An adaptive data enhancement graph collaborative filtering recommendation method

Abstract

The application relates to the technical field of machine learning, in particular to a graph collaborative filtering recommendation method based on adaptive data enhancement. The method comprises the following steps: based on a user-item original bipartite graph, using a GCN to generate an embedding representation of the user-item original bipartite graph, and performing preference prediction; based on the user-item original bipartite graph and a noise measurement function, performing edge dropping to obtain a noise mitigation view and a noise perception view respectively, and using the GCN to obtain an embedding representation of the noise mitigation view and an embedding representation of the noise perception view respectively; based on the embedding representation of the user-item original bipartite graph, the embedding representation of the noise mitigation view and the embedding representation of the noise perception view, performing contrastive learning; and based on a recommendation loss function of the preference prediction and a contrastive learning loss function, training a collaborative filtering recommendation model based on adaptive data enhancement. The graph collaborative filtering recommendation method can filter noise data while avoiding overfitting.

Description

Technical Field

[0001] The present application relates to the field of machine learning technology, and in particular to a graph collaborative filtering recommendation method with adaptive data enhancement. Background Art

[0002] As a core technology for alleviating information overload and improving user experience, recommendation systems are widely used in social platforms, e-commerce, and content recommendation. Their core goal is to predict user interests and generate personalized recommendation lists by analyzing user historical behaviors (such as clicks, purchases, and ratings).

[0003] Among many recommendation technologies, collaborative filtering (CF) achieves recommendations by mining the interaction patterns between users and items, while graph collaborative filtering (GCF) further models users and items as a bipartite graph structure, using a graph convolutional network (GCN) to aggregate multi-hop neighbor information, capture high-order collaborative signals, and significantly improve recommendation performance.

[0004] For example, NGCF introduces graph structure information through the message passing mechanism, LightGCN simplifies the GCN architecture to improve efficiency, and the GCF model that introduces the attention mechanism can weightedly aggregate neighbor information to enhance representation capabilities.

[0005] However, existing methods rely on random or manually designed augmentation strategies (such as edge dropping with a fixed probability), which are difficult to dynamically adapt to different data distributions. They may destroy key connections in the graph structure or introduce irrelevant noise, resulting in the generated contrastive views lacking information and discrimination, thereby weakening the effectiveness of the supervisory signal. Furthermore, existing contrastive learning frameworks typically focus only on enhancing a single view and fail to fully consider the coordinated optimization of noise filtering and sparsity mitigation, limiting the model's robustness and generalization capabilities. Summary of the Invention

[0006] In order to address the deficiencies in the prior art, the purpose of this application is to provide a graph collaborative filtering recommendation method with adaptive data enhancement, which can filter noisy data while avoiding overfitting caused by sparse data.

[0007] To achieve the above objectives, this application provides an adaptive data-enhanced graph collaborative filtering recommendation method, comprising:

[0008] Based on the original bipartite graph of user-item, GCN is used to generate the embedding representation of the original bipartite graph of user-item and perform preference prediction;

[0009] Based on the original bipartite graph of user-items and the noise metric function, edge dropout is performed to obtain the noise-mitigated view and the noise-aware view, and GCN is used to obtain the embedding representation of the noise-mitigated view and the noise-aware view, respectively.

[0010] Perform contrastive learning based on the embedding representation of the original bipartite graph of user-item, the embedding representation of the noise-mitigated view, and the embedding representation of the noise-aware view;

[0011] Based on the preference prediction recommendation loss function and contrastive learning loss function, a collaborative filtering recommendation model with adaptive data enhancement is trained.

[0012] Furthermore, the specific steps of using GCN to generate an embedding representation of the original bipartite graph of user-item based on the original bipartite graph of user-item and performing preference prediction are as follows:

[0013] ;

[0014] ;

[0015] in, is the original bipartite graph of user-items Embedding representation, GCN () is a graph neural network, are the trainable parameters of the graph neural network, is the predicted preference of user u for item i, is the embedding representation of item i, is the transpose of the embedding representation of user u.

[0016] Furthermore, the graph neural network is a simplified graph neural network that omits nonlinear activation and feature transformation.

[0017] Furthermore, the specific steps of performing edge drop based on the original bipartite graph of user-items and the noise metric function to obtain a noise mitigation view and a noise perception view, and using GCN to obtain an embedding representation of the noise mitigation view and an embedding representation of the noise perception view include:

[0018] Based on the mask matrix and the adjacency matrix of the original bipartite graph of user-items, the adjacency matrices of the noise-mitigated view and the noise-aware view are generated by element-by-element dot product, and then the noise-mitigated view and the noise-aware view are created. The formula is as follows:

[0019] ;

[0020] in, is the adjacency matrix of the original bipartite graph of users and items, is the first mask matrix, is the second mask matrix, is the adjacency matrix of the noise-mitigated view, is the adjacency matrix of the noise-aware view;

[0021] Among them, the mask of each edge of the first mask matrix and the second mask matrix and Follow the and 1- Bernoulli distribution of :

[0022] ;

[0023] ;

[0024] in, is the noise measurement function of user u regarding item i, and the range of the noise measurement function is [0,1]. For noise mitigation view, for noise-aware views;

[0025] Based on the noise perception view and the noise mitigation view, GCN is used to obtain an embedding representation of the noise mitigation view and an embedding representation of the noise perception view, respectively.

[0026] Furthermore, the specific steps of performing contrastive learning based on the embedding representation of the original bipartite graph of user-items, the embedding representation of the noise mitigation view, and the embedding representation of the noise perception view include:

[0027] Perform first contrastive learning based on the embedding representation of the original bipartite graph of user-item and the embedding representation of the noise-mitigated view;

[0028] Perform second contrastive learning based on the embedding representation of the original bipartite graph of user-item and the embedding representation of the noise-aware view;

[0029] Based on the embedding representation of the noise-aware view and the embedding representation of the noise-mitigated view, a third contrastive learning is performed.

[0030] Furthermore, the specific steps of training and generating the adaptive data-enhanced collaborative filtering recommendation model based on the preference prediction recommendation loss function and the contrastive learning loss function include:

[0031] In the upper layer, a multi-task training strategy is used to jointly optimize the preference prediction task and the self-supervised contrastive learning task for training. The formula is as follows:

[0032] ;

[0033] in, is the recommendation loss function for preference prediction, is the upper layer training loss function, and are hyperparameters that control the strength of contrastive learning and regularization, respectively. is the loss function for contrastive learning;

[0034] In the lower layer, the recommendation loss function for preference prediction using the embedded representations of the noise-mitigated view and the noise-aware view is trained using the following formula:

[0035] ;

[0036] in, A proposed loss function for preference prediction of embedded representations of noise-mitigated views; Recommendation loss function for preference prediction of noise-aware view embedding representations, Training loss function for the lower layer;

[0037] Based on the upper-layer training loss function and the lower-layer training loss function, an adaptive data-enhanced collaborative filtering recommendation model is generated through iterative training.

[0038] To achieve the above-mentioned purpose, the electronic device provided by this application includes:

[0039] processor;

[0040] a memory having stored thereon one or more computer program instructions executed on the processor;

[0041] When the processor runs the computer instructions, the above-mentioned graph collaborative filtering recommendation method with adaptive data enhancement is executed.

[0042] To achieve the above objectives, the present application provides a computer-readable storage medium having computer instructions stored thereon. When the computer instructions are executed by a processor, the steps of the above-mentioned adaptive data enhanced graph collaborative filtering recommendation method are executed.

[0043] This application proposes an adaptive data-enhanced graph collaborative filtering recommendation method that provides beneficial regularization by introducing dynamic enhancements of noise perception and noise mitigation, thereby addressing these limitations of existing contrastive learning models and more effectively learning user and item embedding representations to achieve better performance.

[0044] The present application discloses an adaptive data-enhanced graph collaborative filtering recommendation method, which can effectively filter noise data in user interaction graphs by generating a noise mitigation graph and avoid model overfitting on sparse data by generating a noise perception graph.

[0045] The present invention proposes an adaptive data-enhanced graph collaborative filtering recommendation method that can accurately mine user preferences for items and improve the accuracy of recommendations.

[0046] Other features and advantages of the present application will be set forth in the following description, and in part will be apparent from the description, or may be learned by practicing the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings:

[0048] Figure 1 This is a flowchart of the adaptive data-enhanced graph collaborative filtering recommendation method of this application;

[0049] Figure 2 Schematic diagram of the structure of the collaborative filtering recommendation model with adaptive data enhancement;

[0050] Figure 3 This is a schematic diagram of the content of the experimental data set of Example 1 of this application;

[0051] Figure 4 Schematic diagram of the recommendation performance of each recommendation model. DETAILED DESCRIPTION

[0052] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are for illustrative purposes only and are not intended to limit the scope of protection of the present application.

[0053] It should be understood that the various steps described in the method embodiments of the present application can be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0054] As used herein, the term "including" and its variations are open-ended, i.e., "including but not limited to." The term "based on" means "based, at least in part, on." The term "one embodiment" means "at least one embodiment," the term "another embodiment" means "at least one additional embodiment," and the term "some embodiments" means "at least some embodiments." Other terms are defined in the following description.

[0055] It should be noted that the modifications of "one" and "plurality" mentioned in this application are illustrative rather than restrictive. Those skilled in the art will understand that unless the context clearly indicates otherwise, they should be understood as "one or more." "Plurality" should be understood as two or more.

[0056] Graph Convolutional Network (GCN) is a deep learning model specifically designed to process graph-structured data. Its core goal is to learn effective node representations by aggregating information about nodes and their neighbors, thereby capturing the complex relationships in graph data. Unlike traditional convolutional neural networks (CNNs) that focus on Euclidean spatial data (such as images and text), GCN extends the convolution operation to graph data with non-Euclidean structures, and can directly process graph-like data such as user-item interactions, social networks, and molecular structures. The basic idea of ​​GCN is derived from spectral graph theory in graph signal processing. It updates the embedding representation of nodes layer by layer through a local information transfer mechanism. In a recommendation system, the interaction between users and items can be modeled as a bipartite graph. GCN generates more expressive user embeddings by aggregating the features of items that users have interacted with (first-order neighbors) and the features of other users associated with these items (second-order neighbors), thereby improving recommendation accuracy.

[0057] For example, the user set is , the project set is , the observed user-item interaction matrix is ​​represented as .in Represents a user and projects There is interaction between them, otherwise it is expressed as . and Represent the number of users and items respectively. In the interaction matrix Based on the adjacency matrix for:

[0058]

[0059] Typically, GCN-based CF methods construct user-item interactions as a primitive bipartite graph of user-item ,in is the set of all nodes, Represents the observed edge set. GCN is used for each user and projects Assign an initialized embedding vector , the set of user and item embeddings is defined as and , is the embedding dimension.

[0060] Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0061] Example 1

[0062] An embodiment of the present application provides a graph collaborative filtering recommendation method with adaptive data enhancement, which will be referred to below. Figures 1-4 The adaptive data-enhanced graph collaborative filtering recommendation method of this application is described in detail.

[0063] Step S101: Based on the original bipartite graph of user-item, use GCN to generate an embedding representation of the original bipartite graph of user-item and perform preference prediction.

[0064] In this implementation, a simplified GCN (GCN omitting nonlinear activation and feature transformation) is used to generate the embedding representation of the original bipartite graph of user-item. The formula is as follows:

[0065] ;

[0066] in, is the original bipartite graph of user-items Embedding representation, GCN () is a graph neural network, are the trainable parameters of the graph neural network.

[0067] Specifically, first, aggregate the information of multi-hop neighbors for graph-level node representation. The formula is as follows:

[0068] ;

[0069] ;

[0070] in and Represents the first Layer adjacent items and users to the target node and Aggregate information. Indicates from Normalized user-item interaction matrix, is a concrete element representation, where:

[0071] ;

[0072] ;

[0073] in, and is the angle matrix, Represents user Adjacent user sets, Representation and Project A collection of adjacent items.

[0074] Then, in order to better preserve the original information of the nodes, residual connections are implemented in each layer of GCN. Layer to The message propagation process of the layer is:

[0075] ;

[0076] ;

[0077] in, It is The embedding representation of user u at the layer, No. The embedding representation of item i at layer.

[0078] Finally, we use the sum function as the readout function to combine the embeddings of all layers to generate the embedding representation of the interaction bipartite graph, as follows:

[0079] ;

[0080] ;

[0081] is the embedding representation of user u, is the embedding representation of item i.

[0082] In this embodiment, inner product operation is used to predict user About the project The preferences are as follows:

[0083]

[0084] in, is the predicted preference of user u for item i, is the embedding representation of item i, is the transpose of the embedding representation of user u.

[0085] In this implementation, Bayesian Personalized Ranking (BPR) loss is used as the recommendation loss function for preference prediction to optimize model parameters, which encourages the score prediction of observed interactions to be higher than that of unobserved interactions, as shown below:

[0086]

[0087] in, is the recommendation loss function for preference prediction, Is a sigmoid function to output probability distribution. Display items With users Interaction, and project and users There is no interaction between them.

[0088] Step S102: Based on the original bipartite graph of user-items and the noise metric function, edge drop is performed to obtain a noise-mitigated view and a noise-aware view, and GCN is used to obtain an embedding representation of the noise-mitigated view and an embedding representation of the noise-aware view.

[0089] The specific steps include:

[0090] Based on the mask matrix and the adjacency matrix of the original bipartite graph of user-items, the adjacency matrices of the noise-mitigated view and the noise-aware view are generated by element-by-element dot product, and then the noise-mitigated view and the noise-aware view are created. The formula is as follows:

[0091] ;

[0092] in, is the adjacency matrix of the original bipartite graph of users and items, is the first mask matrix, is the second mask matrix, is the adjacency matrix of the noise-mitigated view, is the adjacency matrix of the noise-aware view;

[0093] Among them, the mask of each edge of the first mask matrix and the second mask matrix and Follow the and 1- Bernoulli distribution of :

[0094] ;

[0095] ;

[0096] in, is the noise measurement function of user u regarding item i, and the range of the noise measurement function is [0,1]. For noise mitigation view, for noise-aware views;

[0097] It can be understood that the value of the noise metric function reflects the extent to which the interaction is affected by the noise factor, with a value of 1 indicating the user's true preference for the item, while a score of 0 indicates a noisy edge, i.e., a noise mitigation view. and noise-aware view It is obtained by actively filtering out the noise edges and preference edges in the original bipartite graph of users and items through a parameterized network.

[0098] Based on the noise perception view and the noise mitigation view, GCN is used to obtain the embedding representation of the noise mitigation view and the embedding representation of the noise perception view, respectively, as follows:

[0099] ;

[0100] ;

[0101] in, View for noise mitigation The embedding representation of Noise-aware view Embedding representation of .

[0102] Step S103: performing comparative learning based on the embedding representation of the original bipartite graph of user-item, the embedding representation of the noise mitigation view, and the embedding representation of the noise perception view;

[0103] In this embodiment, multiple self-supervised contrastive learning tasks are provided, including:

[0104] Perform first contrastive learning based on the embedding representation of the original bipartite graph of user-item and the embedding representation of the noise-mitigated view;

[0105] Perform second contrastive learning based on the embedding representation of the original bipartite graph of user-item and the embedding representation of the noise-aware view;

[0106] Performing third contrastive learning based on the embedding representation of the noise-aware view and the embedding representation of the noise-mitigated view;

[0107] In this implementation, InfoNCE loss is used as an auxiliary supervisory signal to control the invariance between views. The specific formula is as follows:

[0108] ;

[0109] in, is the loss function of contrastive learning, is the loss function of the first contrastive learning, is the loss function of the second contrastive learning; is the loss function for the third contrastive learning.

[0110] Step S104: Based on the preference prediction recommendation loss function and the contrastive learning loss function, a collaborative filtering recommendation model with adaptive data enhancement is trained and generated;

[0111] The specific steps include:

[0112] In the upper layer, a multi-task training strategy is used to jointly optimize the preference prediction task and the self-supervised contrastive learning task for training. The formula is as follows:

[0113] ;

[0114] in, is the recommendation loss function for preference prediction, is the upper layer training loss function, and are hyperparameters that control the strength of contrastive learning and regularization, respectively;

[0115] In the lower layer, the recommendation loss function for preference prediction using the embedded representations of the noise-mitigated view and the noise-aware view is trained using the following formula:

[0116] ;

[0117] in, View for noise mitigation Embedded representation of Recommendation loss function for preference prediction; Noise-aware view Embedded representation of Recommendation loss function for preference prediction;

[0118] Based on the upper layer training loss function and the lower layer training loss function, iterative training is performed to generate an adaptive data enhanced collaborative filtering recommendation model. The structure of the adaptive data enhanced collaborative filtering recommendation model is as follows: Figure 2 shown.

[0119] See Figure 3 , Figure 3 This is a schematic diagram of the content of the experimental data set of Example 1 of this application, such as Figure 3 As shown, in this embodiment, experiments are conducted using three datasets: Last FM, Yelp, and BeerAdvocate.

[0120] In this implementation, Recall@K and NDCG@K are used as evaluation metrics, and the latest recommendation model is selected as our baseline model, where K∈[20, 40]. Figure 4 , Figure 4 is a schematic diagram of the recommendation performance of each recommendation model, such as Figure 4 As shown in the figure, the adaptive data-augmented collaborative filtering recommendation model (AGCF) of this application always achieves the best performance under the Top-20 and Top-40 settings, with an improvement of 1%-4% compared to other benchmark models.

[0121] Example 2

[0122] One embodiment of the present application further provides an adaptive data augmented graph collaborative filtering model. The adaptive data augmented graph collaborative filtering model (AGCF) of the present application is described in detail below, including:

[0123] The first module, original representation learning, is as follows:

[0124] The GCN-based method uses a bipartite graph to The message passing approach is used to model user and item representations. A simplified GCN (omitting nonlinear activation and feature transformation) is used to aggregate information about multi-hop neighbors for graph-level node representation. The formula is as follows:

[0125] ;

[0126] ;

[0127] in and Indicates the Layer adjacent items and users to the target node and Aggregate information. Indicates from Normalized user-item interaction matrix, is a concrete element representation. The calculation is as follows:

[0128] ;

[0129] ;

[0130] in and is the angle matrix, and the user and projects The adjacent item and user sets are represented as and Next, in order to better preserve the original information of the nodes, residual connections are implemented in each layer. Layer to The message propagation process of the layer is:

[0131] ;

[0132] ;

[0133] It is Layer embedding representation, After layer propagation, the summation function is used as the readout function to combine the embeddings of all layers and obtain the final raw representation as follows:

[0134] ;

[0135] ;

[0136] Formally speaking, the information aggregation process of the user-item bipartite graph can be summarized as follows:

[0137]

[0138] in, is an embedding set, represents all trainable parameters. We use the original representation for the recommendation task and use the inner product operation to predict the user About the project The preferences are as follows:

[0139]

[0140] Note that the inner product is also used as a similarity measure to retrieve top recommendations based on the generated scores. Then, the pairwise Bayesian personalized ranking (BPR) loss is adopted as the recommendation loss to optimize the model parameters, which encourages the score prediction of observed interactions to be higher than that of unobserved interactions, as shown below:

[0141]

[0142] in The sigmoid function is convenient for outputting probability distribution. Display items With users Interaction, and project and users There is no interaction between them.

[0143] The second module, adaptive data enhancement, is as follows:

[0144] In this embodiment, noise mitigation views and noise awareness views are dynamically generated to enhance the performance of the model on noisy data. In this implementation, a learnable noise metric function based on a linear attention mechanism is defined to account for the noise amplified by the message passing mechanism of the graph convolutional network. , which represents the interaction noise level between users and items:

[0145]

[0146] Among them, the range of the noise metric function is [0,1], reflecting the degree to which the interaction is affected by noise factors. A score of 1 indicates the user's true preference for the item, while a score of 0 indicates a noise edge. The main idea of ​​this case is to actively filter out the noise edges and preference edges in the input graph through a parameterized network to obtain a noise-mitigated view. and noise-aware view Specifically, these views are constructed on the original bipartite graph of users and items. On the basis of and 1- Obtained by performing edge dropout. Formally, for each layer of the graph convolutional network, their respective adjacency matrices are defined as:

[0147]

[0148] in, The original two graphs are put into the adjacency matrix, and is a specific mask matrix, and the element-wise product ⊙ implements the edge discarding operation. is the adjacency matrix of the noise mitigation graph, is the adjacency matrix of the noise perception graph. For efficient gradient propagation, the mask of each edge (u, i) in the mask matrix is and Follow the parameters and 1- Bernoulli distribution of :

[0149] ;

[0150] ;

[0151] At the beginning of each training epoch, the mask matrix is ​​sampled from a Bernoulli distribution and then a dynamic noise-mitigated view is created. and noise-aware view During training, the graph convolutional network encoder (parameters ) are applied to these two subgraphs individually and multi-layered aggregation is performed to obtain enhanced user and item representations:

[0152] ;

[0153] ;

[0154] Next, we use InfoNCE loss as an auxiliary supervisory signal to control the invariance between representation views. In other words, we encourage the consistency of representations of the same node across views while distinguishing representations of different nodes. AGCF has multiple self-supervised contrastive learning tasks, and the specific formula is as follows:

[0155]

[0156] Noise mitigation and noise-aware representations can improve the original representation's denoising and generalization capabilities, respectively. Encouraging consistency between noise mitigation and noise-aware representations allows the model to capture invariant signals while ignoring variations introduced by noise. This helps extract essential user preferences while mitigating the impact of noise on representation learning.

[0157] The third module, model training, is as follows:

[0158] The training of the AGCF model consists of two parts. In the upper layer, a multi-task training strategy is used to jointly optimize the original recommendation task and the self-supervised contrastive learning task:

[0159] ;

[0160] in, is the recommendation loss function for preference prediction, is the upper layer training loss function, and are hyperparameters that control the strength of contrastive learning and regularization, respectively.

[0161] In the lower layer, the node embeddings of the two enhanced views are used to calculate the BPR loss to optimize noise mitigation and noise-aware view generation. The formula is as follows:

[0162] ;

[0163] in, View for noise mitigation Embedded representation of Recommendation loss function for preference prediction; Noise-aware view Embedded representation of Recommendation loss function for preference prediction.

[0164] Example 3

[0165] This embodiment also provides an electronic device comprising a processor and a memory. The memory is configured to store non-transitory computer-readable instructions. The processor is configured to execute the non-transitory computer-readable instructions. When the processor executes the non-transitory computer-readable instructions, the processor can perform one or more steps of the adaptive data-enhanced graph collaborative filtering recommendation method described above. The memory and processor can be interconnected via a bus system and / or other connection mechanisms.

[0166] For example, the processor can be a central processing unit (CPU), a digital signal processor (DSP), or other forms of processing units with data processing capabilities and / or program execution capabilities, such as a field programmable gate array (FPGA); for example, the central processing unit (CPU) can be an X86 or ARM architecture, etc.

[0167] For example, the memory may include any combination of one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), a hard disk, an erasable programmable read-only memory (EPROM), a portable compact disk read-only memory (CD-ROM), a USB memory, a flash memory, etc. One or more computer program modules may be stored on the computer-readable storage medium, and the processor may execute one or more computer program modules to implement various functions of the electronic device. Various applications and various data, as well as various data used and / or generated by the application, may also be stored in the computer-readable storage medium.

[0168] It should be noted that, in the embodiments of the present application, the specific functions and technical effects of the electronic device can be referred to the above description of the graph collaborative filtering recommendation method with adaptive data enhancement, which will not be repeated here.

[0169] Example 4

[0170] This embodiment further provides a computer-readable storage medium for storing non-transitory computer-readable instructions. For example, when the non-transitory computer-readable instructions are executed by a computer, one or more steps of the graph collaborative filtering recommendation method for adaptive data enhancement described above may be performed.

[0171] For example, the storage medium can be applied to the above-mentioned electronic device. For example, the storage medium can be the memory in the electronic device of Example 2. For example, the relevant description of the storage medium can refer to the corresponding description of the memory in the electronic device of Example 3, and will not be repeated here.

[0172] It should be noted that the storage medium (computer-readable medium) mentioned above in the present application may be a computer-readable signal medium or a non-transitory computer-readable storage medium or any combination of the above two. The non-transitory computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of non-transitory computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0173] In this application, a non-transitory computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a non-transitory computer-readable storage medium that can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0174] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0175] Computer program code for carrying out the operations of the present application may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may execute entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server.

[0176] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, program segment or a part of code, and the module, program segment or a part of code includes one or more executable instructions for realizing the prescribed logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented using a dedicated hardware-based system that performs the prescribed function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0177] The units involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of a unit does not, in some cases, constitute a limitation on the unit itself.

[0178] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0179] The above description is only a partial embodiment of the present application and an illustration of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in this application is not limited to the technical solutions formed by the specific combination of the above technical features, but also includes other technical solutions formed by any combination of the above technical features or their equivalents without departing from the above disclosed concepts. For example, the above features can be replaced with (but not limited to) technical features with similar functions disclosed in this application.

[0180] In addition, although adopting specific order to describe each operation, this should not be interpreted as requiring these operations to be performed in the specific order shown or in sequential order. Under certain environment, multitasking and parallel processing may be advantageous. Similarly, although comprising some specific implementation details in the above discussion, these should not be interpreted as limiting the scope of the application. Some features described in the context of separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment also can be implemented in multiple embodiments individually or in the mode of any suitable subcombination.

[0181] Although the subject matter has been described in language specific to structural features and / or methodological logical acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms of implementing the claims.

Claims

1. A graph collaborative filtering recommendation method with adaptive data augmentation, comprising: Based on the original bipartite graph of user-item, GCN is used to generate the embedding representation of the original bipartite graph of user-item and perform preference prediction. The specific steps are as follows: ； ； in, is the original bipartite graph of user-items Embedding representation, GCN () is a graph neural network, are the trainable parameters of the graph neural network, is the predicted preference of user u for item i, is the embedding representation of item i, is the transpose of the embedding representation of user u; Based on the original bipartite graph of user-items and the noise metric function, edge dropout is performed to obtain the noise-mitigated view and the noise-aware view, and GCN is used to obtain the embedding representation of the noise-mitigated view and the noise-aware view, respectively. Perform contrastive learning based on the embedding representation of the original bipartite graph of user-item, the embedding representation of the noise-mitigated view, and the embedding representation of the noise-aware view; Based on the preference prediction recommendation loss function and contrastive learning loss function, a collaborative filtering recommendation model with adaptive data enhancement is trained.

2. The adaptive data-enhanced graph collaborative filtering recommendation method according to claim 1, characterized in that: The graph neural network is a simplified graph neural network that omits nonlinear activation and feature transformation.

3. The adaptive data-enhanced graph collaborative filtering recommendation method according to claim 1, characterized in that: The specific steps of performing edge discarding based on the original bipartite graph of user-items and the noise metric function to obtain the noise mitigation view and the noise perception view, and using GCN to obtain the embedding representation of the noise mitigation view and the embedding representation of the noise perception view include: Based on the mask matrix and the adjacency matrix of the original bipartite graph of user-items, the adjacency matrices of the noise-mitigated view and the noise-aware view are generated by element-by-element dot product, and then the noise-mitigated view and the noise-aware view are created. The formula is as follows: ； in, is the adjacency matrix of the original bipartite graph of users and items, is the first mask matrix, is the second mask matrix, is the adjacency matrix of the noise-mitigated view, is the adjacency matrix of the noise-aware view; Among them, the mask of each edge of the first mask matrix and the second mask matrix and Follow the and 1- Bernoulli distribution of : ； ； in, is the noise measurement function of user u regarding item i, and the range of the noise measurement function is [0,1]. For noise mitigation view, for noise-aware views; Based on the noise perception view and the noise mitigation view, GCN is used to obtain an embedding representation of the noise mitigation view and an embedding representation of the noise perception view, respectively.

4. The adaptive data-enhanced graph collaborative filtering recommendation method according to claim 1, characterized in that: The specific steps of performing contrastive learning based on the embedding representation of the original bipartite graph of user-items, the embedding representation of the noise mitigation view, and the embedding representation of the noise perception view include: Perform first contrastive learning based on the embedding representation of the original bipartite graph of user-item and the embedding representation of the noise-mitigated view; Perform second contrastive learning based on the embedding representation of the original bipartite graph of user-item and the embedding representation of the noise-aware view; Based on the embedding representation of the noise-aware view and the embedding representation of the noise-mitigated view, a third contrastive learning is performed.

5. The adaptive data-enhanced graph collaborative filtering recommendation method according to claim 1, characterized in that: The specific steps of training and generating the adaptive data-enhanced collaborative filtering recommendation model based on the preference prediction recommendation loss function and the contrastive learning loss function include: In the upper layer, a multi-task training strategy is used to jointly optimize the preference prediction task and the self-supervised contrastive learning task for training. The formula is as follows: ； in, is the recommendation loss function for preference prediction, is the upper layer training loss function, and are hyperparameters that control the strength of contrastive learning and regularization, respectively. is the loss function for contrastive learning; In the lower layer, the recommendation loss function for preference prediction using the embedded representations of the noise-mitigated view and the noise-aware view is trained using the following formula: ； in, A proposed loss function for preference prediction of embedded representations of noise-mitigated views; Recommendation loss function for preference prediction of noise-aware view embedding representations, Training loss function for the lower layer; Based on the upper-layer training loss function and the lower-layer training loss function, an adaptive data-enhanced collaborative filtering recommendation model is generated through iterative training.

6. An electronic device, characterized in that: include: processor; a memory having stored thereon one or more computer program instructions executed on the processor; When the processor runs the computer program instructions, it executes the adaptive data-enhanced graph collaborative filtering recommendation method according to any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that Computer instructions are stored thereon, and when the computer instructions are executed by a processor, the steps of the adaptive data enhanced graph collaborative filtering recommendation method according to any one of claims 1 to 5 are executed.

Images