CN107609700A - A kind of customer value model optimization method based on machine learning - Google Patents
A kind of customer value model optimization method based on machine learning Download PDFInfo
- Publication number
- CN107609700A CN107609700A CN201710807555.0A CN201710807555A CN107609700A CN 107609700 A CN107609700 A CN 107609700A CN 201710807555 A CN201710807555 A CN 201710807555A CN 107609700 A CN107609700 A CN 107609700A
- Authority
- CN
- China
- Prior art keywords
- learner
- data
- customer value
- sample
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Complex Calculations (AREA)
Abstract
The present invention relates to a kind of customer value model optimization method based on machine learning, including the steps:Step 1:The customer value model data of N number of client's main body different times is extracted by stochastical sampling method, obtains initial model data sample Si (i=1,2,3...N);Step 2:To using bagging machine learning methods respectively to individual initial model data sample Si (i=1,2,3...n), N number of independent individual weak learner Hi (i=1,2,3...N) is accordingly trained;Step 3:Described individual weak learner Hi (i=1,2,3...N) is combined into by learner H one strong by stacking combinations strategy;Step 4:Using strong learner H as optimal models rule, and existing customer value models data sample is input to strong learner H, the result that strong learner H is drawn is optimal result model.
Description
Technical field
The present invention relates to a kind of processing method of transaction data, more particularly to a kind of customer value mould based on machine learning
Type optimization method.
Background technology
At present, traditional model optimization mode, verified using Experimental comparison.For target identification class model, root
According to needing Optimized model application scenarios, comparative selection data, a part of target data and other interference data are included in data.Will
Test data imports model running, checks the identification quantity of target data in model output result, carries out modelling effect judgement.Mould
Type effect is judged mainly by the way that the recall ratio and precision ratio of target data, two indices are weighed:
Recall ratio, refer in model calculation result, comprising target data number of samples, account for target data sample in detection data
This percentage.
Precision ratio, refer in model calculation result, comprising target data number of samples, account for whole Model Identification number of samples
Percentage.
Class model is predicted for index, it is same to select historical data to import model, according to model calculation result and actual number
According to being compared, calculation error scope, if error range meets model accuracy design requirement, model need not optimize;If by mistake
Poor scope then needs to carry out model optimization more than model accuracy requirement.
The optimization process of the same model of sector application at present, it is consistent, it is necessary to re-start substantially with the newly-built process of model
Mode input data are associated analysis, import new data field and replace legacy data information.Then it is root in terms of model algorithm
According to optimization at that time, overall social base algorithm research present situation, more preferable algorithm is selected to substitute original algorithm.
Passing through the above-mentioned explanation optimized to current business models, it can be seen that the mode of existing model optimization is more traditional,
Labor intensive, time cost are higher, less efficient.Existing sector application model optimization simultaneously, it is necessary under experimental conditions could
Complete, real-time optimization can not be carried out under real running environment automatically, delay practical, commercial, if model application is some
The core mechanism of enterprise, model optimization process, also larger interests can be brought to lose to enterprise.Therefore, majority is also actually caused
Enterprise, it is reluctant to spend so high cost to carry out model optimization, still continues to use old model, equally also have impact on the actual effect of model
Fruit.
The content of the invention
In order to solve the above technical problems, it is an object of the invention to provide a kind of customer value model based on machine learning is excellent
Change method, the customer value model optimization method can reduce manpower, time cost, improve data-optimized efficiency, while also protect
The effect of model of a syndrome application, improves utilization benefit.
A kind of customer value model optimization method based on machine learning of the present invention, its feature are that this method includes
The steps:
Step 1:The customer value model data of N number of client's main body different times is extracted by stochastical sampling method, is obtained just
Beginning model data sample Si (i=1,2,3...N);
Step 2:Bagging machine learning sides are used respectively to each initial model data sample Si (i=1,2,3...n)
Method, accordingly train N number of independent individual weak learner Hi (i=1,2,3...N);
Step 3:Strategy is combined by the individual weak learner Hi (i=1,2,3...N) described in step 2 by stacking
It is combined into learner H one strong;
Step 4:Using the strong learner H that step 3 obtains as optimal models rule, and by existing customer value models data
Sample is input to strong learner H, and the result that strong learner H is drawn is optimal result model.
Further, the stochastical sampling method in step 1 is self-service sampling method (Bootstap sampling), i.e., for N number of
The original training set of sample, each first one sample of random acquisition are put into sampling set, then the sample are put back to, so gathers N
It is secondary, untill obtaining the sampling set of N number of sample.
Further, the stacking in step 3 includes the steps with reference to strategy:
First concentrated from customer value model data and randomly select 45%-55% data samples as training set, while from visitor
20%-30% data samples are randomly selected in the value models data set of family as test set;
One secondary learner of retraining, during secondary learner is trained by each weak learner Hi (i=1,
2nd, 3...N) input of the learning outcome as secondary learner, the output using the result of training set as secondary learner;
Finally test set is predicted once with primary learner, obtains the input sample of secondary learner, then learned with secondary
Practise device and forecast sample is once obtained to test set prediction, while the data correlation between input sample and forecast sample is matched and closed
The continuous training of system, best model input and the procedure parameter span being optimal under output result are strong so as to obtain
Learner H.
Further, described data correlation matching relationship includes customer value mode input data, procedure parameter and defeated
The association matching relationship gone out between result three, described procedure parameter be customer value model data in each index weight or
Person divides the span of client's classification index, and described output result is regular for the value label or customer segmentation of client.
Further, described customer value model data includes data field, index weights, the model in index system
Algorithm and model result.
Further, concentrated from customer value model data and randomly select 50% data sample as training set, while from
Customer value model data is concentrated and randomly selects 25% data sample as test set.
By such scheme, the present invention at least has advantages below:The present invention constantly uses according to user, in combination with
Different user, for same industry application scenarios, the data mining model of the differentiation of use so that sector application model possesses
Automatic study, the ability of real-time optimization, i.e., complete from model construction, from practical application that time, is just constantly learning automatically,
Automatic Optimal, including mode input data and model algorithm, it is ensured that model all in optimum state, has evaded conventional model at any time
Optimize the manpower brought, time, interests loss, while also ensure model application effect, being connected in client brings huge income.
Described above is only the general introduction of technical solution of the present invention, in order to better understand the technological means of the present invention,
And can be practiced according to the content of specification, below with presently preferred embodiments of the present invention and coordinate accompanying drawing describe in detail as after.
Brief description of the drawings
Fig. 1 is the workflow diagram of the present invention.
Embodiment
With reference to the accompanying drawings and examples, the embodiment of the present invention is described in further detail.Implement below
Example is used to illustrate the present invention, but is not limited to the scope of the present invention.
Referring to a kind of customer value model optimization side based on machine learning described in Fig. 1 a preferred embodiment of the present invention
Method, including the steps:
Step 1:The customer value model data of N number of client's main body different times is extracted by stochastical sampling method, is obtained just
Beginning model data sample Si (i=1,2,3...N);
Step 2:Bagging machine learning sides are used respectively to each initial model data sample Si (i=1,2,3...n)
Method, accordingly train N number of independent individual weak learner Hi (i=1,2,3...N);
Step 3:Strategy is combined by the individual weak learner Hi (i=1,2,3...N) described in step 2 by stacking
It is combined into learner H one strong;
Step 4:Using the strong learner H that step 3 obtains as optimal models rule, and by existing customer value models data
Sample is input to strong learner H, and the result that strong learner H is drawn is optimal result model.
As a further improvement on the present invention, the stochastical sampling method in step 1 is self-service sampling method (Bootstap
Sampling), i.e., for the original training set of N number of sample, each first one sample of random acquisition is put into sampling set, then this
Sample is put back to, and so gathers n times, untill obtaining the sampling set of N number of sample.
As a further improvement on the present invention, the stacking in step 3 includes the steps with reference to strategy:
First concentrated from customer value model data and randomly select 45%-55% data samples as training set, while from visitor
20%-30% data samples are randomly selected in the value models data set of family as test set;
One secondary learner of retraining, during secondary learner is trained by each weak learner Hi (i=1,
2nd, 3...N) input of the learning outcome as secondary learner, the output using the result of training set as secondary learner;
Finally test set is predicted once with primary learner, obtains the input sample of secondary learner, then learned with secondary
Practise device and forecast sample is once obtained to test set prediction, while the data correlation between input sample and forecast sample is matched and closed
The continuous training of system, best model input and the procedure parameter span being optimal under output result are strong so as to obtain
Learner H.
As a further improvement on the present invention, described data correlation matching relationship includes customer value mode input number
According to the association matching relationship between, procedure parameter and output result three, described procedure parameter is customer value model data
In each index weight or divide client's classification index span, described output result for client value label or
Customer segmentation rule.
As a further improvement on the present invention, described customer value model data includes the data word in index system
Section, index weights, model algorithm and model result.
As a further improvement on the present invention, concentrated from customer value model data and randomly select 50% data sample conduct
Training set, while concentrated from customer value model data and randomly select 25% data sample as test set.
Described above is only the preferred embodiment of the present invention, is not intended to limit the invention, it is noted that for this skill
For the those of ordinary skill in art field, without departing from the technical principles of the invention, can also make it is some improvement and
Modification, these improvement and modification also should be regarded as protection scope of the present invention.
Claims (6)
- A kind of 1. customer value model optimization method based on machine learning, it is characterised in that including the steps:Step 1:The customer value model data of N number of client's main body different times is extracted by stochastical sampling method, obtains introductory die Type data sample Si (i=1,2,3...N);Step 2:Bagging machine learning methods are used respectively to each initial model data sample Si (i=1,2,3...n), Accordingly train N number of independent individual weak learner Hi (i=1,2,3...N);Step 3:Strategy is combined by stacking to combine the individual weak learner Hi (i=1,2,3...N) described in step 2 Into learner H one strong;Step 4:Using the strong learner H that step 3 obtains as optimal models rule, and by existing customer value models data sample Strong learner H is input to, the result that strong learner H is drawn is optimal result model.
- 2. the customer value model optimization method according to claim 1 based on integrated study Bagging algorithms, its feature It is:Stochastical sampling method in step 1 is self-service sampling method (Bootstap sampling), the i.e. original instruction for N number of sample Practice collection, each first one sample of random acquisition is put into sampling set, then the sample is put back to, so gathers n times, until obtaining N Untill the sampling set of individual sample.
- 3. the customer value model optimization method according to claim 1 based on integrated study Bagging algorithms, its feature It is:Stacking in step 3 includes the steps with reference to strategy:First concentrated from customer value model data and randomly select 45%-55% data samples as training set, while from client's valency Value model data is concentrated and randomly selects 20%-30% data samples as test set;One secondary learner of retraining, during secondary learner is trained by each weak learner Hi (i=1,2, 3...N input of the learning outcome) as secondary learner, the output using the result of training set as secondary learner;Finally test set is predicted once with primary learner, obtains the input sample of secondary learner, then with secondary learner Forecast sample is once obtained to test set prediction, while to the data correlation matching relationship between input sample and forecast sample Constantly training, best model input and the procedure parameter span being optimal under output result, so as to be learnt by force Device H.
- 4. the customer value model optimization method according to claim 3 based on integrated study Bagging algorithms, its feature It is:Described data correlation matching relationship include customer value mode input data, procedure parameter and output result three it Between association matching relationship, described procedure parameter be customer value model data in each index weight or division customer class The span of other index, described output result are regular for the value label or customer segmentation of client.
- 5. the customer value model optimization method according to claim 1 based on integrated study Bagging algorithms, its feature It is:Described customer value model data includes data field, index weights, model algorithm and the model knot in index system Fruit.
- 6. the customer value model optimization method according to claim 1 based on integrated study Bagging algorithms, its feature It is:Concentrated from customer value model data and randomly select 50% data sample as training set, while from customer value model 25% data sample is randomly selected in data set as test set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710807555.0A CN107609700A (en) | 2017-09-08 | 2017-09-08 | A kind of customer value model optimization method based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710807555.0A CN107609700A (en) | 2017-09-08 | 2017-09-08 | A kind of customer value model optimization method based on machine learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107609700A true CN107609700A (en) | 2018-01-19 |
Family
ID=61061966
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710807555.0A Pending CN107609700A (en) | 2017-09-08 | 2017-09-08 | A kind of customer value model optimization method based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107609700A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108459997A (en) * | 2018-02-07 | 2018-08-28 | 深圳市微埃智能科技有限公司 | High skewness data value probability forecasting method based on deep learning and neural network |
CN110405343A (en) * | 2019-08-15 | 2019-11-05 | 山东大学 | A kind of laser welding process parameter optimization method of the prediction model integrated based on Bagging and particle swarm optimization algorithm |
CN113095511A (en) * | 2021-04-16 | 2021-07-09 | 广东电网有限责任公司 | Method and device for judging in-place operation of automatic master station |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102289682A (en) * | 2011-05-18 | 2011-12-21 | 华北电力大学 | Transformer fault diagnosis method based on integrated learning Bagging algorithm |
US8200514B1 (en) * | 2006-02-17 | 2012-06-12 | Farecast, Inc. | Travel-related prediction system |
CN106934493A (en) * | 2017-02-28 | 2017-07-07 | 北京科技大学 | A kind of construction method of power customer appraisal Model |
-
2017
- 2017-09-08 CN CN201710807555.0A patent/CN107609700A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8200514B1 (en) * | 2006-02-17 | 2012-06-12 | Farecast, Inc. | Travel-related prediction system |
CN102289682A (en) * | 2011-05-18 | 2011-12-21 | 华北电力大学 | Transformer fault diagnosis method based on integrated learning Bagging algorithm |
CN106934493A (en) * | 2017-02-28 | 2017-07-07 | 北京科技大学 | A kind of construction method of power customer appraisal Model |
Non-Patent Citations (1)
Title |
---|
刘龙飞: "基于卷积神经网络的在线商品评论情感倾向性研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108459997A (en) * | 2018-02-07 | 2018-08-28 | 深圳市微埃智能科技有限公司 | High skewness data value probability forecasting method based on deep learning and neural network |
CN110405343A (en) * | 2019-08-15 | 2019-11-05 | 山东大学 | A kind of laser welding process parameter optimization method of the prediction model integrated based on Bagging and particle swarm optimization algorithm |
CN110405343B (en) * | 2019-08-15 | 2021-06-29 | 山东大学 | Laser welding process parameter optimization method based on Bagging integrated prediction model and particle swarm optimization algorithm |
CN113095511A (en) * | 2021-04-16 | 2021-07-09 | 广东电网有限责任公司 | Method and device for judging in-place operation of automatic master station |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103632168B (en) | Classifier integration method for machine learning | |
CN107766929B (en) | Model analysis method and device | |
CN103679132B (en) | A kind of nude picture detection method and system | |
CN110070067A (en) | The training method of video classification methods and its model, device and electronic equipment | |
CN108509976A (en) | The identification device and method of animal | |
CN106951825A (en) | A kind of quality of human face image assessment system and implementation method | |
CN109215028A (en) | A kind of multiple-objection optimization image quality measure method based on convolutional neural networks | |
CN104182474A (en) | Method for recognizing pre-churn users | |
CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
CN110135231A (en) | Animal face recognition methods, device, computer equipment and storage medium | |
CN103218832B (en) | Based on the vision significance algorithm of global color contrast and spatial distribution in image | |
CN104966105A (en) | Robust machine error retrieving method and system | |
CN105719045A (en) | Retention risk determiner | |
CN107609700A (en) | A kind of customer value model optimization method based on machine learning | |
CN108960264A (en) | The training method and device of disaggregated model | |
CN105654196A (en) | Adaptive load prediction selection method based on electric power big data | |
CN104850868A (en) | Customer segmentation method based on k-means and neural network cluster | |
CN116821698B (en) | Wheat scab spore detection method based on semi-supervised learning | |
CN104616005A (en) | Domain-self-adaptive facial expression analysis method | |
CN103886030A (en) | Cost-sensitive decision-making tree based physical information fusion system data classification method | |
CN103279944A (en) | Image division method based on biogeography optimization | |
Gawade et al. | Early-stage apple leaf disease prediction using deep learning | |
CN111144462A (en) | Unknown individual identification method and device for radar signals | |
CN105320720B (en) | Dependency rule analytical equipment and dependency rule analysis method | |
CN109583712A (en) | A kind of data target analysis method and device, storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180119 |