module constructs a hypergraph H through the "student-course" interaction relationship, where the nodes represent courses and the hyperedges represent all the courses learned by the student, for modeling the high-order associations between courses. In addition, the module introduces a mean shift clustering mechanism to dynamically cluster the embeddings of course nodes, generate global clustering centers, and enhance the global feature expression ability of nodes by calculating the similarity between nodes and clustering centers. By combining the structural information of the adjacency graph and the hypergraph, the graph construction and enhancement module can comprehensively capture the local and global relationships between courses, providing rich structural information for subsequent graph convolution operations.

[0107] In some embodiments, the recommendation device can capture the high-order association relationships between courses by modeling the learning behavior of students as a hypergraph. Hypergraph definition: The node set V of represents courses, and the hyperedge set represents the session data of students, and each hyperedge connects all the courses learned by a student. The adjacency matrix H of the hypergraph is defined as:

[0108]

[0109] where, if the course , then ; if the course , then .

[0110] By statistically analyzing the co-occurrence relationship of courses in the user session, an adjacency graph is constructed to capture the local aggregation characteristics between courses. The element of the adjacency matrix A represents the co-occurrence frequency of courses and in the student session:

[0111]

[0112] where, if the course and the course are jointly selected by the same student, then ; if the course and the course are not jointly selected by the same student, then .

[0113] In some embodiments, in order to enhance the global distribution perception ability of node embeddings, the recommendation device can use the mean shift clustering method to extract the global clustering centers in the course embeddings and dynamically enhance the node features. According to the distribution characteristics of the node embeddings, the bandwidth h of the mean shift is dynamically adjusted:

[0114]

[0115] Among them, is the mean of node embeddings;

[0116] Generate the clustering center C and clustering label L through mean shift operation:

[0117]

[0118] Among them, MS represents the mean shift operation;

[0119] Generate the clustering feature F by calculating the similarity between the node embedding X and the clustering center C and through weighting:

[0120]

[0121] Among them, is a learnable weight matrix, is a bias term.

[0122] Step S50: Perform graph convolution operations based on the adjacency graph and the hypergraph to generate the target embedding representation vectors of each course node.

[0123] It can be understood that in this embodiment, the graph convolutional network can respectively perform information aggregation on the adjacency graph and the hypergraph, combine local features and global features, and generate the final embedding representation of the course node.

[0124] In some embodiments, the recommendation device can perform information aggregation and relationship modeling on the features of the course nodes through the graph convolution processing module. The module respectively applies graph convolution operations to the adjacency graph and the hypergraph to extract the local features and high-order associations between courses. For the adjacency graph, the module captures the short-range dependence relationships of courses through neighborhood aggregation operations; for the hypergraph, the module uses hyperedge information to model the interactions between courses participated by the same students, thereby capturing the potential connections of courses in the learning path. Subsequently, the module fuses the output features of the adjacency graph convolution and the hypergraph convolution to generate the final embedding representation of the course node. The design of the graph convolution processing module aims to make up for the deficiencies in modeling the relationships between courses in traditional methods and provides a more comprehensive node feature representation for the recommendation model.

[0125] Step S60: Generate the attention weights of each course node, perform weighted aggregation on the target embedding representation vectors based on the attention weights, and perform course session recommendation based on the weighted aggregated target embedding representation vectors.

[0126] It can be understood that in this embodiment, the dynamic attention mechanism can be used to optimize the weight distribution between course nodes, combine node embeddings to generate the prediction results of course recommendations, and output a personalized course recommendation list.

[0127] In some embodiments, the recommendation device can further optimize the embedding representation of course nodes by combining the mean shift clustering mechanism and the dynamic attention mechanism through the node embedding enhancement module. First, the module uses mean shift clustering to generate the cluster center of the course embedding, and dynamically enhances the node embedding by calculating the similarity between the node and the cluster center, thereby improving the node's perception of global characteristics. Secondly, the module introduces a position-aware dynamic attention mechanism, which dynamically adjusts the weight distribution between nodes by combining the semantic features and position encoding information of the nodes to capture implicit relationships and semantic differences. Finally, the module uses the enhanced node embedding as the input of the recommendation model. The design of the node embedding enhancement module effectively improves the node expression ability in sparse data scenarios and provides a more accurate feature representation for course recommendations.

[0128] In some embodiments, the recommendation device can predict the score of the node embedding through the recommendation generation module to generate a personalized course recommendation list for the user. The module first uses the node embedding and user features to calculate the predicted score of the course, and converts the score into a probability distribution through the softmax function. Subsequently, the module selects the courses that best suit the user's interests based on the scoring results and generates a recommendation list. In addition, the module introduces a cross entropy loss function to optimize the model, minimize the divergence between the predicted value and the true label, and thus improve the accuracy of the recommendation results. The design of the recommendation generation module aims to convert the high-quality node embeddings generated by the aforementioned modules into actual recommendation results, providing users with accurate course recommendation services.

[0129] In this embodiment, the text description data of the courses is input into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors, and a semantic embedding matrix is generated based on the semantic feature vectors; user behavior analysis is performed on the user's historical learning interaction data, and a user behavior feature matrix is constructed according to the user behavior analysis results; the semantic embedding matrix and the user behavior feature matrix are concatenated in channels to generate a multi-dimensional feature matrix; an adjacency graph and a hypergraph between courses are constructed according to the multi-dimensional feature matrix, where the adjacency graph includes the adjacency relationships between each course node, and the hypergraph includes the course nodes and the interaction relationships between the user and each course node; graph convolution operations are performed based on the adjacency graph and the hypergraph to generate target embedding representation vectors for each course node; attention weights for each course node are generated, and the target embedding representation vectors are weighted and aggregated based on the attention weights, and course session recommendations are made based on the weighted and aggregated target embedding representation vectors; since this embodiment improves the feature expression ability by fusing multi-source information by fusing course semantic features and user behavior features, an adjacency graph and a hypergraph between courses are constructed according to the multi-dimensional feature matrix, thereby effectively improving the feature expression ability of the course nodes in the graph structure, effectively capturing the multi-node interaction relationships in the hypergraph structure and the neighborhood local feature interaction relationships of each course node in the adjacency graph, performing graph convolution operations based on the adjacency graph and the hypergraph, thereby accurately mining the explicit feature relationships and implicit feature relationships between the course nodes, realizing accurate capture of the dynamic changes of the user's interests, and thus accurately making course session recommendations.

[0130] Reference Figure 3 , Figure 3 is a schematic flowchart of the second embodiment of the course session recommendation method of the present invention.

[0131] Based on the above first embodiment, in the second embodiment of the course session recommendation method of the present invention, the step S10 further includes:

[0132] Step S101: Collect the text description data of the courses and preprocess the text description data.

[0133] It should be noted that the preprocessing includes word segmentation, stop word screening, and special character screening.

[0134] It can be understood that this embodiment can collect the text description data of the courses, including course titles, introductions, and other relevant content. Perform preprocessing operations on the text data, such as word segmentation, stop word removal, special character removal, etc., to ensure the standardization and consistency of the data, and provide high-quality input for subsequent semantic embedding generation.

[0135] Step S102: Input the preprocessed text description data into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors.

[0136] It should be understood that in this embodiment, a pre-trained deep language model (such as the BERT model) can be used to encode the preprocessed course text data and embed it into a high-dimensional semantic space. The model processes the text description of each course to generate a semantic embedding vector , which is used to represent the semantic features of the course.

[0137] Step S103: Semantically embed and integrate the semantic feature vectors of each course to generate an initial semantic matrix.

[0138] It can be understood that in this embodiment, the semantic embedding vectors of all courses can be integrated into a semantic embedding matrix E, where each row represents the semantic features of a course. The dimension of this matrix is , where m represents the total number of courses, d represents the dimension of the semantic embedding vector.

[0139] Step S104: Normalize the semantic feature vectors of each dimension in the initial semantic matrix to generate a semantic embedding matrix.

[0140] It should be understood that in this embodiment, the generated initial semantic matrix E can be normalized to ensure the consistency of the semantic features of different courses in the feature space. The normalization operation can improve the sensitivity of the model to features and avoid the degradation of model performance caused by differences in feature scales.

[0141] In this embodiment, by collecting the text description data of the course and preprocessing the text description data, the preprocessing includes word segmentation, stop word filtering, and special character filtering, inputting the preprocessed text description data into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors, semantically embedding and integrating the semantic feature vectors of each course to generate an initial semantic matrix, and normalizing the semantic feature vectors of each dimension in the initial semantic matrix to generate a semantic embedding matrix, thereby accurately capturing the rich semantic features in the course content, providing rich semantic information support for node feature fusion and graph convolution operations, and effectively improving the accuracy of course recommendation.

[0142] Reference Figure 4 , Figure 4 is the flowchart of the third embodiment of the course session recommendation method of the present invention.

[0143] Based on the above embodiments, in the third embodiment of the course session recommendation method of the present invention, the step S20 further includes:

[0144] Step S201: Model the user's learning behavior as a course learning session sequence in chronological order based on the user's historical learning interaction data.

[0145] It should be noted that the course learning session sequence includes the user's learning course set.

[0146] It can be understood that in this embodiment, the learning behavior of each student can be modeled as a course learning session sequence in chronological order according to the historical interaction records of the user's course learning: Define the student set , each student 's session is represented as , where represents the -th course that the student i learns in chronological order. Each session contains a masked validation set , which is used to predict the subsequent recommendation target of the last course.

[0147] Step S202: Obtain the user's course learning chronological order features according to the sequence position encoding information of the course learning session sequence.

[0148] It should be understood that in this embodiment, the chronological order features of course learning can be captured through sequence position encoding, and the time sequence information can be incorporated into the representation of the behavior sequence to generate the course learning chronological order features.

[0149] Step S203: Analyze the interaction frequency between the user and each course according to the course learning session sequence, and generate an explicit relationship weight matrix based on the interaction frequency.

[0150] It should be noted that the elements in the explicit relationship weight matrix represent the occurrence frequencies of different courses in the same session.

[0151] It can be understood that in this embodiment, the co-occurrence frequencies of courses in the sequence can be counted to generate an explicit relationship weight matrix , where represents the frequency of occurrence of course and

[0152] in the same session.

[0153] Step S204: Map the course ID corresponding to each course to an initial behavior embedding vector, and construct an initial behavior feature matrix based on the initial behavior embedding vector. It should be understood that in this embodiment, the course ID can be mapped to an initial behavior embedding vector to form a behavior feature matrix

[0154] Step S205: Based on the chronological feature of the course learning time and the explicit relationship weight matrix, perform weighted aggregation on the initial behavior feature matrix to generate enhanced behavior features.

[0155] It can be understood that in this embodiment, time series coding and co-occurrence weights can be combined to generate enhanced behavior features through weighted aggregation:

[0156]

[0157] Among them, represents the enhanced behavior features, represents the set of neighbor courses that appear in the same session as course represents the position encoding of course represents course and course in the same session represents the frequency of occurrence of course and course

[0158] Step S206: Based on the enhanced behavior features of each course, construct a user behavior feature matrix.

[0159] It should be understood that in this embodiment, a sliding window mechanism can be used to extract local sequence patterns, capture short-term changes in user interests, and perform feature compression on long sequences through a gating mechanism to generate a time series-aware user behavior feature matrix .

[0160] In this embodiment, based on the user's historical learning interaction data, the user's learning behavior is modeled as a course learning session sequence in chronological order, and the course learning session sequence includes the set of the user's learning courses;

[0161] According to the sequence position encoding information of the course learning session sequence, obtain the chronological feature of the user's course learning time, analyze the interaction frequency between the user and each course according to the course learning session sequence, generate an explicit relationship weight matrix based on the interaction frequency, map the course ID corresponding to each course to an initial behavior embedding vector, construct an initial behavior feature matrix based on the initial behavior embedding vector, perform weighted aggregation on the initial behavior feature matrix based on the chronological feature of the course learning time and the explicit relationship weight matrix to generate enhanced behavior features, and construct a user behavior feature matrix based on the enhanced behavior features of each course, thereby providing basic data support for subsequent graph construction and enhancement, and at the same time providing dynamic features of the user's interests for the course recommendation model, realizing accurate analysis of the user's dynamic behavior and interest features, and improving the accuracy of course recommendation.

[0162] ​​Reference Figure 5 , Figure 5 is a schematic flowchart of the fourth embodiment of the course session recommendation method of the present invention.

[0163] Based on the above embodiments, in the fourth embodiment of the course session recommendation method of the present invention, step S50 further includes:[[]]

[0164] Step S501: Perform a neighborhood aggregation operation on the adjacency graph through a graph convolutional network to obtain the short-range dependency relationship between each course;

[0165] Step S502: Obtain the local feature information of each course according to the short-range dependency relationship.

[0166] It should be noted that the adjacency graph is used to capture the local characteristics and short-range dependency relationships between courses. Through the adjacency graph convolutional operation, the neighborhood information of the course nodes is aggregated to update the node feature representation. Using a graph convolutional network to process the adjacency graph, the update rule is:

[0167]

[0168] Among them, is the degree matrix of the adjacency matrix, is a learnable weight matrix, is an activation function, represents the output feature of the graph convolution of the adjacency graph.

[0169] Step S503: Perform a graph convolution operation on the hypergraph through a graph convolutional network to obtain hyperedge information;

[0170] Step S504: Obtain the interaction relationship between the student and each course based on the hyperedge information, and obtain the potential feature information of the course in the learning path based on the interaction relationship.

[0171] It should be noted that the hypergraph is used to capture the high-order association relationships between courses and model the potential connections in the student's learning path. Through the hypergraph convolutional operation, the feature expression ability of the course nodes is further enhanced using the hyperedge information. The hypergraph convolution updates the node features through the following formula:

[0172]

[0173] Among them, and are the degree matrices of the hypergraph nodes and hyperedges respectively, represents the feature output by the hypergraph performing a graph convolution operation.

[0174] Step S505: Fuse the local feature information and the potential feature information to obtain a fused feature.

[0175] It should be noted that after the convolutional processing of the adjacency graph and the hypergraph is completed, the output features of the two parts are fused to generate the final embedding representation of the course nodes. The output of the hypergraph convolution The output of the adjacency graph convolution are weighted and fused as follows:

[0176]

[0177] where α is an adjustable weight parameter used to balance the contributions of local features and global features, represents the fused feature.

[0178] Step S506: Activate and normalize the fused feature to generate the target embedding representation vector for each course node.

[0179] It can be understood that in this embodiment, by activating and normalizing the fused feature, the stability and robustness of the embedding representation are ensured.

[0180] In this embodiment, the graph convolutional network performs a neighborhood aggregation operation on the adjacency graph to obtain the short-range dependence relationship between each course, and obtains the local feature information of each course according to the short-range dependence relationship. The graph convolutional network performs a graph convolution operation on the hypergraph to obtain hyperedge information, and based on the hyperedge information, obtains the interaction relationship between the student and each course, and based on the interaction relationship, obtains the potential feature information of the course in the learning path. The local feature information and the potential feature information are fused to obtain a fused feature, and the fused feature is activated and normalized to generate the target embedding representation vector for each course node. Since this embodiment captures the short-range dependence relationship of the course through the neighborhood aggregation operation and models the interaction between the courses participated by the same student based on the hyperedge information, thereby capturing the potential connection of the course in the learning path, and fusing the output features of the adjacency graph convolution and the hypergraph convolution to generate the final embedding representation of the course node, so as to combine the local feature information and the global feature information of the course for embedding feature representation, providing a more comprehensive node feature representation for the recommendation model and effectively improving the accuracy of course recommendation.

[0181] Reference Figure 6 , Figure 6 is the flowchart of the fifth embodiment of the course session recommendation method of the present invention.

[0182] Based on the above embodiments, in the fifth embodiment of the course session recommendation method of the present invention, the step S60 further includes:

[0183] Step S601: Determine the cosine similarity between each course node according to the semantic feature vector of each course node;

[0184] Step S602: Determine the semantic difference information between each course node according to the cosine similarity.

[0185] It should be noted that in this embodiment, through the calculation of dynamic attention weights, based on the dynamic attention mechanism, the weight distribution between course nodes can be optimized, latent relationships and semantic differences can be captured. First, the similarity between nodes is calculated through cosine similarity. i and j The semantic difference information between course nodes is captured based on the cosine similarity:

[0186]

[0187] Among them, represents the cosine similarity between node i and node j , and respectively represent the semantic feature vectors of node i and node j .

[0188] Step S603: Generate the attention weights of each course node based on the semantic difference information and the position encoding information of each course node.

[0189] It should be noted that in this embodiment, by introducing position encoding, combining the semantic features of nodes and position encoding information, the weight distribution can be dynamically adjusted:

[0190]

[0191] Among them, represents the attention weight, represents the position encoding information;

[0192] Step S604: Generate the clustering center of the target embedding representation vector.

[0193] In some embodiments, the recommendation device can generate the clustering center C and the clustering label L through mean shift operation:

[0194]

[0195] Among them, MS represents the mean shift operation.

[0196] Step S605: Calculate the clustering similarity between each course node and the clustering center.

[0197] In some embodiments, the recommendation device can calculate the similarity between the node embedding X and the clustering center C, and generate the clustering feature F through weighting:

[0198]

[0199] Among them, is a learnable weight matrix, is the bias term.

[0200] Step S606: Weighted aggregation is performed on the target embedded representation vector based on the clustering similarity and the attention weight.

[0201] It can be understood that in this embodiment, weighted aggregation can be performed on the node embedding through the attention weight to generate a session embedding, capturing the dynamic changes of the user's interests. The weighted aggregation formula is as follows:

[0202]

[0203] Among them, represents the target embedded representation vector after weighted aggregation.

[0204] Step S607: Generate the behavior feature vector of the user based on the user's historical learning interaction data.

[0205] In some embodiments, the historical learning interaction data may include the course IDs learned by the user, the duration of the user's learning of each course, the speech activity of the user in the comment area of each course, the user's historical learning search records, the content of the user's favorite notes, etc.

[0206] In some embodiments, the recommendation device may extract behavior features based on the user's historical learning interaction data. The behavior features include: features related to learning progress, interactivity features, learning performance features, interest preference features, learning habit features, etc. The behavior features are encoded and transformed, and then normalized or standardized to ensure that data of different scales can be processed within a comparable range, thereby generating the behavior feature vector.

[0207] Step S608: Input the target embedded representation vector after weighted aggregation and the behavior feature vector into a pre-trained recommendation model for course session recommendation.

[0208] In some embodiments, the recommendation device may input the target embedded representation vector after weighted aggregation and the behavior feature vector into a pre-trained recommendation model for course prediction scoring, and convert the score into a probability distribution through the softmax function, and perform course session recommendation based on the probability distribution information.

[0209] Furthermore, in order to improve the recommendation accuracy performance of the recommendation model and provide more accurate recommended course push for users, in some embodiments, the above step S608 may include:

[0210] Step S6081: Input the weighted aggregated target embedding representation vector and the behavior feature vector into a pre-trained recommendation model to calculate the predicted scores of each course node;

[0211] Step S6082: Determine the label difference information between the predicted scores and the true labels based on the cross-entropy loss function;

[0212] Step S6083: Update the model parameters of the recommendation model through the backpropagation algorithm;

[0213] Step S6084: Optimize the recommendation model according to the updated model parameters, and perform course session recommendations based on the optimized recommendation model.

[0214] It should be noted that the recommendation device can predict the user's preference scores for courses based on the generated session embeddings. The score calculation is through the interaction of the user embedding feature matrix and the course embedding feature matrix to generate predicted scores , referring to the following formula:

[0215]

[0216] Where represents the predicted score, represents the behavior feature vector, represents the course embedding feature matrix constructed from the weighted aggregated target embedding representation vector.

[0217] It can be understood that the recommendation device can optimize the model parameters through the loss function to generate a personalized recommendation list. The cross-entropy loss function is used to measure the gap between the predicted scores and the true labels:

[0218]

[0219] Where represents the label difference information, represents the true label, represents the predicted score.

[0220] It should be understood that the recommendation device can adjust the model parameters through the backpropagation algorithm:

[0221]

[0222] Where η represents the learning rate, represents the gradient of the loss function, represents the model parameters before update, represents the model parameters after update.

[0223] In this embodiment, the cosine similarity between each course node is determined according to the semantic feature vectors of each course node, the semantic difference information between each course node is determined according to the cosine similarity, the attention weights of each course node are generated based on the semantic difference information and the position encoding information of each course node, the clustering center of the target embedded representation vector is generated, the clustering similarity between each course node and the clustering center is calculated, the target embedded representation vector is weighted and aggregated based on the clustering similarity and the attention weight, the behavior feature vector of the user is generated based on the historical learning interaction data of the user, and the weighted and aggregated target embedded representation vector and the behavior feature vector are input into a pre-trained recommendation model for course session recommendation; since in this embodiment, by combining the semantic features and position encoding information of the nodes, the weight distribution between the nodes is dynamically adjusted, the implicit relationships and semantic differences are captured, and the node expression ability in the sparse data scenario is effectively improved, providing a more accurate feature representation for course recommendation.

[0224] In addition, an embodiment of the present invention also proposes a computer-readable storage medium, on which a course session recommendation program is stored, and when the course session recommendation program is executed by a processor, the steps of the course session recommendation method described above are implemented.

[0225] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems, or devices, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM: Random Access Memory), read-only memory (ROM: Read Only Memory), erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium can be transmitted by any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0226] The above computer-readable storage medium can be included in the course session recommendation device; it can also exist separately without being assembled into the course session recommendation device.

[0227] In addition, an embodiment of the present invention also provides a computer program product, including a course session recommendation program, and when the course session recommendation program is executed by a processor, the steps of the above-mentioned course session recommendation method are implemented.

[0228] The specific implementation manner of the computer program product of the present invention is basically the same as that of the embodiments of the above-mentioned course session recommendation method, and will not be elaborated here.

[0229] Refer to Figure 7 , Figure 7 which is a structural block diagram of the first embodiment of the course session recommendation device of the present invention.

[0230] As Figure 7 shown, the course session recommendation device proposed by the embodiment of the present invention includes:

[0231] A semantic analysis module 10, configured to input text description data of a course into a pre-trained deep language model for semantic quantization, obtain a semantic feature vector, and generate a semantic embedding matrix based on the semantic feature vector;

[0232] A user behavior analysis module 20, configured to perform user behavior analysis on historical learning interaction data of a user, and construct a user behavior feature matrix according to the user behavior analysis result;

[0233] A feature splicing module 30, configured to perform channel splicing on the semantic embedding matrix and the user behavior feature matrix to generate a multi-dimensional feature matrix;

[0234] A graph construction module 40, configured to construct an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix, where the adjacency graph includes adjacency relationships between each course node, and the hypergraph includes course nodes and interaction relationships between a user and each course node;

[0235] A graph convolution processing module 50, configured to perform graph convolution operations based on the adjacency graph and the hypergraph to generate a target embedding representation vector of each course node;

[0236] A course recommendation module 60, configured to generate an attention weight for each course node, perform weighted aggregation on the target embedding representation vector based on the attention weight, and perform course session recommendation based on the weighted aggregated target embedding representation vector.

[0237] Further, the semantic analysis module 10 is also used to collect the text description data of the courses, and preprocess the text description data. The preprocessing includes word segmentation, stop word filtering, and special character filtering; input the preprocessed text description data into a pre-trained deep language model for semantic quantization to obtain semantic feature vectors; perform semantic embedding integration on the semantic feature vectors of each course to generate an initial semantic matrix; perform normalization processing on the semantic feature vectors of each dimension in the initial semantic matrix to generate a semantic embedding matrix.

[0238] Further, the user behavior analysis module 20 is also used to model the learning behavior of the user into a course learning session sequence in chronological order based on the user's historical learning interaction data. The course learning session sequence includes the set of courses learned by the user; obtain the chronological feature of the user's course learning according to the sequence position encoding information of the course learning session sequence; analyze the interaction frequency between the user and each course according to the course learning session sequence, and generate an explicit relationship weight matrix based on the interaction frequency. The elements in the explicit relationship weight matrix represent the occurrence frequencies of different courses in the same session; map the course ID corresponding to each course to an initial behavior embedding vector, and construct an initial behavior feature matrix based on the initial behavior embedding vector; perform weighted aggregation on the initial behavior feature matrix based on the course learning chronological feature and the explicit relationship weight matrix to generate enhanced behavior features:

[0239]

[0240] Wherein, represents the enhanced behavior feature, represents the set of neighbor courses that appear in the same session as course ; represents the position encoding of course ; represents course and course in the same session The occurrence frequency of; represents course The initial behavior embedding vector of;

[0241] Construct a user behavior feature matrix based on the enhanced behavior features of each course.

[0242] Further, the graph convolution processing module 50 is further configured to perform a neighborhood aggregation operation on the adjacency graph through a graph convolutional network to obtain the short-range dependence relationship between each course; obtain the local feature information of each course according to the short-range dependence relationship; perform a graph convolution operation on the hypergraph through the graph convolutional network to obtain hyperedge information; obtain the interaction relationship between the student and each course based on the hyperedge information, and obtain the potential feature information of the course in the learning path based on the interaction relationship; fuse the local feature information and the potential feature information to obtain a fused feature; perform activation and normalization processing on the fused feature to generate the target embedding representation vector of each course node.

[0243] Further, the course recommendation module 60 is further configured to determine the cosine similarity between each course node according to the semantic feature vectors of each course node:

[0244]

[0245] wherein, represents the cosine similarity between node i and node j , and respectively represent the semantic feature vectors of node i and node j ;

[0246] Determine the semantic difference information between each course node according to the cosine similarity;

[0247] Generate the attention weight of each course node based on the semantic difference information and the position encoding information of each course node:

[0248]

[0249] wherein, represents the attention weight, represents the position encoding information;

[0250] Generate the clustering center of the target embedding representation vector;

[0251] Calculate the clustering similarity between each course node and the clustering center;

[0252] Perform weighted aggregation on the target embedding representation vector based on the clustering similarity and the attention weight:

[0253]

[0254] wherein, represents the target embedding representation vector after weighted aggregation;

[0255] Generate the behavior feature vector of the user based on the user's historical learning interaction data;

[0256] Input the weighted aggregated target embedding representation vector and the behavior feature vector into a pre-trained recommendation model for course session recommendation.

[0257] Furthermore, the course recommendation module 60 is also used to input the weighted aggregated target embedding representation vector and the behavior feature vector into a pre-trained recommendation model to calculate the predicted scores of each course node:

[0258]

[0259] Among them, represents the predicted score, represents the behavior feature vector, represents the course embedding feature matrix constructed by the weighted aggregated target embedding representation vector;

[0260] Determine the label difference information between the predicted score and the true label based on the cross-entropy loss function:

[0261]

[0262] Among them, represents the label difference information, represents the true label, represents the predicted score;

[0263] Update the model parameters of the recommendation model through the backpropagation algorithm:

[0264]

[0265] Among them, η represents the learning rate, represents the gradient of the loss function, represents the model parameters before update, represents the model parameters after update;

[0266] Optimize the recommendation model according to the updated model parameters, and perform course session recommendation based on the optimized recommendation model.

[0267] In this embodiment, the text description data of the course is input into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector, and a semantic embedding matrix is generated based on the semantic feature vector; the historical learning interaction data of the user is analyzed for user behavior, and a user behavior feature matrix is constructed according to the results of the user behavior analysis; the semantic embedding matrix and the user behavior feature matrix are concatenated in channels to generate a multi-dimensional feature matrix; an adjacency graph and a hypergraph between courses are constructed according to the multi-dimensional feature matrix, where the adjacency graph includes the adjacency relationships between each course node, and the hypergraph includes the course nodes and the interaction relationships between the user and each course node; graph convolution operations are performed based on the adjacency graph and the hypergraph to generate a target embedding representation vector for each course node; attention weights for each course node are generated, and the target embedding representation vector is weighted and aggregated based on the attention weights, and course session recommendations are made based on the weighted and aggregated target embedding representation vector; since this embodiment improves the feature expression ability by fusing multi-source information by integrating the course semantic features and user behavior features, an adjacency graph and a hypergraph between courses are constructed according to the multi-dimensional feature matrix, thereby effectively improving the feature expression ability of the course nodes in the graph structure, effectively capturing the multi-node interaction relationships in the hypergraph structure and the neighborhood local feature interaction relationships between each course node in the adjacency graph, performing graph convolution operations based on the adjacency graph and the hypergraph, thereby accurately mining the explicit feature relationships and implicit feature relationships between the course nodes, realizing accurate capture of the dynamic changes of the user's interests, and thus making accurate course session recommendations.

[0268] The course session recommendation device provided by this application adopts the course session recommendation method in the above embodiment and can solve the technical problems of course session recommendation. Compared with the prior art, the beneficial effects of the course session recommendation device provided by this application are the same as those of the course session recommendation method provided by the above embodiment, and other technical features in the course session recommendation device are the same as the features disclosed in the method of the above embodiment and will not be elaborated here.

[0269] It should be understood that the above is only an example for illustration and does not constitute any limitation to the technical solution of the present invention. In specific applications, those skilled in the art can set according to needs, and the present invention does not limit this.

[0270] It should be noted that the above-described work process is only illustrative and does not limit the protection scope of the present invention. In actual applications, those skilled in the art can select some or all of them according to actual needs to achieve the purpose of the solution of this embodiment, and no limitation is made here.

[0271] In addition, for the technical details not described in detail in this embodiment, reference can be made to the course session recommendation method provided in any embodiment of the present invention, which will not be elaborated here.

[0272] It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, such that a process, method, article or system including a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article or system including such element.

[0273] The serial numbers of the embodiments of the present invention above are only for description and do not represent the superiority or inferiority of the embodiments.

[0274] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that makes a contribution to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as a read-only memory / random access memory, magnetic disk, optical disc), and includes several instructions to enable a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in various embodiments of the present invention.

[0275] The above are only the preferred embodiments of the present invention and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A method for recommending course sessions, characterized in that, The course session recommendation method includes: Inputting the text description data of the course into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector, and generating a semantic embedding matrix based on the semantic feature vector; Performing user behavior analysis on the user's historical learning interaction data, and constructing a user behavior feature matrix according to the results of the user behavior analysis; Performing channel splicing on the semantic embedding matrix and the user behavior feature matrix to generate a multi-dimensional feature matrix; Constructing an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix, where the adjacency graph includes the adjacency relationships between each course node, and the hypergraph includes the course nodes and the interaction relationships between the user and each course node; Performing graph convolution operations based on the adjacency graph and the hypergraph to generate a target embedding representation vector for each course node; Generating an attention weight for each course node, and performing weighted aggregation on the target embedding representation vector based on the attention weight, and performing course session recommendation based on the weighted aggregated target embedding representation vector; The performing user behavior analysis on the user's historical learning interaction data and constructing a user behavior feature matrix according to the results of the user behavior analysis includes: Modeling the learning behavior of the user into a course learning session sequence in chronological order based on the user's historical learning interaction data, where the course learning session sequence includes the set of courses learned by the user; Obtaining the course learning chronological order feature of the user according to the sequence position encoding information of the course learning session sequence; Analyzing the interaction frequency between the user and each course according to the course learning session sequence, and generating an explicit relationship weight matrix based on the interaction frequency, where the elements in the explicit relationship weight matrix represent the occurrence frequencies of different courses in the same session; Mapping the course ID corresponding to each course into an initial behavior embedding vector, and constructing an initial behavior feature matrix based on the initial behavior embedding vector; Performing weighted aggregation on the initial behavior feature matrix based on the course learning chronological order feature and the explicit relationship weight matrix to generate an enhanced behavior feature; Among them, represents enhanced behavioral features, represents the set of neighbor courses that appear in the same conversation as the course, represents the position encoding of the course , represents the frequency of occurrence of the course in the same conversation as the course , represents the initial behavioral embedding vector of the course . Constructing a user behavior feature matrix based on the enhanced behavior features of each course.

2. The course session recommendation method according to claim 1, wherein The inputting the text description data of the course into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector, and generating a semantic embedding matrix based on the semantic feature vector includes: Collecting the text description data of the course, and performing preprocessing on the text description data, where the preprocessing includes word segmentation, stop word filtering, and special character filtering; Inputting the preprocessed text description data into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector; Performing semantic embedding integration on the semantic feature vectors of each course to generate an initial semantic matrix; Performing normalization processing on the semantic feature vectors of each dimension in the initial semantic matrix to generate a semantic embedding matrix.

3. The course session recommendation method according to claim 1, wherein The performing graph convolution operations based on the adjacency graph and the hypergraph to generate a target embedding representation vector for each course node includes: Performing neighborhood aggregation operations on the adjacency graph through a graph convolutional network to obtain the short-range dependence relationships between each course; Obtain the local feature information of each course according to the short-term dependence relationship; Perform graph convolution operations on the hypergraph through a graph convolutional network to obtain hyperedge information; Based on the hyperedge information, obtain the interaction relationship between students and each course, and based on the interaction relationship, obtain the potential feature information of the course in the learning path; Fuse the local feature information and the potential feature information to obtain a fused feature; Perform activation and normalization processing on the fused feature to generate the target embedding representation vector of each course node.

4. The course session recommendation method according to any one of claims 1 to 3, characterized in that, Generate the attention weights of each course node, and perform weighted aggregation on the target embedding representation vector based on the attention weights. Course session recommendation based on the weighted aggregated target embedding representation vector includes: Determine the cosine similarity between each course node according to the semantic feature vector of each course node: Among them, represents the cosine similarity i between node j and node and respectively represent the semantic feature vectors i of node j and node Determine the semantic difference information between each course node according to the cosine similarity; Generate the attention weights of each course node based on the semantic difference information and the position encoding information of each course node: Among them, represents the attention weight, represents the position encoding information; Generate the clustering center of the target embedding representation vector; Calculate the clustering similarity between each course node and the clustering center; Perform weighted aggregation on the target embedding representation vector based on the clustering similarity and the attention weights: Among them, represents the target embedding representation vector after weighted aggregation; Generate the behavior feature vector of the user based on the user's historical learning interaction data; Input the weighted aggregated target embedding representation vector and the behavior feature vector into a pre-trained recommendation model for course session recommendation.

5. The course session recommendation method according to claim 4, wherein, The inputting the weighted aggregated target embedding representation vector and the behavior feature vector into a pre-trained recommendation model for course session recommendation includes: Input the weighted aggregated target embedding representation vector and the behavior feature vector into a pre-trained recommendation model to calculate the predicted scores of each course node: Among them, represents the predicted score, represents the behavioral feature vector, represents the course embedding feature matrix constructed by the target embedding representation vector after weighted aggregation; Determine the label difference information between the predicted scores and the true labels based on the cross-entropy loss function: Among them, represents the label difference information, represents the true label, represents the predicted score; Update the model parameters of the recommendation model through the backpropagation algorithm: Among them, η represents the learning rate, represents the gradient of the loss function, represents the model parameters before update, represents the model parameters after update; Optimize the recommendation model according to the updated model parameters, and perform course session recommendation based on the optimized recommendation model.

6. A course session recommendation device, characterized in that, The course session recommendation device includes: A semantic analysis module for inputting the text description data of the course into a pre-trained deep language model for semantic quantization to obtain a semantic feature vector, and generating a semantic embedding matrix based on the semantic feature vector; A user behavior analysis module for performing user behavior analysis on the user's historical learning interaction data and constructing a user behavior feature matrix according to the user behavior analysis results; A feature splicing module for splicing the semantic embedding matrix and the user behavior feature matrix in channels to generate a multi-dimensional feature matrix; A graph construction module for constructing an adjacency graph and a hypergraph between courses according to the multi-dimensional feature matrix, where the adjacency graph includes the adjacency relationship between each course node, and the hypergraph includes course nodes and the interaction relationship between the user and each course node; A graph convolution processing module for performing graph convolution operations based on the adjacency graph and the hypergraph to generate the target embedding representation vector of each course node; The course recommendation module is used to generate the attention weights of each course node, perform weighted aggregation on the target embedded representation vector based on the attention weights, and perform course session recommendation based on the weighted aggregated target embedded representation vector; The user behavior analysis module is further configured to model the learning behavior of the user into a course learning session sequence in chronological order based on the user's historical learning interaction data, where the course learning session sequence includes the set of courses learned by the user; obtain the chronological feature of the user's course learning according to the sequence position encoding information of the course learning session sequence; analyze the interaction frequency between the user and each course according to the course learning session sequence, and generate an explicit relationship weight matrix based on the interaction frequency, where the elements in the explicit relationship weight matrix represent the occurrence frequencies of different courses in the same session; map the course ID corresponding to each course to an initial behavior embedding vector, and construct an initial behavior feature matrix based on the initial behavior embedding vector; perform weighted aggregation on the initial behavior feature matrix based on the course learning chronological feature and the explicit relationship weight matrix to generate enhanced behavior features: Among them, represents enhanced behavioral features, represents the neighbor course set that appears in the same session as the course, represents the position encoding of the course represents the initial behavioral embedding vector of the course and the course in the same session; represents the occurrence frequency of the course ​​ Construct a user behavior feature matrix based on the enhanced behavior features of each course.

7. A course session recommendation device, characterized in that, The course session recommendation device includes: a memory, a processor, and a course session recommendation program stored on the memory and executable on the processor, where the course session recommendation program is configured to implement the course session recommendation method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, A course session recommendation program is stored on the computer-readable storage medium, and when the course session recommendation program is executed by a processor, it implements the course session recommendation method according to any one of claims 1 to 5.

9. A computer program product, characterized in that, The computer program product includes a course session recommendation program, and when the course session recommendation program is executed by a processor, it implements the steps of the course session recommendation method according to any one of claims 1 to 5.

Images