Data classification method and device based on linear feature enhancement, equipment and storage medium

Abstract

The invention discloses a data classification method and device based on linear feature enhancement, equipment and a storage medium. The method comprises the following steps: respectively carrying out linear propagation on node features and corresponding neighbor node features to obtain a first linear feature and a second linear feature; performing nonlinear mapping on the first linear feature and the second linear feature through KAN nonlinear mapping to obtain a first mapping node feature and a second mapping node feature; determining the embedding of the first node in each layer of the graph convolutional network by combining the first mapping node feature and the second mapping node feature through the graph convolutional network; superposing the embedding of each layer to obtain a target node feature representation of the first node; based on the Sigmoid activation function, the # imgabs0 # activation function and the target node feature representation, determining an enhancement feature of the first node; and converting the enhanced features through a softmax function to obtain a classification result. The method effectively guarantees the calculation efficiency of graph data classification, and improves the classification capability of complex graph data.

Description

Technical Field

[0001] The present invention relates to the technical field of data classification, and particularly to a data classification method, device, equipment and storage medium based on linear feature enhancement. Background Art

[0002] In the prior art, when classifying graph data sets in the medical field, a non-linear model is usually used to represent the complex patterns of the data. However, although the non-linear model has obvious advantages in capturing complex relationships, it brings the problem of increased computational complexity, and the training time becomes correspondingly longer. How to improve the classification ability of the model for complex graph data while ensuring computational efficiency has become an important issue in current research. Summary of the Invention

[0003] Based on this, in view of the above problems, it is necessary to propose a data classification method, device, equipment and storage medium based on linear feature enhancement.

[0004] A data classification method based on linear feature enhancement is used to classify graph data. The graph data contains multiple nodes, each node has a neighbor node, each node corresponds to a node feature, and each neighbor node corresponds to a neighbor node feature. The method includes: Performing linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; Performing non-linear mapping on the first linear feature and the second linear feature respectively through KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; Determining the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; Stacking the embeddings of each layer to obtain the target node feature representation of the first node; Based on the Sigmoid activation function and the activation function and the target node feature representation to determine the enhanced feature of the first node; Converting the enhanced feature through the softmax function to obtain a classification result.

[0005] In one embodiment, the performing linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature is implemented through the following expression: Where is the node feature is the neighbor node feature, is the first linear feature, is the second linear feature, is the adjacency matrix, is the power of the adjacency matrix.

[0006] In one embodiment, the first mapped node feature and the second mapped node feature obtained by non-linearly mapping the first linear feature and the second linear feature respectively through KAN non-linear mapping are implemented by the following expression: Wherein, is the first mapped node feature, is the second mapped node feature, and are trainable coefficients, is the first linear feature, is the second linear feature, is the number of grids of the Fourier basis, is the frequency index.

[0007] In one embodiment, the embedding of the first node in each layer of the graph convolutional network is determined by combining the first mapped node feature and the second mapped node feature through the following expression: Wherein, represents the embedding of node at the th layer, is the first mapped node feature, is the second mapped node feature, is the neighbor set, is the transfer node basis KAN activation function, is the element-wise multiplication.

[0008] In one embodiment, the stacking of the embeddings of each layer to obtain the target node feature representation of the first node is implemented by the following expression: Wherein, is the target node feature representation of the first node u, , ,..., respectively represent the features of the first node u at different layers, is the total number of layers, layer is one of them.

[0009] In one embodiment, based on the Sigmoid activation function and the activation function and the target node feature representation, determining the enhanced feature of the first node is achieved through the following expression: = Wherein, is the weighted feature, is the first mapped node feature, is the adaptive weight, is the Sigmoid activation function, is the first linear transformation, is the second linear transformation, is the activation function, is the third mapped node feature of the target node feature representation, is the target node feature representation of the first node u, and are trainable coefficients, is the number of grids of the Fourier basis, is the frequency index.

[0010] In one embodiment, converting the enhanced feature through the softmax function to obtain the classification result is achieved through the following expression: Wherein, is the classification result, is the weight matrix, is the third mapped node feature of the target node feature representation.

[0011] A data classification device based on linear feature enhancement, the device includes: A propagation module, configured to perform linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; A mapping module, configured to perform non - linear mapping on the first linear feature and the second linear feature respectively through KAN non - linear mapping to obtain a first mapped node feature and a second mapped node feature; A combination module, configured to determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; An overlay module, configured to overlay the embeddings of each layer to obtain the target node feature representation of the first node; A determination module, configured to determine the enhanced feature of the first node based on the Sigmoid activation function and the activation function and the target node feature representation; An acquisition module, configured to convert the enhanced feature through the softmax function to obtain a classification result.

[0012] A computer device, comprising a memory and a processor, where the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to execute the following steps: Perform linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; Perform non-linear mapping on the first linear feature and the second linear feature respectively through the KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; Determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; Stack the embeddings of each layer to obtain the target node feature representation of the first node; Based on the Sigmoid activation function and the activation function and the target node feature representation to determine the enhanced feature of the first node; Convert the enhanced feature through the softmax function to obtain a classification result.

[0013] A computer-readable storage medium, storing a computer program, and when the computer program is executed by a processor, the processor is caused to execute the following steps: Perform linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; Perform non-linear mapping on the first linear feature and the second linear feature respectively through the KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; Determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; Stack the embeddings of each layer to obtain the target node feature representation of the first node; Based on the Sigmoid activation function and the activation function and the target node feature representation to determine the enhanced feature of the first node; The enhanced features are transformed through the softmax function to obtain a classification result.

[0014] In this application, the node features and the corresponding neighbor node features are linearly propagated respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node features is the first node; the first linear feature and the second linear feature are non-linearly mapped through the KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; the embedding of the first node in each layer of the graph convolutional network is determined by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; the embeddings of each layer are stacked to obtain the target node feature representation of the first node; based on the Sigmoid activation function and the activation function and the target node feature representation, the enhanced features of the first node are determined; the enhanced features are transformed through the softmax function to obtain a classification result. While effectively ensuring the computational efficiency of graph data classification, the classification ability for complex graph data is improved. Description of the Drawings

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0016] Among them: Figure 1 It is an application environment diagram of a data classification method based on linear feature enhancement in an embodiment; Figure 2 It is a flowchart of a data classification method based on linear feature enhancement in an embodiment; Figure 3 It is a structural block diagram of a data classification device based on linear feature enhancement in an embodiment; Figure 4 It is a comparison diagram of the influence of the position relationship between the linear layer and the KAN layer on the accuracy rate in an embodiment; Figure 5 It is a comparison diagram of the running time of KSGC and SGC in an embodiment; Figure 6 It is an internal structure diagram of a computer device in an embodiment. Detailed Embodiments

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] In the prior art, when classifying a graph data set in the medical field, a non-linear model is usually used to represent the complex patterns of the data. However, although the non-linear model has obvious advantages in capturing complex relationships, it brings the problem of increased computational complexity, and the training time is also correspondingly lengthened. How to improve the classification ability of the model for complex graph data while ensuring computational efficiency has become an important issue in current research. To solve the above technical problems, the present application provides a data classification method based on linear feature enhancement.

[0019] Figure 1 It is an application environment diagram of the data classification method based on linear feature enhancement in an embodiment. Refer to Figure 1 , the data classification method based on linear feature enhancement is applied to a data classification system based on linear feature enhancement. The data classification system based on linear feature enhancement includes a terminal 110 and a server 120. The terminal 110 and the server 120 are connected through a network. The terminal 110 may specifically be a desktop terminal or a mobile terminal, and the mobile terminal may specifically be at least one of a mobile phone, a tablet computer, a laptop computer, etc. The server 120 may be implemented by an independent server or a server cluster composed of multiple servers. The terminal 110 is used to perform linear propagation on the node features and the corresponding neighbor node features respectively to obtain a first linear feature and a second linear feature. The server 120 is used to perform non-linear mapping on the first linear feature and the second linear feature respectively through a KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature; stack the embeddings of each layer to obtain the target node feature representation of the first node; determine the enhanced feature of the first node based on the Sigmoid activation function and activation function and the target node feature representation; and convert the enhanced feature through a softmax function to obtain a classification result.

[0020] As Figure 2As shown, in one embodiment, a data classification method based on linear feature enhancement is provided. This method can be applied to both terminals and servers. In this embodiment, an example of applying it to a terminal is given. The data classification method based on linear feature enhancement is used to classify graph data in the medical field. The graph data contains multiple nodes, each node has a neighbor node, each node corresponds to a node feature, and each neighbor node corresponds to a neighbor node feature. For example, Figure 2 as shown, the method includes: S10: Perform linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; S20: Perform non-linear mapping on the first linear feature and the second linear feature respectively through KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; S30: Determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; S40: Stack the embeddings of each layer to obtain the target node feature representation of the first node; S50: Based on the Sigmoid activation function and the activation function and the target node feature representation, determine the enhanced feature of the first node; S60: Convert the enhanced feature through the softmax function to obtain the classification result.

[0021] In this application, linear propagation is performed on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; non-linear mapping is performed on the first linear feature and the second linear feature respectively through KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; the embedding of the first node in each layer of the graph convolutional network is determined by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; the embeddings of each layer are stacked to obtain the target node feature representation of the first node; based on the Sigmoid activation function and the activation function and the target node feature representation, determine the enhanced feature of the first node; the enhanced feature is converted through the softmax function to obtain the classification result. While effectively ensuring the computational efficiency of graph data classification, it improves the classification ability for complex graph data.

[0022] In one embodiment, for the step S10, the linear propagation of the node features and the corresponding neighbor node features respectively to obtain the first linear feature and the second linear feature is realized through the following expressions: (1) (2) Wherein, is the node feature, is the neighbor node feature, is the first linear feature, is the second linear feature, is the adjacency matrix, is the power of the adjacency matrix.

[0023] Specifically, after adding self-loops, the spectralization process can avoid the interference of high-frequency noise on the node features and the corresponding neighbor node features. The low-pass filter is used to adjust the node features and the corresponding neighbor node features in the graph data.

[0024] In one embodiment, for the step S20, the first mapped node feature and the second mapped node feature obtained by nonlinearly mapping the first linear feature and the second linear feature respectively through the KAN non-linear mapping are realized through the following expressions: (3) (4) Wherein, is the first mapped node feature, is the second mapped node feature, and are trainable coefficients, is the first linear feature, is the second linear feature, is the number of grids of the Fourier basis, is the frequency index.

[0025] Specifically, the Kolmogorov–Arnold network (KAN) is introduced to nonlinearly map the first mapped node feature and the second mapped node feature. KAN guarantees the approximation ability for any continuous multivariate function. In this application, by designing the KAN mapping layer, a learnable unary function is used to replace the linear transformation of the traditional neural network. To solve this problem, this paper proposes to reconstruct the feature extraction process through the non-linear mapping of the enhanced node basis. The core idea of this KAN design is to use the linear combination of cosine and sine basis functions with different frequencies to comprehensively approximate complex non-linear functions. For example, when When the basis function captures low-frequency features, by adaptively adjusting and weights, the feature expression ability of each channel is dynamically adjusted.

[0026] In one embodiment, for the determination of the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature in step S30, it is achieved through the following expression: (5) where, represents the embedding of node in the -th layer, is the first mapped node feature, is the second mapped node feature, is the neighbor set, is the transfer node basis KAN activation function, is the element-wise multiplication.

[0027] Specifically, the first mapped node feature and the second mapped node feature obtained through the KAN non-linear mapping are integrated into the SGC framework (Simple Graph Convolution). As shown in formula (5), this formula strengthens the information expression ability through the non-linear transformation of the transfer node basis: the first term represents the combination of the first mapped node feature of node with the second mapped node feature (self-product) of its neighbor node , forming the basis for information propagation; the second term represents the interaction between node and neighbor node .

[0028] In one embodiment, for the superposition of the embeddings of each layer in step S40 to obtain the target node feature representation of the first node, it is achieved through the following expression: (6) where, is the target node feature representation of the first node u, , ,..., respectively represent the features of the first node u in different layers, is the total number of layers, -th layer is one of the layers.

[0029] Specifically, this design retains the lightweight architecture of SGC, and at the same time significantly improves the adaptability of the model to graph data through the non-linear ability introduced by the transfer node basis KAN.

[0030] In one embodiment, for the above-mentioned based on the Sigmoid activation function and the activation function and the target node feature representation to determine the enhanced feature of the first node is implemented through the following expression: (7) (8) = (9) Wherein, is the weighted feature, is the first mapped node feature, is the adaptive weight, is the Sigmoid activation function, is the first linear transformation, is the second linear transformation, is the activation function, is the third mapped node feature of the target node feature representation, is the target node feature representation of the first node u, and are trainable coefficients, is the number of grids of the Fourier basis, is the frequency index.

[0031] In one embodiment, the conversion of the enhanced feature through the softmax function to obtain the classification result is implemented through the following expression: (10) Wherein, is the classification result, is the weight matrix, is the third mapped node feature of the target node feature representation. SGC avoids the introduction of non-linear activation functions, significantly improves the computational efficiency, and is suitable for processing large-scale graph data. In this way, the present application can process complex graph data, not only maintaining high efficiency in computing, but also effectively enhancing the expression ability of the model, overcoming the deficiencies of the simplified graph convolutional network.

[0032] Compared with the traditional model with an affine transformation combined activation function, this structure of the KAN combined with SGC breaks the limitation of linear weights, allowing each dimension of features to be adjusted by a specially trained non-linear function without destroying the original simplified linear structure model. This enables the model to have stronger expressive power and capture highly non-linear interaction relationships between neighborhood features and its own features. In this method, the node representation after KAN mapping is denoted as Z = KAN(H), which will be used as the input representation for downstream tasks (such as node classification). The introduction of the KAN module provides a solid theoretical basis and expressive potential for the model, making up for the deficiency of the linear propagation of SGC.

[0033] In summary, this method ensures good performance when dealing with complex graph data, especially when processing large-scale data. L2 regularization effectively constrains the complexity of the model, message dropout enhances robustness, and the low-pass filter suppresses high-frequency noise, ensuring an efficient training process. The present invention is verified as follows: In the experimental verification section, the method based on the non-linear feature enhancement and dynamic screening module using Fourier transform was mainly evaluated. The commonly used graph neural network datasets, Cora, Pubmed, and Citeseer, were utilized, and verification experiments and comparative experiments were designed. ACC and AdaBoost were used as the main evaluation metrics to analyze and verify the effectiveness of the method in this chapter for clustering tasks in high-dimensional data.

[0034] Compare multiple graph neural network models with the traditional SGC method. First is GCN, which performs convolutional operations on the graph structure through the neighborhood information of nodes and captures the high-order relationships of nodes in the graph through multiple layers of graph convolutions. GAT introduces a self-attention mechanism, enabling each node to assign different weights according to the different importance of its neighbors when calculating the influence of its neighbors. FastGCN is an improved GCN method that approximates the graph convolution calculation process as a sampling method, reducing the computational complexity of graph convolution operations and thus improving the training efficiency of graph neural networks. GIN further enhances the learning ability of the graph structure by introducing a graph isomorphism network structure with strong representation ability. LNet is a graph neural network model that introduces local information in the graph and improves the graph representation ability by modeling the local structure of nodes. AdaLNet further enhances the learning of important information in the graph by combining an adaptive weighting mechanism, thereby improving the processing ability for complex graph structures. DGI uses the local and global information of the graph for training, optimizes the graph representation ability, and enhances the effect of graph embedding. Finally, SGC simplifies the convolutional calculation process of traditional GCN. By directly applying a linear transformation, it reduces the computational amount, improves the efficiency, and preserves the structural information of the graph. The introduction of these methods provides multiple different reference criteria for comparing and optimizing models. Compared with complex network models such as GAT and GIN, the traditional SGC sacrifices some representation ability, but by simplifying the calculation, it reduces the risk of overfitting and also avoids training problems caused by high-frequency components. The simplicity of SGC makes it easier to debug and deploy and is suitable for tasks with high requirements for computational efficiency. Nevertheless, SGC is still limited by its simple structure and may not be able to compete with more advanced graph neural network methods (such as GAT and GIN) in complex tasks. However, its high efficiency and low complexity make it a very attractive choice in some application scenarios. The method in this chapter uses all the above-mentioned GCN, GAT, FastGCN, GIN, LNet, DGI, and SGC algorithms as comparison baselines, utilizes two knowledge enhancement modules, namely Fourier basis-enhanced nonlinear mapping and channel adaptive focusing unit, to perform node classification tasks under different settings, and uses two evaluation metrics, ACC and ADABOOST, to evaluate the performance of the method to prove the superiority of the method in this chapter.

[0035] The feature extraction methods adopted are as follows: Fourier transform: The KSGC algorithm (of the present invention) performs nonlinear enhancement on the input node features and the corresponding neighbor node features by introducing the Fourier transform. Through the Fourier transform, the input signal can be transformed from the time domain to the frequency domain, enabling the algorithm to mine more useful information from the frequency features. The Fourier transform can help capture the frequency characteristics in the data and enhance the network's learning ability for local features of the data.

[0036] Adaptive Focus: KSGC incorporates CAFU, which adaptively adjusts the focus degree of features through weights to further optimize the feature extraction process. The adaptive focus module can dynamically enhance certain key features during feature learning while suppressing unimportant information.

[0037] Graph Convolution: During the feature extraction process, the KSGC algorithm also utilizes graph convolutional layers to further enhance feature representation. Graph convolution operations learn features by leveraging graph structure information, enabling the model to effectively capture the relationships and structural information between graph nodes.

[0038] Feature Enhancement: Through the feature enhancement mechanism, KSGC strengthens the expressiveness of input features and enhances the model's robustness to data by introducing more feature information. This enhancement mechanism combines the above-mentioned Fourier transform and graph convolutional network, aiming to improve the depth and accuracy of data representation.

[0039] Based on the experimental results in 1, the following conclusions can be drawn: The node classification accuracy of KSGC is higher than that of other models. The experimental results of each method on the Cora, Pubmed, and Citeseer datasets are compared with the classification accuracies of KSGC and the baseline model on the three datasets. KSGC is higher than other models on Cora, Citeseer, and PubMed respectively.

[0040] Based on the ablation experiment results in Table 2, the following conclusions can be drawn from the ablation experiment, verifying the contributions of the Graph KAN and CAFU modules. After removing GraphKAN, the accuracies of PubMed, Cora, and Citeseer and the performance after removing CAFU both further decrease. This indicates that GraphKAN enhances neighborhood feature interaction through the Fourier kernel, while CAFU optimizes feature fusion through channel attention, and neither can be absent.

[0041] Based on the experimental results in Table 3, the following conclusions can be drawn: SGC + AdaBoost performs significantly worse than independent SGC on the test set, and the training ACC is much higher than the test ACC, indicating severe overfitting. This is directly related to the conflict between the linear characteristics of SGC and the non-linear integration objective of AdaBoost, further demonstrating the advantages of the linear model of SGC. Figure 1Shows the performance impact of the GraphKAN layer before and after the linear layer. When GraphKAN is placed before the linear layer, the ACC of PubMed, Cora, and Citeseer all increases respectively. This is because the Fourier kernel transform of GraphKAN can a priori enhance the frequency-domain information of the input features, providing a more discriminative feature basis for subsequent linear propagation. Subsequent work also needs to place it before the linear layer to achieve better results. Figure 2 Compares the training time of KSGC and SGC. The training time of KSGC increases respectively on Cora, PubMed, and Citeseer. This is due to the additional overhead introduced by the Fourier kernel calculation of GraphKAN, but it is still within an acceptable range. In the future, it can be further optimized and improved through other methods such as sparsification or parallelization.

[0042] According to Figure 4 The experimental results can draw the following conclusions: The position of KAN and the linear layer will affect the accuracy of node classification. Adding a linear layer before the KAN layer fails to improve the performance of the model and may even affect the model's effectiveness. By comparing the changes in accuracy with different positions of the linear layer and KAN in the figure on different datasets. For the PubMed, Citeseer, and Cora datasets, there are significant differences in accuracy before and after adding the linear layer. The accuracy of KAN is lower on all datasets before adding the linear layer, especially on the PubMed dataset, with a more obvious gap. After adding the linear layer to KAN, the accuracy has increased significantly, especially on the Cora dataset, with a more prominent performance. This indicates that the structure with KAN before the linear layer can more effectively improve the accuracy. Therefore, in this model design, placing the KAN layer before the linear layer may be more conducive to improving the model performance, and there is more room for exploration on other models.

[0043] According to Figure 5The experimental results can draw the following conclusions: The introduction of KAN has far lower computational efficiency than the original model. As can be seen from the figure, there are differences in the time consumption of the KSGC model and the SGC model on different datasets. For the PubMed, Cora, and Citeseer datasets, the training time of the KSGC model (red and green bars) is significantly higher than that of the SGC model (blue and purple bars). Especially on the PubMed and Cora datasets, the training time of KSGC is significantly higher than that of SGC, indicating that the KSGC model has a certain disadvantage in terms of time consumption compared to the SGC model. Both the training time and the total time of the KSGC model (green bar and blue bar) are higher than those of the SGC model, showing that it is relatively more complex in the processing process. This may also be due to the introduction of the KAN layer, which increases the computational amount, because existing experiments have proved that KAN is less efficient than some basic models in some cases. Although the KAN layer can enhance the performance and feature extraction ability of the model, its computational efficiency still needs to be improved compared with traditional models such as MLP. Therefore, although the KAN model shows strong performance in some tasks, there is still a certain gap in efficiency compared with the traditional SGC model and MLP model, and there is great room for development in the future, and its efficiency can be improved through further optimization.

[0044] The present invention also provides a data classification device based on linear feature enhancement, as Figure 3 shown, the device includes: A propagation module 10 for linearly propagating the node features and the corresponding neighbor node features respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node features is the first node; A mapping module 20 for non-linearly mapping the first linear feature and the second linear feature respectively through KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; A combination module 30 for determining the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; A stacking module 40 for stacking the embeddings of each layer to obtain the target node feature representation of the first node; A determination module 50 for determining the enhanced feature of the first node based on the Sigmoid activation function and the activation function and the target node feature representation; An acquisition module 60 for converting the enhanced feature through the softmax function to obtain a classification result.

[0045] The present invention also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs the following steps: S10: Perform linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; S20: Perform non - linear mapping on the first linear feature and the second linear feature respectively through KAN non - linear mapping to obtain a first mapped node feature and a second mapped node feature; S30: Determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; S40: Stack the embeddings of each layer to obtain the target node feature representation of the first node; S50: Based on the Sigmoid activation function and the activation function and the target node feature representation, determine the enhanced feature of the first node; S60: Convert the enhanced feature through the softmax function to obtain the classification result.

[0046] A computer - readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor performs the following steps: S10: Perform linear propagation on the node feature and the corresponding neighbor node feature respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node feature is the first node; S20: Perform non - linear mapping on the first linear feature and the second linear feature respectively through KAN non - linear mapping to obtain a first mapped node feature and a second mapped node feature; S30: Determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through the graph convolutional network; S40: Stack the embeddings of each layer to obtain the target node feature representation of the first node; S50: Based on the Sigmoid activation function and the activation function and the target node feature representation, determine the enhanced feature of the first node; S60: Convert the enhanced feature through the softmax function to obtain the classification result.

[0047] Figure 6 shows the internal structure diagram of a computer device in an embodiment. The computer device can specifically be a terminal or a server. As Figure 6As shown in the figure, the computer device includes a processor, a memory, and a network interface connected via a system bus. Among them, the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and can also store a computer program. When the computer program is executed by the processor, the processor can implement a data classification method based on linear feature enhancement. The internal memory can also store a computer program. When the computer program is executed by the processor, the processor can execute a data classification method based on linear feature enhancement. Those skilled in the art can understand that Figure 6 The structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0048] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to memory, storage, database, or other media used in the various embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0049] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0050] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A data classification method based on linear feature enhancement for classifying graph data. The graph data contains multiple nodes, each node has a neighbor node, each node corresponds to a node feature, and each neighbor node corresponds to a neighbor node feature. It is characterized in that The method includes: Performing linear propagation on the node features and the corresponding neighbor node features respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node features is the first node; Performing non-linear mapping on the first linear feature and the second linear feature respectively through KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; Determining the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through a graph convolutional network; Stacking the embeddings of each layer to obtain the target node feature representation of the first node; Based on the Sigmoid activation function and determine the enhanced feature of the first node based on the activation function and the target node feature representation; Converting the enhanced feature through a softmax function to obtain a classification result.

2. The data classification method based on linear feature enhancement according to claim 1, wherein The performing of linear propagation on the node features and the corresponding neighbor node features respectively to obtain a first linear feature and a second linear feature is implemented through the following expression: Among them, is the node feature, is the neighbor node feature, is the first linear feature, is the second linear feature, is the adjacency matrix, is the power of the adjacency matrix.

3. The data classification method based on linear feature enhancement according to claim 2, wherein The obtaining of the first mapped node feature and the second mapped node feature by performing non-linear mapping on the first linear feature and the second linear feature respectively through KAN non-linear mapping is implemented through the following expression: Among them, is the first mapped node feature, is the second mapped node feature, and are trainable coefficients, is the first linear feature, is the second linear feature, is the number of grids of the Fourier basis, is the frequency index.

4. The data classification method based on linear feature enhancement according to claim 1, wherein The determining of the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through a graph convolutional network is implemented through the following expression: Among them, is expressed as a node in the layer embedding, is the first mapped node feature, is the second mapped node feature, is the neighbor set, is the transfer node base KAN activation function, is the element-wise multiplication.

5. The data classification method based on linear feature enhancement according to claim 4, wherein The stacking of the embeddings of each layer to obtain the target node feature representation of the first node is implemented through the following expression: Among them, is the target node feature representation of the first node u, , ,..., respectively represent the features of the first node u at different layers, is the total number of layers, The l-th layer is one of the layers.

6. The data classification method based on linear feature enhancement according to claim 5, wherein The above-mentioned based on the Sigmoid activation function and determine the enhanced feature of the first node by the following expression according to the activation function and the target node feature representation: = Among them, is the weighted feature, is the first mapping node feature, is the adaptive weight, is the Sigmoid activation function, is the first linear transformation, is the second linear transformation, is the activation function, is the third mapping node feature of the target node feature representation, is the target node feature representation of the first node u, and are the trainable coefficients, is the number of grids of the Fourier basis, is the frequency index.

7. The data classification method based on linear feature enhancement according to claim 6, wherein The converting of the enhanced feature through a softmax function to obtain a classification result is implemented through the following expression: Among them, is the classification result, is the weight matrix, is the third mapped node feature of the target node feature representation.

8. A data classification device based on linear feature enhancement, characterized in that, The apparatus includes: A propagation module, configured to perform linear propagation on the node features and the corresponding neighbor node features respectively to obtain a first linear feature and a second linear feature; the node corresponding to the node features is the first node; A mapping module, configured to perform non-linear mapping on the first linear feature and the second linear feature respectively through KAN non-linear mapping to obtain a first mapped node feature and a second mapped node feature; A combining module, configured to determine the embedding of the first node in each layer of the graph convolutional network by combining the first mapped node feature and the second mapped node feature through a graph convolutional network; A stacking module, configured to stack the embeddings of each layer to obtain the target node feature representation of the first node; A determination module, configured to determine an enhanced feature of the first node based on a Sigmoid activation function and the activation function and the target node feature representation; An obtaining module, configured to convert the enhanced feature through a softmax function to obtain a classification result.

9. A computer device, characterized in that, It includes a memory and a processor. When the computer program stored in the memory is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored. When the computer program is executed by a processor, the processor executes the steps of the method according to any one of claims 1 to 7.

Images