Similar work order searching method, device and nonvolatile computer readable storage medium

Abstract

The present disclosure relates to a similar work order retrieval method and device, and a non-volatile computer readable storage medium, and relates to the technical field of computers. The method comprises: constructing a first work order graph structure of a first work order and a second work order graph structure of a second work order; connecting the first work order graph structure and the second work order graph structure to construct a third work order graph structure; determining a vector representation matrix corresponding to the third work order graph structure; and calculating the similarity between the first work order and the second work order according to the vector representation matrix corresponding to the third work order graph structure. The technical solution of the present disclosure can calculate the similarity between two work orders based on the graph structure, accurately find similar work orders, and reduce the difficulty of finding similar work orders.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a similar work order retrieval method, a similar work order retrieval device, and a non-volatile computer-readable storage medium. Background Technology

[0002] Similar work orders refer to work orders with the same fault. If we can retrieve some historical similar work orders for a newly pushed work order, we can provide the operation and maintenance personnel with solutions, reduce the "difficulty" of processing the newly pushed work order, and increase the work efficiency of the operation and maintenance personnel.

[0003] In related technologies, maintenance personnel search for similar work orders based on textual information such as work order type and work order number recorded in the fault work order. This search is time-consuming, laborious, and inaccurate. Summary of the Invention

[0004] In view of this, this disclosure proposes a similar work order retrieval method, which can calculate the similarity between two work orders based on a graph structure, accurately find similar work orders, and reduce the difficulty of finding similar work orders.

[0005] According to some embodiments of this disclosure, a similar work order retrieval method is provided, including: constructing a first work order graph structure for a first work order and a second work order graph structure for a second work order; connecting the first work order graph structure and the second work order graph structure to construct a third work order graph structure; determining the vector representation matrix corresponding to the third work order graph structure; and calculating the similarity between the first work order and the second work order based on the vector representation matrix corresponding to the third work order graph structure.

[0006] In some embodiments, where the edges of the third work order graph structure have weights, determining the vector representation matrix corresponding to the third work order graph structure includes: walking through each node in the third work order graph structure according to the weights of each edge of the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure; grouping the neighboring nodes according to the type of the neighboring nodes of each node, grouping neighboring nodes of the same type into one group; encoding one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node; determining a second feature of each node according to the weights of each edge of the third work order graph structure and the first vector representation of each node; aggregating the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node; and aggregating the second features of each node and the second vector representation of each group of each node to obtain a third vector representation of each node, wherein the third vector representations of each node in the third work order graph structure constitute the vector representation matrix corresponding to the third work order graph structure.

[0007] In some embodiments, the step of walking through each node in the third work order graph structure according to the weights of each edge of the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure includes: calculating the transition probability of each node in the third work order graph structure according to the weights of the edges in the third work order graph structure, and walking through the node according to the transition probability starting from any node in the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure.

[0008] In some embodiments, encoding one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node includes: pre-training different types of first features of each node using different methods; inputting different types of third features obtained after pre-training into different feature transformation layers; inputting the fourth features output by each feature transformation layer into a Bi-LSTM model, encoding the fourth features to obtain a fifth feature; and performing pooling processing on each fifth feature to obtain a first vector representation of each node.

[0009] In some embodiments, determining the second feature of each node based on the weights of each edge of the third work order graph structure and the first vector representation of each node includes: inputting the first vector representation of each node into a Bi-LSTM model to obtain the sixth feature of each node; and weighting the sixth feature of each node using the weights of the edges between each node and its neighboring nodes to obtain the second feature of each node.

[0010] In some embodiments, the aggregation of the second features of each neighboring node in the same group of each node to obtain the second vector representation of each group of each node includes: pooling the second features of each neighboring node in the same group of each node to obtain the second vector representation of each group of each node.

[0011] In some embodiments, the aggregation of the second feature of each node and the second vector representation of each group of each node to obtain the third vector representation of each node includes: determining the aggregation coefficient of each node and its group using an attention mechanism based on the second feature of each node and the second vector representation of each group of each node; and aggregating the second feature of each node and the second vector representation of each group of each node based on the aggregation coefficient of each node and its group to obtain the third vector representation of each node.

[0012] In some embodiments, constructing the first work order graph structure of the first work order and the second work order graph structure of the second work order includes: the method of constructing the work order graph structure is as follows: establishing connection relationships between information items in the work order that have causal relationships; determining the weight of the connection relationship based on the causal relationship value of the information items with connection relationships; and establishing connection relationships between information items in the work order that have physical relationships or record relationships; wherein the work order includes the first work order or the second work order, and the work order graph structure includes the first work order graph structure or the second work order graph structure.

[0013] In some embodiments, calculating the similarity between the first work order and the second work order based on the vector representation matrix corresponding to the third work order graph structure includes: inputting the vector representation matrix of the third work order graph structure into a graph convolutional neural network (GCN) to obtain the similarity between the first work order and the second work order.

[0014] In some embodiments, the method further includes: if the first work order and the second work order are determined to be similar based on similarity, the first work order is fault-handled according to the fault-handling method of the second work order.

[0015] According to some other embodiments of this disclosure, a similar work order retrieval device is provided, comprising: a construction module configured to construct a first work order graph structure for a first work order and a second work order graph structure for a second work order; a connection module configured to connect the first work order graph structure and the second work order graph structure to construct a third work order graph structure; a determination module configured to determine a vector representation matrix corresponding to the third work order graph structure; and a calculation module configured to calculate the similarity between the first work order and the second work order based on the vector representation matrix corresponding to the third work order graph structure.

[0016] In some embodiments, the edges of the third work order graph structure have weights, and the determining module is configured to: walk through each node in the third work order graph structure according to the weights of each edge of the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure; group the neighboring nodes according to the type of the neighboring nodes of each node, and group neighboring nodes of the same type into one group; encode one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node; determine a second feature of each node according to the weights of each edge of the third work order graph structure and the first vector representation of each node; aggregate the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node; and aggregate the second features of each node and the second vector representation of each group of each node to obtain a third vector representation of each node, wherein the third vector representations of each node in the third work order graph structure constitute the vector representation matrix corresponding to the third work order graph structure.

[0017] In some embodiments, the determining module is configured to calculate the transition probability of each node in the third work order graph structure based on the weight of the edges in the third work order graph structure, and to perform a walk based on the transition probability starting from any node in the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure.

[0018] In some embodiments, the determining module is configured to: pre-train different types of first features of each node using different methods; input different types of third features obtained after pre-training into different feature transformation layers; input the fourth features output by each feature transformation layer into a Bi-LSTM model, encode the fourth features to obtain a fifth feature; and perform pooling processing on each fifth feature to obtain a first vector representation of each node.

[0019] In some embodiments, the determining module is configured to: input a first vector representation of each node into a Bi-LSTM model to obtain a sixth feature of each node; and weight the sixth feature of each node using the weights of the edges between each node and its neighboring nodes to obtain a second feature of each node.

[0020] In some embodiments, the determining module is configured to perform pooling on the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node.

[0021] In some embodiments, the determining module is configured to: determine the aggregation coefficients of each node and its group based on the second feature of each node and the second vector representation of each group of each node using an attention mechanism; and aggregate the second feature of each node and the second vector representation of each group of each node based on the aggregation coefficients of each node and its group to obtain a third vector representation of each node.

[0022] In some embodiments, the construction module is configured to establish a connection relationship between information items in the work order that have a causal relationship; determine the weight of the connection relationship based on the causal relationship value of the information items with the connection relationship; and establish a connection relationship between information items in the work order that have a physical relationship or a record relationship; wherein the work order includes the first work order or the second work order, and the work order graph structure includes the first work order graph structure or the second work order graph structure.

[0023] In some embodiments, the calculation module is configured to input the vector representation matrix of the third work order graph structure into a graph convolutional neural network (GCN) to obtain the similarity between the first work order and the second work order.

[0024] In some embodiments, if the first work order and the second work order are determined to be similar based on similarity, the first work order is fault-handled according to the fault-handling method of the second work order.

[0025] According to further embodiments of this disclosure, a similar work order retrieval device is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the similar work order retrieval method of any of the above embodiments based on instructions stored in the memory.

[0026] According to further embodiments of the present disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the similar work order retrieval method of any of the above embodiments.

[0027] In the above embodiments, by constructing a work order graph structure, a vector representation matrix of the work order graph structure is obtained. The similarity between work orders is calculated using this vector matrix, enabling accurate identification of similar work orders and reducing the difficulty of finding them. Furthermore, by retrieving processed historical work orders with high similarity to the work order to be processed, a solution is provided for maintenance personnel to handle the pending work orders, reducing the difficulty of processing them and improving the efficiency and accuracy of processing. Attached Figure Description

[0028] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.

[0029] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein:

[0030] Figure 1 Flowcharts illustrating some embodiments of the similar work order retrieval method of this disclosure are shown.

[0031] Figure 2 A schematic diagram showing some embodiments of the work order diagram structure of this disclosure is provided.

[0032] Figure 3 Flowcharts illustrating some embodiments of the method for determining the vector representation matrix corresponding to the third work order diagram structure of this disclosure are shown.

[0033] Figure 4 Schematic diagrams illustrating some embodiments of the present disclosure for calculating the similarity between a first work order and a second work order;

[0034] Figure 5 Block diagrams illustrating some embodiments of the similar work order retrieval device of this disclosure;

[0035] Figure 6 Block diagrams illustrating other embodiments of the similar work order retrieval device of this disclosure are shown. Detailed Implementation

[0036] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0037] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0038] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0039] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0040] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0041] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0042] This disclosure provides a method for retrieving similar work orders. It constructs a work order graph structure, obtains a vector representation matrix of the graph structure, and calculates the similarity between work orders using this matrix. This method accurately identifies similar work orders, reducing the difficulty of finding them. By retrieving processed historical work orders with high similarity to the work order to be processed, it provides maintenance personnel with a solution for handling pending work orders, reducing the difficulty of processing them and improving their efficiency and accuracy.

[0043] Figure 1 Flowcharts illustrating some embodiments of the similar work order retrieval method of this disclosure are shown.

[0044] like Figure 1 As shown, the similar work order retrieval method includes steps 110-140, and may also include step 150 if necessary.

[0045] In step 110, the first work order diagram structure of the first work order and the second work order diagram structure of the second work order are constructed.

[0046] In some embodiments, the method for constructing a work order graph structure includes: establishing connections between information items in a work order that have causal relationships; determining the weight of the connections based on the causal relationship values ​​of the information items with connections; and establishing connections between information items in a work order that have physical or record relationships. The work order includes either the first work order or the second work order, and the work order graph structure includes either the first work order graph structure or the second work order graph structure.

[0047] Figure 2 A schematic diagram showing some embodiments of the work order diagram structure of this disclosure is provided.

[0048] In some embodiments, such as Figure 2 As shown, the work order diagram structure includes the work order's valid information, which includes the work order fault description, fault category, fault subcategory, and alarms (alarm type, alarm network element, and alarm port).

[0049] In some embodiments, such as Figure 2 As shown, the nodes and relationships that constitute the work order diagram structure include: information items with causal relationships, such as alarm-alarm (causal relationship); information items with record relationships, such as work order-fault category, work order-alarm; and information items with physical relationships, such as alarm-network element, network element-network element (physical connection), etc.

[0050] In some embodiments, information items in the work order that have causal relationships, record relationships, and physical relationships are embedded as nodes, and the association relationships between information items are embedded as edges into the work order graph structure.

[0051] In some embodiments, the Node2vector algorithm is used to embed alarm type nodes into the work order graph structure.

[0052] In some embodiments, determining the weight of a connection relationship based on the causal relationship value of information items with a connection relationship includes: calculating the causal relationship value between information items with a causal relationship in the work order, and using it as the weight of the edge between nodes with a causal relationship in the work order graph structure, so that the work order graph structure becomes a weighted graph.

[0053] In some embodiments, the PrefixSpan (Prefix-Projected Pattern Growth) algorithm is used to calculate the causal relationship value between alarm types, which is used as the weight of the edge between alarm types in the work order graph structure.

[0054] In some embodiments, the key to constructing the work order diagram structure is establishing the causal relationships between alarm types. For example, alarm types can be port down, CPU (Central Processing Unit) failure, and OTN (Optical Transport Network) failure.

[0055] In some embodiments, for example, if port down and CPU failure always occur together, then port down and CPU failure are causally related. If the probability of them occurring together is 0.96, then the causal relationship between port down and CPU failure is 0.96.

[0056] In some embodiments, for example, if the probability of port DOWN and OTN failures occurring together is very small, then port DOWN and OTN failures are not causally related. If the probability of them occurring together is 0.01, then the causal relationship between port DOWN and OTN failures is 0.01.

[0057] Furthermore, for edges between information items in a work order that have physical or record relationships, the weight is one if the edge exists, and zero if the edge does not exist.

[0058] In some embodiments, during the construction of the work order graph structure, a causal relationship graph is first constructed based on causal relationships. Then, other relationships, such as record relationships and physical relationships, are added to the causal relationship graph to obtain the work order graph structure. The construction process of the causal relationship graph includes initially constructing a causal relationship graph for the alarm type using the PrefixSpan algorithm and the Node2vector algorithm, and this graph should be a directed graph. Then, an unsupervised training process is performed on the causal relationship graph for the alarm type using a GCNN (Graph Convolutional Neural Network) model to obtain the final causal relationship graph for the alarm type.

[0059] In step 120, the first work order diagram structure and the second work order diagram structure are connected to construct the third work order diagram structure.

[0060] In some embodiments, a third work order diagram structure is obtained by connecting the first work order diagram structure and the second work order diagram structure according to their association.

[0061] This association is, for example, such that if a node in the first work order diagram structure is the same as a node in the second work order diagram structure, then the first and second work order diagram structures can be connected through that node to obtain the third work order diagram structure. For example... Figure 2 As shown, this node can be, for example, a fault category, a fault subcategory, an alarm type, a network element, or a network element port.

[0062] In step 130, the vector representation matrix corresponding to the third work order diagram structure is determined.

[0063] Figure 3 Flowcharts illustrating some embodiments of the method for determining the vector representation matrix corresponding to the third work order diagram structure of this disclosure are shown.

[0064] like Figure 3 As shown, the method for determining the vector representation matrix corresponding to the third work order graph structure includes steps 310-360. The edges of the third work order graph structure have weights.

[0065] In some embodiments, a graph embedding algorithm is used to calculate the third work order graph structure in order to determine the vector representation matrix corresponding to the third work order graph structure.

[0066] In step 310, each node in the third work order graph structure is walked according to the weight of each edge of the third work order graph structure in order to sample the neighboring nodes of each node in the third work order graph structure.

[0067] In some embodiments, the node to be traversed includes several neighboring nodes. There is a connection between the node before the traversal and the node to be traversed. If there is no connection between the node before the traversal and the node to be traversed, the traversal will not proceed.

[0068] In some embodiments, the softmax algorithm is used to calculate the transition probability and to walk through each node in the third work order graph structure. The calculation formula is as follows:

[0069] E(n i )={j∈{1,2,...,u}|e i，j}

[0070] W(n i )={j∈{1,2,...,u}|W i，j}

[0071] P(n i ) = softmax(W(n i ))={j∈{1,2,...,u}|p i，j}

[0072] In some embodiments, n i Indicates the node to be visited; e i，j Indicates n i Edges with head nodes; E(n) i ) represents all n i Edges with head nodes; w i，j e i，j The corresponding weights; W(n) i ) represents all e i，j The corresponding weight; p i，j e i，j The transition probability; P(n) i ) represents all e i，j The transition probability.

[0073] In some embodiments, a walk begins from any node in the third work order graph structure and proceeds according to the transition probability. During the walk, the neighboring nodes of the node are repeatedly visited or the node is returned with a certain probability. The walk stops after a fixed number of neighboring nodes have been collected.

[0074] In some embodiments, the transition probability of each node in the third work order graph structure is calculated based on the weight of the edges in the third work order graph structure, and a walk is performed starting from any node in the third work order graph structure according to the transition probability to sample the neighboring nodes of each node in the third work order graph structure.

[0075] In some embodiments, for a certain type of neighboring nodes of a sampled node, the neighboring node of that type that ranks highest in frequency is selected as the neighboring node of that type of node.

[0076] In the above embodiments, the nodes in the third work order graph structure are walked according to the transition probability, which makes the sampling more accurate.

[0077] In step 320, the neighboring nodes are grouped according to the type of the neighboring nodes of each node, and neighboring nodes of the same type are grouped together.

[0078] In step 330, one or more first features of each node in the third work order graph structure are encoded to obtain a first vector representation of each node.

[0079] In some embodiments, step 330 includes steps (1)-(4).

[0080] (1) Different methods are used to pre-train the first features of different types of each node.

[0081] In some embodiments, each node may include different types of first features, such as images, text, and categories.

[0082] In some embodiments, different methods are used for pre-training of different types of first features. For example, one-hot encoding is used for categorical features, the par2vec method is used for text features, and CNN (Convolutional Neural Network) is used for image features.

[0083] (2) Input the different types of third features obtained after pre-training into different feature transformation layers.

[0084] In some embodiments, different types of third features obtained after pre-training are input into different feature transformation layers to unify the output dimension. Feature transformation layers are, for example, identity transformations or fully connected layers.

[0085] (3) Input the fourth feature output by each feature transformation layer into the Bi-LSTM (Bi-directional Long Short-Term Memory) model to encode the fourth feature and obtain the fifth feature.

[0086] In some embodiments, the fourth feature output by each feature transformation layer is input into the Bi-LSTM model to encode the fourth feature, thereby enhancing its expressive power and obtaining the fifth feature.

[0087] (4) Pool each fifth feature to obtain the first vector representation of each node.

[0088] In some embodiments, each fifth feature is pooled, for example, by average pooling, to obtain a first vector representation of each node.

[0089] In step 340, the second feature of each node is determined based on the weights of each edge of the third work order graph structure and the first vector representation of each node.

[0090] In some embodiments, step 340 includes steps (1)-(2).

[0091] (1) Input the first vector representation of each node into the Bi-LSTM model to obtain the sixth feature of each node. The calculation formula is as follows:

[0092] H=BiLSTM(X), X={i∈{1,...,n}|x i}

[0093] H = {i ∈ {1, ..., n} | h i}

[0094] Where, x i h is the first vector representation of each. i It is the sixth feature of each node.

[0095] (2) Using the weights of the edges between each node and its neighboring nodes, the sixth feature of each node is weighted to obtain the second feature of each node. The calculation formula is as follows:

[0096] F = {i∈{1,...,n}|f i}

[0097] f i =w i ×h i

[0098] Among them, f iw represents the second feature of each node. i This represents the weight of the edges between each node and its neighboring nodes.

[0099] In step 350, the second features of each neighboring node in the same group of each node are aggregated to obtain the second vector representation of each group of each node.

[0100] In some embodiments, the second features of each neighboring node in the same group of each node are pooled, for example, by average pooling, to obtain a second vector representation of each group of each node.

[0101] In step 360, the second feature of each node and the second vector representation of each group of each node are aggregated to obtain the third vector representation of each node. The third vector representations of each node in the third work order graph structure constitute the vector representation matrix corresponding to the third work order graph structure.

[0102] In some embodiments, step 360 includes steps (1)-(2).

[0103] (1) Based on the second feature of each node and the second vector representation of each group of neighbors of each node, the aggregation coefficient of each node and its group is determined by using an attention mechanism.

[0104] The aggregation coefficient of each group of neighboring nodes is determined based on the attention weight of a certain group of neighboring nodes to that node.

[0105] (2) Aggregate the second feature of each node and the second vector representation of each group of each node according to the aggregation coefficient of each node and its group to obtain the third vector representation of each node. The aggregation method includes, for example, weighted summation.

[0106] In some embodiments, the model for determining the vector representation matrix corresponding to the third work order graph structure is optimized and trained in an unsupervised manner.

[0107] In step 140, the similarity between the first work order and the second work order is calculated based on the vector representation matrix corresponding to the third work order diagram structure.

[0108] In some embodiments, the vector representation matrix of the third work order graph structure is input into a GCN (Graph Convolutional Network) to obtain the similarity between the first work order and the second work order.

[0109] Figure 4 Schematic diagrams illustrating some embodiments of the present disclosure for calculating the similarity between a first work order and a second work order.

[0110] like Figure 4 As shown, the GCN network includes hidden layers, fully connected layers (fc layers), and ReLU activation functions. The first and second work order graph structures are connected to obtain a third work order graph structure. The vector representation matrix corresponding to the third work order graph structure is input into the GCN network. After processing through hidden layers, ReLU activation functions, and fully connected layers, the similarity between the first and second work orders is finally obtained.

[0111] In step 150, let the first work order represent the work order to be processed and the second work order represent the historical work order that has been processed. If the first work order and the second work order are similar based on similarity, the first work order is processed according to the fault handling method of the second work order.

[0112] In some embodiments, based on the similarity between the first work order and the second work order, a specified number (one or more) of second work orders with a high similarity to the first work order are recommended to the maintenance personnel.

[0113] In the above embodiments, by constructing a work order graph structure, a vector representation matrix of the work order graph structure is obtained. The similarity between work orders is calculated using this vector matrix, enabling accurate identification of similar work orders and reducing the difficulty of finding them. Furthermore, by retrieving processed historical work orders with high similarity to the work order to be processed, a solution is provided for maintenance personnel to handle the pending work orders, reducing the difficulty of processing them and improving the efficiency and accuracy of processing.

[0114] Figure 5 Block diagrams illustrating some embodiments of the similar work order retrieval device of this disclosure are shown.

[0115] like Figure 5 As shown, the similar work order retrieval device 50 includes a construction module 51, a connection module 52, a determination module 53, and a calculation module 54, and may also include a processing module 55 as needed.

[0116] Module 51 is configured to construct the first work order diagram structure for the first work order and the second work order diagram structure for the second work order.

[0117] In some embodiments, the construction module 51 is further configured to: establish connection relationships between information items in the work order that have causal relationships; determine the weight of the connection relationship based on the causal relationship value of the information items with connection relationships; and establish connection relationships between information items in the work order that have physical relationships or record relationships. The work order includes the first work order or the second work order, and the work order graph structure includes the first work order graph structure or the second work order graph structure.

[0118] The connection module 52 is configured to connect the first work order diagram structure and the second work order diagram structure to construct a third work order diagram structure.

[0119] The determination module 53 is configured to determine the vector representation matrix corresponding to the third work order diagram structure.

[0120] In some embodiments, the edges of the third work order graph structure have weights.

[0121] In some embodiments, the determining module 53 is further configured to: walk through each node in the third work order graph structure according to the weights of each edge of the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure; group the neighboring nodes according to the type of the neighboring nodes of each node, and group neighboring nodes of the same type into one group; encode one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node; determine a second feature of each node according to the weights of each edge of the third work order graph structure and the first vector representation of each node; aggregate the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node; and aggregate the second features of each node and the second vector representation of each group of each node to obtain a third vector representation of each node, wherein the third vector representations of each node in the third work order graph structure constitute the vector representation matrix corresponding to the third work order graph structure.

[0122] In some embodiments, the determining module 53 is further configured to calculate the transition probability of each node in the third work order graph structure based on the weight of the edges in the third work order graph structure, and to perform a walk based on the transition probability starting from any node in the third work order graph structure to sample the neighboring nodes of each node in the third work order graph structure.

[0123] In some embodiments, the determining module 53 is further configured to: pre-train different types of first features of each node using different methods; input different types of third features obtained after pre-training into different feature transformation layers; input the fourth features output by each feature transformation layer into a Bi-LSTM model, encode the fourth features to obtain fifth features; and perform pooling processing on each fifth feature to obtain a first vector representation of each node.

[0124] In some embodiments, the determining module 53 is further configured to: input the first vector representation of each node into the Bi-LSTM model to obtain the sixth feature of each node; and use the weights of the edges between each node and its neighboring nodes to weight the sixth feature of each node to obtain the second feature of each node.

[0125] In some embodiments, the determining module 53 is further configured to perform pooling processing on the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node.

[0126] In some embodiments, the determining module 53 is further configured to: determine the aggregation coefficients of each node and its group based on the second feature of each node and the second vector representation of each group of each node using an attention mechanism; and aggregate the second feature of each node and the second vector representation of each group of each node based on the aggregation coefficients of each node and its group to obtain a third vector representation of each node.

[0127] The calculation module 54 is configured to calculate the similarity between the first work order and the second work order based on the vector representation matrix corresponding to the third work order graph structure.

[0128] In some embodiments, the calculation module 54 is further configured to input the vector representation matrix of the third work order graph structure into a graph convolutional neural network (GCN) to obtain the similarity between the first work order and the second work order.

[0129] The processing module 55 is configured to set a first work order to represent a work order to be processed and a second work order to represent a processed historical work order. If the first work order and the second work order are similar based on similarity, the first work order is processed according to the fault handling method of the second work order.

[0130] In the above embodiments, by constructing a work order graph structure, a vector representation matrix of the work order graph structure is obtained. The similarity between work orders is calculated using this vector matrix, enabling accurate identification of similar work orders and reducing the difficulty of finding them. Furthermore, by retrieving processed historical work orders with high similarity to the work order to be processed, a solution is provided for maintenance personnel to handle the pending work orders, reducing the difficulty of processing them and improving the efficiency and accuracy of processing.

[0131] Figure 6 Block diagrams illustrating other embodiments of the similar work order retrieval device of this disclosure are shown.

[0132] like Figure 6As shown, the similar work order retrieval device 60 of this embodiment includes: a memory 61 and a processor 62 coupled to the memory 61. The processor 62 is configured to execute the similar work order retrieval method in any of the foregoing embodiments based on instructions stored in the memory 61.

[0133] The memory 61 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory stores, for example, the operating system, application programs, the boot loader, and other programs.

[0134] The similar work order retrieval device 60 may also include an input / output interface 63, a network interface 64, and a storage interface 65. These interfaces 63, 64, and 65, as well as the memory 61 and processor 62, can be connected, for example, via a bus 66. The input / output interface 63 provides a connection interface for input / output devices such as monitors, mice, keyboards, touchscreens, microphones, and speakers. The network interface 64 provides a connection interface for various networked devices. The storage interface 65 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0135] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0136] This concludes the detailed description of the similar work order retrieval method, apparatus, and non-volatile computer-readable storage medium according to this disclosure. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art will fully understand how to implement the technical solutions disclosed herein based on the above description.

[0137] The methods and systems of this disclosure may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of this disclosure are not limited to the specific order described above unless otherwise specifically stated. Furthermore, in some embodiments, this disclosure may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the methods according to this disclosure. Thus, this disclosure also covers recording media storing programs for performing the methods according to this disclosure.

[0138] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A method for retrieving similar work orders, comprising: Construct the first work order diagram structure for the first work order and the second work order diagram structure for the second work order; Connect the first work order diagram structure and the second work order diagram structure to construct the third work order diagram structure; Determining the vector representation matrix corresponding to the third work order graph structure includes: walking through each node in the third work order graph structure according to the weights of each edge of the third work order graph structure to sample the neighboring nodes of each node; grouping the neighboring nodes according to the type of each node, grouping neighboring nodes of the same type into one group; encoding one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node; determining a second feature of each node according to the weights of each edge of the third work order graph structure and the first vector representation of each node; aggregating the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node; and aggregating the second features of each node and the second vector representation of each group of each node to obtain a third vector representation of each node, wherein the third vector representations of each node in the third work order graph structure constitute the vector representation matrix corresponding to the third work order graph structure, wherein the edges of the third work order graph structure have weights; and The similarity between the first work order and the second work order is calculated based on the vector representation matrix corresponding to the third work order graph structure.

2. The similar work order retrieval method according to claim 1, wherein, The step of walking through each node in the third work order graph structure according to the weights of each edge of the third work order graph structure, and sampling the neighbor nodes of each node in the third work order graph structure, includes: Based on the weights of the edges in the third work order graph structure, the transition probability of each node in the third work order graph structure is calculated. Starting from any node in the third work order graph structure, a walk is performed according to the transition probability to sample the neighboring nodes of each node in the third work order graph structure.

3. The similar work order retrieval method according to claim 1, wherein, Encoding one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node includes: Different methods are used to pre-train the first features of different types for each node; Different types of third features obtained after pre-training are input into different feature transformation layers; The fourth feature output from each feature transformation layer is input into the Bi-LSTM model, and the fourth feature is encoded to obtain the fifth feature; and Pooling is performed on each of the fifth features to obtain the first vector representation of each node.

4. The similar work order retrieval method according to any one of claims 1-3, wherein, The step of determining the second feature of each node based on the weights of each edge of the third work order graph structure and the first vector representation of each node includes: The first vector representation of each node is input into the Bi-LSTM model to obtain the sixth feature of each node; and By using the weights of the edges between each node and its neighboring nodes, the sixth feature of each node is weighted to obtain the second feature of each node.

5. The similar work order retrieval method according to any one of claims 1-3, wherein, The aggregation of the second features of each neighboring node in the same group of each node to obtain the second vector representation of each group of each node includes: The second features of each neighboring node in the same group of each node are pooled to obtain the second vector representation of each group of each node.

6. The similar work order retrieval method according to any one of claims 1-3, wherein, The aggregation of the second feature of each node and the second vector representation of each group of each node to obtain the third vector representation of each node includes: Based on the second feature of each node and the second vector representation of each group of each node, an attention mechanism is used to determine the aggregation coefficients of each node and its group; and The second feature of each node and the second vector representation of each group of each node are aggregated based on the aggregation coefficients of each node and each group to obtain the third vector representation of each node.

7. The similar work order retrieval method according to any one of claims 1-3, wherein, The construction of the first work order diagram structure for the first work order and the second work order diagram structure for the second work order includes: The method for constructing the work order diagram structure is as follows: Establish connections between information items in the work order that have causal relationships; The weights of the connections are determined based on the causal relationship values ​​of the information items with connection relationships; and Establish connections between information items in the work order that have physical or record relationships; The work order includes either the first work order or the second work order, and the work order diagram structure includes either the first work order diagram structure or the second work order diagram structure.

8. The similar work order retrieval method according to any one of claims 1-3, wherein, The step of calculating the similarity between the first work order and the second work order based on the vector representation matrix corresponding to the third work order diagram structure includes: The vector representation matrix of the third work order graph structure is input into the graph convolutional neural network (GCN) to obtain the similarity between the first work order and the second work order.

9. The similar work order retrieval method according to any one of claims 1-3 further includes: If the first work order and the second work order are similar based on similarity, the first work order is processed according to the fault handling method of the second work order.

10. A similar work order retrieval device, comprising: The construction module is configured to construct the first work order diagram structure for the first work order and the second work order diagram structure for the second work order. The connection module is configured to connect the first work order diagram structure and the second work order diagram structure to construct a third work order diagram structure; The determining module is configured to determine the vector representation matrix corresponding to the third work order graph structure, including: traversing each node in the third work order graph structure according to the weights of each edge of the third work order graph structure to sample the neighboring nodes of each node; grouping the neighboring nodes according to the type of each node, grouping neighboring nodes of the same type into one group; encoding one or more first features of each node in the third work order graph structure to obtain a first vector representation of each node; determining a second feature of each node according to the weights of each edge of the third work order graph structure and the first vector representation of each node; aggregating the second features of each neighboring node in the same group of each node to obtain a second vector representation of each group of each node; and aggregating the second features of each node and the second vector representation of each group of each node to obtain a third vector representation of each node, wherein the third vector representations of each node in the third work order graph structure constitute the vector representation matrix corresponding to the third work order graph structure, wherein the edges of the third work order graph structure have weights; and The calculation module is configured to calculate the similarity between the first work order and the second work order based on the vector representation matrix corresponding to the third work order graph structure.

11. The similar work order retrieval device according to claim 10, wherein, The determining module is configured to calculate the transition probability of each node in the third work order graph structure based on the weight of the edges in the third work order graph structure, and to perform a walk based on the transition probability starting from any node in the third work order graph structure in order to sample the neighboring nodes of each node in the third work order graph structure.

12. A similar work order retrieval device, comprising: Memory; and A processor coupled to the memory, the processor being configured to execute the similar work order retrieval method of any one of claims 1-9 based on instructions stored in the memory.

13. A non-volatile computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the similar work order retrieval method according to any one of claims 1-9.

Images