A Process Evaluation Method Based on Graph Neural Networks

Abstract

This invention provides a process evaluation method based on graph neural networks, comprising: constructing a planning flowchart for the current work based on user-uploaded work data; using graph neural networks to correct the graph operators corresponding to each process node, generating several logical loop conditions; performing logical analysis on the work data to obtain the node results corresponding to the process nodes; using graph neural networks to capture the process information contained in each node result to obtain the corresponding graph structure information; analyzing each cost input; establishing the overall graph structure information of the planning flowchart; using the bottleneck information and redundancy information contained in the overall graph structure information of the graph neural network to determine bottleneck process nodes and redundant process nodes; generating cost calibration suggestions; locating each bottleneck process node and each redundant process node in the planning flowchart; and generating node modification suggestions to help evaluate the rationality of the process design during the development and design phase.

Description

Technical Field

[0001] This invention relates to the field of transaction process management technology, and in particular to a process step evaluation method based on graph neural networks. Background Technology

[0002] With the arrival of the global digital age, businesses face new and greater challenges. To cope with global competitive pressures and improve their ability to handle various uncertainties, companies often need to dynamically adjust their development strategies and business scope based on market changes and demands. This necessitates changing, restructuring, and optimizing business processes to enhance competitiveness. Ensuring that business processes meet needs in the face of change and uncertainty has become a crucial condition for enterprise development and gaining a competitive advantage. Business process rationality assessment is one important evaluation indicator. Process rationality refers to the ability of a business to ensure task completion when facing threats and uncertainties. Process rationality directly reflects the level of success of business processes in the face of uncertain environments.

[0003] Currently, the only way to address this situation is to analyze the rationality of the flowchart based on the indicator data of each step after it has been running for a period of time. Obviously, this is inefficient, slow to update, and may even affect the stable operation of the enterprise's business.

[0004] Therefore, the present invention provides a process step evaluation method based on graph neural networks. Summary of the Invention

[0005] This invention provides a process step evaluation method based on graph neural networks, which helps to evaluate the rationality of process step design during the development and design phase.

[0006] This invention provides a process step evaluation method based on graph neural networks, comprising:

[0007] Step 1: Construct a planning flowchart for this task based on the work data uploaded by the user, and use a graph neural network to correct the graph operator corresponding to each process node in the planning flowchart;

[0008] Step 2: Generate several logical loop conditions for the planning flowchart based on the graph operator, and use the logical loop conditions to perform logical analysis on the working data to obtain the node results corresponding to the process nodes;

[0009] Step 3: Use the graph neural network to capture the process information contained in the result of each node, obtain the graph structure information corresponding to each process node, and analyze the cost input corresponding to each process node.

[0010] Step 4: Establish the overall graph structure information of the planning flowchart, and use a graph neural network to filter the bottleneck information and redundant information contained in the overall graph structure information to determine the bottleneck process nodes and redundant process nodes of the planning flowchart.

[0011] Step 5: Based on the cost input corresponding to each process node, generate cost calibration suggestions, locate each bottleneck process node and each redundant process node in the planning flowchart, and generate node modification suggestions.

[0012] In one feasible approach

[0013] Step 1 includes:

[0014] Step 11: Obtain the work data uploaded by the user, classify the work data to obtain the process data segment and execution data segment corresponding to this work, establish the matching relationship between the process data segment and the execution data segment, and generate several process steps for this work;

[0015] Step 12: Analyze the interaction relationship between different process data segments, generate corresponding interaction conditions, match the corresponding process steps for each interaction condition, deploy data to the response execution data segment using the interaction conditions, determine the data range corresponding to each process step, adjust the corresponding process steps using the data range, and generate a planning flowchart.

[0016] Step 13: Use the graph neural network to identify the point representation vector corresponding to each process node in the planning flowchart, and obtain the node attributes corresponding to each process node. Determine the interaction logic corresponding to the point representation vector based on the node attributes, and generate several interaction representation vectors for the planning flowchart.

[0017] Step 14: Determine the basic operator information of the corresponding process node based on the point representation vector, determine the depth operator information of the corresponding process node based on the interaction representation vector, match the corresponding operator element for the process node based on the basic operator information, match the corresponding operator condition for the operator element based on the depth operator information, and generate the graph operator corresponding to each process node.

[0018] In one feasible approach

[0019] Step 2 includes:

[0020] Step 21: Generate several flow paths of the planning flowchart based on the graph operator, divide each flow path into several unit stages, count the several unit stages corresponding to each flow node, and generate the logical loop condition corresponding to each flow node in the planning flowchart based on the flow path corresponding to each unit stage.

[0021] Step 22: Establish a logical model of the planning flowchart using the logical loop conditions, input the work data into the logical model for logical analysis, generate the process result corresponding to each process direction, and obtain the data change process of the work data in the logical model;

[0022] Step 23: Based on the data change process and the process results corresponding to each process flow, determine the data processing characteristics and data processing results corresponding to each unit stage in the planning flowchart, generate the node results of the corresponding process nodes, and display them.

[0023] In one feasible approach

[0024] Also includes:

[0025] Obtain the characteristics of the user-uploaded work target;

[0026] Filter the output node results corresponding to the output process nodes and generate process result features;

[0027] When the characteristics of the process result are inconsistent with the characteristics of the work objective, the planning flowchart is determined to be abnormal, an abnormality alert is generated and displayed.

[0028] In one feasible approach

[0029] Step 3 includes:

[0030] Step 31: Input each node result into the graph neural network, identify the local structure between different node results, generate the adjacency list of the planning flowchart by combining the presentation data corresponding to each node result, and capture the process information corresponding to each process node in the adjacency list.

[0031] Step 32: Establish the global structure of the planning flowchart based on the local structure between different node results, input the process information into the global structure to obtain the planning role of each process node in the planning flowchart, and establish the graph structure information of the corresponding process node based on the planning role.

[0032] Step 33: Determine the executable workload of the corresponding process node and the transferable workload between different process nodes based on the graph structure information, and determine the cost input corresponding to each process node.

[0033] In one feasible approach

[0034] Step 4 includes:

[0035] Step 41: Map the graph structure information corresponding to each process node to the planning flowchart respectively. Add corresponding external structures to each graph structure information according to the mapping results. Use the external structures to connect the graph structure information to generate the overall graph structure information of the planning flowchart.

[0036] Step 42: Use the graph neural network to traverse the overall graph structure information to obtain several process decision statements of the planning flowchart, generate the decision tree of the planning flowchart, analyze the decision execution information corresponding to each leaf node in the decision tree, and screen and determine the overloaded leaf nodes with bottleneck information and the isolated leaf nodes with redundant information.

[0037] Step 44: Obtain the overload graph structure information corresponding to the overload leaf node, construct the execution flow corresponding to the overload leaf node based on the overload graph structure information, filter the corresponding experimental data in the working data and input it into the execution flow to check whether the execution time of the corresponding overload leaf node is abnormal, and determine the process node corresponding to the overload leaf node as the bottleneck process node.

[0038] Step 45: Obtain the isolated graph structure information corresponding to the isolated node, identify the input execution object and output execution object of the corresponding isolated leaf node in the isolated graph structure information, and determine the process node corresponding to the isolated node as a redundant process node when the isolated leaf node does not contain an output execution object or does not contain an output execution object.

[0039] In one feasible approach

[0040] The process of filtering and identifying overloaded leaf nodes with bottleneck information and isolated leaf nodes with redundancy information includes:

[0041] Obtain the input-output relationship between different leaf nodes in the decision tree and establish several leaf node flows;

[0042] Obtain the last leaf node corresponding to each leaf node flow, and add an isolated identifier to each last leaf node;

[0043] Obtain the node input-output duration and the mean squared error duration corresponding to each leaf node flow, respectively. Identify the overloaded leaf nodes whose node input-output duration is higher than the corresponding mean squared error duration, and add a bottleneck identifier to each overloaded leaf node.

[0044] In one feasible approach

[0045] Step 5 includes:

[0046] Step 51: Obtain the workload characteristics corresponding to each process node, use the workload characteristics to evaluate the rationality of the cost investment corresponding to each process node, determine the first process node whose cost investment is higher than the corresponding usage cost, and the second process node whose cost investment is lower than the corresponding usage cost, and generate cost calibration suggestions.

[0047] Step 52: Locate each bottleneck process node and the redundant process node in the planning flowchart, obtain the diversion pressure corresponding to each bottleneck process node, and generate node modification suggestions based on the deletion suggestions for each redundant process node.

[0048] In one feasible approach

[0049] Also includes:

[0050] A scoring report for the planning flowchart is generated and displayed based on the cost calibration recommendations and the node modification recommendations.

[0051] In one feasible approach

[0052] Also includes:

[0053] The graph neural network is used to evaluate the functions of several nodes corresponding to each process node, and a function score report for each process node is generated and displayed.

[0054] The beneficial effects of the above technical solution are as follows: To create a flowchart that meets user needs, improve user work efficiency and quality, and reduce errors in automatically generated flowcharts, a planning flowchart is first built based on user-uploaded work data. A graph neural network is then used to correct the graph operators of each process node in the planning flowchart. The planning flowchart is further divided into several logical loop conditions, and the node results are determined through logical analysis. The graph neural network is then used again to capture the process information contained in each node result, thereby constructing the graph structure information of each process node. By constructing the graph structure information, the actual working structure of each process node can be visualized, providing an intuitive understanding of the work process of each process node, thus determining the cost input of each process node. Furthermore, by building the overall graph structure information of the planning flowchart, redundant and bottleneck information contained in the planning flowchart is analyzed. Finally, cost calibration suggestions and node modification suggestions for the planning flowchart are generated. In this way, each node in the planning flowchart can be analyzed, unreasonable arrangements in the planning flowchart can be identified, and corresponding solutions can be proposed. Users can adjust the planning flowchart according to the given solutions, which can quickly achieve the adjustment purpose and allow users to appropriately adjust the structure of the planning flowchart according to their own enterprise situation, achieving a balance between human and machine.

[0055] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0056] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0057] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0058] Figure 1 This is a schematic diagram of the workflow of a process evaluation method based on graph neural networks in an embodiment of the present invention.

[0059] Figure 2 This is a schematic diagram of step 2 of a process evaluation method based on graph neural networks in an embodiment of the present invention. Detailed Implementation

[0060] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0061] Example 1

[0062] This embodiment provides a process step evaluation method based on graph neural networks, such as... Figure 1 As shown, it includes:

[0063] Step 1: Construct a planning flowchart for this task based on the work data uploaded by the user, and use a graph neural network to correct the graph operator corresponding to each process node in the planning flowchart;

[0064] Step 2: Generate several logical loop conditions for the planning flowchart based on the graph operator, and use the logical loop conditions to perform logical analysis on the working data to obtain the node results corresponding to the process nodes;

[0065] Step 3: Use the graph neural network to capture the process information contained in the result of each node, obtain the graph structure information corresponding to each process node, and analyze the cost input corresponding to each process node.

[0066] Step 4: Establish the overall graph structure information of the planning flowchart, and use a graph neural network to filter the bottleneck information and redundant information contained in the overall graph structure information to determine the bottleneck process nodes and redundant process nodes of the planning flowchart.

[0067] Step 5: Based on the cost input corresponding to each process node, generate cost calibration suggestions, locate each bottleneck process node and each redundant process node in the planning flowchart, and generate node modification suggestions.

[0068] In this example, the working data refers to data uploaded by users regarding the operation of the enterprise's business;

[0069] In this example, the planning flowchart represents a flowchart generated by arranging the steps of this task in a flow;

[0070] In this example, the graph operator represents the conditions required for process nodes to perform different operations;

[0071] In this example, a logical loop condition represents a logical or cyclical execution condition in the planning flowchart;

[0072] In this example, the node results contain the information generated by the corresponding process node;

[0073] In this example, the graph structure information represents the structure of several process operations involved in a process node;

[0074] In this example, the overall diagram structure information represents the structure of all process nodes involved in the planning flowchart;

[0075] In this example, bottleneck information represents information generated when the progress of the planning flowchart is slow;

[0076] In this example, redundant information refers to information in the planning flowchart that is superfluous and has no specific meaning.

[0077] The working principle and beneficial effects of the above technical solution are as follows: To create a flowchart that meets user needs, improves user work efficiency and quality, and reduces errors in automatically generated flowcharts, a planning flowchart is first built based on user-uploaded work data. A graph neural network is then used to correct the graph operators of each process node in the planning flowchart. The planning flowchart is further divided into several logical loop conditions, and the node results are determined through logical analysis. Again, a graph neural network is used to capture the process information contained in each node result, thereby constructing the graph structure information of each process node. By constructing the graph structure information, the actual working structure of each process node can be visualized, providing an intuitive understanding of the work process of each process node, thus determining the cost input of each process node. Furthermore, by building the overall graph structure information of the planning flowchart, redundant and bottleneck information contained in the planning flowchart is analyzed. Finally, cost calibration suggestions and node modification suggestions for the planning flowchart are generated. In this way, each node in the planning flowchart can be analyzed, unreasonable arrangements in the planning flowchart can be identified, and corresponding solutions can be proposed. Users can adjust the planning flowchart according to the given solutions, which can quickly achieve the adjustment purpose, and users can also appropriately adjust the structure of the planning flowchart according to their own enterprise situation to achieve a balance between human and machine.

[0078] Example 2

[0079] Based on Example 1, the process evaluation method based on graph neural networks, step 1 includes:

[0080] Step 11: Obtain the work data uploaded by the user, classify the work data to obtain the process data segment and execution data segment corresponding to this work, establish the matching relationship between the process data segment and the execution data segment, and generate several process steps for this work;

[0081] Step 12: Analyze the interaction relationship between different process data segments, generate corresponding interaction conditions, match the corresponding process steps for each interaction condition, deploy data to the response execution data segment using the interaction conditions, determine the data range corresponding to each process step, adjust the corresponding process steps using the data range, and generate a planning flowchart.

[0082] Step 13: Use the graph neural network to identify the point representation vector corresponding to each process node in the planning flowchart, and obtain the node attributes corresponding to each process node. Determine the interaction logic corresponding to the point representation vector based on the node attributes, and generate several interaction representation vectors for the planning flowchart.

[0083] Step 14: Determine the basic operator information of the corresponding process node based on the point representation vector, determine the depth operator information of the corresponding process node based on the interaction representation vector, match the corresponding operator element for the process node based on the basic operator information, match the corresponding operator condition for the operator element based on the depth operator information, and generate the graph operator corresponding to each process node.

[0084] In this example, the process data segment represents the data related to the workflow within the work data;

[0085] In this example, the execution data segment represents the work data, which is always related to the execution of the work.

[0086] In this example, a process stage refers to a part of a process generated by combining an execution task with related data;

[0087] In this example, the point representation vector representation uses vectors to represent the results of tasks in the process nodes;

[0088] In this example, the interactive representation vector is a vector generated by combining the representation vectors of multiple points with the same direction in the planning flowchart, which is used to represent a logical result.

[0089] In this example, the basic operator information includes the location and task of the process node, while the depth operator information represents the interaction relationship between the process node and other process nodes.

[0090] In this example, the operator element represents the border corresponding to the process node, and the operator condition represents the conditions required for the operator element to perform its work.

[0091] The working principle and beneficial effects of the above technical solution are as follows: By classifying and matching the work data uploaded by users, several process steps of the current work are generated. Then, interaction conditions are constructed based on the interaction relationships between the work data. Furthermore, corresponding process steps and data ranges are matched for each interaction condition, thereby generating a planning flowchart. Then, the point representation vector corresponding to each process node in the planning flowchart and the interaction representation vector between different process nodes are identified in the graph neural network, thereby determining the basic operator information and depth operator information of each process node. Finally, corresponding operator elements and operator conditions are matched for each process node, generating the graph operator of each process node, laying the foundation for subsequent logical analysis.

[0092] Example 3

[0093] Based on Example 1, the process evaluation method based on graph neural networks, step 2, is as follows: Figure 2 As shown, it includes:

[0094] Step 21: Generate several flow paths of the planning flowchart based on the graph operator, divide each flow path into several unit stages, count the several unit stages corresponding to each flow node, and generate the logical loop condition corresponding to each flow node in the planning flowchart based on the flow path corresponding to each unit stage.

[0095] Step 22: Establish a logical model of the planning flowchart using the logical loop conditions, input the work data into the logical model for logical analysis, generate the process result corresponding to each process direction, and obtain the data change process of the work data in the logical model;

[0096] Step 23: Based on the data change process and the process results corresponding to each process flow, determine the data processing characteristics and data processing results corresponding to each unit stage in the planning flowchart, generate the node results of the corresponding process nodes, and display them.

[0097] In this example, the unit stage represents the process stage between two process nodes;

[0098] In this example, a process node can correspond to one or more logical loop conditions;

[0099] In this example, the data processing feature represents the characteristics of the data that needs to be processed in a unit stage;

[0100] In this example, the data processing result represents the result generated after the data is input into the unit stage and processed.

[0101] The working principle and beneficial effects of the above technical solution are as follows: Since each process node has a corresponding processing flow, it is necessary to perform corresponding analysis and processing on each process node. First, the flow direction of the planning flowchart is determined according to the graph operator. Then, it is divided into several unit stages, and the logical loop conditions of each process node of the planning flowchart are constructed. Furthermore, the logical loop conditions are used to establish the logical model of the planning flowchart. Through logical analysis, the flow result of each process direction is determined, and the data change process of the working data in the logical model is determined. Finally, the data processing characteristics and data processing results corresponding to each node are determined, and the node results of each process node are generated. In this way, multiple process nodes can be analyzed simultaneously. By using the model to simulate the working status of the process nodes, the actual calculation time can be reduced, and the efficiency and quality of node analysis can be improved.

[0102] Example 4

[0103] Based on Example 3, the process evaluation method based on graph neural networks further includes:

[0104] Obtain the characteristics of the user-uploaded work target;

[0105] Filter the output node results corresponding to the output process nodes and generate process result features;

[0106] When the characteristics of the process result are inconsistent with the characteristics of the work objective, the planning flowchart is determined to be abnormal, an abnormality alert is generated and displayed.

[0107] In this example, the output process node represents the process node in the planned process diagram that outputs the content to be executed.

[0108] The working principle and beneficial effects of the above technical solution are as follows: By comparing the characteristics of the work objectives uploaded by the user with the characteristics of the process results of the output process nodes, it can be determined whether a large number of errors have occurred in the planning process diagram, and corresponding reminders can be given.

[0109] Example 5

[0110] Based on Example 1, step 3 of the process evaluation method based on graph neural networks includes:

[0111] Step 31: Input each node result into the graph neural network, identify the local structure between different node results, generate the adjacency list of the planning flowchart by combining the presentation data corresponding to each node result, and capture the process information corresponding to each process node in the adjacency list.

[0112] Step 32: Establish the global structure of the planning flowchart based on the local structure between different node results, input the process information into the global structure to obtain the planning role of each process node in the planning flowchart, and establish the graph structure information of the corresponding process node based on the planning role.

[0113] Step 33: Determine the executable workload of the corresponding process node and the transferable workload between different process nodes based on the graph structure information, and determine the cost input corresponding to each process node.

[0114] In this example, a local structure represents a structure composed of two or more nodes;

[0115] In this example, the adjacency list is a data structure used to represent the relationships between different process nodes in the planning flowchart;

[0116] In this example, the global structure represents the structure composed of all process nodes in the planning flowchart;

[0117] In this example, the planning role represents the function of a process node in the planning flowchart;

[0118] In this example, the executable workload represents the maximum workload that a process node can handle when it performs its tasks.

[0119] In this example, transferable workload represents the maximum amount of work required to transfer data between process nodes.

[0120] The working principle and beneficial effects of the above technical solution are as follows: In order to perform cost analysis on each process node, an adjacency list is first established based on the local structure between node results. The process information corresponding to each process node is captured in the list. Then, by analyzing the global structure of the planning flowchart, the planning role of each process node in the planning flowchart is determined, thereby establishing corresponding graph structure information for the corresponding process node. Further analysis of the executable workload and transferable workload of each process node is conducted, thereby determining the cost input of each process node. In this way, the structure, workload, and structure of the planning flowchart of each process node can be analyzed, and the cost of the planning flowchart can be allocated to each process node, thus determining the cost input of each process node.

[0121] Example 6

[0122] Based on Example 1, the process evaluation method based on graph neural networks, step 4, includes:

[0123] Step 41: Map the graph structure information corresponding to each process node to the planning flowchart respectively. Add corresponding external structures to each graph structure information according to the mapping results. Use the external structures to connect the graph structure information to generate the overall graph structure information of the planning flowchart.

[0124] Step 42: Use the graph neural network to traverse the overall graph structure information to obtain several process decision statements of the planning flowchart, generate the decision tree of the planning flowchart, analyze the decision execution information corresponding to each leaf node in the decision tree, and screen and determine the overloaded leaf nodes with bottleneck information and the isolated leaf nodes with redundant information.

[0125] Step 44: Obtain the overload graph structure information corresponding to the overload leaf node, construct the execution flow corresponding to the overload leaf node based on the overload graph structure information, filter the corresponding experimental data in the working data and input it into the execution flow to check whether the execution time of the corresponding overload leaf node is abnormal, and determine the process node corresponding to the overload leaf node as the bottleneck process node.

[0126] Step 45: Obtain the isolated graph structure information corresponding to the isolated node, identify the input execution object and output execution object of the corresponding isolated leaf node in the isolated graph structure information, and determine the process node corresponding to the isolated node as a redundant process node when the isolated leaf node does not contain an output execution object or does not contain an output execution object.

[0127] In this example, the outer structure represents a graph structure that connects different process nodes;

[0128] In this example, the process decision statements represent the control statements corresponding to the execution of different tasks in the planning flowchart;

[0129] In this example, the process of checking whether the execution time of the corresponding overloaded leaf node is abnormal is as follows: check whether the execution time of the overloaded leaf node exceeds the average execution time of the planning flowchart.

[0130] The working principle and beneficial effects of the above technical solution are as follows: Since the planning flowchart may contain some unreasonable process nodes, the process decision statements contained in the planning flowchart are analyzed through graph structure information to construct a decision tree. Then, based on the decision execution information, overloaded leaf nodes and isolated leaf nodes are filtered out. The input-output relationship of isolated leaf nodes is determined by analysis to further determine whether they are redundant process nodes. And the duration is used to further determine whether overloaded leaf nodes are bottleneck process nodes. In this way, abnormal nodes in the planning flowchart can be quickly filtered out, and the basic work of evaluation is completed.

[0131] Example 7

[0132] Based on Example 6, the process of screening and determining overloaded leaf nodes with bottleneck information and isolated leaf nodes with redundancy information using a graph neural network-based process evaluation method includes:

[0133] Obtain the input-output relationship between different leaf nodes in the decision tree and establish several leaf node flows;

[0134] Obtain the last leaf node corresponding to each leaf node flow, and add an isolated identifier to each last leaf node;

[0135] Obtain the node input-output duration and the mean squared error duration corresponding to each leaf node flow, respectively. Identify the overloaded leaf nodes whose node input-output duration is higher than the corresponding mean squared error duration, and add a bottleneck identifier to each overloaded leaf node.

[0136] The working principle and beneficial effects of the above technical solution are as follows: By using the above method, leaf nodes with isolated features and overloaded leaf nodes with timeout features in the decision tree can be screened out and corresponding labels can be added to them, thereby achieving the purpose of accurate identification and classification labeling.

[0137] Example 8

[0138] Based on Example 1, the process evaluation method based on graph neural networks, step 5 includes:

[0139] Step 51: Obtain the workload characteristics corresponding to each process node, use the workload characteristics to evaluate the rationality of the cost investment corresponding to each process node, determine the first process node whose cost investment is higher than the corresponding usage cost, and the second process node whose cost investment is lower than the corresponding usage cost, and generate cost calibration suggestions.

[0140] Step 52: Locate each bottleneck process node and the redundant process node in the planning flowchart, obtain the diversion pressure corresponding to each bottleneck process node, and generate node modification suggestions based on the deletion suggestions for each redundant process node.

[0141] The working principle and beneficial effects of the above technical solution are as follows: Each process node is evaluated for its rationality, the process nodes with unreasonable costs are identified, and corresponding suggestions are configured for bottleneck process nodes and redundant process nodes. Cost calibration suggestions and node modification suggestions are generated for users to refer to.

[0142] Example 9

[0143] Based on Example 1, the process evaluation method based on graph neural networks further includes:

[0144] A scoring report for the planning flowchart is generated and displayed based on the cost calibration recommendations and the node modification recommendations.

[0145] The working principle and beneficial effects of the above technical solution are as follows: After making suggestions, the planning flowchart is scored, allowing users to intuitively understand the shortcomings of the planning flowchart.

[0146] Example 10

[0147] Based on Example 1, the process evaluation method based on graph neural networks further includes:

[0148] The graph neural network is used to evaluate the functions of several nodes corresponding to each process node, and a function score report for each process node is generated and displayed.

[0149] The working principle and beneficial effects of the above technical solution are as follows: by evaluating each process node separately, users can better understand the function and defects of each node.

[0150] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A process step evaluation method based on graph neural networks, characterized in that, include: Step 1: Construct a planning flowchart for this task based on the work data uploaded by the user, and use a graph neural network to correct the graph operator corresponding to each process node in the planning flowchart; Step 2: Generate several logical loop conditions for the planning flowchart based on the graph operator, and use the logical loop conditions to perform logical analysis on the working data to obtain the node results corresponding to the process nodes; Step 3: Use the graph neural network to capture the process information contained in the result of each node, obtain the graph structure information corresponding to each process node, and analyze the cost input corresponding to each process node. Step 4: Establish the overall graph structure information of the planning flowchart, and use a graph neural network to filter the bottleneck information and redundant information contained in the overall graph structure information to determine the bottleneck process nodes and redundant process nodes of the planning flowchart. Step 5: Based on the cost input corresponding to each process node, generate cost calibration suggestions, locate each bottleneck process node and each redundant process node in the planning flowchart, and generate node modification suggestions. Step 1 includes: Step 11: Obtain the work data uploaded by the user, classify the work data to obtain the process data segment and execution data segment corresponding to this work, establish the matching relationship between the process data segment and the execution data segment, and generate several process steps for this work; Step 12: Analyze the interaction relationship between different process data segments, generate corresponding interaction conditions, match the corresponding process steps for each interaction condition, deploy data to the response execution data segment using the interaction conditions, determine the data range corresponding to each process step, adjust the corresponding process steps using the data range, and generate a planning flowchart. Step 13: Use the graph neural network to identify the point representation vector corresponding to each process node in the planning flowchart, and obtain the node attributes corresponding to each process node. Determine the interaction logic corresponding to the point representation vector based on the node attributes, and generate several interaction representation vectors for the planning flowchart. Step 14: Determine the basic operator information of the corresponding process node based on the point representation vector, determine the depth operator information of the corresponding process node based on the interaction representation vector, match the corresponding operator element for the process node based on the basic operator information, match the corresponding operator condition for the operator element based on the depth operator information, and generate the graph operator corresponding to each process node.

2. The process evaluation method based on graph neural networks as described in claim 1, characterized in that, Step 2 includes: Step 21: Generate several flow paths of the planning flowchart based on the graph operator, divide each flow path into several unit stages, count the several unit stages corresponding to each flow node, and generate the logical loop condition corresponding to each flow node in the planning flowchart based on the flow path corresponding to each unit stage. Step 22: Establish a logical model of the planning flowchart using the logical loop conditions, input the work data into the logical model for logical analysis, generate the process result corresponding to each process direction, and obtain the data change process of the work data in the logical model; Step 23: Based on the data change process and the process results corresponding to each process flow, determine the data processing characteristics and data processing results corresponding to each unit stage in the planning flowchart, generate the node results of the corresponding process nodes, and display them.

3. The process evaluation method based on graph neural networks as described in claim 2, characterized in that, Also includes: Obtain the characteristics of the user-uploaded work target; Filter the output node results corresponding to the output process nodes and generate process result features; When the characteristics of the process result are inconsistent with the characteristics of the work objective, the planning flowchart is determined to be abnormal, an abnormality alert is generated and displayed.

4. The process evaluation method based on graph neural networks as described in claim 1, characterized in that, Step 3 includes: Step 31: Input each node result into the graph neural network, identify the local structure between different node results, generate the adjacency list of the planning flowchart by combining the presentation data corresponding to each node result, and capture the process information corresponding to each process node in the adjacency list. Step 32: Establish the global structure of the planning flowchart based on the local structure between different node results, input the process information into the global structure to obtain the planning role of each process node in the planning flowchart, and establish the graph structure information of the corresponding process node based on the planning role. Step 33: Determine the executable workload of the corresponding process node and the transferable workload between different process nodes based on the graph structure information, and determine the cost input corresponding to each process node.

5. The process evaluation method based on graph neural networks as described in claim 1, characterized in that, Step 4 includes: Step 41: Map the graph structure information corresponding to each process node to the planning flowchart respectively. Add corresponding external structures to each graph structure information according to the mapping results. Use the external structures to connect the graph structure information to generate the overall graph structure information of the planning flowchart. Step 42: Use the graph neural network to traverse the overall graph structure information to obtain several process decision statements of the planning flowchart, generate the decision tree of the planning flowchart, analyze the decision execution information corresponding to each leaf node in the decision tree, and screen and determine the overloaded leaf nodes with bottleneck information and the isolated leaf nodes with redundant information. Step 44: Obtain the overload graph structure information corresponding to the overload leaf node, construct the execution flow corresponding to the overload leaf node based on the overload graph structure information, filter the corresponding experimental data in the working data and input it into the execution flow to check whether the execution time of the corresponding overload leaf node is abnormal, and determine the process node corresponding to the overload leaf node as the bottleneck process node. Step 45: Obtain the isolated graph structure information corresponding to the isolated leaf node, identify the input execution object and output execution object of the corresponding isolated leaf node in the isolated graph structure information, and determine the process node corresponding to the isolated node as a redundant process node when the isolated leaf node does not contain an output execution object or does not contain an output execution object.

6. The process evaluation method based on graph neural networks as described in claim 5, characterized in that, The process of screening and identifying overloaded leaf nodes with bottleneck information and isolated leaf nodes with redundancy information includes: Obtain the input-output relationships between different leaf nodes in the decision tree, and establish several leaf node flows; Obtain the last leaf node corresponding to each leaf node flow, and add an isolated identifier to each last leaf node; Obtain the node input-output duration and the mean squared error duration corresponding to each leaf node flow, respectively. Identify the overloaded leaf nodes whose node input-output duration is higher than the corresponding mean squared error duration, and add a bottleneck identifier to each overloaded leaf node.

7. The process evaluation method based on graph neural networks as described in claim 1, characterized in that, Step 5 includes: Step 51: Obtain the workload characteristics corresponding to each process node, use the workload characteristics to evaluate the rationality of the cost investment corresponding to each process node, determine the first process node whose cost investment is higher than the corresponding usage cost, and the second process node whose cost investment is lower than the corresponding usage cost, and generate cost calibration suggestions. Step 52: Locate each bottleneck process node and the redundant process node in the planning flowchart, obtain the diversion pressure corresponding to each bottleneck process node, and generate node modification suggestions based on the deletion suggestions for each redundant process node.

8. The process evaluation method based on graph neural networks as described in claim 1, characterized in that, Also includes: A scoring report for the planning flowchart is generated and displayed based on the cost calibration recommendations and the node modification recommendations.

9. The process evaluation method based on graph neural networks as described in claim 1, characterized in that, Also includes: The graph neural network is used to evaluate the functions of several nodes corresponding to each process node, and a function score report for each process node is generated and displayed.

Images