A data processing method, system, device and medium based on graph neural network compression

Abstract

The application relates to a data processing method, system, device and medium based on graph neural network compression, comprising the following steps: obtaining a historical data set and preprocessing the historical data set to obtain a graph structure data set; based on the graph structure data set, a hybrid training method is used to accelerate training of a pre-constructed graph neural network model to obtain a trained graph neural network model; and the trained graph neural network model is used to process a target graph structure to obtain a data processing result. The application uses a hybrid training method to train the pre-constructed graph neural network model, and uses a hierarchical compression method to accelerate training of the graph neural network model, so that the graph neural network model converges quickly; then, the graph neural network model is switched to a normal training mode without compression to fine-tune parameters of the graph neural network model, so that faster convergence speed and optimal accuracy are achieved, and the application can be widely applied to the field of big data processing.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of big data processing, and particularly relates to a data processing method, system, device and medium based on graph neural network compression. BACKGROUND

[0002] As a core technology for processing graph-structured data, graph neural networks (GNNs) have shown remarkable advantages in recent years in fields such as social network analysis, recommendation systems, biological molecule modeling, and natural language processing.

[0003] The core operations of GNNs include propagation and transformation, where propagation is used to aggregate neighbor node information, and transformation is used to perform nonlinear conversion on node features through a neural network to capture complex relationships in graph structures. However, with the explosive growth of graph data (such as trillion-edge-level social graphs and biological molecule networks), the training of GNNs faces severe efficiency challenges.

[0004] Existing GNN frameworks (such as PyG and DGL) mainly improve efficiency through hardware acceleration (such as cuBLAS / cuSparse libraries) and distributed training, but still have the following shortcomings:

[0005] 1) Coupling of computation and storage: The intermediate results of propagation operations (such as edge features in GAS) occupy a large amount of memory, especially in large-scale graphs, which can easily cause out-of-memory (OOM) problems.

[0006] 2) Redundant computation not fully utilized: Repetitive computation in transformation operations is not identified, resulting in low GPU utilization.

[0007] 3) Difficulty in balancing accuracy and efficiency: Methods such as sampling and quantization may improve speed but sacrifice model accuracy, especially in scenarios requiring high-fidelity features (such as biomedical graph analysis). SUMMARY

[0008] To address the above problems, the present application provides a data processing method, system, device and medium based on graph neural network compression, which accelerates propagation and transformation operations in graph neural networks based on hierarchical compression, thereby improving data processing efficiency.

[0009] To achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0010] In a first aspect, the present application provides a data processing method based on graph neural network compression, comprising the following steps:

[0011] Obtain a historical data set and preprocess it to obtain a graph-structured data set;

[0012] Based on the graph structure data set, a pre-constructed graph neural network model is accelerated by using a hybrid training method, and a trained graph neural network model is obtained.

[0013] The trained graph neural network model is used to process the target graph structure, and a data processing result is obtained.

[0014] Further, the graph structure data set is based on the graph structure data set, and the pre-constructed graph neural network model is accelerated by using a hybrid training method, and a trained graph neural network model is obtained, comprising:

[0015] The given graph structure is compressed, and the generated compressed graph structure is applied to the propagation operator of the graph neural network model to obtain a compressed-based propagation operator;

[0016] The node features are compressed, and the generated compressed features are applied to the conversion operator of the graph neural network model to obtain a compressed-based conversion operator;

[0017] According to the actual demand, the obtained compressed-based propagation operator and conversion operator are arranged to obtain a graph neural network model that has been preliminarily trained;

[0018] The graph neural network model that has been preliminarily trained is fine-tuned by using a conventional training mode to obtain a trained graph neural network model.

[0019] Further, the given graph structure is compressed, and the generated compressed graph structure is applied to the propagation operator of the graph neural network model to obtain a compressed-based propagation operator, comprising:

[0020] The given graph structure is compressed by using the CompressGraph method to generate a preliminary compressed graph, and the compressed graph contains original nodes and virtual nodes;

[0021] The virtual vertices and edges with low performance contribution in the preliminary compressed graph are filtered by using a double-layer filtering method to obtain a filtered compressed graph;

[0022] A multi-step propagation engine is used to apply the propagation operator of the graph neural network model to the filtered compressed graph to obtain a compressed-based propagation operator.

[0023] Further, the virtual vertices and edges with low performance contribution in the preliminary compressed graph are filtered by using a double-layer filtering method to obtain a filtered compressed graph, comprising:

[0024] The virtual nodes in the preliminary compressed graph are filtered based on the contribution according to a preset contribution threshold value until there is no virtual node in the compressed graph whose contribution value is less than the contribution threshold value;

[0025] Starting from the maximum depth of the compressed graph, delete all virtual nodes layer by layer until the maximum depth of the compressed graph reaches the preset depth threshold.

[0026] Furthermore, the multi-step propagation engine is employed to apply the propagation operator of the graph neural network model to the filtered compressed graph, resulting in a compression-based propagation operator, including:

[0027] Based on preset rules, determine the depth of each node and edge in the filtered compressed image;

[0028] Based on the depth of each node and edge in the compressed graph, nodes and edges with the same depth are divided into the same subgraph, and the number of all subgraphs is denoted as N.

[0029] Based on the depth, for subgraphs with depths from 1 to N, the propagation operator is applied sequentially to propagate the features.

[0030] Furthermore, the compression of node features and the application of the generated compressed features to the transformation operator of the graph neural network model includes the following steps:

[0031] Obtain the input nodes at the current level, and apply a clustering algorithm to compress the features of the input nodes to obtain compressed features;

[0032] Repeat the previous step to compress the node features at other levels to obtain the corresponding compressed features;

[0033] A multi-level data reuse mechanism is adopted, and transformation operators are applied to the compression features at each level.

[0034] Furthermore, the step of obtaining the input nodes at the current level and applying a clustering algorithm to compress the features of the input nodes to obtain compressed features includes:

[0035] Apply a clustering algorithm to the input nodes of the current level to obtain the node index and a set of representative features for each original node;

[0036] The representative features are input into the original transformation operation to obtain the representative output;

[0037] Based on the node indices representing the output and the original nodes, the node output is reconstructed as a compression feature.

[0038] Secondly, the present invention provides a data processing system based on graph neural network compression, comprising:

[0039] The data acquisition module is used to acquire historical datasets and preprocess them to obtain graph structure datasets;

[0040] The model training module is used to accelerate the training of a pre-built graph neural network model based on a graph structure dataset using a hybrid training method, so as to obtain a trained graph neural network model.

[0041] The data processing module is used to process the target graph structure using a trained graph neural network model to obtain the data processing results.

[0042] Thirdly, the present invention provides a computer-readable storage medium for storing one or more programs, said one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any method.

[0043] Fourthly, the present invention provides a computing device comprising: one or more processors and a memory, wherein the memory stores one or more programs and is configured to be executed by the one or more processors, the one or more programs including instructions for performing any method.

[0044] The present invention has the following advantages due to the adoption of the above technical solutions:

[0045] 1. In training graph neural networks, this invention adopts a hybrid training method, which accelerates the training process based on hierarchical compression to enable rapid convergence. Then, the parameters of the graph neural network model are fine-tuned using a conventional training mode, achieving faster convergence speed and accuracy, and making it suitable for application fields with high accuracy requirements.

[0046] 2. This invention compresses the propagation operator and transformation operator of the graph neural network separately. The independent compression and application of the two operators makes this invention applicable to various types of graph neural network structures, thus broadening its scope of application.

[0047] Therefore, this invention can be widely applied in the field of big data processing. Attached Figure Description

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0049] Figure 1 This is the existing graph neural network processing flow;

[0050] Figure 2a and Figure 2b This is a schematic diagram of redundant operations in propagation and transformation in existing graph neural networks;

[0051] Figure 3This is a flowchart of the data processing method based on graph neural network compression provided in this embodiment of the invention;

[0052] Figure 4 This is a flowchart of a compression-based graph neural network acceleration method provided in an embodiment of the present invention;

[0053] Figure 5 This is the compression conversion operator provided in the embodiments of the present invention;

[0054] Figure 6 This is an example of locality-sensitive hashing provided in the embodiments of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0056] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0057] like Figure 1 The diagram shows the execution flow of an existing graph neural network. The input to a typical graph neural network includes a graph A and the feature information X corresponding to each node in graph A. (0) The training process of a graph neural network includes two operators: propagation and transformation. In the propagation phase, each node in the graph propagates its features to its neighbors. In the transformation phase, the node features undergo a non-linear transformation through a deep neural network (DNN) layer. A graph neural network model is composed of propagation and transformation operators, and the order of propagation and transformation varies slightly depending on the graph neural network model architecture.

[0058] like Figure 2a and Figure 2b As shown, this invention discovers that in the execution flow of existing graph neural networks, both the propagation and transformation processes are redundant, including:

[0059] 1) Data redundancy during the transmission process

[0060] like Figure 2aAs shown, the core of the propagation operation is that nodes aggregate information through edges. Its data redundancy mainly stems from repeated propagation by common neighbors in the graph structure. Specifically, when multiple nodes transmit information to the same group of common neighbor nodes, duplicate message paths are generated. For example, if nodes 1 and 2 are both connected to nodes 3, 4, and 5, traditional propagation requires 6 transmissions: 1→3, 1→4, 1→5, 2→3, 2→4, and 2→5. However, in practice, the features of nodes 1 and 2 can be aggregated to a virtual node first, and then transmitted to nodes 3, 4, and 5 respectively through the virtual node, requiring only 5 transmissions (reducing the amount of operation by 16.7%).

[0061] 2) Computational redundancy during the conversion process

[0062] like Figure 2b As shown, the transformation operation uses a neural network (such as a linear layer and activation function in a GCN) to transform node features. Its computational redundancy mainly stems from the repeated computation of similar features. Specifically, this manifests as computational redundancy due to feature similarity: when multiple nodes have the same or highly similar features, their computations through the same transformation layer are completely identical. For example, if nodes 1, 2, and 3 all have the input feature x, after the linear transformation W and activation function, their output features are completely identical. In this case, only one transformation needs to be computed, and the remaining nodes can directly reuse the results, eliminating 2 / 3 of the computational load.

[0063] Based on this, some embodiments of the present invention provide a data processing method based on graph neural network compression. This method utilizes a graph neural network model to process data. During the training process of the graph neural network model, compression is first applied to generate a compressed graph structure and compressed features. Then, the compressed graph structure and compressed features are applied to propagation and transformation operators. Finally, these two operators are arranged to form the desired graph neural network model. The present invention accelerates the training process of graph neural networks based on compression methods, reducing training time and cost. The trained graph neural network model can be applied to multiple technical fields, including social networks and biomedical molecular modeling. In practical applications, it can greatly improve data processing efficiency.

[0064] Correspondingly, in other embodiments of the present invention, a data processing system, device, and medium based on graph neural network compression are provided.

[0065] Example 1

[0066] like Figure 3 and Figure 4 As shown, the present invention provides a data processing method based on graph neural network compression, which includes the following steps:

[0067] 1) Obtain historical datasets and preprocess them to obtain graph structure datasets;

[0068] 2) Based on the graph structure dataset, a hybrid training method is used to accelerate the training of the pre-built graph neural network model to obtain a trained graph neural network model;

[0069] 3) Use the trained graph neural network model to process the target graph structure and obtain the data processing results.

[0070] Furthermore, in step 1) above, the historical dataset in this embodiment can be data from social networks, biomedical molecular modeling, etc., which includes several users and the relationships between them. Each user also includes some attribute information, such as age, gender, and other attribute information. After preprocessing the collected data, a corresponding graph structure is obtained, including nodes and node feature information.

[0071] Furthermore, in step 2) above, when training the pre-built graph neural network model using a hybrid training method in this embodiment, firstly, a hierarchical compression-based method is used to accelerate the training of the graph neural network model, enabling it to converge quickly; then, the model is switched to an uncompressed conventional training mode to fine-tune the parameters of the graph neural network model. This hybrid training method combines faster convergence speed with the ability to achieve optimal accuracy, ensuring suitability for applications requiring extremely high accuracy.

[0072] Specifically, it includes the following steps:

[0073] 2.1) Compress the given graph structure and apply the resulting compressed graph structure to the propagation operator of the graph neural network model to obtain the compression-based propagation operator;

[0074] 2.2) The node features are compressed, and the generated compressed features are applied to the transformation operator of the graph neural network model to obtain the compression-based transformation operator;

[0075] 2.3) Arrange the obtained compression-based propagation operators and transformation operators according to actual needs to obtain a preliminarily trained graph neural network model;

[0076] 2.4) Using the conventional training mode, the pre-trained graph neural network model is fine-tuned to obtain a trained graph neural network model that meets the preset requirements.

[0077] Furthermore, in step 2.1) above, the graph compression method of this invention draws inspiration from the CompressGraph method. In the CompressGraph method, shared neighbors in the graph are recursively identified as virtual nodes, thereby generating a graph containing both original vertices and virtual vertices. While this invention is inspired by CompressGraph and employs a virtual node extraction method, directly applying it to GNNs is not feasible. In CompressGraph, vertices lack features, while feature computation is the primary time overhead in GNNs, requiring efficient GPU operators for both node features and topology. Graph traversal in CompressGraph involves multiple iterations and uses a state array to record node states, while the number of nodes processed in each iteration of a GNN is fixed. Moreover, in GNN models dominated by transformation operators, the propagation optimization of CompressGraph has limitations. Therefore, this invention performs data and computation compression on propagation and transformation separately, and develops a multi-step propagation engine to achieve efficient compressed graph propagation.

[0078] In this embodiment, the compression-based propagation operator includes two steps: offline and online. Since the graph structure can be reused after compression, the graph structure is compressed in the offline stage after being given, and the compression-based propagation operator is applied to the compressed graph structure in the online stage.

[0079] Specifically, it includes the following steps:

[0080] 2.1.1) The CompressGraph method is used to compress the given graph structure to generate a preliminary compressed graph, which contains the original nodes and virtual nodes;

[0081] 2.1.2) A two-layer filtering method is used to filter virtual vertices and edges with low performance contribution in the initial compressed graph to obtain the filtered compressed graph;

[0082] 2.1.3) A multi-step propagation engine is used to apply the propagation operator of the graph neural network model to the filtered compressed graph to obtain a compression-based propagation operator.

[0083] Furthermore, in step 2.1.2) above, since the multi-step propagation engine is affected by the number of parallel rounds and the number of virtual nodes, this embodiment filters the virtual nodes in the initial compressed graph based on two dimensions to further improve propagation efficiency. First, the concept of virtual node contribution is introduced to measure its impact on performance; the higher the contribution, the better. The contribution value of a virtual node is the product of its number of incoming neighbors and its number of outgoing neighbors. For example, for a virtual node with 2 incoming neighbors and 3 outgoing neighbors, its contribution value is 6. Second, the maximum depth of the compressed graph is considered, which corresponds to the total number of parallel rounds.

[0084] Specifically, it includes the following steps:

[0085] 2.1.2.1) Based on the preset contribution threshold, the virtual nodes in the preliminary compressed graph G are filtered according to their contribution until there are no virtual nodes in the compressed graph G with a contribution value less than the contribution threshold.

[0086] Specifically, it includes the following steps:

[0087] a) For each virtual node v in the initial compressed graph G, calculate its contribution value and insert it into an ascending priority queue;

[0088] b) Extract the virtual node with the smallest contribution value from the queue and compare its contribution value with the preset contribution threshold. If its contribution value is less than the contribution threshold, delete the virtual node and its corresponding edge from the compressed graph G, and update the contribution values ​​of its in-neighbor and out-neighbor in the queue.

[0089] c) Repeat step b) until there are no virtual nodes in the queue with a contribution value less than the contribution threshold, then the contribution-based filtering process terminates.

[0090] 2.1.2.2) Starting from the maximum depth of the compressed graph G, delete all virtual nodes layer by layer until the maximum depth of the compressed graph G reaches the preset depth threshold.

[0091] Furthermore, in step 2.1.3) above, to ensure that the propagation results on the compressed graph are consistent with those on the original graph, this embodiment applies the propagation operator to the compressed graph based on a multi-step propagation engine. The specific process is as follows:

[0092] 2.1.3.1) Determine the depth of each node and edge on the filtered compressed graph according to the preset rules.

[0093] In this embodiment, the depth of each node and edge in the filtered compressed graph is denoted as D, and D is defined as follows:

[0094] a) For the original node in the graph, its depth is denoted as 1;

[0095] b) For a virtual node, its depth is the maximum depth of all its neighbors plus 1;

[0096] c) For all edges, the depth is the same as the depth of the starting node.

[0097] 2.1.3.2) Based on the depth of each node and edge in the compressed graph, nodes and edges with the same depth are divided into the same subgraph, and the number of all subgraphs is denoted as N.

[0098] 2.1.3.3) Based on the depth size, for subgraphs with depths from 1 to N, the propagation operator is applied sequentially to propagate the features.

[0099] In this embodiment, the propagation operator is the same as the original propagation operator. Their inputs are a graph structure and a set of features. When propagating features, a space is first created to temporarily store the features of virtual vertices. If the propagation result is an original vertex, it is placed in the space of the original vertex; if it is a virtual vertex, it is placed in the temporarily created space.

[0100] Further, in step 2.2) above, the node features are compressed, and the generated compressed features are applied to the transformation operator of the graph neural network model, including the following steps:

[0101] 2.2.1) Obtain the input nodes of the current level, and apply a clustering algorithm to compress the features of the input nodes to obtain compressed features;

[0102] 2.2.2) Repeat step 2.2.1) to compress the node features at other levels to obtain the corresponding compressed features;

[0103] 2.2.3) A multi-level data reuse mechanism is adopted, and transformation operators are applied to the compression features of each level.

[0104] Furthermore, in step 2.2.1) above, as Figure 5 As shown, it includes the following steps:

[0105] 2.2.1.1) Apply a clustering algorithm to the input nodes of the current level to obtain the node index and a set of representative features for each original node;

[0106] 2.2.1.2) Input the representative features into the original transformation operation to obtain the representative output;

[0107] 2.2.1.3) Based on the node index of the representative output and the original node, the node output is restored as a compression feature.

[0108] Furthermore, in step 2.2.1.1) above, the clustering algorithm in this embodiment adopts a projection-based locality-sensitive hashing algorithm, including the following steps:

[0109] a) Randomly generate H vectors with a length equal to the width F of the vertex features, and transpose them to obtain a random matrix R of size F*H;

[0110] b) Multiply the original node features by the random matrix R to obtain the output matrix;

[0111] In this embodiment, the original node features are represented as N*F, where N is the number of nodes. The output matrix obtained by multiplying the original vertex features by the random matrix R has a size of N*H.

[0112] c) Divide all elements in the output matrix, marking values ​​greater than 0 as 1 and values ​​less than or equal to zero as 0, to obtain a matrix P in which all values ​​are either 0 or 1, with the size still N*H;

[0113] d) Combine each row of this 0 / 1 matrix into a 32-bit / 64-bit integer of size N*1;

[0114] e) For each row, if its numerical value is equal to that of another row, then they are considered to be in the same class;

[0115] f) Complete clustering.

[0116] like Figure 6 The diagram illustrates an example of Locality Sensitive Hashing (LSH). In the diagram, x1, x2, and x3 are three input features with a dimension of 4, and h1 and h2 are two random vectors. CompressGNN first obtains the projection results of x1, x2, and x3 onto the random vectors, and then converts the projection results into bit vectors. Since the bit vectors of x1 and x2 are the same, they are clustered into one class, while x3 belongs to a different class.

[0117] Furthermore, in step 2.2.3) above, the multi-level data reuse mechanism adopted in this embodiment includes:

[0118] The first level of reuse occurs within successive transformation operations. In this case, instead of clustering for every transformation operation, the node indices obtained from the first clustering are reused within these transformations.

[0119] The second level of reuse involves reusing data across multiple rounds. During training, clustering is not performed in every round, but rather after several rounds. At this point, the frequency of the second-level reuse is determined in real-time based on the convergence speed. Assuming a convergence speed threshold of x, if the convergence speed is greater than x, then re-clustering is performed; if it is less than x, the original features are used.

[0120] The update frequency is dynamically adjusted during training. Specifically, CompressGNN monitors the degree of loss change between training epochs. As the loss gradually converges, the update frequency is reduced accordingly.

[0121] The third level of reuse involves reusing node indices. This invention uses a method similar to the second level of reuse to control whether to update the indices. The frequency of the second level of reuse can also be dynamically determined by inferring the changes in hidden layer features during training based on the convergence speed.

[0122] Example 2

[0123] The above-described embodiment 1 provides a data processing method based on graph neural network compression. Correspondingly, this embodiment provides a data processing system based on graph neural network compression. The system provided in this embodiment can implement the data processing method based on graph neural network compression of embodiment 1. The system can be implemented by software, hardware, or a combination of both. For example, the system may include integrated or separate functional modules or units to execute the corresponding steps in the methods of embodiment 1. Since the system in this embodiment is basically similar to the method embodiment, the description process in this embodiment is relatively simple. For relevant details, please refer to the description of embodiment 1. The system embodiment provided in this embodiment is merely illustrative.

[0124] The data processing system based on graph neural network compression provided in this embodiment includes:

[0125] The data acquisition module is used to acquire historical datasets and preprocess them to obtain graph structure datasets;

[0126] The model training module is used to accelerate the training of a pre-built graph neural network model based on a graph structure dataset using a hybrid training method, so as to obtain a trained graph neural network model.

[0127] The data processing module is used to process the target graph structure using a trained graph neural network model to obtain the data processing results.

[0128] Example 3

[0129] This embodiment provides a processing device corresponding to the data processing method based on graph neural network compression provided in Embodiment 1. The processing device can be a client-side processing device, such as a mobile phone, laptop, tablet computer, desktop computer, etc., to execute the method of Embodiment 1.

[0130] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the data processing method based on graph neural network compression provided in Embodiment 1.

[0131] Preferably, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0132] Preferably, the processor can be any type of general-purpose processor such as a central processing unit (CPU) or a digital signal processor (DSP), and there is no limitation herein.

[0133] Example 4

[0134] The data processing method based on graph neural network compression in Embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the data processing method based on graph neural network compression described in Embodiment 1 are loaded.

[0135] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0136] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0137] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0138] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0139] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A data processing method based on graph neural network compression, characterized in that, Includes the following steps: Obtain historical datasets and preprocess them to obtain graph structure datasets; Based on a graph structure dataset, a hybrid training method is used to accelerate the training of a pre-built graph neural network model, resulting in a trained graph neural network model, including: The given graph structure is compressed, and the resulting compressed graph structure is applied to the propagation operator of the graph neural network model to obtain the compression-based propagation operator; The node features are compressed, and the resulting compressed features are applied to the transformation operator of the graph neural network model to obtain the compression-based transformation operator; The obtained compression-based propagation and transformation operators are arranged according to actual needs to obtain a preliminarily trained graph neural network model. Using a conventional training mode, the pre-trained graph neural network model is fine-tuned to obtain a well-trained graph neural network model. The trained graph neural network model is used to process the target graph structure to obtain the data processing results.

2. The data processing method based on graph neural network compression as described in claim 1, characterized in that, The process of compressing a given graph structure and applying the resulting compressed graph structure to the propagation operator of a graph neural network model to obtain a compression-based propagation operator includes: The CompressGraph method is used to compress a given graph structure to generate a preliminary compressed graph, which contains original nodes and virtual nodes. A two-layer filtering method is used to filter virtual vertices and edges with low performance contribution in the initial compressed graph to obtain the filtered compressed graph. A multi-step propagation engine is used to apply the propagation operator of the graph neural network model to the filtered compressed graph, resulting in a compression-based propagation operator.

3. The data processing method based on graph neural network compression as described in claim 2, characterized in that, The method employs a two-layer filtering approach to filter virtual vertices and edges with low performance contribution in the initial compressed graph, resulting in a filtered compressed graph, including: Based on a preset contribution threshold, the virtual nodes in the initial compressed graph are filtered according to their contribution until there are no virtual nodes in the compressed graph with a contribution value less than the contribution threshold. Starting from the maximum depth of the compressed graph, delete all virtual nodes layer by layer until the maximum depth of the compressed graph reaches the preset depth threshold.

4. The data processing method based on graph neural network compression as described in claim 2, characterized in that, The multi-step propagation engine is employed to apply the propagation operator of the graph neural network model to the filtered compressed graph, resulting in a compression-based propagation operator, including: Based on preset rules, determine the depth of each node and edge in the filtered compressed image; Based on the depth of each node and edge in the compressed graph, nodes and edges with the same depth are divided into the same subgraph, and the number of all subgraphs is denoted as N. Based on the depth, for subgraphs with depths from 1 to N, the propagation operator is applied sequentially to propagate the features.

5. The data processing method based on graph neural network compression as described in claim 1, characterized in that, The process of compressing node features and applying the resulting compressed features to the transformation operator of the graph neural network model includes the following steps: Obtain the input nodes at the current level, and apply a clustering algorithm to compress the features of the input nodes to obtain compressed features; Repeat the previous step to compress the node features at other levels to obtain the corresponding compressed features; A multi-level data reuse mechanism is adopted, and transformation operators are applied to the compression features at each level.

6. The data processing method based on graph neural network compression as described in claim 5, characterized in that, The process of obtaining the input nodes at the current level and applying a clustering algorithm to compress the features of the input nodes to obtain compressed features includes: Apply a clustering algorithm to the input nodes of the current level to obtain the node index and a set of representative features for each original node; The representative features are input into the original transformation operation to obtain the representative output; Based on the node indices representing the output and the original nodes, the node output is reconstructed as a compression feature.

7. A data processing system based on graph neural network compression, characterized in that, include: The data acquisition module is used to acquire historical datasets and preprocess them to obtain graph structure datasets; The model training module is used to accelerate the training of a pre-built graph neural network model based on a graph structure dataset using a hybrid training method, resulting in a trained graph neural network model, including: The given graph structure is compressed, and the resulting compressed graph structure is applied to the propagation operator of the graph neural network model to obtain the compression-based propagation operator; The node features are compressed, and the resulting compressed features are applied to the transformation operator of the graph neural network model to obtain the compression-based transformation operator; The obtained compression-based propagation and transformation operators are arranged according to actual needs to obtain a preliminarily trained graph neural network model. Using a conventional training mode, the pre-trained graph neural network model is fine-tuned to obtain a well-trained graph neural network model. The data processing module is used to process the target graph structure using a trained graph neural network model to obtain the data processing results.

8. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 6.

9. A computing device, characterized in that, include: One or more processors and a memory, the memory storing one or more programs and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 6.

Images