Data relationship reconstruction method, device, computer equipment and storage medium

The data relationship reconstruction method improves e-commerce platform efficiency by reconstructing transaction node relationships using a node feature graph, adjacency matrix, and graph neural network, revealing hidden interactions for better product recommendations.

CN113222023BActive Publication Date: 2024-12-03GUANGZHOU HUADUO NETWORK TECH
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202110535518.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-05-17
Publication Date
2024-12-03
Estimated Expiration
2041-05-17

AI Technical Summary

Technical Problem

Existing data analysis methods fail to effectively explore and utilize the hidden logical relationships between transaction nodes in e-commerce platforms, limiting the development and utilization of e-commerce platforms.

Method used

A data relationship reconstruction method involving the construction of a node feature graph, conversion to an adjacency matrix, and utilization of a graph neural network to generate a reconstruction matrix, followed by similarity calculations and clustering to identify target transaction nodes.

Benefits of technology

Enhances the depth and efficiency of data analysis by revealing and utilizing hidden transaction relationships, enabling better product recommendations and platform development.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113222023B_ABST
    Figure CN113222023B_ABST
Patent Text Reader

Abstract

The present application discloses a data relationship reconstruction method, device, computer equipment and storage medium, including: obtaining a node feature graph that represents the interaction relationship between each transaction node in a transaction network; converting the node feature graph into an adjacency matrix according to the interaction relationship between each transaction node; inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, and reconstructing the relationship of each transaction node in the reconstruction matrix based on a preset similarity calculation formula; and screening out target transaction nodes that meet preset conditions according to the reconstructed relationship between each transaction node. In the process of breaking up and reorganizing the data node feature graph and the adjacency matrix, the graph neural network verifies the correctness of the hidden association relationship between each transaction node that it has learned, and deeply interprets the logical relationship behind the data through graphics processing technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data logic analysis, and in particular to a data relationship reconstruction method, device, computer equipment and storage medium. Background Art

[0002] The purpose of data analysis is to collect and extract information hidden in a large amount of seemingly chaotic data, so as to find out the internal laws of the object of study. In practical applications, data analysis can help people make judgments so that they can take appropriate actions. Data analysis is the process of collecting and analyzing data in an organized and purposeful manner to turn it into information.

[0003] The inventor of the present invention found in the research that in the current data analysis method, the data required for analysis is often directly extracted from the huge data resources by data extraction, and then the data is analyzed and interpreted according to the data relationship of the data representation to obtain the analysis conclusion. However, the connection relationship or interaction relationship of the data representation often has some hidden logical relationships, and these hidden logical relationships help us understand the real scene reflected by the data.

[0004] Specifically in e-commerce application scenarios, there are certain interactive relationships hidden between multiple transaction nodes defined by transaction relationships, such as the relationship between the regions where different sellers are located and the hot-selling products that buyers desire. How to effectively explore and use such transaction relationships is related to the technical support capabilities and development potential of the e-commerce platform. Summary of the invention

[0005] The embodiments of the present invention provide a data relationship reconstruction method, device, computer equipment and storage medium capable of reconstructing data logical relationships and revealing data hidden relationships.

[0006] In order to solve the above technical problems, a technical solution adopted by an embodiment of the present invention is to provide a data relationship reconstruction method, comprising:

[0007] Obtaining a node feature graph representing the interaction relationship between transaction nodes in the transaction network;

[0008] According to the interaction relationship between the transaction nodes, the node feature graph is converted into an adjacency matrix;

[0009] Inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, wherein the reconstruction matrix records the hidden association relationships between the transaction nodes;

[0010] Based on a preset similarity calculation formula, reconstructing the relationship of each transaction node in the reconstruction matrix;

[0011] According to the reconstructed relationships between the transaction nodes, target transaction nodes that meet preset conditions are screened out.

[0012] Optionally, the acquiring of a node feature graph representing an interaction relationship between transaction nodes in the transaction network includes:

[0013] Collecting log information in the transaction network, wherein the log information includes flow data of each transaction node;

[0014] Comparing the traffic data of each transaction node with a preset traffic threshold;

[0015] When the flow data of a transaction node is greater than the flow threshold, collecting regional flow data in a preset area of ​​the transaction node;

[0016] The node feature graph is constructed based on the interactive relationship between each transaction node represented in the regional traffic data.

[0017] Optionally, the transaction node includes: associated parties and / or associated items, and the step of constructing the node feature graph based on the interaction relationship between the transaction nodes represented in the regional traffic data includes:

[0018] Converting the associated parties and / or associated items recorded in the regional traffic data into a preset node image;

[0019] According to the interactive relationship and / or subordinate relationship recorded in the regional traffic data, the node images are connected to generate the node feature graph.

[0020] Optionally, the step of inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph includes:

[0021] Performing an addition operation on the adjacency matrix and a preset unit matrix;

[0022] Normalizing the adjacency matrix after the addition operation to generate a normalized adjacency matrix;

[0023] The normalized adjacency matrix and the node feature graph are input into a preset graph neural network to generate a reconstruction matrix of the node feature graph.

[0024] Optionally, reconstructing the relationship of each transaction node in the reconstruction matrix based on a preset similarity calculation formula includes:

[0025] Performing a transpose operation on the reconstructed matrix to generate a transposed matrix;

[0026] Performing a matrix inner product operation on the reconstruction matrix and the transposed matrix to generate a reconstructed feature map;

[0027] Based on a preset loss function, calculating a feature distance between the node feature map and the reconstructed feature map;

[0028] When the characteristic distance is less than a preset distance threshold, the relationship of each transaction node in the reconstruction matrix is ​​reconstructed according to the similarity calculation formula.

[0029] Optionally, reconstructing the relationship of each transaction node in the reconstruction matrix according to the similarity calculation formula includes:

[0030] Based on the similarity calculation formula, calculate the similarity between the elements in the reconstruction matrix;

[0031] The relationship between the transaction nodes is reconstructed according to the similarities between the elements.

[0032] Optionally, the screening out target transaction nodes meeting preset conditions according to the reconstructed relationships between the transaction nodes includes:

[0033] Clustering the elements whose similarity is greater than a preset similarity threshold and their corresponding transaction nodes to generate a cluster set;

[0034] The transaction nodes in the cluster set with preset role information are determined as target transaction nodes, and the transaction nodes in the cluster set that meet preset traffic conditions are pushed to the target transaction nodes.

[0035] In order to solve the above technical problems, an embodiment of the present invention further provides a data relationship reconstruction device, comprising:

[0036] An acquisition module, used to acquire a node feature graph representing the interaction relationship between transaction nodes in the transaction network;

[0037] A conversion module, used for converting the node feature graph into an adjacency matrix according to the interaction relationship between the transaction nodes;

[0038] A processing module, used for inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, wherein the reconstruction matrix records the hidden association relationships between the transaction nodes;

[0039] A calculation module, used for reconstructing the relationship of each transaction node in the reconstruction matrix based on a preset similarity calculation formula;

[0040] The execution module is used to screen out target transaction nodes that meet preset conditions according to the reconstructed relationships between the transaction nodes.

[0041] Optionally, the data relationship reconstruction device further includes:

[0042] A first collection submodule is used to collect log information in the transaction network, wherein the log information includes flow data of each transaction node;

[0043] A first comparison submodule, used for comparing the flow data of each transaction node with a preset flow threshold;

[0044] A first processing submodule, configured to collect regional flow data in a preset area of ​​a transaction node when flow data of a transaction node is greater than the flow threshold;

[0045] The first execution submodule is used to construct the node feature graph based on the interaction relationship between each transaction node represented in the regional traffic data.

[0046] Optionally, the transaction node includes: associated parties and / or associated items, and the data relationship reconstruction device further includes:

[0047] A first conversion submodule, used to convert the associated parties and / or associated items recorded in the regional traffic data into a preset node image;

[0048] The second execution submodule is used to connect the node images to generate the node feature graph according to the interactive relationship and / or subordinate relationship recorded in the regional traffic data.

[0049] Optionally, the data relationship reconstruction device further includes:

[0050] A first operator module, used for performing an addition operation on the adjacency matrix and a preset unit matrix;

[0051] A second processing submodule is used to perform normalization processing on the adjacency matrix after the addition operation to generate a normalized adjacency matrix;

[0052] The first generating submodule is used to input the normalized adjacency matrix and the node feature graph into a preset graph neural network to generate a reconstruction matrix of the node feature graph.

[0053] Optionally, the data relationship reconstruction device further includes:

[0054] A first transposition submodule, used for performing a transposition operation on the reconstructed matrix to generate a transposed matrix;

[0055] A second operator module is used to perform a matrix inner product operation on the reconstruction matrix and the transposed matrix to generate a reconstructed feature map;

[0056] A third operator module, used for calculating a feature distance between the node feature map and the reconstructed feature map based on a preset loss function;

[0057] The third execution submodule is used to reconstruct the relationship of each transaction node in the reconstruction matrix according to the similarity calculation formula when the characteristic distance is less than a preset distance threshold.

[0058] Optionally, the data relationship reconstruction device further includes:

[0059] A fourth operator module, used for calculating the similarity between the elements in the reconstructed matrix based on the similarity calculation formula;

[0060] The fourth execution submodule is used to reconstruct the relationship between the transaction nodes according to the similarity between the elements.

[0061] Optionally, the data relationship reconstruction device further includes:

[0062] A first clustering submodule is used to cluster elements whose similarity is greater than a preset similarity threshold and their corresponding transaction nodes to generate a cluster set;

[0063] The fifth execution submodule is used to determine the transaction node with preset role information in the cluster set as the target transaction node, and push the transaction node in the cluster set that meets the preset traffic condition to the target transaction node.

[0064] To solve the above technical problem, an embodiment of the present invention further provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the above-mentioned data relationship reconstruction method.

[0065] In order to solve the above technical problem, an embodiment of the present invention further provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the above-mentioned data relationship reconstruction method.

[0066] Compared with the prior art, the beneficial effects of the embodiments of the present invention are:

[0067] By constructing the data relationship between the collected transaction nodes into a node feature graph and a corresponding adjacency matrix, and using a graph neural network to extract features from the node feature graph and the adjacency matrix, the relationship is reconstructed according to the extracted features. When reconstructing the relationship, the graph neural network reconstructs the data based on the hidden logical relationship between the learned data. The process of relationship reconstruction is the process of restoring the data node feature graph and the adjacency matrix. In the process of breaking up and reorganizing the data node feature graph and the adjacency matrix, the graph neural network verifies the correctness of the hidden association relationship between each transaction node it has learned. Then, the learned hidden relationship is used to reconstruct the relationship between each transaction node, and the target transaction nodes that meet the conditions are selected for subsequent use. Through the graphic processing technology, the logical relationship behind the data is deeply interpreted, which improves the depth and efficiency of data analysis. Applying the data mining principle of the present invention, the e-commerce platform data can be rapidly developed and utilized in depth, and related data can be provided for each transaction node, such as providing hot-selling product recommendations to buyers. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0069] Figure 1 A basic flow chart of a data relationship reconstruction method according to a specific embodiment of the present application;

[0070] Figure 2 A schematic diagram of the application of a node feature graph of a specific embodiment of the present application;

[0071] Figure 3 A schematic diagram of a process of constructing a node feature graph through log information according to a specific embodiment of the present application;

[0072] Figure 4 A schematic diagram of a process for constructing a node feature graph according to a specific embodiment of the present application;

[0073] Figure 5 A schematic diagram of a process of generating a reconstruction matrix according to a specific embodiment of the present application;

[0074] Figure 6 A schematic diagram of a process of reconstructing a transaction node relationship through verification according to a specific embodiment of the present application;

[0075] Figure 7 A schematic diagram of a process of reconstructing the relationship between interactive nodes by similarity according to a specific embodiment of the present application;

[0076] Figure 8 This is a schematic diagram of the process of information push by reconstructing a relationship according to a specific embodiment of the present application;

[0077] Fig. 9 A schematic diagram of the basic structure of a data relationship reconstruction device according to an embodiment of the present application;

[0078] Fig.10 This is a basic structural block diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0079] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0080] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0081] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art to which this application belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as here.

[0082] It will be understood by those skilled in the art that the term "terminal" as used herein includes both devices with wireless signal receivers, which are devices with only wireless signal receivers without transmission capability, and devices with receiving and transmitting hardware, which are devices with receiving and transmitting hardware capable of performing two-way communication on a two-way communication link. Such devices may include: cellular or other communication devices, which have a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; PCS (Personal Communications Service), which may combine voice, data processing, fax and / or data communication capabilities; PDA (Personal Digital Assistant), which may include a radio frequency receiver, a pager, Internet / intranet access, a web browser, a notepad, a calendar and / or a GPS (Global Positioning System) receiver; conventional laptop and / or palmtop computers or other devices, which have and / or include a conventional laptop and / or palmtop computer or other device with a radio frequency receiver. The "terminal" used here can be portable, transportable, installed in a vehicle (air, sea and / or land), or suitable and / or configured to operate locally, and / or in a distributed form, at any other location on the earth and / or in space. The "terminal" used here can also be a communication terminal, an Internet terminal, a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device) and / or a mobile phone with music / video playback function, or a smart TV, a set-top box and other devices.

[0083] See also Figure 1 , Figure 1 Schematic diagram of the basic flow of the data relationship reconstruction method of this embodiment.

[0084] like Figure 1 As shown, a data relationship reconstruction method includes:

[0085] S1100, obtaining a node feature graph representing the interaction relationship between transaction nodes in the transaction network;

[0086] A node feature graph is obtained that characterizes the interaction relationship between each transaction node in the transaction network. In this implementation, each terminal node in the Internet node and the product targeted by the data are determined as a transaction node. The data interaction or subordinate relationship between the transaction nodes serves as a bridge to connect the data relationship between each transaction node, and a node feature graph is established.

[0087] The node feature graph can represent (but not limited to): buying and selling relationships, transportation relationships, article citation relationships, book borrowing relationships, and any other data interaction relationships that occur in reality and are recorded in the network.

[0088] In some embodiments, the node feature graph is used to characterize the product transaction relationship in the network. Figure 2 , Figure 2 FIG. 1 is a schematic diagram showing an application of the node feature graph of this embodiment. Figure 2 As shown, the node feature graph is used to construct a topological graph of Internet product transactions. In the graph, a node feature graph for product transactions is established within the regional scope through the ownership relationship, purchase relationship and distance relationship between the product and the seller. It should be pointed out that the transaction nodes in the node feature graph in this embodiment are not limited to this. Depending on the specific application scenario, the number of transaction nodes recorded in the node feature graph can be a positive integer greater than or equal to 2. The way of constructing the connection relationship between transaction nodes is not limited to ownership, purchase relationship and distance relationship. Depending on the specific application scenario, in some embodiments, the connection relationship can also be (not limited to): the transaction platform, transaction payment method, link entry channel, express transportation, return and exchange information, etc. Any relationship that can occur during the online transaction process.

[0089] S1200, converting the node feature graph into an adjacency matrix according to the interaction relationship between the transaction nodes;

[0090] After completing the node feature graphs between each transaction node, it is necessary to convert the node feature graphs into the corresponding adjacency matrix according to the connection relationship between the node feature graphs. The adjacency matrix is ​​a matrix that represents the adjacent relationship between vertices. Specifically, each transaction node is taken as the first of a column or a row, and the connection relationship between it and other transaction nodes is converted into the elements of the corresponding column or row in turn. For example, the corresponding elements of the transaction nodes connected to it are converted to 1, and the corresponding elements of the transaction nodes not connected to it are converted to 0.

[0091] The adjacency matrix in this implementation can be an undirected adjacency matrix or a directed adjacency matrix. When the adjacency matrix is ​​a directed adjacency matrix, the corresponding node feature diagram also needs to indicate the direction of data interaction.

[0092] In this embodiment, when the adjacency matrix is ​​converted, the elements corresponding to the transaction nodes with a connection relationship are converted to 1, and the matrix elements corresponding to the transaction nodes without a connection relationship are converted to 0. However, the corresponding relationship of the conversion is not limited to this. To further subdivide the interaction data between the transaction nodes, the corresponding conversion relationship can be set according to the type of connection relationship between the transaction nodes in the node feature graph. For example, the conversion value of the purchase relationship is set to 3, the conversion value of the affiliation relationship is set to 2, the conversion value of the distance relationship is set to 1, and the conversion value of the no connection relationship is set to 0. It should be pointed out that when the conversion value is set according to the connection relationship type, the size of the value can be set according to the actual needs of the user, and is not limited to the contents listed in this embodiment.

[0093] S1300, inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, wherein the reconstruction matrix records the hidden association relationships between the transaction nodes;

[0094] The generated node feature graph and adjacency matrix are input into a preset graph neural network. The graph neural network in this embodiment can be (but not limited to): one of: Graph Convolution Networks (GCN), Graph Attention Networks, Graph Autoencoders, Graph Generative Networks and Graph Spatial-temporal Networks.

[0095] In some embodiments, the graph neural network is a graph convolution network (GCN), which learns a function mapping to aggregate the transaction node in the node feature graph with its neighbor features to generate a new representation of the node. The new representation is the hidden relationship between the transaction nodes.

[0096] After the output of the graph neural network is processed, a reconstruction matrix of the node feature graph and the adjacency matrix is ​​generated, where each element in the reconstruction matrix belongs to the embedding of the transaction node corresponding to the node feature graph, that is, the reconstruction matrix is ​​the reconstruction matrix corresponding to the node feature graph.

[0097] S1400, reconstructing the relationship of each transaction node in the reconstruction matrix based on a preset similarity calculation formula;

[0098] Each element in the reconstruction matrix belongs to the embedding of the transaction node corresponding to the node feature graph. The similarity between each element in the reconstruction matrix is ​​calculated through the preset similarity calculation formula. The similarity between different transaction nodes is obtained through the similarity calculation, which completes the reconstruction of the relationship between different transaction nodes.

[0099] In this embodiment, the similarity calculation formula is a cosine similarity calculation formula, but the similarity calculation formula is not limited to this. Depending on the specific application scenario, in some embodiments, the similarity calculation formula can be (not limited to): Pearson correlation coefficient calculation formula, Jaccard similarity coefficient calculation formula or Tanimoto coefficient calculation formula, etc.

[0100] S1500: Filter out target transaction nodes that meet preset conditions according to the reconstructed relationships between the transaction nodes.

[0101] After the similarity between each transaction node is calculated using the similarity calculation formula, different transaction nodes are clustered according to the similarity. Different cluster sets are obtained by clustering. In the same cluster set, nodes that meet the preset screening conditions are selected as target transaction nodes. The preset conditions are: screening according to the type of transaction node.

[0102] In some implementation scenarios, the method shown in this embodiment is used to screen sellers or buyers corresponding to hot-selling products, and the cluster set where the hot-selling products are located is determined through clustering. Then, transaction nodes labeled as sellers or buyers are screened as target transaction nodes in the cluster set where the hot-selling products are located, and the hot-selling products are pushed to the target transaction nodes.

[0103] In some implementation scenarios, the method shown in this embodiment is used to screen the seller's hot-selling products, and the cluster set where the transaction node corresponding to the selected seller is located is determined through clustering. Then, the transaction nodes labeled as products are screened in the cluster set as target transaction nodes, and the products represented by the target transaction nodes are selected as the hot-selling products of the selected seller.

[0104] The above implementation method constructs the data relationship between the collected transaction nodes into a node feature graph and a corresponding adjacency matrix, and uses a graph neural network to extract features from the node feature graph and the adjacency matrix, and reconstructs the relationship based on the extracted features. When reconstructing the relationship, the graph neural network reconstructs the data based on the learned hidden logical relationship between the data. The process of relationship reconstruction is the process of restoring the data node feature graph and the adjacency matrix. In the process of breaking up and reorganizing the data node feature graph and the adjacency matrix, the graph neural network verifies the correctness of the hidden association relationship between the transaction nodes it has learned. Then, the learned hidden relationship is used to reconstruct the relationship between the transaction nodes, and the target transaction nodes that meet the conditions are selected for subsequent use. Through graphics processing technology, the logical relationship behind the data is deeply interpreted, which improves the depth and efficiency of data analysis.

[0105] In some implementations, it is necessary to construct a node feature graph by collecting log information from transaction nodes. Figure 3 , Figure 3 This is a schematic diagram of the process of constructing a node feature graph through log information in this embodiment.

[0106] like Figure 3 As shown, S1100 includes:

[0107] S1110, collecting log information in the transaction network, wherein the log information includes flow data of each transaction node;

[0108] Collect log information in the transaction network. The transaction network refers to the network involved in the selected scenario in the specific application scenario of this embodiment. The transaction network is the full network of the network selected by the user. For example, the user selects the transaction network on the XX platform within the administrative jurisdiction of XX Province as the transaction network, and the transaction network is limited to the transaction network of the XX platform within the administrative jurisdiction of XX Province. The nodes involved in the transaction network can be nodes such as sellers, buyers, and goods to which the sellers belong whose registered places are in the administrative region. The transaction network generates transaction logs in daily work, and the transaction logs record the interaction information and traffic data between each network node. Interaction information refers to the relationship between sellers, buyers, goods and regions between each transaction node, and traffic data refers to the number of goods bought and sold in each transaction node or the number of transactions for each commodity.

[0109] S1120, comparing the traffic data of each transaction node with a preset traffic threshold;

[0110] First, the transaction nodes belonging to commodities in the transaction nodes are counted, and then the transaction data of each commodity is compared with the preset flow data. In this embodiment, the flow threshold is the threshold of the commodity transaction quantity.

[0111] It should be pointed out that the transaction nodes in this embodiment can also be all nodes in the transaction network, and the traffic data can be (but not limited to): transaction quantity, transaction price, number of views, number of recommendations or number of uses, etc., which can represent the popularity of a type of transaction nodes or all transaction nodes.

[0112] When the traffic data is greater than the traffic threshold, it means that the data volume of the product or seller has reached the standard of a hot-selling product. Otherwise, it means that the data volume of the product or seller is average and does not meet the requirements for screening hot-selling products.

[0113] S1130: When the flow data of a transaction node is greater than the flow threshold, collecting regional flow data in a preset area of ​​the transaction node;

[0114] When the traffic data of a transaction node in the traffic data is greater than the traffic threshold, it means that a hot-selling product has appeared in the current trading network. Considering that when a product becomes a hot-selling product, its sales volume will often increase in a small area in the area where it is located. Therefore, it is necessary to continue to promote the hot-selling product in the area where the seller of the hot-selling product is located.

[0115] The preset area of ​​the seller where the hot-selling product is located refers to: an area with a radius of 10 kilometers centered on the seller's registration or actual location. However, the scope of the preset area is not limited to this. Depending on the specific application scenario, the radius area of ​​the preset area can be (not limited to): 100m, 1Km, 5Km, 100Km or other radius ranges set by the user. In some embodiments, the preset area can also be an administrative area, for example, the administrative area where the seller of the hot-selling product is located: township, county, city, province or even country.

[0116] After determining the preset area corresponding to the hot-selling product, filter out the regional traffic data of the corresponding area from the full traffic data.

[0117] S1140. Construct the node feature graph based on the interaction relationship between each transaction node represented in the regional traffic data.

[0118] After collecting the regional traffic data, the corresponding transaction nodes in the regional traffic data are extracted, and each transaction node is converted into a graph, with each transaction node as an origin. It should be pointed out that the shape representation of the transaction node graph is not limited to a circle. Depending on the specific application scenario, the shape of the transaction node after graphing can be: a point, a triangle, a quadrilateral, a pentagon, a hexagon, or a regular or irregular image set by the user.

[0119] After the transaction nodes contained in the regional traffic are visualized, each transaction node is connected according to the interactive relationship between the transaction nodes recorded in the regional traffic data. Specifically, the transaction relationship or affiliation between sellers, buyers, commodities and regions. The transaction node images with the above relationship are connected with lines to generate a node feature map.

[0120] In the process of drawing the node feature map, we screen out the transaction nodes that meet the requirements from the full amount of log information, and then use the transaction nodes to extract regional traffic data as the data source of the node feature map. This not only conforms to the common sense of commodity transactions, but also can accurately reduce the amount of data, greatly reduce the amount of calculation in the later processing, and effectively improve efficiency.

[0121] In some embodiments, a transaction node includes: associated parties and / or associated items. In the process of constructing a node feature graph, it is necessary to construct the node feature graph based on the interaction relationship and / or the subordinate relationship. Figure 4 , Figure 4 A schematic diagram of the process of constructing a node feature graph in this embodiment.

[0122] like Figure 4 As shown, S1140 includes:

[0123] S1141, converting the associated parties and / or associated items recorded in the regional traffic data into a preset node image;

[0124] The transaction nodes in this embodiment include related parties and / or related items in the transaction process. The related parties refer to the parties involved in the transaction, such as buyers, sellers, platform parties or payers. The related items are the goods or services involved in the transaction.

[0125] After collecting the regional traffic data, the corresponding transaction nodes in the regional traffic data are extracted, and each transaction node is converted into a graph, with each transaction node as an origin. It should be pointed out that the shape representation of the transaction node graph is not limited to a circle. Depending on the specific application scenario, the shape of the transaction node after graphing can be: a point, a triangle, a quadrilateral, a pentagon, a hexagon, or a regular or irregular image set by the user.

[0126] S1142. Connect the node images according to the interactive relationship and / or subordinate relationship recorded in the regional traffic data to generate the node feature graph.

[0127] After the transaction nodes are graphed, it is necessary to establish connections between the transaction node graphs based on the interaction relationship and / or subordinate relationship of the transaction nodes recorded in the regional traffic data.

[0128] The interactive relationship refers to the buying and selling relationship formed during the transaction process. The subordinate relationship includes the subordinate relationship between goods and sellers, the subordinate relationship between regions, the subordinate relationship between platforms and merchants, and the subordinate relationship between platforms and transaction methods, etc.

[0129] After the transaction nodes included in the regional traffic are visualized, the transaction nodes are connected according to the interactive relationship between the transaction nodes recorded in the regional traffic data, specifically, the transaction relationship or subordinate relationship between sellers, buyers, commodities and regions, and the transaction node images with the above relationship are connected with lines to generate a node feature graph. When two transaction nodes have the above transaction relationship or subordinate relationship, the two nodes are connected with a line segment, otherwise, they are not connected.

[0130] In some implementations, in order to avoid the phenomenon that the diagonal of the adjacency matrix is ​​0 due to the distance between the transaction node and the transaction node itself being 0 when establishing the adjacency matrix, it is necessary to perform data processing on the adjacency matrix. Figure 5 , Figure 5 The following is a flow chart of generating a reconstruction matrix in this embodiment.

[0131] like Figure 5 As shown, S1300 includes:

[0132] S1310, performing an addition operation on the adjacency matrix and a preset unit matrix;

[0133] If the adjacency matrix converted from the node feature graph is directly used in subsequent calculations, since the diagonal of the adjacency matrix is ​​all 0, when multiplied with the feature matrix, only the weighted sum of the features of all neighbors of a transaction node will be calculated, and the features of the transaction node itself will be ignored. Therefore, it is necessary to add a unit matrix to the adjacency matrix. The specifications of the unit matrix are the same as those of the adjacency matrix, both of which are N*M matrices, and all the elements on its diagonal are 1. After adding the unit matrix, the diagonal positions of the adjacency matrix are all 1. When multiplied with the feature matrix later, the features of the transaction node itself will not be ignored.

[0134] The feature matrix refers to the feature matrix of the node feature graph extracted by the graph neural network.

[0135] S1320, normalizing the adjacency matrix after the addition operation to generate a normalized adjacency matrix;

[0136] The adjacency matrix plus the identity matrix is ​​not a normalized matrix, so multiplying it with the feature matrix will change the original distribution of the features and produce some calculation deviations. Therefore, further normalization is required.

[0137] First we need to make the adjacency matrix plus the identity matrix add up to 1 for each row, so we can multiply it by , D is the degree matrix. We can further split Multiply with A to get a symmetric and normalized matrix .

[0138] Among them, A is the adjacency matrix with the identity matrix added, and D is the degree matrix calculated by A. After the above calculation, the normalized adjacency matrix is ​​generated.

[0139] S1330. Input the normalized adjacency matrix and the node feature graph into a preset graph neural network to generate a reconstruction matrix of the node feature graph.

[0140] The normalized adjacency matrix and node feature map are input into the preset graph neural network. The graph neural network extracts the features of the node feature map through the convolution layer. Then, after multiplying the extracted features with the normalized adjacency matrix, the calculated matrix is ​​mapped to generate a reconstructed matrix.

[0141] In some implementations, the graph neural network is a semi-supervised model. In order to ensure that the reconstruction matrix reconstructed by it is as consistent as possible with the feature of the node feature graph, the output results of the graph neural network need to be verified during calculation. Figure 6 , Figure 6 This is a flow chart of reconstructing the relationship between transaction nodes through verification in this embodiment.

[0142] like Figure 6 As shown, S1400 includes:

[0143] S1410, performing a transposition operation on the reconstructed matrix to generate a transposed matrix;

[0144] The reconstruction matrix is ​​transposed, and the purpose of the transposition operation is to transpose the reconstruction matrix Z (n rows and m columns) into a transposed matrix with n columns and m rows. That is, the transposed matrix in this embodiment is to convert the rows and columns in the regrouping matrix.

[0145] S1420, performing a matrix inner product operation on the reconstruction matrix and the transposed matrix to generate a reconstructed feature map;

[0146] After the transposed matrix is ​​calculated, a matrix inner product operation is performed on the reconstructed matrix and the transposed matrix, that is, the reconstructed matrix and the transposed matrix are matrix multiplied to generate a new matrix, which we define as the inner product matrix.

[0147] The inner product matrix records the connection relationship between each transaction node. By performing the inverse operation of generating the adjacency matrix, the reconstructed feature graph corresponding to the inner product matrix is ​​calculated.

[0148] S1430, calculating a feature distance between the node feature graph and the reconstructed feature graph based on a preset loss function;

[0149] Based on a preset loss function, the feature distance between the node feature map and the reconstructed feature map is calculated. Specifically, the features of the node feature map and the reconstructed feature map are extracted respectively through the convolution layer of the graph neural network, and then the feature distance between the two feature matrices is calculated through the loss function. However, the calculation method of the feature distance is not limited to this. In some embodiments, the loss function calculates the feature distance between the feature matrix of the node feature map and the inner product matrix.

[0150] The loss function in this embodiment is: cross entropy loss function. However, the type of loss function is not limited to this. Depending on the specific application scenario, in some embodiments, the loss function can also be: absolute value loss function, log loss function, square loss function, exponential loss function or Hinge loss function, one or a mixed loss function of at least two.

[0151] S1440: When the characteristic distance is less than a preset distance threshold, reconstruct the relationship of each transaction node in the reconstruction matrix according to the similarity calculation formula.

[0152] When the calculated feature distance is less than the preset distance threshold, the similarity between the Embeddings representing each transaction node in the reconstructed evidence is calculated according to the similarity calculation formula.

[0153] In this embodiment, the similarity calculation formula is a cosine similarity calculation formula, but the similarity calculation formula is not limited to this. Depending on the specific application scenario, in some embodiments, the similarity calculation formula can be (not limited to): Pearson correlation coefficient calculation formula, Jaccard similarity coefficient calculation formula or Tanimoto coefficient calculation formula, etc.

[0154] When the calculated feature distance is greater than or equal to the set distance threshold, the weights in the graph neural network need to be corrected through the set feedback function. In this way, the training is iterated until the calculated feature distance is less than the preset distance threshold.

[0155] The distance threshold in this implementation can be set according to the actual needs of the user. The smaller the value, the higher the required accuracy of the model, and vice versa.

[0156] By calculating the distance between the node feature map and the reconstructed feature map, we can obtain the reconstruction restoration rate of the node feature map by the graph neural network. The smaller the feature distance, the higher the restoration rate of the node feature map by the graph neural network, and the more correct the extraction of hidden relationships in the node feature map, thus improving the accuracy of model processing.

[0157] In some implementations, it is necessary to reconstruct the relationship between each transaction node by calculating the similarity between each element in the reconstruction matrix. Figure 7 , Figure 7 This is a schematic diagram of the process of reconstructing the interactive node relationship through similarity in this embodiment.

[0158] like Figure 7 As shown, S1430 includes:

[0159] S1431, calculating the similarity between the elements in the reconstruction matrix based on the similarity calculation formula;

[0160] In this embodiment, the similarity calculation formula is a cosine similarity calculation formula, but the similarity calculation formula is not limited to this. Depending on the specific application scenario, in some embodiments, the similarity calculation formula can be (not limited to): Pearson correlation coefficient calculation formula, Jaccard similarity coefficient calculation formula or Tanimoto coefficient calculation formula, etc.

[0161] Take the cosine similarity calculation formula as an example: in the reconstruction matrix, each element is the embedding of the corresponding transaction node in it, and the embedding is a vector of fixed length. Therefore, the cosine angle between two vectors can be calculated through the cosine similarity calculation formula. The smaller the cosine angle, the more similar the two vectors are, and vice versa.

[0162] S1432: Reconstruct the relationship between the transaction nodes according to the similarities between the elements.

[0163] After calculating the similarity between the corresponding vectors of each transaction node, the similarity relationship between the transaction nodes can be obtained. Then, different clustering sets are established according to the similarity relationship, thereby completing the reconstruction of the relationship between the transaction nodes.

[0164] In some implementations, it is necessary to filter out target transaction nodes based on similarity relationships for product push. Figure 8 , Figure 8 The figure is a schematic diagram of the process of pushing information by reconstructing the relationship in this embodiment.

[0165] like Figure 8 As shown, S1500 includes:

[0166] S1510, clustering elements whose similarity is greater than a preset similarity threshold and their corresponding transaction nodes to generate a cluster set;

[0167] After the similarity between each transaction node is calculated using the similarity calculation formula, different transaction nodes are clustered according to the similarity to obtain different cluster sets.

[0168] S1520: Determine a transaction node in the cluster set having preset role information as a target transaction node, and push the transaction nodes in the cluster set that meet preset traffic conditions to the target transaction node.

[0169] After generating different cluster sets, the target cluster set is determined according to the specific implementation scenario. For example, if the target is to recommend popular products, the cluster set containing the popular products is determined as the target cluster set; if the target is to predict the best-selling products of merchants, the cluster set containing the selected merchants is determined as the target cluster set; if the target is to predict the best-selling products of platforms, the cluster set containing the selected platforms is determined as the target cluster set. Therefore, the selection of the target cluster set can be flexibly set according to the specific applicable application scenario.

[0170] After determining the target cluster set, determine which one or which transaction nodes in the target cluster set are the target transaction nodes based on the identity tags of each transaction node in the cluster set. Similarly, the screening of role information also needs to be set according to the specific scenario, for example: when the method of this embodiment is used to recommend hot-selling products to sellers, the preset role is selected as the "seller" role; when the method of this embodiment is used to recommend hot-selling products to buyers, the preset role is selected as the "buyer"; when the method of this embodiment is used to predict the hot-selling products of merchants, the preset role is selected as the "product"; when the method of this embodiment is used to predict the hot-selling products of the platform, the preset role is selected as the "product".

[0171] After the target transaction node is obtained through screening, the transaction node that meets the preset traffic condition is pushed to the target transaction node. The preset traffic condition in this embodiment is a hot-selling or hot-selling product with a sales volume greater than a set threshold.

[0172] In order to solve the above technical problems, the embodiment of the present invention also provides a data relationship reconstruction device. Fig. 9 , Fig. 9 Schematic diagram of the basic structure of the data relationship reconstruction device in this embodiment.

[0173] like Fig. 9As shown, a data relationship reconstruction device includes: an acquisition module, a conversion module, a processing module, a calculation module and an execution module. The acquisition module is used to acquire a node feature graph that represents the interaction relationship between each transaction node in the transaction network; the conversion module is used to convert the node feature graph into an adjacency matrix according to the interaction relationship between each transaction node; the processing module is used to input the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, wherein the reconstruction matrix records the hidden association relationship between each transaction node; the calculation module is used to reconstruct the relationship between each transaction node in the reconstruction matrix based on a preset similarity calculation formula; the execution module is used to screen out target transaction nodes that meet preset conditions according to the reconstructed relationship between each transaction node.

[0174] The data relationship reconstruction device constructs the data relationship between the collected transaction nodes into a node feature graph and a corresponding adjacency matrix, and uses a graph neural network to extract features from the node feature graph and the adjacency matrix, and reconstructs the relationship based on the extracted features. When reconstructing the relationship, the graph neural network reconstructs the data based on the learned hidden logical relationship between the data. The process of relationship reconstruction is the process of restoring the data node feature graph and the adjacency matrix. In the process of breaking up and reorganizing the data node feature graph and the adjacency matrix, the graph neural network verifies the correctness of the hidden association relationship between the transaction nodes it has learned, and then uses the learned hidden relationship to reconstruct the relationship between the transaction nodes, and selects the target transaction nodes that meet the conditions for subsequent use. Through graphics processing technology, the logical relationship behind the data is deeply interpreted, which improves the depth and efficiency of data analysis.

[0175] In some embodiments, the data relationship reconstruction device further includes: a first collection submodule, a first comparison submodule, a first processing submodule and a first execution submodule. The first collection submodule is used to collect log information in the transaction network, wherein the log information includes the flow data of each transaction node; the first comparison submodule is used to compare the flow data of each transaction node with a preset flow threshold; the first processing submodule is used to collect regional flow data in a preset area of ​​the transaction node when the flow data of a transaction node is greater than the flow threshold; the first execution submodule is used to construct the node feature graph based on the interaction relationship between each transaction node represented in the regional flow data.

[0176] In some embodiments, the transaction node includes: associated parties and / or associated items, and the data relationship reconstruction device further includes: a first conversion submodule and a second execution submodule. The first conversion submodule is used to convert the associated parties and / or associated items recorded in the regional flow data into a preset node image; the second execution submodule is used to connect the node images to generate the node feature graph according to the interactive relationship and / or subordinate relationship recorded in the regional flow data.

[0177] In some embodiments, the data relationship reconstruction device further includes: a first operation submodule, a second processing submodule and a first generation submodule. The first operation submodule is used to perform an addition operation on the adjacency matrix and a preset unit matrix; the second processing submodule is used to perform a normalization process on the adjacency matrix after the addition operation to generate a normalized adjacency matrix; the first generation submodule is used to input the normalized adjacency matrix and the node feature graph into a preset graph neural network to generate a reconstruction matrix of the node feature graph.

[0178] In some embodiments, the data relationship reconstruction device further includes: a first transposition submodule, a second operation submodule, a third operation submodule and a third execution submodule. The first transposition submodule is used to perform a transposition operation on the reconstruction matrix to generate a transposed matrix; the second operation submodule is used to perform a matrix inner product operation on the reconstruction matrix and the transposed matrix to generate a reconstructed feature graph; the third operation submodule is used to calculate the feature distance between the node feature graph and the reconstructed feature graph based on a preset loss function; the third execution submodule is used to reconstruct the relationship of each transaction node in the reconstruction matrix according to the similarity calculation formula when the feature distance is less than a preset distance threshold.

[0179] In some embodiments, the data relationship reconstruction device further includes: a fourth operator module and a fourth execution submodule. The fourth operator module is used to calculate the similarity between the elements in the reconstruction matrix based on the similarity calculation formula; the fourth execution submodule is used to reconstruct the relationship between the transaction nodes according to the similarity between the elements.

[0180] In some embodiments, the data relationship reconstruction device further includes: a first clustering submodule and a fifth execution submodule. The first clustering submodule is used to cluster elements with similarity greater than a preset similarity threshold and their corresponding transaction nodes to generate a cluster set; the fifth execution submodule is used to determine the transaction node with preset role information in the cluster set as the target transaction node, and push the transaction node in the cluster set that meets the preset traffic condition to the target transaction node.

[0181] To solve the above technical problems, the embodiment of the present invention also provides a computer device. Fig.10 , Fig.10 This is a basic structural block diagram of the computer device in this embodiment.

[0182] like Fig.10 As shown, a schematic diagram of the internal structure of a computer device. The computer device includes a processor, a non-volatile storage medium, a memory, and a network interface connected via a system bus. Among them, the non-volatile storage medium of the computer device stores an operating system, a database, and computer-readable instructions. The database may store a control information sequence. When the computer-readable instructions are executed by the processor, the processor can implement a data relationship reconstruction method. The processor of the computer device is used to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute a data relationship reconstruction method. The network interface of the computer device is used to connect and communicate with a terminal. Those skilled in the art can understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0183] In this embodiment, the processor is used to execute Fig. 9 The specific functions of the acquisition module 1100, the conversion module 1200, the processing module 1300, the calculation module 1400 and the execution module 1500 are stored in the memory, and the program code and various data required to execute the above modules are stored. The network interface is used to transmit data between user terminals or servers. The memory in this embodiment stores the program code and data required to execute all sub-modules in the data relationship reconstruction device, and the server can call the program code and data of the server to execute the functions of all sub-modules.

[0184] The computer equipment constructs the data relationship between the collected transaction nodes into a node feature graph and a corresponding adjacency matrix, and uses a graph neural network to extract features from the node feature graph and the adjacency matrix, and reconstructs the relationship based on the extracted features. When reconstructing the relationship, the graph neural network reconstructs the data based on the learned hidden logical relationship between the data. The process of relationship reconstruction is the process of restoring the data node feature graph and the adjacency matrix. In the process of breaking up and reorganizing the data node feature graph and the adjacency matrix, the graph neural network verifies the correctness of the hidden association relationship between the transaction nodes it has learned, and then reconstructs the relationship between the transaction nodes using the learned hidden relationship, and selects the target transaction nodes that meet the conditions for subsequent use. Through graphics processing technology, the logical relationship behind the data is deeply interpreted, which improves the depth and efficiency of data analysis.

[0185] The present invention also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the data relationship reconstruction method in any of the above embodiments.

[0186] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0187] The present invention also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the data relationship reconstruction method in any of the above embodiments.

[0188] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0189] Those skilled in the art will appreciate that the various operations, methods, steps, measures, and schemes in the processes discussed in this application may be alternated, altered, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art that are similar to those disclosed in this application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

[0190] The above description is only a partial implementation method of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A data relationship reconstruction method, characterized in that: Used in e-commerce transaction scenarios, including: Obtaining a node characteristic graph representing the interaction relationship between each transaction node in a transaction network, including: collecting log information in the transaction network, wherein the log information includes flow data of each transaction node; comparing the flow data of each transaction node with a preset flow threshold; when the flow data of a transaction node is greater than the flow threshold, collecting regional flow data in a preset area of ​​the transaction node; and constructing the node characteristic graph based on the interaction relationship between each transaction node represented by the regional flow data; According to the interaction relationship between the transaction nodes, the node feature graph is converted into an adjacency matrix; Inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, including: adding the adjacency matrix to a preset unit matrix; normalizing the adjacency matrix after the addition operation to generate a normalized adjacency matrix; inputting the normalized adjacency matrix and the node feature graph into a preset graph neural network to generate a reconstruction matrix of the node feature graph, wherein the reconstruction matrix records the hidden associations between the transaction nodes; Based on a preset similarity calculation formula, the relationship of each transaction node in the reconstruction matrix is ​​reconstructed, including: performing a transposition operation on the reconstruction matrix to generate a transposed matrix; performing a matrix inner product operation on the reconstruction matrix and the transposed matrix to generate a reconstructed feature graph; based on a preset loss function, calculating the feature distance between the node feature graph and the reconstructed feature graph; when the feature distance is less than a preset distance threshold, the relationship of each transaction node in the reconstruction matrix is ​​reconstructed according to the similarity calculation formula; According to the reconstructed relationship between the transaction nodes, the target transaction nodes meeting the preset conditions are screened out, and the transaction nodes meeting the preset traffic conditions are pushed to the target transaction nodes.

2. The data relationship reconstruction method according to claim 1, characterized in that: The transaction nodes include: associated parties and / or associated items, and the node feature graph constructed based on the interactive relationship between the transaction nodes represented in the regional traffic data includes: Converting the associated parties and / or associated items recorded in the regional traffic data into a preset node image; According to the interactive relationship and / or subordinate relationship recorded in the regional traffic data, the node images are connected to generate the node feature graph.

3. The data relationship reconstruction method according to claim 1, characterized in that: The reconstructing the relationship of each transaction node in the reconstruction matrix according to the similarity calculation formula includes: Based on the similarity calculation formula, calculate the similarity between the elements in the reconstruction matrix; The relationship between the transaction nodes is reconstructed according to the similarities between the elements.

4. The data relationship reconstruction method according to claim 3, characterized in that: The step of screening out target transaction nodes that meet preset conditions according to the reconstructed relationships between the transaction nodes includes: Clustering the elements whose similarity is greater than a preset similarity threshold and their corresponding transaction nodes to generate a cluster set; The transaction nodes in the cluster set with preset role information are determined as target transaction nodes, and the transaction nodes in the cluster set that meet preset traffic conditions are pushed to the target transaction nodes.

5. A data relationship reconstruction device, characterized in that: Used in e-commerce transaction scenarios, including: An acquisition module, used for acquiring a node characteristic graph representing the interaction relationship between each transaction node in the transaction network, comprising: collecting log information in the transaction network, wherein the log information includes flow data of each transaction node; comparing the flow data of each transaction node with a preset flow threshold; when the flow data of a transaction node is greater than the flow threshold, collecting regional flow data in a preset area of ​​the transaction node; and constructing the node characteristic graph based on the interaction relationship between each transaction node represented in the regional flow data; A conversion module, used for converting the node feature graph into an adjacency matrix according to the interaction relationship between the transaction nodes; A processing module, used for inputting the node feature graph and the adjacency matrix into a preset graph neural network to generate a reconstruction matrix of the node feature graph, including: adding the adjacency matrix to a preset unit matrix; normalizing the adjacency matrix after the addition operation to generate a normalized adjacency matrix; inputting the normalized adjacency matrix and the node feature graph into a preset graph neural network to generate a reconstruction matrix of the node feature graph, wherein the reconstruction matrix records the hidden associations between the transaction nodes; A calculation module, used for reconstructing the relationship of each transaction node in the reconstruction matrix based on a preset similarity calculation formula, including: performing a transposition operation on the reconstruction matrix to generate a transposed matrix; performing a matrix inner product operation on the reconstruction matrix and the transposed matrix to generate a reconstructed feature graph; calculating a feature distance between the node feature graph and the reconstructed feature graph based on a preset loss function; when the feature distance is less than a preset distance threshold, reconstructing the relationship of each transaction node in the reconstruction matrix according to the similarity calculation formula; The execution module is used to screen out target transaction nodes that meet preset conditions according to the reconstructed relationships between the transaction nodes, and push the transaction nodes that meet the preset traffic conditions to the target transaction nodes.

6. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the data relationship reconstruction method according to any one of claims 1 to 4.

7. A storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to perform the steps of the data relationship reconstruction method according to any one of claims 1 to 4.

Citation Information

Patent Citations

  • Graph data category prediction method based on adaptive fractional order gradient

    CN111353534A

  • Personnel association method and device, and graph convolution network training method and device

    CN111553215A

  • Data processing method and device, electronic equipment and storage medium

    CN111581488A

  • Multi-attribute image semantic analysis method and device

    CN112215837A

  • Hyperedge link prediction method based on hypergraph convolution

    CN112417219A