Power customer appeal work order classification method, system and device and medium

Abstract

The invention discloses a power customer appeal work order classification method, system and device and a medium, mainly relates to the technical field of work order classification, and is used for solving the problems of insufficient feature extraction and poor classification effect in a power work order classification process. Comprising the following steps: preprocessing obtained customer appeal work order data, and removing abnormal data; performing serialization operation on the processed work order data to obtain a serialized work order text vector; inputting the serialized work order text vector into a feature relationship extraction model to obtain a feature relationship in the work order data; obtaining sequence information from the feature relationship by using an attention mechanism; and inputting the sequence information into a softmax function to obtain a classification label corresponding to the customer appeal work order data.

Description

Technical Field

[0001] The present application relates to the technical field of work order classification, and in particular to a method, system, device and medium for classifying work orders requested by power customers. Background Art

[0002] With the rapid development of the power industry and the increasing diversification of power customer needs, the power customer service system, as an important window for power supply companies to communicate with power customers, bears the important responsibility of handling customer demands and solving power problems. As a direct basis for reflecting customer power demand and service quality, the effective analysis and processing of power customer demand work order data is of great significance for improving customer satisfaction, optimizing service processes, and reducing operating costs.

[0003] In recent years, with the rapid progress of deep learning technology, especially in the field of natural language processing, a new path has been opened up for the analysis of power customer complaint work orders. First, deep text mining is carried out on power customer complaint work orders, and the sentence structure is accurately divided with the help of advanced word segmentation technology. Then, word frequency statistics and TF-IDF algorithm are used to scientifically calculate the importance weight of each keyword, and high-weight keywords are selected as representatives of the core information of the text. Then, word cloud analysis technology is used to intuitively display these key information, making the analysis results clear at a glance.

[0004] However, the current classification method is still simple and crude, facing bottleneck problems such as incomplete feature extraction and low classification accuracy, which makes it difficult to accurately classify power work orders during processing and the classification effect is poor, affecting service efficiency and customer satisfaction. Therefore, there is an urgent need for a power customer demand work order classification method, system, equipment and medium to solve the problem of insufficient feature extraction and poor classification effect in the power work order classification process. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the present application provides a method, system, device and medium for classifying electric power customer demand work orders to solve the problems of insufficient feature extraction and poor classification effect in the process of electric power work order classification.

[0006] In the first aspect, the present application provides a method for classifying electric power customer complaint work orders, the method comprising: preprocessing the obtained customer complaint work order data to remove abnormal data; serializing the processed work order data to obtain a serialized work order text vector; inputting the serialized work order text vector into a feature relationship extraction model to obtain feature relationships in the work order data; using an attention mechanism to obtain sequence information from the feature relationships; and inputting the sequence information into a softmax function to obtain a classification label corresponding to the customer complaint work order data.

[0007] Furthermore, the processed work order data is serialized to obtain a serialized work order text vector, which specifically includes: Get the processed work order data, add a preset start identifier at the starting position; input the work order data into the preset word segmenter, and divide the work order data into single-word sequences; input the single-word sequence into the preset model to obtain the output data. The preset model formula is as follows: , Where D is the input word sequence, and are preset vector weights and bias parameters respectively, GELU is the activation function, and the activation function formula is , for The cumulative function of the Gaussian normal distribution is: , in, Refers to random variables The probability of is the mean, is the standard deviation, is the base of natural logarithms.

[0008] Furthermore, the serialized work order text vector is input into the feature relationship extraction model to obtain the feature relationship in the work order data, specifically including: Use bidirectional LSTM to capture the forward and backward output hidden states of each ticket text vector in the text; By formula: , get the output hidden state of time step t; where, represents the forward output hidden state, represents the backward output hidden state, represents the previous forward output hidden state, Represents the previous backward output hidden state; By formula: , obtain the output vector H representing the feature relationship; where n represents the preset total number of time steps.

[0009] Furthermore, the attention mechanism is used to obtain sequence information from feature relationships, including: By formula: , , , , Calculate sequence information ;in, represents the activation function that maps the vector to -1 ~ 1, , represents the output vector of the feature relationship, X represents the preset balance matrix, Represents the preset attention weight matrix.

[0010] Furthermore, the sequence information is input into the softmax function to obtain the classification labels corresponding to the customer request work order data, including: By formula: , calculate the conditional probability on each classification label, and then determine the classification label with a conditional probability greater than a preset threshold as the classification label corresponding to the customer request work order data; where, is the preset weight parameter, is the bias parameter.

[0011] On the second aspect, the present application provides a power customer complaint work order classification system, the system including: a removal module, used to pre-process the obtained customer complaint work order data and remove abnormal data; an acquisition module, used to serialize the processed work order data to obtain a serialized work order text vector; input the serialized work order text vector into a feature relationship extraction model to obtain the feature relationship in the work order data; use the attention mechanism to obtain sequence information from the feature relationship; input the sequence information into the softmax function to obtain the classification label corresponding to the customer complaint work order data.

[0012] Furthermore, the obtaining module comprises a first obtaining unit, Used to obtain the processed work order data and add a preset start identifier at the starting position; Input the work order data into the preset word segmenter to segment the work order data into single word sequences; Input the single word sequence into the hidden layer of the preset model to obtain output data; Input the output data into the Transformer encoder to obtain the serialized work order text vector.

[0013] Furthermore, the obtaining module includes a second obtaining unit, Used to use bidirectional LSTM to capture the forward and backward output hidden states of each work order text vector in the text; By formula: , get the output hidden state of time step t; where, represents the forward output hidden state, represents the backward output hidden state, represents the previous forward output hidden state, Represents the previous backward output hidden state; By formula: , obtain the output vector H representing the feature relationship; where n represents the preset total number of time steps.

[0014] In a third aspect, the present application provides a device for classifying electricity customer complaint work orders, the device comprising: a processor; and a memory having executable code stored thereon, and when the executable code is executed, the processor executes a method for classifying electricity customer complaint work orders such as any one of the above items.

[0015] In a fourth aspect, the present application provides a non-volatile computer storage medium having computer instructions stored thereon, which when executed implement a method for classifying work orders for power customer demands as described in any one of the above items.

[0016] Those skilled in the art can understand that the present application has at least the following beneficial effects: This application combines the attention mechanism with the feature relationship extraction model to better capture the key information and context dependencies in the power customer demand work order, thereby improving the accuracy of classification and solving the problem of insufficient feature extraction and poor classification effect in the power work order classification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Some embodiments of the present disclosure are described below with reference to the accompanying drawings, in which: Figure 1 It is a flow chart of a method for classifying work orders for power customer demands provided in an embodiment of the present application.

[0018] Figure 2 It is a schematic diagram of the internal structure of an electricity customer demand work order classification system provided in an embodiment of the present application.

[0019] Figure 3 It is a schematic diagram of the internal structure of an electricity customer demand work order classification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] It should be understood by those skilled in the art that the embodiments described below are only preferred embodiments of the present disclosure, and do not mean that the present disclosure can only be implemented through the preferred embodiments. The preferred embodiments are only used to explain the technical principles of the present disclosure, and are not used to limit the protection scope of the present disclosure. Based on the preferred embodiments provided by the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work should still fall within the protection scope of the present disclosure.

[0021] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0022] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0023] The present application embodiment provides a method for classifying electric power customer demand work orders, such as Figure 1 As shown, the method provided in the embodiment of the present application mainly includes the following steps: Step 110: pre-process the obtained customer request work order data to remove abnormal data.

[0024] It should be noted that the customer demand work order data is obtained and pre-processed, including cleaning blank work orders and defect work orders, removing stop words, and deleting meaningless modal particles and symbols.

[0025] Step 120: perform serialization operation on the processed work order data to obtain a serialized work order text vector.

[0026] This step can be specifically as follows: Get the processed work order data, add a preset start identifier at the starting position; input the work order data into the preset word segmenter, and divide the work order data into single-word sequences; input the single-word sequence into the preset model to obtain the output data. The preset model formula is as follows: , Where D is the input word sequence, and are preset vector weights and bias parameters respectively. When initializing the weight matrix, random initialization is usually used to make full use of the network's adaptive ability. When initializing the bias vector, a zero vector is usually used for initialization. GELU is the activation function, and its formula is , for The cumulative function of the Gaussian normal distribution is: , in, Refers to random variables The probability of is the center position of the data and the average value of all data points. The formula is: ,in represents the value of each data point, Represents the number of data points, i represents the i-th data point, and i∈[1,n]; is the standard deviation, calculated as: , i represents the i-th data point. It should be noted that the preset start identifier may be [CLS].

[0027] Step 130: Input the serialized work order text vector into the feature relationship extraction model to obtain the feature relationship in the work order data; and use the attention mechanism to obtain sequence information from the feature relationship.

[0028] The serialized work order text vector is input into the feature relationship extraction model to obtain the feature relationship in the work order data, which can be specifically: A bidirectional LSTM is used to capture the forward and backward output hidden states of each ticket text vector in the text.

[0029] By formula: , Get the output hidden state at time step t; where, represents the forward output hidden state, represents the backward output hidden state, represents the previous forward output hidden state, Represents the previous backward output hidden state.

[0030] By formula: , Obtain an output vector H representing the feature relationship, wherein n represents the preset total number of time steps.

[0031] Among them, the attention mechanism is used to obtain sequence information from feature relationships, which can be specifically: By formula: , , , , Calculate sequence information ;in, represents the activation function that maps the vector to -1 ~ 1, , The output vector represents the feature relationship. For example, the input of the Bi-LSTM layer is a one-dimensional vector [1 5 3 8 10] containing 5 words. The word embedding dimension is set to 128, the hidden layer dimension is set to 256, and the final output H is [-0.0171 0.0285 -0.1338 0.0147 0.0305]. The H value in the above formula is consistent; X represents the preset balance matrix, using the Xavier / Glorot initialization strategy. The specific formula is: , represents a normal distribution, are the number of input units and output units in the weight tensor respectively. For the one-dimensional vector [1 5 3 8 10] in the above example, the initialization value of X can be [0.35655944 -0.46668932 0.13031723 0.51798067-0.28582504]. Represents the preset attention weight matrix, whose value is given by Calculated.

[0032] Step 140: Input the sequence information into the softmax function to obtain the classification label corresponding to the customer request work order data.

[0033] This step can be specifically as follows: By formula: , Calculate the conditional probability on each classification label, and then determine the classification label with a conditional probability greater than a preset threshold as the classification label corresponding to the customer request work order data; is the preset weight parameter, initialized to a random number matrix with a mean of 0, is the bias parameter, which is initially set to 0.

[0034] besides, Figure 2 A power customer demand work order classification system is provided in an embodiment of the present application. Figure 2 As shown, the system provided in the embodiment of the present application mainly includes: The removal module 210 is used to pre-process the obtained customer demand work order data and remove abnormal data.

[0035] The acquisition module 220 is used to serialize the processed work order data to obtain a serialized work order text vector; input the serialized work order text vector into the feature relationship extraction model to obtain the feature relationship in the work order data; use the attention mechanism to obtain sequence information from the feature relationship; input the sequence information into the softmax function to obtain the classification label corresponding to the customer demand work order data.

[0036] The acquisition module 220 includes a first acquisition unit, which is used to obtain the processed work order data, add a preset start identifier at the starting position; input the work order data into a preset word segmenter, and divide the work order data into single-word sequences; input the single-word sequence into a hidden layer of a preset model to obtain output data; input the output data into a Transformer encoder to obtain a serialized work order text vector.

[0037] The acquisition module 220 includes a second acquisition unit for capturing the forward and backward output hidden states of each work order text vector in the text using a bidirectional LSTM.

[0038] By formula: , Get the output hidden state at time step t; where, represents the forward output hidden state, represents the backward output hidden state, represents the previous forward output hidden state, Represents the previous backward output hidden state.

[0039] By formula: , Obtain an output vector H representing the feature relationship, wherein n represents the preset total number of time steps.

[0040] The above is a method embodiment of the present application. Based on the same inventive concept, the present application embodiment also provides a power customer demand work order classification device. Figure 3 As shown, the device includes: a processor; and a memory, on which executable codes are stored. When the executable codes are executed, the processor executes a method for classifying electric power customer demand work orders as described in the above embodiment.

[0041] Specifically, the server preprocesses the obtained customer request work order data to remove abnormal data; serializes the processed work order data to obtain a serialized work order text vector; inputs the serialized work order text vector into the feature relationship extraction model to obtain the feature relationship in the work order data; uses the attention mechanism to obtain sequence information from the feature relationship; and inputs the sequence information into the softmax function to obtain the classification label corresponding to the customer request work order data.

[0042] In addition, an embodiment of the present application further provides a non-volatile computer storage medium on which executable instructions are stored. When the executable instructions are executed, a method for classifying electric power customer demand work orders as described above is implemented.

[0043] So far, the technical solutions of the present disclosure have been described in combination with the above multiple embodiments, but it is easy for those skilled in the art to understand that the protection scope of the present disclosure is not limited to these specific embodiments. Without departing from the technical principles of the present disclosure, those skilled in the art can split and combine the technical solutions in the above-mentioned various embodiments, and can also make equivalent changes or replacements to the relevant technical features. Any changes, equivalent replacements, improvements, etc. made within the technical concept and / or technical principle of the present disclosure will fall within the protection scope of the present disclosure.

Claims

1. A method for classifying work orders for power customers, characterized in that: The method comprises: Pre-process the obtained customer request work order data to remove abnormal data; Serialize the processed work order data to obtain a serialized work order text vector; Input the serialized work order text vector into the feature relationship extraction model to obtain the feature relationship in the work order data; use the attention mechanism to obtain sequence information from the feature relationship; Input the sequence information into the softmax function to obtain the classification label corresponding to the customer request work order data.

2. The method for classifying electric power customer demand work orders according to claim 1, characterized in that: The processed work order data is serialized to obtain a serialized work order text vector, which includes: Get the processed work order data and add a preset start identifier at the starting position; Input the work order data into the preset word segmenter to segment the work order data into single word sequences; Input the word sequence into the preset model to obtain output data. The preset model formula is as follows: ； Where D is the input word sequence, and are preset vector weights and bias parameters respectively, and GELU is the activation function; Input the output data into the Transformer encoder to obtain the serialized work order text vector.

3. The method for classifying electric power customer demand work orders according to claim 1, characterized in that: The serialized work order text vector is input into the feature relationship extraction model to obtain the feature relationship in the work order data, including: Use bidirectional LSTM to capture the forward and backward output hidden states of each ticket text vector in the text; By formula: Get the output hidden state at time step t; where, represents the forward output hidden state, represents the backward output hidden state, represents the previous forward output hidden state, Represents the previous backward output hidden state; By formula: Obtain an output vector H representing the feature relationship, wherein n represents the preset total number of time steps.

4. The method for classifying electric power customer demand work orders according to claim 1, characterized in that: The attention mechanism is used to obtain sequence information from feature relationships, including: By formula: Calculate sequence information ; in, represents the activation function that maps the vector to -1 ~ 1, , represents the output vector of the feature relationship, X represents the preset balance matrix, Represents the preset attention weight matrix.

5. The method for classifying electric power customer demand work orders according to claim 1, characterized in that: Input the sequence information into the softmax function to obtain the classification labels corresponding to the customer request work order data, including: By formula: Calculate the conditional probability on each classification label, and then determine the classification label with a conditional probability greater than a preset threshold as the classification label corresponding to the customer request work order data; is the preset weight parameter, is the bias parameter.

6. A power customer demand work order classification system, characterized in that: The system comprises: The removal module is used to pre-process the obtained customer demand work order data and remove abnormal data; The acquisition module is used to serialize the processed work order data to obtain a serialized work order text vector; input the serialized work order text vector into the feature relationship extraction model to obtain the feature relationship in the work order data; use the attention mechanism to obtain sequence information from the feature relationship; input the sequence information into the softmax function to obtain the classification label corresponding to the customer demand work order data.

7. The power customer demand work order classification system according to claim 6, characterized in that: The acquisition module comprises a first acquisition unit, Used to obtain the processed work order data and add a preset start identifier at the starting position; Input the work order data into the preset word segmenter to segment the work order data into single word sequences; Input the single word sequence into the hidden layer of the preset model to obtain output data; Input the output data into the Transformer encoder to obtain the serialized work order text vector.

8. The power customer demand work order classification system according to claim 6, characterized in that: The obtaining module comprises a second obtaining unit, Used to use bidirectional LSTM to capture the forward and backward output hidden states of each work order text vector in the text; By formula: Get the output hidden state at time step t; where, represents the forward output hidden state, represents the backward output hidden state, represents the previous forward output hidden state, Represents the previous backward output hidden state; By formula: Obtain an output vector H representing the feature relationship, wherein n represents the preset total number of time steps.

9. A power customer demand work order classification device, characterized in that: The device comprises: processor; and a memory having executable codes stored thereon, which, when executed, causes the processor to execute a method for classifying electric power customer demand work orders as described in any one of claims 1 to 5.

10. A non-volatile computer storage medium, characterized in that: Computer instructions are stored thereon, and when the computer instructions are executed, a method for classifying electric power customer demand work orders as described in any one of claims 1-5 is implemented.

Images