CN107818376A - Customer loss Forecasting Methodology and device - Google Patents
Customer loss Forecasting Methodology and device Download PDFInfo
- Publication number
- CN107818376A CN107818376A CN201610818511.3A CN201610818511A CN107818376A CN 107818376 A CN107818376 A CN 107818376A CN 201610818511 A CN201610818511 A CN 201610818511A CN 107818376 A CN107818376 A CN 107818376A
- Authority
- CN
- China
- Prior art keywords
- user data
- feature
- user
- extraction
- unit
- 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
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
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Resources & Organizations (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of customer loss Forecasting Methodology and device, it is related to data processing field.Customer loss Forecasting Methodology therein includes:User data is classified, user data is marked as being lost in user and is not lost in user;Portion of user data is extracted respectively from each classification of user data;According to the features training disaggregated model of the user data of extraction, grader is obtained;Predict whether user to be measured can be lost in using grader.Classify by using to user data, partial data is extracted respectively from each classification again, with training be used for predict user whether the disaggregated model that can be lost in, the characteristics of data of train classification models, which can be used in, can fully demonstrate original user data, so as to more accurately predict whether user can be lost in.
Description
Technical field
The present invention relates to data processing field, more particularly to a kind of customer loss Forecasting Methodology and device.
Background technology
Currently, enterprise will more pay attention to existing user while focusing on attracting and developing new user.And keep existing use here
The premise at family is to understand existing user, predicts the possibility of customer loss, is taken before user embodies and is lost in sign corresponding
Keep measure, customer loss can be prevented, improve the effectiveness of operation of enterprise.
During existing customer loss forecast model models, it is modeled using the sample data randomly selected.
However, because the randomness of sample makes the accuracy of forecast model relatively low, so reduce prediction user whether be lost in it is accurate
Rate.
The content of the invention
A technical problem to be solved of the embodiment of the present invention is:How the accuracy of customer loss prediction is improved.
One side according to embodiments of the present invention, there is provided a kind of customer loss Forecasting Methodology, including:To number of users
According to being classified, user data is marked as being lost in user and is not lost in user;From each classification of user data respectively
Extract portion of user data;According to the features training disaggregated model of the user data of extraction, grader is obtained;It is pre- using grader
Survey whether user to be measured can be lost in.
In one embodiment, carrying out classification to user data includes:According to related to customer loss in user data
Feature, user data is classified using the method for cluster.
In one embodiment, part user data package is extracted from each classification of user data to include:From user data
Each classification in extract the user data of preset ratio respectively.
In one embodiment, according to the features training disaggregated model of the user data of extraction, obtaining grader includes:From
Some groups of user data are randomly choosed in the user data of extraction;Decision tree is respectively trained using some groups of user data of selection
Model;Some decision trees that some groups of user data using selection are trained to obtain are collectively as grader.
In one embodiment, included according to the features training disaggregated model of the user data of extraction:Count the use extracted
The species of all values of each feature in user data;If the species of all values of feature is more than preset value, by feature
Each value is respectively as the new feature of the user data of extraction, and the species for deleting all values is more than the spy of preset value
Sign;According to the features training disaggregated model of the user data of the extraction after processing.
In one embodiment, included according to the features training disaggregated model of the user data of extraction:From the user of extraction
In data, using lasso trick algorithms selection feature;According to the features training disaggregated model of the selection of the user data of extraction.
In one embodiment, the user data of mark is obtained using following methods:Obtain the tool in very first time unit
There is the user data of some features;The shape whether user data in very first time unit is lost in the second time quantum
State, mark the user data in very first time unit;Wherein, very first time unit is the adjacent first time of the second time quantum
Unit.
Second aspect according to embodiments of the present invention, there is provided a kind of customer loss prediction meanss, including:User data point
Generic module, for classifying to user data, user data is marked as being lost in user and is not lost in user;User data
Abstraction module, for extracting portion of user data respectively from each classification of user data;Model training module, for basis
The features training disaggregated model of the user data of extraction, obtain grader;Customer loss prediction module, for pre- using grader
Survey whether user to be measured can be lost in.
In one embodiment, user data sort module is further used for according to related to customer loss in user data
Feature, user data is classified using the method for cluster.
In one embodiment, user data abstraction module is further used for taking out respectively from each classification of user data
Take the user data of preset ratio.
In one embodiment, model training module includes:Grouped data extracting unit, for the user data from extraction
Some groups of user data of middle random selection;Decision tree training unit, for being respectively trained using some groups of user data of selection
Decision-tree model;Grader forms unit, for will train obtained some decision trees using some groups of user data of selection
Collectively as grader.
In one embodiment, model training module includes:Feature value species statistic unit, for counting the use extracted
The species of all values of each feature in user data;Feature Conversion unit, the species for all values when feature are more than
During preset value, using each value of feature as the new feature of the user data of extraction, and all values are deleted
Species is more than the feature of preset value;First model training unit, the feature for the user data according to the extraction after processing are instructed
Practice disaggregated model.
In one embodiment, model training module includes:Feature selection unit, for from the user data of extraction,
Using lasso trick algorithms selection feature;Second model training unit, the features training for the selection of the user data according to extraction
Disaggregated model.
In one embodiment, in addition to user data acquisition module, user data acquisition module include:User data obtains
Unit is taken, for obtaining the user data with some features in very first time unit;User data indexing unit, for root
The state whether being lost in the second time quantum according to the user data in very first time unit, mark the use in very first time unit
User data;Wherein, very first time unit is the adjacent first time quantum of the second time quantum.
The present invention classifies by using to user data, then extracts partial data respectively from each classification, to train use
In prediction user whether the disaggregated model that can be lost in, can be used in the data of train classification models can fully demonstrate original use
The characteristics of user data, so as to more accurately predict whether user can be lost in.
By referring to the drawings to the present invention exemplary embodiment detailed description, further feature of the invention and its
Advantage will be made apparent from.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the required accompanying drawing used in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is the flow chart of one embodiment of customer loss Forecasting Methodology of the present invention.
Fig. 2 is the flow chart of another embodiment of customer loss Forecasting Methodology of the present invention.
Fig. 3 is the structure chart of one embodiment of customer loss prediction meanss of the present invention.
Fig. 4 is the structure chart of another embodiment of customer loss prediction meanss of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, rather than whole embodiments.Below
Description only actually at least one exemplary embodiment is illustrative, is never used as to the present invention and its application or makes
Any restrictions.Based on the embodiment in the present invention, those of ordinary skill in the art are not making creative work premise
Lower obtained every other embodiment, belongs to the scope of protection of the invention.
The customer loss Forecasting Methodology of one embodiment of the invention is described below with reference to Fig. 1.
Fig. 1 is the flow chart of one embodiment of customer loss Forecasting Methodology of the present invention.As shown in figure 1, the embodiment
Method includes:
Step S102, classifies to user data, and user data is marked as being lost in user and is not lost in user.
In step s 102, user data is the data for training.User data is multidimensional data, user data it is each
Individual dimension is each feature of user data.
Data source used in being predicted to customer loss can include user attribute data and user behavior data.With
Family attribute data is mainly the static data of user, includes some essential informations of user;User behavior data mainly includes using
The communication data at family and service data etc..
Wherein it is possible to obtain user data using following methods:First, obtaining in very first time unit has some spies
The user data of sign;Then, the state that whether user data in very first time unit is lost in the second time quantum, mark
Remember the user data in very first time unit.Wherein, very first time unit is the adjacent first time quantum of the second time quantum.
For example, the user data of August can be obtained, the use of the August then obtained according to user in the status indication whether being lost in of September
User data.So as to, it is stronger using the feature in user data and the relevance of the attrition status of user, it is more suitable so as to establish
Answer the grader of practical business scene.
During classifying to user data, user data can be classified according to class of service, also may be used
To be classified according to the part feature related to customer loss to user data.
One embodiment is, according to feature related to customer loss in user data, using the method for cluster to user
Data are classified.
For example, for the user of operator, the feature related to customer loss can be preferential classification, set meal class
Type, the preferential subsidy amount of money, total flow etc., features described above can reflect the service condition of user and the service scenario of operator,
Therefore can be as the feature related to customer loss.
It is thus possible to form multiple classifications according to the feature of user data in itself, user can be described exactly and forms feelings
Condition, it is consistent with actual conditions.
Step S104, portion of user data is extracted respectively from each classification of user data.
The user data extracted can be distributed from each classification according to the data total amount required for train classification models
Amount, the user of each type can be covered thereby using the data in train classification models.
A kind of extraction mode is to extract the user data of preset ratio respectively from each classification of user data.For example,
10% user data can be extracted from each classification.The user data for being designed for train classification models can also be calculated
Quantity take user data total amount ratio, and extract from each classification the user data of the ratio.
It is thus possible to it is used in the composition and the composition of actual user data of the data of train classification models
Unanimously, the accuracy of prediction is further improved.
Step S106, according to the features training disaggregated model of the user data of extraction, obtain grader.
The problem of whether being lost in prediction user, is converted to classification problem, i.e. user belongs to " loss " class still " no
It is lost in " class.It is thus possible to according to the data train classification models for being marked as being lost in or not being lost in, obtain and used for predicting
Family whether the grader that can be lost in.
Disaggregated model can for example use decision tree, logistic regression, SVMs, neutral net etc..
Step S108, predict whether user to be measured can be lost in using grader.
Classify by using to user data, then partial data is extracted respectively from each classification, be used to predict with training
User whether the disaggregated model that can be lost in, can be used in the data of train classification models can fully demonstrate original user data
The characteristics of, so as to more accurately predict whether user can be lost in.
Preferably, random forests algorithm can also be used in step s 106, and training process is mainly:
1. some groups of user data are randomly choosed from the user data of extraction.
Wherein, each group when selecting user data be one by one, have the selection put back to.I.e. for some group, in user
One is randomly choosed in data and adds the group, then randomly chooses a data in user data again, the data of selection can
With the data including having added the group, some data are selected to add the group using the above method.
2. decision-tree model is respectively trained using some groups of user data of selection.
During decision tree is trained, some features are selected from all features of user data first, then from selection
Feature determine segmentation feature, to decision tree carry out branch.
3. some decision trees that some groups of user data using selection are trained to obtain are collectively as grader.
, can be respectively using each decision tree to testing data when being predicted using grader caused by the above method
Classification prediction is carried out, using the classification results of majority decision tree as final classification knot.In addition it is also possible to mark the classification of each tree
End value, for example, loss is designated as into 1, is not lost in and is designated as -1, then using the average value of the classification results value of each decision tree as
Prediction result represents the possibility of customer loss, or compared with default threshold value, judges classification results.
In order to comprehensively describe the feature of user, user data often has more feature, so that training process
Amount of calculation is very big.And by using the above method, the data with big measure feature can be handled.Also, due to every decision tree
Used training dataset is different, therefore the grader that can avoid obtaining produces the situation of over-fitting, so as to be lifted
The accuracy of attrition prediction.
In train classification models and before obtaining grader, user data can also be pre-processed.For example, can be right
The feature of user data is screened, and only retains the feature related to customer loss;The value of the feature of classifying type can also be turned
It is changed to readable, easy-to-handle data, such as this feature of mobile phone model, can be by each manufacturer, each model
Mobile phone uses Digital ID.
Further, it is also possible to the feature of user data is optimized.Another embodiment of the present invention is described below with reference to Fig. 2
Customer loss Forecasting Methodology.
Fig. 2 is the flow chart of another embodiment of customer loss Forecasting Methodology of the present invention.As shown in Fig. 2 the embodiment
Method include:
Step S202, classifies to user data, and user data is marked as being lost in user and is not lost in user.
Step S204, portion of user data is extracted respectively from each classification of user data.
Step S202~S204 embodiment is referred to step S102~S104.
Step S2052~S2054 can be used to carry out conversion process to the feature of the user data of extraction.
Step S2052, count the species of all values of each feature in the user data of extraction.
For example, for " whether opening online " this feature, only comprising two kinds of values for representing "Yes" and "No".And for
" mobile phone model ", due to there is numerous cell phone manufacturers at present, each manufacturer has some product types again, therefore " mobile phone model "
Value species may be a lot.
Therefore, if do not handled feature, larger amount of calculation may be produced in the training process.
Step S2054, if the species of all values of feature is more than preset value, using each value of feature as
The new feature of the user data of extraction, and the species for deleting all values is more than the feature of preset value.
For example, the value species for setting " mobile phone model " is more than preset value, value includes A models, Type B number, c-type number, D models
Etc., new feature that can be by " A models ", " Type B number ", " c-type number ", " D models " etc. respectively as user data, each feature
Value include representing two kinds of values of "Yes" and "No", and delete original " mobile phone model " feature.
Further, it is also possible to feature selecting is carried out to user data using step S2056.
Step S2056, from the user data of extraction, using lasso trick (LASSO, Least Absolute Shrinkage
And Selection Operator) algorithms selection feature.
For example, R language provides glmnet bags, lasso trick algorithm can be implemented using glmnet bags.In lasso trick algorithm
In output, each feature has corresponding weights, i.e. λ value.Can by weights for 0 or weights be more than threshold value feature make
For the feature of selection.
It is thus possible to be screened to feature, reducing multiple features may disturb to caused by disaggregated model.
Step S206, according to the features training disaggregated model of the user data of the extraction after processing, obtain grader.
Step S208, predict whether user to be measured can be lost in using grader.
By using the above method, the amount of calculation in training process can be reduced, mitigates computational load.
The customer loss prediction meanss of one embodiment of the invention are described below with reference to Fig. 3.
Fig. 3 is the structure chart of one embodiment of customer loss prediction meanss of the present invention.As shown in figure 3, the embodiment
Device includes:User data sort module 31, for classifying to user data, user data is marked as being lost in user
User is not lost in;User data abstraction module 32, for extracting part number of users respectively from each classification of user data
According to;Model training module 33, for the features training disaggregated model of the user data according to extraction, obtain grader;User flows
Prediction module 34 is lost, for predicting whether user to be measured can be lost in using grader.
Wherein, user data sort module 31 can be further used for according to spy related to customer loss in user data
Sign, is classified using the method for cluster to user data.
Wherein, user data abstraction module 32 can be further used for from each classification of user data extracting respectively in advance
If the user data of ratio.
The customer loss prediction meanss of another embodiment of the present invention are described below with reference to Fig. 4.
Fig. 4 is the structure chart of another embodiment of customer loss prediction meanss of the present invention.As shown in figure 4, the embodiment
Model training module 33 can include:Grouped data extracting unit 431, if for being randomly choosed from the user data of extraction
Dry group user data;Decision tree training unit 432, for decision tree mould to be respectively trained using some groups of user data of selection
Type;Grader forms unit 433, and some decision trees for training to obtain by some groups of user data using selection make jointly
For grader.
In addition, model training module 33 can also include:Feature value species statistic unit 434, for counting what is extracted
The species of all values of each feature in user data;Feature Conversion unit 435, the species for all values when feature
During more than preset value, using each value of feature as the new feature of the user data of extraction, and all take is deleted
The species of value is more than the feature of preset value;First model training unit 436, for the user data according to the extraction after processing
Features training disaggregated model.
In addition, model training module 33 can also include:Feature selection unit 437, for from the user data of extraction,
Using lasso trick algorithms selection feature;Second model training unit 438, the feature instruction for the selection of the user data according to extraction
Practice disaggregated model.
In addition, device can also include user data acquisition module 45, user data acquisition module 45 includes:User data
Acquiring unit 451, for obtaining the user data with some features in very first time unit;User data indexing unit
452, the state whether being lost in the second time quantum for the user data in very first time unit, mark the very first time
User data in unit.Wherein, very first time unit is the adjacent first time quantum of the second time quantum.
In addition, the method according to the invention is also implemented as a kind of computer program product, the computer program product
Including computer-readable medium, be stored with the computer-readable medium for perform the present invention method in limit it is above-mentioned
The computer program of function.Those skilled in the art will also understand is that, various exemplary with reference to described by disclosure herein
Logical block, module, circuit and algorithm steps may be implemented as the combination of electronic hardware, computer software or both.
The foregoing is only presently preferred embodiments of the present invention, be not intended to limit the invention, it is all the present invention spirit and
Within principle, any modification, equivalent substitution and improvements made etc., it should be included in the scope of the protection.
Claims (14)
- A kind of 1. customer loss Forecasting Methodology, it is characterised in that including:User data is classified, the user data is marked as being lost in user and is not lost in user;Portion of user data is extracted respectively from each classification of user data;According to the features training disaggregated model of the user data of extraction, grader is obtained;Predict whether user to be measured can be lost in using the grader.
- 2. according to the method for claim 1, it is characterised in that described classification is carried out to user data to include:According to feature related to customer loss in user data, user data is classified using the method for cluster.
- 3. according to the method for claim 1, it is characterised in that part is extracted in each classification from user data and is used User data includes:Extract the user data of preset ratio respectively from each classification of user data.
- 4. according to the method for claim 1, it is characterised in that the features training classification of the user data according to extraction Model, obtaining grader includes:Some groups of user data are randomly choosed from the user data of the extraction;Decision-tree model is respectively trained using some groups of user data of selection;Some decision trees that some groups of user data using selection are trained to obtain are collectively as grader.
- 5. according to the method for claim 1, it is characterised in that the features training classification of the user data according to extraction Model includes:Count the species of all values of each feature in the user data extracted;If the species of all values of feature is more than preset value, using each value of the feature as the user of extraction The new feature of data, and the species for deleting all values is more than the feature of preset value;According to the features training disaggregated model of the user data of the extraction after processing.
- 6. method according to claim 1 or 5, it is characterised in that the features training of the user data according to extraction Disaggregated model includes:From the user data of the extraction, using lasso trick algorithms selection feature;According to the features training disaggregated model of the selection of the user data of extraction.
- 7. according to the method for claim 1, it is characterised in that the user data of mark is obtained using following methods:Obtain the user data with some features in very first time unit;The state whether user data in the very first time unit is lost in the second time quantum, mark described first User data in time quantum;Wherein, very first time unit is the adjacent first time quantum of the second time quantum.
- A kind of 8. customer loss prediction meanss, it is characterised in that including:User data sort module, for classifying to user data, the user data be marked as being lost in user and User is not lost in;User data abstraction module, for extracting portion of user data respectively from each classification of user data;Model training module, for the features training disaggregated model of the user data according to extraction, obtain grader;Customer loss prediction module, for predicting whether user to be measured can be lost in using the grader.
- 9. device according to claim 8, it is characterised in that the user data sort module be further used for according to The feature related to customer loss in user data, is classified using the method for cluster to user data.
- 10. device according to claim 8, it is characterised in that the user data abstraction module be further used for from The user data of preset ratio is extracted in each classification of user data respectively.
- 11. device according to claim 8, it is characterised in that the model training module includes:Grouped data extracting unit, for randomly choosing some groups of user data from the user data of the extraction;Decision tree training unit, for decision-tree model to be respectively trained using some groups of user data of selection;Grader forms unit, and some decision trees for training to obtain by some groups of user data using selection are common As grader.
- 12. device according to claim 8, it is characterised in that the model training module includes:Feature value species statistic unit, for counting the species of all values of each feature in the user data extracted;Feature Conversion unit, when the species for all values when feature is more than preset value, by each value of the feature Respectively as the new feature of the user data of extraction, and the species for deleting all values is more than the feature of preset value;First model training unit, the features training disaggregated model for the user data according to the extraction after processing.
- 13. the device according to claim 8 or 12, it is characterised in that the model training module includes:Feature selection unit, for from the user data of the extraction, using lasso trick algorithms selection feature;Second model training unit, the features training disaggregated model for the selection of the user data according to extraction.
- 14. device according to claim 8, it is characterised in that also including user data acquisition module, the user data Acquisition module includes:User data acquiring unit, for obtaining the user data with some features in very first time unit;User data indexing unit, whether flowed in the second time quantum for the user data in the very first time unit The state of mistake, mark the user data in the very first time unit;Wherein, very first time unit is the adjacent first time quantum of the second time quantum.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610818511.3A CN107818376A (en) | 2016-09-13 | 2016-09-13 | Customer loss Forecasting Methodology and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610818511.3A CN107818376A (en) | 2016-09-13 | 2016-09-13 | Customer loss Forecasting Methodology and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107818376A true CN107818376A (en) | 2018-03-20 |
Family
ID=61600667
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610818511.3A Pending CN107818376A (en) | 2016-09-13 | 2016-09-13 | Customer loss Forecasting Methodology and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107818376A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190796A (en) * | 2018-08-02 | 2019-01-11 | 北京天元创新科技有限公司 | A kind of telecom client attrition prediction method, system and electronic equipment |
CN109377058A (en) * | 2018-10-26 | 2019-02-22 | 中电科新型智慧城市研究院有限公司 | The enterprise of logic-based regression model moves outside methods of risk assessment |
CN110070392A (en) * | 2019-04-17 | 2019-07-30 | 北大方正集团有限公司 | Customer churn method for early warning and device |
CN111401591A (en) * | 2018-12-17 | 2020-07-10 | ***通信集团北京有限公司 | Quality difference user determination method and device and readable medium |
CN111833086A (en) * | 2019-04-22 | 2020-10-27 | 百度在线网络技术(北京)有限公司 | Account classification model training method and device and account classification method and device |
CN112990425A (en) * | 2019-12-18 | 2021-06-18 | ***通信集团浙江有限公司 | Automatic classification method of 5G network slices, device thereof, electronic equipment and computer storage medium |
CN117575675A (en) * | 2023-11-17 | 2024-02-20 | 中电智恒信息科技服务有限公司 | Telecommunication user loss prediction method, device, equipment and medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1795462A (en) * | 2003-05-22 | 2006-06-28 | 珀欣投资有限责任公司 | Customer revenue prediction method and system |
US20070156673A1 (en) * | 2005-12-30 | 2007-07-05 | Accenture S.P.A. | Churn prediction and management system |
CN102955894A (en) * | 2012-05-24 | 2013-03-06 | 华东师范大学 | Customer segmentation-based method for controlling churn rate prediction |
CN102982077A (en) * | 2012-10-30 | 2013-03-20 | 中国联合网络通信集团有限公司 | User data processing method and device |
CN105069534A (en) * | 2015-08-18 | 2015-11-18 | 广州华多网络科技有限公司 | Customer loss prediction method and device |
-
2016
- 2016-09-13 CN CN201610818511.3A patent/CN107818376A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1795462A (en) * | 2003-05-22 | 2006-06-28 | 珀欣投资有限责任公司 | Customer revenue prediction method and system |
US20070156673A1 (en) * | 2005-12-30 | 2007-07-05 | Accenture S.P.A. | Churn prediction and management system |
CN102955894A (en) * | 2012-05-24 | 2013-03-06 | 华东师范大学 | Customer segmentation-based method for controlling churn rate prediction |
CN102982077A (en) * | 2012-10-30 | 2013-03-20 | 中国联合网络通信集团有限公司 | User data processing method and device |
CN105069534A (en) * | 2015-08-18 | 2015-11-18 | 广州华多网络科技有限公司 | Customer loss prediction method and device |
Non-Patent Citations (1)
Title |
---|
孙良君 等: "基于Group Lasso的多源电信数据离网用户分析", 《南京师范大学学报(工程技术版)》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109190796A (en) * | 2018-08-02 | 2019-01-11 | 北京天元创新科技有限公司 | A kind of telecom client attrition prediction method, system and electronic equipment |
CN109190796B (en) * | 2018-08-02 | 2021-03-02 | 北京天元创新科技有限公司 | Telecommunication customer loss prediction method, system and electronic equipment |
CN109377058A (en) * | 2018-10-26 | 2019-02-22 | 中电科新型智慧城市研究院有限公司 | The enterprise of logic-based regression model moves outside methods of risk assessment |
CN111401591A (en) * | 2018-12-17 | 2020-07-10 | ***通信集团北京有限公司 | Quality difference user determination method and device and readable medium |
CN110070392A (en) * | 2019-04-17 | 2019-07-30 | 北大方正集团有限公司 | Customer churn method for early warning and device |
CN110070392B (en) * | 2019-04-17 | 2022-02-22 | 北大方正集团有限公司 | User loss early warning method and device |
CN111833086A (en) * | 2019-04-22 | 2020-10-27 | 百度在线网络技术(北京)有限公司 | Account classification model training method and device and account classification method and device |
CN111833086B (en) * | 2019-04-22 | 2024-04-23 | 百度在线网络技术(北京)有限公司 | Account classification model training method and device and account classification method and device |
CN112990425A (en) * | 2019-12-18 | 2021-06-18 | ***通信集团浙江有限公司 | Automatic classification method of 5G network slices, device thereof, electronic equipment and computer storage medium |
CN117575675A (en) * | 2023-11-17 | 2024-02-20 | 中电智恒信息科技服务有限公司 | Telecommunication user loss prediction method, device, equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107818376A (en) | Customer loss Forecasting Methodology and device | |
WO2020143233A1 (en) | Method and device for building scorecard model, computer apparatus and storage medium | |
CN103064987B (en) | A kind of wash sale information identifying method | |
CN109857862B (en) | Text classification method, device, server and medium based on intelligent decision | |
CN108833139B (en) | OSSEC alarm data aggregation method based on category attribute division | |
CN106095939B (en) | The acquisition methods and device of account authority | |
CN109543925A (en) | Risk Forecast Method, device, computer equipment and storage medium based on machine learning | |
CN109582849A (en) | A kind of Internet resources intelligent search method of knowledge based map | |
WO2015039478A1 (en) | Method and apparatus for recognizing junk messages | |
CN109919436A (en) | A kind of promise breaking user's probability forecasting method based on sparse features insertion | |
CN107368856A (en) | Clustering method and device, the computer installation and readable storage medium storing program for executing of Malware | |
CN107438050A (en) | Identify the method and system of the potential malicious user of website | |
CN109359137A (en) | Based on user's growth of Feature Selection and semi-supervised learning portrait construction method | |
CN113409555B (en) | Real-time alarm linkage method and system based on Internet of things | |
CN112750030A (en) | Risk pattern recognition method, risk pattern recognition device, risk pattern recognition equipment and computer readable storage medium | |
CN112084761A (en) | Hydraulic engineering information management method and device | |
CN104346411B (en) | The method and apparatus that multiple contributions are clustered | |
CN113255263B (en) | Particle band dividing method, device, computer equipment and storage medium | |
CN113852204A (en) | Three-dimensional panoramic monitoring system and method for transformer substation with digital twin | |
CN116909534B (en) | Operator flow generating method, operator flow generating device and storage medium | |
CN107291922B (en) | A kind of data distributing method and device | |
US11227288B1 (en) | Systems and methods for integration of disparate data feeds for unified data monitoring | |
CN116842174A (en) | Agricultural resource database platform building method based on network data | |
WO2024103436A1 (en) | Device defect data rule base construction method and device defect correlation analysis method | |
CN103778210B (en) | Method and device for judging specific file type of file to be analyzed |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180320 |
|
RJ01 | Rejection of invention patent application after publication |