Artificial intelligence-based graph data processing method and related device
By preprocessing graph data and constructing a graph neural network model, and filtering nodes to generate thumbnail data, the problem of existing technologies failing to effectively consider graph structure and node importance is solved, and efficient dimensionality reduction and analysis of graph data is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2023-04-20
- Publication Date
- 2026-06-02
AI Technical Summary
Existing graph data processing methods fail to effectively consider the structure of the graph and the importance of the nodes when reducing the graph size, resulting in high computational complexity or insufficient information, making them unsuitable for large-scale graph data.
By preprocessing the graph data to obtain node feature vectors, constructing the operational elements of the graph neural network model, filtering nodes to generate thumbnail data, and calculating the full graph vector, the dimensionality of the graph data is continuously reduced by using multiple operational elements, thereby achieving graph data coarsening.
It improves the efficiency and timeliness of graph data processing, effectively reduces the dimensionality of graph data while retaining key information, and enhances the performance of subsequent analysis.
Smart Images
Figure CN116484061B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of artificial intelligence and digital medical technology, and in particular to a graph data processing method, apparatus, electronic device and storage medium based on artificial intelligence. Background Technology
[0002] Graph coarsening is a technique that reduces the size of a graph while preserving its basic properties (for example, if the original graph data has 10,000 nodes, considering data redundancy in real-world applications, we can retain 5,000 or fewer nodes with high information content and representativeness for modeling). Although some graph coarsening techniques have been proposed, traditional coarsening often resembles the design of Convolutional Neural Networks (CNNs) in the field of image recognition, using averaging or summing methods. This simple approach treats the importance of each node equally, without considering the structure of the graph and the importance of each node.
[0003] Currently, to address this issue, pooling methods such as node drop and node clustering have been proposed. Among these, node clustering methods have high computational complexity, making them difficult to apply to even slightly large-scale graph data. On the other hand, node drop methods often select the k highest-scoring nodes to remove nodes with little or no information. However, these methods typically do not consider the graph's topological structure or information about surrounding nodes; they only rank nodes based on their features. Summary of the Invention
[0004] In view of the above, it is necessary to provide an artificial intelligence-based graph data processing method and related equipment to solve the technical problem of how to improve the efficiency of graph data processing. The related equipment includes an artificial intelligence-based graph data processing device, electronic equipment, and storage medium.
[0005] This application provides an artificial intelligence-based graph data processing method, the method comprising:
[0006] The collected graph data is preprocessed to obtain the feature vector of each node in the graph data;
[0007] Multiple computational elements are constructed based on multiple preset graph neural network models, and the multiple computational elements are combined to construct a graph data representation model;
[0008] The feature vector of each node in the graph data is input into the graph data representation model to filter the nodes in the graph data using the multiple operational elements and obtain multiple thumbnail data. The number of thumbnail data is the same as the number of operational elements.
[0009] Calculate the full-image vector corresponding to the image data based on the thumbnail data.
[0010] In some embodiments, the preprocessing of the acquired graph data to obtain the feature vector of each node in the graph data specifically includes:
[0011] Construct an initial feature extraction model, which includes an encoder and a decoder;
[0012] The initial feature extraction model is trained using all the node data in the graph data to obtain a node feature extraction model, which is used for data dimensionality reduction.
[0013] Each node data is sequentially input into the encoder of the node feature extraction model to obtain a feature vector corresponding to each node data, wherein the dimension of the feature vector is smaller than that of the node data.
[0014] In some embodiments, the step of constructing multiple operational elements based on multiple preset graph neural network models and combining the multiple operational elements to construct a graph data representation model specifically includes:
[0015] Each of the preset graph neural network models is used as a single operational element to obtain multiple operational elements;
[0016] Each of the preset graph neural network models corresponds to input data of different dimensions. The multiple operational elements are chained together to form a graph data representation model according to the order of the dimensions of the input data from largest to smallest.
[0017] In some embodiments, the preset graph neural network model in the computational unit is used to receive graph structure data and output the representation vector corresponding to each node in the graph structure data. The step of using the multiple computational units to filter nodes in the graph data and obtain multiple thumbnail data specifically includes:
[0018] The feature vectors of all nodes in the graph data are input into the first operand of the graph data representation model to obtain the representation vector of each node in the graph data.
[0019] The weight of each node is calculated based on the representation vector, and thumbnail data is filtered from all nodes of the graph data based on the weight.
[0020] For each operand in the graph data representation model other than the first operand, the thumbnail data output by the previous operand is input into the operand to obtain the thumbnail data corresponding to each operand.
[0021] In some embodiments, the step of calculating the weight of each node based on the representation vector and filtering thumbnail data from all nodes of the graph data based on the weight specifically includes:
[0022] Calculate the attribute similarity and structural similarity between each node and its neighboring nodes based on the representation vector, and calculate the overall similarity between the node and its neighboring nodes based on the attribute similarity and structural similarity.
[0023] For each node, the average of the overall similarity between the node and all its neighboring nodes is used as the weight of the node, and nodes of equal number to the preset filtering parameters are selected as thumbnail data according to the order of the weights from largest to smallest.
[0024] In some embodiments, calculating the full-image vector corresponding to the image data based on the thumbnail data specifically includes:
[0025] For each of the thumbnail data, the mean of the feature vectors of all nodes in the thumbnail data is calculated as the thumbnail vector;
[0026] Calculate the mean of all the abbreviated vectors to obtain the full graph vector corresponding to the graph data.
[0027] In some embodiments, the method further includes:
[0028] The full graph vector is input into a pre-trained graph data classification model to obtain the category of the full graph vector.
[0029] This application embodiment also provides an artificial intelligence-based graph data processing device, the device comprising:
[0030] The preprocessing unit is used to preprocess the collected graph data to obtain the feature vector of each node in the graph data.
[0031] The construction unit is used to construct multiple operational elements based on multiple preset graph neural network models, and combine the multiple operational elements to construct a graph data representation model;
[0032] A filtering unit is used to input the feature vector of each node in the graph data into the graph data representation model, so as to use the multiple operational elements to filter the nodes in the graph data and obtain multiple thumbnail data, wherein the number of thumbnail data is the same as the number of operational elements;
[0033] A calculation unit is used to calculate the full-image vector corresponding to the image data based on the thumbnail data.
[0034] This application embodiment also provides an electronic device, the electronic device comprising:
[0035] Memory, which stores computer-readable instructions; and
[0036] The processor executes computer-readable instructions stored in the memory to implement the AI-based graph data processing method.
[0037] This application also provides a computer-readable storage medium storing computer-readable instructions, which are executed by a processor in an electronic device to implement the AI-based graph data processing method.
[0038] The aforementioned AI-based graph data processing method preprocesses the collected graph data to obtain the feature vector of each node, unifying the dimension of the data stored in each node and facilitating the subsequent calculation of the full graph vector. Multiple computational units are constructed based on several pre-defined graph neural network models, and these units are combined to build a graph data representation model, providing model support for the subsequent generation of thumbnail data. Subsequently, the multiple computational units in the graph data representation model are used to continuously filter the nodes in the graph data, continuously reducing the dimensionality of the graph data to achieve coarsening, thereby obtaining multiple thumbnail data. Finally, the full graph vector corresponding to the graph data is calculated based on the thumbnail data, thus improving the timeliness of subsequent graph data analysis. Attached Figure Description
[0039] Figure 1 This is a flowchart of a preferred embodiment of an artificial intelligence-based graph data processing method involved in this application.
[0040] Figure 2 This is a schematic diagram of the structure of a graph data representation model with two operands involved in the embodiments of this application.
[0041] Figure 3 This is a schematic diagram of the process of selecting nodes using the computational elements involved in the embodiments of this application.
[0042] Figure 4 This is a functional block diagram of a preferred embodiment of the graph data processing apparatus based on artificial intelligence involved in this application.
[0043] Figure 5 This is a schematic diagram of the structure of an electronic device that is a preferred embodiment of the graph data processing method based on artificial intelligence involved in this application. Detailed Implementation
[0044] To better understand the purpose, features, and advantages of this application, a detailed description of the application is provided below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of this application can be combined with each other. Numerous specific details are set forth in the following description to provide a thorough understanding of this application; the described embodiments are only a part of the embodiments of this application, and not all of them.
[0045] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.
[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.
[0047] This application provides an artificial intelligence-based graph data processing method that can be applied to one or more electronic devices. The electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.
[0048] The electronic device can be any electronic product that can interact with the user, such as a personal computer, tablet computer, smartphone, personal digital assistant (PDA), game console, interactive network television (IPTV), smart wearable device, etc.
[0049] The electronic device may also include network devices and / or user devices. The network devices include, but are not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.
[0050] The networks in which the electronic devices are located include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, and virtual private networks (VPNs).
[0051] like Figure 1The diagram shown is a flowchart of a preferred embodiment of the graph data processing method based on artificial intelligence according to this application. The order of the steps in this flowchart can be changed, and some steps can be omitted, depending on different requirements.
[0052] S10, preprocess the collected graph data to obtain the feature vector of each node in the graph data.
[0053] In an optional embodiment, the preprocessing of the acquired graph data to obtain the feature vector of each node in the graph data specifically includes:
[0054] Construct an initial feature extraction model, which includes an encoder and a decoder;
[0055] The initial feature extraction model is trained using all the node data in the graph data to obtain a node feature extraction model, which is used for data dimensionality reduction.
[0056] Each node data is sequentially input into the encoder of the node feature extraction model to obtain a feature vector corresponding to each node data, wherein the dimension of the feature vector is smaller than that of the node data.
[0057] In this optional embodiment, the graph data refers to pre-collected graph structure data, which consists of nodes and edges. Each node stores a piece of data from the graph data, and the data stored in the node can be in any format, such as a constant, vector, or string. The edges represent the connection relationship between any two nodes.
[0058] For example, the graph data can be a medical knowledge graph, where nodes can be used to represent medical industry-related terms, and edges can be used to represent the similarity between any two medical industry-related terms; the graph data can also be used to represent protein structures, where nodes can represent amino acids in proteins, and edges can represent the connection relationships between amino acids in proteins; the graph data can also be used to represent multiple CT images collected during a patient's physical examination, where each node can represent one of the CT images, and edges can represent the similarity between any two CT images.
[0059] In this optional embodiment, the initial feature extraction model is used to self-encode all node data in the graph data. The initial feature extraction model can be any existing neural network model, such as a graph neural network model or a graph convolutional neural network model, and this application does not limit it. The initial feature extraction model includes an encoder and a decoder. The input of the encoder is the node data in the graph data, and the output of the encoder is the feature vector corresponding to the node data. The input of the decoder is the feature vector, and the output of the decoder is the restored vector corresponding to the feature vector. It should be noted that the dimension of the node data is the same as the dimension of the restored vector, and the dimension of the feature vector is lower than the dimension of the node data.
[0060] In this optional embodiment, the node data can be sequentially input into the initial feature extraction network as training data to obtain the restored vector corresponding to the node data. The restored vector and the node data are then simultaneously input into a preset loss function to obtain the loss value of the initial feature extraction model. The preset loss function can be any existing loss function such as the Euclidean distance function, cosine distance function, or Hamming distance function, and this application does not limit it.
[0061] In this optional embodiment, the initial feature extraction model can be continuously updated using the gradient descent method until the loss value of the initial feature extraction model no longer changes. Then, the update is stopped and a node feature extraction model is obtained, which is used for data dimensionality reduction.
[0062] In this optional embodiment, the node data can be sequentially input into the encoder of the node feature extraction model to obtain the feature vector corresponding to each node data output by the encoder.
[0063] In this way, by training a node feature extraction model that matches the features of the graph data based on all the node data in the graph data, and then using the encoder in the node feature extraction model to extract the features of each node data, the dimensionality of each node data can be reduced, thereby improving the efficiency of subsequent calculation of the full graph vector of the graph data.
[0064] S11, construct multiple operational elements based on multiple preset graph neural network models, and combine the multiple operational elements to construct a graph data representation model.
[0065] In an optional embodiment, the step of constructing multiple operational elements based on multiple preset graph neural network models, and combining the multiple operational elements to construct a graph data representation model, specifically includes:
[0066] Each of the preset graph neural network models is used as a single operational element to obtain multiple operational elements;
[0067] Each of the preset graph neural network models corresponds to input data of different dimensions. The multiple operational elements are chained together to form a graph data representation model according to the order of the dimensions of the input data from largest to smallest.
[0068] In this optional embodiment, the preset graph neural network model is used to receive graph structure data and output a set of vectors corresponding to the graph structure data. The set of vectors includes multiple vectors, each vector representing a feature of a node in the graph structure data received by the graph neural network model. The preset graph neural network model can be an existing graph neural network model such as GNN or GCN, and this application does not limit it. It should be noted that the input data of each preset graph neural network model corresponds to a different dimension, where the dimension refers to the number of nodes in the graph data.
[0069] In this optional embodiment, each of the graph neural networks can be used as a single operational unit, and the multiple operational units are concatenated into a graph data representation model according to the descending order of the dimension of the input data. The input of the first operational unit in the graph data representation model is the feature vector of all nodes in the graph data, and the output of each operational unit in the graph data representation model is a thumbnail data. It should be noted that the input of each operational unit in the graph data representation model other than the first operational unit is the thumbnail data output by the previous operational unit.
[0070] like Figure 2 The diagram shows a schematic of a graph data representation model with two operands, wherein the input graph data is the feature vector of all nodes in the graph data.
[0071] In this way, a graph data representation model is constructed by using multiple pre-defined graph neural network models with different input data dimensions, providing model support for the subsequent generation of graph data representation data.
[0072] S12, input the feature vector of each node in the graph data into the graph data representation model, so as to use the multiple operational elements to filter the nodes in the graph data and obtain multiple thumbnail data, the number of thumbnail data is the same as the number of operational elements.
[0073] In an optional embodiment, the pre-defined graph neural network model in the computational unit is used to receive graph structure data and output the representation vector corresponding to each node in the graph structure data. The step of using the multiple computational units to filter nodes in the graph data and obtain multiple thumbnail data specifically includes:
[0074] The feature vectors of all nodes in the graph data are input into the first operand of the graph data representation model to obtain the representation vector of each node in the graph data.
[0075] The weight of each node is calculated based on the representation vector, and thumbnail data is filtered from all nodes of the graph data based on the weight.
[0076] For each operand in the graph data representation model other than the first operand, the thumbnail data output by the previous operand is input into the operand to obtain the thumbnail data corresponding to each operand.
[0077] In this optional embodiment, the feature vectors of all nodes in the graph data are input into the graph neural network of the first operator in the graph data representation model, and the representation vector of each node in the graph data is obtained by the output of the graph neural network model. The representation vector is used to represent the global information corresponding to the feature vector of each node in the graph data after being processed by the graph neural network.
[0078] In an optional embodiment, the weight of each node is calculated based on the representation vector, and thumbnail data is filtered from all nodes of the graph data based on the weights, specifically including:
[0079] Calculate the attribute similarity and structural similarity between each node and its neighboring nodes based on the representation vector, and calculate the overall similarity between the node and its neighboring nodes based on the attribute similarity and structural similarity.
[0080] For each node, the average of the overall similarity between the node and all its neighboring nodes is used as the weight of the node, and nodes of equal number to the preset filtering parameters are selected as thumbnail data according to the order of the weights from largest to smallest.
[0081] In this optional embodiment, for any node in the graph data, if the features of all nodes in the neighborhood are relatively similar, the neighborhood of that node can be considered as a region with less information change, and the node can be removed to reduce the dimension of the graph data; if the feature similarity between the neighboring nodes of the node and the node is low, it indicates that the differences between the nodes in the neighborhood are large, so the node can be retained, thereby achieving a better graph data coarsening effect with better information preservation.
[0082] In this optional embodiment, the attribute similarity is used to characterize the similarity between two nodes in the graph data. For example, in a medical knowledge graph, the attribute similarity can be used to characterize the similarity between any two medical terms; in protein structures, the attribute similarity can be used to characterize the similarity between any two amino acids. The calculation method of the attribute similarity satisfies the following relationship:
[0083]
[0084] Where A(i,j) represents the attribute similarity between node i and node j in the graph data; A i and A j Let i and j be the representation vectors of node i and node j, respectively.
[0085] For example, when the representation vector of node i is [0,1,2,3] and the representation vector of node j is [1,2,3,4], the similarity of attributes between node i and node j is calculated as follows:
[0086]
[0087] In this optional embodiment, for each node in the initial graph data, the remaining nodes connected to the node are taken as the neighboring nodes of the node, and the neighboring node set of each node is obtained. For example, the neighboring node set of node i can be denoted as N(i), and the neighboring node set of node j can be denoted as N(j).
[0088] In this optional embodiment, for any two nodes, the intersection of the neighboring node sets of the two nodes can be used as candidate nodes, and the sum of the inverse logarithms of the number of candidate nodes is calculated as the structural similarity between the two nodes. The structural similarity is calculated in a way that satisfies the following relationship:
[0089]
[0090] Where i and j represent any two nodes with edges in the graph data; N(i) represents the set of neighboring nodes of node i; N(j) represents the set of neighboring nodes of node j; u represents any node in the intersection of -N(i) and N(j); |N(u)| represents the number of neighboring nodes of node u. The larger the structural similarity B(i,j) between node i and node j, the higher the similarity between node i and node j; conversely, the smaller B(i,j), the lower the similarity between node i and node j.
[0091] For example, when node i and node j have two identical neighboring nodes, u and v, and node u has 3 neighboring nodes and node v has 2 neighboring nodes, then the structural similarity B(i,j) between node i and node j is calculated as follows:
[0092]
[0093] In this optional embodiment, the overall similarity between any two nodes is calculated according to the following formula:
[0094] C(i,j)=α*A(i,j)+β*B(i,j)
[0095] Wherein, C(i,j) represents the overall similarity between node i and node j; A(i,j) represents the attribute similarity between node i and node j; B(i,j) represents the structural similarity between node i and node j; α represents the preset attribute similarity weight, and β represents the preset structural similarity weight. Preferably, in this embodiment, both α and β can be set to 0.5.
[0096] For example, when the attribute similarity between node i and node j is 0.97 and the structural similarity is 0.78, the overall similarity between node i and node j is calculated as follows:
[0097] C(i,j)=0.5*0.97+0.5*0.78=0.875
[0098] In this optional embodiment, for each node, the mean of the overall similarity between the node and all neighboring nodes can be calculated. The larger the mean, the more likely the node belongs to a high-frequency region and needs to be retained. The smaller the mean, the more likely the node belongs to a low-frequency region and can be considered for deletion.
[0099] In this optional embodiment, the dimension of the input data of the next operand can be used as a filtering parameter, and nodes of equal quantity to the filtering parameter can be selected from the plurality of nodes according to the order of the overall similarity mean from large to small as thumbnail data.
[0100] like Figure 3 The diagram shows a flowchart of node selection using the aforementioned operands. The graph data is processed by the graph neural network in the operands to generate a representation vector corresponding to each node. The representation vectors are then filtered by nodes to generate thumbnail data.
[0101] In this optional embodiment, for each of the remaining operands, the thumbnail data output by the previous operand is used as input data, and the input data is input into the graph neural network model in the operand to obtain the thumbnail data corresponding to the input data.
[0102] Ultimately, each computational element in the graph data representation model outputs a thumbnail data, and the dimensions of the thumbnail data output by each computational element are different, that is, the number of nodes contained in each thumbnail data is different.
[0103] In this way, by continuously filtering nodes in the graph data through multiple operational elements in the graph data representation model, the scale of the graph data can be continuously reduced, thereby achieving graph data coarsening, reducing the dimensionality of the graph data, and improving the efficiency of subsequent calculation of the full graph vector.
[0104] S13, Calculate the full-image vector corresponding to the image data based on the thumbnail data.
[0105] In an optional embodiment, calculating the full-image vector corresponding to the image data based on the thumbnail data specifically includes:
[0106] For each of the thumbnail data, the mean of the feature vectors of all nodes in the thumbnail data is calculated as the thumbnail vector;
[0107] Calculate the mean of all the abbreviated vectors to obtain the full graph vector corresponding to the graph data.
[0108] In this optional embodiment, the thumbnail data contains multiple nodes, each node corresponding to a feature vector. For each piece of thumbnail data, the mean of the feature vectors of all nodes in the thumbnail data is calculated as the thumbnail vector. For example, when a piece of thumbnail data contains 3 feature vectors and they are [0,1,2,3], [1,2,3,4], and [2,3,4,5], the method for calculating the thumbnail vector corresponding to the thumbnail data is as follows:
[0109]
[0110] In this optional embodiment, the mean of all thumbnail vectors can be used as the full image vector corresponding to the image data. For example, if there are three thumbnail vectors [1,2,3,4], [2,3,4,5], and [4,5,6,7], the full image vector is calculated as follows:
[0111]
[0112] In an optional embodiment, the method further includes:
[0113] The full graph vector is input into a pre-trained graph data classification model to obtain the category of the full graph vector.
[0114] In this optional embodiment, the full graph vector is used to represent the global information of the graph data, and therefore can be used for graph data classification tasks. The full graph vector can be input into a pre-trained classification model to obtain the category of the graph data. The pre-trained graph data classification model can be an existing classification model such as a multilayer perceptron model, a fully connected neural network model, or a decision tree model, and this application does not limit it to this.
[0115] For example, in a graph data classification task, when the graph data is used to store a medical knowledge graph, and the nodes in the graph data are used to represent medical-related terms corresponding to a certain specialty, the category of the full graph vector can be: surgery, internal medicine, otolaryngology, dentistry, etc.; when the graph data is used to store protein structures, and the nodes in the graph data are used to represent amino acids, the category corresponding to the full graph vector can be the type of protein; when the graph data is used to store medical images, and each node in the graph data is used to store CT images obtained during a patient's physical examination, the category of the full graph vector can be the patient's health level predicted by the graph data classification model, to assist physicians in making deeper diagnoses.
[0116] Thus, using lower-dimensional full-graph vectors to represent global information in graph data can improve the timeliness of graph data processing tasks.
[0117] The aforementioned AI-based graph data processing method preprocesses the collected graph data to obtain the feature vector of each node, unifying the dimension of the data stored in each node and facilitating the subsequent calculation of the full graph vector. Multiple computational units are constructed based on several pre-defined graph neural network models, and these units are combined to build a graph data representation model, providing model support for the subsequent generation of thumbnail data. Subsequently, the multiple computational units in the graph data representation model are used to continuously filter the nodes in the graph data, continuously reducing the dimensionality of the graph data to achieve coarsening, thereby obtaining multiple thumbnail data. Finally, the full graph vector corresponding to the graph data is calculated based on the thumbnail data, thus improving the timeliness of subsequent graph data analysis.
[0118] like Figure 4 The diagram shown is a functional block diagram of a preferred embodiment of the AI-based graph data processing device provided in this application. The AI-based graph data processing device 11 includes a preprocessing unit 110, a construction unit 111, a filtering unit 112, and a calculation unit 113. The module / unit referred to in this application refers to a series of computer program segments that can be executed by the processor 13 and perform a fixed function, and are stored in the memory 12. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.
[0119] The preprocessing unit 110 is used to preprocess the collected graph data to obtain the feature vector of each node in the graph data.
[0120] The construction unit 111 is used to construct multiple operational elements based on multiple preset graph neural network models, and combine the multiple operational elements to construct a graph data representation model.
[0121] The filtering unit 112 is used to input the feature vector of each node in the graph data into the graph data representation model, so as to use the multiple operational elements to filter the nodes in the graph data and obtain multiple thumbnail data, wherein the number of thumbnail data is the same as the number of operational elements.
[0122] The calculation unit 113 is used to calculate the full-image vector corresponding to the image data based on the thumbnail data.
[0123] In an optional embodiment, the preprocessing unit 110 is specifically used for:
[0124] Construct an initial feature extraction model, which includes an encoder and a decoder;
[0125] The initial feature extraction model is trained using all the node data in the graph data to obtain a node feature extraction model, which is used for data dimensionality reduction.
[0126] Each node data is sequentially input into the encoder of the node feature extraction model to obtain a feature vector corresponding to each node data, wherein the dimension of the feature vector is smaller than that of the node data.
[0127] In an optional embodiment, the building unit 111 is specifically used for:
[0128] Each of the preset graph neural network models is used as a single operational element to obtain multiple operational elements;
[0129] Each of the preset graph neural network models corresponds to input data of different dimensions. The multiple operational elements are chained together to form a graph data representation model according to the order of the dimensions of the input data from largest to smallest.
[0130] In an optional embodiment, the filtering unit 112 is specifically used for:
[0131] The feature vectors of all nodes in the graph data are input into the first operand of the graph data representation model to obtain the representation vector of each node in the graph data.
[0132] The weight of each node is calculated based on the representation vector, and thumbnail data is filtered from all nodes of the graph data based on the weight.
[0133] For each operand in the graph data representation model other than the first operand, the thumbnail data output by the previous operand is input into the operand to obtain the thumbnail data corresponding to each operand.
[0134] In an optional embodiment, the filtering unit 112 is further configured to:
[0135] Calculate the attribute similarity and structural similarity between each node and its neighboring nodes based on the representation vector, and calculate the overall similarity between the node and its neighboring nodes based on the attribute similarity and structural similarity.
[0136] For each node, the average of the overall similarity between the node and all its neighboring nodes is used as the weight of the node, and nodes of equal number to the preset filtering parameters are selected as thumbnail data according to the order of the weights from largest to smallest.
[0137] In an optional embodiment, the computing unit 113 is specifically used for:
[0138] For each of the thumbnail data, the mean of the feature vectors of all nodes in the thumbnail data is calculated as the thumbnail vector;
[0139] Calculate the mean of all the abbreviated vectors to obtain the full graph vector corresponding to the graph data.
[0140] In an optional embodiment, the computing unit 113 is further configured to:
[0141] The full graph vector is input into a pre-trained graph data classification model to obtain the category of the full graph vector.
[0142] like Figure 5 The diagram shown is a structural schematic of an electronic device provided in an embodiment of this application. The electronic device 1 includes a memory 12 and a processor 13. The memory 12 stores computer-readable instructions, and the processor 13 executes the computer-readable instructions stored in the memory to implement the AI-based graph data processing method of any of the above embodiments.
[0143] In an alternative embodiment, the electronic device 1 further includes a bus and a computer program stored in memory 12 and executable on processor 13, such as an AI-based graph data processing program.
[0144] Figure 5 Only electronic device 1 with memory 12 and processor 13 is shown. It will be understood by those skilled in the art that... Figure 5 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.
[0145] Combination Figure 1 The memory 12 in electronic device 1 stores multiple computer-readable instructions to implement an artificial intelligence-based graph data processing method, and the processor 13 can execute multiple instructions to achieve:
[0146] The collected graph data is preprocessed to obtain the feature vector of each node in the graph data;
[0147] Multiple computational elements are constructed based on multiple preset graph neural network models, and the multiple computational elements are combined to construct a graph data representation model;
[0148] The feature vector of each node in the graph data is input into the graph data representation model to filter the nodes in the graph data using the multiple operational elements and obtain multiple thumbnail data. The number of thumbnail data is the same as the number of operational elements.
[0149] Calculate the full-image vector corresponding to the image data based on the thumbnail data.
[0150] Specifically, the processor 13's implementation method for the above instructions can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.
[0151] Those skilled in the art will understand that the schematic diagram is merely an example of electronic device 1 and does not constitute a limitation on electronic device 1. Electronic device 1 can be either a bus topology or a star topology. Electronic device 1 may also include more or fewer other hardware or software than shown in the diagram, or different component arrangements. For example, electronic device 1 may also include input / output devices, network access devices, etc.
[0152] It should be noted that electronic device 1 is only an example. Other existing or future electronic products that are suitable for this application should also be included within the scope of protection of this application and are incorporated herein by reference.
[0153] The memory 12 includes at least one type of readable storage medium, which can be non-volatile or volatile. The readable storage medium includes flash memory, portable hard drives, multimedia cards, card-type memory (e.g., SD or DX memory), magnetic storage, magnetic disks, optical disks, etc. In some embodiments, the memory 12 can be an internal storage unit of the electronic device 1, such as a portable hard drive of the electronic device 1. In other embodiments, the memory 12 can be an external storage device of the electronic device 1, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc., equipped on the electronic device 1. Furthermore, the memory 12 can include both internal storage units and external storage devices of the electronic device 1. The memory 12 can be used not only to store application software and various types of data installed on the electronic device 1, such as the code of graph data processing programs based on artificial intelligence, but also to temporarily store data that has been output or will be output.
[0154] In some embodiments, the processor 13 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 13 is the control unit of the electronic device 1, connecting various components of the electronic device 1 through various interfaces and lines. It executes programs or modules stored in the memory 12 (e.g., executing AI-based graph data processing programs) and calls data stored in the memory 12 to perform various functions and process data in the electronic device 1.
[0155] The processor 13 executes the operating system of the electronic device 1 and various installed applications. The processor 13 executes these applications to implement the steps described in the various embodiments of the artificial intelligence-based graph data processing methods, for example... Figure 1 The steps are shown.
[0156] For example, the computer program may be divided into one or more modules / units, which are stored in memory 12 and executed by processor 13 to complete this application. The one or more modules / units may be a series of computer-readable instruction segments capable of performing specific functions, which describe the execution process of the computer program in electronic device 1. For example, the computer program may be divided into a preprocessing unit 110, a construction unit 111, a filtering unit 112, and a calculation unit 113.
[0157] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute portions of the artificial intelligence-based graph data processing method described in the various embodiments of this application.
[0158] If the modules / units integrated in electronic device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware devices. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above.
[0159] The computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory, and other memory.
[0160] Furthermore, the computer-readable storage medium may primarily include a stored program area and a stored data area, wherein the stored program area may store the operating system, an application program required for at least one function, etc.; and the stored data area may store data created based on the use of blockchain nodes, etc.
[0161] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, in... Figure 5 The symbol is represented by only one arrow, but this does not indicate that there is only one bus or one type of bus. The bus is configured to enable communication between the memory 12 and at least one processor 13, etc.
[0162] This application also provides a computer-readable storage medium (not shown), which stores computer-readable instructions. These computer-readable instructions are executed by a processor in an electronic device to implement the AI-based graph data processing method described in any of the above embodiments.
[0163] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices described in the specification may also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.
[0164] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.
Claims
1. A graph data processing method based on artificial intelligence, characterized in that, The method includes: The collected graph data is preprocessed to obtain the feature vector of each node in the graph data; the graph data is used to represent multiple CT images collected during the patient's physical examination, each node in the graph data is used to represent any one of the multiple CT images, and the edges in the graph data are used to represent the similarity between any two CT images; Multiple computational units are constructed based on multiple preset graph neural network models, and the multiple computational units are combined to construct a graph data representation model. This includes: treating each preset graph neural network model as a computational unit to obtain multiple computational units; each preset graph neural network model corresponds to input data of different dimensions; the multiple computational units are concatenated into a graph data representation model according to the order of the dimensions of the input data from largest to smallest; the input of the first computational unit in the graph data representation model is the feature vector of all nodes in the graph data, and the input of the remaining computational units in the graph data representation model (excluding the first computational unit) is the thumbnail data output by the previous computational unit. The feature vector of each node in the graph data is input into the graph data representation model to filter the nodes in the graph data using the multiple operational elements and obtain multiple thumbnail data. The number of thumbnail data is the same as the number of operational elements. The full graph vector corresponding to the graph data is calculated based on the thumbnail data. The full graph vector represents the global information of the graph data and is used for graph data classification tasks.
2. The graph data processing method based on artificial intelligence as described in claim 1, characterized in that, The preprocessing of the collected graph data to obtain the feature vector of each node in the graph data specifically includes: Construct an initial feature extraction model, which includes an encoder and a decoder; The initial feature extraction model is trained using all the node data in the graph data to obtain a node feature extraction model, which is used for data dimensionality reduction. Each node data is sequentially input into the encoder of the node feature extraction model to obtain a feature vector corresponding to each node data, wherein the dimension of the feature vector is smaller than that of the node data.
3. The graph data processing method based on artificial intelligence as described in claim 1, characterized in that, The pre-defined graph neural network model in the computational unit is used to receive graph structure data and output the representation vector corresponding to each node in the graph structure data. The step of using the multiple computational units to filter nodes in the graph data and obtain multiple thumbnail data specifically includes: The feature vectors of all nodes in the graph data are input into the first operand of the graph data representation model to obtain the representation vector of each node in the graph data. The weight of each node is calculated based on the representation vector, and thumbnail data is filtered from all nodes of the graph data based on the weight. For each operand in the graph data representation model other than the first operand, the thumbnail data output by the previous operand is input into the operand to obtain the thumbnail data corresponding to each operand.
4. The graph data processing method based on artificial intelligence as described in claim 3, characterized in that, The step of calculating the weight of each node based on the representation vector, and filtering thumbnail data from all nodes of the graph data based on the weights, specifically includes: Calculate the attribute similarity and structural similarity between each node and its neighboring nodes based on the representation vector, and calculate the overall similarity between the node and its neighboring nodes based on the attribute similarity and structural similarity. For each node, the average of the overall similarity between the node and all its neighboring nodes is used as the weight of the node, and nodes of equal number to the preset filtering parameters are selected as thumbnail data according to the order of the weights from largest to smallest.
5. The graph data processing method based on artificial intelligence as described in claim 1, characterized in that, The step of calculating the full-image vector corresponding to the image data based on the thumbnail data specifically includes: For each of the thumbnail data, the mean of the feature vectors of all nodes in the thumbnail data is calculated as the thumbnail vector; Calculate the mean of all the abbreviated vectors to obtain the full graph vector corresponding to the graph data.
6. The graph data processing method based on artificial intelligence as described in claim 5, characterized in that, The method further includes: The full graph vector is input into a pre-trained graph data classification model to obtain the category of the full graph vector.
7. A graph data processing device based on artificial intelligence, characterized in that, The device includes: The preprocessing unit is used to preprocess the acquired graph data to obtain the feature vector of each node in the graph data; the graph data is used to represent multiple CT images acquired by the patient during the physical examination, each node in the graph data is used to represent any one of the multiple CT images, and the edges in the graph data are used to represent the similarity between any two CT images. A construction unit is used to construct multiple operational elements based on multiple preset graph neural network models, and to combine the multiple operational elements to construct a graph data representation model. The unit includes: treating each preset graph neural network model as a single operational element to obtain multiple operational elements; each preset graph neural network model corresponds to input data of different dimensions; and concatenating the multiple operational elements into a graph data representation model according to the order of the dimensions of the input data from largest to smallest; the input of the first operational element in the graph data representation model is the feature vector of all nodes in the graph data, and the input of the remaining operational elements in the graph data representation model (excluding the first operational element) is the thumbnail data output by the previous operational element. A filtering unit is used to input the feature vector of each node in the graph data into the graph data representation model, so as to use the multiple operational elements to filter the nodes in the graph data and obtain multiple thumbnail data, wherein the number of thumbnail data is the same as the number of operational elements; The computing unit is used to calculate the full graph vector corresponding to the graph data based on the thumbnail data. The full graph vector represents the global information of the graph data and is used for graph data classification tasks.
8. An electronic device, characterized in that, The electronic device includes: Memory, which stores computer-readable instructions; and The processor executes computer-readable instructions stored in the memory to implement the AI-based graph data processing method as described in any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which are executed by a processor in an electronic device to implement the AI-based graph data processing method as described in any one of claims 1 to 6.