CN108269012A - Construction method, device, storage medium and the terminal of risk score model - Google Patents
Construction method, device, storage medium and the terminal of risk score model Download PDFInfo
- Publication number
- CN108269012A CN108269012A CN201810030179.3A CN201810030179A CN108269012A CN 108269012 A CN108269012 A CN 108269012A CN 201810030179 A CN201810030179 A CN 201810030179A CN 108269012 A CN108269012 A CN 108269012A
- Authority
- CN
- China
- Prior art keywords
- account number
- risk score
- division
- sample database
- abnormal
- 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
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/08—Insurance
Abstract
The present invention provides a kind of construction method of risk score model, the construction method includes:Blacklist sample database and white list sample database are built according to preset Account Data, the blacklist sample database includes abnormal account number, and the white list sample database includes normal account number;Decision tree GBDT algorithms are promoted based on gradient, cluster training is carried out to the normal account number in the abnormal account number and white list sample database in the blacklist sample database, filter out abnormal account number characteristic of division;The abnormal account number characteristic of division is trained based on random forest RF algorithms, obtains each corresponding contribution degree of exception account number characteristic of division;According to the abnormal account number characteristic of division and its corresponding contribution degree, risk score model is built, the risk score model is used to identify abnormal account number.Risk score model constructed by the present invention improves the timeliness for removing abnormal account number, reduces the noise jamming that abnormal account number is brought, improves the computational accuracy of all multi objectives of APP.
Description
Technical field
The invention belongs to field of communication technology more particularly to a kind of construction method of risk score model, device, storage Jie
Matter and terminal.
Background technology
There are a large amount of abnormal fake registrations users in life insurance class APP at present, it is active that these fake registrations user has brush
The behaviors such as degree, brush activity.And the rate of enlivening of APP is to log in ratio of the number with user's total amount of APP, conversion ratio is obtains objective quantity
Ratio with the number of users for buying product.Existing exception account number recognition methods is the passive discernings mechanism such as ex-post analysis, clearly
Except the timeliness and precision of abnormal account number are low, the assessment of rate and conversion ratio is enlivened to APP so as to cause fake registrations user and is carried
It rises and brings great noise jamming, reduce the computational accuracy of all multi objectives of APP.Further, since the need of fake registrations user
Ask stimulation, criminal also using in APP loophole complete the registration of abnormal account number, derive diversified abnormal user
Version is registered, there are larger safety.
Invention content
An embodiment of the present invention provides a kind of construction method, device, storage medium and the terminal of risk score model, to carry
High definition removes the timeliness of abnormal account number, reduces the noise jamming that abnormal account number is brought, promotes the computational accuracy of all multi objectives of APP.
An embodiment of the present invention provides a kind of construction method of risk score model, the construction method includes:
Blacklist sample database and white list sample database are built according to preset Account Data, wrapped in the blacklist sample database
Abnormal account number is included, the white list sample database includes normal account number;
Decision tree GBDT algorithms are promoted to the abnormal account number and white list sample in the blacklist sample database based on gradient
Normal account number in library carries out cluster training, filters out abnormal account number characteristic of division;
The abnormal account number characteristic of division is trained based on random forest RF algorithms, obtains each abnormal account number point
The corresponding contribution degree of category feature;
According to the abnormal account number characteristic of division and its corresponding contribution degree, risk score model is built, the risk is commented
Sub-model is used to identify abnormal account number.
Further, the structure risk score model, further includes later:
The corresponding risk of each normal account number in the white list sample database is calculated using the risk score model
Scoring;
Normal account number of the risk score greater than or equal to preset risk threshold value is filtered, retains risk score less than default
Risk threshold value normal account number;
Each corresponding risk of exception account number in the blacklist sample database is calculated using the risk score model
Scoring;
According to the risk score of the normal account number retained and the risk score of abnormal account number, the risk threshold value is corrected.
Further, it is described that decision tree GBDT algorithms are promoted to the abnormal account number in the blacklist sample database based on gradient
Cluster training is carried out with the normal account number in white list sample database to include:
Obtain each abnormal account number, the corresponding sample value of characteristic information of normal account number;
According to the corresponding sample value of the characteristic information, according to the characteristic condition of prediction by the abnormal account number, normal account
Number distribution is to first regression tree in GBDT models, until each abnormal account number, normal account number, which are divided equally, is assigned to each leaf
Child node;
Loss function is obtained, initializes the constant value of loss function minimization;
For each leaf node, according to each abnormal account number of the loss function and constant value estimation, normal
The corresponding residual error approximation of account number;
Based on all lower regression trees of residual error approximation repetitive exercise.
Further, the abnormal account number characteristic of division includes:
Continuous high frequency binding feature, continuous cipher binding feature, continuous IP binding feature, IP high diverging rates feature, account number
The same regional feature of business personnel, registration binding time difference feature, registration binding bury point feature without front end.
Further, it is described that the abnormal account number characteristic of division is trained based on random forest RF algorithms, it obtains every
The corresponding contribution degree of one exception account number characteristic of division includes:
Arbitrary combination is carried out by predetermined number to the abnormal account number characteristic of division by random forest RF models, is obtained more
A categorised decision tree;
The abnormal account number characteristic of division not being combined in each categorised decision tree is obtained, based on the abnormal account number point
Category feature estimates the error in classification of the categorised decision tree;
The categorised decision tree of error in classification minimum is chosen, and each abnormal account number point is calculated based on the categorised decision tree
The contribution degree of category feature.
Further, the structure risk score model includes structure risk score formula:
F (X)=F0+β1T1(X)+β2T2(X)+……++βmTm(X)
Wherein, risk scores of the F (X) for user account number X, F0For regression residuals item, Ti(X) it is abnormal account number characteristic of division
Feature combinatorial formula, βiIt is characterized combinatorial formula Ti(X) the sum of contribution degree of each exception account number characteristic of division in.
The embodiment of the present invention additionally provides a kind of construction device of risk score model, and the construction device includes:
Sample database builds module, for building blacklist sample database and white list sample database according to preset Account Data,
The blacklist sample database includes abnormal account number, and the white list sample database includes normal account number;
Feature Selection module, for promoting decision tree GBDT algorithms to the exception in the blacklist sample database based on gradient
Normal account number in account number and white list sample database carries out cluster training, filters out abnormal account number characteristic of division;
Contribution degree training module is trained the abnormal account number characteristic of division for being based on random forest RF algorithms,
Obtain each corresponding contribution degree of exception account number characteristic of division;
Model construction module, for according to the abnormal account number characteristic of division and its corresponding contribution degree, structure risk to be commented
Sub-model, the risk score model are used to identify abnormal account number.
Further, the construction device further includes:
First grading module, for each in the use risk score model calculating white list sample database just
The corresponding risk score of normal account number;
Account number filtering module for normal account number of the risk score greater than or equal to preset risk threshold value to be filtered, is protected
Risk score is stayed to be less than the normal account number of preset risk threshold value;
Second grading module, for different using each in the risk score model calculating blacklist sample database
The corresponding risk score of normal account number;
Threshold correction module, for the risk score according to the normal account number retained and the risk score of abnormal account number,
Correct the risk threshold value.
The embodiment of the present invention additionally provides a kind of computer readable storage medium, is stored thereon with computer program, the journey
The step described in the construction method of risk score model as described above is realized when sequence is performed as processor.
The embodiment of the present invention additionally provides a kind of terminal, and the terminal includes memory, processor and is stored in memory
Computer program that is upper and can running on a processor, the processor are realized as described above when performing the computer program
Step described in the construction method of risk score model.
Compared with prior art, the embodiment of the present invention introduces gradient and promotes decision tree GBDT algorithms, is determined by gradient promotion
Plan tree GBDT algorithms concentrate the normal account number in the abnormal account number and white list sample database in the blacklist sample database
Training, to eliminate invalid characteristic of division, obtains the higher abnormal account number characteristic of division of precision;Then random forest RF is introduced
Algorithm is trained the abnormal account number characteristic of division by random forest RF algorithms, obtains each abnormal account number classification
The corresponding contribution degree of feature;Finally according to the abnormal account number characteristic of division and its corresponding contribution degree, risk score mould is built
Type.Since the arithmetic speed of RF algorithms is fast, precision is higher, the timeliness for removing abnormal account number is effectively improved, based on the wind
Dangerous Rating Model can rapidly identify abnormal account number in time, and the noise generated so as to significantly reduce abnormal account number is done
It disturbs, improves storage user quality, and then improve the computational accuracy of all multi objectives of APP.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention, for those of ordinary skill in the art, without creative efforts, can be with
Other accompanying drawings are obtained according to these attached drawings.
Fig. 1 is the first realization flow chart of the construction method of risk score model provided in an embodiment of the present invention;
Fig. 2 is the second realization flow chart of the construction method of risk score model provided in an embodiment of the present invention;
Fig. 3 is that the third of the construction method of risk score model provided in an embodiment of the present invention realizes flow chart;
Fig. 4 is the 4th realization flow chart of the construction method of risk score model provided in an embodiment of the present invention;
Fig. 5 is risk score distribution map provided in an embodiment of the present invention;
Fig. 6 is the composition structure chart of the construction device of risk score model provided in an embodiment of the present invention;
Fig. 7 is the schematic diagram of terminal provided in an embodiment of the present invention.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, it is right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Fig. 1 shows that the first of the construction method of risk score model provided in an embodiment of the present invention realizes flow.This hair
The construction method for the risk score model that bright embodiment provides is applied to terminal.Refering to Fig. 1, the method includes:
In step S101, blacklist sample database and white list sample database are built according to preset Account Data, it is described black
List sample database includes abnormal account number, and the white list sample database includes normal account number.
The embodiment of the present invention obtains a collection of Account Data from specified channel in advance, these Account Datas are by manual verification
Abnormal account number or normal account number.Herein, the specified channel can have multiple;For the abnormal account number, including but not
It is limited to false list that mechanism confirms and report, walks the false list of former interface registration that new edition APP do not use, from network
The false list that can not be put through, the outside platform of purchase drag the false list that library is come and the false name obtained by reptile mode
It is single;For the normal account number, the declaration form user and interior employee of product had including but not limited to been bought.
Herein, the embodiment of the present invention cleans the Account Data, to filter out the account number sample of feature missing,
Then classification is divided, blacklist sample database and white list sample database are built, wherein the blacklist sample database includes several
Abnormal account number, the white list sample database include several normal account numbers.Each abnormal account number or normal account number are wrapped
Include multiple characteristic informations.Optionally, the characteristic information includes but not limited to:
Occupation, gender, age, registion time, binding time, number attribution, business personnel's work number of binding, account number are close
Code, registration IP, binding IP, accumulative business personnel bind number, accumulative cryptographic binding frequency of usage, accumulative ip bindings frequency of usage, industry
Business person's cell-phone number ownership place, business personnel bind interval, password using interval, ip using interval, ip diverging rates, whether there is front end to bury
Point.
In step s 102, decision tree GBDT algorithms are promoted to the abnormal account number in the blacklist sample database based on gradient
Cluster training is carried out with the normal account number in white list sample database, filters out abnormal account number characteristic of division.
In embodiments of the present invention, the blacklist sample database includes abnormal account number, and white list sample database is included normally
Account number, each exception account number or normal account number include several characteristic informations, and total quantity is more, need to these characteristic informations into
Row screening.Herein, the embodiment of the present invention promotes decision tree GBDT algorithms come to the abnormal account number and normal account using gradient
Number carry out combined training, to reject some useless characteristic informations.Wherein, a kind of decision Tree algorithms of iteration of GBDT algorithms, should
Algorithm is made of more decision trees.Each newly-built decision tree is the gradient in the residual error reduction of previous decision tree
(Gradient) it is set up on direction, that is, the foundation of every decision tree is in order to enable the residual error of decision tree is past before
Gradient direction is reduced, and so as to more accurately select validity feature and weed out invalid feature, obtains higher different of precision
Normal account number characteristic of division, improve for build risk score model characteristic of division accuracy.
In step s 103, the abnormal account number characteristic of division is trained based on random forest RF algorithms, obtained every
The corresponding contribution degree of one exception account number characteristic of division.
After abnormal account number characteristic of division is got, the corresponding contribution degree of the training abnormal account number characteristic of division,
That is weights.Wherein, calculate the sorting algorithm of contribution degree, the embodiment of the present invention uses random forest RF algorithms, this be compare ID3,
It is chosen after the computational efficiency and precision of the Decision Tree Algorithms such as CART and Logic recurrence sorting algorithms.The RF algorithms pair
Faster, nicety of grading higher, error in classification is lower for the arithmetic speed of abnormal account number characteristic of division that the embodiment of the present invention is inputted.
In step S104, according to the abnormal account number characteristic of division and its corresponding contribution degree, risk score model is built,
The risk score model is used to identify abnormal account number.
It is after the abnormal account number characteristic of division is obtained, the abnormal account number characteristic of division and its corresponding contribution degree is defeated
Enter preset model, obtain risk score model.Optionally, the risk score model is Logic Regression Models.
In conclusion the embodiment of the present invention promotes decision tree GBDT algorithms by introducing gradient, decision is promoted based on gradient
Tree GBDT algorithms carry out concentration instruction to the normal account number in the abnormal account number and white list sample database in the blacklist sample database
Practice, to eliminate invalid characteristic of division, obtain abnormal account number characteristic of division, reduce the sum of characteristic information;Then introduce with
Machine forest RF algorithms are trained the abnormal account number characteristic of division based on random forest RF algorithms, obtain each exception
The corresponding contribution degree of account number characteristic of division;Finally according to the abnormal account number characteristic of division and its corresponding contribution degree, wind is built
Dangerous Rating Model.Since the arithmetic speed of RF algorithms is fast, precision is higher, is effectively improved the timeliness for removing abnormal account number,
Abnormal account number can rapidly be identified based on the risk score model in time, generated so as to significantly reduce abnormal account number
Noise jamming, improve storage user quality, and then improve the computational accuracy of all multi objectives of APP.
Further, it on the basis of the first realization flow of the construction method based on Fig. 1 risk score models provided, carries
Go out the construction method of risk score model provided in an embodiment of the present invention second realizes flow.
As shown in Fig. 2, be the construction method of risk score model provided in an embodiment of the present invention second realization flow show
It is intended to.In embodiments of the present invention, the step S102 is based on gradient and promotes decision tree GBDT algorithms to the blacklist sample
The normal account number in abnormal account number and white list sample database in library be trained including:
In step S1021, each abnormal account number, the corresponding sample value of characteristic information of normal account number are obtained.
Herein, the embodiment of the present invention is mixed described different based on the abnormal account number and normal account number obtained in step S101
Normal account number and normal account number are trained.Characteristic information in each abnormal account number or normal account number has corresponding sample
Value.Wherein, the sample value is characterized the value of information, for example, abnormal account number and age characteristics, gender in normal account number it is special
Sign, the age of a certain exception account number a are the sample value that 40 years old is the age characteristics, and gender is that male is the sex character
Sample value.
It, will be described different according to the characteristic condition of prediction according to the corresponding sample value of the characteristic information in step S1022
Normal account number, normal account number distribute the first regression tree into GBDT models, until each abnormal account number, normal account number are divided equally
It is assigned to each leaf node.
For each characteristic information, the sample value of all abnormal account numbers and normal account number is obtained, by the characteristic information pair
It is trained in all sample values input GBDT models answered.The GBDT models include several regression trees, every
The node of regression tree except for the leaf nodes all corresponds to a preset characteristic condition, includes one or more under the node
A branch, each branch correspond to a division threshold value/threshold range of the characteristic condition.The leaf node is returns decision
The endpoint node of tree.According to the characteristic condition of present node, associated characteristic information and sample value are obtained, compares each institute
State division threshold value/threshold range of sample value and the characteristic condition.If the sample value meets a certain division threshold value/threshold value model
It encloses, then the corresponding abnormal account number of the sample value or normal account number is divided to the branch.Illustratively, it is assumed that there are the age
Characteristic information, the age sample value of abnormal account number a is 10 years old, and the age sample value of normal account number b is 30 years old, if present node
Characteristic condition is the age, and division threshold range is more than 20 years old and less than or equal to 20 years old, if sample value meets condition " division threshold
Value is ranging from more than 20 years old ", then distribute the corresponding abnormal account number of the sample value or normal account number to the right branch to the node,
Otherwise it distributes to the left branch under the node;Therefore, the exception account number a is distributed to the left branch of the node, the normal account
Number b is distributed to the right branch of the node.Lower level node similarly, until each abnormal account number and normal account number fall on a certain leaf section
Point on.
In step S1023, loss function is obtained, initializes the constant value of loss function minimization.
In step S1024, for each leaf node, according to the loss function and constant value estimation are each
The corresponding residual error approximation of abnormal account number, normal account number.
Optionally, the loss function includes but not limited to absolute error loss function, Huber loss functions, square differential loss
Lose function.
By taking difference of two squares loss function as an example, what each regression tree learnt is the conclusion and residual error of all trees before,
Obtain current residual error regression tree.Initialization makes the constant value of loss function minimization, for difference of two squares loss function, constant value
It is 0, the calculation formula of residual error is residual error=actual value-predicted value, and predicted value is the average value of sample.If statistics " division threshold value
Ranging from it is more than 20 years old " abnormal account number included by the corresponding leaf node of branch and normal account number obtain number N, N number of year altogether
Age sample asks for the average value of the sum of described N number of age sample and N number of age sample.Then it is directed to the leaf node
In each abnormal account number or normal account number, ask for the difference between corresponding age sample and average value, obtain described
The residual error of age sample.As previously described, it is assumed that " division threshold range is more than 20 years old " corresponding leaf node A of branch includes
Abnormal account number and normal account number totally 4, age sample difference A1, A2, A3, A4, ask for the sum of 4 age samples Asum
The average value A of=A1+A2+A3+A4 and 4 age samplesaverage=Asum/4.Then in the leaf node
The age sample A1 asks for the age sample A1 and average value AaverageBetween difference, obtain the age sample A1
Residual error A1residual=A1-Aaverage, and so on, the residual error A2 of the age sample A2residual=A2-Aaverage, it is described
The residual error A3 of age sample A3residual=A3-Aaverage, the residual error A4 of the age sample A4residual=A4-Aaverage.It is false
Abnormal account number and normal account number that " if division threshold range is less than or equal to 20 years old " corresponding leaf node B of branch includes are total to
3, age sample is respectively B1, B2, B3, asks for the sum of 3 age samples Bsum=B1+B2+B3 and 3 ages
The average value of sample, Baverage=Bsum/3.Then for the age sample B 1 in the leaf node, the year is asked for
Age sample B 1 and average value BaverageBetween difference, obtain the residual error B1 of the age sample B 1residual=B1-Baverage,
And so on, the residual error B2 of the age sample B 2residual=B2-Baverage, the residual error B3 of the age sample B 3residual=
B3-Baverage。
In step S1025, based on all lower regression trees of residual error approximation repetitive exercise.
After upper one tree trains to obtain residual error approximation, the sample value that these residual error approximations substitute former account number is held,
It goes to learn to next tree, a new decision regression tree is established on the gradient direction of residual error reduction.
Illustratively, as previously mentioned, the residual error of 4 age samples is respectively A1 in leaf node Aresidual、A2residual、
A3residual、A4residual;The residual error of 3 age samples is respectively B1 in leaf node Bresidual、B2residual、B3residual;
Then it is based on the A1residual、A2residual、A3residual、A4residual、B1residual、B2residual、B3residualIt is next to train
Regression tree.And so on, carry out successive ignition training.
Optionally, in embodiments of the present invention, for life insurance class APP, decision tree GBDT algorithms are promoted to institute based on gradient
State the exception obtained after the normal account number in abnormal account number and white list sample database in blacklist sample database is trained
Account number characteristic of division includes but not limited to:
Continuous high frequency binding feature, continuous cipher binding feature, continuous IP binding feature, IP high diverging rates feature, account number
The same regional feature of business personnel, registration binding time difference feature, registration binding bury point feature without front end.
As it can be seen that the residual error of regression tree is past before the foundation of every regression tree is provided to cause in GBDT algorithms
Gradient direction is reduced, so as to more accurately filter out useful spy from blacklist sample database and white list sample database
Sign, rejects useless feature, improves the accuracy of Feature Selection, reduce characteristic dimension in risk score model construction process and
Sum.
Further, it on the basis of the first realization flow of the construction method based on Fig. 1 risk score models provided, carries
The third for going out the construction method of risk score model provided in an embodiment of the present invention realizes flow.
As shown in figure 3, be the construction method of risk score model provided in an embodiment of the present invention third realize flow show
It is intended to.In embodiments of the present invention, the step S103 be based on random forest RF algorithms to the abnormal account number characteristic of division into
Row training obtains each corresponding contribution degree of exception account number characteristic of division and includes:
In step S1031, the abnormal account number characteristic of division is carried out by predetermined number by random forest RF models
Arbitrary combination, obtains multiple categorised decision trees.
It is assumed herein that the exception account number characteristic of division has M, the random forest RF models use bootstrap
(bootstrap) resampling technique repeats to randomly select k sample from described M abnormal account number characteristic of division with putting back to
New training sample set is generated, it is then random gloomy into m categorised decision tree composition according to each training sample set symphysis
Woods.
In step S1032, the abnormal account number characteristic of division not being combined in each categorised decision tree is obtained, is based on
The exception account number characteristic of division estimates the error in classification of the categorised decision tree.
Herein, random forest RF algorithms merge m categorised decision tree, and the foundation of each tree depends on one
The sample of independent draws, each tree in forest have identical distribution, and error in classification depends on the classification capacity per one tree
And the correlation between them.
When randomly selecting boostrap training sample k from original M abnormal account number characteristic of division in step S1031, also
There is (M-k) a abnormal account number characteristic of division not to be selected, this part of not selected abnormal account number characteristic of division can be used for estimating
Count the error in classification of single categorised decision tree.
In step S1033, the categorised decision tree of error in classification minimum is chosen, and calculate often based on the categorised decision tree
The contribution degree of one abnormal account number characteristic of division.
Wherein, error in classification is smaller, and random forest classification accuracy is higher.The embodiment of the present invention is by comparing random forest
In all categorised decision trees error in classification, obtain error in classification minimum one tree, be then based on the error in classification minimum
Categorised decision tree comes out the contribution degree of each abnormal account number characteristic of division.
The embodiment of the present invention is to pass through using random forest RF algorithms in the weights for calculating abnormal account number characteristic of division
Compare the Decision Tree Algorithms such as id3, cart and logic and return what is chosen after the computational efficiency and precision of sorting algorithm, in speed
RF algorithms are to the data operation speed of input on degree, and error is relatively low from the classifying quality of test set in precision,
It is effectively improved the accuracy of the contribution degree of abnormal account number characteristic of division.
Optionally, the risk that the step S104 is built according to the abnormal account number characteristic of division and its corresponding contribution degree
In Rating Model, including risk score formula:
F (X)=F0+β1T1(X)+β2T2(X)+……++βmTm(X)
Wherein, risk scores of the F (X) for user X, F0For regression residuals item, Ti(X) spy for abnormal account number characteristic of division
Levy combinatorial formula, βiIt is characterized combinatorial formula Ti(X) the sum of contribution degree of each exception account number characteristic of division in, i are positive integer, 1≤
i≤m。
Illustratively, it is assumed that the exception account number characteristic of division has 5, respectively x1、x2、x3、x4、x5;It is described random gloomy
Woods RF models use bootstrap resampling technique, repeat to randomly select 3 sample lifes from described 5 abnormal account number characteristic of division
The training sample set of Cheng Xin.In first time samples, it is assumed that randomly selected x1、x2、x3Totally 3 abnormal account number characteristic of division,
Categorised decision tree is formed, calculates the error in classification of the categorised decision tree, the exception account number characteristic of division x1、x2、x3Then form spy
Levy combinatorial formula T1(X);In secondary sample, it is assumed that randomly selected x1、x2、x4Totally 3 abnormal account number characteristic of division, group
Constituent class decision tree calculates the error in classification of the categorised decision tree, the exception account number characteristic of division x1、x2、x4Then composition characteristic
Combinatorial formula T2(X) ... and so on, m sampling is carried out, m categorised decision tree composition random forest is generated, obtains m spy
Levy combinatorial formula Ti(X).Each abnormal account number characteristic of division x is calculated according to the categorised decision tree of error in classification minimum1、x2、
x3、x4、x5Contribution degree, to feature combinatorial formula Ti(X) contribution degree of each exception account number characteristic of division is summed to obtain β ini,
Such as β1It is characterized combinatorial formula T1(X) abnormal account number characteristic of division x in1、x2、x3The sum of corresponding contribution degree, β2For spy
Levy combinatorial formula T2(X) abnormal account number characteristic of division x in1、x2、x4The sum of corresponding contribution degree ... is last, according to described
The sum of the contribution degree degree of feature combinatorial formula and its corresponding abnormal account number characteristic of division structure risk score model, obtains risk
Score formula:
F (X)=F0+β1T1(X)+β2T2(X)+……++βmTm(X)。
The embodiment of the present invention passes through the sum of contribution degree degree of feature combinatorial formula and its corresponding abnormal account number characteristic of division
Risk score model is built, the performance of risk score model is effectively improved, improves and use the risk score model evaluation
Accuracy when as a result.
Further, it on the basis of the first realization flow of the construction method based on Fig. 1 risk score models provided, carries
Go out the construction method of risk score model provided in an embodiment of the present invention the 4th realizes flow.
As shown in figure 4, be the construction method of risk score model provided in an embodiment of the present invention the 4th realization flow show
It is intended to.In embodiments of the present invention, the structure risk score model, further includes later:
In step S105, each normal account in the white list sample database is calculated using the risk score model
Number corresponding risk score.
After risk score model is obtained, the embodiment of the present invention also based on the risk score model, calculates white list sample
The corresponding risk score of each normal account number in this library, to test normal account number existing in white list sample database
Card.
In step s 106, normal account number of the risk score greater than or equal to preset risk threshold value is filtered, retains wind
Danger scoring is less than the normal account number of preset risk threshold value.
Each normal account number can obtain a wind after the risk score formula in the scoring Rating Model calculates
Danger scoring.The risk score of each normal account number is compared the embodiment of the present invention with preset risk threshold value, obtains
It is associated greater than or equal to the normal account number of preset risk threshold value, and by it with tying up the white lists features such as card, purchase declaration form,
To filter out the noise in white list sample database, the degree of purity of white list sample database is improved.
In step s 107, each abnormal account in the blacklist sample database is calculated using the risk score model
Number corresponding risk score.
In step S108, according to the risk score of the normal account number retained and the risk score of abnormal account number, correction
The risk threshold value.
After normal account number risk score corresponding with abnormal account number is obtained, structure scoring distribution map, and calculate normal account
Number corresponding risk score mean value risk score mean value corresponding with abnormal account number.Scoring is excavated based on the scoring distribution map
Phenomenon is clustered, then in conjunction with the corresponding risk score mean value of normal account number risk score mean value corresponding with abnormal account number, correction
The risk threshold value.
Illustratively, it is assumed that in correction course, the risk score mean value that normal account number is calculated is 0.0825, abnormal account
Number risk score mean value for 0.5602, risk score distribution map is as shown in figure 5, by comparing abnormal account number and normal account number
Risk score can be found that apparent Clustering Effect, and normal account number is more biased towards being gathered in 0~0.05 score value region, and abnormal account number is more
Concentrate on 0.53~0.59 score value region.Based on error in classification is reduced, more abnormal account number is marked off as possible, it can be by risk
Threshold value is determined as 0.5.Empirical tests, when risk threshold value is determined as 0.5, normal account number divides accuracy rate with abnormal account number and is respectively
98.4% and 96.7%.
In conclusion the embodiment of the present invention is come further by using existing blacklist sample database and white list sample database
The division threshold value of normal account number and abnormal account number is corrected, abnormal account number can be utmostly being found out and reduce to normal account number
Identification error further improves the accuracy of risk score Model Identification exception account number.
It should be understood that in the above-described embodiments, the size of the serial number of each step is not meant to the priority of execution sequence, it is each to walk
Rapid execution sequence should determine that the implementation process without coping with the embodiment of the present invention forms any limit with its function and internal logic
It is fixed.
Embodiment 2
Fig. 6 shows the composition structure chart of the construction device of risk score model provided in an embodiment of the present invention, in order to just
In explanation, illustrate only and the relevant part of the embodiment of the present invention.
In embodiments of the present invention, the construction device of the risk score model is used to implement above-mentioned Fig. 1, Fig. 2, Fig. 3, figure
The construction method of risk score model described in 4 embodiments, can be the software unit for being built in terminal, hardware cell or
The unit of software and hardware combining.
Refering to Fig. 6, the construction device of the risk score model includes:
Sample database builds module 61, for building blacklist sample database and white list sample according to preset Account Data
Library, the blacklist sample database include abnormal account number, and the white list sample database includes normal account number;
Feature Selection module 62, for promoting decision tree GBDT algorithms to different in the blacklist sample database based on gradient
Normal account number in normal account number and white list sample database carries out cluster training, filters out abnormal account number characteristic of division;
Contribution degree training module 63 instructs the abnormal account number characteristic of division for being based on random forest RF algorithms
Practice, obtain each corresponding contribution degree of exception account number characteristic of division;
Model construction module 64, for according to the abnormal account number characteristic of division and its corresponding contribution degree, building risk
Rating Model.
Optionally, the construction device further includes:
First grading module 65, for calculating each in the white list sample database using the risk score model
The corresponding risk score of normal account number;
Account number filtering module 66, for normal account number of the risk score greater than or equal to preset risk threshold value to be filtered,
Retain the normal account number that risk score is less than preset risk threshold value;
Second grading module 67, for calculating each in the blacklist sample database using the risk score model
The corresponding risk score of abnormal account number;
Threshold correction module 68, the risk for the risk score according to the normal account number retained and abnormal account number are commented
Point, correct the risk threshold value.
Optionally, the Feature Selection module 62 includes:
Acquiring unit 621, for obtaining each abnormal account number, the corresponding sample value of characteristic information of normal account number;
Account number training unit 622, for according to the corresponding sample value of the characteristic information, being incited somebody to action according to the characteristic condition of prediction
The exception account number, normal account number distribute the first regression tree into GBDT models, until each abnormal account number, normal account
Number divide equally be assigned to each leaf node;
Residual error approximation evaluation unit 623 for obtaining loss function, initializes the constant value of loss function minimization,
For each leaf node, corresponded to according to each abnormal account number of the loss function and constant value estimation, normal account number
Residual error approximation;
The account number training unit 622 is additionally operable to, based on all lower regression trees of residual error approximation repetitive exercise.
Optionally, the abnormal account number characteristic of division includes but not limited to:
Continuous high frequency binding feature, continuous cipher binding feature, continuous IP binding feature, IP high diverging rates feature, account number
The same regional feature of business personnel, registration binding time difference feature, registration binding bury point feature without front end.
Optionally, the contribution degree training module 63 further includes:
Feature training unit 631, for passing through random forest RF models to the abnormal account number characteristic of division by default
Number carries out arbitrary combination, obtains multiple categorised decision trees;
Error in classification evaluation unit 632, for obtaining the abnormal account number not being combined in each categorised decision tree point
Category feature estimates the error in classification of the categorised decision tree based on the abnormal account number characteristic of division;
Contribution degree computing unit 633, for choosing the categorised decision tree of error in classification minimum, and based on the categorised decision
Tree calculates the contribution degree of each abnormal account number characteristic of division.
Optionally, the structure risk score model includes structure risk score formula:
F (X)=F0+β1T1(X)+β2T2(X)+……++βmTm(X)
Wherein, risk scores of the F (X) for user account number X, F0For regression residuals item, Ti(X) it is abnormal account number characteristic of division
Feature combinatorial formula, βiIt is characterized combinatorial formula Ti(X) the sum of contribution degree of each exception account number characteristic of division in.
It should be noted that each module/unit in the embodiment of the present invention can be used to implement in above method embodiment
Whole technical solutions, specific work process can refer to the corresponding process in preceding method embodiment, no longer superfluous herein
It states.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
Embodiment 3
The present embodiment provides a computer readable storage medium, computer journey is stored on the computer readable storage medium
Sequence, the construction method of realization 1 risk Rating Model of embodiment, repeats to avoid when which is executed by processor,
Which is not described herein again.Alternatively, the structure dress of 2 risk Rating Model of embodiment is realized when the computer program is executed by processor
The function of each module/unit, repeats, which is not described herein again to avoid in putting.
Embodiment 4
Fig. 7 is a kind of schematic diagram of terminal provided in an embodiment of the present invention, and the terminal includes but not limited to server, moves
Dynamic terminal.As shown in fig. 7, the terminal 7 of the embodiment includes:Processor 70, memory 71 and it is stored in the memory 71
In and the computer program 72 that can be run on the processor 70.The processor 70 performs real during the computer program 72
Step in the construction method embodiment of existing above-mentioned risk score model, such as step S101 to S104 shown in FIG. 1, Fig. 2 are real
The step S1021 to S1025 described in example is applied, in the step S1031 to S1033 and Fig. 4 embodiments described in Fig. 3 embodiments
The step S105 to step S108;Alternatively, the processor 70 realizes above-mentioned risk when performing the computer program 72
The function of each module/unit in the construction device embodiment of Rating Model, such as the function of module 61 to 68 shown in Fig. 6.
Illustratively, the computer program 72 can be divided into one or more module/units, it is one or
Multiple module/units are stored in the memory 71, and are performed by the processor 70, to complete the present invention.Described one
A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, which is used for
Implementation procedure of the computer program 72 in the terminal 7 is described.For example, the computer program 72 can be divided into
Sample database structure module, Feature Selection module, contribution degree training module, model construction module, each module concrete function are as follows:
Sample database builds module, for building blacklist sample database and white list sample database according to preset Account Data,
The blacklist sample database includes abnormal account number, and the white list sample database includes normal account number;
Feature Selection module, for promoting decision tree GBDT algorithms to the exception in the blacklist sample database based on gradient
Normal account number in account number and white list sample database carries out cluster training, filters out abnormal account number characteristic of division;
Contribution degree training module is trained the abnormal account number characteristic of division for being based on random forest RF algorithms,
Obtain each corresponding contribution degree of exception account number characteristic of division;
Model construction module, for according to the abnormal account number characteristic of division and its corresponding contribution degree, structure risk to be commented
Sub-model.
The terminal 7 can be the computing devices such as desktop PC, notebook, palm PC and cloud server.Institute
Stating terminal may include, but be not limited only to, processor 70, memory 71.It will be understood by those skilled in the art that Fig. 7 is only eventually
The restriction of the example, not structure paired terminal 7 at end 7 can be included than illustrating more or fewer components or combining certain portions
Part or different components, such as the terminal can also include input-output equipment, network access equipment, bus etc..
Alleged processor 70 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor can also be any conventional processor
Deng the processor is the control centre of the terminal, utilizes various interfaces and the various pieces of the entire terminal of connection.
The memory 71 can be used for storing the computer program and/or module, and the processor is by running or holding
The computer program and/or module and call the data being stored in memory that row is stored in the memory, realize institute
State the various functions of terminal.The memory can mainly include storing program area and storage data field, wherein, storing program area can
Application program (such as sound-playing function, image player function etc.) needed for storage program area, at least one function etc.;It deposits
Storage data field can be stored uses created data etc. according to terminal.In addition, memory can be deposited including high random access
Reservoir can also include nonvolatile memory, such as hard disk, memory, plug-in type hard disk, intelligent memory card (Smart Media
Card, SMC), safe digital card (Secure Digital, SD), flash card (Flash Card), at least one magnetic disk storage
Part, flush memory device or other volatile solid-state parts.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it can also
That each unit is individually physically present, can also two or more units integrate in a unit.Above-mentioned integrated list
The form that hardware had both may be used in member is realized, can also be realized in the form of SFU software functional unit.
If the integrated module/unit realized in the form of SFU software functional unit and be independent product sale or
In use, it can be stored in a computer readable storage medium.Based on such understanding, the present invention realizes above-described embodiment
All or part of flow in method can also instruct relevant hardware to complete, the calculating by computer program
Machine program can be stored in a computer readable storage medium, and the computer program is when being executed by processor, it can be achieved that above-mentioned
The step of each embodiment of the method.Wherein, the computer program includes computer program code, the computer program code
Can be source code form, object identification code form, executable file or certain intermediate forms etc..The computer-readable storage medium
Matter can include:Can carry the computer program code any entity or device, recording medium, USB flash disk, mobile hard disk,
Magnetic disc, CD, computer storage, read-only memory (ROM, Read-Only Memory), random access memory (RAM,
Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..It is it should be noted that described
It is appropriate that the content that computer readable storage medium includes can be carried out according to legislation in jurisdiction and the requirement of patent practice
Increase and decrease, such as in certain jurisdictions, according to legislation and patent practice, it is electric load that computer readable storage medium, which does not include,
Wave signal and telecommunication signal.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations;Although with reference to aforementioned reality
Example is applied the present invention is described in detail, it will be understood by those of ordinary skill in the art that:It still can be to aforementioned each
Technical solution recorded in embodiment modifies or carries out equivalent replacement to which part technical characteristic;And these are changed
Or replace, the spirit and scope for various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of construction method of risk score model, which is characterized in that the construction method includes:
Blacklist sample database is built according to preset Account Data and white list sample database, the blacklist sample database include different
Normal account number, the white list sample database include normal account number;
Decision tree GBDT algorithms are promoted in the abnormal account number and white list sample database in the blacklist sample database based on gradient
Normal account number carry out cluster training, filter out abnormal account number characteristic of division;
The abnormal account number characteristic of division is trained based on random forest RF algorithms, it is special to obtain each abnormal account number classification
Levy corresponding contribution degree;
According to the abnormal account number characteristic of division and its corresponding contribution degree, risk score model, the risk score mould are built
Type is used to identify abnormal account number.
2. the construction method of risk score model as described in claim 1, which is characterized in that the structure risk score mould
Type further includes later:
The corresponding risk score of each normal account number in the white list sample database is calculated using the risk score model;
Normal account number of the risk score greater than or equal to preset risk threshold value is filtered, retains risk score and is less than preset wind
The normal account number of dangerous threshold value;
Each corresponding risk score of exception account number in the blacklist sample database is calculated using the risk score model;
According to the risk score of the normal account number retained and the risk score of abnormal account number, the risk threshold value is corrected.
3. the construction method of risk score model as claimed in claim 1 or 2, which is characterized in that described to be promoted based on gradient
Decision tree GBDT algorithms gather the normal account number in the abnormal account number and white list sample database in the blacklist sample database
Class training includes:
Obtain each abnormal account number, the corresponding sample value of characteristic information of normal account number;
According to the corresponding sample value of the characteristic information, the abnormal account number, normal account number are divided according to the characteristic condition of prediction
First regression tree being assigned in GBDT models, until each abnormal account number, normal account number, which are divided equally, is assigned to each leaf section
Point;
Loss function is obtained, initializes the constant value of loss function minimization;
For each leaf node, according to each abnormal account number of the loss function and constant value estimation, normal account number
Corresponding residual error approximation;
Based on all lower regression trees of residual error approximation repetitive exercise.
4. the construction method of risk score model as claimed in claim 3, which is characterized in that the exception account number characteristic of division
Including:
Continuous high frequency binding feature, continuous cipher binding feature, continuous IP binding feature, IP high diverging rates feature, account number business
The same regional feature of member, registration binding time difference feature, registration binding bury point feature without front end.
5. the construction method of risk score model as claimed in claim 1 or 2, which is characterized in that described to be based on random forest
RF algorithms are trained the abnormal account number characteristic of division, obtain each corresponding contribution degree packet of exception account number characteristic of division
It includes:
Arbitrary combination is carried out by predetermined number to the abnormal account number characteristic of division by random forest RF models, obtains multiple points
Class decision tree;
The abnormal account number characteristic of division not being combined in each categorised decision tree is obtained, it is special based on the abnormal account number classification
Sign estimates the error in classification of the categorised decision tree;
The categorised decision tree of error in classification minimum is chosen, and each abnormal account number classification is special based on categorised decision tree calculating
The contribution degree of sign.
6. the construction method of risk score model as claimed in claim 1 or 2, which is characterized in that the structure risk score
Model includes structure risk score formula:
F (X)=F0+β1T1(X)+β2T2(X)+……++βmTm(X)
Wherein, risk scores of the F (X) for user account number X, F0For regression residuals item, Ti(X) spy for abnormal account number characteristic of division
Levy combinatorial formula, βiIt is characterized combinatorial formula Ti(X) the sum of contribution degree of each exception account number characteristic of division in.
7. a kind of construction device of risk score model, which is characterized in that the construction device includes:
Sample database builds module, described for building blacklist sample database and white list sample database according to preset Account Data
Blacklist sample database includes abnormal account number, and the white list sample database includes normal account number;
Feature Selection module, for promoting decision tree GBDT algorithms to the abnormal account number in the blacklist sample database based on gradient
Cluster training is carried out with the normal account number in white list sample database, filters out abnormal account number characteristic of division;
Contribution degree training module is trained the abnormal account number characteristic of division for being based on random forest RF algorithms, obtains
Each corresponding contribution degree of exception account number characteristic of division;
Model construction module, for according to the abnormal account number characteristic of division and its corresponding contribution degree, building risk score mould
Type, the risk score model are used to identify abnormal account number.
8. the construction device of risk score model as claimed in claim 7, which is characterized in that the construction device further includes:
First grading module, for calculating each normal account in the white list sample database using the risk score model
Number corresponding risk score;
Account number filtering module for normal account number of the risk score greater than or equal to preset risk threshold value to be filtered, retains wind
Danger scoring is less than the normal account number of preset risk threshold value;
Second grading module, for calculating each abnormal account in the blacklist sample database using the risk score model
Number corresponding risk score;
Threshold correction module, for the risk score according to the normal account number retained and the risk score of abnormal account number, correction
The risk threshold value.
9. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is held by processor
The step described in the construction method of claim 1 to 6 any one of them risk score model is realized during row.
10. a kind of terminal, what the terminal included memory, processor and storage on a memory and can run on a processor
Computer program, which is characterized in that the processor realizes such as the claims 1 to 6 when performing the computer program
Step described in the construction method of risk score model described in one.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810030179.3A CN108269012A (en) | 2018-01-12 | 2018-01-12 | Construction method, device, storage medium and the terminal of risk score model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810030179.3A CN108269012A (en) | 2018-01-12 | 2018-01-12 | Construction method, device, storage medium and the terminal of risk score model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108269012A true CN108269012A (en) | 2018-07-10 |
Family
ID=62775496
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810030179.3A Pending CN108269012A (en) | 2018-01-12 | 2018-01-12 | Construction method, device, storage medium and the terminal of risk score model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108269012A (en) |
Cited By (29)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109064343A (en) * | 2018-08-13 | 2018-12-21 | 中国平安人寿保险股份有限公司 | Risk model method for building up, risk matching process, device, equipment and medium |
CN109167753A (en) * | 2018-07-23 | 2019-01-08 | 中国科学院计算机网络信息中心 | A kind of detection method and device of network intrusions flow |
CN109257354A (en) * | 2018-09-25 | 2019-01-22 | 平安科技(深圳)有限公司 | Abnormal flow analysis method and device, electronic equipment based on model tree algorithm |
CN109284380A (en) * | 2018-09-25 | 2019-01-29 | 平安科技(深圳)有限公司 | Illegal user's recognition methods and device, electronic equipment based on big data analysis |
CN109523412A (en) * | 2018-11-14 | 2019-03-26 | 平安科技(深圳)有限公司 | Intelligent core protects method, apparatus, computer equipment and computer readable storage medium |
CN109544351A (en) * | 2018-10-12 | 2019-03-29 | 平安科技(深圳)有限公司 | Vehicle risk appraisal procedure, device, computer equipment and storage medium |
CN109639633A (en) * | 2018-11-02 | 2019-04-16 | 平安科技(深圳)有限公司 | Abnormal flow data identification method, device, medium and electronic equipment |
CN109685635A (en) * | 2018-09-11 | 2019-04-26 | 深圳平安财富宝投资咨询有限公司 | Methods of risk assessment, air control server-side and the storage medium of financial business |
CN109754157A (en) * | 2018-11-30 | 2019-05-14 | 畅捷通信息技术股份有限公司 | A kind of methods of marking and system for reflecting enterprise's health management, financing and increasing letter |
CN109801151A (en) * | 2019-01-07 | 2019-05-24 | 平安科技(深圳)有限公司 | Financial fraud risk monitoring and control method, apparatus, computer equipment and storage medium |
CN109871954A (en) * | 2018-12-24 | 2019-06-11 | 腾讯科技(深圳)有限公司 | Training sample generation method, method for detecting abnormality and device |
CN110009170A (en) * | 2018-11-05 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Model score bearing calibration, device and server |
CN110231156A (en) * | 2019-06-26 | 2019-09-13 | 山东大学 | Service robot kinematic system method for diagnosing faults and device based on temporal aspect |
CN110348580A (en) * | 2019-06-18 | 2019-10-18 | 第四范式(北京)技术有限公司 | Construct the method, apparatus and prediction technique, device of GBDT model |
CN110503566A (en) * | 2019-07-08 | 2019-11-26 | 中国平安人寿保险股份有限公司 | Air control method for establishing model, device, computer equipment and storage medium |
CN110516910A (en) * | 2019-07-23 | 2019-11-29 | 平安科技(深圳)有限公司 | Declaration form core based on big data protects model training method and core protects methods of risk assessment |
WO2020015089A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技(深圳)有限公司 | Identity information risk assessment method and apparatus, and computer device and storage medium |
WO2020015140A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技(深圳)有限公司 | Passenger rating model generation method and apparatus, and computer device and storage medium |
CN110766040A (en) * | 2019-09-03 | 2020-02-07 | 阿里巴巴集团控股有限公司 | Method and device for risk clustering of transaction risk data |
CN111538908A (en) * | 2020-06-22 | 2020-08-14 | 腾讯科技(深圳)有限公司 | Search ranking method and device, computer equipment and storage medium |
CN111563810A (en) * | 2020-04-28 | 2020-08-21 | 北京云从科技有限公司 | Credit wind control model generation method, credit evaluation system, machine-readable medium and device |
WO2020199692A1 (en) * | 2019-04-04 | 2020-10-08 | 中国科学院深圳先进技术研究院 | Method and apparatus for screening predictive image features for cancer metastasis, and storage medium |
CN111898675A (en) * | 2020-07-30 | 2020-11-06 | 北京云从科技有限公司 | Credit wind control model generation method and device, scoring card generation method, machine readable medium and equipment |
CN112070225A (en) * | 2020-09-01 | 2020-12-11 | 多点(深圳)数字科技有限公司 | Entity card abnormal binding alarm method based on unsupervised learning |
CN112118259A (en) * | 2020-09-17 | 2020-12-22 | 四川长虹电器股份有限公司 | Unauthorized vulnerability detection method based on classification model of lifting tree |
WO2021000572A1 (en) * | 2019-07-01 | 2021-01-07 | 创新先进技术有限公司 | Data processing method and apparatus, and electronic device |
CN113315790A (en) * | 2021-07-29 | 2021-08-27 | 湖南华菱电子商务有限公司 | Intrusion flow detection method, electronic device and storage medium |
CN113488187A (en) * | 2021-08-03 | 2021-10-08 | 南通市第二人民医院 | Anesthesia accident case collecting and analyzing method and system |
WO2021232588A1 (en) * | 2020-05-21 | 2021-11-25 | 平安国际智慧城市科技股份有限公司 | Food safety risk assessment method, apparatus, device, and storage medium |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105069470A (en) * | 2015-07-29 | 2015-11-18 | 腾讯科技(深圳)有限公司 | Classification model training method and device |
CN106778603A (en) * | 2016-12-13 | 2017-05-31 | 中国科学院信息工程研究所 | A kind of pedestrian recognition method that SVM classifier is cascaded based on gradient type |
CN107122890A (en) * | 2017-04-07 | 2017-09-01 | 信雅达系统工程股份有限公司 | Personal credit score optimizing evaluation method |
CN107220732A (en) * | 2017-05-31 | 2017-09-29 | 福州大学 | A kind of power failure complaint risk Forecasting Methodology based on gradient boosted tree |
CN107480839A (en) * | 2017-10-13 | 2017-12-15 | 深圳市博安达信息技术股份有限公司 | The classification Forecasting Methodology of high-risk pollution sources based on principal component analysis and random forest |
CN108108743A (en) * | 2016-11-24 | 2018-06-01 | 百度在线网络技术(北京)有限公司 | Abnormal user recognition methods and the device for identifying abnormal user |
-
2018
- 2018-01-12 CN CN201810030179.3A patent/CN108269012A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105069470A (en) * | 2015-07-29 | 2015-11-18 | 腾讯科技(深圳)有限公司 | Classification model training method and device |
CN108108743A (en) * | 2016-11-24 | 2018-06-01 | 百度在线网络技术(北京)有限公司 | Abnormal user recognition methods and the device for identifying abnormal user |
CN106778603A (en) * | 2016-12-13 | 2017-05-31 | 中国科学院信息工程研究所 | A kind of pedestrian recognition method that SVM classifier is cascaded based on gradient type |
CN107122890A (en) * | 2017-04-07 | 2017-09-01 | 信雅达系统工程股份有限公司 | Personal credit score optimizing evaluation method |
CN107220732A (en) * | 2017-05-31 | 2017-09-29 | 福州大学 | A kind of power failure complaint risk Forecasting Methodology based on gradient boosted tree |
CN107480839A (en) * | 2017-10-13 | 2017-12-15 | 深圳市博安达信息技术股份有限公司 | The classification Forecasting Methodology of high-risk pollution sources based on principal component analysis and random forest |
Non-Patent Citations (2)
Title |
---|
中国环境监测总站: "《环境空气质量预报预警方法技术指南》", 30 June 2017 * |
田茂再著: "《分位回归与复杂分层结构数据分析》", 30 June 2015 * |
Cited By (46)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020015140A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技(深圳)有限公司 | Passenger rating model generation method and apparatus, and computer device and storage medium |
WO2020015089A1 (en) * | 2018-07-18 | 2020-01-23 | 平安科技(深圳)有限公司 | Identity information risk assessment method and apparatus, and computer device and storage medium |
CN109167753A (en) * | 2018-07-23 | 2019-01-08 | 中国科学院计算机网络信息中心 | A kind of detection method and device of network intrusions flow |
CN109064343B (en) * | 2018-08-13 | 2023-09-26 | 中国平安人寿保险股份有限公司 | Risk model building method, risk matching device, risk model building equipment and risk matching medium |
CN109064343A (en) * | 2018-08-13 | 2018-12-21 | 中国平安人寿保险股份有限公司 | Risk model method for building up, risk matching process, device, equipment and medium |
CN109685635A (en) * | 2018-09-11 | 2019-04-26 | 深圳平安财富宝投资咨询有限公司 | Methods of risk assessment, air control server-side and the storage medium of financial business |
CN109257354A (en) * | 2018-09-25 | 2019-01-22 | 平安科技(深圳)有限公司 | Abnormal flow analysis method and device, electronic equipment based on model tree algorithm |
CN109284380A (en) * | 2018-09-25 | 2019-01-29 | 平安科技(深圳)有限公司 | Illegal user's recognition methods and device, electronic equipment based on big data analysis |
CN109257354B (en) * | 2018-09-25 | 2021-11-12 | 平安科技(深圳)有限公司 | Abnormal flow analysis method and device based on model tree algorithm and electronic equipment |
WO2020062803A1 (en) * | 2018-09-25 | 2020-04-02 | 平安科技(深圳)有限公司 | Abnormal traffic analysis method and apparatus based on model tree algorithm, and electronic device and non-volatile readable storage medium |
CN109284380B (en) * | 2018-09-25 | 2023-04-25 | 平安科技(深圳)有限公司 | Illegal user identification method and device based on big data analysis and electronic equipment |
CN109544351A (en) * | 2018-10-12 | 2019-03-29 | 平安科技(深圳)有限公司 | Vehicle risk appraisal procedure, device, computer equipment and storage medium |
CN109544351B (en) * | 2018-10-12 | 2024-05-07 | 平安科技(深圳)有限公司 | Vehicle risk assessment method, device, computer equipment and storage medium |
CN109639633A (en) * | 2018-11-02 | 2019-04-16 | 平安科技(深圳)有限公司 | Abnormal flow data identification method, device, medium and electronic equipment |
CN109639633B (en) * | 2018-11-02 | 2021-11-12 | 平安科技(深圳)有限公司 | Abnormal flow data identification method, abnormal flow data identification device, abnormal flow data identification medium, and electronic device |
CN110009170A (en) * | 2018-11-05 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Model score bearing calibration, device and server |
CN110009170B (en) * | 2018-11-05 | 2023-04-07 | 蚂蚁金服(杭州)网络技术有限公司 | Model scoring correction method and device and server |
CN109523412A (en) * | 2018-11-14 | 2019-03-26 | 平安科技(深圳)有限公司 | Intelligent core protects method, apparatus, computer equipment and computer readable storage medium |
CN109754157A (en) * | 2018-11-30 | 2019-05-14 | 畅捷通信息技术股份有限公司 | A kind of methods of marking and system for reflecting enterprise's health management, financing and increasing letter |
CN109871954A (en) * | 2018-12-24 | 2019-06-11 | 腾讯科技(深圳)有限公司 | Training sample generation method, method for detecting abnormality and device |
CN109871954B (en) * | 2018-12-24 | 2022-12-02 | 腾讯科技(深圳)有限公司 | Training sample generation method, abnormality detection method and apparatus |
CN109801151B (en) * | 2019-01-07 | 2023-09-05 | 平安科技(深圳)有限公司 | Financial falsification risk monitoring method, device, computer equipment and storage medium |
CN109801151A (en) * | 2019-01-07 | 2019-05-24 | 平安科技(深圳)有限公司 | Financial fraud risk monitoring and control method, apparatus, computer equipment and storage medium |
WO2020199692A1 (en) * | 2019-04-04 | 2020-10-08 | 中国科学院深圳先进技术研究院 | Method and apparatus for screening predictive image features for cancer metastasis, and storage medium |
CN110348580B (en) * | 2019-06-18 | 2022-05-10 | 第四范式(北京)技术有限公司 | Method and device for constructing GBDT model, and prediction method and device |
CN110348580A (en) * | 2019-06-18 | 2019-10-18 | 第四范式(北京)技术有限公司 | Construct the method, apparatus and prediction technique, device of GBDT model |
CN110231156A (en) * | 2019-06-26 | 2019-09-13 | 山东大学 | Service robot kinematic system method for diagnosing faults and device based on temporal aspect |
WO2021000572A1 (en) * | 2019-07-01 | 2021-01-07 | 创新先进技术有限公司 | Data processing method and apparatus, and electronic device |
CN110503566A (en) * | 2019-07-08 | 2019-11-26 | 中国平安人寿保险股份有限公司 | Air control method for establishing model, device, computer equipment and storage medium |
CN110503566B (en) * | 2019-07-08 | 2024-02-09 | 中国平安人寿保险股份有限公司 | Wind control model building method and device, computer equipment and storage medium |
CN110516910B (en) * | 2019-07-23 | 2023-05-26 | 平安科技(深圳)有限公司 | Insurance policy and insurance model training method and insurance risk assessment method based on big data |
CN110516910A (en) * | 2019-07-23 | 2019-11-29 | 平安科技(深圳)有限公司 | Declaration form core based on big data protects model training method and core protects methods of risk assessment |
CN110766040A (en) * | 2019-09-03 | 2020-02-07 | 阿里巴巴集团控股有限公司 | Method and device for risk clustering of transaction risk data |
CN110766040B (en) * | 2019-09-03 | 2024-02-06 | 创新先进技术有限公司 | Method and device for risk clustering of transaction risk data |
CN111563810A (en) * | 2020-04-28 | 2020-08-21 | 北京云从科技有限公司 | Credit wind control model generation method, credit evaluation system, machine-readable medium and device |
WO2021232588A1 (en) * | 2020-05-21 | 2021-11-25 | 平安国际智慧城市科技股份有限公司 | Food safety risk assessment method, apparatus, device, and storage medium |
CN111538908A (en) * | 2020-06-22 | 2020-08-14 | 腾讯科技(深圳)有限公司 | Search ranking method and device, computer equipment and storage medium |
CN111538908B (en) * | 2020-06-22 | 2020-10-20 | 腾讯科技(深圳)有限公司 | Search ranking method and device, computer equipment and storage medium |
CN111898675B (en) * | 2020-07-30 | 2021-04-23 | 北京云从科技有限公司 | Credit wind control model generation method and device, scoring card generation method, machine readable medium and equipment |
CN111898675A (en) * | 2020-07-30 | 2020-11-06 | 北京云从科技有限公司 | Credit wind control model generation method and device, scoring card generation method, machine readable medium and equipment |
CN112070225B (en) * | 2020-09-01 | 2023-10-10 | 多点(深圳)数字科技有限公司 | Entity card abnormal binding alarm method based on unsupervised learning |
CN112070225A (en) * | 2020-09-01 | 2020-12-11 | 多点(深圳)数字科技有限公司 | Entity card abnormal binding alarm method based on unsupervised learning |
CN112118259A (en) * | 2020-09-17 | 2020-12-22 | 四川长虹电器股份有限公司 | Unauthorized vulnerability detection method based on classification model of lifting tree |
CN113315790A (en) * | 2021-07-29 | 2021-08-27 | 湖南华菱电子商务有限公司 | Intrusion flow detection method, electronic device and storage medium |
CN113488187A (en) * | 2021-08-03 | 2021-10-08 | 南通市第二人民医院 | Anesthesia accident case collecting and analyzing method and system |
CN113488187B (en) * | 2021-08-03 | 2024-02-20 | 南通市第二人民医院 | Anesthesia accident case collecting and analyzing method and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108269012A (en) | Construction method, device, storage medium and the terminal of risk score model | |
US11748758B2 (en) | Method for improving operating profits with better automated decision making with artificial intelligence | |
CN108304853B (en) | Game correlation obtaining method and device, storage medium and electronic device | |
CN107689008A (en) | A kind of user insures the method and device of behavior prediction | |
CN103198161B (en) | Microblog water army recognition methods and equipment | |
Kwon et al. | Crime scene reconstruction: Online gold farming network analysis | |
Sifa et al. | Customer lifetime value prediction in non-contractual freemium settings: Chasing high-value users using deep neural networks and SMOTE | |
CN109523412A (en) | Intelligent core protects method, apparatus, computer equipment and computer readable storage medium | |
CN106803039B (en) | A kind of homologous determination method and device of malicious file | |
CN109858919B (en) | Abnormal account number determining method and device, and online ordering method and device | |
CN110166344B (en) | Identity identification method, device and related equipment | |
CN107291672A (en) | The treating method and apparatus of tables of data | |
CN108066987B (en) | Distribution method and device of application scene level combination and server | |
CN105574544A (en) | Data processing method and device | |
Lee et al. | In-game action sequence analysis for game bot detection on the big data analysis platform | |
CN112750030B (en) | Risk pattern recognition method, apparatus, device and computer readable storage medium | |
CN111199474A (en) | Risk prediction method and device based on network diagram data of two parties and electronic equipment | |
CN111428217A (en) | Method and device for identifying cheat group, electronic equipment and computer readable storage medium | |
Jamalian et al. | A hybrid data mining method for customer churn prediction | |
CN107908796A (en) | E-Government duplicate checking method, apparatus and computer-readable recording medium | |
CN108197795A (en) | The account recognition methods of malice group, device, terminal and storage medium | |
CN108596271A (en) | Appraisal procedure, device, storage medium and the terminal of fingerprint developing algorithm | |
Malinowski et al. | An approach to property valuation based on market segmentation with crisp and fuzzy clustering | |
CN108777749A (en) | A kind of fraudulent call recognition methods and device | |
CN111160647B (en) | Money laundering behavior prediction method and device |
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: 20180710 |
|
RJ01 | Rejection of invention patent application after publication |