Self-service-oriented intelligent operation and maintenance customer service response method

Abstract

The application provides a self-service-oriented intelligent operation and maintenance customer service response method, which comprises the following steps: (1) annotating fault data of an operation and maintenance system, extracting fault features, and obtaining a fault data relationship graph G; (2) inputting the fault features and the fault data relationship graph G into a graph neural network model for training; (3) when an operation and maintenance fault occurs, requiring an operation and maintenance customer service personnel to select the most possible fault type and the least possible fault type; (4) calculating a relationship graph G' of a current sample and training samples, correcting G', inputting the corrected G' and the current fault features into the trained graph neural network model, and outputting a fault type. The application has the beneficial effect that the operation of the operation and maintenance system is analyzed, a plurality of options are provided to the customer service personnel, the operation and maintenance system fault diagnosis is completed in a semi-automatic manner, the customer service response speed is greatly improved, and the accuracy of the operation and maintenance system fault diagnosis is also improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance, and mainly to an intelligent operation and maintenance customer service response method oriented towards self-service. Background Technology

[0002] Currently, intelligent operation and maintenance (O&M) systems are primarily responsible for the management and maintenance of target networks and computer systems, including project implementation, technical assistance, customer service and support; ensuring the normal operation of desktop computers, servers, internal and external networks, and related hardware and software devices. During the operation and work of O&M systems, while their intelligence and automation levels are becoming increasingly widespread, a portion of the work in most O&M systems is still primarily performed manually – this is customer service. When an O&M system malfunctions, most service providers need to send specialists to the site to investigate, troubleshoot, and determine solutions. Furthermore, existing methods manually define the similarity between data (including between faulty data and faulty data, between faulty data and normal data, and between normal data), without considering the aggregability of similar data, i.e., whether they can be connected to form a complete graph. Therefore, this patent proposes a semi-automated O&M customer service solution. Summary of the Invention

[0003] To address the problems in customer service of existing operation and maintenance systems, this invention provides an intelligent operation and maintenance customer service response method oriented towards self-service. The method analyzes the operation status of the operation and maintenance system, provides several options to operation and maintenance customer service personnel, and the system semi-automatically completes the fault diagnosis of the operation and maintenance system.

[0004] The objective of this invention is achieved through the following technical solution: An intelligent operation and maintenance customer service response method for self-service, comprising the following steps: (1) Label the fault data of the operation and maintenance system, extract fault features from the labeled fault data, and obtain the operation and maintenance fault data relationship graph G; (2) Training of the fault model of the operation and maintenance system: Input the fault feature and operation and maintenance fault data relationship graph G into the graph neural network model to train the graph neural network model; (3) Self-service customer service type selection in case of operation and maintenance failure: When an operation and maintenance failure occurs, the operation and maintenance system will pop up a window to remind the operation and maintenance customer service personnel, and at the same time, require the operation and maintenance customer service personnel to select the most likely failure type and the least likely failure type. (4) Semi-automated operation and maintenance fault detection: (4.1) Extract the fault features under the current operation and maintenance fault conditions, calculate them with the existing training samples, and calculate the relationship graph G' between the current sample and the training samples; (4.2) Based on the data of the most likely fault type and the least likely fault type, the relationship diagram G' is modified; (4.3) Input the corrected relation graph G' and the current fault features into the trained graph neural network model and output the fault type.

[0005] Furthermore, step (1) includes: (1.1) Collect fault data from the operation and maintenance system and label the fault data with type information; (1.2) Extract fault features from the labeled fault data, including the bandwidth at the time of system failure, the network speed at the time of system failure, the number of applications running at the time of system failure, the memory occupied at the time of system failure, the CPU utilization rate at the time of system failure, the average bandwidth when the system is running normally, and the maximum bandwidth when the system is running. (1.3) Generate an operation and maintenance fault data relationship diagram G based on the fault characteristics; The operation and maintenance fault data relationship graph G is defined as G=(V, E), where V={v1, v2, ..., vi, ..., vj, ..., vn} is the set of fault data nodes. Each fault data is considered as a node. vi represents the fault data at node i, vj represents the fault data at node j, and vn represents the fault data at node n. E={eij, ...} is the set of undirected edges. eij represents the undirected edge between node i and node j. If vi and vj are close, then i and j have a direct edge, and eij=1; otherwise, eij=0. Vi and vj are close if and only if the Euclidean distance between the eigenvectors of vi and vj is less than the average of the pairwise Euclidean distances between all nodes.

[0006] Furthermore, the correction method in step (4.2) includes: randomly selecting N% of the nodes from the data of the most likely fault types, and setting the two to have an edge regardless of whether there is an edge between the selected node and the current sample; randomly selecting M% of the nodes from the data of the least likely fault types, and setting the two to have no edge regardless of whether there is an edge between the selected node and the current sample, where N and M are variable parameters.

[0007] The beneficial effects of this invention are as follows: 1. This invention solves the problem that customer service work in existing operation and maintenance systems is all done manually. It analyzes the operation status of the operation and maintenance system and provides several options for operation and maintenance customer service personnel. The system completes the fault diagnosis of the operation and maintenance system in a semi-automatic manner. The invention incorporates a self-service manual assistance mechanism, which greatly improves the customer service response speed and helps to achieve better detection accuracy under the experience judgment of operation and maintenance customer service personnel.

[0008] 2. This invention uses a graph neural network method. The advantage of graph neural networks is that they attribute the features of each node's neighboring nodes to the current node, making the results more accurate and comprehensive. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the process of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings and embodiments: like Figure 1 As shown, this invention provides an intelligent operation and maintenance customer service response method for self-service, which includes the following steps: In a first aspect, the present invention provides a method for identifying fault data in an operation and maintenance system, specifically including: (1) Collect fault data of the operation and maintenance system and label the fault data with type information. It can be labeled as Class I, Class II, etc., or as A, B, C, etc. It is determined manually by the operation and maintenance personnel. This patent does not make any special requirements. (2) Extract fault features from the labeled fault data, including but not limited to: bandwidth at the time of system failure, network speed at the time of system failure, number of applications running at the time of system failure, memory occupied at the time of system failure, CPU utilization at the time of system failure, average bandwidth during normal system operation, maximum bandwidth during system operation, etc. These vectors constitute feature vector I=x1,x2,x3…xm (which means there are m-dimensional features).

[0011] (3) Generate an operation and maintenance fault data relationship diagram G based on the fault characteristics; The operation and maintenance fault data relationship graph G is defined as G=(V, E), where V={v1, v2, ..., vi, ..., vj, ..., vn} is the set of fault data nodes. Each fault data is considered as a node. vi represents the fault data of node i, vj represents the fault data of node j, and vn represents the fault data of node n. E={eij, ...} is the set of undirected edges. eij represents the undirected edge between node i and node j. If vi and vj are close, then node i and node j have a direct edge (eij=1), otherwise node i and node j have no edge (eij=0).

[0012] Here, vi and vj are similar if and only if the Euclidean distance between the eigenvectors of vi and vj is less than the average of the pairwise Euclidean distances between all nodes.

[0013] The calculation of Euclidean distance is a well-known problem in the field, and this method does not provide a special definition.

[0014] Secondly, this invention provides a method for training a fault model for an operation and maintenance system, specifically including: The graph G showing the relationship between the labeled fault data and the operation and maintenance fault data is input into the graph neural network model to train the graph neural network model. Graph Neural Networks (GNNs) are a well-known method in the field of artificial intelligence. Common GNN models include GCN, GraphSage, and GAT, which will not be defined here. The model needs to be input with a sample set and a graph of sample relationships. The format of the samples in the sample set is [label, ID number, feature]. Among them, the label is as shown in the first aspect (1); the number is the sample sequence number, the feature is as shown in the first aspect (2), and the sample relationship diagram is as shown in the first aspect (3).

[0015] Thirdly, this invention provides a method for selecting self-service customer service types in the event of an operational or maintenance failure, specifically including: When an operational failure occurs, a pop-up notification will appear for the operations and maintenance (O&M) support staff. The notification will include, but is not limited to, a description of the on-site failure, the bandwidth at the time of the failure, the network speed at the time of the failure, and the number of applications running at the time of the failure. The O&M support staff will also be asked to select the most likely failure type (multiple selections are allowed) and the least likely failure type.

[0016] Fourthly, this invention provides a semi-automated operation and maintenance fault detection method, specifically including: (1) Extract the features of the operation and maintenance system under the current fault environment according to the first aspect (2), calculate the relationship graph G' between the current sample and the training sample, and construct the relationship graph G' in the same way as the operation and maintenance fault data relationship graph G.

[0017] (2) According to the third aspect, the relationship graph G' is modified as follows: From the data (nodes) of the most likely fault type, 60% of the nodes are randomly selected, and regardless of whether there is an edge between the selected node and the node in the current sample, the two are set to have an edge. From the data (nodes) of the least likely fault type, 60% of the nodes are randomly selected, and regardless of whether there is an edge between the selected node and the node in the current sample, the two are set to have no edge.

[0018] (3) Input the corrected relation graph G' and the current fault features into the trained graph neural network model, and output the fault type. This is provided by the graph neural network model, and this patent can directly input it without special explanation.

[0019] Fifthly, based on the output results, the fault type is obtained, and the appropriate measures can be selected based on the fault type.

[0020] It is understood that, for those skilled in the art, any equivalent substitutions or modifications to the technical solutions and inventive concepts of this invention should fall within the scope of protection of the appended claims.

Claims

1. A smart operation and maintenance customer service response method for self-service, characterized in that: The method includes the following steps: (1) Label the fault data of the operation and maintenance system, extract fault features from the labeled fault data, and obtain the operation and maintenance fault data relationship graph G; (2) Training of the fault model of the operation and maintenance system: Input the fault feature and operation and maintenance fault data relationship graph G into the graph neural network model to train the graph neural network model; (3) Self-service customer service type selection in case of operation and maintenance failure: When an operation and maintenance failure occurs, the operation and maintenance system will pop up a window to remind the operation and maintenance customer service personnel, and at the same time, require the operation and maintenance customer service personnel to select the most likely failure type and the least likely failure type. (4) Semi-automated operation and maintenance fault detection: (4.1) Extract the fault features under the current operation and maintenance fault conditions, calculate them with the existing training samples, and calculate the relationship graph G' between the current sample and the training samples; (4.2) Based on the data of the most likely fault type and the least likely fault type, the relationship diagram G' is modified; (4.3) Input the corrected relation graph G' and the current fault features into the trained graph neural network model, and output the fault type; Step (1) includes: (1.1) Collect fault data from the operation and maintenance system and label the fault data with type information; (1.2) Extract fault features from the labeled fault data, including the bandwidth at the time of system failure, the network speed at the time of system failure, the number of applications running at the time of system failure, the memory occupied at the time of system failure, the CPU utilization rate at the time of system failure, the average bandwidth when the system is running normally, and the maximum bandwidth when the system is running. (1.3) Generate an operation and maintenance fault data relationship diagram G based on the fault characteristics; The operation and maintenance fault data relationship graph G is defined as G=(V,E), where V={v1,v2, …vi,…vj, …vn} is the set of fault data nodes, vi represents fault data at node i, vj represents fault data at node j, and vn represents fault data at node n; E={eij,…} is the set of undirected edges, eij represents the undirected edge between node i and node j. If vi and vj are close, then node i and node j have a direct edge, i.e., eij=1; otherwise, eij=0. Vi and vj are close if the Euclidean distance between the eigenvectors of vi and vj is less than the average of the pairwise Euclidean distances between all nodes. The correction method for step (4.2) is as follows: randomly select N% of the nodes from the data of the most likely fault types, and set the two nodes to have an edge regardless of whether there is an edge between the selected node and the current sample; randomly select M% of the nodes from the data of the least likely fault types, and set the two nodes to have no edge regardless of whether there is an edge between the selected node and the current sample, where N and M are variable parameters.

2. The intelligent operation and maintenance customer service response method for self-service as described in claim 1, characterized in that: The graph neural network models include GCN, GraphSage, and GAT.

Images