CN107730286A - A kind of target customer's screening technique and device - Google Patents
A kind of target customer's screening technique and device Download PDFInfo
- Publication number
- CN107730286A CN107730286A CN201610652832.0A CN201610652832A CN107730286A CN 107730286 A CN107730286 A CN 107730286A CN 201610652832 A CN201610652832 A CN 201610652832A CN 107730286 A CN107730286 A CN 107730286A
- Authority
- CN
- China
- Prior art keywords
- sample
- target customer
- screening
- client
- probability
- 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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0202—Market predictions or forecasting for commercial activities
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0207—Discounts or incentives, e.g. coupons or rebates
- G06Q30/0224—Discounts or incentives, e.g. coupons or rebates based on user history
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a kind of target customer's screening technique and device, target customer's screening technique includes:Pre- this target customer for carrying out pushing presupposed information is filtered out in client's sample to be sorted;Obtain and the push feedback result after presupposed information push is carried out to this described target customer;The screening model for screening this target customer is modified according to the push feedback result, to carry out the screening of next target customer according to revised screening model.Such scheme, by using marketing feedback result dynamic adjusting data mining model, improve the accuracy that marketing client screens, substantially increase the hit rate and precision of marketing, marketing effectiveness is improved, considerably reduces cost of marketing, improves marketing input-output ratio and marketing income.
Description
Technical field
The present invention relates to data warehouse technology field, more particularly to a kind of target customer's screening technique and device.
Background technology
In the case where market competition is growing more intense, how to improve target customers selection precision, improve marketing into
Power, reduction cost of marketing are each electric business operator concerns always.At this stage, generally be directed to the marketing of specialized field
Activity, establish data mining model, and the mode such as matching characteristic and user preference screens customers.Data mining model is usually
Built based on historical data analysis, find out the screening rule of relative users group, then obtain the client that markets using corresponding rule
Group, the process of structure do not use newest marketing feedback result.
Prior art is primarily present following 2 defects:
1st, the hysteresis quality of data mining model adjustment.
The rule that data mining filters out tends to rely on output-index, the completeness of dimension, and data mining team
Experience.Model training is fixed up model result after checking, go out potential user group by corresponding Rules Filtering.
And the input pointer of model, the weight of index keep immobilizing, can not with the change of business adjust automatically.Until mould
Type hydraulic performance decline to a certain extent, causes the attention of marketing personnel, and then puts into a large amount of human and material resources and time to data digging
Dig model and carry out double optimization.
2nd, marketing feedback can not be fully used.
Traditional marketing feedback information is used merely to evaluate the effect of marketing activity, or the quality for assessment models,
There is no making full use of for actual feedback information, be not carried out marketing closed loop truly.
With the change of market environment, user behavior can also change.Original static models rule or client's mark
Label, can not increasingly adapt to the change in market, customer action, cause that customers' screening is not accurate, and marketing effectiveness is also increasingly
Difference.
The content of the invention
It is existing to solve the technical problem to be solved in the present invention is to provide a kind of target customer's screening technique and device
Static data mining model can not combine marketing feedback result dynamic and adjust, and cause marketing client to screen inaccurate, marketing effectiveness
The problem of poor.
In order to solve the above-mentioned technical problem, the embodiment of the present invention provides a kind of target customer's screening technique, including:
Pre- this target customer for carrying out pushing presupposed information is filtered out in client's sample to be sorted;
Obtain and the push feedback result after presupposed information push is carried out to this described target customer;
The screening model for screening this target customer is modified according to the push feedback result, after according to amendment
Screening model carry out the screening of next target customer.
Further, it is described that pre- this target customer for carrying out pushing presupposed information is filtered out in client's sample to be sorted
The step of include:
Obtain screening model;
According to the screening model, it is general in each classification that each client's sample in client's sample to be sorted is calculated respectively
Rate value;
According to the probable value of each client's sample, the sheet for pushing presupposed information is selected in client's sample to be sorted
Secondary target customer.
Further, when the screening model is Naive Bayes Classifier, described the step of obtaining screening model, includes:
Obtain the training data and test data for building Naive Bayes Classifier;
Build to obtain initial grader using the training data;
The grader is selected using the test data, obtains Naive Bayes Classifier.
Further, described the step of being built using the training data and obtained initial grader, includes:
The prior probability and conditional probability of initial grader are obtained using training data;
Classification according to belonging to prior probability and conditional probability obtain each sample in training data, test data, and count
Training error, test error are calculated, chooses the minimum grader of test error as initial grader.
Further, it is described initial grader is obtained using training data prior probability and conditional probability the step of wrap
Include:
Utilize formula:Prior probability of the training data in each classification is calculated;
Utilize formula:The condition of training data is calculated
Probability;
Wherein, training data T={ (x1,y1),(x2,y2),...,(xN,yN), N represents to include sample in the training data
This x total number;P (Y=ck) represent the prior probability on classification Y;yiFor Y target variable, sample class, and y are representedi
∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability function;P(X(j)=ajl| Y=ck) table
Show the conditional probability of training data;J-th of feature of i-th of sample is represented, andajlIt is jth
Individual feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k=1,2 ..., K.
Further, it is described that each sample institute in training data, test data is obtained according to prior probability and conditional probability
The classification of category, and training error, test error are calculated, choose step of the minimum grader of test error as initial grader
Suddenly include:
Utilize formula:By each sample in training data
Classification of the classification corresponding to maximum tried to achieve in pre-set categories as the sample;Wherein,
Y represents the sample class with maximum;P (Y=ck) represent the prior probability on classification Y;Y represents sample class
Not;ckFor sample class value;P(X(j)=x(j)| Y=ck) represent conditional probability, x(j)For x target variable, x represents that sample is real
Example, and x=(x(1),x(2),...,x(n))T。
Further, it is described that the grader is selected using the test data, obtain Naive Bayes Classification
The step of device, includes:
The concentration of target variable and the parameter of Bayesian Estimation in multiple adjusting training data, according to the mesh after each adjustment
The concentration of variable and the parameter of Bayesian Estimation are marked, evaluation index value is calculated in test data, obtains an assessment of maximum
The parameter of object event concentration and Bayesian Estimation corresponding to desired value, obtains Naive Bayes Classifier;
Wherein, the prior probability of the Naive Bayes Classifier is:
Conditional probability is:
Wherein, Pλ(Y=ck) be Naive Bayes Classifier prior probability;yiFor Y target variable, sample class is represented
Not, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;λ is Bayesian Estimation
Parameter;Pλ(X(j)=ajl| Y=ck) be Naive Bayes Classifier conditional probability;Represent j-th of spy of i-th of sample
Sign, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k
=1,2 ..., K.
Further, the evaluation index value utilizes formula:It is calculated;
Wherein, F1 is evaluation index value, and P actual is positive example, represents the precision ratio of initial grader in test data,
And P=is predicted as the number of samples of the number of samples of positive example/be predicted as positive example;R is initial grader in test data
Recall ratio, and R=is predicted as the number of samples of number of samples/actual positive example of positive example.
Further, the probable value of each client's sample of the basis, selects in client's sample to be sorted and pushes away
The step of this target customer for sending presupposed information, includes:
Calculate client's sample to be sorted and belong to different classes of probable value, it is general more than non-target class to choose target class probable value
The sample of rate value, and the descending descending of probable value is arranged, choose the client's sample conduct for being arranged in predetermined number above
This target customer.
Further, it is described that the screening model for screening this target customer is modified according to the push feedback result
The step of include:
By the push feedback result compared with the class label of this target customer, obtained in this target customer
Push feedback result is taken to push failed first sample;
Screening model is adjusted according to the first sample.
Further, when the screening model is Naive Bayes Classifier, it is described according to the first sample to sieve
The step of modeling type is adjusted includes:
According to formula:Recalculate the elder generation of Naive Bayes Classifier
Test probability;
According to formula:Again
Calculate the conditional probability of Naive Bayes Classifier;
Wherein, P1(Y=ck) represent Naive Bayes Classifier prior probability;yiFor Y target variable, sample is represented
Classification, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;N represents training data
In include sample x total number;P1(X(j)=ajl| Y=ck) represent Naive Bayes Classifier conditional probability;Represent the
J-th of feature of i sample, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ...,
N, l=1,2 ..., Sj, k=1,2 ..., K;N1For the number of first sample.
The embodiment of the present invention provides a kind of target customer's screening plant, including:
Screening module, for filtering out pre- this target visitor for carrying out pushing presupposed information in client's sample to be sorted
Family;
Feedback result acquisition module, it is anti-to the push after this described target customer progress presupposed information push for obtaining
Present result;
Correcting module, for being repaiied according to the push feedback result to the screening model for screening this target customer
Just, with the screening according to the next target customer of revised screening model progress.
Further, the screening module includes:
Model acquisition submodule, for obtaining screening model;
Probability calculation submodule, for according to the screening model, calculating each client in client's sample to be sorted respectively
Probable value of the sample in each classification;
Submodule is chosen, for the probable value according to each client's sample, is selected in client's sample to be sorted
Push this target customer of presupposed information.
Further, the correcting module includes:
Feedback result screens submodule, for the push feedback result and the class label of this target customer to be carried out
Compare, push feedback result is obtained in this target customer to push failed first sample;
Submodule is adjusted, for being adjusted according to the first sample to screening model.
The beneficial effects of the invention are as follows:
Such scheme, by using marketing feedback result dynamic adjusting data mining model, improve marketing client's screening
Accuracy, substantially increase the hit rate and precision of marketing, improve marketing effectiveness, considerably reduce cost of marketing,
Improve marketing input-output ratio and marketing income.
Brief description of the drawings
Fig. 1 represents the schematic flow sheet of target customer's screening technique of the embodiment of the present invention one.
Fig. 2 represents the specific implementation flow chart of step 100;
Fig. 3 represents the specific implementation flow chart of step 110;
Fig. 4 represents the specific implementation flow chart of step 112;
Fig. 5 represents the specific implementation flow chart of step 300;
Fig. 6 represents the structural representation of the incremental learning model of the embodiment of the present invention one;
Fig. 7 represent the embodiment of the present invention one based on naive Bayesian incremental learning system architecture diagram;
Fig. 8 represents the structural representation of target customer's screening plant of the embodiment of the present invention two.
Embodiment
It is right below in conjunction with the accompanying drawings and the specific embodiments to make the object, technical solutions and advantages of the present invention clearer
The present invention is described in detail.
The present invention can not combine marketing feedback result dynamic for existing static data mining model and adjust, and cause to market
Client screen it is inaccurate, the problem of marketing effectiveness difference, there is provided a kind of target customer's screening technique and device.
Embodiment one
As shown in figure 1, target customer's screening technique of the embodiment of the present invention, including:
Step 100, pre- this target customer for carrying out pushing presupposed information is filtered out in client's sample to be sorted;
It should be noted that the presupposed information is commonly referred to as marketing message, for example, be pushed to user 4G change planes information,
Subscribe to information of a certain function services etc..
Step 200, obtain and the push feedback result after presupposed information push is carried out to this described target customer;
User can make corresponding selection after presupposed information push is received according to the information content, receive the default letter
Cease or refuse the presupposed information, the push feedback result is the marketing feedback result marketed, generally include marketing into
Work((user receives marketing Push Service) and the result of marketing unsuccessful (user refuses marketing Push Service).
Step 300, the screening model for screening this target customer is modified according to the push feedback result, with root
The screening of next target customer is carried out according to revised screening model.
The embodiment of the present invention, by using marketing feedback result dynamic adjusting data mining model, improve marketing client
The accuracy of screening, substantially increase the hit rate and precision of marketing, improve marketing effectiveness, considerably reduce marketing into
This, improves marketing input-output ratio and marketing income.
Alternatively, as shown in Fig. 2 step 100 is in specific implementation, including:
Step 110, screening model is obtained;
It should be noted that in the present embodiment, using screening model of the Naive Bayes Classifier as client's sample.
Step 120, according to the screening model, each client's sample is calculated in client's sample to be sorted respectively in each class
Probable value on not;
Step 130, according to the probable value of each client's sample, it is default that push is selected in client's sample to be sorted
This target customer of information.
The step 100 is mainly accomplished that is filtering out target customer (i.e. marketing visitor using Naive Bayes Classifier
Family), it is generally the case that it regard the larger client's sample of the probable value obtained by the use of Naive Bayes Classifier as marketing client.
Further, as shown in figure 3, step 110 is in specific implementation, including:
Step 111, the training data and test data for building Naive Bayes Classifier are obtained;
Step 112, build to obtain initial grader using the training data;
Step 113, the grader is selected using the test data, obtains Naive Bayes Classifier.
It should be noted that when carrying out the structure of Naive Bayes Classifier, typically first with a part of client's sample
This builds original grader, then recycles test data tested original grader, be excellent as training data
Change, finally obtain Naive Bayes Classifier.
Specifically, as shown in figure 4, step 112 is when realizing, including:
Step 1121, the prior probability and conditional probability of initial grader are obtained using training data;
Step 1122, according to belonging to prior probability and conditional probability obtain each sample in training data, test data
Classification, and training error, test error are calculated, the minimum grader of test error is chosen as initial grader.
Under normal circumstances, the specific implementation of step 1121 is:
Utilize formula:Prior probability of the training data in each classification is calculated;
Utilize formula:The condition of training data is calculated
Probability;
Wherein, training data T={ (x1,y1),(x2,y2),...,(xN,yN), N represents to include sample in the training data
This x total number;P (Y=ck) represent the prior probability on classification Y;yiFor Y target variable, sample class, and y are representedi
∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability function;P(X(j)=ajl| Y=ck) table
Show the conditional probability of training data;J-th of feature of i-th of sample is represented, andajlIt is jth
Individual feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k=1,2 ..., K, K represent sample class
Number.
Under normal circumstances, the specific implementation of step 1122 is:
Utilize formula:By each sample in training data
Classification of the classification corresponding to maximum tried to achieve in pre-set categories as the sample;Wherein,
Y represents the sample class with maximum;P (Y=ck) represent the prior probability on classification Y;Y represents sample class
Not;ckFor sample class value;P(X(j)=x(j)| Y=ck) represent conditional probability, x(j)For x target variable, x represents that sample is real
Example, and x=(x(1),x(2),...,x(n))T。
Specifically, the specific implementation of step 113 is:
The concentration of target variable and the parameter of Bayesian Estimation in multiple adjusting training data, according to the mesh after each adjustment
The concentration of variable and the parameter of Bayesian Estimation are marked, evaluation index value is calculated in test data, obtains an assessment of maximum
The parameter of object event concentration and Bayesian Estimation corresponding to desired value, obtains Naive Bayes Classifier;
Wherein, the prior probability of the Naive Bayes Classifier is:
Conditional probability is:
Wherein, Pλ(Y=ck) be Naive Bayes Classifier prior probability;yiFor Y target variable, sample class is represented
Not, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;λ is Bayesian Estimation
Parameter, and λ >=0;Pλ(X(j)=ajl| Y=ck) be Naive Bayes Classifier conditional probability;Represent i-th sample
J-th of feature, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ..., n, l=1,
2,...,Sj, k=1,2 ..., K.
It should be noted that for Maximum-likelihood estimation it is possible that the probable value to be estimated be 0 in the case of, this
When influence whether the result of calculation of posterior probability, classification is produced deviation.In order to solve this problem, above-mentioned pattra leaves need to be used
The parameter of this estimation comes design conditions probability and prior probability.
Above-mentioned evaluation index value utilizes formula:It is calculated;
Wherein, F1 is evaluation index value, and P actual is positive example, represents the precision ratio of initial grader in test data,
And P=is predicted as the number of samples of the number of samples of positive example/be predicted as positive example;R is initial grader in test data
Recall ratio, and R=is predicted as the number of samples of number of samples/actual positive example of positive example.
Alternatively, the specific implementation of step 130 is:
Calculate client's sample to be sorted and belong to different classes of probable value, it is general more than non-target class to choose target class probable value
The sample of rate value, and the descending descending of probable value is arranged, choose the client's sample conduct for being arranged in predetermined number above
This target customer.
By taking two points of problems as an example, each sample is calculated and belongs to positive example, the probability of negative example, is more than in the probability of positive example negative
Probability is chosen in the sample of the probability of example and comes multiple client's samples above as this target customer, carries out marketing message
Push.
Specifically, as shown in figure 5, step 300 is in implementation, including:
Step 310, by the push feedback result compared with the class label of this target customer, in this target
Push feedback result is obtained in client to push failed first sample;
It should be noted that the meaning that collection pushes failed client's sample is contrast successfully and failed client's sample
Difference between this, the adjustment to Naive Bayes Classifier is further realized by difference therebetween.
Step 320, screening model is adjusted according to the first sample.
Further, the specific implementation of the step 320 is:
According to formula:Recalculate the elder generation of Naive Bayes Classifier
Test probability;
According to formula:Again
Calculate the conditional probability of Naive Bayes Classifier;
Wherein, P1(Y=ck) represent Naive Bayes Classifier prior probability;yiFor Y target variable, sample is represented
Classification, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;N represents training data
In include sample x total number;P1(X(j)=ajl| Y=ck) represent Naive Bayes Classifier conditional probability;Represent the
J-th of feature of i sample, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ...,
N, l=1,2 ..., Sj, k=1,2 ..., K;N1For the number of first sample.
It should be noted that the major issue that data mining process faces is the new data constantly to develop, it is existing
Grader needs constantly to adapt to it.When being handled for large batch of data set, if newly-increased sample closes with known sample
And post-process, on the one hand can increase the difficulty of study, still further aspect also because sample set is excessive and consume the excessive time and
Memory space.One effective solution method is that newly-increased sample set is respectively trained, and is progressively provided with the accumulation of sample set
Learn precision.Marketing feedback data belong to markd incremental data.Therefore, marketing feedback data are utilized in the embodiment of the present invention
Carry out the feature weight of Optimized model.Specially:The push feedback result (i.e. incremental data) that will be obtained, first with existing classification
Device examines its class label, and current class device is retained if matching, otherwise with the current grader of new samples amendment, selection it is new
Increase sample and need facilitate the nicety of grading for improving current class device, to adjust the prior probability of Naive Bayes Classifier and bar
Part probability.It should be noted simultaneously that the addition due to new samples so that sample information is added in prior probability, posteriority is general
Rate is also just determined jointly by prior probability and newly-increased sample information.
As shown in fig. 6, the principle of the embodiment of the present invention is:Priori determines sample knowledge, and the new samples of addition are carried out
The adjustment of sample knowledge, aposterior knowledge are determined by sample knowledge and priori, while aposterior knowledge adjusts priori again, certainly
The prediction result for determining test sample is generated by aposterior knowledge.
The application of the embodiment of the present invention in practice is specifically described with reference to Fig. 7 as follows:
By taking potential 4G renewal users incremental learning as an example:Extraction may change planes related to user first from data warehouse
Behavior property data set, such as the ARPU of user (Average Revenue Per User, i.e., per user's average income, are used for
Weigh the index of telecom operators and Internet firm's health service revenue), the duration of call, the data such as internet behavior feature, according to suitable
When data are split as training set, test set by ratio, the structure of Naive Bayes Classifier is carried out on training set, and using surveying
The parameter of examination set pair grader is selected, the Naive Bayes Classifier initialized;Secondly, extracted from data warehouse
The behavioural characteristic data of sample to be sorted, i.e. client, using the Naive Bayes Classifier of initialization, client is calculated can
The probability of changing planes of energy, i.e. target customer;By marketing service device, the modes such as short message, artificial outgoing call are carried out to target customer and entered
Field headquarters pin contacts;Using marketing service feedback collection device, marketing result is gathered, result data of marketing is put in storage;Seeking
On the basis of selling data acquisition, using incremental learning device, prior probability, posterior probability corresponding to the grader of initialization are entered
Mobile state tuning, and the model after tuning is screened using test set, the grader after being updated, into next round
Iterative cycles.In the process, Naive Bayes Classifier carries out self-renewing constantly according to increment marketing feedback data.
Theoretically, any marketing activity all has target customers, and selection target customers are all conditional, this
A little conditions all can constantly learn to obtain by incremental learning device, so by the embodiment of the present invention based on simple shellfish
Ye Si incremental learning system can improve the precision of target customer, improve marketing effectiveness, and customers' screening restraining factors are got over
It is more, it can more embody the practical value of the present invention.
It should be noted that the present invention not only make it that marketing objectives client positioning is more accurate, and marketing is utilized first
Feedback information dynamic construction Naive Bayes Classifier, the hit rate and precision of marketing are substantially increased, improve marketing effect
Fruit, cost of marketing is considerably reduced, improve marketing input-output ratio, maximize marketing income.
Embodiment two
As shown in figure 8, the embodiment of the present invention provides a kind of target customer's screening plant, including:
Screening module 10, for filtering out pre- this target visitor for carrying out pushing presupposed information in client's sample to be sorted
Family;
Feedback result acquisition module 20, the push after presupposed information push is carried out to this described target customer for obtaining
Feedback result;
Correcting module 30, for being repaiied according to the push feedback result to the screening model for screening this target customer
Just, with the screening according to the next target customer of revised screening model progress.
Specifically, the screening module 10 includes:
Model acquisition submodule, for obtaining screening model;
Probability calculation submodule, for according to the screening model, calculating each client in client's sample to be sorted respectively
Probable value of the sample in each classification;
Submodule is chosen, for the probable value according to each client's sample, is selected in client's sample to be sorted
Push this target customer of presupposed information.
When the screening model is Naive Bayes Classifier, alternatively, the model acquisition submodule includes:
Acquiring unit, for obtaining the training data and test data that are used for building Naive Bayes Classifier;
Grader construction unit, for building to obtain initial grader using the training data;
Selecting unit, for being selected using the test data the grader, obtain Naive Bayes Classification
Device.
Specifically, the grader construction unit is specifically used for:
The prior probability and conditional probability of initial grader are obtained using training data;
Classification according to belonging to prior probability and conditional probability obtain each sample in training data, test data, and count
Training error, test error are calculated, chooses the minimum grader of test error as initial grader.
Alternatively, the prior probability of initial grader and the specific implementation of conditional probability are obtained using training data
For:
Utilize formula:Prior probability of the training data in each classification is calculated;
Utilize formula:The condition of training data is calculated
Probability;
Wherein, training data T={ (x1,y1),(x2,y2),...,(xN,yN), N represents to include sample in the training data
This x total number;P (Y=ck) represent the prior probability on classification Y;yiFor Y target variable, sample class, and y are representedi
∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability function;P(X(j)=ajl| Y=ck) table
Show the conditional probability of training data;J-th of feature of i-th of sample is represented, andajlIt is jth
Individual feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k=1,2 ..., K.
Alternatively, the class according to belonging to prior probability and conditional probability obtain each sample in training data, test data
Not, and training error, test error are calculated, chooses specific implementation of the minimum grader of test error as initial grader
Mode is:
Utilize formula:By each sample in training data
Classification of the classification corresponding to maximum tried to achieve in pre-set categories as the sample;Wherein,
Y represents the sample class with maximum;P (Y=ck) represent the prior probability on classification Y;Y represents sample class
Not;ckFor sample class value;P(X(j)=x(j)| Y=ck) represent conditional probability, x(j)For x target variable, x represents that sample is real
Example, and x=(x(1),x(2),...,x(n))T。
Specifically, the selecting unit is used for:
The concentration of target variable and the parameter of Bayesian Estimation in multiple adjusting training data, according to the mesh after each adjustment
The concentration of variable and the parameter of Bayesian Estimation are marked, evaluation index value is calculated in test data, obtains an assessment of maximum
The parameter of object event concentration and Bayesian Estimation corresponding to desired value, obtains Naive Bayes Classifier;
Wherein, the prior probability of the Naive Bayes Classifier is:
Conditional probability is:
Wherein, Pλ(Y=ck) be Naive Bayes Classifier prior probability;yiFor Y target variable, sample class is represented
Not, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;λ is Bayesian Estimation
Parameter;Pλ(X(j)=ajl| Y=ck) be Naive Bayes Classifier conditional probability;Represent j-th of spy of i-th of sample
Sign, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k
=1,2 ..., K.
Wherein, the evaluation index value utilizes formula:It is calculated;
Wherein, F1 is evaluation index value, and P actual is positive example, represents the precision ratio of initial grader in test data,
And P=is predicted as the number of samples of the number of samples of positive example/be predicted as positive example;R is initial grader in test data
Recall ratio, and R=is predicted as the number of samples of number of samples/actual positive example of positive example.
Alternatively, the selection submodule is specifically used for:
Calculate client's sample to be sorted and belong to different classes of probable value, it is general more than non-target class to choose target class probable value
The sample of rate value, and the descending descending of probable value is arranged, choose the client's sample conduct for being arranged in predetermined number above
This target customer.
Alternatively, the correcting module 30 includes:
Feedback result screens submodule, for the push feedback result and the class label of this target customer to be carried out
Compare, push feedback result is obtained in this target customer to push failed first sample;
Submodule is adjusted, for being adjusted according to the first sample to screening model.
Wherein, when the screening model is Naive Bayes Classifier, the adjustment submodule is specifically used for:
According to formula:Recalculate the elder generation of Naive Bayes Classifier
Test probability;
According to formula:Again
Calculate the conditional probability of Naive Bayes Classifier;
Wherein, P1(Y=ck) represent Naive Bayes Classifier prior probability;yiFor Y target variable, sample is represented
Classification, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;N represents training data
In include sample x total number;P1(X(j)=ajl| Y=ck) represent Naive Bayes Classifier conditional probability;Represent the
J-th of feature of i sample, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ...,
N, l=1,2 ..., Sj, k=1,2 ..., K;N1For the number of first sample.
It should be noted that the device embodiment is the device corresponding with the above method, the institute of above method embodiment
There is implementation suitable for the device embodiment, can also reach identical technique effect.
Above-described is the preferred embodiment of the present invention, it should be pointed out that is come for the ordinary person of the art
Say, some improvements and modifications can also be made under the premise of principle of the present invention is not departed from, and these improvements and modifications also exist
In protection scope of the present invention.
Claims (14)
- A kind of 1. target customer's screening technique, it is characterised in that including:Pre- this target customer for carrying out pushing presupposed information is filtered out in client's sample to be sorted;Obtain and the push feedback result after presupposed information push is carried out to this described target customer;The screening model for screening this target customer is modified according to the push feedback result, with according to revised sieve Modeling type carries out the screening of next target customer.
- 2. target customer's screening technique according to claim 1, it is characterised in that described to be sieved in client's sample to be sorted Select it is pre- carry out push presupposed information this target customer the step of include:Obtain screening model;According to the screening model, each probability of client's sample in each classification in client's sample to be sorted is calculated respectively Value;According to the probable value of each client's sample, this mesh for pushing presupposed information is selected in client's sample to be sorted Mark client.
- 3. target customer's screening technique according to claim 2, it is characterised in that the screening model is naive Bayesian During grader, described the step of obtaining screening model, includes:Obtain the training data and test data for building Naive Bayes Classifier;Build to obtain initial grader using the training data;The grader is selected using the test data, obtains Naive Bayes Classifier.
- 4. target customer's screening technique according to claim 3, it is characterised in that described to be built using the training data The step of obtaining initial grader includes:The prior probability and conditional probability of initial grader are obtained using training data;Classification according to belonging to prior probability and conditional probability obtain each sample in training data, test data, and calculate instruction Practice error, test error, choose the minimum grader of test error as initial grader.
- 5. target customer's screening technique according to claim 4, it is characterised in that described to be obtained initially using training data Grader prior probability and conditional probability the step of include:Utilize formula:Prior probability of the training data in each classification is calculated;Utilize formula:The conditional probability of training data is calculated;Wherein, training data T={ (x1,y1),(x2,y2),...,(xN,yN), N represents to include sample x in the training data Total number;P (Y=ck) represent the prior probability on classification Y;yiFor Y target variable, sample class, and y are representedi∈ {c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability function;P(X(j)=ajl| Y=ck) represent The conditional probability of training data;J-th of feature of i-th of sample is represented, andajlIt is j-th Feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k=1,2 ..., K.
- 6. target customer's screening technique according to claim 4, it is characterised in that described general according to prior probability and condition Rate obtains the classification belonging to each sample in training data, test data, and calculates training error, test error, chooses test The minimum grader of error includes as the step of initial grader:Utilize formula:By each sample in training data pre- If classification of the classification as the sample corresponding to the maximum tried to achieve in classification;Wherein,Y represents the sample class with maximum;P (Y=ck) represent the prior probability on classification Y;Y represents sample class;ck For sample class value;P(X(j)=x(j)| Y=ck) represent conditional probability, x(j)For x target variable, x represents sample instance, and x =(x(1),x(2),...,x(n))T。
- 7. target customer's screening technique according to claim 3, it is characterised in that described to utilize the test data to institute The step of stating grader to be selected, obtaining Naive Bayes Classifier includes:The concentration of target variable and the parameter of Bayesian Estimation in multiple adjusting training data, become according to the target after each adjustment The concentration of amount and the parameter of Bayesian Estimation, evaluation index value is calculated in test data, obtain an evaluation index of maximum The parameter of the corresponding object event concentration of value and Bayesian Estimation, obtains Naive Bayes Classifier;Wherein, the prior probability of the Naive Bayes Classifier is:Conditional probability is:Wherein, Pλ(Y=ck) be Naive Bayes Classifier prior probability;yiFor Y target variable, sample class is represented, and yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;λ is the parameter of Bayesian Estimation; Pλ(X(j)=ajl| Y=ck) be Naive Bayes Classifier conditional probability;J-th of feature of i-th of sample is represented, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ..., n, l=1,2 ..., Sj, k=1, 2,...,K。
- 8. target customer's screening technique according to claim 7, it is characterised in that the evaluation index value utilizes formula:It is calculated;Wherein, F1 is evaluation index value, and P actual is positive example, represents the precision ratio of initial grader in test data, and P =number of samples for being predicted as positive example/is predicted as the number of samples of positive example;R is initial grader looking into entirely in test data Rate, and R=is predicted as the number of samples of number of samples/actual positive example of positive example.
- 9. target customer's screening technique according to claim 2, it is characterised in that the institute of each client's sample of basis The step of stating probable value, this target customer for pushing presupposed information is selected in client's sample to be sorted includes:Calculate client's sample to be sorted and belong to different classes of probable value, choose target class probable value and be more than non-target class probable value Sample, and the descending descending of probable value is arranged, chooses and be arranged in client's sample of predetermined number above and be used as this Target customer.
- 10. target customer's screening technique according to claim 1, it is characterised in that described according to the push feedback knot The step of fruit is modified to the screening model for screening this target customer includes:By the push feedback result compared with the class label of this target customer, obtain and push away in this target customer Feedback result is sent to push failed first sample;Screening model is adjusted according to the first sample.
- 11. target customer's screening technique according to claim 10, it is characterised in that when the screening model is simple shellfish During this grader of leaf, described the step of being adjusted according to the first sample to screening model, includes:According to formula:The priori for recalculating Naive Bayes Classifier is general Rate;According to formula:Recalculate The conditional probability of Naive Bayes Classifier;Wherein, P1(Y=ck) represent Naive Bayes Classifier prior probability;yiFor Y target variable, sample class is represented, And yi∈{c1,c2,...,cK, ckFor sample class value;I () represents to seek sample class probability;N represents to include in training data Sample x total number;P1(X(j)=ajl| Y=ck) represent Naive Bayes Classifier conditional probability;Represent i-th of sample This j-th of feature, andajlIt is that j-th of feature obtains l-th of value;J=1,2 ..., n, l= 1,2,...,Sj, k=1,2 ..., K;N1For the number of first sample.
- A kind of 12. target customer's screening plant, it is characterised in that including:Screening module, for filtering out pre- this target customer for carrying out pushing presupposed information in client's sample to be sorted;Feedback result acquisition module, the push feedback knot after presupposed information push is carried out to this described target customer for obtaining Fruit;Correcting module, for being modified according to the push feedback result to the screening model for screening this target customer, with The screening of next target customer is carried out according to revised screening model.
- 13. target customer's screening plant according to claim 12, it is characterised in that the screening module includes:Model acquisition submodule, for obtaining screening model;Probability calculation submodule, for according to the screening model, calculating each client's sample in client's sample to be sorted respectively Probable value in each classification;Submodule is chosen, for the probable value according to each client's sample, push is selected in client's sample to be sorted This target customer of presupposed information.
- 14. target customer's screening plant according to claim 12, it is characterised in that the correcting module includes:Feedback result screens submodule, for the push feedback result and the class label of this target customer to be compared Compared with it is to push failed first sample that push feedback result is obtained in this target customer;Submodule is adjusted, for being adjusted according to the first sample to screening model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610652832.0A CN107730286A (en) | 2016-08-10 | 2016-08-10 | A kind of target customer's screening technique and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610652832.0A CN107730286A (en) | 2016-08-10 | 2016-08-10 | A kind of target customer's screening technique and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107730286A true CN107730286A (en) | 2018-02-23 |
Family
ID=61199431
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610652832.0A Pending CN107730286A (en) | 2016-08-10 | 2016-08-10 | A kind of target customer's screening technique and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107730286A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109191210A (en) * | 2018-09-13 | 2019-01-11 | 厦门大学嘉庚学院 | A kind of broadband target user's recognition methods based on Adaboost algorithm |
CN110046770A (en) * | 2019-04-23 | 2019-07-23 | 中国科学技术大学 | Grain mildew prediction technique and device |
CN110062023A (en) * | 2019-03-12 | 2019-07-26 | 阿里巴巴集团控股有限公司 | A kind of safety education information-pushing method, device and equipment |
CN110119466A (en) * | 2019-03-29 | 2019-08-13 | 五渡(杭州)科技有限责任公司 | A kind of big data intelligent marketing system and method |
CN110457566A (en) * | 2019-08-15 | 2019-11-15 | 腾讯科技(武汉)有限公司 | Method, device, electronic equipment and storage medium |
CN110619534A (en) * | 2018-06-19 | 2019-12-27 | 华为技术有限公司 | Marketing information pushing method and device and readable storage medium |
CN111612492A (en) * | 2019-02-26 | 2020-09-01 | 北京奇虎科技有限公司 | User online accurate marketing method and device based on multi-feature fusion |
CN111768040A (en) * | 2020-07-01 | 2020-10-13 | 深圳前海微众银行股份有限公司 | Model interpretation method, device, equipment and readable storage medium |
CN111797942A (en) * | 2020-07-23 | 2020-10-20 | 深圳壹账通智能科技有限公司 | User information classification method and device, computer equipment and storage medium |
CN112581191A (en) * | 2020-08-14 | 2021-03-30 | 支付宝(杭州)信息技术有限公司 | Training method and device of behavior prediction model |
CN113065885A (en) * | 2021-03-01 | 2021-07-02 | 苏宁金融科技(南京)有限公司 | Method and system for intelligent marketing |
CN114418376A (en) * | 2022-01-14 | 2022-04-29 | 上海明胜品智人工智能科技有限公司 | Enterprise task issuing method and device, electronic equipment and storage medium |
CN114757724A (en) * | 2022-06-14 | 2022-07-15 | 湖南三湘银行股份有限公司 | Precise information pushing system and method based on genetic algorithm |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102982042A (en) * | 2011-09-07 | 2013-03-20 | 中国移动通信集团公司 | Personalization content recommendation method and platform and system |
CN103150696A (en) * | 2011-12-06 | 2013-06-12 | 中兴通讯股份有限公司 | Method and device for selecting potential customer of target value-added service |
CN104391860A (en) * | 2014-10-22 | 2015-03-04 | 安一恒通(北京)科技有限公司 | Content type detection method and device |
CN105608592A (en) * | 2014-08-21 | 2016-05-25 | 深圳市深讯信息科技发展股份有限公司 | Telecom user intelligent analyzing and pushing method and telecom user intelligent analyzing and pushing system |
-
2016
- 2016-08-10 CN CN201610652832.0A patent/CN107730286A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102982042A (en) * | 2011-09-07 | 2013-03-20 | 中国移动通信集团公司 | Personalization content recommendation method and platform and system |
CN103150696A (en) * | 2011-12-06 | 2013-06-12 | 中兴通讯股份有限公司 | Method and device for selecting potential customer of target value-added service |
CN105608592A (en) * | 2014-08-21 | 2016-05-25 | 深圳市深讯信息科技发展股份有限公司 | Telecom user intelligent analyzing and pushing method and telecom user intelligent analyzing and pushing system |
CN104391860A (en) * | 2014-10-22 | 2015-03-04 | 安一恒通(北京)科技有限公司 | Content type detection method and device |
Non-Patent Citations (2)
Title |
---|
唐炉亮 等: "一种基于朴素贝叶斯分类的车道数量探测", 《中国公路学报》 * |
高影繁 等: "一种结合参数优化的贝叶斯文本分类算法", 《计算机研究与发展》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110619534A (en) * | 2018-06-19 | 2019-12-27 | 华为技术有限公司 | Marketing information pushing method and device and readable storage medium |
CN109191210A (en) * | 2018-09-13 | 2019-01-11 | 厦门大学嘉庚学院 | A kind of broadband target user's recognition methods based on Adaboost algorithm |
CN111612492A (en) * | 2019-02-26 | 2020-09-01 | 北京奇虎科技有限公司 | User online accurate marketing method and device based on multi-feature fusion |
CN110062023A (en) * | 2019-03-12 | 2019-07-26 | 阿里巴巴集团控股有限公司 | A kind of safety education information-pushing method, device and equipment |
CN110062023B (en) * | 2019-03-12 | 2021-08-31 | 创新先进技术有限公司 | Safety education information pushing method, device and equipment |
CN110119466A (en) * | 2019-03-29 | 2019-08-13 | 五渡(杭州)科技有限责任公司 | A kind of big data intelligent marketing system and method |
CN110046770B (en) * | 2019-04-23 | 2021-04-23 | 中国科学技术大学 | Grain mildew prediction method and device |
CN110046770A (en) * | 2019-04-23 | 2019-07-23 | 中国科学技术大学 | Grain mildew prediction technique and device |
CN110457566A (en) * | 2019-08-15 | 2019-11-15 | 腾讯科技(武汉)有限公司 | Method, device, electronic equipment and storage medium |
CN111768040A (en) * | 2020-07-01 | 2020-10-13 | 深圳前海微众银行股份有限公司 | Model interpretation method, device, equipment and readable storage medium |
CN111797942A (en) * | 2020-07-23 | 2020-10-20 | 深圳壹账通智能科技有限公司 | User information classification method and device, computer equipment and storage medium |
CN112581191A (en) * | 2020-08-14 | 2021-03-30 | 支付宝(杭州)信息技术有限公司 | Training method and device of behavior prediction model |
CN113065885A (en) * | 2021-03-01 | 2021-07-02 | 苏宁金融科技(南京)有限公司 | Method and system for intelligent marketing |
CN114418376A (en) * | 2022-01-14 | 2022-04-29 | 上海明胜品智人工智能科技有限公司 | Enterprise task issuing method and device, electronic equipment and storage medium |
CN114757724A (en) * | 2022-06-14 | 2022-07-15 | 湖南三湘银行股份有限公司 | Precise information pushing system and method based on genetic algorithm |
CN114757724B (en) * | 2022-06-14 | 2022-09-20 | 湖南三湘银行股份有限公司 | Precise information pushing system and method based on genetic algorithm |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107730286A (en) | A kind of target customer's screening technique and device | |
Idris et al. | Intelligent churn prediction for telecom using GP-AdaBoost learning and PSO undersampling | |
CN103166830B (en) | A kind of Spam Filtering System of intelligent selection training sample and method | |
CN101877837B (en) | Method and device for short message filtration | |
CN108540320A (en) | The appraisal procedure of user satisfaction is excavated based on signaling | |
CN107766929B (en) | Model analysis method and device | |
CN103617435B (en) | Image sorting method and system for active learning | |
CN109299344A (en) | The generation method of order models, the sort method of search result, device and equipment | |
CN109242135A (en) | A kind of model method for running, device and service server | |
CN105373606A (en) | Unbalanced data sampling method in improved C4.5 decision tree algorithm | |
CN107862022A (en) | Cultural resource commending system | |
CN105550295B (en) | Disaggregated model optimization method and device | |
CN109344738A (en) | The recognition methods of crop diseases and pest crop smothering and device | |
CN105491599A (en) | Novel regression system for predicting LTE network performance indexes | |
CN111598143A (en) | Credit evaluation-based defense method for federal learning poisoning attack | |
CN108022589A (en) | Aiming field classifier training method, specimen discerning method, terminal and storage medium | |
CN107832132A (en) | Application control method, apparatus, storage medium and electronic equipment | |
CN106960248A (en) | A kind of method and device that customer problem is predicted based on data-driven | |
CN106991577A (en) | A kind of method and device for determining targeted customer | |
CN110070452A (en) | Model training method, calculates equipment and computer readable storage medium at device | |
CN107808346A (en) | A kind of appraisal procedure and apparatus for evaluating of potential target object | |
CN107239964A (en) | User is worth methods of marking and system | |
CN116775879A (en) | Fine tuning training method of large language model, contract risk review method and system | |
CN108563786A (en) | Text classification and methods of exhibiting, device, computer equipment and storage medium | |
Sangaiah et al. | Automatic fault detection and diagnosis in cellular networks and beyond 5g: intelligent network management |
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: 20180223 |