CN116702982A - User complaint prediction method, device and storage medium - Google Patents
User complaint prediction method, device and storage medium Download PDFInfo
- Publication number
- CN116702982A CN116702982A CN202310675071.0A CN202310675071A CN116702982A CN 116702982 A CN116702982 A CN 116702982A CN 202310675071 A CN202310675071 A CN 202310675071A CN 116702982 A CN116702982 A CN 116702982A
- Authority
- CN
- China
- Prior art keywords
- user
- complaint
- data
- hidden layer
- model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 80
- 238000000605 extraction Methods 0.000 claims abstract description 62
- 230000009467 reduction Effects 0.000 claims abstract description 62
- 238000004891 communication Methods 0.000 claims abstract description 25
- 210000002569 neuron Anatomy 0.000 claims description 76
- 238000006243 chemical reaction Methods 0.000 claims description 74
- 238000012545 processing Methods 0.000 claims description 41
- 238000003062 neural network model Methods 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 30
- 239000011159 matrix material Substances 0.000 claims description 26
- 230000009466 transformation Effects 0.000 claims description 22
- 230000001174 ascending effect Effects 0.000 claims description 14
- 238000013527 convolutional neural network Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 abstract description 23
- 230000006870 function Effects 0.000 description 28
- 230000004913 activation Effects 0.000 description 12
- 238000009434 installation Methods 0.000 description 9
- 238000005259 measurement Methods 0.000 description 8
- 230000003287 optical effect Effects 0.000 description 8
- 238000012423 maintenance Methods 0.000 description 7
- 230000003213 activating effect Effects 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 239000004973 liquid crystal related substance Substances 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000008439 repair process Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 239000013307 optical fiber Substances 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000001308 synthesis method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/01—Customer relationship services
- G06Q30/015—Providing customer assistance, e.g. assisting a customer within a business location or via helpdesk
- G06Q30/016—After-sales
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Development Economics (AREA)
- Databases & Information Systems (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application provides a user complaint prediction method, a device and a storage medium, relates to the technical field of communication, and can predict the complaint probability of a user. The method comprises the following steps: determining user complaint data; inputting user portrait data of a user to be predicted into a first model to extract hidden layer characteristics, and determining first characteristic data; the first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data; inputting the first characteristic data into a second model for convolutional layer characteristic extraction, and determining second characteristic data; the second model is used for extracting the characteristics of the convolution layer at least once for the data; and inputting the second characteristic data into a third model for complaint prediction, determining the complaint probability of the user, wherein the third model is used for carrying out at least one hidden layer characteristic dimension reduction on the data to determine the third characteristic data, and predicting the complaint probability of the user based on the third characteristic data. The embodiment of the application is used in the user complaint prediction process.
Description
Technical Field
The present application relates to the field of communications technologies, and in particular, to a method and apparatus for predicting complaints of users, and a storage medium.
Background
Complaints may be made when the user is not satisfied with the network or service of the operator, and excessive user complaints may lead to a decrease in the public praise of the operator and loss of the user. Therefore, operators need to predict potential complaint users in time, actively know the user demands before the complaint of the users, and solve the problems of the users. The main mode of the current prediction complaint user depends on manual prediction, and staff with abundant working experience is required to analyze the network experience parameters of the user in a polling mode during manual prediction, so that the process is time-consuming and labor-consuming.
Disclosure of Invention
The application provides a user complaint prediction method, a device and a storage medium, which solve the problem that the manual prediction of user complaints in the prior art is time-consuming and labor-consuming.
In order to achieve the above purpose, the application adopts the following technical scheme:
in a first aspect, a method for predicting complaints of a user is provided, the method comprising: determining user portrait data of a user to be predicted; wherein the user profile data is used to identify data related to user complaints from a plurality of dimensions; inputting user portrait data of a user to be predicted into a first model to extract hidden layer characteristics, and determining first characteristic data; the first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data; inputting the first characteristic data into a second model for convolutional layer characteristic extraction, and determining second characteristic data; the second model is used for extracting the characteristics of the convolution layer at least once for the data; and inputting the second characteristic data into a third model for complaint prediction, determining the complaint probability of the user to be predicted, wherein the third model is used for carrying out at least one hidden layer characteristic dimension reduction on the data to determine the third characteristic data, and predicting the complaint probability of the user based on the third characteristic data.
With reference to the first aspect, in one possible implementation manner, inputting user portrait data of a user to be predicted into a first model to perform hidden layer feature extraction, determining first feature data includes: inputting user portrait data of a user to be predicted into a first hidden layer of a first model to perform ascending and nonlinear conversion to obtain characteristics of the first hidden layer; the first hidden layer comprises n neurons, wherein n is a positive integer; inputting the first hidden layer characteristics into a second hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain the second hidden layer characteristics; the second hidden layer comprises m neurons, m is a positive integer less than N; inputting the second hidden layer characteristics into a third hidden layer of the first model to perform ascending and nonlinear conversion to obtain third hidden layer characteristics; the third hidden layer includes n neurons; inputting the third hidden layer characteristics into a fourth hidden layer of the first model for dimension reduction and nonlinear conversion to obtain fourth hidden layer characteristics; the fourth hidden layer includes m neurons; inputting the characteristics of the fourth hidden layer into a fifth hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain first characteristic data; the fifth hidden layer includes n neurons.
With reference to the first aspect, in one possible implementation manner, inputting the first feature data into the second model to perform convolutional layer feature extraction, and determining the second feature data includes: inputting the first characteristic data into a first convolution layer of a second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a first convolution layer characteristic; the first convolution layer comprises a convolution kernel of a×a; a is a positive integer; inputting the first characteristic data into a second convolution layer of a second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain second convolution layer characteristics; the second convolution layer includes a bXb convolution kernel; b is a positive integer greater than a; inputting the first characteristic data into a third convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a third convolution layer characteristic; the third convolution layer includes a c×c convolution kernel; c is a positive integer greater than b; channel splicing is carried out on the first convolution layer characteristics, the second convolution layer characteristics and the third convolution layer characteristics, and a target characteristic matrix is determined; inputting the target feature matrix into a fourth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fourth convolution layer features; a convolution kernel of b x b is included in the fourth convolution layer; inputting the target feature matrix into a fifth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fifth convolution layer features; the fifth convolution layer includes a convolution kernel of a×a; inputting the target feature matrix into a sixth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain second feature data; the sixth convolution layer includes a bxb convolution kernel.
With reference to the first aspect, in one possible implementation manner, the inputting the second feature data into the third model to perform complaint prediction, and determining the complaint probability of the user includes: inputting the second characteristic data into a sixth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the characteristics of the sixth hidden layer; the sixth hidden layer includes m neurons; inputting the sixth hidden layer characteristics into a seventh hidden layer of the third model to perform ascending and nonlinear transformation, and determining the seventh hidden layer characteristics; the seventh hidden layer includes n neurons; inputting the seventh hidden layer characteristics into an eighth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the eighth hidden layer characteristics; the eighth hidden layer includes m neurons; inputting the eighth hidden layer characteristics into a ninth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the ninth hidden layer characteristics; the ninth hidden layer includes r neurons; r is a positive integer less than m; inputting the ninth hidden layer characteristics into a tenth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the tenth hidden layer characteristics; the tenth hidden layer includes s neurons; s is a positive integer less than r; and inputting the tenth hidden layer characteristics into an output layer of the third model to carry out probability prediction, and determining the complaint probability of the user.
With reference to the first aspect, in one possible implementation manner, before the user portrait data of the user to be predicted is input into the first model to perform hidden layer feature extraction, determining first feature data further includes: determining user image data of complaint users and user image data of non-complaint users; based on the complaint type of the complaint user, determining a complaint weight value of the complaint user; based on the complaint weight value of the complaint user, carrying out weighted calculation on the user portrait data of the complaint user to determine the weighted user portrait data of the complaint user; training a first fully connected neural network model, a convolutional neural network model and a second fully connected neural network model based on weighted user portrayal data of a complaint user and non-complaint user data, and determining a first model, a second model and a third model; the method comprises the steps of training a first model through a first full-connection neural network model, training a second model through a convolution neural network model, and training a third model through a second full-connection neural network model.
With reference to the first aspect, in one possible implementation manner, determining user image data of a complaint user includes: acquiring user image data of a history complaint user; determining whether the number of user image data of the historic complaint user reaches a first number; if the preset number is not reached, determining user image data of a second number of virtual complaint users according to the user image data of the historical complaint users; the second number is determined based on a difference between the number of user representation data of the historic complaint user and the first number; the user profile data for the determined complaint user includes user profile data for the historical complaint user and user profile data for a second number of virtual complaint users.
With reference to the first aspect, in one possible implementation manner, determining, according to user image data of the historical complaint user, user image data of a second number of virtual complaint users includes: performing the following target operations a plurality of times to determine user image data for a second number of virtual complaint users; the target operations include: acquiring user image data of a first historical complaint user and user image data of a second historical complaint user; the user image data of the first historical complaint user and the user image data of the second historical complaint user are data in the user image data of the historical complaint user; and adjusting the data of the target dimension in the user image data of the first historical complaint user to the data of the target dimension in the user image data of the second historical complaint user, and determining the user image data of the first virtual complaint user.
In a second aspect, there is provided a user complaint predicting apparatus comprising: a processing unit; the processing unit is used for determining user complaint data; the processing unit is also used for inputting user portrait data of the user to be predicted into the first model to extract hidden layer characteristics and determine first characteristic data; the first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data; the processing unit is also used for inputting the first characteristic data into the second model to perform convolutional layer characteristic extraction and determine second characteristic data; the second model is used for extracting the characteristics of the convolution layer at least once for the data; the processing unit is further used for inputting the second characteristic data into a third model to conduct complaint prediction, determining the complaint probability of the user, wherein the third model is used for conducting at least one hidden layer characteristic dimension reduction on the data to determine the third characteristic data, and predicting the complaint probability of the user based on the third characteristic data.
With reference to the second aspect, in one possible implementation manner, the processing unit is specifically configured to: inputting user portrait data of a user to be predicted into a first hidden layer of a first model to perform ascending and nonlinear conversion to obtain characteristics of the first hidden layer; the first hidden layer comprises n neurons, wherein n is a positive integer; inputting the first hidden layer characteristics into a second hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain the second hidden layer characteristics; the second hidden layer comprises m neurons, m is a positive integer less than N; inputting the second hidden layer characteristics into a third hidden layer of the first model to perform ascending and nonlinear conversion to obtain third hidden layer characteristics; the third hidden layer includes n neurons; inputting the third hidden layer characteristics into a fourth hidden layer of the first model for dimension reduction and nonlinear conversion to obtain fourth hidden layer characteristics; the fourth hidden layer includes m neurons; inputting the characteristics of the fourth hidden layer into a fifth hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain first characteristic data; the fifth hidden layer includes n neurons.
With reference to the second aspect, in one possible implementation manner, the processing unit is specifically configured to: inputting the first characteristic data into a first convolution layer of a second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a first convolution layer characteristic; the first convolution layer comprises a convolution kernel of a×a; a is a positive integer; inputting the first characteristic data into a second convolution layer of a second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain second convolution layer characteristics; the second convolution layer includes a bXb convolution kernel; b is a positive integer greater than a; inputting the first characteristic data into a third convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a third convolution layer characteristic; the third convolution layer includes a c×c convolution kernel; c is a positive integer greater than b; channel splicing is carried out on the first convolution layer characteristics, the second convolution layer characteristics and the third convolution layer characteristics, and a target characteristic matrix is determined; inputting the target feature matrix into a fourth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fourth convolution layer features; a convolution kernel of b x b is included in the fourth convolution layer; inputting the target feature matrix into a fifth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fifth convolution layer features; the fifth convolution layer includes a convolution kernel of a×a; inputting the target feature matrix into a sixth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain second feature data; the sixth convolution layer includes a bxb convolution kernel.
With reference to the second aspect, in one possible implementation manner, the processing unit is specifically configured to: inputting the second characteristic data into a sixth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the characteristics of the sixth hidden layer; the sixth hidden layer includes m neurons; inputting the sixth hidden layer characteristics into a seventh hidden layer of the third model to perform ascending and nonlinear transformation, and determining the seventh hidden layer characteristics; the seventh hidden layer includes n neurons; inputting the seventh hidden layer characteristics into an eighth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the eighth hidden layer characteristics; the eighth hidden layer includes m neurons; inputting the eighth hidden layer characteristics into a ninth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the ninth hidden layer characteristics; the ninth hidden layer includes r neurons; r is a positive integer less than m; inputting the ninth hidden layer characteristics into a tenth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the tenth hidden layer characteristics; the tenth hidden layer includes s neurons; s is a positive integer less than r; and inputting the tenth hidden layer characteristics into an output layer of the third model to carry out probability prediction, and determining the complaint probability of the user.
With reference to the second aspect, in a possible implementation manner, the processing unit is further configured to: determining user image data of complaint users and user image data of non-complaint users; based on the complaint type of the complaint user, determining a complaint weight value of the complaint user; based on the complaint weight value of the complaint user, carrying out weighted calculation on the user portrait data of the complaint user to determine the weighted user portrait data of the complaint user; training a first fully connected neural network model, a convolutional neural network model and a second fully connected neural network model based on weighted user portrayal data of a complaint user and non-complaint user data, and determining a first model, a second model and a third model; the method comprises the steps of training a first model through a first full-connection neural network model, training a second model through a convolution neural network model, and training a third model through a second full-connection neural network model.
With reference to the second aspect, in one possible implementation manner, the apparatus further includes: a communication unit; the communication unit is also used for acquiring user image data of the history complaint user; the processing unit is also used for determining whether the number of the user image data of the history complaint user reaches a first number; the processing unit is further used for determining user image data of a second number of virtual complaint users according to the user image data of the historical complaint users if the preset number is not reached; the second number is determined based on a difference between the number of user representation data of the historic complaint user and the first number; the processing unit is further configured to determine that the user profile data of the complaint user includes user profile data of a historical complaint user and user profile data of a second number of virtual complaint users.
With reference to the second aspect, in one possible implementation manner, the processing unit is specifically configured to: performing the following target operations a plurality of times to determine user image data for a second number of virtual complaint users; the target operations include: acquiring user image data of a first historical complaint user and user image data of a second historical complaint user; the user image data of the first historical complaint user and the user image data of the second historical complaint user are data in the user image data of the historical complaint user; and adjusting the data of the target dimension in the user image data of the first historical complaint user to the data of the target dimension in the user image data of the second historical complaint user, and determining the user image data of the first virtual complaint user.
In a third aspect, the present application provides a user complaint predicting apparatus comprising: a processor and a communication interface; the communication interface is coupled to a processor for running a computer program or instructions to implement the user complaint prediction method as described in any one of the possible implementations of the first aspect and the first aspect.
In a fourth aspect, the present application provides a computer readable storage medium having instructions stored therein which, when run on a terminal, cause the terminal to perform a method of predicting a user complaint as described in any one of the possible implementations of the first aspect and the first aspect.
In a fifth aspect, embodiments of the present application provide a computer program product comprising instructions that, when run on a user complaint prediction apparatus, cause the user complaint prediction apparatus to perform a user complaint prediction method as described in any one of the possible implementations of the first aspect and the first aspect.
In a sixth aspect, embodiments of the present application provide a chip comprising a processor and a communication interface, the communication interface and the processor being coupled, the processor being for running a computer program or instructions to implement a method of customer complaint prediction as described in any one of the possible implementations of the first aspect and the first aspect.
Specifically, the chip provided in the embodiment of the application further includes a memory, which is used for storing a computer program or instructions.
The scheme at least brings the following beneficial effects: in the technical scheme provided by the application, the user complaint prediction device inputs user portrait data of a user to be predicted into the extraction of the hidden layer characteristics, the extraction of the convolution layer characteristics and the extraction of the second hidden layer characteristics in sequence, and predicts the complaint probability of the user based on the finally determined characteristics. Therefore, the user complaint prediction device can directly predict the complaint probability of the user according to the portrait data of the user and the preset network model, and the problem that time and labor are wasted during manual prediction in the prior art is avoided.
Drawings
FIG. 1 is a schematic diagram of a system architecture of a customer complaint prediction system according to the present application;
FIG. 2 is a schematic diagram of a device for predicting complaints of users according to the present application;
FIG. 3 is a flow chart of a method for predicting customer complaints provided by the present application;
FIG. 4 is a flow chart of another method for predicting customer complaints provided by the present application;
FIG. 5 is a flow chart of yet another method for predicting customer complaints provided by the present application;
FIG. 6 is a flow chart of yet another method for predicting customer complaints provided by the present application;
fig. 7 is a schematic structural diagram of a user complaint predicting device provided by the application.
Detailed Description
The user complaint prediction device, device and storage medium provided by the embodiment of the application are described in detail below with reference to the accompanying drawings.
The term "and/or" is herein merely an association relationship describing an associated object, meaning that there may be three relationships, e.g., a and/or B, may represent: a exists alone, A and B exist together, and B exists alone.
The terms "first" and "second" and the like in the description and in the drawings are used for distinguishing between different objects or between different processes of the same object and not for describing a particular order of objects.
Furthermore, references to the terms "comprising" and "having" and any variations thereof in the description of the present application are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed but may optionally include other steps or elements not listed or inherent to such process, method, article, or apparatus.
It should be noted that, in the embodiments of the present application, words such as "exemplary" or "such as" are used to mean serving as an example, instance, or illustration. Any embodiment or design described herein as "exemplary" or "e.g." in an embodiment should not be taken as preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "such as" is intended to present related concepts in a concrete fashion.
Current network operators are increasingly paying attention to the network experience of users. User complaints are taken as important indexes of network experience of users, so that dissatisfaction of network experience (network quality experience, user service experience and the like) of operators of the users is intuitively expressed, and excessive user complaints can cause the public praise of the operators to be reduced. Therefore, the current operator needs to predict the possible complaint users before complaint of the users, solve the possible complaint hidden danger of the users in advance, and improve the service quality and the service level. Therefore, how to predict complaint users is a technical problem to be solved currently.
In the related art, a manual prediction method is generally adopted when predicting complaint users. The method specifically comprises the following steps: the staff tests the network quality of the user on site, such as broadband speed measurement, wireless signal intensity measurement, wireless coverage and the like, judges whether the network quality meets the standard or not based on the tested signals, and complaints and hidden dangers exist if the network quality does not meet the standard. However, the method requires a staff to have abundant working experience, and the judging process is time-consuming and labor-consuming, so that the method is difficult to popularize and maintain.
In order to solve the problem of trouble and effort in the process of manually predicting user complaints in the related art, the embodiment of the application provides a complaint user complaint prediction method, wherein a user complaint prediction device sequentially inputs user portrait data of a user to be predicted into extraction of hidden layer features, extraction of convolution layer features and extraction of secondary hidden layer features, and predicts the complaint probability of the user based on the finally determined features. Therefore, the user complaint prediction device can directly predict the complaint probability of the user according to the portrait data of the user and the preset network model, and the problem that time and labor are wasted during manual prediction in the prior art is avoided.
The user complaint prediction method provided by the embodiment of the application can be applied to the user complaint prediction system shown in figure 1. As shown in fig. 1, the user complaint prediction system 10 includes: user complaint prediction apparatus 101 and data server 102.
Wherein, the user complaint predicting device 101 is used for obtaining user portrait data related to user complaints from the data server 102; the user portrait data are then input into a preset model to determine the complaint probability of the user.
Data server 102 is used to provide user profile data associated with user complaints to user complaint prediction device 101.
In one possible implementation of an embodiment of the present application, the basic hardware structure of user complaint prediction apparatus 101 in a user complaint prediction system includes elements included in user complaint prediction apparatus 200 shown in fig. 2. The hardware configuration of user complaint predicting apparatus 200 will be described below using user complaint predicting apparatus 200 shown in fig. 2 as an example.
As shown in FIG. 2, the user complaint prediction device 200 includes at least one processor 201, a communication line 202, and at least one communication interface 204, and may also include a memory 203. The processor 201, the memory 203, and the communication interface 204 may be connected through a communication line 202.
The processor 201 may be a central processing unit (central processing unit, CPU), an application specific integrated circuit (application specific integrated circuit, ASIC), or one or more integrated circuits configured to implement embodiments of the present application, such as: one or more digital signal processors (digital signal processor, DSP), or one or more field programmable gate arrays (field programmable gate array, FPGA).
Communication line 202 may include a path for communicating information between the above-described components.
The communication interface 204, for communicating with other devices or communication networks, may use any transceiver-like device, such as ethernet, radio access network (radio access network, RAN), WLAN, etc.
The memory 203 may be, but is not limited to, a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (random access memory, RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (electrically erasable programmable re ad-only memory, EEPROM), a compact disc read-only memory (compact disc read-only memory) or other optical disc storage, a compact disc storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to include or store the desired program code in the form of instructions or data structures and that can be accessed by a computer.
In one possible design, the memory 203 may exist independent of the processor 201, that is, the memory 203 may be a memory external to the processor 201, where the memory 203 may be connected to the processor 201 through a communication line 202, for storing execution instructions or application program codes, and the execution is controlled by the processor 201, to implement a user complaint prediction method provided by the following embodiments of the present application. In yet another possible design, the memory 203 may be integrated with the processor 201, i.e., the memory 203 may be an internal memory of the processor 201, e.g., the memory 203 may be a cache, may be used to temporarily store some data and instruction information, etc.
As one implementation, processor 201 may include one or more CPUs, such as CPU0 and CPU1 in fig. 2. As another implementation, user complaint prediction device 200 may include multiple processors, such as processor 201 and processor 207 in FIG. 2. As yet another implementation, user complaint predicting apparatus 200 may further include an output device 205 and an input device 206.
The method aims to solve the problem that time and labor are wasted when complaint users are predicted manually in the related technology. The application provides a user complaint prediction method which can be applied to a user complaint prediction device as shown in fig. 2. As shown in fig. 3, the user complaint prediction method may be implemented by the following steps 301 to 304.
Step 301, a user complaint prediction device determines user portrait data of a user to be predicted.
Wherein the user profile data is used to identify data related to user complaints from multiple dimensions.
In one possible implementation manner, taking the user to be predicted as a broadband internet user as an example, the user complaint prediction device may divide user portrait data of the user to be predicted into: network data, user data, dimension data, environment data, and reference data.
Wherein the network data comprises various data indexes of broadband provided by the broadband operator, and the data indexes are used for representing objective evaluation of network quality of the broadband provided by the operator to the user. The user data is used for representing personal related data of the user and objective association relation between the user and an operator. The installation and maintenance data are data such as installation and maintenance of the user broadband and are used for representing subjective association relation between the user and the operator. The environmental data is the data of the user and the adjacent users, and is used for representing the geographic environmental factors of the user broadband. The reference data is a data for reflecting the quality of the broadband from the quality according to the IPTV, thereby providing a reference for the accuracy of the prediction data. It should be noted that, in the embodiment of the present application, the user portrait data of the user to be predicted may be divided from other dimensions, which is not limited in the present application. In addition, the user to be predicted may be other types of users, such as a mobile network user, a fixed phone network user, etc., which is not limited by the present application.
Step 302, the user complaint prediction device inputs user portrait data of a user to be predicted into a first model to extract hidden layer features and determine first feature data.
The first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data.
Optionally, the first model is a fully connected neural network model.
It is noted that the first model performs at least one time of hidden layer feature lifting and hidden layer feature dimension reduction on the user image data of the user to be predicted, so that hidden data features in the user image data can be better extracted, and the accuracy of final prediction of user complaint probability is improved.
Step 303, the user complaint prediction device inputs the first feature data into the second model to perform convolutional layer feature extraction, and determines second feature data.
The second model is used for carrying out at least one convolution layer feature extraction on the data.
Optionally, the second model is a convolutional neural network model.
It should be noted that the data processed by the first model of the second model is convolved, so that the data characteristics of the data can be further mined, and the accuracy of final prediction of the complaint probability of the user is improved.
And 304, the user complaint prediction device inputs the second characteristic data into a third model to conduct complaint prediction, and the complaint probability of the user to be predicted is determined.
The third model is used for carrying out at least one hidden layer feature dimension reduction on the data to determine third feature data, and predicting the complaint probability of the user based on the third feature data.
In a possible implementation manner, the third model includes at least one hidden layer for dimension reduction, and an output layer for outputting a prediction result. After the user complaint prediction device inputs the second feature data into the third model, the third model invokes the at least one hidden layer to dimensionality reduce the data. And then, the third model calls the output layer to predict the data after the dimension reduction, and predicts the complaint probability of the user.
The scheme at least brings the following beneficial effects: the embodiment of the application provides a complaint user complaint prediction method, wherein a user complaint prediction device inputs user portrait data of a user to be predicted into a hidden layer feature extraction mode, a convolution layer feature extraction mode and a second hidden layer feature extraction mode in sequence, and predicts the complaint probability of the user based on the finally determined features. Therefore, the user complaint prediction device can directly predict the complaint probability of the user according to the portrait data of the user and the preset network model, and the problem that time and labor are wasted during manual prediction in the prior art is avoided.
In a possible implementation manner, before the method for predicting the complaint probability of the user in the steps 301 to 304, the user complaint prediction device in the embodiment of the present application may first classify the image data of the user and determine the assignment rule, so as to better describe the image data of the complaint user from multiple dimensions.
Optionally, taking the user to be predicted as the broadband internet user as an example, the user portrait data determined by the user complaint predicting device is shown in the following table 1:
table 1, user Portrait data sheet for broadband Internet users
/>
/>
/>
The user portrait data of each entry in table 1 is described as follows:
(1) The broadband uplink speed measurement speed is used for representing and reflecting the uplink network speed provided by an operator to a user, corresponding values are given according to the section where the highest speed is located, the highest speed is given with a value 1 in the section of 0-10MB, the highest speed is given with a value 2 in the section of 10-20MB, the highest speed is given with a value 3 in the section of 20-30MB, the highest speed is given with a value 4 in the section of 30-40MB, the highest speed is given with a value 5 in the section of 40-50MB, the highest speed is given with a value 6 in the section of 50-60MB, the highest speed is given with a value 7 in the section of 60-70MB, the highest speed is given with a value 8 in the section of 70-80MB, the highest speed is given with a value 9 in the section of 80-90MB, the highest speed is given with a value 10 in the section of 90-100MB, and the highest speed is given with a value 11 in the section of 100MB or above.
(2) The broadband downlink speed measurement is used for representing downlink network speeds provided by operators to users, corresponding values are given according to the section where the highest speed is located, the highest speed is given a value 1 in the section of 0-100MB, the highest speed is given a value 2 in the section of 100-200MB, the highest speed is given a value 3 in the section of 200-300MB, the highest speed is given a value 4 in the section of 300-400MB, the highest speed is given a value 5 in the section of 400-500MB, the highest speed is given a value 6 in the section of 500-600MB, the highest speed is given a value 7 in the section of 600-700MB, the highest speed is given a value 8 in the section of 700-800MB, the highest speed is given a value 9 in the section of 800-900MB, the highest speed is given a value 10 in the section of 900-1000MB, and the highest speed is given a value 11 in the section of 1000MB and above.
(3) The optical power is used for representing the optical power of the access network optical fiber provided by an operator, and the optical power is given with the following numerical values: a value greater than-27 dB is assigned 1; -27dB is assigned a value of 3; -28dB is assigned a value of 5; -29dB is assigned a value of 10; -30dB is assigned 13; a value of less than-30 dB is assigned 15.
(4) The GPON port peak flow, which is used to reflect the maximum value of the GPON port flow, is given a numerical value using the ratio of the GPON port peak flow and the maximum flow that the GPON port can carry, as shown in equation 1.
Wherein G is the given value of the peak flow of the GPON port, F MAX For the maximum flow which can be borne by the GPON port, F max Is the GPON port peak flow.
(5) The end network industry mismatch refers to a large gap between the broadband bandwidth provided by the user and the maximum bandwidth that can be supported by the router used by the user, and a numerical value is given by using the ratio of the broadband bandwidth provided by the user and the maximum bandwidth that can be supported by the router used by the user, as shown in formula 2.
Wherein D is a value given by end network mismatch, B MAX Broadband bandwidth provided to subscribers for operators, B max Is the maximum bandwidth that the user router can support.
(6) The number of the line cascades from the OLT to the BASE refers to the number of the devices passing between the OLT device accessed by the user and the BASE server, and the number of the devices passing between the OLT device accessed by the user and the BASE server is directly used for giving a numerical value.
(7) CDN resources refer to whether the CDN resources of the websites accessed by the users are in the province, and according to the preference of the users, CDN resources frequently accessed by the first ten users are arranged, whether the CDN resources are in the province is judged, and the quantity used in the province is given with a numerical value.
(8) The age is used for adding user characteristics into the prediction model, so that the model learns the relation between more users and broadband internet complaints as much as possible, and the numerical value is the quotient obtained by dividing the real age numerical value of the users by 10.
(9) Gender, which is used for helping the model learn the relationship between the user gender and the broadband internet complaint, different values are given according to the gender, the male gives a value of 1, and the female gives a value of 2.
(10) Package monthly rentals are used to represent the amount of money a user spends in the broadband for each month, and a value is assigned to the average monthly amount of consumption of the broadband by the user.
(11) The usage time is used for representing the time of using the broadband by a user, and the number of years is directly taken as a numerical value by calculating the usage time in a year less than one year.
(12) Application preferences reflecting the user's usage preferences for using the broadband, assigning values according to the different usage preferences, navigating the application: 1. communication application: 2. video application: 3. financial application: 4. music application: 5. gaming application: 6. web browsing application: 7. downloading an application: 8. other applications: 9.
(13) The user uplink and downlink speed measurement is used for representing downlink speeds when the user uses the broadband internet daily, corresponding values are assigned according to the interval where the highest speed is located, the highest speed is assigned with a value 1 in the interval of 0-100MB, the highest speed is assigned with a value 2 in the interval of 100-200MB, the highest speed is assigned with a value 3 in the interval of 200-300MB, the highest speed is assigned with a value 4 in the interval of 300-400MB, the highest speed is assigned with a value 5 in the interval of 400-500MB, the highest speed is assigned with a value 6 in the interval of 500-600MB, the highest speed is assigned with a value 7 in the interval of 600-700MB, the highest speed is assigned with a value 8 in the interval of 700-800MB, the highest speed is assigned with a value 9 in the interval of 800-900MB, the highest speed is assigned with a value 10 in the interval of 900-1000MB and the highest speed is assigned with a value 11 in the interval of 1000MB and above.
(14) The uplink speed measurement of the user on the internet is used for representing that the user daily uses the broadband internet, corresponding values are assigned according to the interval of the highest speed, the highest speed is assigned with a value 1 in the interval of 0-10MB, the highest speed is assigned with a value 2 in the interval of 10-20MB, the highest speed is assigned with a value 3 in the interval of 20-30MB, the highest speed is assigned with a value 4 in the interval of 30-40MB, the highest speed is assigned with a value 5 in the interval of 40-50MB, the highest speed is assigned with a value 6 in the interval of 50-60MB, the highest speed is assigned with a value 7 in the interval of 60-70MB, the highest speed is assigned with a value 8 in the interval of 70-80MB, the highest speed is assigned with a value 9 in the interval of 80-90MB, the highest speed is assigned with a value 10 in the interval of 90-100MB, and the highest speed is assigned with a value 11 in the interval of 100MB or above.
(15) Whether the number-carrying and network-transferring (carrying-out) is processed or not, and the given value is processed or not according to the user: is processed: 5, have not been handled: 0.
(16) Historical complaints, which are used for representing complaints of the user, representing tendencies of the complaints, classifying the complaints into three types and giving different values: the letter department complains: 5. group customer service or government hot line complaints (override complaints): 4. common complaints: and 1, adding previous complaints of the user to give a numerical value.
(17) The repeated faults are used for representing the repeated repair situation of the user, and the fault repair work orders with 4 times or more in one month are set as repeated faults, and the number of the fault repair work orders with the repeated faults is directly used for giving a numerical value.
(18) The current time length of the installation is used for reflecting the overlong time length of the current installation, the time length of the installation is set to be one-time record of the installation which is more than 3 days, and the number of days of the installation which is not completed is directly used for giving a numerical value.
(19) The current obstacle repairing non-statement is used for reflecting the condition that the current obstacle repairing duration is overlong, setting the obstacle repairing non-statement exceeding 3 days as one-time record, and directly using the number of days of the non-statement to give a numerical value.
(20) The customer service return visit satisfaction degree of the maintenance service is used for reflecting the satisfaction degree of the user in the events of installation, maintenance and the like, and the average value is given according to the score calculation average value of the customer service return visit satisfaction degree of the maintenance work order of the user.
(21) The complaint condition of the neighbor user refers to other users on the same optical splitter with the user, and the complaint condition also reflects the possibility of complaint of the user, and values are given according to different complaint types: the letter department complains: 5. group customer service or government hot line complaints (override complaints): 4. common complaints: and 1, adding previous complaints of the neighbor users and giving a numerical value by averaging.
(22) And the customer service return visit satisfaction degree of the business service is used for reflecting the satisfaction degree of the user on the business personnel, and the average value is given according to the score calculation average value of the satisfaction degree of the customer service return visit user on the business personnel.
(23) The customer service return visit satisfaction of the adjacent user is used for reflecting the satisfaction degree of the adjacent user in the events of installation, maintenance and the like, and the average value is given according to the customer service to score the customer service return visit satisfaction degree of the once-used maintenance work order.
(24) Whether other operators network is covered in the area reflects whether other operators provide broadband network access in the area where the user is located, and the value is given according to different conditions: non-access: 0, accessed 1.
(25) Broadband access region, which is used to reflect the region where the access broadband is located, and give numerical values according to the different regions: rural area: 1. city: 2.
(26) The IPTV quality difference rate is used for reflecting the IPIV quality, indirectly reflecting the broadband quality and directly using the value of the IPTV quality difference rate to give a numerical value.
(27) The IPTV quality difference rate of the neighbor users is reflected by the users, the broadband quality of the neighbor users is indirectly reflected by the users, and the average value is calculated by using the value of the IPTV quality difference rate to give a numerical value.
Above, the image data related to the user complaints are reflected by 27 parameters in an exemplary way, so that the influence of different parameters on the user complaints can be more comprehensively reflected, and the image data of the user is more accurate.
It should be noted that the parameters in table 1 above are merely exemplary, and more or fewer parameters may be used to describe the user image in a specific application, which is not limited by the present application.
In yet another possible implementation, after the user complaint prediction means classifies the user representation data and determines the assignment rule. The user complaint prediction device can also acquire historical user portrait data, perform assignment, and further perform model training according to the assigned historical user portrait data to obtain the first model, the second model and the third model.
As shown in fig. 4, the process of model training by the user complaint prediction device may be specifically implemented by the following steps 401 to 404, which are described in detail below:
step 401, the user complaint prediction device determines user image data of a complaint user and user image data of a non-complaint user.
In a specific implementation manner, the user complaint prediction device acquires historical data, determines a complaint user and a non-complaint user based on whether the user has complained, and further determines user image data of the complaint user and user image data of the non-complaint user.
It should be noted that the user image data of the complaint user and the user image data of the non-complaint user are both assigned values. Optionally, the user complaint predicting device obtains historical user portrait data as shown in table 1 of the plurality of users, and performs assignment based on assignment rules as shown in table 1.
Step 402, the user complaint predicting device determines a complaint weight value of the complaint user based on the complaint type of the complaint user.
In one possible implementation manner, the user complaint prediction device is based on the complaint condition of the user, and assigns values according to three kinds of complaints of the working department, group customer service complaints and common complaints: 5. and 4, carrying out normalization processing on the samples 1. And determining normalized complaint probability. Illustratively, the normalized complaint probability satisfies the following equation 3:
wherein y is i Represents normalized complaint probability, x i Representing the value before normalization, max () represents the maximum value in x.
Based on the above formula 3, it can be determined that the complaint probability of the user complaining from the normalized letter portion is 1, the complaint probability of the user complaining from the group customer service is 0.8, and the complaint probability of the general complaint user is 0.2.
The user complaint prediction device determines that the complaint probability of the user is recorded as a complaint weight value.
Step 403, the user complaint predicting device performs weighted calculation on the user portrait data of the complaint user based on the complaint weight value of the complaint user to determine weighted user portrait data of the complaint user.
Step 404, the user complaint prediction device trains the first fully-connected neural network model, the convolutional neural network model and the second fully-connected neural network model based on the weighted user portrait data of the complaint user and the non-complaint user data, and determines a first model, a second model and a third model.
The method comprises the steps of training a first model through a first full-connection neural network model, training a second model through a convolution neural network model, and training a third model through a second full-connection neural network model.
In one possible implementation manner, the user complaint prediction device divides total data (including user image data of complaint users and non-complaint users) into a training set and a test set according to a preset proportion (for example, 4:1), trains a first full-connection neural network model, a convolutional neural network model and a second full-connection neural network model based on the training set and the test set, and determines the first model, the second model and the third model.
Based on the first model, the second model and the third model are obtained by training the user complaint prediction device based on the user portrait data of the complaint users and the non-complaint users in the historical data, so that the user complaint prediction model can accurately predict the complaint probability of the users according to the first model, the second model and the third model.
In one possible implementation manner, in the model training process, since the number of non-complaint users in the users is generally much greater than the number of complaint users, in order to avoid the problem of unbalanced training data caused by the small number of complaint users. The user complaint prediction device can generate a large number of user image data of virtual complaint users by adopting a randomly exchanged data synthesis method so that the number of the user image data of the complaint users and the number of the user image data of the non-complaint users are equivalent.
In this case, as shown in fig. 5 in conjunction with fig. 4, before the user complaint prediction device determines the user image data of the complaint user in step 401, the user complaint prediction device may specifically generate the user image data of a plurality of virtual complaint users through the following steps 501 to 504.
Step 501, a user complaint prediction device acquires user image data of a history complaint user.
Step 502, the user complaint predicting device determines whether the number of user image data of the history complaint user reaches the first number.
In one possible implementation, the first number of values is determined based on the number of non-complaint users in the historical data. Illustratively, the historical data includes 10 ten thousand pieces of user image data of the non-complaint user, and the user complaint predicting device may determine the first number of values to be the same as or similar to 10 ten thousand.
Step 503, if the user complaint prediction device does not reach the preset number, determining user image data of a second number of virtual complaint users according to the user image data of the historical complaint users.
Wherein the second number is determined based on a difference between the number of user portrayal data of the historic complaint user and the first number.
In one possible implementation, the user complaint predicting device determines the user image data of the second number of virtual complaint users by performing the following target operations a plurality of times; the target operations include:
step one, a user complaint prediction device acquires user image data of a first historical complaint user and user image data of a second historical complaint user; the user image data of the first historical complaint user and the user image data of the second historical complaint user are both data in the user image data of the historical complaint user.
Step two, the user complaint prediction device adjusts the data of the target dimension in the user image data of the first historical complaint user to the data of the target dimension in the user image data of the second historical complaint user, and determines the user image data of the first virtual complaint user.
For example, in combination with table 1, the user complaint predicting device determines that the target dimension data in the user image data of the first historical complaint user is the broadband uplink speed, and the value of the broadband uplink speed is 5. The first historic complaint user has a value of 8 for the broadband uplink speed measurement. At this time, the user complaint prediction device adjusts the assignment of the broadband uplink speed measurement speed of the first historical complaint user to 8, and the data of the other 26 dimensions are unchanged, so as to obtain the user image data of the first virtual complaint user.
Step 504, the user complaint prediction means determines that the user profile data of the complaint user includes user profile data of the historical complaint user and user profile data of a second number of virtual complaint users.
The scheme at least brings the following beneficial effects: the user complaint prediction device generates user portrait data of a plurality of virtual complaint users in a data exchange mode, ensures that the complaint users are similar to the user portrait data of non-complaint users, ensures that the data amount used in the model training process is more balanced, and improves the accuracy of the model training.
The above description has been given of the classification assignment of user portrait data, the generation process of training data, and the model training process.
Hereinafter, a process of predicting by the user complaint predicting device based on the user portrait data of the user to be predicted will be described in detail.
Referring to fig. 3, as shown in fig. 6, in the above step 302, the user complaint prediction device inputs the user portrait data of the user to be predicted into the first model to perform hidden layer feature extraction, and the process of determining the first feature data may be specifically implemented by the following steps 601 to 605.
And 601, the user complaint prediction device inputs user portrait data of a user to be predicted into a first hidden layer of a first model to perform ascending and nonlinear conversion, so as to obtain the characteristics of the first hidden layer.
The first hidden layer comprises n neurons, wherein n is a positive integer.
In one example, n has a value of 100, that is, the first hidden layer is a hidden layer containing 100 neurons.
In this case, step 601 may be specifically implemented as: the user complaint predicting device inputs the values of the 27-dimensional attributes in the user portrait data of the predicted user to the trained first full connection godAnd calling 100 neurons of a first hidden layer of the first fully-connected neural network model in the network model to perform characteristic dimension lifting on the user portrait data. After the dimension is increased, nonlinear transformation is carried out according to the Tanh activation function, and the characteristic of the first hidden layer is marked as L 1 。
Exemplary, first hidden layer feature L 1 The following equation 4 is satisfied:
L 1 ＝T(f 100 (F) Equation 4)
Wherein F is 27 attributes of the personalized portrayal of the user, F 100 () T () is a Tanh activation function, which is a hidden layer containing 100 neurons.
Step 602, the user complaint prediction device inputs the first hidden layer feature to a second hidden layer of the first model to perform dimension reduction and nonlinear conversion, so as to obtain the second hidden layer feature.
Wherein the second hidden layer includes m neurons, m being a positive integer less than N.
An example, m has a value of 25, that is, the second hidden layer is a hidden layer containing 25 neurons.
In this case, step 602 may be specifically implemented as: the user complaint prediction device inputs the first hidden layer feature generated in the step 601 into the second hidden layer, and performs feature dimension reduction on the first hidden layer feature through 25 neurons of the second hidden layer. After the dimension reduction, nonlinear conversion is carried out according to the Tanh activation function, and the characteristic of the second hidden layer is marked as L 2 。
Exemplary, second hidden layer feature L 2 The following equation 5 is satisfied:
L 2 ＝T(f 25 (L 1 ) Equation 5)
Wherein F is 25 () Is a hidden layer containing 25 neurons.
And step 603, the user complaint prediction device inputs the second hidden layer characteristics to a third hidden layer of the first model to perform ascending and nonlinear transformation, so as to obtain the third hidden layer characteristics.
Wherein the third hidden layer includes n neurons.
That is, the third hidden layer is a hidden layer containing 100 neurons.
In this case, step 603 may be specifically implemented as: the user complaint predicting device inputs the second hidden layer features generated in step 602 into 100 neurons of the third hidden layer to perform feature up-dimension on the user portrait data. After the dimension is increased, nonlinear transformation is carried out according to the Tanh activation function, and a third hidden layer characteristic L is obtained 3 。
Exemplary, third hidden layer feature L 3 The following equation 6 is satisfied:
L 3 ＝T(f 100 (L 2 ) Equation 6)
Step 604, the user complaint prediction device inputs the third hidden layer feature to the fourth hidden layer of the first model to perform the dimension reduction and nonlinear transformation, so as to obtain the fourth hidden layer feature.
Wherein the fourth hidden layer comprises m neurons;
that is, the fourth hidden layer is a hidden layer containing 25 neurons.
In this case, step 604 may be specifically implemented as: the user complaint prediction device inputs the third hidden layer feature generated in the step 603 into the fourth hidden layer, and performs feature dimension reduction on the second hidden layer feature through 25 neurons of the fourth hidden layer. After the dimension reduction, nonlinear conversion is carried out according to the Tanh activation function, and the characteristic of the fourth hidden layer is marked as L 4 。
Exemplary, fourth hidden layer feature L 4 The following equation 7 is satisfied:
L 4 ＝T(F 25 (L 3 ) Equation 7)
Step 605, the user complaint prediction device inputs the characteristics of the fourth hidden layer to the fifth hidden layer of the first model to perform dimension reduction and nonlinear transformation, so as to obtain first characteristic data.
Wherein the fifth hidden layer includes n neurons.
That is, the fifth hidden layer is a hidden layer including 100 neurons.
In this case, step 605 may be implemented specifically as: the user complaint predicting device inputs the fourth hidden layer feature generated in step 604 into 100 neurons of the fifth hidden layer to perform feature up-dimension on the user portrait data. After the dimension is increased, nonlinear transformation is carried out according to the Tanh activation function, and the obtained first characteristic data is recorded as L 5 。
Exemplary, first characteristic data L 5 The following equation 8 is satisfied:
L 5 ＝T(F 100 (L 4 ) Equation 8)
In a possible implementation manner, since the fifth hidden layer includes 100 neurons, the first characteristic data L output by the fifth hidden layer 5 Including 100 eigenvalues. The user complaint prediction device predicts the first characteristic data L 5 Is converted into a 10 x 10 feature matrix G 1 For subsequent processing. User complaint prediction apparatus determines G 1 Is the first characteristic data.
The process of extracting the hidden layer features of the user portrait data of the user to be predicted by the user complaint prediction device through the first model is described in detail above.
The process of extracting the convolution layer characteristics of the output result of the first model by the user complaint prediction device through the second model will be described in detail below.
Referring to fig. 3, as shown in fig. 6, in step 303, the user complaint prediction device inputs the first feature data into the second model to perform convolutional layer feature extraction, and the process of determining the second feature data may be specifically implemented by the following steps 606-612. The following is a detailed description:
step 606, the user complaint prediction device inputs the first feature data into the first convolution layer of the second model to perform convolution layer feature extraction and nonlinear transformation, and obtains the first convolution layer feature.
Wherein the first convolution layer comprises a x a convolution kernel; a is a positive integer.
An example, a has a value of 1, that is, the first convolution layer includes a 1 x 1 convolution kernel.
In this case, step 606 may be implemented specifically as: the user complaint prediction device predicts the first characteristic data G 1 And (3) inputting the features into a trained convolutional neural network model, and calling a 1 multiplied by 1 convolutional kernel of a first convolutional layer of the convolutional neural network model to perform feature extraction. After extracting the features, performing nonlinear conversion according to the function of activating Mish to obtain a first convolution layer feature marked as G 2 。
Exemplary, first convolution layer characteristics G 2 The following equation 9 is satisfied:
where M () is a Mish () activation function,is a convolution operation with a convolution kernel size of 1 x 1.
And step 607, the user complaint prediction device inputs the first characteristic data into a second convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion, so as to obtain the second convolution layer characteristic.
Wherein the second convolution layer comprises a bXb convolution kernel; b is a positive integer greater than a.
An example, b has a value of 3, that is, the second convolution layer includes a 3 x 3 convolution kernel.
In this case, step 607 may be specifically implemented as: the user complaint prediction device predicts the first characteristic data G 1 And inputting the features into the trained convolutional neural network model, and calling a 3 multiplied by 3 convolutional kernel in a second convolutional layer of the convolutional neural network model to perform feature extraction. After extracting the features, performing nonlinear conversion according to the function of activating Mish to obtain a second convolution layer feature marked as G 3 。
Exemplary, second convolution layer characteristics G 3 The following equation 10 is satisfied:
wherein, the liquid crystal display device comprises a liquid crystal display device,is a convolution operation with a convolution kernel size of 3 x 3.
And 608, the user complaint prediction device inputs the first characteristic data into a third convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion, so as to obtain the third convolution layer characteristic.
Wherein the third convolution layer comprises a c×c convolution kernel; c is a positive integer greater than b;
for example, the value of c is 5, that is, the second convolution layer includes a convolution kernel of 5×5.
In this case, step 608 may be implemented specifically as: the user complaint prediction device predicts the first characteristic data G 1 And (3) inputting the features into the trained convolutional neural network model, and calling a 5 multiplied by 5 convolutional kernel in a third convolutional layer of the convolutional neural network model to perform feature extraction. After extracting the features, performing nonlinear conversion according to the function of activating Mish to obtain a third convolution layer feature marked as G 4 。
Exemplary, third convolution layer feature G 4 The following equation 11 is satisfied:
wherein, the liquid crystal display device comprises a liquid crystal display device,is a convolution operation with a convolution kernel size of 5 x 5.
And step 609, the user complaint prediction device performs channel splicing on the first convolution layer characteristics, the second convolution layer characteristics and the third convolution layer characteristics to determine a target characteristic matrix.
In a possible implementation manner, the user complaint predicting device predicts the first convolution layer characteristic G obtained in the step 606 2 Step (a) of607, a second convolution layer feature G 3 Third convolution layer feature G obtained in step 608 4 Performing feature stitching to obtain a target feature matrix denoted as G 5 。
Exemplary, target feature matrix G 5 Satisfy the following formula 12
G 5 ＝Concat(G 2 ，G 3 ，G 4 ) Equation 12
Wherein Concat () is a channel splicing operation.
And 610, the user complaint prediction device inputs the target feature matrix into a fourth convolution layer of the second model to extract the feature of the convolution layer and perform nonlinear transformation to obtain the feature of the fourth convolution layer.
Wherein the fourth convolution layer includes a bxb convolution kernel.
That is, the fourth convolution layer includes a 3×3 convolution kernel.
In this case, step 610 may be implemented specifically as: the user complaint prediction device predicts the target feature matrix G 5 And inputting the features into a fourth convolution layer, and calling a 3 multiplied by 3 convolution kernel of the fourth convolution layer to perform feature extraction. After extracting the features, performing nonlinear conversion according to the function of activating Mish to obtain a fourth convolution layer feature marked as G 6 。
Exemplary, fourth convolution layer feature G 6 The following equation 13 is satisfied:
and 611, the user complaint prediction device inputs the target feature matrix into a fifth convolution layer of the second model to extract the feature of the convolution layer and perform nonlinear transformation to obtain the feature of the fifth convolution layer.
Wherein the fifth convolution layer includes a a×a convolution kernel.
That is, the fifth convolution layer includes a 1×1 convolution kernel.
In this case, step 611 may be specifically implemented as: user complaint prediction deviceFourth convolution layer feature G 6 And inputting the features into the third convolution layer, and calling a 1 multiplied by 1 convolution kernel of the fourth convolution layer to perform feature extraction. After extracting the features, performing nonlinear conversion according to the function of using Mish to obtain a fifth convolution layer feature marked as G 7 。
Exemplary, fifth convolution layer feature G 7 The following equation 14 is satisfied:
and step 612, the user complaint prediction device inputs the target feature matrix into a sixth convolution layer of the second model to perform convolution layer feature extraction and nonlinear transformation, so as to obtain second feature data.
Wherein the sixth convolution layer includes a bxb convolution kernel.
That is, the sixth convolution layer includes a 3×3 convolution kernel.
In this case, step 612 may be implemented specifically as: the user complaint prediction device predicts the fifth convolution layer characteristic G 7 And inputting the features into a sixth convolution layer, and calling a 3 multiplied by 3 convolution kernel of the sixth convolution layer to perform feature extraction. After extracting the features, performing nonlinear conversion according to the function of using Mish to obtain a sixth convolution layer feature marked as G 8 。
Exemplary, sixth convolution layer feature G 8 The following equation 15 is satisfied:
in a possible implementation, due to the first characteristic data G 1 For a 10×10 feature matrix, the sixth convolution layer feature G is obtained after the convolution processing 8 Also a 10 x 10 feature matrix. The user complaint prediction device predicts the sixth convolution layer characteristic G 8 Feature data L converted into 100 features 6 For subsequent processing. User complaint prediction device determines L 6 Is the second characteristicData.
The process of convolutional layer feature extraction of the user image data of the user to be predicted by the user complaint prediction device through the second model is described in detail above.
The process of extracting hidden layer features and predicting the probability of user complaints from the output result of the third model by the user complaint predicting device through the third model will be described in detail below.
Referring to fig. 3, as shown in fig. 6, in step 304, the user complaint prediction device inputs the second feature data into the third model to perform complaint prediction, and the process of determining the complaint probability of the user to be predicted may be specifically implemented by the following steps 613-618.
Step 613, the user complaint prediction device inputs the second feature data into the sixth hidden layer of the third model to perform the dimension reduction and nonlinear transformation, and determines the features of the sixth hidden layer.
Wherein the sixth hidden layer includes m neurons.
That is, the sixth hidden layer is a hidden layer containing 25 neurons.
In this case, step 6513 may be specifically implemented as: the user complaint prediction device predicts the second characteristic data L 6 And inputting the feature up-dimension data into a second full-connection neural network model, and calling 25 neurons of a sixth hidden layer of the second full-connection neural network model to perform feature up-dimension on the user portrait data. After the dimension is increased, nonlinear transformation is carried out according to the Tanh activation function, and a sixth hidden layer characteristic L is obtained 7 。
Illustratively, a sixth hidden layer feature L 7 The following equation 16 is satisfied:
L 7 ＝T(F 25 (L 6 ) Equation 16)
Step 614, the user complaint prediction device inputs the sixth hidden layer feature into the seventh hidden layer of the third model to perform up-scaling and nonlinear transformation, and determines the seventh hidden layer feature.
Wherein the seventh hidden layer includes n neurons.
That is, the seventh hidden layer is a hidden layer including 100 neurons.
In this case, step 614 may be implemented specifically as: the user complaint predicting device inputs the sixth hidden layer feature generated in step 613 to 100 neurons of the seventh hidden layer to perform feature up-scaling on the user portrait data. After the dimension is increased, nonlinear transformation is carried out according to the Tanh activation function, and the obtained first characteristic data is recorded as L 8 。
Exemplary, first characteristic data L 8 The following equation 17 is satisfied:
L 8 ＝T(F 100 (L 8 ) Equation 17)
Step 615, the user complaint prediction device inputs the seventh hidden layer feature into the eighth hidden layer of the third model to perform the dimension reduction and nonlinear transformation, and determines the eighth hidden layer feature.
Wherein the eighth hidden layer includes m neurons.
That is, the eighth hidden layer is a hidden layer including 25 neurons.
In this case, step 615 may be implemented specifically as: the user complaint prediction device inputs the seventh hidden layer feature generated in step 614 into the eighth hidden layer, and performs feature dimension reduction on the seventh hidden layer feature through 25 neurons of the eighth hidden layer. After the dimension reduction, nonlinear conversion is carried out according to the Tanh activation function, and the eighth hidden layer characteristic is marked as L 9 。
Illustratively, an eighth hidden layer feature L 9 The following equation 18 is satisfied:
L 9 ＝T(f 25 (L 7 ) Equation 18)
Step 616, the user complaint prediction device inputs the eighth hidden layer feature into the ninth hidden layer of the third model to perform the dimension reduction and nonlinear transformation, and determines the ninth hidden layer feature.
Wherein the ninth hidden layer includes r neurons; r is a positive integer less than m.
An example, r has a value of 10, that is, the ninth hidden layer is a hidden layer containing 10 neurons.
In this case, step 616 may be implemented specifically as: the user complaint prediction device inputs the eighth hidden layer feature generated in step 615 into the ninth hidden layer, and performs feature dimension reduction on the eighth hidden layer feature through 10 neurons of the ninth hidden layer. After the dimension reduction, nonlinear conversion is carried out according to the Tanh activation function, and a ninth hidden layer characteristic is obtained and is marked as L 10 。
Illustratively, a ninth hidden layer feature L 10 The following formula 19 is satisfied:
L 10 ＝T(f 10 (L 9 ) Equation 19)
Wherein F is 10 () Is a hidden layer containing 10 neurons.
Step 617, the user complaint prediction device inputs the ninth hidden layer feature into a tenth hidden layer of the third model to perform dimension reduction and nonlinear transformation, and determines the tenth hidden layer feature.
Wherein the tenth hidden layer comprises s neurons; s is a positive integer less than r.
For example, s has a value of 5, that is, the tenth hidden layer is a hidden layer containing 5 neurons.
In this case, step 617 may be implemented specifically as: the user complaint prediction device inputs the ninth hidden layer feature generated in step 616 into the tenth hidden layer, and performs feature dimension reduction on the ninth hidden layer feature through 5 neurons of the tenth hidden layer. After the dimension reduction, nonlinear conversion is carried out according to the Tanh activation function, and the tenth hidden layer characteristic is obtained and marked as L 11 。
Illustratively, a tenth hidden layer feature L 11 The following equation 20 is satisfied:
L 11 ＝T(F 5 (L 10 ) Equation 20)
Wherein F is 5 () Is a hidden layer containing 5 neurons.
Step 618, the user complaint prediction device inputs the tenth hidden layer feature into the output layer of the third model to perform probability prediction, so as to determine the user complaint probability.
By way of example only, and not by way of limitation,the output layer of the third model comprises 1 neuron, and the output layer activates the function through Sigmoid to the tenth hidden layer characteristic L 11 And (3) calculating, wherein a calculated result is used as the user complaint probability.
In a possible implementation manner, in the embodiment of the application, the user with the user complaint probability exceeding 0.7 is taken as the complaint user with great attention. The method is characterized in that the complaint reasons are determined according to the requirements of the users, the complaint reasons are processed, and the complaint hidden danger of the users is eliminated in advance.
In the above, the process of determining the user complaint probability is described by processing the user portrait data of the user to be predicted through the first model, the second model and the third model. Through the technical scheme, the embodiment of the application extracts the characteristics of the hidden layer and the convolution layer through a plurality of models, and finally predicts the complaint probability of the user, so that the accuracy of the prediction result can be greatly improved.
According to the embodiment of the application, the function modules or the function units of the user complaint prediction device can be divided according to the method example, for example, each function module or each function unit can be divided corresponding to each function, and two or more functions can be integrated in one processing module. The integrated modules may be implemented in hardware, or in software functional modules or functional units. The division of the modules or units in the embodiment of the present application is schematic, which is merely a logic function division, and other division manners may be implemented in practice.
As shown in fig. 7, a schematic structural diagram of a user complaint predicting device 70 according to an embodiment of the present application includes: a processing unit 701.
A processing unit 701 for determining user complaint data; the processing unit 701 is further configured to input user portrait data of a user to be predicted into the first model to perform hidden layer feature extraction, and determine first feature data; the first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data; the processing unit 701 is further configured to input the first feature data into the second model to perform convolutional layer feature extraction, and determine second feature data; the second model is used for extracting the characteristics of the convolution layer at least once for the data; the processing unit 701 is further configured to input the second feature data into a third model to perform complaint prediction, determine a complaint probability of the user, and the third model is configured to perform at least one hidden layer feature dimension reduction on the data to determine the third feature data, and predict the complaint probability of the user based on the third feature data.
In one possible implementation, the processing unit 701 is specifically configured to: inputting user portrait data of a user to be predicted into a first hidden layer of a first model to perform ascending and nonlinear conversion to obtain characteristics of the first hidden layer; the first hidden layer comprises n neurons, wherein n is a positive integer; inputting the first hidden layer characteristics into a second hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain the second hidden layer characteristics; the second hidden layer comprises m neurons, m is a positive integer less than N; inputting the second hidden layer characteristics into a third hidden layer of the first model to perform ascending and nonlinear conversion to obtain third hidden layer characteristics; the third hidden layer includes n neurons; inputting the third hidden layer characteristics into a fourth hidden layer of the first model for dimension reduction and nonlinear conversion to obtain fourth hidden layer characteristics; the fourth hidden layer includes m neurons; inputting the characteristics of the fourth hidden layer into a fifth hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain first characteristic data; the fifth hidden layer includes n neurons.
In one possible implementation, the processing unit 701 is specifically configured to: inputting the first characteristic data into a first convolution layer of a second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a first convolution layer characteristic; the first convolution layer comprises a convolution kernel of a×a; a is a positive integer; inputting the first characteristic data into a second convolution layer of a second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain second convolution layer characteristics; the second convolution layer includes a bXb convolution kernel; b is a positive integer greater than a; inputting the first characteristic data into a third convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a third convolution layer characteristic; the third convolution layer includes a c×c convolution kernel; c is a positive integer greater than b; channel splicing is carried out on the first convolution layer characteristics, the second convolution layer characteristics and the third convolution layer characteristics, and a target characteristic matrix is determined; inputting the target feature matrix into a fourth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fourth convolution layer features; a convolution kernel of b x b is included in the fourth convolution layer; inputting the target feature matrix into a fifth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fifth convolution layer features; the fifth convolution layer includes a convolution kernel of a×a; inputting the target feature matrix into a sixth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain second feature data; the sixth convolution layer includes a bxb convolution kernel.
In one possible implementation, the processing unit 701 is specifically configured to: inputting the second characteristic data into a sixth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the characteristics of the sixth hidden layer; the sixth hidden layer includes m neurons; inputting the sixth hidden layer characteristics into a seventh hidden layer of the third model to perform ascending and nonlinear transformation, and determining the seventh hidden layer characteristics; the seventh hidden layer includes n neurons; inputting the seventh hidden layer characteristics into an eighth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the eighth hidden layer characteristics; the eighth hidden layer includes m neurons; inputting the eighth hidden layer characteristics into a ninth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the ninth hidden layer characteristics; the ninth hidden layer includes r neurons; r is a positive integer less than m; inputting the ninth hidden layer characteristics into a tenth hidden layer of the third model for dimension reduction and nonlinear conversion, and determining the tenth hidden layer characteristics; the tenth hidden layer includes s neurons; s is a positive integer less than r; and inputting the tenth hidden layer characteristics into an output layer of the third model to carry out probability prediction, and determining the complaint probability of the user.
In a possible implementation manner, the processing unit 701 is further configured to: determining user image data of complaint users and user image data of non-complaint users; based on the complaint type of the complaint user, determining a complaint weight value of the complaint user; based on the complaint weight value of the complaint user, carrying out weighted calculation on the user portrait data of the complaint user to determine the weighted user portrait data of the complaint user; training a first fully connected neural network model, a convolutional neural network model and a second fully connected neural network model based on weighted user portrayal data of a complaint user and non-complaint user data, and determining a first model, a second model and a third model; the method comprises the steps of training a first model through a first full-connection neural network model, training a second model through a convolution neural network model, and training a third model through a second full-connection neural network model.
In one possible implementation, the apparatus further includes: a communication unit 702; a communication unit 702, configured to obtain user image data of a history complaint user; a processing unit 701, configured to determine whether the number of user image data of the historic complaint user reaches the first number; the processing unit 701 is further configured to determine, if the preset number of virtual complaint users does not reach the preset number of virtual complaint users, user image data of a second number of virtual complaint users according to the user image data of the historical complaint users; the second number is determined based on a difference between the number of user representation data of the historic complaint user and the first number; the processing unit 701 is further configured to determine that the user profile data of the complaint user includes user profile data of a history complaint user and user profile data of a second number of virtual complaint users.
In one possible implementation, the processing unit 701 is specifically configured to: performing the following target operations a plurality of times to determine user image data for a second number of virtual complaint users; the target operations include: acquiring user image data of a first historical complaint user and user image data of a second historical complaint user; the user image data of the first historical complaint user and the user image data of the second historical complaint user are data in the user image data of the historical complaint user; and adjusting the data of the target dimension in the user image data of the first historical complaint user to the data of the target dimension in the user image data of the second historical complaint user, and determining the user image data of the first virtual complaint user.
When implemented in hardware, the communication unit 702 in the embodiments of the present application may be integrated on a communication interface, and the processing unit 701 may be integrated on a processor.
From the foregoing description of the embodiments, it will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of functional modules is illustrated, and in practical application, the above-described functional allocation may be implemented by different functional modules according to needs, i.e. the internal structure of the apparatus is divided into different functional modules to implement all or part of the functions described above. The specific working processes of the above-described systems, devices and units may refer to the corresponding processes in the foregoing method embodiments, which are not described herein.
Embodiments of the present application provide a computer program product comprising instructions which, when run on a computer, cause the computer to perform the user complaint prediction method of the method embodiments described above.
The embodiment of the application also provides a computer readable storage medium, wherein the computer readable storage medium stores instructions which, when run on a computer, cause the computer to execute the user complaint prediction method in the method flow shown in the method embodiment.
The computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the computer-readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access Memory (Random Access Memory, RAM), a Read-Only Memory (ROM), an erasable programmable Read-Only Memory (Erasable Programmable Read Only Memory, EPROM), a register, a hard disk, an optical fiber, a portable compact disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing, or any other form of computer readable storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium may be integral to the processor. The processor and the storage medium may reside in an application specific integrated circuit (Application Specific Integrated Circuit, ASIC). In embodiments of the present application, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
Embodiments of the present application provide a computer program product comprising instructions which, when run on a computer, cause the computer to perform a method of predicting customer complaints as described in figures 3 to 6.
Since the user complaint prediction apparatus, the computer-readable storage medium and the computer program product according to the embodiments of the present application can be applied to the above-mentioned method, the technical effects obtained by the method can also refer to the above-mentioned method embodiments, and the embodiments of the present application are not described herein again.
In the several embodiments provided by the present application, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some interface, indirect coupling or communication connection of devices or units, electrical, mechanical, or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The present application is not limited to the above embodiments, and any changes or substitutions within the technical scope of the present application should be covered by the scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.
Claims (10)
1. A method of predicting complaints of a user, the method comprising:
determining user portrait data of a user to be predicted; wherein the user profile data is used to identify data related to user complaints from multiple dimensions;
Inputting the user portrait data of the user to be predicted into a first model to extract hidden layer characteristics, and determining first characteristic data; the first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data;
inputting the first characteristic data into a second model for convolutional layer characteristic extraction, and determining second characteristic data; the second model is used for extracting the characteristics of the convolution layer of the data at least once;
and inputting the second characteristic data into a third model for complaint prediction, determining the complaint probability of the user to be predicted, wherein the third model is used for carrying out at least one hidden layer characteristic dimension reduction on the data to determine third characteristic data, and predicting the complaint probability of the user based on the third characteristic data.
2. The method according to claim 1, wherein the inputting the user portrait data of the user to be predicted into the first model for hidden layer feature extraction, and determining the first feature data includes:
inputting the user portrait data of the user to be predicted into a first hidden layer of the first model for ascending and nonlinear conversion to obtain a first hidden layer characteristic; the first hidden layer comprises n neurons, wherein n is a positive integer;
Inputting the first hidden layer characteristics into a second hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain second hidden layer characteristics; the second hidden layer comprises m neurons, and m is a positive integer smaller than N;
inputting the second hidden layer characteristics into a third hidden layer of the first model to perform ascending and nonlinear transformation to obtain third hidden layer characteristics; the third hidden layer includes n neurons;
inputting the third hidden layer characteristics into a fourth hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain fourth hidden layer characteristics; the fourth hidden layer includes m neurons;
inputting the characteristics of the fourth hidden layer into a fifth hidden layer of the first model to perform dimension reduction and nonlinear conversion to obtain the first characteristic data; the fifth hidden layer includes n neurons.
3. The method of claim 2, wherein the inputting the first feature data into a second model for convolutional layer feature extraction, determining second feature data, comprises:
inputting the first characteristic data into a first convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a first convolution layer characteristic; the first convolution layer comprises a x a convolution kernel; a is a positive integer;
Inputting the first characteristic data into a second convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a second convolution layer characteristic; the second convolution layer comprises a bXb convolution kernel; b is a positive integer greater than a;
inputting the first characteristic data into a third convolution layer of the second model to perform convolution layer characteristic extraction and nonlinear conversion to obtain a third convolution layer characteristic; the third convolution layer comprises a c×c convolution kernel; c is a positive integer greater than b;
performing channel splicing on the first convolution layer characteristics, the second convolution layer characteristics and the third convolution layer characteristics to determine a target characteristic matrix;
inputting the target feature matrix into a fourth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fourth convolution layer features; the fourth convolution layer comprises a bXb convolution kernel;
inputting the target feature matrix into a fifth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain fifth convolution layer features; the fifth convolution layer comprises a x a convolution kernel;
inputting the target feature matrix into a sixth convolution layer of the second model to perform convolution layer feature extraction and nonlinear conversion to obtain second feature data; the sixth convolution layer includes a bxb convolution kernel.
4. A method according to claim 3, wherein said inputting the second feature data into a third model for complaint prediction, determining a probability of a user complaint, comprises:
inputting the second characteristic data into a sixth hidden layer of the third model to perform dimension reduction and nonlinear conversion, and determining the characteristics of the sixth hidden layer; the sixth hidden layer includes m neurons;
inputting the sixth hidden layer characteristics into a seventh hidden layer of the third model to perform ascending and nonlinear transformation, and determining the seventh hidden layer characteristics; the seventh hidden layer includes n neurons;
inputting the seventh hidden layer characteristics into an eighth hidden layer of the third model to perform dimension reduction and nonlinear conversion, and determining eighth hidden layer characteristics; the eighth hidden layer includes m neurons;
inputting the eighth hidden layer characteristics into a ninth hidden layer of the third model to perform dimension reduction and nonlinear conversion, and determining the ninth hidden layer characteristics; the ninth hidden layer includes r neurons; r is a positive integer less than m;
inputting the ninth hidden layer characteristics into a tenth hidden layer of the third model for performing dimension reduction and nonlinear conversion, and determining tenth hidden layer characteristics; the tenth hidden layer includes s neurons; s is a positive integer less than r;
And inputting the tenth hidden layer characteristics into an output layer of the third model to perform probability prediction, and determining the user complaint probability.
5. The method according to any one of claims 1-4, further comprising, before said inputting the user representation data of the user to be predicted into the first model for hidden layer feature extraction, determining first feature data:
determining user image data of complaint users and user image data of non-complaint users;
based on the complaint type of the complaint user, determining a complaint weight value of the complaint user;
based on the complaint weight value of the complaint user, carrying out weighted calculation on the user portrait data of the complaint user to determine weighted user portrait data of the complaint user;
training a first fully connected neural network model, a convolutional neural network model and a second fully connected neural network model based on the weighted user representation data of the complaint user and the non-complaint user data, and determining the first model, the second model and the third model; the first model is obtained through training of the first full-connection neural network model, the second model is obtained through training of the convolution neural network model, and the third model is obtained through training of the second full-connection neural network model.
6. The method of claim 5, wherein the determining user image data of the complaint user comprises:
acquiring user image data of a history complaint user;
determining whether the number of user image data of the historic complaint user reaches a first number;
if the preset number is not reached, determining user image data of a second number of virtual complaint users according to the user image data of the historical complaint users; the second number is determined based on a difference between the number of user image data of the historic complaint user and the first number;
determining that the user profile data for the complaint user includes user profile data for the historical complaint user and user profile data for the second number of virtual complaint users.
7. The method of claim 6, wherein determining the second number of user image data for the virtual complaint user based on the user image data for the historical complaint user comprises:
performing the following target operations a plurality of times to determine user image data of the second number of virtual complaint users; the target operation includes:
acquiring user image data of a first historical complaint user and user image data of a second historical complaint user; the user portrait data of the first historical complaint user and the user portrait data of the second historical complaint user are data in the user portrait data of the historical complaint user;
And adjusting the data of the target dimension in the user image data of the first historical complaint user to the data of the target dimension in the user image data of the second historical complaint user, and determining the user image data of the first virtual complaint user.
8. A user complaint predicting apparatus, the apparatus comprising: a processing unit;
the processing unit is used for determining user complaint data;
the processing unit is further used for inputting the user portrait data of the user to be predicted into a first model to perform hidden layer feature extraction and determine first feature data; the first model is used for carrying out at least one hidden layer characteristic lifting and hidden layer characteristic dimension reduction on the data;
the processing unit is further used for inputting the first characteristic data into a second model to perform convolutional layer characteristic extraction and determine second characteristic data; the second model is used for extracting the characteristics of the convolution layer of the data at least once;
the processing unit is further configured to input the second feature data into a third model for complaint prediction, determine a complaint probability of the user, and the third model is configured to perform at least one hidden layer feature dimension reduction on the data to determine third feature data, and predict the complaint probability of the user based on the third feature data.
9. A user complaint predicting apparatus, comprising: a processor and a communication interface; the communication interface being coupled to the processor for executing a computer program or instructions to implement a method of predicting a user complaint as claimed in any one of claims 1 to 7.
10. A computer readable storage medium having instructions stored therein, wherein when executed by a computer, the computer performs the method of predicting complaints by a user as claimed in any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310675071.0A CN116702982A (en) | 2023-06-06 | 2023-06-06 | User complaint prediction method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310675071.0A CN116702982A (en) | 2023-06-06 | 2023-06-06 | User complaint prediction method, device and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116702982A true CN116702982A (en) | 2023-09-05 |
Family
ID=87835247
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310675071.0A Pending CN116702982A (en) | 2023-06-06 | 2023-06-06 | User complaint prediction method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116702982A (en) |
-
2023
- 2023-06-06 CN CN202310675071.0A patent/CN116702982A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108876600A (en) | Warning information method for pushing, device, computer equipment and medium | |
| CN109165840A (en) | Risk profile processing method, device, computer equipment and medium | |
| CN109670940A (en) | Credit Risk Assessment Model generation method and relevant device based on machine learning | |
| CN106126597A (en) | User property Forecasting Methodology and device | |
| CN110647696B (en) | Business object sorting method and device | |
| US8788438B2 (en) | Method performed in a computer system for aiding the assessment of an influence of a user in or interacting with a communication system by applying social network analysis, SNA, functions, a computer system, computer program and computer program product | |
| CN111898247B (en) | Landslide displacement prediction method, landslide displacement prediction equipment and storage medium | |
| CN114638633A (en) | Abnormal flow detection method and device, electronic equipment and storage medium | |
| CN110717509A (en) | Data sample analysis method and device based on tree splitting algorithm | |
| KR20110034174A (en) | Method and server apparatus for providing advertizing service | |
| CN109325781A (en) | Client's Quality Analysis Methods, device, computer equipment and storage medium | |
| CN108510003A (en) | Car networking big data air control assemblage characteristic extracting method, device and storage medium | |
| CN116702982A (en) | User complaint prediction method, device and storage medium | |
| CN112231299A (en) | Method and device for dynamically adjusting feature library | |
| Paul et al. | Decoding the Divide: Analyzing Disparities in Broadband Plans Offered by Major US ISPs | |
| CN116797235A (en) | Method and device for processing consumption information, storage medium and computer equipment | |
| US20210366048A1 (en) | Methods and systems for reacting to loss reporting data | |
| CN101620701A (en) | Application of KPI analysis in income guarantee system of telecommunication industry based on stratification method | |
| CN115660101A (en) | Data service providing method and device based on service node information | |
| CN113806682A (en) | Information processing method, information processing device, electronic equipment and storage medium | |
| CN107392408A (en) | The prompt message output intent and device of a kind of credit score | |
| CN112750023A (en) | Consumption financial user income estimation method based on factor analysis | |
| CN111738818A (en) | Method, equipment and storage medium for re-detection of credit after credit | |
| Wang et al. | Image analysis considering textual correlations enables accurate user switching tendency prediction | |
| CN113723522B (en) | Abnormal user identification method and device, electronic equipment and storage 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 |