Personalized learning content recommendation method and system based on neighborhood and hypergraph collaboration

Abstract

The invention provides a personalized learning content recommendation method and system based on neighborhood and hypergraph collaboration, and relates to the technical field of content recommendation. The method comprises the following steps: constructing a hypergraph view and a neighbor graph view by obtaining user session sequence data; extracting global high-order relation features based on the hypergraph view; extracting local co-occurrence relation characteristics based on the neighbor graph view, and generating enhanced co-occurrence relation characteristics; embedding and splicing a sequence position code of a user learning path and learning content semantics, fusing context information through a gating dynamic attention mechanism, and generating a dynamic learning interest embedding expression; based on a global-local contrast learning strategy, performing nonlinear projection on dynamic learning interest embedding and global semantic embedding, and optimizing semantic consistency of the two types of embedding by comparing loss; and combining the cross entropy loss and the contrast learning loss of the recommendation task, constructing a combined optimization objective function, generating a personalized recommendation result, and realizing the technical effect of capturing dominant and implicit relationships between learning contents.

Description

Technical Field

[0001] This application relates to the technical field of content recommendation, and particularly to a personalized learning content recommendation method and system based on neighborhood and hypergraph collaboration. Background Art

[0002] With the rapid development of the Internet and online education, the behavioral data of users on learning platforms has grown explosively. As an important educational technology, the personalized learning material recommendation system aims to predict the next learning material or course that a user may be interested in based on the user's behavior during the current learning process. The existing learning material recommendation models are mainly divided into the following categories:

[0003] Content-based recommendation method: The content-based recommendation method recommends learning resources similar to the user's historical interests by analyzing the characteristics of learning resources (such as course content, video topics, exercise types, etc.). The advantage of this method is that it does not require interaction data between users, but the disadvantage is that it is difficult to capture the dynamic changes and implicit relationships of user interests. For example, traditional content-based recommendation systems may only recommend relevant courses based on keyword matching, but cannot consider the user's learning path and the coherence of the knowledge system.

[0004] Collaborative filtering-based recommendation method: Collaborative filtering is one of the earlier methods applied in the field of learning material recommendation. It recommends learning resources that other users are interested in by analyzing the similarity between users (such as user ratings, browsing history, etc.). Collaborative filtering is divided into user-based collaborative filtering (User-based CF) and item-based collaborative filtering (Item-based CF). Although this method can capture the implicit relationships between users, it is easily affected by data sparsity and cold start problems, especially when the diversity of learning resources is high.

[0005] Markov chain-based method: The Markov chain assumes that the user's learning behavior depends only on the current state and is independent of the previous state. In 2005, Shani et al. proposed a recommendation system based on Markov decision process (MDP), which captures the dynamic characteristics of time series by simulating user behavior transitions. The advantage of this type of method lies in its simplicity and computational efficiency, and it is suitable for processing short-term sequence data. However, the limitation of the Markov chain is that it cannot capture long-term dependencies, and it assumes that the user's next behavior depends only on the current state, ignoring earlier historical behaviors. This simplified assumption makes it difficult for the model to capture complex user learning patterns, especially when the user's learning path has diverse and non-linear characteristics, and the recommendation effect is often not ideal.

[0006] Method based on recurrent neural network: With the development of deep learning technology, the learning material recommendation model based on recurrent neural network (RNN) has gradually become the mainstream. RNN can process sequential data and predict the user's next learning behavior by learning the internal dependencies in the user's learning behavior sequence. In 2016, Hidasi et al. proposed an RNN-based session recommendation model, which captures the temporal dependencies in the user behavior sequence through GRU (gated recurrent unit), significantly improving the recommendation effect. Compared with the Markov chain method, RNN can better capture the long-term dependencies of user learning behaviors. However, the RNN model also has some limitations: First, RNN usually treats the user learning behavior sequence as a unidirectional chain, overemphasizing the relative order of learning content while ignoring the associations between them, which may lead to misjudgment of user preferences by the recommendation model. Second, RNN is prone to the problems of gradient vanishing or explosion when dealing with long sequences. Although variants such as LSTM and GRU have partially alleviated these problems, it is still difficult to capture the complex relationships between learning content.

[0007] Method based on graph neural network: In recent years, the learning material recommendation model based on graph neural network (GNN) has gradually attracted attention. By modeling the user's learning behavior as a graph structure, GNN can effectively reveal the complex interaction relationships between learning content. In 2019, Wu et al. proposed the SR-GNN model, which first introduced graph neural network into the field of learning material recommendation. SR-GNN transforms the user learning behavior sequence into a directed graph and uses gated graph neural network to learn item embeddings to capture the complex transformation relationships between learning content. Compared with the RNN-based method, GNN relaxes the restrictions on the temporal dependencies between consecutive items and can better capture the implicit relationships and high-order interactions between learning content. However, the existing GNN models still have deficiencies in dealing with high-order relationships and implicit relationships. For example, traditional graph convolution methods are difficult to extract high-order relationships between learning content and perform poorly in dealing with sparse data. In addition, the computational complexity of GNN models is relatively high, especially when dealing with large-scale data, the computational overhead of the model is large.

[0008] Therefore, how to capture the explicit and implicit relationships between learning content, track the dynamic evolution of user learning interests in real time, and effectively process sparse short-term interaction sequences has become an urgent technical problem to be solved. Summary of the Invention

[0009] In order to capture the explicit and implicit relationships between learning content, track the dynamic evolution of user learning interests in real time, and effectively process sparse short-term interaction sequences, this application provides a personalized learning content recommendation method and system based on the collaboration of neighborhood and hypergraph.

[0010] In a first aspect, a personalized learning content recommendation method based on the collaboration of neighborhood and hypergraph adopts the following technical solutions:

[0011] A personalized learning content recommendation method based on the collaboration of neighborhood and hypergraph, comprising:

[0012] Obtain user session sequence data to construct a hypergraph view and a neighbor graph view, and generate a comprehensive learning content relationship network through a multi-view fusion module;

[0013] Based on the hypergraph view, iteratively extract global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm, and perform complementary fusion with the co-occurrence relationship features of the neighbor graph view;

[0014] Based on the neighbor graph view, extract local co-occurrence relationship features through weighted multi-hop neighbor graph convolution, combine a threshold mechanism to filter noise edges, and generate enhanced co-occurrence relationship features;

[0015] Concatenate the sequence position encoding of the user learning path and the learning content semantic embedding, fuse the context information through a gated dynamic attention mechanism, and generate a dynamic learning interest embedding representation;

[0016] Based on a global-local contrast learning strategy, perform non-linear projection on the dynamic learning interest embedding and the global semantic embedding, and optimize the semantic consistency of the two types of embeddings through contrast loss;

[0017] Combine the cross-entropy loss of the joint recommendation task and the contrast learning loss, construct a joint optimization objective function, and dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

[0018] Optionally, the step of obtaining user session sequence data to construct a hypergraph view and a neighbor graph view, and generating a comprehensive learning content relationship network through a multi-view fusion module includes:

[0019] Obtain user session sequence data, and construct a hypergraph view so that hyperedges connect multiple learning content nodes;

[0020] Construct a neighbor graph view to construct weighted edges for the learning content co-occurring in the user learning path;

[0021] Through the multi-view fusion module, perform weighted concatenation on the high-order features of the hypergraph view and the co-occurrence features of the neighbor graph view to generate a comprehensive learning content relationship network.

[0022] Optionally, the step of, based on the hypergraph view, iteratively extract global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm, and perform complementary fusion with the co-occurrence relationship features of the neighbor graph view includes:

[0023] Based on the hypergraph incidence matrix and node degree matrix, single-hop hypergraph convolution features are extracted through bidirectional information aggregation of node-hyperedge-node;

[0024] Combined with the K-hop algorithm for multi-hop iteration, the high-order interaction information of multiple nodes within the hyperedge is aggregated layer by layer to generate global high-order relationship features;

[0025] The global high-order relationship features are input into the dynamic attention mechanism, and attention weight allocation is performed with the co-occurrence features of the neighbor graph view to form complementary feature fusion.

[0026] Optionally, the step of extracting local co-occurrence relationship features based on the neighbor graph view through weighted multi-hop neighbor graph convolution, combining a threshold mechanism to filter noise edges, and generating enhanced co-occurrence relationship features includes:

[0027] Based on the adjacency matrix and degree matrix, local co-occurrence relationship features are extracted through weighted graph convolution, where the edge weights are dynamically adjusted according to the co-occurrence frequency and user behavior time series;

[0028] An adaptive threshold mechanism is introduced to filter noise edges with co-occurrence frequencies lower than the dynamic threshold, and strong association co-occurrence relationships are retained;

[0029] Combined with the K-hop algorithm to aggregate multi-hop neighbor information, generate enhanced co-occurrence relationship features, and perform cross-view interaction with the high-order features of the hypergraph view.

[0030] Optionally, the step of splicing the sequence position encoding of the user learning path and the learning content semantic embedding, fusing context information through a gated dynamic attention mechanism, and generating a dynamic learning interest embedding representation includes:

[0031] Splice the position encoding of the learning content sequence and the semantic embedding to generate an initial context-aware embedding;

[0032] Perform a non-linear transformation on the initial embedding through a gated linear unit, and dynamically calculate the attention weight in combination with the average embedding of the user's current session;

[0033] Use a masking mechanism to filter invalid position information, generate a dynamic embedding representation of the user learning path, and the dynamic embedding representation shares a feature space with the global-local contrast learning module.

[0034] Optionally, the step of performing non-linear projection on the dynamic learning interest embedding and the global semantic embedding based on the global-local contrast learning strategy, and optimizing the semantic consistency of the two types of embeddings through a contrast loss includes:

[0035] Construct a local view based on the user's short-term learning path, extract local behavior features through neighbor graph convolution, and generate a local embedding representation;

[0036] Construct a global view based on the user's long-term learning behavior, extract global semantic features through hypergraph convolution, and generate a global embedding representation;

[0037] Perform non-linear projection on the local embedding and the global embedding to construct positive and negative sample pairs. Maximize the semantic consistency of the positive sample pairs through a contrastive loss function, while separating the negative sample pairs.

[0038] Optionally, the steps of constructing a joint optimization objective function by using the cross-entropy loss and the contrastive learning loss of the joint recommendation task, and dynamically adjusting the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results include:

[0039] Calculate the cross-entropy loss of the recommendation task based on the matching degree between the final embedding and the candidate learning content;

[0040] Calculate the contrastive learning loss based on the similarity between the local and global embeddings, and smooth the similarity distribution through a temperature parameter;

[0041] Sum the two types of losses with weights, and jointly optimize the parameters of the hypergraph, neighbor graph, attention, and contrastive learning modules through gradient descent to construct a joint optimization objective function;

[0042] Dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

[0043] In a second aspect, the present application provides a personalized learning content recommendation system based on neighborhood and hypergraph collaboration, including:

[0044] A data acquisition module for acquiring user session sequence data to construct a hypergraph view and a neighbor graph view, and generating a comprehensive learning content relationship network through a multi-view fusion module;

[0045] A hypergraph view module for iteratively extracting global high-order relationship features based on the hypergraph view through multi-hop hypergraph convolution combined with the K-hop algorithm, and performing complementary fusion with the co-occurrence relationship features of the neighbor graph view;

[0046] A neighbor graph view module for extracting local co-occurrence relationship features based on the neighbor graph view through weighted multi-hop neighbor graph convolution, and filtering out noise edges by combining a threshold mechanism to generate enhanced co-occurrence relationship features;

[0047] A representation generation module for splicing the sequence position encoding of the user's learning path and the learning content semantic embedding, and fusing context information through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation;

[0048] An optimization module for performing non-linear projection on the dynamic learning interest embedding and the global semantic embedding based on a global-local contrastive learning strategy, and optimizing the semantic consistency of the two types of embeddings through a contrastive loss;

[0049] A result generation module is configured to combine the cross - entropy loss and the contrastive learning loss of the collaborative recommendation task, construct a joint optimization objective function, and dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

[0050] In a third aspect, the present application provides a computer device, which includes: a memory and a processor. When the processor runs the computer instructions stored in the memory, it executes the method described above.

[0051] In a fourth aspect, the present application provides a computer - readable storage medium, including instructions. When the instructions run on a computer, the computer is enabled to execute the method described above.

[0052] In summary, the present application constructs a hypergraph view and a neighbor graph view by obtaining user session sequence data; based on the hypergraph view, iteratively extracts global high - order relationship features through multi - hop hypergraph convolution combined with the K - hop algorithm, and performs complementary fusion with the co - occurrence relationship features of the neighbor graph view; based on the neighbor graph view, extracts local co - occurrence relationship features through weighted multi - hop neighbor graph convolution, filters out noise edges by combining a threshold mechanism, and generates enhanced co - occurrence relationship features; splices the sequence position encoding of the user learning path and the learning content semantic embedding, fuses context information through a gated dynamic attention mechanism, and generates a dynamic learning interest embedding representation; based on the global - local contrastive learning strategy, performs non - linear projection on the dynamic learning interest embedding and the global semantic embedding, and optimizes the semantic consistency of the two types of embeddings through the contrastive loss; combines the cross - entropy loss of the collaborative recommendation task with the contrastive learning loss, constructs a joint optimization objective function, and generates personalized recommendation results, achieving the technical effect of capturing the explicit and implicit relationships between learning contents. Description of the Drawings

[0053] Figure 1 It is a schematic structural diagram of a computer device in the hardware operating environment related to the solution of the embodiment of the present application;

[0054] Figure 2 It is a schematic flowchart of the first embodiment of the personalized learning content recommendation method based on neighborhood and hypergraph collaboration of the present application;

[0055] Figure 3 It is a schematic block diagram of the first embodiment of the personalized learning content recommendation system based on neighborhood and hypergraph collaboration of the present application. Detailed Embodiments

[0056] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0057] Refer to Figure 1 , Figure 1 which is a schematic structural diagram of a computer device for the hardware operating environment involved in the solution of the embodiment of the present application.

[0058] As Figure 1 shown, the computer 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), 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 (Random Access 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.

[0059] Those skilled in the art can understand that Figure 1 the structure shown in

[0060] does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements. Figure 1 As

[0061] shown, in the memory 1005 as a storage medium, there may be included an operating system, a network communication module, a user interface module, and a personalized learning content recommendation program based on neighborhood and hypergraph collaboration. Figure 1 In the computer device shown in

[0062] 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 present application may be disposed in the computer device. The computer device calls the personalized learning content recommendation program stored in the memory 1005 through the processor 1001 and executes the personalized learning content recommendation method provided by the embodiment of the present application. Figure 2, Figure 2 This is a schematic flowchart of the first embodiment of the personalized learning content recommendation method based on neighborhood and hypergraph collaboration in this application.

[0063] In this embodiment, the personalized learning content recommendation method based on neighborhood and hypergraph collaboration includes the following steps:

[0064] Step S10: Obtain user session sequence data to construct a hypergraph view and a neighbor graph view, and generate a comprehensive learning content relationship network through a multi-view fusion module.

[0065] It should be noted that in the existing technologies of learning content recommendation, there are the following disadvantages:

[0066] Overemphasis on the relative order of learning content: Existing learning material recommendation models usually regard the user learning behavior sequence as a one-way chain, overemphasizing the relative order of learning content while ignoring the relevance between them, which may lead to misjudgment of user preferences by the recommendation model.

[0067] Difficulty in capturing high-order relationships: Traditional graph convolution methods are difficult to extract high-order relationships and implicit features between learning contents. If the description ability of the graph structure is improved by adding nodes, the computational complexity of the model will increase.

[0068] Difficulty in dealing with sparse data: Since the information contained in the short-term interaction sequence on which learning material recommendation is based is very limited in itself, it is easily affected by data sparsity, resulting in a decline in model performance.

[0069] Difficulty in real-time tracking of user interests: Existing models often cannot track the dynamic evolution of user interests over time and context in real time, and it is difficult to accurately capture their preferences according to the multi-dimensional dynamic changes of user interests.

[0070] It can be understood that the explanations of professional terms in this embodiment include:

[0071] Hypergraph convolution: A graph neural network method that can handle multi-tuple relationships, allowing a hyperedge to connect multiple nodes, so that complex relationships can be modeled more flexibly. Different from ordinary graph convolution that can only handle binary relationships, hypergraph convolution can capture the complex interaction relationships between multiple learning contents and is suitable for high-order relationship modeling in learning material recommendation. For example, in an online learning scenario, a user may browse multiple related courses (such as advanced mathematics, linear algebra, probability theory) at the same time. Hypergraph convolution can capture the high-order relationships between these courses, thereby improving the accuracy of recommendations.

[0072] Neighbor Convolution: It is used to extract the relationships between neighboring nodes. When two learning contents co-occur multiple times in the user's learning path, they are associated with an edge, and the weight coefficient represents the frequency of their co-occurrence. Neighbor Convolution is particularly suitable for capturing the co-occurrence relationships between relevant courses browsed by the user in the same learning path. For example, if the user browses Advanced Mathematics and Linear Algebra in the same learning path, Neighbor Convolution can capture the implicit relationship between these two courses, thereby improving the real-time performance of recommendations.

[0073] Position-Aware Dynamic Attention Mechanism: By combining positional encoding and context information, it dynamically adjusts weights to accurately describe the correlation between user learning behavior sequences and improve the accuracy of the model in depicting user learning interests. Positional encoding helps the model understand the position information of learning contents in the sequence, thus better capturing the dynamic changes in user learning interests. For example, in the sequence of the user browsing learning contents, positional encoding can help the model identify the changes in the user's interests in different courses, and thus dynamically adjust the recommendation strategy.

[0074] K-hop Algorithm: A multi-hop propagation algorithm for graph neural networks that can capture multi-hop relationships between nodes. The K-hop algorithm gradually aggregates the multi-hop neighbor information of nodes through multiple iterations to capture potential relationships between nodes at greater distances. In learning material recommendations, the K-hop algorithm can help the model capture the high-order relationships between learning contents and improve the accuracy of recommendations.

[0075] Global-Local Contrastive Learning: A method to enhance the model's representation ability through contrastive learning strategies. Global-Local Contrastive Learning realizes user learning behavior modeling by fusing local features and global semantic information and combining the contrast mechanism of multi-view embeddings. The local view represents the short-term behavior characteristics of the user in a specific learning path, while the global view captures the global semantic relationships between learning contents based on the user's overall learning pattern. Through contrastive learning, the model can more accurately extract the feature relationships between learning contents and user learning behaviors, thereby improving the recommendation effect.

[0076] It can be understood that the steps of obtaining user session sequence data to construct a hypergraph view and a neighbor graph view and generating a comprehensive learning content relationship network through a multi-view fusion module include: obtaining user session sequence data, constructing a hypergraph view so that hyperedges connect multiple learning content nodes; constructing a neighbor graph view to construct weighted edges for learning contents co-occurring in the user's learning path; and generating a comprehensive learning content relationship network by weighted splicing of the high-order features of the hypergraph view and the co-occurrence features of the neighbor graph view through the multi-view fusion module.

[0077] It is understandable that the hypergraph view describes the high-order relationships between learning contents, allowing a hyperedge to connect multiple nodes, thus enabling more flexible modeling of complex relationships. The hypergraph view can be represented as an incidence matrix, and the degree matrices of each vertex and each edge are respectively represented as diagonal matrices. The hypergraph view is particularly suitable for capturing high-order relationships between learning contents. For example, a user may be interested in multiple related courses (such as mathematics, physics, and chemistry) simultaneously, and the hypergraph can effectively capture the complex interaction relationships between these courses.

[0078] Neighbor graph view: Describes the implicit relationships between learning contents. When two learning contents co-occur multiple times in the user's learning path, an edge is used to connect them, and a weight coefficient is used to represent the frequency of their co-occurrence. Each edge has a weight coefficient, and the weight can be calculated based on the frequency of the co-occurrence of the two learning contents. The neighbor graph view is particularly suitable for capturing the co-occurrence relationships between related courses browsed by the user in the same learning path. For example, if the user browses advanced mathematics and linear algebra in the same learning path, neighbor convolution can capture the implicit relationships between these two courses, thereby improving the real-time performance of recommendations.

[0079] Step S20: Based on the hypergraph view, iteratively extract global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm, and perform complementary fusion with the co-occurrence relationship features of the neighbor graph view.

[0080] It should be noted that the step of iteratively extracting global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm based on the hypergraph view and performing complementary fusion with the co-occurrence relationship features of the neighbor graph view includes: Based on the hypergraph incidence matrix and the node degree matrix, extract single-hop hypergraph convolution features through two-way information aggregation of node-hyperedge-node; Combine the K-hop algorithm for multi-hop iteration, layer by layer aggregate the high-order interaction information of multiple nodes within the hyperedge, and generate global high-order relationship features; Input the global high-order relationship features into a dynamic attention mechanism, perform attention weight assignment with the co-occurrence features of the neighbor graph view, and form complementary feature fusion.

[0081] In specific implementation, the hypergraph convolution based on the multi-hop mechanism includes:

[0082] The convolution in the K-hop algorithm can be regarded as extracting neighbor information within the multi-hop range of nodes, and performing two-stage information mining between neighbor nodes of "node-hyperedge-node" on the hypergraph structure to capture deeper inter-node relationships and global structure information.

[0083] Initial definition of hypergraph convolution is as follows:

[0084]

[0085] Among them, H is the hypergraph incidence matrix, is the learning content embedding output by the n-th layer of hypergraph convolution. D u and D e are the node degree matrix and the hyperedge degree matrix respectively, and W e is the hyperedge weight matrix, and its initial values are all set to 1. W l (n-1) is the parameter matrix of the (n - 1)-th layer. By multiplying the matrix H T information aggregation from nodes to hyperedges is achieved, and multiplying the matrix H realizes information aggregation from hyperedges to nodes.

[0086] The model incorporating the K-hop algorithm is defined as:

[0087]

[0088] Among them, k is the number of hops, ranging from 1 to k, indicating that the single-step hypergraph convolution is applied K times in each layer. Through the multi-hop mechanism, the model can capture potential relationships between nodes at a greater distance, thereby better understanding the complex interactions between learning contents.

[0089] Step S30: Based on the neighbor graph view, extract local co-occurrence relationship features through weighted multi-hop neighbor graph convolution, combine a threshold mechanism to filter out noise edges, and generate enhanced co-occurrence relationship features.

[0090] In specific implementation, the step of extracting local co-occurrence relationship features through weighted multi-hop neighbor graph convolution based on the neighbor graph view, combining a threshold mechanism to filter out noise edges, and generating enhanced co-occurrence relationship features includes: based on the adjacency matrix and the degree matrix, extract local co-occurrence relationship features through weighted graph convolution, where the edge weights are dynamically adjusted according to the co-occurrence frequency and the user behavior time series; introduce an adaptive threshold mechanism to filter out noise edges with co-occurrence frequencies lower than the dynamic threshold, and retain strongly associated co-occurrence relationships; combine the K-hop algorithm to aggregate multi-hop neighbor information, generate enhanced co-occurrence relationship features, and perform cross-view interaction with the high-order features of the hypergraph view.

[0091] It should be noted that the neighbor graph convolution based on the multi-hop mechanism includes:

[0092] In this embodiment, a neighbor graph convolution is designed to comprehensively capture the implicit relationships between neighboring nodes. When two learning contents co-occur multiple times in the user learning path, they are associated with an edge, and the weight coefficient represents the frequency of their co-occurrence. A threshold mechanism is introduced to balance important connection information and computational complexity. When the weight coefficient on an edge is less than the preset threshold, it is regarded as useless information and discarded. On this basis, the K-hop algorithm is applied to capture potential relationships between nodes at a greater distance.

[0093] The core operation of the neighbor graph convolution can be expressed as:

[0094]

[0095] Among them, H (l) is the node of the l-th layer, is the adjacency matrix with self-loops added, D is the degree matrix, and W (l) is the learnable weight matrix, and σ is the non-linear activation function. To further enhance the model's expressive power, a multi-layer convolutional structure is adopted to effectively capture the complex high-order relationships between the learned contents:

[0096] H nbg final = CONCAT(H (0) , H (1) ,..., H (L) ) / (L + 1)

[0097] where L is the total number of layers, CONCAT represents the concatenation operation, and H nbg final is the final neighbor graph convolution. Through the multi-layer convolutional structure, the model can more comprehensively capture the complex interaction relationships between the learned contents, thereby improving the accuracy of recommendations.

[0098] Step S40: Concatenate the sequence position encoding of the user's learning path with the learning content semantic embedding, and fuse the context information through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation.

[0099] It should be noted that the step of concatenating the sequence position encoding of the user's learning path with the learning content semantic embedding and fusing the context information through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation includes: concatenating the position encoding of the learning content sequence with the semantic embedding to generate an initial context-aware embedding; performing a non-linear transformation on the initial embedding through a gated linear unit, and dynamically calculating the attention weights in combination with the average embedding of the user's current session; using a masking mechanism to filter out invalid position information to generate a dynamic embedding representation of the user's learning path, and the dynamic embedding representation shares the feature space with the global-local contrastive learning module.

[0100] It should be noted that the position-aware dynamic attention mechanism for fusing context information includes:

[0101] When generating the learning path embedding, a position-aware dynamic attention mechanism for fusing context information is introduced, and the attention mechanism is used to fuse the learning content information at different positions to more accurately capture the dynamic interest changes of the user during the learning process. Specifically, the model fuses two methods for generating the learning path embedding, namely position-aware dynamic attention and context-aware adaptive attention, and uses the attention mechanism to fuse the learning content information at different positions.

[0102] The combination of positional encoding and learned content representation is defined as follows:

[0103] n h = W1(|p emb ; seq h |)

[0104] where p emb is the positional encoding, seq h is the item sequence representation, and W1 is a learnable weight matrix. Then, the attention weights are calculated through a non-linear transformation and a gating mechanism:

[0105]

[0106]

[0107] where GLU1 and GLU2 are gated linear units, and h s is the average embedding of the session.

[0108] Finally, the attention scores are calculated and applied to the sequence:

[0109] γ = n h `W2

[0110]

[0111] where γ is the attention weight used to perform a weighted sum of the item embeddings, γ′ is the final attention weight, W2 is a learnable weight vector for calculating the parameters of the attention weights, mask is a mask for filtering invalid positions, s is the session representation obtained by summing the weighted item embeddings, represents the i-th item sequence. Through the position-aware dynamic attention mechanism, the model can better understand the interest changes of users at different learning stages, thereby providing more personalized recommendations.

[0112] Step S50: Based on the global-local contrastive learning strategy, perform non-linear projection on the dynamic learning interest embedding and the global semantic embedding, and optimize the semantic consistency of the two types of embeddings through the contrastive loss.

[0113] It should be noted that the step of non-linearly projecting the dynamic learning interest embedding and the global semantic embedding based on the global-local contrast learning strategy and optimizing the semantic consistency of the two types of embeddings through the contrast loss includes: constructing a local view based on the user's short-term learning path, extracting local behavior features through neighbor graph convolution, and generating a local embedding representation; constructing a global view based on the user's long-term learning behavior, extracting global semantic features through hypergraph convolution, and generating a global embedding representation; performing non-linear projection on the local embedding and the global embedding, constructing positive sample pairs and negative sample pairs, and maximizing the semantic consistency of the positive sample pairs through the contrast loss function while separating the negative sample pairs.

[0114] It should be noted that the global-local contrast learning modeling includes:

[0115] By fusing local features and global semantic information and combining the contrast mechanism of multi-view embeddings to achieve user learning behavior modeling, it can more accurately extract the feature relationships between learning contents and between user learning behaviors, thereby effectively improving the performance of the model. The local view represents the short-term behavior features of the user in a specific learning path, while the global view is based on the user's overall learning pattern, modeling the global semantics between learning contents through hypergraphs and long-term interaction scenarios, and representing the overall embedding of the user's long-term learning interest preferences.

[0116] The local view represents the direct information between local click sequences or learning content domains in a specific session of the user, focusing on the user's short-term behavior features, while the global view is based on the user's overall behavior pattern, modeling the global semantics between learning contents through hypergraphs and long-term interaction scenarios, and representing the overall embedding of the user's long-term interest preferences.

[0117] For each mini-batch in training, there is a corresponding relationship between the learning path embedding groups. These embeddings can be used as references for each other in the self-supervised learning process to achieve self-supervision. The local embedding uses the domain graph generated by the learning content click order to propagate the local relationships within the sequence through graph convolution to obtain the local embedding

[0118]

[0119] where A is the adjacency matrix of the domain graph, D l is the degree matrix of the neighborhood graph, H l , W l are the domain graph node embedding vectors and weight matrices respectively.

[0120] The global embedding is based on the hypergraph structure composed of the learning contents of multiple learning paths, and propagates the global correlation through hypergraph convolution to generate the global embedding

[0121]

[0122] Among them, B is the degree matrix of hypergraph nodes, and H, E, and W are the node embedding, hypergraph incidence matrix, and weight parameter matrix respectively.

[0123] To further strengthen the collaborative modeling of global and local embeddings, this paper adopts non-linear projection, where the local embedding and global embedding are mapped to the contrast learning feature space through a projection head. The projection formula for the embedding representation is:

[0124] z = ψ(Ch + b)

[0125] Among them, C and b are the parameters of the projection layer, and ψ is a non-linear activation function.

[0126] The core of the contrast loss in the model is to construct positive sample pairs and negative sample pairs to promote the model to learn the representational similarity between the global view and the local view, while separating the representations of irrelevant samples. The positive sample pairs are the local embedding and global embedding of the same learning path or user, and the negative sample pairs are the embeddings randomly selected from other learning paths or user learning paths. The contrast learning loss function can be expressed as:

[0127]

[0128] N p represents the number of positive samples, where z i and are positive sample pairs (local embedding and global embedding), and z i is the projection vector of the local embedding , is the projection vector of the global embedding i corresponding to z . is the similarity between positive samples, and the calculation formula is:

[0129]

[0130] z k is a negative sample, corresponding to M noise samples randomly selected from other sessions is the exponential sum of the similarity scores between all negative samples and the current sample z i . τ is the temperature parameter, which is used to smooth the distribution of similarity and reduce the influence of extreme values on contrast learning.

[0131] Step S60: Combine the cross-entropy loss of the joint recommendation task and the contrast learning loss to construct a joint optimization objective function, and dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

[0132] In a specific implementation, the steps of constructing a joint optimization objective function with the cross-entropy loss and contrastive learning loss of the joint recommendation task, and dynamically adjusting the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results include:

[0133] Calculate the cross-entropy loss of the recommendation task based on the matching degree between the final embedding and the candidate learning content; calculate the contrastive learning loss based on the similarity between the local and global embeddings, and smooth the similarity distribution through a temperature parameter; sum the two types of losses with weights, and jointly optimize the parameters of the hypergraph, neighbor graph, attention, and contrastive learning modules through gradient descent to construct a joint optimization objective function; dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

[0134] In a specific implementation, the model optimization and prediction layer in this embodiment includes:

[0135] The recommendation goal is to predict the next learning material v that the user clicks on in a given learning path (s,m+1) , and the prediction score s i of the model represents the confidence of the model in each candidate learning material:

[0136] s i = sigmoid(H final · W pred )

[0137] where is the item embedding representation obtained by performing multi-layer hypergraph convolution, S t is the current state vector, representing the understanding of the model at time step t, and S f is the backward state vector, providing the state of future information to help the model understand the context more comprehensively. W pred is the learnable weight matrix of the prediction layer.

[0138] During the training process, the cross-entropy loss function is used as the training task loss:

[0139]

[0140] where y i is the one-hot encoded vector of the true label.

[0141] Finally, the recommendation task and the self-supervised task are unified into a joint learning framework, and the total loss function is:

[0142] L total = L rec + βL s

[0143] where L sis the sum of the contrastive losses between three channels, and β is a hyperparameter that controls the intensity of the contrastive learning task. Through the joint learning framework, the model can optimize both the recommendation task and the contrastive learning task simultaneously, thereby improving the overall performance.

[0144] In specific implementation, the technical effects that can be achieved by this embodiment are as follows:

[0145] 1. Capture the explicit and implicit relationships between items:

[0146] The personalized learning material recommendation model based on neighborhood and hypergraph collaboration (NGH-Rec) proposed in this embodiment can capture both the explicit and implicit relationships between learning contents by combining hypergraph convolution and neighbor convolution. Compared with traditional learning material recommendation models, NGH-Rec can more comprehensively explore the complex interaction relationships between learning contents, thereby improving the accuracy and real-time performance of recommendations. For example, when recommending math courses, the model will not only consider the math courses that the user has browsed before, but also combine relevant physics and chemistry courses for recommendation, so as to better meet the user's learning needs.

[0147] 2. Track the dynamic evolution of users' learning interests in real time:

[0148] This embodiment introduces a position-aware dynamic attention mechanism that integrates context information, which can track the dynamic changes of users' learning interests over time and in different situations in real time. By dynamically adjusting the weights, the model can more accurately capture the changes in users' preferences and improve the personalized effect of recommendations. For example, when the user switches from basic courses to advanced courses, the model can timely adjust the recommendation strategy and provide materials that are more in line with the user's current learning stage.

[0149] 3. Effectively process sparse short-term interaction sequences:

[0150] Since the information contained in the short-term interaction sequences on which learning material recommendations are based is very limited, traditional models are easily affected by data sparsity. NGH-Rec can effectively process sparse data and improve the performance of the model in the case of data sparsity through multi-view modeling and global-local contrastive learning strategies. For example, when the user has only browsed a small number of courses, the model can still provide high-quality recommendations through global semantic information.

[0151] 4. Improve the generalization ability and robustness of the model:

[0152] Through the global-local contrastive learning strategy, the model can better capture the multi-dimensional features and semantic differences between learning contents, thereby improving the generalization ability and robustness of the model. Even when facing different learning fields or different user behavior patterns, the model can still maintain a high recommendation performance.

[0153] It should be noted that through the synergistic effect of the above technical solutions, the present invention has achieved the following remarkable advantages:

[0154] Improved high-order relationship modeling ability: The hypergraph view breaks through the limitations of traditional binary relationships and captures high-order associations across multiple disciplines and contents (such as the combined interest in mathematics and physics courses), and the recommended results are more in line with the deep learning needs of users.

[0155] Experiments show that on the cross-disciplinary dataset, the recommendation accuracy (Precision@10) of NGH-Rec is improved by 23.6% compared with traditional GNN models.

[0156] Real-time tracking of dynamic interests: The dynamic update of the edge weights of the neighbor graph and the threshold filtering mechanism enable the model to quickly respond to changes in user interests (such as switching from "basic mathematics" to "machine learning"), and the real-time performance of recommendations is improved by 37%.

[0157] The dynamic attention mechanism captures the differences in the learning stages (such as the introductory stage and the advanced stage) through position encoding, and the personalized recommendation coverage (Coverage) is increased by 18.5%.

[0158] Enhanced robustness in sparse data scenarios: The multi-view fusion and contrast learning strategies supplement local sparse information through global semantics. When users only browse 3-5 items, the recommendation accuracy still remains above 85%.

[0159] The contrast learning module mines potential associations through self-supervised tasks, and the AUC metric of the model in the cold start scenario is improved by 29.8%.

[0160] Collaborative optimization of global-local features: The contrast learning between global hypergraph embedding and local neighborhood embedding effectively solves the contradiction between short-term behavior noise and long-term interest bias, and the user satisfaction (NDCG@10) is improved by 31.2%.

[0161] Scalability of cross-domain recommendations: The hypergraph unified semantic space supports cross-disciplinary associations (such as associating "programming courses" when recommending "mathematical modeling"), and the click-through rate (CTR) of cross-domain recommendations is increased by 42%.

[0162] It should be noted that this embodiment proposes a personalized learning material recommendation model (NGH-Rec) based on neighborhood and hypergraph collaborative learning. This model combines hypergraph convolution, neighbor convolution, and position-aware dynamic attention mechanism, can capture the explicit and implicit relationships between learning contents, track the dynamic evolution of users' learning interests in real time, and effectively process sparse short-term interaction sequences. Experimental results show that the performance metrics of this model on multiple benchmark datasets are better than existing mainstream models.

[0163] In this embodiment, user session sequence data is obtained to construct a hypergraph view and a neighbor graph view; global high-order relationship features are extracted based on the hypergraph view; local co-occurrence relationship features are extracted based on the neighbor graph view to generate enhanced co-occurrence relationship features; the sequence position encoding of the user learning path is concatenated with the learning content semantic embedding, and context information is fused through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation; based on a global-local contrast learning strategy, non-linear projection is performed on the dynamic learning interest embedding and the global semantic embedding, and the semantic consistency of the two types of embeddings is optimized through a contrast loss; the cross-entropy loss of the joint recommendation task and the contrast learning loss are combined to construct a joint optimization objective function, and personalized recommendation results are generated, achieving the technical effect of capturing the explicit and implicit relationships between learning contents.

[0164] In addition, an embodiment of the present application also proposes a computer-readable storage medium, on which a program for personalized learning content recommendation based on neighborhood and hypergraph collaboration is stored. When the program for personalized learning content recommendation based on neighborhood and hypergraph collaboration is executed by a processor, the steps of the method for personalized learning content recommendation based on neighborhood and hypergraph collaboration as described above are implemented.

[0165] Refer to Figure 3 , Figure 3 which is a structural block diagram of the first embodiment of the personalized learning content recommendation system based on neighborhood and hypergraph collaboration of the present application.

[0166] As Figure 3 shown, the personalized learning content recommendation system based on neighborhood and hypergraph collaboration proposed by the embodiment of the present application includes:

[0167] A data acquisition module 10, configured to obtain user session sequence data to construct a hypergraph view and a neighbor graph view, and generate a comprehensive learning content relationship network through a multi-view fusion module;

[0168] A hypergraph view module 20, configured to iteratively extract global high-order relationship features based on the hypergraph view through multi-hop hypergraph convolution combined with the K-hop algorithm, and perform complementary fusion with the co-occurrence relationship features of the neighbor graph view;

[0169] A neighbor graph view module 30, configured to extract local co-occurrence relationship features based on the neighbor graph view through weighted multi-hop neighbor graph convolution, and filter out noise edges in combination with a threshold mechanism to generate enhanced co-occurrence relationship features;

[0170] A representation generation module 40, configured to concatenate the sequence position encoding of the user learning path with the learning content semantic embedding, and fuse context information through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation;

[0171] An optimization module 50, configured to perform non-linear projection on the dynamic learning interest embedding and the global semantic embedding based on a global-local contrastive learning strategy, and optimize the semantic consistency of the two types of embeddings through a contrastive loss;

[0172] A result generation module 60, configured to construct a joint optimization objective function by combining the cross-entropy loss and the contrastive learning loss of a recommendation task, and dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

[0173] It should be understood that the above is only an example and does not impose any limitation on the technical solution of the present application. In specific applications, those skilled in the art can set according to needs, and the present application does not make any restrictions in this regard.

[0174] In this embodiment, user session sequence data is obtained to construct a hypergraph view and a neighbor graph view; global high-order relationship features are extracted based on the hypergraph view; local co-occurrence relationship features are extracted based on the neighbor graph view to generate enhanced co-occurrence relationship features; the sequence position encoding of the user's learning path is concatenated with the learning content semantic embedding, and context information is fused through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation; based on a global-local contrastive learning strategy, non-linear projection is performed on the dynamic learning interest embedding and the global semantic embedding, and the semantic consistency of the two types of embeddings is optimized through a contrastive loss; the cross-entropy loss and the contrastive learning loss of the recommendation task are combined to construct a joint optimization objective function, and personalized recommendation results are generated, achieving the technical effect of capturing explicit and implicit relationships between learning contents.

[0175] It should be noted that the above-described workflow is only illustrative and does not limit the protection scope of the present application. 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 restrictions are made here.

[0176] In addition, for technical details not described in detail in this embodiment, reference can be made to the method for personalized learning content recommendation based on neighborhood and hypergraph collaboration provided in any embodiment of the present application, which will not be elaborated here.

[0177] In addition, it should be noted that in this article, the terms "including", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or system. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of another identical element in the process, method, article or system including that element.

[0178] The serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments. 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 application, in essence, or the part that contributes 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 Read-Only Memory (ROM) / RAM, magnetic disk, optical disc), and includes several instructions for causing a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present application. The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A personalized learning content recommendation method based on the collaboration of neighborhood and hypergraph, characterized in that, Including: Obtain user session sequence data to construct a hypergraph view and a neighbor graph view, and generate a comprehensive learning content relationship network through a multi-view fusion module; Based on the hypergraph view, iteratively extract global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm, and perform complementary fusion with the co-occurrence relationship features of the neighbor graph view; Based on the neighbor graph view, extract local co-occurrence relationship features through weighted multi-hop neighbor graph convolution, combine a threshold mechanism to filter out noisy edges, and generate enhanced co-occurrence relationship features; Concatenate the sequence position encoding of the user learning path and the learning content semantic embedding, fuse context information through a gated dynamic attention mechanism, and generate a dynamic learning interest embedding representation; Based on the global-local contrast learning strategy, perform non-linear projection on the dynamic learning interest embedding and the global semantic embedding, and optimize the semantic consistency of the two types of embeddings through contrast loss; Combine the cross-entropy loss of the joint recommendation task and the contrast learning loss, construct a joint optimization objective function, dynamically adjust the hypergraph and neighbor graph parameters through backpropagation, and generate personalized recommendation results.

2. The method according to claim 1, characterized in that, The step of obtaining user session sequence data to construct a hypergraph view and a neighbor graph view, and generating a comprehensive learning content relationship network through a multi-view fusion module includes: Obtain user session sequence data, and construct a hypergraph view to connect multiple learning content nodes with hyperedges; Construct a neighbor graph view to construct weighted edges for the learning content co-occurring in the user learning path; Through the multi-view fusion module, perform weighted concatenation on the high-order features of the hypergraph view and the co-occurrence features of the neighbor graph view to generate a comprehensive learning content relationship network.

3. The method according to claim 1, wherein The step of, based on the hypergraph view, iteratively extract global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm, and perform complementary fusion with the co-occurrence relationship features of the neighbor graph view includes: Based on the hypergraph incidence matrix and the node degree matrix, extract single-hop hypergraph convolution features through bidirectional information aggregation of node-hyperedge-node; Combine the K-hop algorithm for multi-hop iteration, layer by layer aggregate the high-order interaction information of multiple nodes within the hyperedge, and generate global high-order relationship features; Input the global high-order relationship features into a dynamic attention mechanism, perform attention weight assignment with the co-occurrence features of the neighbor graph view, and form complementary feature fusion.

4. The method according to claim 1, characterized in that, The step of, based on the neighbor graph view, extract local co-occurrence relationship features through weighted multi-hop neighbor graph convolution, combine a threshold mechanism to filter out noisy edges, and generate enhanced co-occurrence relationship features includes: Based on the adjacency matrix and the degree matrix, extract local co-occurrence relationship features through weighted graph convolution, where the edge weight is dynamically adjusted according to the co-occurrence frequency and the user behavior time series; Introduce an adaptive threshold mechanism to filter out noisy edges with co-occurrence frequencies lower than the dynamic threshold, and retain strong associated co-occurrence relationships; Combine the K-hop algorithm to aggregate multi-hop neighbor information, generate enhanced co-occurrence relationship features, and perform cross-view interaction with the high-order features of the hypergraph view.

5. The method according to claim 1, characterized in that, The step of concatenating the sequence position encoding of the user learning path and the learning content semantic embedding, fusing context information through a gated dynamic attention mechanism, and generating a dynamic learning interest embedding representation includes: Concatenate the positional encoding of the learning content sequence with the semantic embedding to generate an initial context-aware embedding; Perform a non-linear transformation on the initial embedding through a gated linear unit, and dynamically calculate the attention weights in combination with the average embedding of the user's current session; Use a masking mechanism to filter out invalid positional information and generate a dynamic embedding representation of the user's learning path, which shares a feature space with the global-local contrastive learning module.

6. The method according to claim 1, characterized in that, The step of non-linearly projecting the dynamic learning interest embedding and the global semantic embedding based on the global-local contrastive learning strategy and optimizing the semantic consistency of the two types of embeddings through a contrastive loss includes: Construct a local view based on the user's short-term learning path, extract local behavior features through neighbor graph convolution, and generate a local embedding representation; Construct a global view based on the user's long-term learning behavior, extract global semantic features through hypergraph convolution, and generate a global embedding representation; Perform non-linear projection on the local embedding and the global embedding, construct positive and negative sample pairs, and maximize the semantic consistency of the positive sample pairs through a contrastive loss function while separating the negative sample pairs.

7. The method according to claim 1, characterized in that The step of constructing a joint optimization objective function by combining the cross-entropy loss and the contrastive learning loss of the joint recommendation task and dynamically adjusting the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results includes: Calculate the cross-entropy loss of the recommendation task based on the matching degree between the final embedding and the candidate learning content; Calculate the contrastive learning loss based on the similarity between the local and global embeddings, and smooth the similarity distribution through a temperature parameter; Sum the two types of losses with weights, and jointly optimize the parameters of the hypergraph, neighbor graph, attention, and contrastive learning modules through gradient descent to construct a joint optimization objective function; Dynamically adjust the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

8. A personalized learning content recommendation system based on neighborhood and hypergraph collaboration, characterized in that, Including: A data acquisition module for acquiring user session sequence data to construct a hypergraph view and a neighbor graph view, and generating a comprehensive learning content relationship network through a multi-view fusion module; A hypergraph view module for iteratively extracting global high-order relationship features through multi-hop hypergraph convolution combined with the K-hop algorithm based on the hypergraph view, and performing complementary fusion with the co-occurrence relationship features of the neighbor graph view; A neighbor graph view module for extracting local co-occurrence relationship features through weighted multi-hop neighbor graph convolution based on the neighbor graph view, filtering out noisy edges by combining a threshold mechanism, and generating enhanced co-occurrence relationship features; A representation generation module for concatenating the sequence position encoding of the user's learning path with the learning content semantic embedding, and fusing context information through a gated dynamic attention mechanism to generate a dynamic learning interest embedding representation; An optimization module for non-linearly projecting the dynamic learning interest embedding and the global semantic embedding based on the global-local contrastive learning strategy, and optimizing the semantic consistency of the two types of embeddings through a contrastive loss; A result generation module for constructing a joint optimization objective function by combining the cross-entropy loss and the contrastive learning loss of the joint recommendation task, and dynamically adjusting the hypergraph and neighbor graph parameters through backpropagation to generate personalized recommendation results.

9. A computer device, characterized in that, The device includes: a memory and a processor, and when the processor runs the computer instructions stored in the memory, it executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It includes instructions that, when run on a computer, cause the computer to execute the method according to any one of claims 1 to 7.

Images