A power supply customer satisfaction evaluation method and system based on a machine learning model

Abstract

The application provides a power supply customer satisfaction evaluation method and system based on a machine learning model, relates to the technical field of evaluation, and comprises an acquisition module, a feature extraction module, a parameter adjustment module, an index identification module and a satisfaction evaluation module. The application respectively constructs a complaint risk identification model for customer service complaint risk and a service quality identification model for customer service indexes in the light of two dimensions of customers and customer service, optimizes the parameters of the complaint risk identification model and the service quality identification model according to user sensitivity, identifies more accurate complaint risk values and service quality values through the optimized model, and further calculates more accurate customer satisfaction.

Description

Technical Field

[0001] This invention relates to the field of evaluation technology, and in particular to a method and system for evaluating power supply customer satisfaction based on a machine learning model. Background Technology

[0002] Existing customer satisfaction and service quality evaluation systems for power supply companies mostly rely on the Analytic Hierarchy Process (AHP). The basic idea of ​​AHP is to decompose complex problems into several levels, based on people's experience and judgment, and to determine weights using a combination of qualitative and quantitative methods. Essentially, it uses integers from 1 to 9 and their reciprocals as scales to construct a judgment matrix. Rough set theory is a new mathematical tool for handling fuzzy and uncertain knowledge. One important idea is that it can reduce the knowledge of the system while maintaining classification ability. In classification rules, customers specify one or more behaviors in the dataset as classification decision behaviors. Based on the different values ​​of these behaviors, the data is divided into different categories, and classification rules are discovered and obtained. The main drawback of existing technologies is their inaccurate calculation of customer satisfaction.

[0003] Machine learning (ML) is a multidisciplinary field that involves probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structures to continuously improve their performance, and machine learning models can be well integrated into user evaluation.

[0004] Therefore, a pressing technical problem that needs to be solved by those skilled in the art is: how to find a new set of power supply customer satisfaction evaluation methods based on machine learning models that can solve the current shortcomings of power supply companies in calculating customer satisfaction and improve the accuracy of customer satisfaction calculation. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides a method and system for evaluating power supply customer satisfaction based on a machine learning model, thus solving the problem of insufficient accuracy in calculating customer satisfaction in the prior art.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] Firstly, a method for evaluating power supply customer satisfaction based on a machine learning model includes:

[0008] Obtain voice call information related to complaints within a specific time period, as well as the call time information corresponding to the voice call information;

[0009] The voice call information is subjected to speech recognition to obtain first customer text features, second customer text features and customer service text features, and customer voiceprint features and customer service voiceprint features are extracted from the voice call information. Combined with the second customer text features, the first sensitivity feature is extracted from the call time information.

[0010] The first sensitivity feature, customer voiceprint feature, and first customer text feature are input into the sensitivity recognition model to obtain customer sensitivity. The threshold parameters of the complaint risk recognition model and the service quality recognition model are then fine-tuned based on the customer sensitivity.

[0011] The second customer text feature is input into the adjusted complaint risk identification model to obtain the customer complaint risk. The customer service text feature and customer service voiceprint feature are input into the adjusted service quality identification model to obtain the customer service quality.

[0012] Customer satisfaction is determined based on the customer complaint risk and the quality of customer service.

[0013] Preferably, the process of obtaining first customer text features, second customer text features, and customer service text features through speech recognition of the voice call information includes:

[0014] The voice call information is converted into text information, and the text information is preprocessed.

[0015] The preprocessed text information was segmented using a stop dictionary, a sentiment dictionary, and an electricity dictionary to obtain customer sentiment corpus, customer request corpus, and customer service response corpus.

[0016] The customer emotional data, customer request data, and customer service response data are converted into word vector matrices.

[0017] The first customer text features, the second customer text features, and the customer service text features are extracted based on the word vector matrix.

[0018] Preferably, the voiceprint features include tone of voice, speech rate, and emotional inclination.

[0019] Preferably, the first sensitivity feature is extracted from the call time information by combining the second customer text features, including:

[0020] Input the second customer text features into the complaint type recognition model to obtain the complaint item and the time when the complaint item occurred;

[0021] A first sensitivity feature is extracted based on the occurrence time and the call duration information. The first sensitivity feature includes the call duration and the complaint interval duration.

[0022] Preferably, the first sensitivity feature, the customer voiceprint feature, and the first customer text feature are input into the sensitivity recognition model to obtain customer sensitivity, including:

[0023] The customer's voiceprint features are processed according to the correlation between voiceprint features and text features to obtain the first customer's voiceprint features that are correlated with the first customer's text features.

[0024] Input the first customer's text features and the first customer's voiceprint features into the emotion recognition model to obtain the customer's emotion value;

[0025] Obtain the difference between the emotion value and the preset emotion threshold, and use the difference as a second sensitivity feature;

[0026] Input the first sensitivity feature and the second sensitivity feature into the sensitivity recognition model to obtain customer sensitivity.

[0027] Preferably, fine-tuning the threshold parameters of the complaint risk identification model and the service quality identification model based on the customer sensitivity includes:

[0028] The target threshold parameter is obtained from the parameter adjustment table based on the customer sensitivity. The parameter adjustment table is used to store the mapping relationship between customer sensitivity and the threshold parameters of the complaint risk identification model and the threshold parameters of the service quality identification model.

[0029] The threshold parameters of the complaint risk identification model and the service quality identification model are adjusted according to the target threshold parameters.

[0030] Preferably, the formula for determining customer satisfaction based on the customer complaint risk and the customer service quality is as follows:

[0031]

[0032] in, For customer satisfaction, To mitigate the risk of customer complaints, For customer service quality, Weighting coefficients for customer complaint analysis.

[0033] Secondly, a power supply customer satisfaction evaluation system based on a machine learning model includes:

[0034] The acquisition module is used to acquire voice call information related to complaints within a specific time period and the call time information corresponding to the voice call information;

[0035] The feature extraction module is used to perform speech recognition on the voice call information to obtain first customer text features, second customer text features and customer service text features, and extract customer voiceprint features and customer service voiceprint features from the voice call information. Combined with the second customer text features, the module extracts first sensitivity features from the call time information.

[0036] The parameter adjustment module is used to input the first sensitivity feature, customer voiceprint feature and first customer text feature into the sensitivity recognition model to obtain customer sensitivity, and to fine-tune the threshold parameters of the complaint risk recognition model and service quality recognition model according to the customer sensitivity.

[0037] The indicator recognition model is used to input the second customer text features into the adjusted complaint risk recognition model to obtain customer complaint risk, and to input the customer service text features and customer service voiceprint features into the adjusted service quality recognition model to obtain customer service quality.

[0038] The customer satisfaction evaluation module is used to determine customer satisfaction based on the customer complaint risk and the customer service quality.

[0039] The beneficial effects of this invention are as follows: This invention provides a power supply customer satisfaction evaluation method based on a machine learning model. The system constructs a complaint risk identification model for customer service complaint risk and a service quality identification model for customer service indicators, focusing on two dimensions: the customer and the customer's perception of customer service. The parameters of the complaint risk identification model and the service quality identification model are optimized based on user sensitivity. The optimized model can identify more accurate complaint risk values ​​and service quality values, thereby calculating more accurate customer satisfaction. Attached Figure Description

[0040] Figure 1 A flowchart illustrating a power supply customer satisfaction evaluation method based on a machine learning model, provided as an embodiment of the present invention.

[0041] Figure 2 A basic principle diagram of a power supply customer satisfaction evaluation method based on a machine learning model provided in an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of a power supply customer satisfaction evaluation method based on a machine learning model provided in an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of a power supply customer satisfaction evaluation system based on a machine learning model, provided in an embodiment of the present invention. Detailed Implementation

[0044] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0045] Example 1

[0046] like Figure 1 As shown, this embodiment of the invention provides a method for evaluating power supply customer satisfaction based on a machine learning model, including:

[0047] Step 1: Obtain the voice call information related to the complaint within a specific time period, as well as the call time information corresponding to the voice call information.

[0048] Specifically, the specific time period can be an hour, a day, a month, or a quarter; this embodiment does not impose any restrictions. Voice call information, voice call time information, and customer information are stored together in the voice database.

[0049] Step two: Perform speech recognition on the voice call information to obtain the first customer text feature, the second customer text feature, and the customer service text feature; extract the customer voiceprint feature and the customer service voiceprint feature from the voice call information; and combine the second customer text feature to extract the first sensitivity feature from the call time information.

[0050] In this embodiment of the invention, the voiceprint features include tone of voice, speech rate, and emotional inclination.

[0051] In this embodiment of the invention, the process of obtaining first customer text features, second customer text features, and customer service text features by performing speech recognition on the voice call information includes: converting the voice call information into text information and preprocessing the text information; performing word segmentation on the preprocessed text information using a stop dictionary, an emotion dictionary, and an electricity dictionary to obtain customer emotion corpus, customer request corpus, and customer service response corpus; converting the customer emotion corpus, customer request corpus, and customer service response corpus into a word vector matrix, and extracting the first customer text features, second customer text features, and customer service text features based on the word vector matrix.

[0052] Specifically, preprocessing includes data cleaning, which can be done through data cleaning alignment, deletion and annotation, or by extracting content through rules, regular expression matching, extraction based on parts of speech and named entities, or batch processing with scripts or code.

[0053] Of course, to ensure the accuracy of the preprocessed data, error correction can be performed on the processed text. As an example, the built-in power terminology can be used to identify text errors in the cleaned text information. Some common error identifications can be pre-stored in the power terminology, such as common fault types, complaint types, and power information types.

[0054] Because the preprocessed text contains many industry-specific terms, such as power outages and coal-to-electricity conversions, general word segmentation dictionaries have difficulty correctly segmenting these terms.

[0055] This application uses Jieba Chinese word segmentation in R language as the stop word dictionary. Jieba combines two word segmentation methods: statistical and string matching, which can achieve basic word segmentation functions. In addition, a custom electricity dictionary is provided, which can perform word segmentation of electricity-specific terms. Due to the different business needs for customer and customer service identification, the electricity dictionary includes a customer service electricity dictionary and a customer electricity dictionary. The customer electricity dictionary mainly includes words related to the type, time, and degree of electricity complaints, while the customer service electricity dictionary mainly includes words related to the solutions for each type of electricity complaint. In addition, an emotion dictionary is also provided. Due to the different business needs for customer and customer service emotion identification, the emotion dictionary includes a customer service emotion dictionary and a customer emotion dictionary. Similarly, the customer service emotion dictionary is a general emotion dictionary, which only includes some general emotion words, such as friendly, peaceful, and angry, while the customer emotion dictionary is a specific emotion dictionary, which only includes words related to urgent emotions.

[0056] In this embodiment of the invention, the extraction of a first sensitivity feature from the call time information in conjunction with the second customer text features includes: inputting the second customer text features into a complaint type recognition model to obtain a complaint item and the occurrence time of the complaint item; and extracting a first sensitivity feature based on the occurrence time and the call time information, wherein the first sensitivity feature includes the call duration and the complaint interval duration.

[0057] Step 3: Input the first sensitivity feature, customer voiceprint feature and first customer text feature into the sensitivity recognition model to obtain customer sensitivity, and fine-tune the threshold parameters of the complaint risk recognition model and service quality recognition model based on the customer sensitivity.

[0058] In this embodiment of the invention, the customer sensitivity is obtained by inputting the first sensitivity feature, the customer voiceprint feature, and the first customer text feature into a sensitivity recognition model. This includes: processing the customer voiceprint feature according to the correlation between the voiceprint feature and the text feature to obtain a first customer voiceprint feature that is correlated with the first customer text feature; inputting the first customer text feature and the first customer voiceprint feature into an emotion recognition model to obtain the customer's emotion value; obtaining the difference between the emotion value and a preset emotion threshold, and using the difference as a second sensitivity feature; and inputting the first sensitivity feature and the second sensitivity feature into a sensitivity recognition model to obtain the customer sensitivity.

[0059] It should be noted that, in order to improve the accuracy of the sensitivity identification model, customer identity features can also be input into the sensitivity identification model. This is because different customers have different sensitivities to different types of requests. For example, government customers are more sensitive to power outages than residential customers.

[0060] In this embodiment of the invention, fine-tuning the threshold parameters of the complaint risk identification model and the service quality identification model based on the customer sensitivity includes: obtaining a target threshold parameter from a parameter adjustment table based on the customer sensitivity, wherein the parameter adjustment table is used to store the mapping relationship between customer sensitivity and the threshold parameters of the complaint risk identification model and the service quality identification model; and adjusting the threshold parameters of the complaint risk identification model and the service quality identification model based on the target threshold parameter.

[0061] like Figure 2 , 3 As shown, customer satisfaction evaluation is mainly evaluated from two perspectives: the customer themselves and the customer's evaluation of customer service. Based on this idea, this application extracts features from the voice information of the customer and the customer service.

[0062] Unlike existing technologies, most of which directly input extracted features into pre-built machine learning models to obtain customer service satisfaction, this sampling method suffers from problems such as large data volume, high data requirements, and long learning time. If the learning requirements are not met, the accuracy will be insufficient. Therefore, the machine learning model is split into two parts: a customer risk complaint index based on the customer's own dimension and its corresponding machine learning-based customer complaint risk identification model; and a customer service quality index based on the customer's perception of customer service and its corresponding machine learning-based customer service quality identification model. Furthermore, the impact of customer sensitivity on the indicators corresponding to the customer's own dimension and the indicators corresponding to the customer's perception of service is considered. The parameters of the corresponding models are optimized by using customer sensitivity, so that the classification threshold of the classifier in the model is adjusted when identifying customer complaint risks and service quality, thereby obtaining a more targeted model and improving the accuracy of customer satisfaction calculation results. For example, the complaint risk type identification result can be 1, 2, 3, 4, 5, which correspond to level 1 risk, level 2 risk, level 3 risk, level 4 risk, and level 5 risk, respectively. The service quality identification model identification result can be 1, 2, 3, 4, 5, which correspond to low quality, slightly low quality, medium quality, slightly high quality, and high quality, respectively.

[0063] Step four: Input the second customer text features into the adjusted complaint risk identification model to obtain customer complaint risk; input the customer service text features and customer service voiceprint features into the adjusted service quality identification model to obtain customer service quality.

[0064] After the model is adjusted, the evaluation index values ​​can be obtained through the model. In this application, the evaluation index values, namely customer complaint risk and customer service quality, are obtained through the complaint risk identification model and the service quality identification model. Since the model is adjusted for customer sensitivity, the obtained customer complaint risk and customer service quality are more accurate.

[0065] It should also be noted that, since customer service representatives should maintain their professionalism when evaluating them, their tone and manner of speaking should be standardized and relatively stable. If changes occur, the customer service quality can be considered poor. Therefore, this embodiment uses voiceprint features and text features related to customer service emotions as features for identifying customer service quality. However, for customers, they are individualized, and their daily tone and manner of speaking are not standardized. Using conventional methods to use voiceprint features and text features related to customer emotions as features for identifying customer complaint risks not only increases the number of customer complaint risks identified but also reduces the accuracy of customer complaint risk identification. Therefore, this embodiment of the invention proposes to characterize the sensitivity feature by the difference between the customer's emotion value and its emotion threshold. The emotion threshold can be a preset emotion mean. If the difference is large, it indicates that the customer is more sensitive; if the difference is small, it indicates that the customer is not so sensitive.

[0066] Step 5: Determine customer satisfaction based on the customer complaint risk and the customer service quality.

[0067] In this embodiment of the invention, the formula for determining customer satisfaction based on the customer complaint risk and the customer service quality is as follows:

[0068]

[0069] in, For customer satisfaction, To mitigate the risk of customer complaints, For customer service quality, Weighting coefficients for customer complaint analysis.

[0070] Specifically, a questionnaire was designed regarding two evaluation indicators. The importance of customer complaint risk and customer service quality indicators was obtained from the questionnaire, and weighting coefficients were determined based on the importance.

[0071] In summary, this invention provides a method for evaluating customer satisfaction with electricity supply based on a machine learning model. The method involves acquiring voice call information related to complaints within a specific time period, as well as the corresponding call time information. Voice recognition is performed on the voice call information to obtain first customer text features, second customer text features, and customer service text features. Customer voiceprint features and customer service voiceprint features are extracted from the voice call information. Combined with the second customer text features, a first sensitivity feature is extracted from the call time information. The first sensitivity feature, customer voiceprint features, and first customer text features are input into a sensitivity recognition model to obtain customer sensitivity. Threshold parameters of the complaint risk recognition model and service quality recognition model are fine-tuned based on the customer sensitivity. The second customer text features are input into the adjusted complaint risk recognition model to obtain customer complaint risk. The customer service text features and customer service voiceprint features are input into the adjusted service quality recognition model to obtain customer service quality. Customer satisfaction is determined based on the customer complaint risk and the customer service quality. This invention constructs a complaint risk identification model and a service quality identification model for customer service indicators, focusing on two dimensions: the customer and the customer's perception of customer service. The parameters of the complaint risk identification model and the service quality identification model are optimized based on user sensitivity. The optimized models can identify more accurate complaint risk values ​​and service quality values, thereby enabling the calculation of more accurate customer satisfaction.

[0072] Example 2

[0073] As shown in Figure 4, this embodiment of the invention provides a power supply customer satisfaction evaluation system based on a machine learning model, comprising: an acquisition module 100, used to acquire voice call information related to complaints within a specific time period and the call time information corresponding to the voice call information; a feature extraction module 200, used to perform voice recognition on the voice call information to obtain a first customer text feature, a second customer text feature, and a customer service text feature, and extract customer voiceprint features and customer service voiceprint features from the voice call information, and extract a first sensitivity feature from the call time information in combination with the second customer text feature; a parameter adjustment module 300, used to input the first sensitivity feature, customer voiceprint features, and first customer text features into a sensitivity recognition model to obtain customer sensitivity, and fine-tune the threshold parameters of the complaint risk recognition model and the service quality recognition model according to the customer sensitivity; an indicator recognition model 400, used to input the second customer text features into the adjusted complaint risk recognition model to obtain customer complaint risk, and input the customer service text features and customer service voiceprint features into the adjusted service quality recognition model to obtain customer service quality; and a satisfaction evaluation module 500, used to determine customer satisfaction based on the customer complaint risk and the customer service quality.

[0074] It should be understood that the power supply customer satisfaction evaluation system based on a machine learning model provided in this embodiment of the invention and the power supply customer satisfaction evaluation method based on a machine learning model provided in the above embodiments are based on the same inventive concept. For more specific working principles of each module in this embodiment of the invention, please refer to the above embodiments, which will not be repeated in this embodiment of the invention.

[0075] Those skilled in the art will understand that although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the machine equivalents of the claims, the invention also intends to include these modifications and modifications.

Claims

1. A method for evaluating power supply customer satisfaction based on a machine learning model, characterized in that, include: Obtain voice call information related to complaints within a specific time period, as well as the call time information corresponding to the voice call information; The voice call information is subjected to speech recognition to obtain first customer text features, second customer text features and customer service text features, and customer voiceprint features and customer service voiceprint features are extracted from the voice call information. Combined with the second customer text features, the first sensitivity feature is extracted from the call time information. The first sensitivity feature, customer voiceprint feature, and first customer text feature are input into the sensitivity recognition model to obtain customer sensitivity. The threshold parameters of the complaint risk recognition model and the service quality recognition model are then fine-tuned based on the customer sensitivity. The second customer text feature is input into the adjusted complaint risk identification model to obtain the customer complaint risk. The customer service text feature and customer service voiceprint feature are input into the adjusted service quality identification model to obtain the customer service quality. Customer satisfaction is determined based on the customer complaint risk and the quality of customer service.

2. The power supply customer satisfaction evaluation method based on a machine learning model according to claim 1, characterized in that, The voice call information is processed by speech recognition to obtain the first customer text features, the second customer text features, and the customer service text features, including: The voice call information is converted into text information, and the text information is preprocessed. The preprocessed text information was segmented using a stop dictionary, a sentiment dictionary, and an electricity dictionary to obtain customer sentiment corpus, customer request corpus, and customer service response corpus. The customer emotional data, customer request data, and customer service response data are converted into word vector matrices. The first customer text features, the second customer text features, and the customer service text features are extracted based on the word vector matrix.

3. The power supply customer satisfaction evaluation method based on a machine learning model according to claim 1, characterized in that, The voiceprint features include tone of voice, speaking speed, and emotional inclination.

4. The power supply customer satisfaction evaluation method based on a machine learning model according to claim 1, characterized in that, Combining the second customer text features, a first sensitivity feature is extracted from the call time information, including: Input the second customer text features into the complaint type recognition model to obtain the complaint item and the time when the complaint item occurred; A first sensitivity feature is extracted based on the occurrence time and the call duration information. The first sensitivity feature includes the call duration and the complaint interval duration.

5. The power supply customer satisfaction evaluation method based on a machine learning model according to claim 1, characterized in that, The first sensitivity feature, customer voiceprint feature, and first customer text feature are input into the sensitivity recognition model to obtain customer sensitivity, including: The customer's voiceprint features are processed according to the correlation between voiceprint features and text features to obtain the first customer's voiceprint features that are correlated with the first customer's text features. Input the first customer's text features and the first customer's voiceprint features into the emotion recognition model to obtain the customer's emotion value; Obtain the difference between the emotion value and the preset emotion threshold, and use the difference as a second sensitivity feature; Input the first sensitivity feature and the second sensitivity feature into the sensitivity recognition model to obtain customer sensitivity.

6. The power supply customer satisfaction evaluation method based on a machine learning model according to claim 1, characterized in that, Fine-tuning the threshold parameters of the complaint risk identification model and service quality identification model based on the aforementioned customer sensitivity includes: The target threshold parameter is obtained from the parameter adjustment table based on the customer sensitivity. The parameter adjustment table is used to store the mapping relationship between customer sensitivity and the threshold parameters of the complaint risk identification model and the threshold parameters of the service quality identification model. The threshold parameters of the complaint risk identification model and the service quality identification model are adjusted according to the target threshold parameters.

7. The power supply customer satisfaction evaluation method based on a machine learning model according to claim 1, characterized in that, The formula for determining customer satisfaction based on the aforementioned customer complaint risk and customer service quality is as follows: in, For customer satisfaction, To mitigate the risk of customer complaints, For customer service quality, Weighting coefficients for customer complaint analysis.

8. A power supply customer satisfaction evaluation system based on a machine learning model, characterized in that, include: The acquisition module is used to acquire voice call information related to complaints within a specific time period and the call time information corresponding to the voice call information; The feature extraction module is used to perform speech recognition on the voice call information to obtain first customer text features, second customer text features and customer service text features, and extract customer voiceprint features and customer service voiceprint features from the voice call information. Combined with the second customer text features, the module extracts first sensitivity features from the call time information. The parameter adjustment module is used to input the first sensitivity feature, customer voiceprint feature and first customer text feature into the sensitivity recognition model to obtain customer sensitivity, and to fine-tune the threshold parameters of the complaint risk recognition model and service quality recognition model according to the customer sensitivity. The indicator recognition model is used to input the second customer text features into the adjusted complaint risk recognition model to obtain customer complaint risk, and to input the customer service text features and customer service voiceprint features into the adjusted service quality recognition model to obtain customer service quality. The customer satisfaction evaluation module is used to determine customer satisfaction based on the customer complaint risk and the customer service quality.

Images