Information prediction method, device, storage medium and computer equipment

Abstract

The present application discloses an information prediction method, apparatus, storage medium and computer equipment; the present application is related to the field of machine learning of artificial intelligence, as well as fields such as data storage and reading. The present application can obtain structural representation data of a target graph structure, the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure; based on the structural representation data, multiple node data of each node of the target graph structure are generated; the multiple node data of each node are aggregated to obtain mapping data of the target graph structure; based on the mapping data, the attribute information of the target graph structure is predicted to obtain a prediction result; the present application can effectively save computer equipment resources.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and storage, and specifically to an information prediction method, apparatus, storage medium, and computer equipment. Background Art

[0002] A lot of data in daily life can be represented using graph structures. A graph structure can contain multiple nodes and multiple edges, and edges can represent the relationship between nodes. With the advancement of technology, graph neural networks can be used to analyze and predict graph structures, as well as the nodes, edges, and other contents within the graph structure. Existing graph neural networks include graph convolutional networks (GCN) and APPNP (a type of graph neural network).

[0003] During the research and practice of the existing technology, the inventors of this application discovered that due to the network characteristics of the existing graph neural network, the existing graph neural network will perform multiple data processing with small differences on the graph structure data, thereby occupying a large amount of storage resources and computing resources of the computer equipment. Summary of the Invention

[0004] The embodiments of the present application provide an information prediction method, apparatus, storage medium, and computer equipment, which can effectively save computer equipment resources.

[0005] The present invention provides an information prediction method, including:

[0006] Acquire structural representation data of a target graph structure, wherein the target graph structure includes a plurality of nodes, and the structural representation data represents the plurality of nodes of the target graph structure;

[0007] generating a plurality of node data for each node of the target graph structure according to the structure representation data;

[0008] Aggregating multiple node data of each node to obtain mapping data of the target graph structure;

[0009] Based on the mapping data, attribute information of the target graph structure is predicted to obtain a prediction result.

[0010] Accordingly, the present application provides an information prediction device, comprising:

[0011] An acquisition module, configured to acquire structural representation data of a target graph structure, wherein the target graph structure includes a plurality of nodes, and the structural representation data represents the plurality of nodes of the target graph structure;

[0012] A generating module, configured to generate a plurality of node data for each node of the target graph structure according to the structure representation data;

[0013] An aggregation module, configured to aggregate multiple node data of each node to obtain mapping data of the target graph structure;

[0014] The prediction module is used to predict the attribute information of the target graph structure based on the mapping data to obtain a prediction result.

[0015] In some embodiments, the aggregation module includes an aggregation submodule and a merging submodule, wherein:

[0016] Aggregation submodule, used to aggregate multiple node data of each node to obtain aggregated data of each node;

[0017] The merging submodule is used to merge the aggregated data of all nodes to obtain the mapping data of the target graph structure.

[0018] In some embodiments, the aggregation submodule is specifically configured to:

[0019] All node data of the target node are fused and calculated to obtain the aggregated data of the target node.

[0020] In some embodiments, the aggregation submodule is specifically configured to:

[0021] Get the aggregation weight information corresponding to each node data of the target node;

[0022] A fusion calculation is performed based on each node data of the target node and the aggregation weight information corresponding to each node data to obtain the aggregated data of the target node.

[0023] In some embodiments, the target graph structure further includes a plurality of edges, the edges connecting nodes having an association relationship, the structure representation data includes node representation data of each node, and the generation module includes a determination submodule and a generation submodule, wherein,

[0024] A determination submodule, configured to determine at least one node on the same edge as the target node;

[0025] The generating submodule is configured to generate a plurality of node data of the target node according to the node representation data of the target node and the node representation data of the nodes on the same edge.

[0026] In some embodiments, the target graph structure includes a target number of nodes, and the generation submodule is specifically configured to:

[0027] Performing a fusion calculation on the node representation data of the target node and the node representation data of the nodes on the same edge to obtain node data of the target node;

[0028] A step of cyclically determining a new same-side node located on the same side as the same-side node, fusing the node representation data of the target node, the node representation data of the same-side node, and the node representation data of the new same-side node to obtain node data of the target node, until the new same-side node is the target number of same-side nodes of the target node, thereby obtaining multiple node data of the target node.

[0029] In some embodiments, the prediction module is specifically configured to:

[0030] The mapping data is input into the trained prediction model to predict the attribute information of the target graph structure to obtain a prediction result.

[0031] In some embodiments, the prediction device further comprises:

[0032] a determination module, configured to determine a plurality of sample node data of each sample node in the sample graph structure according to the graph sample representation data of the sample graph structure;

[0033] A sampling module, configured to sample a plurality of sample nodes of the sample graph structure and determine a plurality of target sample nodes required for training a prediction model;

[0034] The training module is used to train the prediction model based on multiple sample node data of each target sample node to obtain a trained prediction model.

[0035] In some embodiments, the prediction device further comprises:

[0036] A receiving module, configured to receive graph sample representation data of a sample graph structure sent by a distributed storage system;

[0037] At this point, determine the module specifically for:

[0038] Determining, according to the graph sample representation data, a plurality of sample node data of a first sample node in the sample graph structure;

[0039] The multiple sample node data of the first sample node are transmitted back to the distributed storage system, thereby obtaining a distributed storage system including the multiple sample node data of each sample node in the sample graph structure.

[0040] The data acquisition module is used to acquire multiple sample node data of each target sample node from the distributed storage system.

[0041] In some embodiments, the training module includes an acquisition submodule, an aggregation submodule, and a training submodule, wherein:

[0042] An acquisition submodule, configured to acquire model parameters of the prediction model from a parameter storage device;

[0043] an aggregation submodule, configured to aggregate the multiple sample node data of each target sample node to obtain first sample mapping data of the sample graph structure;

[0044] A training submodule is used to train the prediction model based on the first sample mapping data and the model parameters to obtain a trained prediction model.

[0045] In some embodiments, the aggregation submodule is specifically configured to:

[0046] Performing a first aggregation process on all sample node data of each target sample node to obtain aggregation weight data of each target sample node;

[0047] Performing a second aggregation process based on the aggregation weight data of each target sample node and all sample node data to obtain sample aggregation data of each target sample node;

[0048] The sample aggregation data of all target sample nodes are merged to obtain first sample mapping data of the sample graph structure.

[0049] In some embodiments, the training submodule is specifically configured to:

[0050] Training the prediction model based on the first sample mapping data and the model parameters to obtain initial updated model parameters of the prediction model;

[0051] The initial updated model parameters are sent to the parameter storage device so that the parameter storage device synchronously updates the model parameters of the prediction model based on the initial updated model parameters received from multiple training devices to obtain updated model parameters, so as to obtain a trained prediction model including the updated model parameters.

[0052] Correspondingly, an embodiment of the present application further provides a storage medium, which stores a computer program. The computer program is suitable for loading by a processor to execute any information prediction method provided in the embodiment of the present application.

[0053] Correspondingly, an embodiment of the present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, any one of the information prediction methods provided in the embodiment of the present application is implemented.

[0054] The present application can obtain structural representation data of a target graph structure, where the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure; based on the structural representation data, multiple node data of each node of the target graph structure are generated; the multiple node data of each node are aggregated to obtain mapping data of the target graph structure; based on the mapping data, the attribute information of the target graph structure is predicted to obtain a prediction result.

[0055] After obtaining the structural representation data of the target graph structure, the present application can generate multiple node data for each node of the target graph structure based on the structural representation data, thereby obtaining the mapping data of the target graph structure, and predicting the attribute information of the target graph structure based on the mapping data. Compared with the prior art, when analyzing and predicting the relevant information of the target graph structure, it is necessary to perform the step of generating multiple node data of the node multiple times. Before predicting the attribute information of the target graph structure, the present application generates multiple node data for each node of the target graph structure, which significantly reduces the resource consumption of the computer equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 This is a schematic diagram of a scenario of an information prediction system provided by an embodiment of the present application;

[0058] Figure 2 Schematic diagram of the information prediction method provided in the embodiment of the present application;

[0059] Figure 3 This is another flow chart of the information prediction method provided in an embodiment of the present application;

[0060] Figure 4 This is a schematic diagram of the overall architecture of the information prediction method provided in the embodiment of the present application;

[0061] Figure 5 Schematic diagram of the implementation structure of the information prediction method provided in the embodiment of the present application;

[0062] Figure 6 This is another schematic diagram of the implementation structure of the information prediction method provided in the embodiment of the present application;

[0063] Figure 7 This is another schematic diagram of the implementation structure of the information prediction method provided in the embodiment of the present application;

[0064] Figure 8 Schematic diagram of the training process of the information prediction method provided in the embodiment of the present application;

[0065] Figure 9 is a schematic structural diagram of an information prediction device provided in an embodiment of the present application;

[0066] Figure 10 It is a structural diagram of the computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described in this application are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of this application.

[0068] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also involves studying the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0069] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is at the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0070] The information prediction method in this application can be applied to the field of neural networks in machine learning. For example, the mapping data of the target graph structure is input into the graph neural network to obtain the prediction results of the attribute information of the target graph structure, etc., which will be specifically introduced in the embodiments of this application.

[0071] The information prediction method of the present application can be integrated into an information prediction system, which can be integrated into one or more computer devices, which can include terminals or servers, etc., wherein the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited to these. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited in this application.

[0072] See also Figure 1 The information prediction system may include an information prediction device, wherein the information prediction device can obtain structural representation data of a target graph structure, the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure; based on the structural representation data, generate multiple node data of each node of the target graph structure; aggregate the multiple node data of each node to obtain mapping data of the target graph structure; based on the mapping data, predict the attribute information of the target graph structure to obtain a prediction result.

[0073] It should be noted that Figure 1 The scenario diagram of the information prediction system shown is only an example. The information prediction system and scenario described in the embodiment of the present application are intended to more clearly illustrate the technical solution of the embodiment of the present application, and do not constitute a limitation on the technical solution provided by the embodiment of the present application. Ordinary technicians in this field can know that with the evolution of information prediction devices and the emergence of new business scenarios, the technical solution provided by the embodiment of the present application is also applicable to similar technical problems.

[0074] In this embodiment, the information prediction method is described in detail. The information prediction method can be integrated in a terminal or a server, such as Figure 2 As shown, Figure 2 : is a flow chart of the information prediction method provided in the embodiment of the present application. The information prediction method may include:

[0075] 101. Obtain structural representation data of a target graph structure, where the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure.

[0076] The target graph structure can include multiple nodes and multiple edges. An edge can represent the relationship between at least two nodes it connects. Multiple entities that have a relationship with each other within a specific range can be represented by the target graph structure. The knowledge graph that is increasingly widely used today is a graph structure. Therefore, the target graph structure can represent data in multiple fields, such as biochemistry (such as molecular structure), transportation (such as route maps), communications (such as base station distribution), etc.

[0077] Specifically, for example, a social network can be represented by a target graph structure, where the nodes can be users and the edges can be the relationships between users. For another example, e-shopping can be represented by a target graph structure, where the nodes can be products and buyers and the edges can be the relationships between products and buyers. For another example, financial transactions can be represented by a target graph structure, where the nodes can be lenders and borrowers and the edges can be the relationships between borrowers and payees, and so on.

[0078] Among them, the structural representation data may include data representing the target graph structure, and the structural representation data may represent the nodes and edges in the target graph structure. The structural representation data may include multiple types. For example, the structural representation data may be a data set, and the data set may include a node set and an edge set. The node set includes all nodes of the target graph structure, and the edge set includes all edges of the target graph structure. For another example, the structural representation data may include an adjacency matrix, an adjacency linked list, etc., and a Laplace matrix may be obtained based on the adjacency matrix and degree matrix of the target graph structure, and the Laplace matrix may be used as the structural representation data of the target graph structure, and so on.

[0079] Specifically, there are many ways to obtain the structural representation data of the target graph structure, such as directly obtaining it from a local computer device or other computer devices (such as a server, etc.), or integrating or calculating the structural representation data from the original graph data (such as a schematic diagram of the target graph structure) obtained based on a local computer device or other computer devices, and so on.

[0080] For example, the structural representation data representing the target graph structure 1: matrix M can be obtained from the server.

[0081] 102. Generate multiple node data for each node of the target graph structure according to the structural representation data.

[0082] Among them, node data can be a summary expression of the information contained in the node. The information contained in the node can include the node itself, and can also include the association relationship of the node. The association relationship of the node can include the edges connecting the node with other nodes, and can also include other nodes connected to the node. Other nodes connected to the node can include other nodes directly connected to the node through edges, and can also include other nodes indirectly connected to the node through multiple edges, such as neighbor nodes directly connected to the node, second-generation neighbor nodes directly connected to the neighbor nodes of the node (second-generation neighbor nodes are indirectly connected to the node through two edges), and so on.

[0083] The structural representation data can represent the nodes and edges of the target graph structure. The node data may include a summary expression of at least one of the information such as the node, the edges connecting the node with other nodes, other nodes directly connected to the node through edges, and other nodes indirectly connected to the node through multiple edges. Therefore, multiple node data of each node can be generated through the structural representation data. For example, elements in the structural representation data (elements such as nodes and edges) can be randomly sampled, and the sampling results can be fused to obtain the node data of the node.

[0084] For example, based on the matrix M, multiple node data of each node of the target graph structure 1 can be generated, and the target graph structure 1 includes 5 nodes.

[0085] In some embodiments, the target graph structure further includes multiple edges, the edges connecting nodes having an association relationship, the structural representation data includes node representation data of each node, and the step of "generating multiple node data for each node of the target graph structure based on the structural representation data" includes:

[0086] At least one node on the same edge as the target node is determined; and multiple node data of the target node are generated according to the node representation data of the target node and the node representation data of the node on the same edge.

[0087] The node representation data may include data representing the node. The node representation data may be in the form of an array, a vector, a matrix, a string, etc., which may be determined specifically according to the nature of the node and is not limited here.

[0088] Among them, the same-edge nodes may include nodes that have an association relationship with the target node, that is, nodes connected by edges. Here, the same-edge nodes and target nodes are relative concepts. The target node can be a randomly determined node in the target graph structure, and the same-edge nodes connected to the target node through edges can be determined. When the target node changes, the same-edge nodes will also change.

[0089] In some embodiments, the process of determining the same-side nodes may include: determining the target edge where the target node is located, determining other nodes other than the target node among the nodes connected to the target edge, the other nodes being the same-side nodes of the target node, and then obtaining multiple node representation data of the target node based on the node representation data of the target node and the same-side node. For example, the number of same-side nodes can be multiple, and each same-side node can be grouped with the target node, and the same-side nodes and target nodes in the group can be fused to obtain multiple node data of the target node.

[0090] For example, determine two nodes on the same edge as the target node: node 1: node 2 and node 3, and then generate multiple node data of node 1 based on the node representation data of node 1, node 2 and node 3.

[0091] In some embodiments, the step of “generating multiple node data of the target node based on the node representation data of the target node and the node representation data of the nodes on the same edge” includes:

[0092] Perform fusion calculation on the node representation data of the target node and the node representation data of the nodes on the same edge to obtain a node data of the target node;

[0093] A new same-edge node located on the same edge as the same-edge node is determined cyclically, and node representation data of the target node, the node representation data of the same-edge node, and the node representation data of the new same-edge node are fused and calculated to obtain a node data of the target node, until the new same-edge node meets the preset termination condition, thereby obtaining multiple node data of the target node.

[0094] Specifically, the fusion calculation of the obtained node representation data may include multiple methods, such as splicing or merging all node representation data to obtain node data, or performing similar processing such as calculating expectation, variance, etc. on all node representation data.

[0095] The preset termination condition can be flexibly determined according to actual application requirements. For example, the preset termination condition may be that the node representation data of the new same-edge node is zero, or that the new same-edge node is only located in one edge, etc.

[0096] The target graph structure may include a target number of nodes, and the preset termination condition may be: the new same-edge node is the target number of same-edge nodes of the target node

[0097] In some embodiments, multiple levels of neighbor nodes of the target node can be determined, and each level of neighbor nodes is a same-side node connected to the previous level of neighbor nodes through an edge, until a final level of neighbor nodes with no new same-side nodes is obtained. For example, a same-side node directly connected to the target node through an edge is a first-level neighbor node, a same-side node connected to the first-level neighbor node through an edge is a second-level neighbor node, and a same-side node connected to the second-level neighbor node through an edge is a third-level neighbor node.

[0098] The node data can be obtained by fusing and calculating the target node with neighbor nodes at each level in sequence. For example, the node data may include: the node representation data of the target node, the node representation data of the target node and the first-level neighbor node are fused and calculated, the node representation data of the target node, the first-level neighbor node and the second-level neighbor node are fused and calculated, and the node representation data of the target node, the first-level neighbor node, the second-level neighbor node and the third-level neighbor node are fused and calculated.

[0099] 103. Aggregate multiple node data of each node to obtain mapping data of the target graph structure.

[0100] Among them, the mapping data may include intermediate data obtained by processing the target graph structure during the information prediction process. Specifically, the process of aggregating node data to obtain mapping data may include multiple methods. For example, hierarchical extraction can be performed from multiple node data of all nodes to obtain multiple target node data, and then the target node data can be fused to obtain mapping data of the target graph structure.

[0101] For example, multiple node data of each node may be aggregated to obtain mapping data 1 of target graph structure 1.

[0102] In some embodiments, the step of “aggregating multiple node data of each node to obtain mapping data of the target graph structure” may include:

[0103] Aggregate multiple node data of each node to obtain aggregate data of each node; merge the aggregate data of all nodes to obtain mapping data of the target graph structure.

[0104] In some embodiments, a node includes multiple node data, and the multiple node data of a node can be aggregated to obtain the aggregated data of a node, where the aggregation processing can include multiple methods, such as calculating the expectation, variance, maximum value, median, etc. of all node data of a node.

[0105] Then, the aggregated data of all nodes are merged to obtain mapping data. The merging process may include various methods, such as connecting all the aggregated data, merging all the aggregated data into a matrix of a preset size, and so on.

[0106] For example, multiple node data of each node can be aggregated to obtain aggregate data of nodes 1 to 5 respectively; the aggregate data of all nodes can be merged to obtain mapping data 1 of target graph structure 1.

[0107] In some embodiments, the step of “aggregating multiple node data of each node to obtain aggregated data of each node” includes:

[0108] All node data of the target node are fused and calculated to obtain the aggregated data of the target node.

[0109] Among them, the fusion calculation can include calculating the expectation, variance, maximum value, minimum value, median, etc. For example, if the target node includes 10 node data, then these 10 node data can be sorted, the maximum value among them can be determined, and the maximum value can be used as the aggregated data of the target node.

[0110] In some embodiments, the step of “aggregating multiple node data of each node to obtain aggregated data of each node” includes:

[0111] Get the aggregation weight information corresponding to each node data of the target node;

[0112] Based on each node data of the target node and the aggregation weight information corresponding to each node data, a fusion calculation is performed to obtain the aggregated data of the target node.

[0113] The aggregation weight information may include information used in the aggregation process. The aggregation weight information may be obtained from a local computer device or other computer devices, or may be obtained through methods such as table lookup and model output.

[0114] By performing fusion calculation based on the node data and its aggregation weight information, the aggregation data of the target node can be obtained. This process can include multiple implementation methods, such as weighted operations (sum, average, etc.), or the product of node data and aggregation weight information can be sorted, and the maximum value obtained can be used as the aggregation data of the target node, etc.

[0115] 104. Based on the mapping data, predict the attribute information of the target graph structure to obtain a prediction result.

[0116] The attribute information may include information related to the target graph structure, such as the target graph structure, nodes and edges within the target graph structure, and the like.

[0117] Among them, the prediction results may include corresponding to the attribute information, such as the prediction results may include the target graph structure, the nodes in the target graph structure, the type of edges, whether there are nodes at specific positions in the target graph structure, etc. The graph structure can represent data in different fields, and the prediction results obtained by performing attribute prediction on graph structures in different fields can represent different information. For example, in the field of biochemistry, when predicting the graph structure representing the molecular structure, the prediction results obtained may include whether there are atoms at specific positions in the molecular structure, the types of atoms at specific positions, whether there are bonds between two specific atoms, the types of bonds between two specific atoms, the type of molecular structure, etc. The characteristics of the molecular structure can be more clearly defined based on the obtained prediction results.

[0118] For example, in financial risk control scenarios, prediction results can include the node type at a specific location in the target graph structure, which can be used to determine whether the person or organization corresponding to that node is legitimate. Prediction results can also include the edge type between specific nodes, which can be used to determine the financial operations between at least two people or organizations corresponding to those nodes. These operations include cashing out, gambling, and multiple loans. Based on the prediction results, intervention can be taken against abnormal individuals or organizations.

[0119] In some embodiments, the step of “predicting attribute information of the target graph structure based on the mapping data to obtain a prediction result” may include:

[0120] The mapping data is input into the trained prediction model to predict the attribute information of the target graph structure and obtain the prediction results.

[0121] This application can predict the attribute information of the target graph structure through a trained prediction model constructed and trained based on the principles of graph neural networks. For example, the mapping data can be input into the trained prediction model, and the trained prediction model can output the prediction results.

[0122] For example, the attribute information of the target graph structure is predicted by the trained model 1, and the mapping data of the target graph structure is input into the trained model 1 to obtain the prediction result.

[0123] In some embodiments, the information prediction method further includes:

[0124] Determining multiple sample node data of each sample node in the sample graph structure according to the graph sample representation data of the sample graph structure;

[0125] Sampling multiple sample nodes of the sample graph structure to determine multiple target sample nodes required for training the prediction model;

[0126] Based on multiple sample node data of each target sample node, the prediction model is trained to obtain a trained prediction model.

[0127] Specifically, after the prediction model is built, it needs to be trained to obtain a trained prediction model. The process of prediction model construction and training can be carried out based on the actual application process and relevant information in the field of graph neural networks. It can be flexibly grasped. For example, the sample nodes of the sample graph structure can be sampled to obtain the target sample nodes required for training, and then multiple sample node data of the target sample nodes can be used for model training to obtain a trained prediction model.

[0128] During the model training process, a loss function can be selected to calculate the loss value, and the model parameters of the prediction model can be updated according to the loss value. Specifically, the loss function can include the cross entropy loss function, the Hinge loss function (a loss function), the softmax function (a loss function), etc. The model parameter update can be based on the stochastic gradient descent (SGD), the stochastic gradient descent using momentum (Momentum SGD), the adaptive gradient (AdaGrad), and other algorithms.

[0129] In some embodiments, the information prediction method further includes:

[0130] Receiving graph sample representation data of a sample graph structure sent by a distributed storage system;

[0131] At this time, the step of "determining multiple sample node data for each sample node in the sample graph structure based on the graph sample representation data of the sample graph structure" may include:

[0132] Determining, according to the graph sample representation data, a plurality of sample node data of a first sample node in the sample graph structure;

[0133] The multiple sample node data of the first sample node are transmitted back to the distributed storage system, thereby obtaining a distributed storage system including the multiple sample node data of each sample node in the sample graph structure.

[0134] Before the step of "training the prediction model based on the multiple sample node data of each target sample node to obtain a trained prediction model", the following steps are also included:

[0135] A plurality of sample node data of each target sample node is obtained from the distributed storage system.

[0136] In some embodiments, relevant data of the sample graph structure can be stored in a distributed storage system to better manage data and achieve more efficient and rapid model training. Graph sample representation data can be obtained from the distributed storage system, and the obtained sample node data of the sample node can be sent to the distributed storage system. Multiple sample node data of the target sample node can be obtained from the distributed storage system.

[0137] The sample graph result includes multiple sample nodes. In order to speed up efficiency, multiple sample node data of the sample node can be determined separately by multiple computer devices. After each computer device generates sample node data, it can be transmitted to the distributed storage system for storage. Finally, the distributed storage system can store all sample node data of all sample nodes in the sample graph structure. For example, the computer device can generate multiple sample node data of the first sample node and transmit these multiple sample node data back to the distributed storage system.

[0138] In some embodiments, the step of “training a prediction model based on multiple sample node data of each target sample node to obtain a trained prediction model” includes:

[0139] Obtain model parameters of the prediction model from a parameter storage device;

[0140] Aggregating multiple sample node data of each target sample node to obtain first sample mapping data of a sample graph structure;

[0141] The prediction model is trained based on the first sample mapping data and the model parameters to obtain a trained prediction model.

[0142] In order to speed up the efficiency of model training, the model parameters that will be continuously updated during the training process can be saved through a parameter storage device. At this time, the process of model training may include obtaining the model parameters of the prediction model from the parameter storage device, obtaining a sample mapping data of the sample graph structure through aggregation processing, that is, the first sample mapping data, and then performing model training to obtain a trained prediction model.

[0143] In some embodiments, the step of “aggregating the multiple sample node data of each target sample node to obtain first sample mapping data of the sample graph structure” includes:

[0144] Performing a first aggregation process on all sample node data of each target sample node to obtain aggregation weight data of each target sample node;

[0145] Performing a second aggregation process based on the aggregation weight data of each target sample node and all sample node data to obtain sample aggregation data of each target sample node;

[0146] The sample aggregation data of all target sample nodes are merged to obtain the first sample mapping data of the sample graph structure.

[0147] Specifically, the aggregation weight data of the target sample node can be obtained based on all sample node data of the target sample node. The first aggregation processing may include at least one of the following methods: calculating the average value, sum and difference product, maximum value, minimum value, median value, etc. of multiple sample node data. Then, the sample aggregation data of each target sample node can be obtained based on the aggregation weight data of each target sample order and multiple sample node data. The second aggregation processing may include weighted operations, such as weighted summation, weighted average, etc. Finally, the sample aggregation data of all target sample nodes can be merged to obtain a sample mapping data of the sample graph structure.

[0148] In some embodiments, the step of “training the prediction model based on the first sample mapping data and the model parameters to obtain a trained prediction model” includes:

[0149] Training the prediction model based on the first sample mapping data and the model parameters to obtain initial updated model parameters of the prediction model;

[0150] Sending initial updated model parameters to the parameter storage device so that the parameter storage device synchronously updates the model parameters of the prediction model based on the initial updated model parameters received from multiple training devices to obtain updated model parameters, so as to obtain a trained prediction model including the updated model parameters.

[0151] During training, the first sample mapping data can be input into the prediction model, and calculations can be performed based on the architecture and model parameters of the prediction model to obtain the loss value of this training and update the parameters based on the loss value to obtain the initial updated model parameters of the prediction model.

[0152] In order to improve the efficiency of model training, the prediction model can be trained multiple times using multiple sample mapping data through multiple computer devices. Each computer device can obtain a device updated model parameter after performing a model training, such as the initial updated model parameter obtained by a computer device performing the first model training. Therefore, the parameter storage device can receive the initial updated model parameters from multiple computer devices, and then perform fusion processing on these parameters. The fusion processing can be, for example, averaging, taking the median, etc., to obtain an updated model parameter of the prediction model, and the updated model parameter is sent to each computer device, so that each computer device performs model training again based on the updated model parameter, and repeats the above process until the model training is completed. The end condition of model training can be flexibly set, such as when the number of training times reaches a preset number, and so on.

[0153] After obtaining the structural representation data of the target graph structure, the present application can generate multiple node data for each node of the target graph structure based on the structural representation data, thereby obtaining the mapping data of the target graph structure, and predicting the attribute information of the target graph structure based on the mapping data. Compared with the prior art, when analyzing and predicting the relevant information of the target graph structure, it is necessary to perform the step of generating multiple node data of the node multiple times. Before predicting the attribute information of the target graph structure, the present application generates multiple node data for each node of the target graph structure, which significantly reduces the resource consumption of the computer equipment.

[0154] The method described in the above embodiment is further described in detail below with examples.

[0155] This application will introduce the information prediction method integrated in the information prediction system as an example. The information prediction system may include multiple computer devices. The computer devices may specifically include training devices, parameter storage devices, distributed storage devices, and computer devices used in the model training process, such as Figure 3 As shown, Figure 3 : is a flow chart of the information prediction method provided in the embodiment of the present application. The information prediction method may include:

[0156] 201. A training device receives graph sample representation data of a sample graph structure, where the sample graph structure includes multiple sample nodes. The graph sample representation data includes node sample representation data of each sample node, and the graph sample representation data is sent by a distributed storage device.

[0157] 202. The training device generates multiple node sample data of the first sample node of the sample graph structure based on the graph sample representation data, and sends the multiple node sample data of the first sample node to the distributed storage device, so that the distributed storage device includes multiple node sample data of each sample node of the sample graph structure.

[0158] 203. The training device obtains multiple node sample data of each target sample node from the distributed storage device. The target sample node is obtained by sampling multiple sample nodes of the sample graph structure. The number of target sample nodes is less than the number of sample nodes.

[0159] 204. The training device obtains model parameters of the preset model from the parameter storage device.

[0160] 205. The training device performs data aggregation processing on multiple sample node data of each target sample node to obtain sample mapping data of a sample graph structure.

[0161] 206. The training device trains the prediction model based on the sample mapping data and the model parameters, and the device updates the model parameters of the prediction model.

[0162] 207. The training device sends device update model parameters to the parameter storage device, so that the parameter storage device performs parameter update according to the received multiple device update model parameters to obtain a trained prediction model.

[0163] 208. The computer device predicts the attribute information of the target graph structure based on the trained prediction model to obtain a prediction result.

[0164] The information propagation architecture used in this application for information prediction can be found in Figure 4 The graph multilayer perceptron (GMLP) obtained according to the information propagation architecture can be used as a prediction model. Specifically, based on the graph multilayer perceptron, node classification, link (i.e., edge) prediction, etc. of the graph structure can be realized. Specifically, the graph multilayer perceptron of this application can be subdivided into three types, a graph multilayer perceptron using a non-adaptive aggregator (i.e., GMLP), a graph multilayer perceptron using an adaptive aggregator (GMLP-GMU), and a graph multilayer perceptron using a non-adaptive aggregator and an adaptive aggregator (i.e., GMLP-GU).

[0165] The information dissemination architecture of the present application will process the data in the graph structure and disseminate information during the processing. Specifically, a graph aggregator can be used to generate multiple node data for each node, and the mapping data of the graph structure can be generated by the information aggregator. The mapping data is updated by the information updater to obtain the updated data of the graph structure. At the operation layer of the information dissemination architecture, the algorithms used for processing the data in the graph structure are specifically explained, such as the attention mechanism algorithm (Attention), the average algorithm Average, specifically normalized adjacency (an attention mechanism algorithm), and the personalized web page ranking algorithm (Personalized PageRank). At the storage layer, the graph structure and the segmented graph information can be stored through a distributed storage system based on the distributed storage of the graph information.

[0166] A structure of the multilayer perceptron in this application can be found in Figure 5 , the target graph structure contains T-order neighbors. For each node in the target graph structure, multiple node data of the node can be determined in an iterative manner. After determination, the node data of all nodes can be aggregated by a non-adaptive aggregator to obtain the mapping data of the target graph structure. The mapping data is input into the multi-layer perceptron for information update, and the updated data is input into the classification loss function to train the multi-layer perceptron, or input into the classifier for information prediction, and so on.

[0167] Another structure of the multilayer perceptron in this application can be found in Figure 6 , the target graph structure contains T-order neighbors. For each node in the target graph structure, multiple node data of the node can be determined iteratively, and the reference vector of the target graph structure can be obtained. The reference vector contains the node weight data of each node in the target graph structure. Then, an adaptive aggregator (such as the self-guided Attention aggregator in the figure) is used to fuse the reference vector and the node data of each node to obtain the mapping data of the target graph structure. The mapping data is then input into the multi-layer perceptron for information update, and the updated data is input into the classification loss function to train the multi-layer perceptron, or input into the classifier for information prediction, and so on.

[0168] Another structure of the multilayer perceptron in this application can be found in Figure 7 , the target graph structure contains T-order neighbors. For each node in the target graph structure, multiple node data of the node can be determined iteratively, and then the node data of all nodes are aggregated through a non-adaptive aggregator to obtain the first mapping data of the target graph structure. The first mapping data is input into the multilayer perceptron for information update, and the updated data is used as the reference vector of the target graph structure. The reference vector contains the node weight data of each node of the target graph structure. Then, an adaptive aggregator (such as the self-guided Attention aggregator in the figure) is used to fuse the reference vector and the node data of each node to obtain the mapping data of the target graph structure. The mapping data is then input into the multilayer perceptron for information update, and the updated data is input into the classification loss function to train the multilayer perceptron, or input into the classifier for information prediction, and so on.

[0169] During the model training process, see Figure 8 , a distributed storage device can be used to store multiple node information (i.e., node data) of each node of the sample graph structure. The process of generating node information can be performed by a training device (i.e., the worker in the figure). A training device can generate multiple node data of the target graph structure node and store them in a distributed storage device, which is the preprocessing process in the figure. In the actual training process, the parameters of the model can be in the parameter storage device (P1, P2, ..., PN in the figure). The training device can receive the parameters in the parameter storage device and the sampled graph information in the distributed storage device to perform model training and transmit the updated parameters after training to the parameter storage device. The parameter storage devices can update the model parameters according to the received updated parameters until the training is completed.

[0170] To facilitate better implementation of the information prediction method provided in the embodiment of the present application, the embodiment of the present application also provides a device based on the above information prediction method. The meanings of the terms are the same as those in the above information prediction method, and the specific implementation details can be referred to the description in the method embodiment.

[0171] like Figure 9 As shown, Figure 9 A schematic diagram of the structure of an information prediction device provided in an embodiment of the present application, wherein the information prediction device may include an acquisition module 301, a generation module 302, an aggregation module 303 and a prediction module 304, wherein:

[0172] An acquisition module 301 is configured to acquire structural representation data of a target graph structure, where the target graph structure includes a plurality of nodes, and the structural representation data represents the plurality of nodes of the target graph structure;

[0173] A generating module 302 is configured to generate a plurality of node data for each node of a target graph structure according to the structure representation data;

[0174] Aggregation module 303, used to aggregate multiple node data of each node to obtain mapping data of the target graph structure;

[0175] The prediction module 304 is used to predict the attribute information of the target graph structure based on the mapping data to obtain a prediction result.

[0176] In some embodiments, the aggregation module includes an aggregation submodule and a merging submodule, wherein:

[0177] Aggregation submodule, used to aggregate multiple node data of each node to obtain aggregated data of each node;

[0178] The merging submodule is used to merge the aggregated data of all nodes to obtain the mapping data of the target graph structure.

[0179] In some embodiments, the aggregation submodule is specifically configured to:

[0180] All node data of the target node are fused and calculated to obtain the aggregated data of the target node.

[0181] In some embodiments, the aggregation submodule is specifically configured to:

[0182] Get the aggregation weight information corresponding to each node data of the target node;

[0183] Based on each node data of the target node and the aggregation weight information corresponding to each node data, a fusion calculation is performed to obtain the aggregated data of the target node.

[0184] In some embodiments, the target graph structure further includes a plurality of edges, the edges connecting nodes having an association relationship, the structure representation data includes node representation data, and the generation module includes a determination submodule and a generation submodule, wherein,

[0185] A determination submodule, configured to determine at least one node on the same edge as the target node;

[0186] The generating submodule is used to generate multiple node data of the target node according to the node representation data of the target node and the node representation data of the nodes on the same edge.

[0187] In some embodiments, the target graph structure includes a target number of nodes, and the generation submodule is specifically used to:

[0188] Perform fusion calculation on the node representation data of the target node and the node representation data of the nodes on the same edge to obtain a node data of the target node;

[0189] A new same-side node located on the same side as the same-side node is cyclically determined, and node representation data of the target node, the node representation data of the same-side node, and the node representation data of the new same-side node are fused and calculated to obtain a node data of the target node, until the new same-side node is the target number of same-side nodes of the target node, thereby obtaining multiple node data of the target node.

[0190] In some embodiments, the prediction module is specifically configured to:

[0191] The mapping data is input into the trained prediction model to predict the attribute information of the target graph structure and obtain the prediction results.

[0192] In some embodiments, the prediction device further comprises:

[0193] a determination module, configured to determine a plurality of sample node data of each sample node in the sample graph structure according to the graph sample representation data of the sample graph structure;

[0194] A sampling module is used to sample multiple sample nodes of the sample graph structure and determine multiple target sample nodes required for training the prediction model;

[0195] The training module is used to train the prediction model based on multiple sample node data of each target sample node to obtain a trained prediction model.

[0196] In some embodiments, the prediction device further comprises:

[0197] A receiving module, configured to receive graph sample representation data of a sample graph structure sent by a distributed storage system;

[0198] At this point, determine the module specifically for:

[0199] Determining, according to the graph sample representation data, a plurality of sample node data of a first sample node in the sample graph structure;

[0200] The multiple sample node data of the first sample node are transmitted back to the distributed storage system, thereby obtaining a distributed storage system including the multiple sample node data of each sample node in the sample graph structure.

[0201] The data acquisition module is used to acquire multiple sample node data of each target sample node from the distributed storage system.

[0202] In some embodiments, the training module includes an acquisition submodule, an aggregation submodule, and a training submodule, wherein:

[0203] An acquisition submodule, used to obtain model parameters of the prediction model from a parameter storage device;

[0204] an aggregation submodule, configured to aggregate the multiple sample node data of each target sample node to obtain first sample mapping data of a sample graph structure;

[0205] The training submodule is used to train the prediction model based on the first sample mapping data and the model parameters to obtain a trained prediction model.

[0206] In some embodiments, the aggregation submodule is specifically configured to:

[0207] Performing a first aggregation process on all sample node data of each target sample node to obtain aggregation weight data of each target sample node;

[0208] Performing a second aggregation process based on the aggregation weight data of each target sample node and all sample node data to obtain sample aggregation data of each target sample node;

[0209] The sample aggregation data of all target sample nodes are merged to obtain the first sample mapping data of the sample graph structure.

[0210] In some embodiments, the training submodule is specifically configured to:

[0211] Training the prediction model based on the first sample mapping data and the model parameters to obtain initial updated model parameters of the prediction model;

[0212] Sending initial updated model parameters to the parameter storage device so that the parameter storage device synchronously updates the model parameters of the prediction model based on the initial updated model parameters received from multiple training devices to obtain updated model parameters, so as to obtain a trained prediction model including the updated model parameters.

[0213] In the present application, the acquisition module 301 can obtain the structural representation data of the target graph structure, the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure. The generation module 302 can generate multiple node data of each node of the target graph structure based on the structural representation data. The aggregation module 303 can aggregate the multiple node data of each node to obtain the mapping data of the target graph structure. The prediction module 304 can predict the attribute information of the target graph structure based on the mapping data to obtain the prediction result.

[0214] After obtaining the structural representation data of the target graph structure, the present application can generate multiple node data for each node of the target graph structure based on the structural representation data, thereby obtaining the mapping data of the target graph structure, and predicting the attribute information of the target graph structure based on the mapping data. Compared with the prior art, when analyzing and predicting the relevant information of the target graph structure, it is necessary to perform the step of generating multiple node data of the node multiple times. Before predicting the attribute information of the target graph structure, the present application generates multiple node data for each node of the target graph structure, which significantly reduces the resource consumption of the computer equipment.

[0215] As in the information prediction method or apparatus of the present application, a plurality of storage devices may be formed into a blockchain, and these storage devices are nodes on the blockchain.

[0216] In addition, the embodiment of the present application also provides a computer device, which can be a terminal or a server, such as Figure 10 , which shows a schematic diagram of the structure of the computer device involved in the embodiment of the present application, specifically:

[0217] The computer device may include one or more processing core processors 401, one or more computer readable storage media memories 402, a power supply 403, an input unit 404 and other components. Those skilled in the art will understand that Figure 10 The computer device structure shown in the figure does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.

[0218] Processor 401 is the control center of the computer device. It utilizes various interfaces and lines to connect the various components of the entire computer device. By running or executing software programs and / or modules stored in memory 402 and accessing data stored in memory 402, it performs various functions of the computer device and processes data, thereby performing overall testing of the computer device. Optionally, processor 401 may include one or more processing cores; preferably, processor 401 may integrate an application processor and a modem processor, wherein the application processor primarily processes the operating system, user interfaces, and application programs, while the modem processor primarily handles wireless communications. It is understood that the modem processor may not be integrated into processor 401.

[0219] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 402 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0220] The computer device also includes a power supply 403 for supplying power to various components. Preferably, the power supply 403 can be logically connected to the processor 401 via a power management system, thereby enabling the power management system to manage charging, discharging, and power consumption. The power supply 403 can also include one or more DC or AC power supplies, a recharging system, a power failure detection circuit, a power converter or inverter, a power status indicator, and other arbitrary components.

[0221] The computer device may further include an input unit 404, which may be configured to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal input related to user settings and function control.

[0222] Although not shown, the computer device may further include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 401 in the computer device will load the executable files corresponding to one or more application processes into the memory 402 according to the following instructions, and the processor 401 will run the application stored in the memory 402 to implement various functions as follows:

[0223] Structural representation data of a target graph structure is obtained, where the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure; multiple node data of each node of the target graph structure are generated based on the structural representation data; the multiple node data of each node are aggregated to obtain mapping data of the target graph structure; based on the mapping data, attribute information of the target graph structure is predicted to obtain a prediction result.

[0224] The specific implementation of the above operations can be found in the previous embodiments and will not be repeated here.

[0225] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various optional implementations of the above embodiments.

[0226] Those skilled in the art will appreciate that all or part of the steps in the various methods of the above embodiments may be accomplished by a computer program, or by controlling related hardware through a computer program. The computer program may be stored in a computer-readable storage medium and loaded and executed by a processor.

[0227] To this end, an embodiment of the present application further provides a storage medium storing a computer program that can be loaded by a processor to execute the steps of any of the information prediction methods provided in the embodiments of the present application. For example, the computer program can execute the following steps:

[0228] Structural representation data of a target graph structure is obtained, where the target graph structure includes multiple nodes, and the structural representation data represents the multiple nodes of the target graph structure; multiple node data of each node of the target graph structure are generated based on the structural representation data; the multiple node data of each node are aggregated to obtain mapping data of the target graph structure; based on the mapping data, attribute information of the target graph structure is predicted to obtain a prediction result.

[0229] The storage medium may include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0230] Since the computer program stored in the storage medium can execute the steps of any information prediction method provided in the embodiments of the present application, the beneficial effects that can be achieved by any information prediction method provided in the embodiments of the present application can be achieved. Please refer to the previous embodiments for details and will not be repeated here.

[0231] The above is a detailed introduction to an information prediction method, device, storage medium and computer equipment provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core ideas. At the same time, for technical personnel in this field, based on the ideas of the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. An information prediction method, characterized in that: include: Obtaining structural representation data of a target graph structure, the target graph structure comprising a plurality of nodes and a plurality of edges, the structural representation data representing the plurality of nodes of the target graph structure, the edges connecting nodes having an association relationship, the structural representation data comprising node representation data of each node, the target graph structure comprising a target number of nodes, and the target graph structure representing any one of a molecular structure, a route map, a base station distribution, a social network, electronic shopping, and a financial transaction; Generate multiple node data of each node of the target graph structure according to the structural representation data, including: determining at least one same-edge node located on the same edge as the target node; fusing the node representation data of the target node and the node representation data of the same-edge node to obtain a node data of the target node; cyclically determining a new same-edge node located on the same edge as the same-edge node, fusing the node representation data of the target node, the node representation data of the same-edge node, and the node representation data of the new same-edge node to obtain a node data of the target node, until the new same-edge node is the target number of same-edge nodes of the target node, thereby obtaining multiple node data of the target node; Aggregating multiple node data of each node to obtain mapping data of the target graph structure; Based on the mapping data, attribute information of the target graph structure is predicted to obtain a prediction result, which includes at least one of the target graph structure, the type of the node in the target graph structure, and the type of the edge in the target graph structure.

2. The method according to claim 1, characterized in that The aggregating multiple node data of each node to obtain mapping data of the target graph structure includes: Aggregate multiple node data of each node to obtain aggregate data of each node; The aggregated data of all nodes are merged to obtain the mapping data of the target graph structure.

3. The method according to claim 2, characterized in that The aggregating multiple node data of each node to obtain aggregated data of each node includes: All node data of the target node are fused and calculated to obtain the aggregated data of the target node.

4. The method according to claim 2, characterized in that The aggregating multiple node data of each node to obtain aggregated data of each node includes: Get the aggregation weight information corresponding to each node data of the target node; A fusion calculation is performed based on each node data of the target node and the aggregation weight information corresponding to each node data to obtain the aggregated data of the target node.

5. The method according to claim 1, characterized in that The step of predicting the attribute information of the target graph structure based on the mapping data to obtain a prediction result includes: The mapping data is input into the trained prediction model to predict the attribute information of the target graph structure to obtain a prediction result.

6. The method according to claim 5, characterized in that The method further comprises: Determining, according to the graph sample representation data of the sample graph structure, a plurality of sample node data of each sample node in the sample graph structure; Sampling a plurality of sample nodes of the sample graph structure to determine a plurality of target sample nodes required for training a prediction model; The prediction model is trained based on multiple sample node data of each target sample node to obtain a trained prediction model.

7. The method according to claim 6, characterized in that The method further comprises: Receiving graph sample representation data of a sample graph structure sent by a distributed storage system; The determining, based on the graph sample representation data of the sample graph structure, a plurality of sample node data of each sample node in the sample graph structure includes: Determining, according to the graph sample representation data, a plurality of sample node data of a first sample node in the sample graph structure; Transmitting multiple sample node data of the first sample node back to the distributed storage system, thereby obtaining a distributed storage system including multiple sample node data of each sample node in the sample graph structure; The method further includes: training the prediction model based on the plurality of sample node data of each target sample node to obtain the trained prediction model; A plurality of sample node data of each target sample node is obtained from the distributed storage system.

8. The method according to claim 6, characterized in that The prediction model is trained based on the multiple sample node data of each target sample node to obtain a trained prediction model, including: Acquire model parameters of the prediction model from a parameter storage device; Aggregating multiple sample node data of each target sample node to obtain first sample mapping data of the sample graph structure; The prediction model is trained based on the first sample mapping data and the model parameters to obtain a trained prediction model.

9. The method according to claim 8, characterized in that The aggregating the multiple sample node data of each target sample node to obtain the first sample mapping data of the sample graph structure includes: Performing a first aggregation process on all sample node data of each target sample node to obtain aggregation weight data of each target sample node; Performing a second aggregation process based on the aggregation weight data of each target sample node and all sample node data to obtain sample aggregation data of each target sample node; The sample aggregation data of all target sample nodes are merged to obtain first sample mapping data of the sample graph structure.

10. The method according to claim 8, characterized in that The step of training the prediction model based on the first sample mapping data and the model parameters to obtain a trained prediction model includes: Training the prediction model based on the first sample mapping data and the model parameters to obtain initial updated model parameters of the prediction model; The initial updated model parameters are sent to the parameter storage device so that the parameter storage device synchronously updates the model parameters of the prediction model based on the initial updated model parameters received from multiple training devices to obtain updated model parameters, so as to obtain a trained prediction model including the updated model parameters.

11. An information prediction device, characterized in that: include: an acquisition module, configured to acquire structural representation data of a target graph structure, the target graph structure comprising a plurality of nodes and a plurality of edges, the structural representation data representing the plurality of nodes of the target graph structure, the edges connecting nodes having an association relationship, the structural representation data comprising node representation data of each node, the target graph structure comprising a target number of nodes, and the target graph structure representing any one of a molecular structure, a route map, a base station distribution, a social network, electronic shopping, and a financial transaction; A generation module, configured to generate a plurality of node data for each node of the target graph structure according to the structural representation data, wherein the generation module includes a determination submodule and a generation submodule, wherein the determination submodule is configured to determine at least one same-edge node located on the same edge as the target node; The generating submodule is configured to perform a fusion calculation on the node representation data of the target node and the node representation data of the same-side node to obtain a piece of node data of the target node; cyclically determine a new same-side node located on the same side as the same-side node, and perform a fusion calculation on the node representation data of the target node, the node representation data of the same-side node, and the node representation data of the new same-side node to obtain a piece of node data of the target node, until the new same-side node is the target number of same-side nodes of the target node, thereby obtaining multiple pieces of node data of the target node; An aggregation module, configured to aggregate multiple node data of each node to obtain mapping data of the target graph structure; A prediction module is used to predict the attribute information of the target graph structure based on the mapping data to obtain a prediction result, wherein the prediction result includes at least one of the target graph structure, the type of the node in the target graph structure, and the type of the edge in the target graph structure.

12. The device according to claim 11, characterized in that The aggregation module includes an aggregation submodule and a merging submodule, wherein: The aggregation submodule is used to aggregate multiple node data of each node to obtain aggregated data of each node; The merging submodule is used to merge the aggregated data of all nodes to obtain the mapping data of the target graph structure.

13. The device according to claim 12, characterized in that The aggregation submodule is specifically used for: All node data of the target node are fused and calculated to obtain the aggregated data of the target node.

14. The device according to claim 13, characterized in that The aggregation submodule is specifically used for: Get the aggregation weight information corresponding to each node data of the target node; A fusion calculation is performed based on each node data of the target node and the aggregation weight information corresponding to each node data to obtain the aggregated data of the target node.

15. The device according to claim 11, characterized in that The prediction module is specifically used for: The mapping data is input into the trained prediction model to predict the attribute information of the target graph structure to obtain a prediction result.

16. The device according to claim 11, characterized in that The prediction device further comprises: a determination module, configured to determine a plurality of sample node data of each sample node in the sample graph structure according to the graph sample representation data of the sample graph structure; A sampling module, configured to sample a plurality of sample nodes of the sample graph structure and determine a plurality of target sample nodes required for training a prediction model; The training module is used to train the prediction model based on multiple sample node data of each target sample node to obtain a trained prediction model.

17. The device according to claim 16, characterized in that The prediction device further comprises: A receiving module, configured to receive graph sample representation data of a sample graph structure sent by a distributed storage system; The determining module is specifically configured to: Determining, according to the graph sample representation data, a plurality of sample node data of a first sample node in the sample graph structure; Transmitting multiple sample node data of the first sample node back to the distributed storage system, thereby obtaining a distributed storage system including multiple sample node data of each sample node in the sample graph structure; The data acquisition module is used to acquire multiple sample node data of each target sample node from the distributed storage system.

18. The device according to claim 16, characterized in that The training module includes an acquisition submodule, an aggregation submodule and a training submodule, wherein: The acquisition submodule is used to acquire the model parameters of the prediction model from a parameter storage device; The aggregation submodule is configured to aggregate the multiple sample node data of each target sample node to obtain the first sample mapping data of the sample graph structure; The training submodule is used to train the prediction model based on the first sample mapping data and the model parameters to obtain a trained prediction model.

19. The device according to claim 18, characterized in that The aggregation submodule is specifically used for: Performing a first aggregation process on all sample node data of each target sample node to obtain aggregation weight data of each target sample node; Performing a second aggregation process based on the aggregation weight data of each target sample node and all sample node data to obtain sample aggregation data of each target sample node; The sample aggregation data of all target sample nodes are merged to obtain first sample mapping data of the sample graph structure.

20. The device according to claim 18, characterized in that The training submodule is specifically used for: Training the prediction model based on the first sample mapping data and the model parameters to obtain initial updated model parameters of the prediction model; The initial updated model parameters are sent to the parameter storage device so that the parameter storage device synchronously updates the model parameters of the prediction model based on the initial updated model parameters received from multiple training devices to obtain updated model parameters, so as to obtain a trained prediction model including the updated model parameters.

21. A storage medium, characterized in that The storage medium stores a plurality of computer programs, and the computer programs are suitable for being loaded by a processor to execute the steps in any one of the methods according to claims 1 to 10.

22. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 10 when executing the computer program.

23. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps of the method according to any one of claims 1 to 10.

Images