Enterprise application complex service node optimization method and system based on graph neural network

Abstract

The invention provides an enterprise application complex service node optimization method and system based on a graph neural network, and relates to the technical field of service semantic network optimization, and the method comprises the steps: defining the internal attributes of nodes and edges of an enterprise application service network, and constructing a graph structure based on the internal attributes of the nodes and edges; inputting the graph structure into a graph neural network model for linear conversion, performing vector conversion on feature vector representation of nodes and edges by adopting an initial parameter matrix, performing graph neural network GNN multi-iterative calculation on the nodes and the edges by utilizing a self-attention mechanism, calculating attention parameter vectors of correlation among the nodes, and obtaining a graph neural network; determining the correlation between the nodes; and performing column normalization on the attention parameter vector of each node and edge, using function conversion to adjust the weight, outputting to obtain feature vectors, aggregating the feature vectors, performing dynamic calculation on the weight of each node, and finally outputting to obtain an optimized graph structure, thereby realizing optimization of the nodes in the enterprise application service scene.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of business semantic network optimization, in particular to an enterprise application complex business node optimization method and system based on a graph neural network. BACKGROUND

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute the prior art.

[0003] For the enterprise application ERP field with limited domain entities, how to organically combine the entity's cognition, relationship logic concept and artificial intelligence and play the maximum role determines the foundation and long-term development of artificial intelligence in the enterprise application field. With the development of artificial intelligence, traditional enterprise application products and architecture are facing the dilemma of being difficult to integrate and natively artificial intelligence capabilities. The factors that bring about this dilemma are multifaceted. First, enterprise applications need a large accumulation of enterprise business best practices, which are difficult to restructure. In addition, the traditional enterprise application architecture is relatively stable, and enterprise users have high stability demands for many core businesses. These two factors also consolidate the user interaction layer and the business logic layer of the enterprise application.

[0004] Based on the business logic layer and the user interaction layer that are not easy to shake, the data layer of the traditional enterprise application three-layer architecture becomes the point of creative innovation. The data layer abstracts business entities and entity relationships. The traditional relational database RDBMS embodies related entities and entity relationships through database tables and foreign key relationships. However, for the entity relationships at this stage, there are only technical semantics, but no business semantics, which means that the users of the data layer are more developers than enterprise business users. At the same time, artificial intelligence cannot achieve the cognition of technical relationships to directly serve enterprise business users. In existing methods, general optimization methods cause enterprise applications to face the problem of chaotic construction of graph structure business entities and the inability to effectively perform secondary optimization, resulting in serious accuracy and performance efficiency problems. SUMMARY

[0005] To solve the above problems, the present disclosure proposes an enterprise application complex business node optimization method and system based on a graph neural network, which uses a graph structure data abstraction definition and storage method to replace a linear data structure, uses natural logic relationships between node entities to replace foreign key linear entity relationships, uses the Transformer mechanism of artificial intelligence LLM, combines Attention attention to balance the weights of nodes and relationships, and realizes the processing and optimization of non-Euclidean data structures through a neural network, achieving accurate and efficient identification and extraction of key node relationships in a complex business graph structure.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions: The enterprise application complex business node optimization method based on a graph neural network comprises the following steps: Defining internal attributes of nodes and edges of an enterprise application business network; Constructing a graph structure based on the internal attributes of the nodes and the edges; Inputting the graph structure into a graph neural network model for linear conversion, performing vector conversion on feature vector representations of the nodes and the edges by using an initial parameter matrix, and performing multiple iteration calculations on the nodes and the edges by using a self-attention mechanism to calculate attention parameter vectors of correlations between the nodes and determine correlation sizes between the nodes; Performing column normalization on the attention parameter vectors of each node and edge, adjusting weights by using a function conversion, outputting feature vectors, aggregating the feature vectors, dynamically calculating weights of each node, and finally outputting an optimized graph structure to realize optimization of nodes in an enterprise application business scenario.

[0007] According to some embodiments, the present disclosure adopts the following technical solution: A graph structure construction module is configured to define internal attributes of nodes and edges of an enterprise application business network and construct a graph structure based on the internal attributes of the nodes and the edges; A node optimization module is configured to input the graph structure into a graph neural network model for linear conversion, perform vector conversion on feature vector representations of the nodes and the edges by using an initial parameter matrix, perform multiple iteration calculations on the nodes and the edges by using a self-attention mechanism to calculate attention parameter vectors of correlations between the nodes and determine correlation sizes between the nodes, perform column normalization on the attention parameter vectors of each node and edge, adjust weights by using a function conversion, output feature vectors, aggregate the feature vectors, dynamically calculate weights of each node, and finally output an optimized graph structure to realize optimization of nodes in an enterprise application business scenario.

[0008] According to some embodiments, the present disclosure adopts the following technical solution: A non-transitory computer readable storage medium is configured to store computer instructions, and the computer instructions are executed by a processor to implement the enterprise application complex business node optimization method based on a graph neural network.

[0009] According to some embodiments, the present disclosure adopts the following technical solution: An electronic device comprises a processor, a memory, and a computer program; the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the enterprise application complex business node optimization method based on a graph neural network.

[0010] Compared with the prior art, the beneficial effects of the present disclosure are: The enterprise application complex business node optimization method based on a graph neural network of the present disclosure constructs and describes business entities and relationships in the enterprise application field through a knowledge graph, and continuously iteratively optimizes the graph through a graph neural network model to realize a leap in the artificial intelligence cognitive system and a substantial improvement in the precision and accuracy of enterprise application artificial intelligence, and ultimately achieve business operation by users of the enterprise application in a natural language manner of artificial intelligence to realize a revolutionary innovation in the MVC architecture, user interface and use of traditional enterprise applications by intelligent native.

[0011] The enterprise application complex business node optimization method based on a graph neural network of the present disclosure innovatively uses a graph structure data abstraction definition and storage method to replace the linear data structure of the previous generation of relational database RDBMS, uses the natural logic relationship theory between node entities to replace the foreign key linear entity relationship of the previous generation, uses the Transformer mechanism of artificial intelligence LLM and the Attention attention principle to balance the weights of nodes and relationships, realizes the processing and optimization of non-Euclidean data structures through a neural network, and realizes accurate and efficient identification and extraction of key node relationships in a complex business graph structure. BRIEF DESCRIPTION OF DRAWINGS

[0012] The accompanying drawings, which form a part of the present disclosure, are intended to provide further understanding of the present disclosure, and the schematic embodiments of the present disclosure and the description thereof are intended to explain the present disclosure, and do not constitute improper limitations on the present disclosure.

[0013] Figure 1 The flowchart of the enterprise application complex business node optimization method based on a graph neural network of the present disclosure. DETAILED DESCRIPTION

[0014] The present disclosure will be further described below in conjunction with the accompanying drawings and embodiments.

[0015] It should be noted that the following detailed description is all exemplary and is intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the present disclosure belongs.

[0016] It should be noted that the terms used herein are only for the purpose of describing specific embodiments, and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form, and furthermore, it should be understood that when the terms "comprise" and / or "include" are used in the present specification, they indicate the presence of a feature, step, operation, device, component and / or combination thereof.

[0017] Embodiment 1 In an embodiment of the present disclosure, a graph neural network-based enterprise application complex business node optimization method is provided, and the method steps include: Step 1: define the internal attributes of the nodes and edges of the enterprise application business network, and build a graph structure based on the internal attributes of the nodes and edges; Step 2: input the graph structure into a graph neural network model for linear conversion, use an initial parameter matrix to perform vector conversion on the feature vector representation of the nodes and edges, and use a self-attention mechanism to perform graph neural network GNN multiple iteration calculation on the nodes and edges, calculate the attention parameter vector of the correlation between the nodes, and determine the correlation size between the nodes; Step 3: column normalize the attention parameter vector of each node and edge, adjust the weight using a function conversion, output the feature vector, aggregate the feature vectors, and dynamically calculate the weight of each node, and finally output the optimized graph structure to realize the optimization of the nodes in the enterprise application business scenario.

[0018] As an embodiment, the graph neural network-based enterprise application complex business node optimization method of the present disclosure replaces the table and foreign key technology relationship in the traditional relational database RDBMS with the nodes and edges of the graph semantic network rich in business semantics and n-n relationship network to empower the interaction mode between artificial intelligence natural language and enterprise applications, and then realizes the continuous optimization and evolution between nodes through the graph neural network GNN and the Transformer mechanism of generative artificial intelligence, ensures the accurate minimum set through the weight parameter determination of artificial intelligence learning in the business practice of massive nodes and edges, and thus enables the enterprise application to realize more efficient and accurate natural language and enterprise application interaction. The specific implementation process is as follows: Step 1: define the internal attributes of the nodes and edges of the enterprise application business network, and build a graph structure based on the internal attributes of the nodes and edges; Specifically, the nodes and edges in the business abstract semantic network graph are abstracted, and the internal attributes of the related nodes and edges are defined according to the enterprise application business scenario, including: abstracting the enterprise application business into nodes and edges in the semantic network graph, defining the internal attributes of the related nodes and edges according to the enterprise application business scenario, and building a graph structure based on the internal attributes of the nodes and edges. As an embodiment, an entity and edge related graph can be constructed through the description of entity and edge metadata. The actual graph data description can be described by OWL in the OWL standard, such as <#Source_Entity, Edge, #Target_Entity>.

[0019] Step 2: Construct the graph structure as the input of the graph neural network model, first perform embedding vectorization, vectorize each node and edge, and each node and edge is represented by a feature vector; Specifically, each entity and edge relationship description attribute is vectorized, and the related attributes are vectorized by the Embedding Model using the large model Transfomer mechanism. Vectorizing the 'description' attribute of each Entity entity and Edge edge relationship can fully utilize the large model Transfomer mechanism, and the related attributes are vectorized by the Embedding Model. The most accurate positioning of related nodes and edges can be achieved in the optimization process of the graph, so as to more effectively optimize the weight parameters Weight of the related nodes and edges.

[0020] Step 3: Input the graph structure into the graph neural network model for linear conversion, use the initial parameter matrix to perform vector conversion on the feature vector representation of the nodes and edges, and use the self-attention mechanism to perform multiple iteration calculations on the nodes and edges Graph neural network GNNLayer (graph neural network layer), calculate the attention parameter vector of the correlation between nodes, and determine the correlation size between nodes; Specifically, the self-attention mechanism of the Transformer in the generative artificial intelligence is used to perform multiple iteration calculations on the nodes and edges in the constructed graph, so as to calculate the attention parameter vector of the correlation between related nodes, and to confirm the correlation size between nodes in the input constructed graph, including: The graph neural network model performs linear conversion on the input vectorized graph structure through the weight matrix, and the latitude of the weight matrix depends on the latitude of the related nodes. If the latitude of the input graph node is x1, x2, x3, then the random matrix of the related weight is W=[a, b, c].

[0021] Further, the graph neural network model performs fitting of the graph neural network parameter weight in the calculation process of the attention parameter:

[0022] wherein, a represents a learnable change weight vector, and LeakyReLU is an activation function, h i and h m represents a transformation parameter, T is a transformation matrix, W is a random matrix.

[0023] Step 4: Column normalization is performed on the attention parameter vector of each node and edge, the weight is adjusted using the function conversion, the feature vector is output, the feature vectors are aggregated, and the weight of each node is dynamically calculated, and finally the optimized graph structure is output to realize the optimization of the nodes in the enterprise application business scenario.

[0024] Specifically, column normalization is performed on the attention parameter matrix of each node and edge, the sum of the attention parameter weights is equal to 1 by using function conversion, the graph neural network outputs an optimized graph, and the weight of each node is calculated according to the feature parameters of the adjacent nodes by using feature vector aggregation. The output graph optimized by the graph neural network GNN is vectorized by the same embedding vectorization model in the enterprise application scenario, and the most accurate and efficient node positioning and operation in the output graph are realized, including: Column normalization is performed on the attention parameter matrix of each node and edge, and the sum of the attention parameter weights is equal to 1 by using function conversion:

[0025] wherein, x m is the input graph node, K represents the neighbor node of the related node, is the weight of the neighbor node of the related node.

[0026] Further, the feature vector is output, the feature vectors are aggregated, and the weight of each node is calculated according to the feature parameters of the adjacent nodes:

[0027] wherein, σ represents an activation function, α im represents a SoftMax function, h m represents a conversion parameter, W is a random matrix.

[0028] Further, the output graph optimized by the graph neural network GNN is vectorized by the same embedding vectorization model in the enterprise application scenario, and the most accurate and efficient node positioning and operation in the output graph are realized.

[0029] The input graph that has undergone vectorization conversion and calculation, the enterprise inputs the correct Tuplet of the related system through verification and test phase, is vectorized through the related Embedding Model, and the existing graph neural network is continuously optimized and fitted, the extended neuron capability of the neural network is fully utilized, the optimization of the enterprise information data correlation is realized, and finally the high-precision output graph is realized, and the efficient and high-precision positioning of the user NLP to the graph is realized.

[0030] As an embodiment, taking "consumer", "consumer order" and relationship edge as an example, the related entity description includes consumer, retailer, sales order, purchase order, production order and bill of materials, and the graph structure is constructed according to the relationship between each entity, the internal attribute of the related node and edge.

[0031] The graph structure is taken as the input of the graph neural network model structure, each entity and edge relationship description attribute is vectorized, the related attribute is vectorized through the Embedding Model by using the large model Transfomer mechanism, the input vectorized graph structure is linearly converted through the weight matrix, the attention parameter calculation process is carried out by using the attention mechanism, the graph neural network parameter weight is fitted, the attention parameter matrix of each node and edge is calculated by column normalization, the sum of the attention parameter weight is equal to 1 by using the function conversion.

[0032] Then the feature vector is output, the feature vector is aggregated, the weight of each node is calculated according to the feature parameter of the adjacent node, and finally the optimized graph structure is output.

[0033] Embodiment 2 An embodiment of the present disclosure provides a graph neural network-based enterprise application complex business node optimization system, comprising: A graph structure construction module is configured to define the internal attributes of the nodes and edges of the enterprise application business network, and construct a graph structure based on the internal attributes of the nodes and edges. A node optimization module is configured to input the graph structure into a graph neural network model for linear conversion, vectorize the feature vector representation of the nodes and edges by using an initial parameter matrix, and perform multiple iteration calculation on the nodes and edges by using a self-attention mechanism, calculate the attention parameter vector of the correlation between the nodes, determine the correlation size between the nodes, perform column normalization on the attention parameter vector of each node and edge, adjust the weight by using a function conversion, output a feature vector, aggregate the feature vector, dynamically calculate the weight of each node, and finally output an optimized graph structure to realize the optimization of the nodes in the enterprise application business scenario.

[0034] As an embodiment, a graph neural network-based enterprise application complex business node optimization system specifically performs the following method steps: Step 1: define the internal attributes of the nodes and edges of the enterprise application business network, and build a graph structure based on the internal attributes of the nodes and edges; Specifically, abstract the nodes and edges in the business semantic network graph, and define the internal attributes of the related nodes and edges according to the enterprise application business scenario, including: abstracting the enterprise application business into nodes and edges in the semantic network graph, and defining the internal attributes of the related nodes and edges according to the enterprise application business scenario, and building a graph structure based on the internal attributes of the nodes and edges. As an embodiment, a related graph combining Entity entities and Edge edges can be constructed through the description of the metadata of entities and edges. The actual graph data description can be described by OWL in the OWL standard, such as <#Source_Entity, Edge, #Target_Entity>.

[0035] Step 2: build the graph structure as the input of the graph neural network model, first perform embedding vectorization, vectorize each node and edge, and each node and edge is represented by a feature vector; Specifically, vectorize each entity and edge relationship description attribute, and use the large model Transfomer mechanism to vectorize the related attributes through the Embedding Model. Vectorizing Embedding each Entity entity and Edge edge relationship 'description' attribute can fully utilize the large model Transfomer mechanism to vectorize the related attributes through the Embedding Model, which can most accurately locate the related nodes and edges in the graph optimization process, thereby more effectively optimizing the weight parameters Weight of the related nodes and edges.

[0036] Step 3: input the graph structure into the graph neural network model for linear conversion, use an initial parameter matrix to perform vector conversion on the feature vector representation of the nodes and edges, and use a self-attention mechanism to perform graph neural network GNN multiple iteration calculation on the nodes and edges, calculate the attention parameter vector of the correlation between nodes, and determine the correlation size between nodes; Specifically, use the self-attention mechanism of the Transformer in generative artificial intelligence to perform graph neural network GNN multiple iteration calculation on the nodes and edges in the constructed graph, thereby calculating the attention parameter vector of the correlation between related nodes, and confirming the correlation size between nodes in the input constructed graph, including: The graph neural network model performs linear conversion on the input vectorized graph structure through a weight matrix, the latitude of the weight matrix depends on the latitude of the relevant node, if the latitude of the input graph node is x1, x2, x3 three dimensions, then the random matrix of the relevant weight is also W =[a, b, c] three dimensions.

[0037] Further, in the calculation process of the attention parameter of the graph neural network model, the fitting of the graph neural network parameter weight is performed:

[0038] wherein, a represents a learnable changeable weight vector, LeakyReLU is an activation function, h i h m represents a conversion parameter, T is a transformation matrix, W is a random matrix.

[0039] Step 4: Column normalization is performed on the attention parameter vector of each node and edge, the weight is adjusted using the function conversion, the feature vector is output, the feature vector is aggregated, and the weight of each node is dynamically calculated, and finally the optimized graph structure is output, realizing the optimization of the nodes in the enterprise application business scenario.

[0040] Specifically, column normalization calculation is performed on the attention parameter matrix of each node and edge, the sum of the attention parameter weight is equal to 1 by using function conversion, the optimized graph is output by the graph neural network, and the weight of each node is calculated according to the feature parameters of the adjacent nodes by using the feature vector aggregation. The output graph optimized by the graph neural network GNN, the enterprise application scenario uses the same embedding vectorization model for vectorization, and realizes the most accurate and efficient node positioning and operation in the output graph, including: Column normalization calculation is performed on the attention parameter matrix of each node and edge, and the sum of the attention parameter weight is equal to 1 by using function conversion:

[0041] wherein, x m is an input graph node, K represents a neighbor node of the relevant node, is the weight of the neighbor node of the relevant node.

[0042] Further, the feature vector is output, the feature vector is aggregated, and the weight of each node is calculated according to the feature parameters of the adjacent nodes: ​

[0043] wherein, σ represents an activation function, α im represents a SoftMax function, h m represents a transformation parameter, W is a random matrix.

[0044] Further, the output graph optimized by the graph neural network GNN is vectorized by the same embedding vectorization model in the enterprise application scenario, and the most accurate and efficient node positioning and operation in the output graph are realized.

[0045] The input graph that has undergone vectorization conversion and calculation inputs the correct Tuplet of the related system through the verification and test stages, is vectorized by the related Embedding Model, and is continuously optimized and fitted in the existing graph neural network, fully utilizes the extended neuron capability of the neural network, realizes continuous optimization of enterprise information data correlation, and finally realizes a high-precision output graph and efficient and high-precision positioning of the user NLP to the graph.

[0046] Embodiment 3 In an embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the method for optimizing complex business nodes of enterprise applications based on a graph neural network.

[0047] Embodiment 4 In an embodiment of the present disclosure, an electronic device is provided, comprising a processor, a memory, and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the method for optimizing complex business nodes of enterprise applications based on a graph neural network.

[0048] The present disclosure is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1means for performing the function specified by the block or blocks.

[0049] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer-implemented process, so that the instructions executed on the computer or other programmable data processing devices provide a process for implementing the flowchart Figure 1 one or more flows and / or blocks Figure 1 steps of a function specified by the block or blocks.

[0050] Although the specific embodiments of the present disclosure are described above with reference to the drawings, the description is not a limitation on the scope of protection of the present disclosure, and those skilled in the art should understand that various modifications or changes made on the basis of the technical solutions of the present disclosure without creative labor are still within the scope of protection of the present disclosure.

Claims

1. A method for optimizing complex business nodes of enterprise applications based on graph neural networks, characterized in that, The method comprises the following steps: Defining internal attributes of nodes and edges of an enterprise application business network; Building a graph structure based on the internal attributes of the nodes and edges; Inputting the graph structure into a graph neural network model for linear conversion, using an initial parameter matrix to perform vector conversion on the feature vector representation of the nodes and edges, and using a self-attention mechanism to perform multiple iterations of graph neural network GNN calculation on the nodes and edges, calculate the attention parameter vector of the correlation between nodes, and determine the correlation size between each node; Column normalization is performed on the attention parameter vector of each node and edge, the weight is adjusted using a function conversion, the feature vector is output, the feature vectors are aggregated, and the weight of each node is dynamically calculated, and finally an optimized graph structure is output, realizing the optimization of the nodes in the enterprise application business scenario.

2. The graph neural network-based enterprise application complex business node optimization method of claim 1, wherein, The graph structure is built based on the internal attributes of the nodes and edges, comprising: abstracting the enterprise application business as nodes and edges in a semantic network graph, defining the internal attributes of the related nodes and edges according to the enterprise application business scenario, and building the graph structure based on the internal attributes of the nodes and edges.

3. The graph neural network-based enterprise application complex business node optimization method of claim 1, wherein, The inputting the graph structure into the graph neural network model for linear conversion comprises: the graph neural network model performing model linear conversion on the input vectorized graph structure through a weight matrix, the latitude of the weight matrix depending on the latitude of the relevant node, and if the latitude of the input graph node is x1, x2, x3, then the random matrix of the relevant weight is W =[a, b, c].

4. The graph neural network-based enterprise application complex business node optimization method of claim 1, wherein, The vector conversion of the feature vector representation of the nodes and edges using the initial parameter matrix comprises: vectorizing each entity and edge relationship description attribute, and using a large model Transfomer mechanism to vectorize the related attributes through an Embedding Model.

5. The graph neural network-based enterprise application complex business node optimization method of claim 1, wherein, The graph neural network GNN multiple iteration calculation on the nodes and edges using the self-attention mechanism comprises: fitting the graph neural network parameter weight in the calculation process of the attention parameter: wherein, a represents a learnable varying weight vector, LeakyReLU is an activation function, h i with h m represents a transformation parameter, T is a transformation matrix, W is a random matrix.

6. The graph neural network-based enterprise application complex business node optimization method of claim 1, wherein, The column normalization of the attention parameter vector of each node and edge comprises: column normalization calculation of the attention parameter matrix of each node and edge, and the sum of the attention parameter weight is equal to 1 through function conversion: wherein, x m is an input graph node, K denotes a neighbor node of the related node, is a weight of the neighbor node of the related node.

7. The graph neural network-based enterprise application complex business node optimization method of claim 1, wherein, The output of the feature vector and the aggregation of the feature vectors comprise: calculating the weight of each node according to the feature parameters of the adjacent nodes: wherein, The method comprises the following steps: represents an activation function, α im represents a SoftMax function, h m represents a transformation parameter, W is a random matrix.

8. The system for optimizing complex business nodes of enterprise applications based on graph neural networks, specifically performing the method for optimizing complex business nodes of enterprise applications based on graph neural networks as claimed in any one of claims 1-7, characterized in that, A graph structure construction module is configured to define internal attributes of nodes and edges of an enterprise application business network; Building a graph structure based on the internal attributes of the nodes and edges; A node optimization module is configured to input the graph structure into a graph neural network model for linear conversion, use an initial parameter matrix to perform vector conversion on the feature vector representation of the nodes and edges, and use a self-attention mechanism to perform multiple iterations of graph neural network GNN calculation on the nodes and edges, calculate the attention parameter vector of the correlation between nodes, and determine the correlation size between each node; column normalization is performed on the attention parameter vector of each node and edge, the weight is adjusted using a function conversion, the feature vector is output, the feature vectors are aggregated, and the weight of each node is dynamically calculated, and finally an optimized graph structure is output, realizing the optimization of the nodes in the enterprise application business scenario. ​ 9. A non-transitory computer-readable storage medium, comprising: The non-transitory computer readable storage medium is configured to store computer instructions, which, when executed by a processor, implement the method for optimizing complex business nodes of enterprise applications based on a graph neural network according to any one of claims 1-7.

10. An electronic device, comprising: Comprise: A processor, a memory, and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the method for optimizing complex business nodes of enterprise applications based on a graph neural network according to any one of claims 1-7.

Images