CN112488743A

CN112488743A - Satisfaction degree prediction method, network equipment and storage medium

Info

Publication number: CN112488743A
Application number: CN201910866736.XA
Authority: CN
Inventors: 钟勇才; 李劲菊
Original assignee: ZTE Corp
Current assignee: ZTE Corp
Priority date: 2019-09-12
Filing date: 2019-09-12
Publication date: 2021-03-12

Abstract

The embodiment of the invention provides a satisfaction prediction method, network equipment and a storage medium, when a mobile operator knows the user satisfaction, the mobile operator does not need to wait for the complaint result of a research or a user, the timeliness of knowing the user satisfaction by the mobile operator is greatly improved, and the comprehensiveness of knowing the user satisfaction by the operator is increased. Furthermore, in the embodiment of the invention, the characteristic that the tolerance of the user to the service is different under different scenes is considered, so that the corresponding user tolerance weight is configured for each scene, and the satisfaction prediction model can be created based on the user tolerance weight corresponding to each scene and the service KPI and investigation satisfaction data of a plurality of investigation users under each scene, so that the user satisfaction can be prevented from being predicted and evaluated according to the unified rigidity standard under all scenes, and the satisfaction prediction accuracy is improved.

Description

Satisfaction degree prediction method, network equipment and storage medium

Technical Field

The present invention relates to the field of networks, and in particular, to a satisfaction prediction method, a network device, and a storage medium.

Background

Each mobile operator is very concerned about the satisfaction degree of the user to the network service in order to improve the experience of the user to the network service and enhance the user viscosity. However, the conventional ways of knowing the satisfaction of the user can only be based on two ways of market research and complaint of the user, but the ways are low in timeliness, so that the mobile operator can only learn the feedback of the user after the user has tolerated the limit and made a negative evaluation; in addition, these methods involve a problem that the range of users is small and the representatives of other users who have not been investigated or complaints are weak.

Therefore, it is needed to provide a scheme that enables a mobile operator to know the satisfaction of each user for a service in time, so that the mobile operator can adjust its own service policy in time to avoid user loss.

Disclosure of Invention

The satisfaction prediction method, the network equipment and the storage medium provided by the embodiment of the invention mainly solve the technical problems that: how to let the mobile operator know the satisfaction degree of the user to the service in time.

To solve the above technical problem, an embodiment of the present invention provides a satisfaction prediction method, including:

acquiring service record data of a user to be predicted;

determining service KPIs (key performance indicators) of the user to be predicted under each scene according to the service record data, wherein each scene has a corresponding user tolerance weight;

inputting the service KPI of the user to be predicted under each scene into a pre-established satisfaction prediction model to obtain the corresponding prediction satisfaction of the user to be predicted; and the satisfaction prediction model is created based on the user tolerance weight corresponding to each scene, the service KPI of a plurality of research users in each scene and the research satisfaction data of each research user, and the service KPI of the research users in each scene is determined according to the service record data of the research users.

The embodiment of the invention also provides network equipment, which comprises a processor, a memory and a communication bus;

the communication bus is used for realizing connection communication between the processor and the memory;

the processor is configured to execute one or more programs stored in the memory to implement the steps of the satisfaction prediction method described above.

Embodiments of the present invention also provide a storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the steps of the satisfaction prediction method described above.

The invention has the beneficial effects that:

according to the satisfaction prediction method, the network equipment and the storage medium provided by the embodiment of the invention, a satisfaction prediction model is created in advance based on the business KPI of a research user and the research satisfaction of the research user, then when the user satisfaction of a certain user to be predicted needs to be known, the business record data of the user to be predicted can be directly obtained, the business KPI of the user to be predicted under each scene is determined according to the business record data, then the business KPI of the user to be predicted under each scene is input into the satisfaction prediction model, and the output of the satisfaction prediction model is the corresponding prediction satisfaction of the user to be predicted. By the method, when the mobile operator knows the user satisfaction, the mobile operator does not need to wait for the investigation or the complaint result of the user, the user satisfaction can be predicted in the process of providing the service for the user, and the timeliness of knowing the user satisfaction by the mobile operator is greatly improved; meanwhile, because the satisfaction degree prediction model can predict the satisfaction degree of any user to the business service, the opinion users known by the mobile operators are not limited to the opinions of the investigation users or the complaint users, and the comprehensiveness of user satisfaction degree understanding can be improved.

Furthermore, in the embodiment of the invention, the characteristic that the tolerance of the user to the service is different under different scenes is considered, so that the corresponding user tolerance weight is configured for each scene, and the satisfaction prediction model can be created based on the user tolerance weight corresponding to each scene and the service KPI and investigation satisfaction data of a plurality of investigation users under each scene, so that the user satisfaction can be prevented from being predicted and evaluated according to the unified rigidity standard under all scenes, and the satisfaction prediction accuracy is improved.

Additional features and corresponding advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Drawings

FIG. 1 is a flow chart of a satisfaction prediction method provided in one embodiment of the present invention;

fig. 2 is a flowchart for determining a service KPI of a user to be predicted in each scene according to service record data of the user to be predicted, according to a first embodiment of the present invention;

fig. 3 is a flowchart of a satisfaction prediction model obtained based on the business record data of the research user and the research satisfaction training provided in the first embodiment of the present invention;

fig. 4 is a flowchart for determining a service KPI of a user to be predicted in each scene according to service record data of the user to be predicted, according to a first embodiment of the present invention;

FIG. 5 is a flowchart of a satisfaction prediction model obtained by training with a random forest classification algorithm according to a first embodiment of the present invention;

FIG. 6 is a schematic diagram illustrating a principle of predicting user satisfaction based on research data of a research user according to a second embodiment of the present invention;

FIG. 7 is a schematic diagram illustrating a detailed principle of predicting user satisfaction based on research data of a research user according to a second embodiment of the present invention;

FIG. 8 is a decision tree diagram of the random forest classification algorithm according to the second embodiment of the present invention;

fig. 9 is a schematic diagram of a hardware structure of a network device according to a third embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The first embodiment is as follows:

with the rapid development of the mobile internet, the proportion of mobile network users in the total network users is greatly increased. The service and product provided by each mobile operator has smaller and smaller difference, the requirement of the user for the mobile service and service is higher and higher, and the complaint amount of the user is increased. An increase in the amount of complaints is necessarily accompanied by a decrease in mobile user satisfaction. The user satisfaction is not only used as an important assessment index of the mobile operator, but also an important means for measuring the network quality and perception of the operator, and an important assessment standard for reflecting the user viscosity.

Currently, for mobile operators, their interest in user satisfaction has gradually shifted from voice services, tariffs, etc. to network quality. The traditional method is that a mobile operator adopts market research and customer complaints to learn the perception satisfaction degree of a customer, but the method has low timeliness and small coverage. In view of the above, the present embodiment provides a satisfaction prediction method, please refer to the flowchart shown in fig. 1:

s102: the network equipment acquires the business record data of the user to be predicted.

In this embodiment, the network device implementing the satisfaction prediction method may include, but is not limited to, a device deployed in a core network, such as a device in a network management system. The "user to be predicted" refers to a user whose satisfaction needs to be predicted, i.e., an object of satisfaction prediction. It is understood that the number of the users to be predicted may be one, or two or more, and for a mobile operator, when the user satisfaction needs to be known, it may be a batch evaluation prediction of the user satisfaction in some areas, in which case, the number of the users to be predicted may include a plurality.

In this embodiment, the service of the user includes, but is not limited to, at least one of video, game, browsing and downloading, mailbox, Voice Over Long Term Evolution (VOLTE), and the like, where the service record data refers to data recorded by the user in the process of implementing the service in the network service process on the network side, for example, at least one of control plane data and user plane data. In some examples of this embodiment, the traffic record data obtained by the network device may be collected via at least one of the s1-mme interface and the s1-u interface.

It should be understood that the user service record data acquired by the network device is not necessarily acquired in real time, for example, in some examples of the embodiment, after the service record data of one user is acquired, the service record data may be stored in a hadoop (hadoop Distributed File system) cluster. When the network device needs to predict the user satisfaction of the user, the service record data of the user can be extracted from the hadoop cluster.

S104: and the network equipment determines the service KPI of the user to be predicted in each scene according to the service record data.

After the service record data of the user to be predicted is obtained, the network device can determine the service KPI of the user to be predicted in each scene according to the obtained service record data, and the service KPI is an index capable of reflecting the service perception of the user. The corresponding service KPIs under several services are shown in table 1:

TABLE 1

In this embodiment, the service KPIs of the user to be predicted in each scene are respectively determined, instead of determining the uniform service KPIs of the user to be predicted in all scenes, or instead of determining the service KPIs of the user to be predicted in part of scenes, mainly because for one user, there is different tolerance psychology for the service in different scenes, for example, when the user performs video service in an office, the service record data is a, and the video comparison is stuck; the service record data of the video service performed by the user at home is also a, and the video is similarly stuck, but the satisfaction degrees in the two scenes are not the same for the user. Since for the user in the office there are many people using the network at the same time, there is a psychological expectation for the network to be stuck in the user's mind, considering that it is normal that some stuck occurs in this environment, but when this degree of stuck occurs at home, there is no psychological expectation for the user because there are few people using the network at home for the user, and it is not understandable that stuck occurs again in this situation. Therefore, the tolerance of the user to video traffic jams at home is lower than that of the user at the company.

Similarly, the tolerance of the users to the same network state is different when the users perform different services. For example, in some examples of the embodiment, the network status presented by the user when downloading the file is substantially consistent with the network status presented by the corresponding service record data when the user plays the game service, but the user may think that the network status is worse when playing the game service.

Two examples of scene division according to the location of the user or the service category performed by the user are given above, but in other examples of the present embodiment, the scene may be divided according to the user rank, for example, two users with a high user rank and a low user rank may have completely different feelings under the same network state. In addition, as will be understood by those skilled in the art, when performing scene division, at least two factors of the three factors, i.e., the location of the user, the service category performed by the user, and the user rating, may also be combined.

Because the user to be predicted has different tolerances to the service in different scenes, in order to ensure that the predicted prediction satisfaction better conforms to the psychological feeling of the user to be predicted in the corresponding scene, in this embodiment, corresponding user tolerance weights may be set for each scene.

Assuming that the scene is divided according to the location of the user, for example, the scene is divided into a work place scene and a residence place scene according to the location of the user, a process of determining the service KPI of the user to be predicted in each scene according to the service record data of the user to be predicted may be performed with reference to the flowchart shown in fig. 2:

s202: and for any scene, determining N resident cells of the user to be predicted in the scene according to the service record data.

For example, assuming that N is equal to 3, the network device may calculate the first 3 resident cells of the user to be predicted in the corresponding scene according to the ticket data and the internet surfing data of the user to be predicted. In this embodiment, assume that the user is at a place of work from 8:00 to 18:00 and at a place of residence between 20:00 and 6:00 the next day. Therefore, the network device can determine which data are generated in the working period and which data are generated in the home period according to the generation time of the user service record data to be predicted. Taking the working place scene as an example, after the network device determines the service record data generated in the working period, it can determine the cells to which the data generation positions belong, and the cells are most likely to be the working places of the users. It should be immediately that the service record data generated by the user to be predicted in the working period is not necessarily generated in the working place, so after determining the cells in which the service record data generated in all working periods are located, the network device may select the first 3 cells generating the most service record data in the working period as 3 resident cells corresponding to the working place scenario. Similarly for the residence, and will not be described in detail here.

S204: and determining a scene area corresponding to the scene based on the coverage areas of the N resident cells.

For the work place, the network device may determine the coverage areas of the first 3 residential cells corresponding to the scene, then determine the center point positions of the coverage areas of the three residential cells, for example, determine the longitude and latitude of the center point, then use the center point position as the center of a circle, and determine a circular range as the scene area of the work place scene with the preset scene area radius. It should be understood that the scene area determined by the network device for a certain scene is not necessarily circular, for example, in some other examples of the embodiment, after determining the central point position, the network device determines a square scene area based on the central point position and the preset side length of the scene area. In other examples, the scene area may also be a triangle, a parallelogram, or even other irregular shape. For a residential area, the network device may also determine a scene area of the scene in a similar manner.

S206: and extracting the business record data in the scene area from the business record data of the user to be predicted.

Continuing to take the working place scene as an example, after the network device determines the scene area corresponding to the working place scene, the service record data belonging to the scene area may be extracted from the service record data of the user to be predicted, and the extracted service record data is used to calculate the service KPI of the user to be predicted in the working place scene. Similarly for the residential site scene, the network device extracts the part generated in the scene area corresponding to the residential site scene from the business record data of the user to be predicted.

S208: and determining the service KPI of the user to be predicted in the scene according to the extracted service record data.

After extracting the service record data generated by the user to be predicted in the scene area corresponding to a certain scene, the network device may perform calculation of the scene service KPI according to the extracted service record data. For example, in this embodiment, the network device determines a first service KPI according to service record data of a user to be predicted in a scene area corresponding to a work place scene, where the first service KPI is a service KPI of the user to be predicted in the work place scene; and the network equipment determines a second service KPI according to the service record data of the user to be predicted in the scene area corresponding to the residence scene, wherein the second service KPI is the service KPI of the user to be predicted in the residence scene.

S106: and the network equipment inputs the service KPI of the user to be predicted in each scene into a pre-established satisfaction prediction model to obtain the prediction satisfaction corresponding to the user to be predicted.

In some examples of this embodiment, the network device may directly input the service KPIs of the user to be predicted into the satisfaction prediction model after obtaining the service KPIs of the user to be predicted in each scene, however, in some examples of this embodiment, before the network device inputs the service KPIs of the user to be predicted in each scene into the satisfaction prediction model, some preprocessing may be performed on the service KPIs of the user to be predicted in each scene, for example, an abnormal value or a null value may be removed therefrom, and for example, control in the service KPIs of the user to be predicted in each scene is filled with a "0" value.

In some examples of this embodiment, when the network device preprocesses the service KPIs of the user to be predicted in each scene, the network device may perform orthogonal normalization processing on the service KPIs, so that occurrence of an over-fitting phenomenon can be effectively avoided.

In this embodiment, the satisfaction degree prediction model is created based on the service KPI of the investigation user and the investigation satisfaction degree data of the investigation user, if the service KPI is used as the independent variable "X", the satisfaction degree data is the dependent variable "Y", and the satisfaction degree prediction model embodies the relationship between X and Y, and in brief, the satisfaction degree prediction model can be regarded as a relationship function between X and Y. The investigation user is a part of all users, so the service KPI of the investigation user can be regarded as X ', and the satisfaction data of the investigation user is Y', in the process of constructing the satisfaction prediction model, a relationship function between X 'and Y' is actually derived based on the corresponding relationship between X 'and Y', and the relationship between the two is used as the relationship function between X and Y, that is, as the satisfaction prediction model.

Because each scene has the corresponding user tolerance weight, when the network equipment constructs the satisfaction prediction model, the network equipment is not only carried out according to the service KPI of the research user and the research satisfaction data, but also combined with the user tolerance weight corresponding to each scene. The process of obtaining a satisfaction prediction model is described below in conjunction with the flow chart shown in fig. 3:

s302: and acquiring the business record data and the research satisfaction data of a plurality of research users.

In this embodiment, the stage of obtaining the satisfaction prediction model is trained, or before this stage, the mobile operator needs to perform investigation, and investigation satisfaction data of part of users are sampled. The satisfaction data refers to data that can represent the satisfaction of the user with the business service, and in some examples of the embodiment, the satisfaction data may include, but is not limited to, NPS (Net Promoter Score), which is also called Net Promoter Score, also called word of mouth, and is an index that measures the possibility that a certain client will recommend a certain business or service to others. In other examples of the present embodiment, the satisfaction data may also be characterized by complaint work order data or the like.

Assuming that the mobile operator is conducting research on 100 people, 100 research reports are obtained, and the 100 research reports can respectively represent research satisfaction data of 100 research users. The network device may obtain the service record data of the 100 research users, for example, the network device extracts the service record data of the 100 research users from the hadoop cluster.

S304: and determining the service KPI of the research user in each scene according to the service record data of the research user.

And after the business record data of each research user is obtained, the network equipment determines the business KPI of each research user in each scene according to the obtained business record data. For example, for a service KPI of a certain research user w in scene 1, the network device may divide the service record data of the research user, acquire the service record data of the research user in scene 1, and then calculate the service KPI of the research user w in scene 1 based on the service record data of the research user w in scene 1. Similarly, the network device may also extract the business record data of the research user w in the scene 2, and then determine the business KPI of the research user w in the scene 2 based on the extracted business record data.

If the scene is divided according to the location of the user, for example, the scene is divided into a work place scene and a residence place scene according to the location of the user, the process of determining the service KPI of the research user in each scene according to the service record data of the research user is specifically similar to the process of determining the service KPI of the user to be predicted in each scene according to the service record data of the user to be predicted in fig. 2, please refer to fig. 4:

s402: for any scene, determining N resident cells of a research user in the scene according to the service record data;

s404: determining a scene area corresponding to a scene based on the coverage areas of the N resident cells;

s406: extracting business record data in scene areas from the business record data of the research users;

s408: and determining the service KPI of the research user in the scene according to the extracted service record data.

For the process of determining the service KPI of the research user in each scene according to the service record data of the research user, the previous process of determining the service KPI of the user to be predicted in each scene may be referred to, and details are not repeated here.

S306: and training according to the user tolerance weight of each scene, the service KPI of the investigation user in each scene and the investigation satisfaction data of the investigation user to obtain a satisfaction prediction model.

After determining the service KPIs of the research user in each scene, the network device can perform model training according to the user tolerance weight corresponding to each scene, the service KPIs of the research user in each scene, and the research satisfaction data of the research user. In this embodiment, the service KPIs of a part of research users in each scene and the research satisfaction data of the research users may be divided into training sets for training the training models; and the service KPI of the other part of investigation users under each scene and the investigation satisfaction data of the investigation users are divided into a test set for testing the training model trained by the training set so as to determine whether the prediction accuracy of the training model meets the requirement and is enough to be used as a satisfaction prediction model.

In some examples of this embodiment, the training set and the test set may be randomly divided when they are divided. For example, in some examples of the present embodiment, the network device may randomly divide the service KPI and research satisfaction data of 1/4 research users in each scenario into a test set, and divide the service KPI and research satisfaction data of the remaining 3/4 research users in each scenario into a training set. For example:

x_train,x_test,y_train,y_test＝train_test_split(X’,Y’,test_size＝0.25,random_state＝1)

wherein, X 'is a service KPI of the investigation user in each scene, Y' is satisfaction data of the investigation user, such as NPS, and "test _ size" refers to the size of the test set, that is, the proportion of the test set data to the total data amount, "random _ state" refers to the division method adopted when the test set and the training set are divided, and "random _ state 1" refers to the random division method.

In some examples of this embodiment, the network device may implement training of the satisfaction prediction model based on a random forest classification algorithm, for example, please refer to a flowchart of fig. 5 that shows training by using the random forest classification algorithm to obtain the satisfaction prediction model:

s502: and (3) adopting a GridSearchCV grid to search training parameters trained by a random forest classification algorithm.

In this embodiment, the training parameters include the number of decision trees and the calculation attributes, and the network device may search the currently optimal number of decision trees and calculation attributes of the random forest classification algorithm by using the GridSearchCV grid.

S504: and determining a training model according to the training parameters, the user tolerance weight of each scene, the service KPI in the training set and the investigation satisfaction data.

It is understood that the number of decision trees and the calculation attribute are both factors for determining the training model, and when any one of the number of decision trees and the calculation attribute is changed, the corresponding village model is also changed.

S506: and inputting the service KPI in the test set into a training model to predict the satisfaction degree.

When a training model is determined, the network equipment can input the service KPI of the centralized testing investigation user in each scene into the training model, and the training model carries out prediction based on the input service KPI, so that the prediction satisfaction degree of the corresponding investigation user is obtained.

S508: and (5) checking the prediction accuracy of the training model according to the investigation satisfaction data in the test set.

After the training model aims at the prediction satisfaction of the investigation user in the test set, the network device can check the prediction accuracy of the training model based on the investigation satisfaction data in the test set.

S510: and when the prediction accuracy of the training model meets the requirement, saving the training model as a satisfaction prediction model.

If the network equipment determines that the prediction accuracy of the satisfaction prediction of the current training model to the test centralized investigation user meets the requirement, the current training model is excellent enough, so that the training model can be directly used as a satisfaction prediction model for subsequently acquiring the satisfaction of the user to be predicted. If the prediction accuracy of the training model does not meet the requirement through detection, the training of the training model is required to be continued until the prediction accuracy meets the requirement.

It should be understood that, when the satisfaction prediction model is obtained by training according to the user tolerance weight of each scene, the service KPI of the investigation user in each scene, and the investigation satisfaction data of the investigation user, the algorithm used is not limited to the random forest algorithm, for example, in some examples of this embodiment, the network device may also use any one of an SVM (Support Vector Machine) algorithm, a bayesian classification algorithm, a BP (Back Propagation) neural network algorithm, and the like, but the satisfaction prediction model determined by the random forest algorithm is more advantageous in terms of accuracy and generalization than the SVM algorithm, the bayesian classification algorithm, the BP neural network algorithm, and the like.

According to the satisfaction degree prediction method provided by the embodiment, when the satisfaction degree of a certain user needs to be determined, the satisfaction degree of the user on the business service can be obtained without waiting for complaints of the user or relying on investigation on the user, the timeliness of obtaining the satisfaction degree of the user by a mobile operator is improved, the mobile operator can find the user with poor experience and can provide a coping strategy in time, and the user experience is improved.

Furthermore, tolerance weights of the user under different scenes are considered when the satisfaction degree prediction model is trained, so that the accuracy of the trained satisfaction degree prediction model in user satisfaction degree prediction is higher.

In addition, the satisfaction degree prediction model is trained based on the random forest classification algorithm, and the accuracy and the generalization of the trained satisfaction degree prediction model are further improved.

Example two:

in order to make the advantages and details of the foregoing satisfaction prediction method more clear to those skilled in the art, the present embodiment will be further described with reference to examples:

the satisfaction prediction model construction based on big data is mainly divided into two stages: a user tolerance model cleaning stage and a user satisfaction model training stage. The specific system algorithm diagram is shown in FIG. 6:

the user tolerance model cleaning stage comprises the following steps:

combining the target 5W principle of communication (whoever, wheever, whenever, whomever, whatever, that is, anyone can make any form of communication with anyone at any time and any place), through pearson correlation analysis, the irrelevant input variable factors of anyone and any time are removed, and the user network tolerance model is determined by the user level, the location and the service category.

And (4) carrying out scene classification based on the position, including a residence place and a work place, counting the cell information of the residence place and the work place of the user according to the call ticket data and the internet surfing data of the user, and selecting a resident top3 cell. For a residential place, the tolerance of a user on internet surfing and conversation is low, and once a network card appears or a call is dropped, the satisfaction of the user is easily influenced, and the work is carried out again. The user tolerance weight values under different scenes are different, so that a user tolerance weight matrix S under different scenes can be constructed.

For the working place scene area, determining the cell in which each user works in the working period, determining the longitude and latitude of the top3 cell, finding out the central longitude and latitude points of the three cells, taking the central longitude and latitude as the center of a circle, extending 10 kilometers to the periphery, and finding out all cell lists under the circle coverage area as the working scene area range. The determination is also carried out on the scene area of the residential area, and the description is omitted here.

Selecting the recorded data of the users in the area, classifying the general classes of the services, such as videos, games, browsing and downloading, finance and economics, application stores, business offices (mailboxes), voice calls and VOLTE (voice over long term evolution) according to group specifications, calculating KPI (key performance indicator) values of the users of different services, and aggregating the KPI values into a service KPI matrix B according to month granularity.

Combining a user tolerance system matrix S and a user service KPI matrix B of a user scene, and performing matrix product S x B on the two matrixes to serve as model input; taking the NPS value in the operator questionnaire as output, the basic satisfaction model is as follows:

(II) a user satisfaction model training stage:

the perception index data under different scenes (any one of different user grades, different service types and different positions) is used as original data of a user satisfaction degree model, data preprocessing is firstly carried out on the perception index data, abnormal data are eliminated, characteristic values are extracted, and a data set is divided into a training set and a testing set.

Inputting the data in the training set into a random forest classification model, and searching the number n _ estimators and the calculation attribute criterion of the optimal decision tree of the random forest classification algorithm through a GridSearchCV grid; and continuously inputting the test set data into the training model for cross validation until a relatively better training model is found as a final satisfaction degree prediction model.

clf＝RandomForestClassifier()

grid_obj＝GridSearchCV(clf,param,scoring＝acc_scorer)

grid_obj＝grid_obj.fit(x_train,y_train.ravel())

clf＝clf.fit(x_train,y_train.ravel())

And inputting the test set data into the training model for prediction verification, wherein the model result can display the minimum mean square error between the prediction result and the real result, and the network equipment can select the training model with the minimum mean square error sufficiently small as the optimal training model, namely the final satisfaction prediction model is stored.

After the satisfaction degree prediction model is trained, the application stage of the satisfaction degree prediction model can be entered:

collecting internet surfing and conversation data of a user with user satisfaction to be predicted;

cleaning and counting the data according to a method of a training stage, and inputting the processed data into a previously trained satisfaction degree prediction model;

the user internet surfing and call data can output the predicted satisfaction (NPS) of the user through calculation of the satisfaction prediction model.

The following describes clearly and completely the technical solution of the satisfaction prediction method adopted in this embodiment with reference to fig. 7:

(1) and (5) classifying scenes.

Selecting a batch of users of the user satisfaction survey questionnaire, inquiring recorded data of control surfaces s1-mme of the batch of users from a hadoop cluster of a list XDR, carrying out statistics according to the activity time rule of the users in a workplace (residence), giving different scores to cells with different activity time rules, calculating scores and values of each cell of each user according to the granularity of days, months and the like, further calculating the confidence score of each user in each cell, clustering all cells of each user according to the scores, and finding out a top3 cell of each user as a workplace (residence) cell record.

(2) The cell is expanded into a scene area.

According to the longitude and latitude of three cells of each user, finding out the central longitude and latitude points of the three cells in a weighting mode, taking the central longitude and latitude as the center of a circle, extending 20 kilometers to the periphery, and finding out all cell lists under the circle coverage range as the scene area range of a working (living) place.

(3) And (4) service KPI statistics of the work (residential) area.

According to the user dimension, the flow and the internet surfing time of the user in the working (residential) area are counted and used as user grade classification reference values, and meanwhile perception indexes of the user under different services in different areas are counted respectively, wherein the specific related services and indexes are shown in table 1.

(4) And (5) constructing a training model.

And setting the calculated service KPI value of the user as an independent variable X 'and setting the corresponding user satisfaction value as a dependent variable Y'. Therefore, the problem is converted into a mathematical problem, the user satisfaction Y 'is determined by the environment variable X', the problem can be regarded as a classification problem, and a random forest classification algorithm in a machine learning algorithm in the classification field, as shown in fig. 8, has better accuracy and generalization compared with an SVM algorithm, a Bayesian classification algorithm, a BP neural network algorithm and the like.

(5) And training the model to optimize the model parameters.

Inputting the training data set into a random forest classification model, and searching the number n _ estimators and the calculation attribute criterion of the optimal decision tree of the random forest classification algorithm through a GridSearchCV grid to obtain the training model.

(6) And measuring a model accuracy mechanism.

And inputting the test set data into the training models for prediction verification, if the error of one training model is small enough, selecting the training model as a satisfaction degree prediction model, and if not, continuously adjusting the model parameters until the error of the training model is small enough.

And (4) verifying the minimum mean square error between the predicted value and the true value of the test set data by cross validation. If the error is smaller, the model is better, otherwise, the error is poor. Each time the accuracy of the prediction data set of the model is recorded, the model with the highest accuracy is selected and stored.

(7) And collecting real-time XDR data of a user with user satisfaction to be predicted.

Randomly selecting a mobile user in an area, selecting control surfaces s1-mme and user surface http, web, video, app and im list data of the user from an XDR list cluster, dividing the scenes of a working place and a living place of the user by using the same training method, and calculating service KPIs (Key Performance indicators) in different scenes.

(8) And preprocessing the service KPI data of the user.

Abnormal or null value data may exist in the service KPI data of the user, the abnormal data are replaced by 0, and a plurality of indexes corresponding to the training model are selected as characteristic values. The characteristic value data is subjected to orthogonal normalization processing, so that the phenomenon of over-fitting can be effectively avoided.

(9) And inputting business KPI data of the user into a random forest model for fitting.

And inputting the user service KPI after statistical processing into a previously trained satisfaction prediction model, and obtaining a satisfaction prediction result aiming at the user through fitting of the satisfaction prediction model.

Example three:

the present embodiment provides a storage medium, in which one or more computer programs that can be read, compiled and executed by one or more processors are stored, and in the present embodiment, the storage medium may store a satisfaction prediction program, which can be executed by one or more processors to implement the flow of any one of the satisfaction prediction methods described in the foregoing embodiments.

In addition, the present embodiment provides a base station, as shown in fig. 9: the network device 90 includes a processor 91, a memory 92, and a communication bus 93 for connecting the processor 91 and the memory 92, wherein the memory 92 may be the aforementioned storage medium storing the satisfaction prediction program. The processor 91 may read the satisfaction prediction program, compile and execute the procedures of implementing the satisfaction prediction methods described in the foregoing embodiments:

the processor 91 obtains service record data of the user to be predicted, determines service KPIs of the user to be predicted in each scene according to the service record data, and then inputs the service KPIs of the user to be predicted in each scene into a pre-established satisfaction prediction model to obtain the prediction satisfaction corresponding to the user to be predicted. In this embodiment, the satisfaction prediction model is created by the processor 91 based on the user tolerance weight corresponding to each scene, the service KPIs of a plurality of research users in each scene, and the research satisfaction data of each research user, where the service KPIs of the research users in each scene is determined by the processor 91 according to the service record data of the research users.

In some examples of this embodiment, before the processor 91 acquires the service record data of the user to be predicted, the service record data and the research satisfaction data of a plurality of research users are acquired, the service KPIs of the research users in each scene are determined according to the service record data of the research users, and the satisfaction prediction model is obtained by training according to the user tolerance weight of each scene, the service KPIs of the research users in each scene, and the research satisfaction data of the research users.

In some examples of this embodiment, the processor 91 may train to obtain the satisfaction prediction model by using a random forest classification algorithm according to the user tolerance weight of each scene, the KPI of the investigation user in each scene, and the investigation satisfaction data of the investigation user.

In some examples, part of the research users belong to a training set under each scene, and the other part of the research users belong to a test set under each scene; the processor 91 may search training parameters for training by using a GridSearchCV grid to search a random forest classification algorithm when training by using the random forest classification algorithm to obtain a satisfaction prediction model according to the user tolerance weight of each scene, the service KPI of the investigation user in each scene, and the investigation satisfaction data of the investigation user, wherein the training parameters include the number of decision trees and the calculation attribute; and then determining a training model according to the training parameters, the user tolerance weight of each scene, the service KPI in the training set and the investigation satisfaction data. After the training model is determined, the processor 91 inputs the service KPI in the test set into the training model for satisfaction prediction, and then checks the prediction accuracy of the training model according to the investigation satisfaction data in the test set; upon determining that the prediction accuracy of the training model meets the requirement, the processor 91 saves the training model as a satisfaction prediction model.

In some examples, before the processor 91 inputs the service KPI of the user to be predicted in each scene into the pre-created satisfaction prediction model to obtain the prediction satisfaction corresponding to the user to be predicted, the processor may perform orthogonal normalization processing on the service KPI of the user to be predicted in each scene.

Optionally, the research satisfaction data comprises a net recommendation NPS.

Optionally, the scene is divided based on at least one of a user level, a location of the user, and a service category.

If the scene is divided according to the location of the user, when the processor 91 determines the service KPI of the user to be predicted in each scene according to the service record data, for any scene, N residential cells of the user to be predicted in the scene may be determined according to the service record data, and then the scene area corresponding to the scene is determined based on the coverage area of the N residential cells. Subsequently, the processor 91 extracts the service record data in the scene area from the service record data of the user to be predicted, and determines the service KPI of the user to be predicted in the scene according to the extracted service record data.

When the network device provided by the embodiment knows the user satisfaction, the timeliness of knowing the user satisfaction by the mobile operator is improved because the network device does not depend on investigation or the complaint result of the user; meanwhile, the satisfaction degree prediction model can be used for predicting the satisfaction degree of any user to the business service, so that the comprehensiveness of the mobile operator in knowing the satisfaction degree of the user is improved. In addition, the characteristic that the tolerance of the user to the service is different in different scenes is considered, and the corresponding user tolerance weight is configured for each scene, so that the satisfaction prediction model can be created based on the user tolerance weight corresponding to each scene and the service KPI and investigation satisfaction data of a plurality of investigation users in each scene, the user satisfaction can be prevented from being predicted and evaluated according to the unified rigidity standard in all scenes, and the satisfaction prediction accuracy is improved.

It will be apparent to those skilled in the art that all or some of the steps of the methods, systems, functional modules/units in the devices disclosed above may be implemented as software (which may be implemented in program code executable by a computing device), firmware, hardware, and suitable combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be performed by several physical components in cooperation. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed over computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media), executed by a computing device, and in some cases may perform the steps shown or described in a different order than here. The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art. Thus, the present invention is not limited to any specific combination of hardware and software.

The foregoing is a more detailed description of embodiments of the present invention, and the present invention is not to be considered limited to such descriptions. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.

Claims

1. A satisfaction prediction method comprising:

acquiring service record data of a user to be predicted;

determining a business key performance indicator KPI of the user to be predicted under each scene according to the business record data, wherein each scene has a corresponding user tolerance weight;

inputting the service KPI of the user to be predicted under each scene into a pre-established satisfaction prediction model to obtain the corresponding prediction satisfaction of the user to be predicted; the satisfaction prediction model is created based on user tolerance weight corresponding to each scene, service KPIs of a plurality of research users in each scene and research satisfaction data of each research user, and the service KPIs of the research users in each scene are determined according to service record data of the research users.

2. The satisfaction prediction method of claim 1, wherein prior to obtaining business record data for a user to be predicted, further comprising:

acquiring service record data and investigation satisfaction data of a plurality of investigation users;

determining the service KPI of the investigation user in each scene according to the service record data of the investigation user;

and training according to the user tolerance weight of each scene, the service KPI of the investigation user in each scene and the investigation satisfaction data of the investigation user to obtain a satisfaction prediction model.

3. The satisfaction prediction method of claim 2, wherein said training according to the user tolerance weight of each of said scenarios, the KPIs of said research user in each scenario, and the research satisfaction data of said research user to obtain a satisfaction prediction model comprises:

and training by adopting a random forest classification algorithm according to the user tolerance weight of each scene, the service KPI of the investigation user in each scene and the investigation satisfaction data of the investigation user to obtain a satisfaction prediction model.

4. The satisfaction prediction method of claim 3, wherein part of said plurality of research users belonging to said training set belong to service KPIs and research satisfaction data of each scene, and another part of said plurality of research users belonging to said testing set belong to said service KPIs and research satisfaction data of each scene; the training by adopting a random forest classification algorithm according to the user tolerance weight of each scene, the service KPI of the investigation user in each scene and the investigation satisfaction data of the investigation user to obtain the satisfaction prediction model comprises the following steps:

training parameters for training by adopting a GridSearchCV grid search random forest classification algorithm, wherein the training parameters comprise the number of decision trees and calculation attributes;

determining a training model according to the training parameters, the user tolerance weight of each scene, the service KPI in the training set and the investigation satisfaction data;

inputting the service KPI in the test set into the training model to predict the satisfaction degree;

checking the prediction accuracy of the training model according to the investigation satisfaction data in the test set;

and when the prediction accuracy of the training model meets the requirement, saving the training model as the satisfaction prediction model.

5. The satisfaction prediction method according to any of claims 1-4, wherein before inputting the service KPI of the user to be predicted under each scene into the pre-created satisfaction prediction model to obtain the prediction satisfaction corresponding to the user to be predicted, further comprising:

and performing orthogonal normalization processing on the service KPI of the user to be predicted in each scene.

6. A satisfaction prediction method according to any of claims 1-4, characterized in that said research satisfaction data comprises the net recommendation NPS.

7. A satisfaction prediction method according to any of claims 1-4, characterized in that said scenes are divided based on at least one of user level, location of the user and traffic class.

8. The satisfaction prediction method of claim 7, wherein if the scenario is divided according to the location of the user, the determining the service KPI of the user to be predicted in each scenario according to the service record data comprises:

for any scene, determining N resident cells of the user to be predicted in the scene according to the service record data;

determining a scene area corresponding to the scene based on the coverage areas of the N resident cells;

extracting service record data in the scene area from the service record data of the user to be predicted;

and determining the service KPI of the user to be predicted under the scene according to the extracted service record data.

9. A network device comprising a processor, a memory, and a communication bus;

the processor is configured to execute one or more programs stored in the memory to implement the steps of the satisfaction prediction method according to any of claims 1 to 8.

10. A storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the steps of the satisfaction prediction method of any of claims 1-8.