CN115719183A - Power customer self-feedback service evaluation method and system based on weight dynamic grading - Google Patents
Power customer self-feedback service evaluation method and system based on weight dynamic grading Download PDFInfo
- Publication number
- CN115719183A CN115719183A CN202211513035.6A CN202211513035A CN115719183A CN 115719183 A CN115719183 A CN 115719183A CN 202211513035 A CN202211513035 A CN 202211513035A CN 115719183 A CN115719183 A CN 115719183A
- Authority
- CN
- China
- Prior art keywords
- service
- evaluation
- client
- content
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
A power customer self-feedback service evaluation method and system based on weight dynamic grading are disclosed, the method comprises: in each specific service in the electric power service APP, combing the operation flow required to be handled by a client, and recording the content volume of data required by the client to handle a specific task; arranging various content amounts obtained by combing and other information in the whole process of customer service into quantifiable data, and constructing an evaluation index model; collecting customer use data, manually cleaning the data, and generating a self-feedback service evaluation in real time by adopting an evaluation index model; training and solving undetermined coefficients in the evaluation index model; will evaluate the index model E n Real-time feedback is pushed toAnd the client himself and optimizes the evaluation model through the feedback of the client. According to the invention, the instant service quality evaluation based on the real-time acquisition of the customer service data is realized by quantifying the customer service flow information in the electric power service APP and dynamically grading the weight influencing the evaluation index.
Description
Technical Field
The invention relates to research of a service evaluation method, in particular to a power customer self-feedback service evaluation method and system based on weight dynamic grading.
Background
The customer service quality is the key point of the marketing task, and the good customer relationship is the basis for stably selling and improving the brand image. In the internet era, traditional offline client services are gradually shifted to non-contact service modes such as a PC end and a mobile phone end. The quality of the intelligent terminal service cannot be directly perceived like the traditional business hall service. Therefore, how to track and analyze the operation behavior of the customer when the customer uses the intelligent terminal to access the service, and accurately evaluate the use experience of the function and the satisfaction degree of receiving the service is very important.
The electric power service APP has wide public users, and in the traditional mode, client side active behaviors such as client message leaving, satellite evaluation, telephone complaint and the like are used for acquiring client service satisfaction information, so that the service quality evaluation covered by a whole person cannot be achieved, and the actual service experience of all users is difficult to obtain.
Disclosure of Invention
The invention mainly solves the technical problems in the prior art and provides a method and a system for evaluating self-feedback service of a power customer based on weight dynamic classification. According to the method, the instant service quality evaluation based on the real-time acquisition of the customer service data is realized by quantifying the customer service flow information in the electric power service APP and dynamically grading the weight influencing the evaluation index, so that the problem that the customer service quality cannot be automatically evaluated in real time according to comprehensive factors such as the content and the flow of customer service in the conventional method is solved, and powerful support is provided for the theory and the application of a customer evaluation method.
The technical problem of the invention is mainly solved by the following technical scheme:
a power customer self-feedback service evaluation method based on weight dynamic grading comprises the following steps:
step one, carding the operation content of a customer: in each specific service in the electric power service APP, combing the operation flow required to be handled by a client, and recording the content volume of data required by the client to handle a specific task;
step two, arranging evaluation indexes: arranging various content amounts obtained by combing in the step one and other information in the whole process of customer service into quantifiable data, and constructing an evaluation index model E n ；
Step three, collecting real-time customer service data: when the customer uses the electric power service APP service function, the customer use data is collected, the data is manually cleaned, part of missing data is removed, and the data is collectedEvaluation index model E constructed by step two n Generating a self-feedback service evaluation in real time;
step four, evaluating the index model E in a semi-supervised weight dynamic grading mode n Undetermined coefficient k in 1 、k 2 、k 3 Training and solving are carried out, and a foundation is laid for subsequent self-feedback evaluation;
step five, self-feedback customer service evaluation: the evaluation index model E obtained in the step two n And real-time feedback is pushed to the client, and the evaluation model is further optimized through the feedback of the client.
Further, the first step specifically includes:
step 1.1, carding the number of pages required to be jumped for each service, wherein each service is represented by S and respectively recorded as: s. the 1 、S 2 、S 3 、…S n Wherein the numbers 1 to n represent the nth service, with S n P all Represents the minimum number of pages, P, that the customer needs to jump to complete the entire process in the nth customer service all All pages representing a certain service;
step 1.2, in each page of each specific service, the content amount which is manually input by a client is combed and recorded, and the input content is divided into 3 types: category 1 is content input by customer selection of options; the 2 nd category is contents which can be quickly and directly filled by a user, such as name, native place, age, mobile phone number and the like; category 3 is the content that the user needs to refer to external data or input through interaction, including verification codes and some information that the user cannot recite directly;
step 1.3, determining that the content of data required by a certain client for handling a specific task in the electric power service APP is the sum of all 3 types of content in all process pages contained in the task.
Further, the content amount of the content required to be input by the type 3 clients is combed and recorded by the following method:
a) The 1 st category is the content which needs to be read and selected by the customer, and is recorded by the number of characters in the content which needs to be read by the customer, wherein, the Chinese characters are recorded by 2 charactersThe mother and the number are recorded by 1 character, the 1 st type content of the 2 nd page in the 3 rd service in the power service APP is represented as S 3 P 2 C 1 ；
b) Class 2 is a category in which the user can fill in the content directly, also using the number of characters as a record, denoted C 2 ；
c) Similarly, type 3 content is recorded in the number of characters, denoted C 3 。
Further, the second step specifically comprises:
step 2.1, establishing a content evaluation model for interaction required by all customers of each specific service according to the content amount combed in the step one, wherein the content evaluation model is used for evaluating the content amount of each specific service and is expressed as:
S n P all C＝k 1 S n P all C 1 +k 2 S n P all C 2 +k 3 S n P all C 3
wherein k is 1 、k 2 、k 3 Linear superposition is adopted for the undetermined coefficient indicating the 3 types of content, and the undetermined coefficient provides guarantee for the dynamic distribution of subsequent weights;
step 2.2, recording the retention time of each page as S n P all T＝S n P 1 T+
S n P 2 T+……+S n P n T；
Step 2.3, regarding a certain page, taking the ratio of the page content and the residence time of a customer in the page in the customer service as one of the page service evaluation indexes, and expressing the ratio as S n P n ＝S n P all C/S n P n And T, the evaluation indexes of all the pages of the whole certain service are expressed as follows:
S n P all ＝S n P 1 +S n P 2 +……+S n P n
step 2.4, sorting the access times of each page, and evaluating the electricity consumption of the client as an influence factorIn the force service APP, whether redundant operation exists or not and further influencing service evaluation are determined by S n P n F represents the redundancy access frequency of the P page in the nth service, and the redundancy access frequency is the actual access frequency minus 1;
step 2.5, S for step 2.3 and step 2.4 n P all And S n P n And F, combining an evaluation index model for evaluating the nth service effect, wherein the evaluation index model is expressed as follows:
E n ＝S n P all F×S n P all
wherein S n P all F＝S n P 1 F+S n P 2 F+……+S n P n F。
Further, the third step specifically comprises:
step 3.1, when a customer uses a specific service, loading attribute data related to the service, including a service number, the number of pages needing to be adjusted in the service and the content of each page;
step 3.2, recording the entering time when the client enters a certain function page; recording the exit time when exiting, and calculating the time difference between the exit time and the exit time to identify the staying time of the client on the page;
3.3, collecting the interactive data of the customers in each page, and recording and calculating the content volume according to the method, wherein the privacy information related to the users only calculates the content volume data and does not record specific content;
3.4, after the current operation of the client is finished, manually cleaning the data, removing part of missing data, and then adopting the evaluation index model E constructed by the steps n And generating self-feedback service evaluation in real time, providing quantitative data support for the service evaluation of the power marketing department according to the evaluation result, and emptying all data of customers.
Further, the fourth step specifically includes:
step 4.1, collecting usage data of a '5-star evaluation' client and a '1-star evaluation' client, respectively identifying good user experience and poor user experience, wherein the collected data comprises all data in the step two;
step 4.2, calibrating the quantitative data used by the client as a training sample, calibrating the 'good evaluation' and the 'poor evaluation' as sample labels, constructing a deep neural network training framework, and training a service evaluation model;
4.3, a deep learning training framework comprises VGG 19 and Inception V3, the training process is divided into two stages, the first stage model is a pre-training model of historical big data, the training of a basic model is realized, and the training is realized by adopting the VGG 19 training framework; the second-stage model adopts a lightweight deep learning framework Inception V3 to realize the rapid and secondary training of the pre-training model in a real-time environment, and a sample set of the secondary training comes from user data of a recent customer;
and 4.4, selecting 1000-5000 groups of small-scale customer data as samples according to different service contents, wherein the samples comprise all the data in the step two, and the labels of the samples are 'good evaluation' and 'poor evaluation'.
Further, the fifth step specifically includes:
step 5.1, according to the evaluation index model E obtained in the step two n Evaluating whether the customer service is satisfied by the customer in real time, and pushing a result to the customer;
step 5.2, pushing self-service functionality and friendliness comprehensive self-evaluation of the electric power service APP to the client, and meanwhile collecting feedback information of the user;
step 5.3, if the feedback information of the client is not collected, the evaluation capability of the default evaluation model is more accurate;
and 5.4, if the difference between the collected feedback information of the client and the evaluation is large, using the new feedback information as a data sample label, and feeding the sample back to the calculation server to optimize the training evaluation model again.
A power customer self-feedback service evaluation system based on weight dynamic grading comprises:
the client operation content combing module is used for combing operation flows needing to be handled by a client in each specific service in the electric power service APP and recording the content of data needed by the client to handle a specific task;
the arrangement evaluation index module is used for arranging various content amounts obtained by combing the client operation content amount combing module and other information in the whole client service process into quantifiable data and constructing an evaluation index model E n ；
The real-time customer service data acquisition module is used for acquiring customer service data when a customer uses an electric power service APP service function, manually cleaning the data, removing partial missing data and adopting a constructed evaluation index model E n Generating a self-feedback service evaluation in real time;
a semi-supervised dynamic weight grading module for grading the evaluation index model E in a semi-supervised dynamic weight grading mode n Undetermined coefficient k in 1 、k 2 、k 3 Training and solving are carried out, and a foundation is laid for subsequent self-feedback evaluation;
a self-feedback customer service evaluation module for obtaining an evaluation index model E n And real-time feedback is pushed to the client, and the evaluation model is further optimized through the feedback of the client.
Furthermore, the client operation content combing module combs the operation flow that the client needs to handle in each item of specific service in the electric power service APP, and records the content of data required for the client to handle a specific task, and specifically includes:
combing the number of pages needing to be jumped for each service, wherein each service is represented by S and is respectively recorded as: s 1 、S 2 、S 3 、…S n Where the numbers 1 to n denote the nth service, in S n P all Represents the minimum number of pages, P, that the customer needs to jump to complete the entire process in the nth customer service all All pages representing a certain service;
in each page of each specific service, a client needs to manually input the content amount and record the content amount, and the input content is divided into 3 types: category 1 is content input by customer selection of options; the 2 nd category is contents which can be quickly and directly filled by a user, such as name, native place, age, mobile phone number and the like; category 3 is the content that the user needs to refer to external data or input through interaction, including verification codes and some information that the user cannot recite directly;
determining that the content of data required by a certain client for handling a specific task in the power service APP is the sum of all 3 types of content in all process pages contained in the task.
Step 1.1, carding the number of pages required to be jumped for each service, wherein each service is represented by S and respectively recorded as: s 1 、S 2 、S 3 、…S n Where the numbers 1 to n denote the nth service, in S n P all Represents the minimum number of pages, P, to be skipped for the client to complete the whole process in the nth client service all All pages representing a certain service.
Further, the arrangement evaluation index module arranges various content amounts obtained by combing the client operation content amount combing module and other information in the whole process of client service into quantifiable data, and constructs an evaluation index model E n The method specifically comprises the following steps:
establishing an interactive content evaluation model for all customers of each specific service according to the content combed by the customer operation content combing module, wherein the content evaluation model is used for evaluating the content of each specific service and is expressed as follows:
S n P all C＝k 1 S n P all C 1 +k 2 S n P all C 2 +k 3 S n P all C 3
wherein k is 1 、k 2 、k 3 Linear superposition is adopted for the undetermined coefficient indicating the 3 types of content, and the undetermined coefficient provides guarantee for the dynamic distribution of subsequent weights;
recording the stay time of each page as S n P all T＝S n P 1 T+
S n P 2 T+……+S n P n T；
For a certain page, the client stops on the page according to the content amount of the page and the service of the clientThe ratio of the remaining time is one of the page service evaluation indexes and is expressed as S n P n ＝S n P all C/S n P n And T, the evaluation indexes of all the pages of the whole service are expressed as follows:
S n P all ＝S n P 1 +S n P 2 +……+S n P n
arranging the access times of each page, using the access times as influence factors to evaluate whether redundant operation exists in the electric power service APP used by the client or not so as to influence service evaluation, and using S n P n F represents the redundancy access frequency of the P page in the nth service, and the redundancy access frequency is the actual access frequency minus 1;
to S n P all And S n P n And F, combining an evaluation index model for evaluating the nth service effect, wherein the evaluation index model is expressed as follows:
E n ＝S n P all F×S n P all
wherein S n P all F＝S n P 1 F+S n P 2 F+……+S n P n F。
The invention has the following beneficial effects:
1. the division of the content of the 3 types of pages is realized;
2. the quantification of the page content is realized, and the user satisfaction is comprehensively evaluated according to the content of each page needing to be accessed in the user service flow and the user access time;
3. using deep learning method to evaluate the coefficient k to be determined in the model En 1 、k 2 、k 3 Dynamic semi-supervised training is carried out, and the generalization capability of the model is increased.
Drawings
FIG. 1 is a schematic flow chart of a power customer self-feedback service evaluation method based on dynamic weight classification according to the present invention;
fig. 2 is a block diagram of a power customer self-feedback service evaluation system based on weight dynamic classification according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
Referring to fig. 1, an embodiment of the present invention provides a method for evaluating a self-feedback service of a power customer based on weight dynamic classification, including the following steps:
step one, carding the operation content of a customer: in each specific service in the electric power service APP, combing the operation flow required to be handled by a client, and recording the content volume of data required by the client to handle a specific task; the concrete implementation steps comprise:
step 1.1, combing the number of pages needing to be jumped for each service, wherein each specific service is represented by S and is respectively recorded as: s 1 、S 2 、S 3 、…S n Where the numbers 1 to n denote the nth service, in S n P all Represents the minimum number of pages, P, that the customer needs to jump to complete the entire process in the nth customer service all All pages representing a certain service;
and step 1.2, combing each page of each specific service, and recording the content amount manually input by the client. The input content is divided into 3 types: category 1 is content input by customer selection of options; the 2 nd category is contents which can be quickly and directly filled by a user, such as name, native place, age, mobile phone number and the like; category 3 is what the user needs to refer to external material or input through interaction, including authentication codes and some information (e.g., bank card number, line number of account opening, etc.) that the user may not be able to recite directly. The content amount of the content which needs to be input by the 3 types of clients is combed and recorded by the following method and steps:
a) Class 1 requiring customer readingAnd the selected content is recorded by the number of characters in the content which is required to be read by a user, wherein Chinese characters are recorded by 2 characters, letters and numbers are recorded by 1 character, and the 1 st type content volume of the 2 nd page in the 3 rd service in the electric power service APP is represented as S 3 P 2 C 1 ；
b) Class 2 is a category in which the user can fill in the content directly, also using the number of characters as a record, denoted C 2 ；
c) Similarly, type 3 content is recorded in the number of characters, denoted C 3 。
Step 1.3, determining that the content of data required by a certain client for handling a specific task in the electric power service APP is the sum of all 3 types of content in all process pages contained in the task.
Step two, arranging evaluation indexes: and (4) sorting the content amounts obtained by combing in the step one and other information in the whole process of customer service into quantifiable data to obtain an evaluation index model. The concrete implementation steps are as follows:
step 2.1, establishing a content evaluation model for interaction required by all customers of each specific service according to the content amount combed in the step one, wherein the content evaluation model is used for evaluating the content amount of each specific service and is expressed as:
S n P all C＝k 1 S n P all C 1 +k 2 S n P all C 2 +k 3 S n P all C 3
wherein k is 1 、k 2 、k 3 The undetermined coefficients indicate that the 3 types of content are linearly superposed, and provide guarantee for the dynamic distribution of subsequent weights;
step 2.2, recording the retention time of each page as S n P all T＝S n P 1 T+
S n P 2 T+……+S n P n T；
Step 2.3, for a certain page, the client stops on the page according to the content amount of the page and the client serviceThe ratio of the remaining time, which is one of the page service evaluation indexes, is expressed as S n P n ＝S n P all C/S n P n And T, the evaluation indexes of all the pages of the whole certain service are expressed as follows:
S n P all ＝S n P 1 +S n P 2 +……+S n P n
step 2.4, the access times of each page are arranged and used as influence factors to evaluate whether redundant operations such as repeated access exist in the electric power service APP used by the customer so as to influence service evaluation, and S is used for n P n F denotes the number of redundant accesses (actual number of accesses minus 1) of the P-th page in the nth service.
Step 2.5, S for step 2.3 and step 2.4 n P all And S n P n And F, combining an evaluation index model for evaluating the nth service effect, wherein the evaluation index model is expressed as follows:
E n ＝S n P all F×S n P all
wherein S n P all F＝S n P 1 F+S n P 2 F+……+S n P n F。
Step three, collecting real-time customer service data: when a customer uses an electric power service APP service function, collecting customer use data, manually cleaning the data, and removing part of missing data (if removing the data which does not successfully obtain the residence time of a user page), wherein the specific implementation steps are as follows:
step 3.1, when a customer uses a specific service, loading attribute data related to the service, including a service number, the number of pages needing to be adjusted in the service and the content of each page;
step 3.2, recording the entering time when the client enters a certain function page; recording the exit time when exiting, and calculating the time difference between the exit time and the exit time to identify the staying time of the client on the page;
step 3.3, collecting the interactive data of the customers in each page, and recording and calculating the content volume according to the method, wherein the privacy information related to the users only calculates the content volume data and does not record specific contents;
3.4, after the current operation of the client is finished, manually cleaning the data, eliminating partial missing data, and then constructing an evaluation index model E by adopting the steps n And generating self-feedback service evaluation in real time, wherein the evaluation result provides quantitative data support for the service evaluation of the power marketing department, and then emptying all data of customers.
Step four, evaluating index model E constructed in step two n Wherein k is 1 、k 2 、k 3 For undetermined coefficients, the client use data acquired in real time are trained and solved in a semi-supervised weight dynamic grading mode to obtain each service evaluation index E n The specific numerical value of (1). The training mode adopts a VGG 19 deep learning model, and the training data is the collected and manually calibrated customer data. The concrete implementation steps are as follows:
and 4.1, collecting use data of the '5-star evaluation' client and the '1-star evaluation' client, respectively identifying good user experience and poor user experience, wherein the collected data comprises all data in the 'sorting evaluation index' step.
And 4.2, calibrating the 'good evaluation' and 'poor evaluation' as sample labels by taking the quantitative data used by the client as a training sample, constructing a deep neural network training framework, and training a service evaluation model.
4.3, the deep learning training framework comprises VGG 19 and IncepotionV 3, the training process is divided into two stages, the first-stage model is a pre-training model of historical big data, the training of the basic model is realized, and the training is realized by adopting the VGG 19 training framework; since the evaluation model trained by historical big data cannot reflect the distribution characteristics of 'recent' update data or real-time data, a second-level model is needed. And the second-stage model adopts a lightweight deep learning framework Inception V3 to realize the rapid and secondary training of the pre-training model in a real-time environment. The sample set for the second training is from recent (1 month or 1 week) customer data.
Step 4.4, to better reflect the real-time performance of the evaluation, the modelThe model needs to be trained quickly, 1000-5000 groups of small-scale customer data are selected as samples according to different service contents, the samples comprise all data in the step of arranging evaluation indexes, and the labels of the samples are 'good evaluation' and 'poor evaluation'. Meanwhile, the mode of secondarily training the evaluation model also reflects the training parameter k 1 、k 2 、k 3 Dynamic nature of (weight assignment).
Step five, self-feedback customer service evaluation: the evaluation index model E obtained in the step two n And real-time feedback is pushed to a client, and the evaluation model is further optimized through the feedback of the client, and the specific implementation steps are as follows:
step 5.1, according to the evaluation index model E obtained in the step two n Real-time evaluation on whether the customer service is satisfied by the customer and pushing the result to the customer, such as "whether our just-served you are satisfied, can you hit 80 points? "
Step 5.2, pushing comprehensive self-evaluation of self-service functionality, friendliness and the like of the electric power service APP to the client, and collecting feedback information of the user;
and 5.3, if the feedback information of the client is not collected, the evaluation capability of the default evaluation model is more accurate.
And 5.4, if the difference between the collected feedback information of the client and the evaluation is large, using the new feedback information as a data sample label, and feeding the sample back to the calculation server to optimize the training evaluation model again.
Referring to fig. 2, an embodiment of the present invention further provides a power customer self-feedback service evaluation system based on dynamic weight classification, which takes the power fee query service in the XX national grid APP as an example, and the system includes:
the client operation content combing module 10 is used for combing operation flows required by clients in various specific services in the electric power service APP, and recording content of data required by the clients for transacting a specific task. The main function of the module is the combing of the client operation content. Taking the electricity fee inquiry service as an example, the service needs to jump to 2 pages in total. Since the client APP is in the login state, the electricity charge is carried outThe page is inquired, no additional identity information is needed to be input, and only the month needing to be inquired is needed to be selected, so that only the type 1 content exists in each 2 pages in the electricity fee inquiry service. The number of the electricity charge inquiry service in all the electric power APP service sequences is 3, and the content quantitative identification of the two pages is S 3 P 1 C 1 And S 3 P 2 C 1 。
Particularly, if a customer inquires the electricity charge information of a non-self-login account, the customer needs to jump to other pages and manually input a username and a user number, and the service is the other service.
A sorting evaluation index module 20 for sorting the various content amounts obtained by the combing of the client operation content amount combing module 10 and other information in the whole process of the client service into quantifiable data and constructing an evaluation index model E n . The main function of the module is to arrange the content calculated by the functional module and other information in the whole process of customer service into quantifiable data, so that the follow-up self-feedback evaluation under the same scale is facilitated. Only the type 1 content amount exists in the electricity charge inquiry service, so the total content amount can be calculated as S 3 P all C＝k 1 (S 3 P 1 C 1 +S 3 P 2 C 1 ) (ii) a The evaluation index of the two page services can be expressed as: s 3 P all ＝S 3 P 1 +S 3 P 2 The evaluation index model is: e 3 ＝S 3 P all F×S 3 P all . Wherein the dwell time S of the page 3 P all T and number of redundant accesses S to page 3 P n F requires real-time acquisition.
The real-time customer service data acquisition module 30 is used for acquiring customer service data when a customer uses the electric power service APP service function, manually cleaning the data, eliminating partial missing data and adopting a constructed evaluation index model E n A self-feedback service rating is generated in real-time. The main functions of the module are as follows: when the customer uses the electric power service APP service function, the customer use data is collected, and the data is cleaned. First, the customer uses electricityWhen the service is queried, attribute data related to the service is loaded, wherein the attribute data comprises a service number, the number of pages needing to be adjusted in the service and the content volume of each page; secondly, recording the entry time when the client enters the 1 st and 2 nd functional pages; and recording the exit time when exiting. The time difference between the two marks the staying time of the client on the page; finally, according to the real-time data collected in each page, the content volume is recorded and calculated according to the method.
In particular, a page dwell threshold is set, and when the dwell time of the client on the page exceeds the threshold, the data is not used as the evaluation basis.
Particularly, the data cleaning also cleans the page access frequency data, and the access frequency data with obviously higher frequency is not used as evaluation basis. The method can be realized by adopting a fixed value mode, for example, when the redundancy access times of a client on a certain page are more than 3, the redundancy access time data are fixed to be 3.
A semi-supervised dynamic weight grading module 40 for grading the evaluation index model E in a semi-supervised dynamic weight grading mode n Undetermined coefficient k in 1 、k 2 、k 3 And training and solving are carried out, and a foundation is laid for subsequent self-feedback evaluation. The main function of the module is based on the evaluation method and the client use data acquired in real time, and a foundation is laid for subsequent self-feedback evaluation in a semi-supervised weight dynamic grading mode. Firstly, constructing a lightweight deep learning framework InceptitionV 3; secondly, loading 1000 pieces of user data of the client in 1 week as a sample set (comprising samples and sample labels), and performing secondary training on a pre-training model taking historical big data as the sample set; and finally obtaining an evaluation model aiming at the customer electricity charge query service.
A self-feedback customer service evaluation module 50 for obtaining an evaluation index model E n And real-time feedback is pushed to the client, and the evaluation model is further optimized through the feedback of the client. Firstly, evaluating whether customer service is satisfied by a customer in real time according to an evaluation model trained by the module, and pushing a result to the customer; secondly, if the difference between the collected feedback information of the client and the evaluation is larger, the client is to be updatedThe feedback information of the data acquisition system is used as a data sample label, and the sample is fed back to a calculation server to re-optimize the training evaluation model; finally, through repeated training of a large amount of real data of customers, self-feedback evaluation can be carried out on customer service in real time, and meanwhile, the performance of an electric power APP service evaluation model can be promoted.
Those of skill would further appreciate that the exemplary elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various exemplary components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the technical solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in random access memory, read only memory, electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
The above description is only an embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. A power customer self-feedback service evaluation method based on weight dynamic classification is characterized by comprising the following steps: the method comprises the following steps:
step one, carding the operation content of a customer: in each specific service in the electric power service APP, combing the operation flow required to be handled by a client, and recording the content volume of data required by the client to handle a specific task;
step two, arranging evaluation indexes: arranging various contents obtained by combing in the step one and other information in the whole process of customer service into quantifiable data, and constructing an evaluation index model E n ；
Step three, collecting real-time customer service data: when the customer uses the electric power service APP service function, collecting customer use data, manually cleaning the data, removing partial missing data, and adopting the evaluation index model E constructed in the second step n Generating a self-feedback service evaluation in real time;
step four, evaluating the index model E in a semi-supervised weight dynamic grading mode n Undetermined coefficient k in 1 、k 2 、k 3 Training and solving are carried out, and a foundation is laid for subsequent self-feedback evaluation;
step five, self-feedback customer service evaluation: the evaluation index model E obtained in the step two n And real-time feedback is pushed to the client, and the evaluation model is further optimized through the feedback of the client.
2. The power customer self-feedback service evaluation method based on weight dynamic ranking of claim 1, wherein: the first step specifically comprises:
step 1.1, combing the number of pages needing to be jumped for each service, wherein each service is represented by S and respectively recorded as: s. the 1 、S 2 、S 3 、…S n Where the numbers 1 to n denote the nth service, in S n P all Represents the minimum number of pages, P, to be skipped for the client to complete the whole process in the nth client service all All pages representing a certain service;
step 1.2, in each page of each specific service, the content amount which is manually input by a client is combed and recorded, and the input content is divided into 3 types: category 1 is content input by customer selection of options; the 2 nd category is contents which can be quickly and directly filled by a user, such as name, native place, age, mobile phone number and the like; type 3 is the content that the user needs to look up external data or input by interaction, including verification codes and some information that the user cannot recite directly;
step 1.3, determining that the content of data required by a certain client for handling a specific task in the electric power service APP is the sum of all 3 types of content in all process pages contained in the task.
3. The power customer self-feedback service evaluation method based on weight dynamic grading according to claim 2, characterized in that: the content amount of the content required to be input by the 3 types of clients is combed and recorded by the following method:
a) The 1 st type is the content that needs the customer to read and select, record with the character number in the content that the user needs to read, wherein the chinese character is recorded with 2 characters, and the letter and digit are recorded with 1 character, then the 1 st type content volume of the 2 nd page in the 3 rd service in electric power service APP is expressed as S 3 P 2 C 1 ；
b) Class 2 is a category in which the user can fill in the content directly, also using the number of characters as a record, denoted C 2 ；
c) Similarly, type 3 content is recorded in the number of characters, denoted C 3 。
4. The power customer self-feedback service evaluation method based on weight dynamic grading according to claim 1, characterized in that: the second step specifically comprises:
step 2.1, establishing a content evaluation model for interaction required by all customers of each specific service according to the content amount combed in the step one, wherein the content evaluation model is used for evaluating the content amount of each specific service and is expressed as:
S n P all C＝k 1 S n P all C 1 +k 2 S n P all C 2 +k 3 S n P all C 3
wherein k is 1 、k 2 、k 3 Linear superposition is adopted for the undetermined coefficient indicating the 3 types of content, and the undetermined coefficient provides guarantee for the dynamic distribution of subsequent weights;
step 2.2, recording the staying time of each page as S n P all T＝S n P 1 T+
S n P 2 T+……+S n P n T；
Step 2.3, regarding a certain page, taking the ratio of the page content and the residence time of a customer in the page in the customer service as one of the page service evaluation indexes, and expressing the ratio as S n P n ＝S n P all C/S n P n And T, the evaluation indexes of all the pages of the whole certain service are expressed as follows:
S n P all ＝S n P 1 +S n P 2 +……+S n P n
step 2.4, sorting the access times of each page, using the access times as influence factors to evaluate whether redundant operation exists in the power service APP used by the customer so as to influence service evaluation, and using S n P n F represents the redundancy access frequency of the P page in the nth service, and the redundancy access frequency is the actual access frequency minus 1;
step 2.5, S for step 2.3 and step 2.4 n P all And S n P n And F, combining an evaluation index model for evaluating the nth service effect, wherein the evaluation index model is expressed as follows:
E n ＝S n P all F×S n P all
wherein S n P all F＝S n P 1 F+S n P 2 F+……+S n P n F。
5. The power customer self-feedback service evaluation method based on weight dynamic grading according to claim 1, characterized in that: the third step specifically comprises:
step 3.1, when a customer uses a specific service, loading attribute data related to the service, including a service number, the number of pages needing to be adjusted in the service and the content of each page;
step 3.2, when collecting that a client enters a certain function page, recording the entering time; recording the exit time when exiting, and calculating the time difference between the exit time and the exit time to identify the staying time of the client on the page;
step 3.3, collecting the interactive data of the customers in each page, and recording and calculating the content volume according to the method, wherein the privacy information related to the users only calculates the content volume data and does not record specific contents;
3.4, after the current operation of the client is finished, manually cleaning the data, removing part of missing data, and then adopting the evaluation index model E constructed by the steps n And generating self-feedback service evaluation in real time, wherein the evaluation result provides quantitative data support for the service evaluation of the power marketing department, and then emptying all data of customers.
6. The power customer self-feedback service evaluation method based on weight dynamic grading according to claim 1, characterized in that: the fourth step specifically comprises:
step 4.1, collecting usage data of a '5-star evaluation' client and a '1-star evaluation' client, respectively identifying good user experience and poor user experience, wherein the collected data comprises all data in the step two;
step 4.2, calibrating the quantitative data used by the client as a training sample, calibrating the 'good evaluation' and the 'poor evaluation' as sample labels, constructing a deep neural network training framework, and training a service evaluation model;
4.3, the deep learning training framework comprises VGG 19 and IncepotionV 3, the training process is divided into two stages, the first-stage model is a pre-training model of historical big data, the training of the basic model is realized, and the training is realized by adopting the VGG 19 training framework; the second-stage model adopts a lightweight deep learning framework Inception V3 to realize the rapid and secondary training of the pre-training model in a real-time environment, and a sample set of the secondary training comes from user data of a recent customer;
and 4.4, selecting 1000-5000 groups of small-scale customer data as samples according to different service contents, wherein the samples comprise all data in the step two, and the labels of the samples are 'good evaluation' and 'poor evaluation'.
7. The power customer self-feedback service evaluation method based on weight dynamic grading according to claim 1, characterized in that: the fifth step specifically comprises:
step 5.1, according to the evaluation index model E obtained in the step two n Evaluating whether the customer service is satisfied by the customer in real time, and pushing a result to the customer;
step 5.2, pushing self-service functionality and friendliness comprehensive self-evaluation of the electric power service APP to the client, and meanwhile collecting feedback information of the user;
step 5.3, if the feedback information of the client is not collected, the evaluation capability of the default evaluation model is more accurate;
and 5.4, if the difference between the collected feedback information of the client and the evaluation is larger, using the new feedback information as a data sample label, and feeding the sample back to the computing server to optimize the training evaluation model again.
8. A power customer self-feedback service evaluation system based on weight dynamic grading is characterized in that: the method comprises the following steps:
the client operation content combing module is used for combing operation flows needing to be handled by a client in each specific service in the electric power service APP and recording the content of data needed by the client to handle a specific task;
the arrangement evaluation index module is used for arranging various content amounts obtained by combing the client operation content amount combing module and other information in the whole client service process into quantifiable data and constructing an evaluation index model E n ；
The real-time customer service data acquisition module is used for acquiring customer service data when a customer uses an electric power service APP service function, manually cleaning the data, removing partial missing data and adopting a constructed evaluation index model E n Generating a self-feedback service evaluation in real time;
semi-supervisionA weight dynamic grading module used for grading the evaluation index model E in a semi-supervised weight dynamic grading mode n Undetermined coefficient k in 1 、k 2 、k 3 Training and solving are carried out, and a foundation is laid for subsequent self-feedback evaluation;
a self-feedback customer service evaluation module for obtaining an evaluation index model E n And real-time feedback is pushed to the client, and the evaluation model is further optimized through the feedback of the client.
9. The power customer self-feedback service evaluation system based on dynamic ranking of weights of claim 8 wherein: the client operation content carding module combs the operation flow that the client needs to handle in each item of specific service in the electric power service APP, and records the content of data required by the client to handle a specific task, and the method specifically comprises the following steps:
combing the number of pages needing to be jumped for each service, wherein each service is represented by S and respectively recorded as: s. the 1 、S 2 、S 3 、…S n Wherein the numbers 1 to n represent the nth service, with S n P all Represents the minimum number of pages, P, to be skipped for the client to complete the whole process in the nth client service all All pages representing a certain service;
in each page for combing each specific service, the client needs to manually input the content amount and record, and the input content is divided into 3 types: category 1 is content input by customer selection of options; the 2 nd category is contents which can be quickly and directly filled by a user, such as name, native place, age, mobile phone number and the like; type 3 is the content that the user needs to look up external data or input by interaction, including verification codes and some information that the user cannot recite directly;
determining that the content of data required by a certain client for handling a specific task in the power service APP is the sum of all 3 types of content in all process pages contained in the task.
Step 1.1, carding the number of pages required to be jumped for each service, wherein each service is represented by S and respectively recorded as: s. the 1 、S 2 、S 3 、…S n Where the numbers 1 to n denote the nth service, in S n P all Represents the minimum number of pages, P, to be skipped for the client to complete the whole process in the nth client service all Representing all pages of a certain service.
10. The power customer self-feedback service evaluation system based on dynamic ranking of weights of claim 8 wherein: the arrangement evaluation index module arranges various content amounts obtained by combing the client operation content amount combing module and other information in the whole process of client service into quantifiable data, and constructs an evaluation index model E n The method specifically comprises the following steps:
establishing a content evaluation model for interaction required by all customers of each specific service for the content amount combed by the customer operation content amount combing module, wherein the content evaluation model is used for evaluating the content amount of each specific service and is expressed as:
S n P all C＝k 1 S n P all C 1 +k 2 S n P all C 2 +k 3 S n P all C 3
wherein k is 1 、k 2 、k 3 Linear superposition is adopted for the undetermined coefficient indicating the 3 types of content, and the undetermined coefficient provides guarantee for the dynamic distribution of subsequent weights;
recording the stay time of each page as S n P all T＝S n P 1 T+
S n P 2 T+……+S n P n T；
For a certain page, the ratio of the content of the page to the stay time of a customer in the page in customer service is taken as one of the page service evaluation indexes and is expressed as S n P n ＝S n P all C/S n P n And T, the evaluation indexes of all the pages of the whole certain service are expressed as follows:
S n P all ＝S n P 1 +S n P 2 +……+S n P n
arranging the access times of each page, using the access times as influence factors to evaluate whether redundant operation exists in the electric power service APP used by the client or not so as to influence service evaluation, and using S n P n F represents the redundancy access frequency of the P page in the nth service, and the redundancy access frequency is the actual access frequency minus 1;
to S n P all And S n P n F is combined into an evaluation index model for evaluating the nth service effect, and the evaluation index model is expressed as:
E n ＝S n P all F×S n P all
wherein S n P all F＝S n P 1 F+S n P 2 F+……+S n P n F。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211513035.6A CN115719183A (en) | 2022-11-29 | 2022-11-29 | Power customer self-feedback service evaluation method and system based on weight dynamic grading |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211513035.6A CN115719183A (en) | 2022-11-29 | 2022-11-29 | Power customer self-feedback service evaluation method and system based on weight dynamic grading |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115719183A true CN115719183A (en) | 2023-02-28 |
Family
ID=85256997
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211513035.6A Pending CN115719183A (en) | 2022-11-29 | 2022-11-29 | Power customer self-feedback service evaluation method and system based on weight dynamic grading |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115719183A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117093782A (en) * | 2023-10-20 | 2023-11-21 | 国网智能科技股份有限公司 | Electric power artificial intelligence model system and method |
-
2022
- 2022-11-29 CN CN202211513035.6A patent/CN115719183A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117093782A (en) * | 2023-10-20 | 2023-11-21 | 国网智能科技股份有限公司 | Electric power artificial intelligence model system and method |
| CN117093782B (en) * | 2023-10-20 | 2024-03-12 | 国网智能科技股份有限公司 | Electric power artificial intelligence model system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| AU2011325860B2 (en) | Choice modelling system and method | |
| CN109784959B (en) | Target user prediction method and device, background server and storage medium | |
| CN113535963B (en) | Long text event extraction method and device, computer equipment and storage medium | |
| CN112001754A (en) | User portrait generation method, device, equipment and computer readable medium | |
| CN115719183A (en) | Power customer self-feedback service evaluation method and system based on weight dynamic grading | |
| CN116091113A (en) | Marketing model data processing method, system and computer readable storage medium | |
| CN111680213B (en) | Information recommendation method, data processing method and device | |
| CN114491255A (en) | Recommendation method, system, electronic device and medium | |
| CN110555749A (en) | credit behavior prediction method and device based on neural network | |
| CN114004691A (en) | Line scoring method, device, equipment and storage medium based on fusion algorithm | |
| CN110209944B (en) | Stock analyst recommendation method and device, computer equipment and storage medium | |
| CN112395499A (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN113836388A (en) | Information recommendation method and device, server and storage medium | |
| Hwang | Would satisfaction with smart speakers transfer into loyalty towards the smart speaker provider? | |
| CN114493686A (en) | Operation content generation and pushing method and device | |
| CN114944219A (en) | Mental scale recommendation method and device based on artificial intelligence and storage medium | |
| CN114741606A (en) | Enterprise recommendation method and device, computer readable medium and electronic equipment | |
| CN113536103B (en) | Information recommendation method and device, electronic equipment and storage medium | |
| CN113128739A (en) | Prediction method of user touch time, prediction model training method and related device | |
| CN113065911A (en) | Recommendation information generation method and device, storage medium and electronic equipment | |
| Sreejesh et al. | Business Research Process | |
| CN113836391B (en) | Resource recommendation method, device, computer equipment and storage medium | |
| Ramaul | Role of AI in marketing through CRM integration with specific reference to chatbots | |
| Krishna et al. | Cultivating Customer Purchase Intent: Leveraging Machine Learning for Precise Predictions | |
| CN114119078A (en) | Target resource determination method, device, electronic equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |