CN107316205A - Recognize humanized method, device, computer-readable medium and the system of holding - Google Patents
Recognize humanized method, device, computer-readable medium and the system of holding Download PDFInfo
- Publication number
- CN107316205A CN107316205A CN201710395643.4A CN201710395643A CN107316205A CN 107316205 A CN107316205 A CN 107316205A CN 201710395643 A CN201710395643 A CN 201710395643A CN 107316205 A CN107316205 A CN 107316205A
- Authority
- CN
- China
- Prior art keywords
- attribute
- consumption
- holder
- tendency
- default
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Marketing (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
It is a kind of to recognize humanized method, device, computer-readable medium and the system of holding.The recognition methods includes:Obtain the corresponding consumption data of holder;Based on default probability graph model, the acquired corresponding consumption data of holder is handled, identification obtains the consumption attribute of the holder.By default probability graph model, to hold it is humanized be identified, the recognition accuracy for holding humanized and coverage rate can be improved.
Description
Technical field
The present invention relates to data processing field, more particularly to a kind of identification hold humanized method, device, computer can
Read medium and system.
Background technology
With the popularization of ecommerce, it is often necessary to recognize the attribute of holder (for example, sex, year based on consumption data
Age section etc.), and different solutions are provided for the crowd of different attribute, so as to meet the crowd demand of different attribute.
And the scheme of its attribute is directly recognized by the industry belonging to the consumption data of holder at present, although it is simple easy
OK, but there is the problem of accuracy rate is low, coverage rate is low, if especially holder only food and drink etc. have no attribute tendency industry
Consumption, then can not judge the attribute of holder according to consumption data completely.
The content of the invention
Present invention solves the technical problem that being the accuracy rate and coverage rate for how improving holder's Attribute Recognition.
The humanized method in order to solve the above technical problems, a kind of identification of offer of the embodiment of the present invention holds, including:Obtain
The corresponding consumption data of holder;Based on default probability graph model, the acquired corresponding consumption data of holder is carried out
Processing, identification obtains the consumption attribute of the holder.
Alternatively, it is described to be based on default probability graph model, at the acquired corresponding consumption data of holder
Reason, the consumption attribute that identification obtains the holder includes:Selection is default to be had the trade company of tendency attribute and marks tendency attribute,
Iteration performs following steps, until reaching the condition of default stopping iteration:According to the tendency for the trade company that marked tendency attribute
Attribute, marks the consumption attribute of the holder corresponding to it;According to the consumption attribute for the holder that marked consumption attribute, mark
The tendency attribute of the trade company of unmarked tendency attribute corresponding to it.
Alternatively, it is described it is default stop iteration condition comprising it is following any one:Reach maximum iteration, own
The consumption attribute of holder and the tendency attribute of all trade companies are labeled.
Alternatively, the tendency attribute according to the trade company that marked tendency attribute, marks holder's corresponding to it
Attribute is consumed, including:Consumption attribute mark value of the attribute for the corresponding holder of trade company of the first attribute will be inclined to and add 0, will be inclined
Add 1 to consumption attribute mark value of the attribute for the corresponding holder of trade company of the second attribute;Calculate the consumption category of the holder
Sex index, the consumption attribute index of the holder is equal to consumption attribute mark value and the business of the total degree of mark;Held when described
When the consumption attribute index for blocking people is less than default consumption attribute the first thresholding of index, the consumption attribute of the holder is marked
For the first attribute;, will be described when the consumption attribute index of the holder is more than default consumption attribute the second thresholding of index
The consumption attribute of holder is labeled as the second attribute, and the second thresholding of the consumption attribute index is more than consumption attribute index first
Limit.
Alternatively, the attribute according to the holder that marked consumption attribute, marks the unmarked tendency corresponding to it
The tendency attribute of the trade company of attribute includes:Tendency attribute mark of the attribute for the corresponding trade company of holder of the first attribute will be consumed
Value Jia 0, will consume tendency attribute mark value of the attribute for the corresponding trade company of holder of the second attribute and adds 1;Calculate the trade company
Tendency attribute index, the tendency attribute index of the trade company is equal to tendency attribute mark value and the business of the total degree of mark;When
When the tendency attribute index of the trade company is less than default tendency attribute the first thresholding of index, by the tendency attribute mark of the trade company
It is designated as the first attribute;, will be described when the tendency attribute index of the trade company is more than default tendency attribute the second thresholding of index
The tendency attribute of trade company is labeled as the second attribute, and the second thresholding of the default tendency attribute index is more than default tendency attribute
The thresholding of index first.
Alternatively, the tendency attribute according to the trade company that marked tendency attribute, marks holder's corresponding to it
Attribute is consumed, including:Attribute will be inclined to and add 1 for the m consumption attribute mark value of the corresponding holder of trade company of m attributes, directly
To travel through it is all marked tendency attribute the corresponding all holders of trade company, 1≤m≤M, wherein:M is the total individual of tendency attribute
Number, M >=1;M is wherein any one tendency attribute;All consumption attribute indexes of the holder are calculated successively, wherein m
Consume attribute index and be equal to the business that m consumes the total degree of attribute mark value and mark;When the m of the holder consumes attribute
When index is more than default consumption attribute three thresholding of index, the consumption attribute of the holder is labeled as m attributes.
Alternatively, the consumption attribute according to the holder that marked consumption attribute, is marked unmarked corresponding to it
Being inclined to the tendency attribute of the trade company of attribute includes:N-th tendency category of the attribute for the corresponding trade company of holder of the n-th attribute will be consumed
Property mark value add 1, until travel through it is all marked consumption attribute the corresponding all trade companies of holder, 1≤n≤N, wherein:N is
Consume the total number of attribute, N >=1;N is wherein any one consumption attribute;All tendency attributes of the trade company are calculated successively
Index, wherein the n-th tendency attribute index is equal to the n-th tendency attribute mark value and the business of the total degree of mark;When the trade company
When n-th tendency attribute index is more than default tendency attribute three thresholding of index, the tendency attribute of the trade company is labeled as n-th
Attribute.
Alternatively, the identification holds humanized method, in addition to:Belong to according to the consumption of the holder recognized
Property, the holder is grouped, wherein t groups crowd is the crowd labeled as t attributes, 1≤t≤T, T is consumption attribute
Total number, T >=1, t is wherein any one consumption attribute;Consumption data based on holder, calculates any under any dimension
Any feature statistical value of index, until traversal obtains all corresponding to all default indexs under all default dimensions
Default characteristic statisticses value, generates the first consumption data characteristic value file;For every group of people, the first consumption number is based respectively on
According to characteristic value file and machine learning model, the real property of the holder is recognized.
Alternatively, the method for building up of the machine learning model includes:Obtain the holder for clearly marking real property
Sample consumption data;Based on the sample consumption data for the holder for clearly marking real property, calculate under any dimension
Any feature statistical value of any index, until traversal is obtained corresponding to all default indexs under all default dimensions
All default characteristic statisticses values, generate the second consumption data characteristic value file;To in the second consumption data characteristic value file
Characteristic value is filtered, and generates the second consumption data validity feature value file;Based on the second consumption data validity feature value file,
Using machine learning algorithm, machine learning model is set up.
Alternatively, the characteristic value in the second consumption data characteristic value file is filtered, including:First order mistake
Filter, removes useless and multiple eigenvalue;The second level is filtered, and removes onrelevant characteristic value.
Alternatively, the first order filtering includes:Using Pearson correlation coefficients, the characteristic value and real property are calculated
Coefficient correlation, remove coefficient correlation be less than the default thresholding of coefficient correlation first characteristic value;Calculate any two characteristic values
Coefficient correlation, when the coefficient correlation of any two characteristic values is higher than default the second thresholding of coefficient correlation, removes and participates in what is calculated
Any feature value.
Alternatively, the second level filtering includes:Using Chi-square Test method, each characteristic value and real property are calculated
Relevance, remove onrelevant characteristic value;Using machine learning algorithm, the multiple characteristic values of assessment are associated with real property
Property, remove the characteristic value of onrelevant.
Alternatively, the machine learning algorithm, including following at least one:Regression algorithm, SVM algorithm, decision Tree algorithms,
Random forests algorithm and Xgboost algorithms.
Alternatively, the default dimension includes following at least one:Time, month, week, hour, industry major class, friendship
Easy channel, province and trade company.
Alternatively, the default index includes following at least one:Number of days, number of times, the amount of money, MCC numbers, single are average
Spending amount, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, the MCC average consumptions amount of money,
MCC average consumptions number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption number of days,
Year, moon number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, monthly average
Spending amount, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
Alternatively, the default characteristic statisticses value includes following at least one:Maximum, minimum value, intermediate value, counting
Value, summing value, average value, standard deviation, ranking value.
Alternatively, it is described hold it is humanized including it is following any one:Age bracket, sex.
The embodiment of the present invention provides a kind of identification and held humanized device, including:Acquiring unit, suitable for obtaining holder
Corresponding consumption data;First recognition unit, it is corresponding to acquired holder to disappear suitable for based on default probability graph model
Expense data are handled, and identification obtains the consumption attribute of the holder.
Alternatively, first recognition unit includes:Iteration subelement, the first identification subelement and the second identification are single
Member, wherein:The iteration subelement, suitable for choose it is default have tendency attribute trade company and mark tendency attribute, iteration perform
First identification subelement and the second identification subelement, until reaching the condition of default stopping iteration;First identification is single
Member, suitable for the tendency attribute according to the trade company that marked tendency attribute, marks the consumption attribute of the holder corresponding to it;It is described
Second identification subelement, suitable for the consumption attribute according to the holder that marked consumption attribute, is marked unmarked corresponding to it
It is inclined to the tendency attribute of the trade company of attribute.
Alternatively, it is described it is default stop iteration condition comprising it is following any one:Reach maximum iteration, own
The consumption attribute of holder and the tendency attribute of all trade companies are labeled.
Alternatively, the first identification subelement includes:First mark module, is the first attribute suitable for will be inclined to attribute
The consumption attribute mark value of the corresponding holder of trade company adds 0, and will be inclined to attribute is the corresponding holder's of trade company of the second attribute
Consumption attribute mark value adds 1;First computing module, the consumption attribute index suitable for calculating the holder, the holder's
Consume attribute index and be equal to consumption attribute mark value and the business of the total degree of mark;First judge module, suitable for being held when described
The consumption attribute index of people is less than default consumption attribute the first thresholding of index, by the consumption attribute of the holder labeled as the
One attribute;When the consumption attribute index of the holder is more than default consumption attribute the second thresholding of index, by the holder
Consumption attribute be labeled as the second attribute, default consumption attribute index second thresholding is more than default consumption attribute index
First thresholding.
Alternatively, the second identification subelement includes:Second mark module, is the first attribute suitable for will consume attribute
The tendency attribute mark value of the corresponding trade company of holder adds 0, and will consume attribute is the corresponding trade company of holder of the second attribute
Tendency attribute mark value adds 1;Second computing module, the tendency attribute index suitable for calculating the trade company, the tendency of the trade company
Attribute index is equal to tendency attribute mark value and the business of the total degree of mark;Second judge module, suitable for inclining when the trade company
It is less than default tendency attribute the first thresholding of index to attribute index, the tendency attribute of the trade company is labeled as the first attribute;
When the tendency attribute index of the trade company is more than default tendency attribute the second thresholding of index, by the tendency attribute mark of the trade company
The second attribute is designated as, the second thresholding of the default tendency attribute index is more than default tendency attribute the first thresholding of index.
Alternatively, the first identification subelement includes:3rd mark module, is the i-th attribute suitable for will be inclined to attribute
The i-th consumption attribute mark value of the corresponding holder of trade company adds 1, until it is corresponding to travel through all trade companies that marked tendency attribute
All holders, 1≤i≤I, wherein:I is the total number of tendency attribute, I >=1;I is wherein any one tendency attribute;3rd
Computing module, all consumption attribute indexes suitable for calculating the holder successively, wherein the i-th consumption attribute index is equal to i-th
Attribute mark value and the business of the total degree of mark;3rd judge module is big suitable for the i-th consumption attribute index as the holder
When default consumption attribute three thresholding of index, the consumption attribute of the holder is labeled as the i-th attribute.
Alternatively, the second identification subelement includes:4th mark module, is jth attribute suitable for will consume attribute
The jth tendency attribute mark value of the corresponding trade company of holder adds 1, until traveling through all holder's correspondences that marked consumption attribute
All trade companies, 1≤j≤J, wherein:J is the total number of consumption attribute, J >=1;J is wherein any one consumption attribute;4th
Computing module, all tendency attribute indexes suitable for calculating the trade company successively, wherein jth tendency attribute index inclines equal to jth
To attribute mark value and the business of the total degree of mark;4th judge module is big suitable for the jth tendency attribute index when the trade company
When default tendency attribute three thresholding of index, the tendency attribute of the trade company is labeled as jth attribute.
Alternatively, it is described to recognize that the humanized device that holds also includes:Grouped element, suitable for being held according to described in being recognized
Block the consumption attribute of people, the holder is grouped, wherein s groups crowd is the crowd labeled as s attributes, 1≤s≤S, S
To consume the total number of attribute, S >=1, s is wherein any one consumption attribute;Computing unit, suitable for the consumption based on holder
Data, calculate any feature statistical value of any index under any dimension, until traversal is obtained under all default dimensions
All default characteristic statisticses values corresponding to all default indexs, generate the first consumption data characteristic value file;Second knows
Other unit, suitable for for every group of people, being based respectively on the first consumption data characteristic value file and machine learning model, recognizes institute
State the real property of holder.
Alternatively, it is described to recognize that the humanized device that holds also includes:Modeling unit, is adapted to set up the machine learning mould
Type, the modeling unit includes:Subelement is obtained, the sample suitable for obtaining the clearly holder of mark real property consumes number
According to;Computation subunit, suitable for the sample consumption data based on the holder for clearly marking real property, is calculated under any dimension
Any index any feature statistical value, until traversal is obtained corresponding to all default indexs under all default dimensions
All default characteristic statisticses values, generate the second consumption data characteristic value file;Subelement is filtered, suitable for the second consumption number
Filtered according to the characteristic value in characteristic value file, generate the second consumption data validity feature value file;Subelement is modeled, is suitable to
Based on the second consumption data validity feature value file, using machine learning algorithm, machine learning model is set up.
Alternatively, the filtering subelement includes:First filtering module, suitable for removing useless and multiple eigenvalue;Second
Filtering module, suitable for removing onrelevant characteristic value.
Alternatively, first filtering module includes:First filter submodule, suitable for utilizing Pearson correlation coefficients, meter
The characteristic value and the coefficient correlation of real property are calculated, the feature that coefficient correlation is less than the default thresholding of coefficient correlation first is removed
Value;Second filter submodule, the coefficient correlation suitable for calculating any two characteristic values, when the coefficient correlation of any two characteristic values is higher than
During default the second thresholding of coefficient correlation, any feature value for participating in calculating is removed.
Alternatively, second filtering module includes:3rd filter submodule, suitable for utilizing Chi-square Test method, is calculated
The relevance of each characteristic value and real property, removes the characteristic value of onrelevant;4th filter submodule, suitable for utilizing machine
Learning algorithm, assesses multiple characteristic values and the relevance of real property, removes the characteristic value of onrelevant.
Alternatively, the machine learning algorithm, including following at least one:Regression algorithm, SVM algorithm, decision Tree algorithms,
Random forests algorithm and Xgboost algorithms.
Alternatively, the default dimension includes following at least one:Time, month, week, hour, industry major class, friendship
Easy channel, province and trade company.
Alternatively, the default index includes following at least one:Number of days, number of times, the amount of money, MCC numbers, single are average
Spending amount, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, the MCC average consumptions amount of money,
MCC average consumptions number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption number of days,
Year, moon number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, monthly average
Spending amount, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
Alternatively, the default characteristic statisticses value includes following at least one:Maximum, minimum value, intermediate value, counting
Value, summing value, average value, standard deviation, ranking value.
Alternatively, it is described hold it is humanized including it is following any one:Age bracket, sex.
The embodiment of the present invention also provides a kind of computer-readable medium, is stored thereon with computer instruction, the computer
Step corresponding to above-described embodiment any methods described is performed during instruction operation.
The embodiment of the present invention also provides a kind of identification and held humanized system, including memory and processor, described to deposit
Be stored with the computer instruction that can be run on the processor on reservoir, when the processor runs the computer instruction
Perform step corresponding to any methods described of above-described embodiment.
Compared with prior art, the technical scheme of the embodiment of the present invention has the advantages that:
The embodiment of the present invention can be obtained the consumption of holder by default probability graph model by the tendency attribute of trade company
Attribute, then the consumption attribute Fan Tui trade companies by holder tendency attribute, by successive ignition, can finally obtain all trade companies
Tendency attribute and all holders consumption attribute, improve holder's Attribute Recognition accuracy rate and coverage rate.
Further, machine learning model is built based on the sample consumption data that marked real property, and based on foundation
Machine learning model the real property of holder is identified, so as to not only identify the consumption attribute of holder, and
The real property of holder is identified, therefore holder's attribute information of multiple dimensions can be obtained, the Genus Homo that holds further is improved
Property identification accuracy rate.
Brief description of the drawings
Fig. 1 is a kind of flow chart for recognizing the humanized method that holds provided in an embodiment of the present invention;
Fig. 2 is a kind of side of the consumption attribute of holder according to merchant identification corresponding to it provided in an embodiment of the present invention
The flow chart of method;
Fig. 3 be it is provided in an embodiment of the present invention it is a kind of marked according to holder the trade company corresponding to it tendency attribute side
The flow chart of method;
Fig. 4 is the consumption attribute of another holder according to merchant identification corresponding to it provided in an embodiment of the present invention
The flow chart of method;
Fig. 5 be it is provided in an embodiment of the present invention it is another marked according to holder the trade company corresponding to it tendency attribute
The flow chart of method;
Fig. 6 is the result of a kind of tendency attribute for marking trade company provided in an embodiment of the present invention and the consumption attribute of holder
Schematic diagram;
Fig. 7 is a kind of flow chart of method for setting up machine learning model provided in an embodiment of the present invention;
Fig. 8 is the ROC curve of the machine learning model of the attribute of correspondence two provided in an embodiment of the present invention;
Fig. 9 is the ROC curve of the machine learning model of the attribute of correspondence two provided in an embodiment of the present invention;
Figure 10 is the ROC curve of the machine learning model of the attribute of correspondence two provided in an embodiment of the present invention;
Figure 11 is the schematic diagram provided in an embodiment of the present invention for recognizing the humanized device that holds.
Embodiment
The scheme of its attribute is directly recognized by the industry belonging to the consumption data of holder at present, although simple and easy to apply,
But there is the problem of accuracy rate is low, coverage rate is low.
The consumption attribute of holder is identified by default probability graph model for the embodiment of the present invention, can be by business
The tendency attribute at family obtains the consumption attribute of holder, then the consumption attribute Fan Tui trade companies by holder tendency attribute, pass through
Successive ignition, can finally obtain the tendency attribute of all trade companies and the consumption attribute of all holders, and raising holds humanized
Recognition accuracy and coverage rate.
It is understandable to enable above-mentioned purpose, feature and beneficial effect of the invention to become apparent, below in conjunction with the accompanying drawings to this
The specific embodiment of invention is described in detail.
Referring to Fig. 1, the embodiment of the present invention proposes a kind of humanized recognition methods that holds, and may include steps of:
S101, obtains the corresponding consumption data of holder.
In specific implementation, the corresponding consumption data of holder can be obtained by the consumer record of POS, can also led to
The consumer record for crossing paying website obtains the corresponding consumption data of holder, as long as resulting in the corresponding consumption data of holder
, specific acquiring way is not limited.
S102, based on default probability graph model, is handled the acquired corresponding consumption data of holder, identification
Obtain the consumption attribute of the holder.
In specific implementation, default probability graph model can be based on, according to the trade company that marked tendency attribute, it is marked
The consumption attribute of corresponding holder, according still further to the consumption attribute for the holder that marked consumption attribute, is marked corresponding to it
Unmarked tendency attribute trade company tendency attribute, successive ignition said process, until reach maximum iteration or institute
There is the consumption attribute of holder and all trade company's attributes are labeled.
In specific implementation, the maximum iteration can correspond to all trade companies and all holders all travel through
The number of times of one time, when reaching maximum iteration, illustrates that the information of all trade companies and holder are all traveled through, changes again
Generation is the result before repeating, and will not produce new output, so reaching that maximum iteration stops iteration and can avoided not
It is necessary to repeat, so as to save resource.When the consumption attribute of all holders and the tendency attribute of all trade companies are marked
Clock, it is contemplated that recognition time, iteration can also be stopped, to improve recognition efficiency.
Using such scheme, by successive ignition, the tendency attribute of all trade companies and all holders can be finally obtained
Consumption attribute, improve holder's Attribute Recognition accuracy rate and coverage rate.
In specific implementation, can also according to the holder recognized consumption attribute, the holder is grouped,
For example, being grouped according to following principle:According to consumption attribute packet, wherein t groups crowd is the crowd labeled as t attributes, 1≤
T≤T, T are the total number of consumption attribute, and T >=1, t is wherein any one consumption attribute.It is then based on the consumption number of holder
According to any feature statistical value of any index under any dimension of calculating, until traversal obtains the institute under all default dimensions
There are all default characteristic statisticses values corresponding to default index, generate the first consumption data characteristic value file.Finally it is directed to
It is described to be based respectively on the first consumption data characteristic value file and machine learning model per group of people, recognize the holder's
Real property.
In specific implementation, the attribute can be gender attribute, the age bracket attribute of only two desired values, can also
For other attributes of multiple desired values.The inventive embodiments are not limited.For example, for only having the sex of two desired values to belong to
Property, desired value is man, female, for there was only the age bracket attribute of two desired values, and desired value is old, young and middle-aged.
In specific implementation, the default dimension can be time, month, week, hour, industry major class, transaction canal
The one or several kinds such as road, province and trade company.
In specific implementation, the default index can be number of days, number of times, the amount of money, MCC numbers, single average consumption
The amount of money, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, the MCC average consumptions amount of money, MCC are put down
Consumption number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption number of days, year,
Month number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, monthly average consumption gold
The one or several kinds such as volume, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
In specific implementation, the default characteristic statisticses value can be maximum, minimum value, intermediate value, count value, summation
The one or several kinds such as value, average value, standard deviation, ranking value, CV values.
The different consumption habits of different holders, generation can be excavated with alternate analysis by dimension, index, characteristic statisticses value
First consumption data characteristic value file, can comprehensively reflect the consumption feature of holder.
In specific implementation, the machine learning algorithm can be regression algorithm, SVM algorithm, decision Tree algorithms, random
The one or several kinds such as forest algorithm or Xgboost algorithms.
The real property of holder is identified machine learning model based on foundation, can obtain the category of multiple dimensions
Property information, further improve holder's Attribute Recognition accuracy rate.
To more fully understand those skilled in the art and realizing the present invention, it is discussed in detail below by way of specific embodiment
According to the method for the consumption attribute of holder of the merchant identification corresponding to it., can be with it is understood that in specific implementation
Using other labeling methods, specific examples below does not constitute limiting the scope of the invention.
In an embodiment of the present invention, as shown in Fig. 2 for there was only the attribute of two values, for example, gender attribute (target
Value:Man, female), the tendency attribute of the trade company by marked tendency attribute marks the consumption attribute of the holder corresponding to it,
It may include steps of:
S201, will be inclined to consumption attribute mark value of the attribute for the corresponding holder of trade company of the first attribute and adds 0, will be inclined to
Attribute adds 1 for the consumption attribute mark value of the corresponding holder of trade company of the second attribute.
S202, calculates the consumption attribute index of the holder, and the consumption attribute index of the holder is equal to consumption and belonged to
Property mark value with mark total degree business.
S203, when the consumption attribute index of the holder is less than default consumption attribute the first thresholding of index, by institute
The consumption attribute for stating holder is labeled as the first attribute;When the consumption attribute index of the holder is more than default consumption attribute
During the second thresholding of index, the consumption attribute of the holder is labeled as the second attribute, the second thresholding of the consumption attribute index
More than the consumption thresholding of attribute index first.
In specific implementation, the first attribute can be set to " man ", the second attribute is set to " female ", as the holder
Consumption attribute index it is smaller when, illustrate mark " man " number of times it is relatively more, when the consumption attribute index of the holder is less than
During default consumption attribute the first thresholding of index, the consumption attribute of the holder is labeled as " man ".When the holder's
When consumption attribute index is larger, illustrate to mark the number of times of " female " relatively more, when the consumption attribute index of the holder is more than in advance
If consumption attribute the second thresholding of index when, by the consumption attribute of the holder be labeled as " female "., can be with using such scheme
By the tendency attribute for the trade company that marked tendency attribute, the consumption attribute of corresponding holder is marked.
In an embodiment of the present invention, as shown in figure 3, for there was only the attribute of two desired values, for example, gender attribute
(desired value:Man, female), according to the consumption attribute for the holder that marked consumption attribute, mark the unmarked tendency corresponding to it
The tendency attribute of the trade company of attribute, may include steps of:
S301, will consume tendency attribute mark value of the attribute for the corresponding trade company of holder of the first attribute and adds 0, will consume
Attribute adds 1 for the tendency attribute mark value of the corresponding trade company of holder of the second attribute.
S302, calculates the tendency attribute index of the trade company, and the tendency attribute index of the trade company is equal to tendency attribute mark
Note value and the business of the total degree of mark.
S303, will be described when the tendency attribute index of the trade company is less than default tendency attribute the first thresholding of index
The tendency attribute of trade company is labeled as the first attribute;When the tendency attribute index of the trade company is more than default tendency attribute index the
During two thresholdings, the tendency attribute of the trade company is labeled as the second attribute, default tendency attribute second thresholding of index is big
In default tendency attribute the first thresholding of index.
In specific implementation, the first attribute can be set to " man ", the second attribute is set to " female ", when the trade company
When tendency attribute index is smaller, illustrate to mark the number of times of " man " relatively more, preset when the tendency attribute index of the trade company is less than
Tendency attribute the first thresholding of index when, by the tendency attribute of the trade company be labeled as " man ".When the tendency attribute of the trade company
When index is larger, illustrate to mark the number of times of " female " relatively more, when the tendency attribute index of the trade company is more than default tendency category
During the second thresholding of sex index, the tendency attribute of the trade company is labeled as " female "., can be by marked consumption using such scheme
The consumption attribute of the holder of attribute, marks the tendency attribute of the trade company of unmarked tendency attribute corresponding to it.
In an embodiment of the present invention, as shown in figure 4, according to the tendency attribute for the trade company that marked tendency attribute, mark
The consumption attribute of holder corresponding to it, may include steps of:
S401, will be inclined to attribute and adds 1 for the m consumption attribute mark value of the corresponding holder of trade company of m attributes, until
All corresponding all holders of trade company that marked tendency attribute of traversal, 1≤m≤M, wherein:M is the total individual of tendency attribute
Number, M >=1;M is wherein any one tendency attribute.
S402, calculates all consumption attribute indexes of the holder successively, wherein m consumption attribute indexes are equal to m
Consume attribute mark value and the business of the total degree of mark.
S403, will when the m consumption attribute indexes of the holder are more than default consumption attribute three thresholding of index
The consumption attribute of the holder is labeled as m attributes.
In an embodiment of the present invention, as shown in figure 5, according to the consumption attribute for the holder that marked consumption attribute, mark
Remember the tendency attribute of the trade company of unmarked tendency attribute corresponding to it, may include steps of:
S501, will consume attribute and adds 1 for the n-th tendency attribute mark value of the corresponding trade company of holder of the n-th attribute, until
All corresponding all trade companies of holder that marked consumption attribute of traversal, 1≤n≤N, wherein:N is the total individual of consumption attribute
Number, N >=1;N is wherein any one consumption attribute.
S502, calculates all tendency attribute indexes of the trade company successively, wherein the n-th tendency attribute index inclines equal to n-th
To attribute mark value and the business of the total degree of mark.
S503, when the n-th tendency attribute index of the trade company is more than default tendency attribute three thresholding of index, by institute
The tendency attribute for stating trade company is labeled as the n-th attribute.
To more fully understand those skilled in the art and realizing the present invention, Fig. 6 gives the tendency attribute of mark trade company
With the result schematic diagram of the consumption attribute of holder.
Referring to Fig. 6, in an embodiment of the present invention, the total number of all trade companies is A, the total number of all holder users
For B.Be first according to it is marked tendency attribute trade company 1, trade company 2, trade company 3 and trade company 4, mark its corresponding holder user 1,
User 2 and the consumption attribute of user 3, the consumption attribute marked according still further to user 1, user 2 and user 3, are marked corresponding to it
The unmarked tendency trade company 5 of attribute, trade company 6, the tendency attribute of trade company 7 and trade company 8, further according to the trade company that marked tendency attribute
5th, the tendency attribute of trade company 6, trade company 7 and trade company 8, marks holder user 4 corresponding to it, user 5, user 6 and user 7
Attribute, successive ignition are consumed, until the consumption attribute of all holders and the tendency attribute of all trade companies are labeled.
In order that those skilled in the art more fully understand and realized the present invention, it is situated between in detail below by way of a specific embodiment
A kind of method for building up of the machine learning model continued in above-described embodiment, as shown in Figure 7.It is understood that in specific implementation
In, the method for building up of other learning models can also be used, specific examples below is not constituted to the scope of the present invention
Limitation.
In an embodiment of the present invention, referring to Fig. 7, the machine learning model can be set up by following step:
S701, has obtained the clearly sample consumption data of the holder of mark real property.
S702, based on the sample consumption data for the holder for clearly marking real property, calculates appointing under any dimension
Any feature statistical value of one index, until traversal obtains the institute corresponding to all default indexs under all default dimensions
There is default characteristic statisticses value, generate the second consumption data characteristic value file.
In specific implementation, the default dimension can be time, month, week, hour, industry major class, transaction canal
The one or several kinds such as road, province and trade company.
In specific implementation, the default index can be number of days, number of times, the amount of money, MCC numbers, single average consumption
The amount of money, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, the MCC average consumptions amount of money, MCC are put down
Consumption number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption number of days, year,
Month number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, monthly average consumption gold
The one or several kinds such as volume, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
In specific implementation, the default characteristic statisticses value can be maximum, minimum value, intermediate value, count value, summation
The one or several kinds such as value, average value, standard deviation, ranking value, CV values.
The different consumption habits of different holders, generation can be excavated with alternate analysis by dimension, index, characteristic statisticses value
Second consumption data characteristic value file, can comprehensively reflect the consumption feature of holder.
In an embodiment of the present invention, dimension selection hour (being specially 10 o'clock), index selection number of days, characteristic statisticses value
Selected and sorted value, generation hour (10) _ number of days _ ranking value, for representing 10 o'clock corresponding consumption number of days in all time clocks
The ranking of the corresponding consumption number of days of point, ranking value is smaller to represent that ranking is more forward.The index is bigger, and its corresponding holder more holds
Easily tend to consumer behavior and occur 10 o'clock after working peak is avoided to the housewife between 11 o'clock a.p.
In an embodiment of the present invention, dimension selection trade company, index selects the daily average consumption amount of money, the choosing of characteristic statisticses value
CV values are selected, the trade company _ amount of money of average consumption daily _ CV values are generated, for representing the CV values of the daily average consumption amount of money under each trade company,
CV values are smaller to show that overall spending amount is more steady, and CV values show that more greatly spending amount fluctuation is bigger.Due to the inclined reason of male's consumption
Property, female consumption easily gets excited, and fluctuation is very big, so CV values are smaller, its corresponding holder more tends to male, and CV values are got over
Greatly, its corresponding holder more tends to women.
In an embodiment of the present invention, dimension selection month, index selection monthly average consumption number of days, index selection standard
Difference, in generation month _ monthly consume number of days _ standard deviation, for representing the fluctuation situation for monthly consuming number of days, standard deviation shows more greatly
Consumption Fluctuation is bigger, and standard deviation is smaller, shows that Consumption Fluctuation is smaller.Because male consumes inclined rationality, female consumption easily gets excited
Fluctuation is very big, so standard deviation is smaller, its corresponding holder more tends to male, and standard deviation is bigger, and its is corresponding to hold
People more tends to women.
In an embodiment of the present invention, can be respectively to male holder for there was only the gender attribute of two desired values
Chinese Word Segmentation is carried out with the name of firm of women holder, trade company's vocabulary frequency and vocabulary inverse document frequency, Ran Houti is calculated
Top10 trade companies keyword is taken as feature, one-hot coding, generation trade company key characteristics value is carried out.
In an embodiment of the present invention, for gender attribute, because different geographical male holder and women holder
Consumption habit is different, so different geographical can select different characteristic attributes to distinguish different gender attributes, for example,
When dimension selects province, for Zhejiang Province, index can select the single average consumption amount of money, for Jiangsu Province, and index can be with
The daily average consumption amount of money is selected, for Beijing, index can select city.
S703, is filtered to the characteristic value in the second consumption data characteristic value file, and the second consumption data of generation is effective
Characteristic value file.
It is described that characteristic value is filtered in specific implementation, two parts can be included:The first order is filtered, and is removed useless
And multiple eigenvalue;The second level is filtered, and removes onrelevant characteristic value.By double-filtration, invalid characteristic value is filtered out, can be kept away
Exempt from the follow-up substantial amounts of invalid computing carried out based on invalid characteristic value, so as to reduce the operand and complexity of machine-learning process
Degree, hoisting machine learning model sets up efficiency.
In an embodiment of the present invention, it is possible to use Pearson correlation coefficients, the characteristic value and real property are calculated
Coefficient correlation, removes the characteristic value that coefficient correlation is less than the default thresholding of coefficient correlation first.Characteristic value rate of change, is not same
Whether the value of the same feature of this individual has resolvability, and characteristic value value changes differentiation energy of the hour to response variable
Power is small, so needing to remove the small feature of those values change.Pearson correlation coefficients are a kind of simplest, can help to understand
The method of relation between feature and response variable, what this method was weighed is the linear dependence between variable, value area as a result
Between be [- 1,1], -1 represents complete negatively correlated, and+1 represents complete positive correlation, and 0 represents no linear correlation.Wherein, negative
Pass refers to that a variable declines in two variables, and another variable will rise.
In an embodiment of the present invention, the coefficient correlation of any two characteristic values can be calculated, when the phase of any two characteristic values
When relation number is higher than default the second thresholding of coefficient correlation, any feature value for participating in calculating is removed.
The substantial amounts of characteristic value of generation can be intersected by dimension, index, characteristic statisticses value, characteristic value two-by-two is calculated successively
Coefficient correlation, and the characteristic value of repetition is removed, the follow-up substantial amounts of invalid computing carried out based on the characteristic value repeated can be avoided.
In an embodiment of the present invention, it is possible to use Chi-square Test method, each characteristic value and real property are calculated
Relevance, removes the characteristic value of onrelevant.Single argument feature selecting can be tested each feature, weigh this feature and
Relation between response variable, the characteristic value of onrelevant is removed according to score.Chi-square Test method is fairly simple, it is easy to run,
It should be readily appreciated that.
In an embodiment of the present invention, it is possible to use machine learning algorithm, the pass of multiple characteristic values and real property is assessed
Connection property, removes the characteristic value of onrelevant.Pass through machine learning method such as regression model, SVM algorithm, decision Tree algorithms or random
Forest algorithm etc., it is easy to given a mark to feature, or be easy to be applied in feature selecting task, therefore can adopt
Take and characteristic value is selected based on machine learning model.
S704, based on the second consumption data validity feature value file, using machine learning algorithm, sets up machine learning mould
Type.
In specific implementation, the machine learning algorithm can be regression algorithm, SVM algorithm, decision Tree algorithms, random
The one or several kinds such as forest algorithm or Xgboost algorithms.
In specific implementation, machine learning model can be set up using decision tree (Decision Tree, DT).Why
Referred to as set, be the developmental process because the similar one tree of its modeling process, i.e., since root, to trunk, to branch, then to thin
The bifurcated of branch minor details, finally grows the leaf of slices.In decision tree, the data sample analyzed was integrated into one before this
Tree root, then by branch layer by layer, ultimately forms several nodes.
In an embodiment of the present invention, using Taxonomy and distribution (Classification and Regression
Tree, CART) algorithm sets up machine learning model.CART segmentation logic is that each layer of division is all based on all features
Inspection and selection on.The decision tree that CART is produced is two points, and each node can only separate two, and in the growth of tree
During, same feature can be reused several times.
Referring to Fig. 8, including subgraph (a) and subgraph (b), in an embodiment of the present invention, for gender attribute, male is utilized
Sample consumption data and CART algorithms set up machine learning model a, and machine is set up using women sample consumption data and CART algorithms
Device learning model b.Wherein:Machine learning model a is corresponding to receive operating characteristics (Receiver Operating
Characteristic, ROC) curve be Fig. 8 (a) in curve 81, the corresponding ROC curve of random reference model be Fig. 8 (a) in
Curve 82, area (Area Under Curve, AUC) value is very high under the corresponding line of curve 81, is 0.926, close to 1, explanation
Machine learning model a has very high accuracy rate.The corresponding ROC curves of machine learning model b are the curve 83 in Fig. 8 (b), with
The corresponding ROC curve of machine reference model is the curve 82 in Fig. 8 (b), and the corresponding AUC of curve 83 is also higher, is 0.869,
Close to 1, illustrate that machine learning model b also has higher accuracy rate.
In specific implementation, machine learning model can be set up using regression algorithm.Regression algorithm mainly describe one because
How variable changes with the change of a collection of independent variable, and its regression equation is exactly that the data of dependent variable and independent variable relation reflect.
The change of dependent variable includes two parts:Systematic change and random change, wherein, systematic change is that independent variable can have been explained
, random change can not be explained by independent variable, commonly referred to as residual value.When target variable is binary variable, logistic regression
Analysis (Logistic Regression, lr) is one highly developed reliable.For the target variable of binary, logistic regression
Purpose seek to predict the probability of " 1 " of the corresponding dependent variable of one group of independent variable numerical value, this probability between [0,1] it
Between.If carrying out probability calculation with linear regression method, it is necessary to use special probability calculation formula, i.e. sigmoid letters
Number, the function may insure the prediction probability of binary target variable between [0,1].
Referring to Fig. 9, including subgraph (a) and subgraph (b), in an embodiment of the present invention, for gender attribute, male is utilized
Sample consumption data and logistic regression algorithm set up machine learning model a, are calculated using women sample consumption data and logistic regression
Method sets up machine learning model b.Wherein:The corresponding ROC curves of machine learning model a are the curve 91 in Fig. 9 (a), random ginseng
It is the curve 92 in Fig. 9 (a) to examine the corresponding ROC curve of model, and the corresponding AUC of curve 91 is higher, is 0.711, close to 1, says
Bright machine learning model a has higher accuracy rate.The corresponding ROC curves of machine learning model b are the curve 93 in Fig. 9 (b),
The corresponding ROC curve of random reference model is the curve 92 in Fig. 9 (b), and the corresponding AUC of curve 93 is very high, is 0.913,
Close to 1, illustrate that machine learning model b has very high accuracy rate.
In specific implementation, machine learning model can be set up using XgBoost algorithms.In general, algorithm be all by
Model, parameter and the part of object function three composition, and it is a machine learning techniques to be lifted, its each step produces a weak prediction
Model (base grader) and weighted accumulation is into total model;If the weak forecast model generation of each step is in accordance with object function
Gradient direction, then referred to as gradient lifted.What Xgboost base grader was selected is CART trees, and adds penalty term Omega
Carry out the complexity of Controlling model and the Maximum-likelihood estimation of regularization plays and prevents the effect of over-fitting.Xgboost algorithms have
Automated characterization is selected, and computation complexity is high, the advantage such as extensive effect is good.
Referring to Figure 10, including subgraph (a) and subgraph (b), in an embodiment of the present invention, for gender attribute, man is utilized
Property sample consumption data and XgBoost algorithms set up machine learning model a, utilize women sample consumption data and XgBoost to calculate
Method sets up machine learning model b.Wherein:The corresponding ROC curves of machine learning model a are the curve 101 in Figure 10 (a), at random
The corresponding ROC curve of reference model is the curve 102 in Figure 10 (a), and the corresponding AUC of curve 101 is very high, is 0.996, connects
Nearly 1, illustrate that machine learning model a has very high accuracy rate.The corresponding ROC curves of machine learning model b is in Figure 10 (b)
Curve 103, the corresponding ROC curve of random reference model be Figure 10 (b) in curve 102, the corresponding AUC of curve 103 nor
Chang Gao, is 0.923, close to 1, illustrates that machine learning model b also has very high accuracy rate.
Can be real to more fully understand those skilled in the art and realizing to additionally provide in the present invention, the embodiment of the present invention
The corresponding device of humanized method that holds is recognized in existing above-described embodiment, is described in detail referring to the drawings.
The device 110 that a kind of identification as shown in figure 11 holds humanized, including:The identification of acquiring unit 111 and first is single
Member 112, wherein:
The acquiring unit 111, suitable for obtaining the corresponding consumption data of holder.
First recognition unit 112, it is corresponding to acquired holder to disappear suitable for based on default probability graph model
Expense data are handled, and identification obtains the consumption attribute of the holder.
In an embodiment of the present invention, first recognition unit 112 includes:Identification of iteration subelement 1121, first
Unit 1122 and second recognizes subelement 1123, wherein:
The iteration subelement 1121, suitable for choose it is default have tendency attribute trade company and mark tendency attribute, iteration
Perform first and recognize subelement and the second identification subelement, until reaching the condition of default stopping iteration.
The first identification subelement 1122, suitable for the tendency attribute according to the trade company that marked tendency attribute, marks it
The consumption attribute of corresponding holder.
The second identification subelement 1123, suitable for the consumption attribute according to the holder that marked consumption attribute, mark
The tendency attribute of the trade company of unmarked tendency attribute corresponding to it.
In specific implementation, it is described it is default stop iteration condition comprising it is following any one:Reach greatest iteration time
The consumption attribute of several, all holders and the tendency attribute of all trade companies are labeled.
In an embodiment of the present invention, the first identification subelement 1122 includes:First mark module (not shown),
One computing module (not shown) and the first judge module (not shown), wherein:
First mark module, suitable for consumption attribute of the attribute for the corresponding holder of trade company of the first attribute will be inclined to
Mark value adds 0, will be inclined to consumption attribute mark value of the attribute for the corresponding holder of trade company of the second attribute and adds 1.
First computing module, the consumption attribute index suitable for calculating the holder, the consumption category of the holder
Sex index is equal to consumption attribute mark value and the business of the total degree of mark.
First judge module, suitable for being less than default consumption attribute index when the consumption attribute index of the holder
First thresholding, the first attribute is labeled as by the consumption attribute of the holder;When the consumption attribute index of the holder is more than
Default consumption attribute the second thresholding of index, the second attribute is labeled as by the consumption attribute of the holder, described default to disappear
Take the thresholding of attribute index second and be more than default consumption attribute the first thresholding of index.
In an embodiment of the present invention, the first identification subelement 1123 includes:Second mark module (not shown),
Two computing module (not shown) and the second judge module (not shown), wherein:
Second mark module, suitable for tendency attribute of the attribute for the corresponding trade company of holder of the first attribute will be consumed
Mark value adds 0, will consume tendency attribute mark value of the attribute for the corresponding trade company of holder of the second attribute and adds 1.
Second computing module, the tendency attribute index suitable for calculating the trade company, the tendency attribute of the trade company refers to
Number is equal to tendency attribute mark value and the business of the total degree of mark.
Second judge module, suitable for being less than default tendency attribute index the when the tendency attribute index of the trade company
One thresholding, the first attribute is labeled as by the tendency attribute of the trade company;When the tendency attribute index of the trade company is more than default
Attribute the second thresholding of index is inclined to, the tendency attribute of the trade company is labeled as the second attribute, the default tendency attribute refers to
Several second thresholdings are more than default tendency attribute the first thresholding of index.
In an embodiment of the present invention, the first identification subelement 1122 includes:3rd mark module (not shown),
Three computing module (not shown) and the 3rd judge module (not shown), wherein:
3rd mark module, belongs to suitable for will be inclined to attribute for the i-th consumption of the corresponding holder of trade company of the i-th attribute
Property mark value add 1, until travel through it is all marked tendency attribute the corresponding all holders of trade company, 1≤i≤I, wherein:I is
It is inclined to the total number of attribute, I >=1;I is wherein any one tendency attribute.
3rd computing module, all consumption attribute indexes suitable for calculating the holder successively, wherein the i-th consumption
Attribute index is equal to the i-th attribute mark value and the business of the total degree of mark.
3rd judge module, is more than default consumption attribute suitable for the i-th consumption attribute index as the holder
During three thresholding of index, the consumption attribute of the holder is labeled as the i-th attribute.
In an embodiment of the present invention, the first identification subelement 1123 includes:4th mark module (not shown),
Four computing module (not shown) and the 4th judge module (not shown), wherein:
4th mark module, belongs to suitable for will consume jth of the attribute for the corresponding trade company of holder of jth attribute and be inclined to
Property mark value add 1, until travel through it is all marked consumption attribute the corresponding all trade companies of holder, 1≤j≤J, wherein:J is
Consume the total number of attribute, J >=1;J is wherein any one consumption attribute.
4th computing module, all tendency attribute indexes suitable for calculating the trade company successively, wherein jth tendency category
Sex index is equal to jth and is inclined to attribute mark value and the business of the total degree of mark.
4th judge module, is more than default tendency attribute suitable for the jth tendency attribute index when the trade company and refers to
During three thresholding of number, the tendency attribute of the trade company is labeled as jth attribute.
In specific implementation, the device 110 for not holding humanized, in addition to:Grouped element 113, computing unit 114
With the second recognition unit 115, wherein:
The grouped element 113, suitable for the consumption attribute according to the holder recognized, by the holder point
Group, wherein s groups crowd is the crowd labeled as s attributes, 1≤s≤S, S is the total number of consumption attribute, and S >=1, s is it
In any one consumption attribute.
The computing unit 114, suitable for the consumption data based on holder, calculates appointing for any index under any dimension
One characteristic statisticses value, until traversal obtains all default spies corresponding to all default indexs under all default dimensions
Statistical value is levied, the first consumption data characteristic value file is generated.
Second recognition unit 115, suitable for for every group of people, being based respectively on the first consumption data characteristic value file
And machine learning model, recognize the real property of the holder.
In specific implementation, the default dimension includes following at least one:Time, month, week, hour, industry
Major class, transaction channel, province and trade company.
In specific implementation, the default index includes following at least one:Number of days, number of times, the amount of money, MCC numbers, list
The secondary average consumption amount of money, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, MCC average consumptions
The amount of money, MCC average consumptions number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption
Number of days, year, moon number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, the moon
The average consumption amount of money, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
In specific implementation, the default characteristic statisticses value includes following at least one:Maximum, minimum value, intermediate value,
Count value, summing value, average value, standard deviation, ranking value.
In specific implementation, the device 110 for not holding humanized, in addition to:Modeling unit 116, is adapted to set up institute
State machine learning model.
In an embodiment of the present invention, the modeling unit 116 includes:Obtain subelement 1161, computation subunit 1162,
Subelement 1163 and modeling subelement 1164 are filtered, wherein:
The acquisition subelement 1161, suitable for having obtained the clearly sample consumption data of the holder of mark real property.
The computation subunit 1162, suitable for the sample consumption data based on the holder for clearly marking real property,
Any feature statistical value of any index under any dimension is calculated, until traversal obtains all pre- under all default dimensions
If index corresponding to all default characteristic statisticses values, generate the second consumption data characteristic value file.
The filtering subelement 1163, it is raw suitable for being filtered to the characteristic value in the second consumption data characteristic value file
Into the second consumption data validity feature value file.
The modeling subelement 1164, suitable for based on the second consumption data validity feature value file, being calculated using machine learning
Method, sets up machine learning model.
In specific implementation, the machine learning algorithm includes following at least one:Regression algorithm, SVM algorithm, decision tree
Algorithm, random forests algorithm and Xgboost algorithms.
In specific implementation, the default dimension includes following at least one:Time, month, week, hour, industry
Major class, transaction channel, province and trade company.
In specific implementation, the default index includes following at least one:Number of days, number of times, the amount of money, MCC numbers, list
The secondary average consumption amount of money, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, MCC average consumptions
The amount of money, MCC average consumptions number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption
Number of days, year, moon number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, the moon
The average consumption amount of money, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
In specific implementation, the default characteristic statisticses value includes following at least one:Maximum, minimum value, intermediate value,
Count value, summing value, average value, standard deviation, ranking value.
In an embodiment of the present invention, the filtering subelement 1163 includes:First filtering module (not shown) and second
Filtering module (not shown), wherein:
First filtering module, suitable for removing useless and multiple eigenvalue.
Second filtering module, suitable for removing onrelevant characteristic value.
In an embodiment of the present invention, first filtering module includes:First filter submodule (not shown) and second
Filter submodule (not shown), wherein:
First filter submodule, suitable for utilizing Pearson correlation coefficients, calculates the characteristic value and real property
Coefficient correlation, removes the characteristic value that coefficient correlation is less than the default thresholding of coefficient correlation first.
Second filter submodule, the coefficient correlation suitable for calculating any two characteristic values, when the phase of any two characteristic values
When relation number is higher than default the second thresholding of coefficient correlation, any feature value for participating in calculating is removed.
In an embodiment of the present invention, second filtering module includes:3rd filter submodule (not shown) and the 4th
Filter submodule (not shown), wherein:
3rd filter submodule, suitable for utilizing Chi-square Test method, calculates each characteristic value and real property
Relevance, removes the characteristic value of onrelevant;
4th filter submodule, suitable for utilizing machine learning algorithm, assesses the pass of multiple characteristic values and real property
Connection property, removes the characteristic value of onrelevant.
In specific implementation, the machine learning algorithm includes following at least one:Regression algorithm, SVM algorithm, decision tree
Algorithm, random forests algorithm and Xgboost algorithms.
In specific implementation, the default dimension includes following at least one:Time, month, week, hour, industry
Major class, transaction channel, province and trade company.
In specific implementation, the default index includes following at least one:Number of days, number of times, the amount of money, MCC numbers, list
The secondary average consumption amount of money, the daily average consumption amount of money, daily average consumption number of times, MCC average consumptions number of times, MCC average consumptions
The amount of money, MCC average consumptions number of days, city number, city average consumption number of times, the city average consumption amount of money, city average consumption
Number of days, year, moon number, annual consumption number of days, annual consumption number of times, annual spending amount, annual consumption moon number, the moon
The average consumption amount of money, monthly average consumption number of days, monthly average consumption number of times, monthly average consumption number of days, trade company's keyword.
In specific implementation, the default characteristic statisticses value includes following at least one:Maximum, minimum value, intermediate value,
Count value, summing value, average value, standard deviation, ranking value.
In specific implementation, it is described hold it is humanized including it is following any one:Age bracket, sex.
The embodiment of the present invention also provides a kind of computer-readable medium, is stored thereon with computer instruction, the computer
Step corresponding to any of the above-described embodiment methods described is performed during instruction operation.
The embodiment of the present invention also provides a kind of identification and held humanized system, including memory and processor, described to deposit
Be stored with the computer instruction that can be run on the processor on reservoir, when the processor runs the computer instruction
Perform step corresponding to any of the above-described embodiment methods described.
Although present disclosure is as above, the present invention is not limited to this.Any those skilled in the art, are not departing from this
In the spirit and scope of invention, it can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
The scope of restriction is defined.
Claims (36)
- A kind of humanized method 1. identification holds, it is characterised in that including:Obtain the corresponding consumption data of holder;Based on default probability graph model, the acquired corresponding consumption data of holder is handled, identification obtains described The consumption attribute of holder.
- The humanized method 2. identification according to claim 1 holds, it is characterised in that described to be based on default probability graph Model, is handled the acquired corresponding consumption data of holder, and identification obtains the consumption attribute of the holder, bag Include:Selection is default to be had the trade company of tendency attribute and marks tendency attribute, and iteration performs following steps, until reaching default Stop the condition of iteration:According to the tendency attribute for the trade company that marked tendency attribute, the consumption attribute of the holder corresponding to it is marked;According to the consumption attribute for the holder that marked consumption attribute, the trade company of unmarked tendency attribute corresponding to it is marked It is inclined to attribute.
- The humanized method 3. identification according to claim 2 holds, it is characterised in that the default stopping iteration Condition comprising it is following any one:Reach the tendency category of maximum iteration, the consumption attribute of all holders and all trade companies Property be labeled.
- 4. identification according to claim 2 holds humanized method, it is characterised in that described according to marked tendency category The tendency attribute of the trade company of property, marks the consumption attribute of the holder corresponding to it, including:Attribute will be inclined to and add 0 for the consumption attribute mark value of the corresponding holder of trade company of the first attribute, be the by tendency attribute The consumption attribute mark value of the corresponding holder of trade company of two attributes adds 1;Calculate the consumption attribute index of the holder, the consumption attribute index of the holder be equal to consumption attribute mark value with The business of the total degree of mark;When the consumption attribute index of the holder is less than default consumption attribute the first thresholding of index, by the holder's Consume attribute and be labeled as the first attribute;When the consumption attribute index of the holder is more than default consumption attribute the second thresholding of index, by the holder's Consume attribute and be labeled as the second attribute, the second thresholding of the consumption attribute index is more than consumption attribute the first thresholding of index.
- 5. identification according to claim 2 holds humanized method, it is characterised in that described according to marked consumption category The attribute of the holder of property, marking the tendency attribute of the trade company of the unmarked tendency attribute corresponding to it includes:Attribute will be consumed and add 0 for the tendency attribute mark value of the corresponding trade company of holder of the first attribute, be the by consumption attribute The tendency attribute mark value of the corresponding trade company of holder of two attributes adds 1;The tendency attribute index of the trade company is calculated, the tendency attribute index of the trade company is equal to tendency attribute mark value and mark Total degree business;When the tendency attribute index of the trade company is less than default tendency attribute the first thresholding of index, by the tendency of the trade company Attribute is labeled as the first attribute;When the tendency attribute index of the trade company is more than default tendency attribute the second thresholding of index, by the tendency of the trade company Attribute is labeled as the second attribute, and the second thresholding of the default tendency attribute index is more than default tendency attribute index first Limit.
- 6. identification according to claim 2 holds humanized method, it is characterised in that described according to marked tendency category The tendency attribute of the trade company of property, marks the consumption attribute of the holder corresponding to it, including:Attribute will be inclined to and add 1 for the m consumption attribute mark value of the corresponding holder of trade company of m attributes, until traversal is all Marked tendency attribute the corresponding all holders of trade company, 1≤m≤M, wherein:M is the total number of tendency attribute, M >=1;m For wherein any one tendency attribute;All consumption attribute indexes of the holder are calculated successively, wherein m consumption attribute indexes are equal to m and consume attribute mark Note value and the business of the total degree of mark;When the m consumption attribute indexes of the holder are more than default consumption attribute three thresholding of index, held described The consumption attribute of people is labeled as m attributes.
- 7. identification according to claim 2 holds humanized method, it is characterised in that described according to marked consumption category The consumption attribute of the holder of property, marking the tendency attribute of the trade company of the unmarked tendency attribute corresponding to it includes:Attribute will be consumed and add 1 for the n-th tendency attribute mark value of the corresponding trade company of holder of the n-th attribute, until traversal is all Marked consumption attribute the corresponding all trade companies of holder, 1≤n≤N, wherein:N is the total number of consumption attribute, N >=1;n For wherein any one consumption attribute;All tendency attribute indexes of the trade company are calculated successively, wherein the n-th tendency attribute index is equal to the n-th tendency attribute mark Value and the business of the total degree of mark;When the n-th tendency attribute index of the trade company is more than default tendency attribute three thresholding of index, by the trade company It is inclined to attribute and is labeled as the n-th attribute.
- The humanized method 8. identification according to claim 1 holds, it is characterised in that also include:According to the consumption attribute of the holder recognized, the holder is grouped, wherein t groups crowd is labeled as the The crowd of t attributes, 1≤t≤T, T is the total number of consumption attribute, and T >=1, t is wherein any one consumption attribute;Consumption data based on holder, calculates any feature statistical value of any index under any dimension, until traveling through Number is consumed in all default characteristic statisticses values corresponding to all default indexs under to all default dimensions, generation first According to characteristic value file;For every group of people, the first consumption data characteristic value file and machine learning model are based respectively on, held described in identification The real property of people.
- 9. identification according to claim 8 holds humanized method, it is characterised in that the machine learning model is built Cube method includes:Obtain the sample consumption data for the holder for clearly marking real property;Based on the sample consumption data for the holder for clearly marking real property, appointing for any index under any dimension is calculated One characteristic statisticses value, until traversal obtains all default spies corresponding to all default indexs under all default dimensions Statistical value is levied, the second consumption data characteristic value file is generated;Characteristic value in second consumption data characteristic value file is filtered, generation the second consumption data validity feature value text Part;Based on the second consumption data validity feature value file, using machine learning algorithm, machine learning model is set up.
- The humanized method 10. identification according to claim 9 holds, it is characterised in that described to the second consumption data Characteristic value in characteristic value file is filtered, including:The first order is filtered, and removes useless and multiple eigenvalue;The second level is filtered, and removes onrelevant characteristic value.
- The humanized method 11. identification according to claim 10 holds, it is characterised in that the first order bag filter Include:Using Pearson correlation coefficients, the characteristic value and the coefficient correlation of real property are calculated, coefficient correlation is removed and is less than in advance If the thresholding of coefficient correlation first characteristic value;The coefficient correlation of any two characteristic values is calculated, when the coefficient correlation of any two characteristic values is higher than default coefficient correlation second During thresholding, any feature value for participating in calculating is removed.
- The humanized method 12. identification according to claim 10 holds, it is characterised in that the second level bag filter Include:Using Chi-square Test method, each characteristic value and the relevance of real property are calculated, the characteristic value of onrelevant is removed;Using machine learning algorithm, multiple characteristic values and the relevance of real property are assessed, the characteristic value of onrelevant is removed.
- The humanized method 13. identification according to claim 12 holds, it is characterised in that the machine learning algorithm, Including following at least one:Regression algorithm, SVM algorithm, decision Tree algorithms, random forests algorithm and Xgboost algorithms.
- The humanized method 14. identification according to claim 8 or claim 9 holds, it is characterised in that the default dimension bag Include following at least one:Time, month, week, hour, industry major class, transaction channel, province and trade company.
- The humanized method 15. identification according to claim 8 or claim 9 holds, it is characterised in that the default index bag Include following at least one:Number of days, number of times, the amount of money, MCC numbers, the single average consumption amount of money, the daily average consumption amount of money, daily Average consumption number of times, MCC average consumptions number of times, the MCC average consumptions amount of money, MCC average consumptions number of days, city number, city are put down Equal consumption number of times, the city average consumption amount of money, city average consumption number of days, year, moon number, annual consumption number of days, annual Consumption number of times, annual spending amount, annual consumption moon number, monthly average spending amount, monthly average consumption number of days, monthly average disappear Take number of times, monthly average consumption number of days, trade company's keyword.
- The humanized method 16. identification according to claim 8 or claim 9 holds, it is characterised in that the default feature system Evaluation includes following at least one:Maximum, minimum value, intermediate value, count value, summing value, average value, standard deviation, ranking value.
- 17. identification according to claim 1 holds humanized method, it is characterised in that it is described hold it is humanized including Below any one:Age bracket, sex.
- The humanized device 18. a kind of identification holds, it is characterised in that including:Acquiring unit, suitable for obtaining the corresponding consumption data of holder;First recognition unit, suitable for based on default probability graph model, being carried out to the acquired corresponding consumption data of holder Processing, identification obtains the consumption attribute of the holder.
- The humanized device 19. identification according to claim 18 holds, it is characterised in that the first recognition unit bag Include:Iteration subelement, the first identification subelement and the second identification subelement, wherein:The iteration subelement, suitable for choose it is default have tendency attribute trade company and mark tendency attribute, iteration perform first Subelement and the second identification subelement are recognized, until reaching the condition of default stopping iteration;The first identification subelement, suitable for the tendency attribute according to the trade company that marked tendency attribute, is marked corresponding to it The consumption attribute of holder;The second identification subelement, suitable for the consumption attribute according to the holder that marked consumption attribute, is marked corresponding to it Unmarked tendency attribute trade company tendency attribute.
- The humanized device 20. identification according to claim 19 holds, it is characterised in that the default stopping iteration Condition comprising it is following any one:Reach maximum iteration, the consumption attribute of all holders and the tendency of all trade companies Attribute is labeled.
- The humanized device 21. identification according to claim 19 holds, it is characterised in that the first identification subelement Including:First mark module, adds suitable for will be inclined to consumption attribute mark value of the attribute for the corresponding holder of trade company of the first attribute 0, consumption attribute mark value of the attribute for the corresponding holder of trade company of the second attribute will be inclined to and add 1;First computing module, the consumption attribute index suitable for calculating the holder, consumption attribute index of the holder etc. In consumption attribute mark value and the business of the total degree of mark;First judge module, suitable for being less than default consumption attribute index first when the consumption attribute index of the holder Limit, the first attribute is labeled as by the consumption attribute of the holder;When the consumption attribute index of the holder is more than default Attribute the second thresholding of index is consumed, the consumption attribute of the holder is labeled as the second attribute, the default consumption attribute The thresholding of index second is more than default consumption attribute the first thresholding of index.
- The humanized device 22. identification according to claim 19 holds, it is characterised in that the second identification subelement Including:Second mark module, adds suitable for will consume tendency attribute mark value of the attribute for the corresponding trade company of holder of the first attribute 0, tendency attribute mark value of the attribute for the corresponding trade company of holder of the second attribute will be consumed and add 1;Second computing module, the tendency attribute index suitable for calculating the trade company, the tendency attribute index of the trade company, which is equal to, to incline To attribute mark value and the business of the total degree of mark;Second judge module, suitable for being less than default tendency attribute the first thresholding of index when the tendency attribute index of the trade company, The tendency attribute of the trade company is labeled as the first attribute;When the tendency attribute index of the trade company is more than default tendency attribute The thresholding of index second, the second attribute, the default tendency attribute index second are labeled as by the tendency attribute of the trade company Limit is more than default tendency attribute the first thresholding of index.
- The humanized device 23. identification according to claim 19 holds, it is characterised in that the first identification subelement Including:3rd mark module, suitable for i-th consumption attribute mark value of the attribute for the corresponding holder of trade company of the i-th attribute will be inclined to Plus 1, until all corresponding all holders of trade company that marked tendency attribute are traveled through, 1≤i≤I, wherein:I is tendency attribute Total number, I >=1;I is wherein any one tendency attribute;3rd computing module, all consumption attribute indexes suitable for calculating the holder successively, wherein the i-th consumption attribute index Equal to the i-th attribute mark value and the business of the total degree of mark;3rd judge module, is more than default consumption attribute index the 3rd suitable for the i-th consumption attribute index as the holder During thresholding, the consumption attribute of the holder is labeled as the i-th attribute.
- The humanized device 24. identification according to claim 19 holds, it is characterised in that the second identification subelement Including:4th mark module, suitable for jth tendency attribute mark value of the attribute for the corresponding trade company of holder of jth attribute will be consumed Plus 1, until traveling through all corresponding all trade companies of holder that marked consumption attribute, 1≤j≤J, J is total for consumption attribute Number, J >=1;J is wherein any one consumption attribute;4th computing module, all tendency attribute indexes suitable for calculating the trade company successively, wherein jth tendency attribute index etc. Attribute mark value and the business of the total degree of mark are inclined in jth;4th judge module, is more than default tendency attribute index the 3rd suitable for the jth tendency attribute index when the trade company In limited time, the tendency attribute of the trade company is labeled as jth attribute.
- The humanized device 25. identification according to claim 18 holds, it is characterised in that also include:Grouped element, suitable for the consumption attribute according to the holder recognized, the holder is grouped, wherein s groups Crowd is the crowd labeled as s attributes, and 1≤s≤S, S is the total number of consumption attribute, and S >=1, s is that wherein any one disappears Take attribute;Computing unit, suitable for the consumption data based on holder, calculates any feature statistics of any index under any dimension Value, until traversal obtains all default characteristic statisticses values corresponding to all default indexs under all default dimensions, Generate the first consumption data characteristic value file;Second recognition unit, suitable for for every group of people, being based respectively on the first consumption data characteristic value file and machine learning Model, recognizes the real property of the holder.
- The humanized device 26. identification according to claim 25 holds, it is characterised in that also include:Modeling unit, is fitted In setting up the machine learning model, the modeling unit includes:Subelement is obtained, suitable for having obtained the clearly sample consumption data of the holder of mark real property;Computation subunit, suitable for the sample consumption data based on the holder for clearly marking real property, calculates any dimension Under any index any feature statistical value, until all default indexs institutes that traversal is obtained under all default dimensions are right All default characteristic statisticses values answered, generate the second consumption data characteristic value file;Subelement is filtered, suitable for being filtered to the characteristic value in the second consumption data characteristic value file, number is consumed in generation second According to validity feature value file;Subelement is modeled, suitable for based on the second consumption data validity feature value file, using machine learning algorithm, setting up engineering Practise model.
- The humanized device 27. identification according to claim 26 holds, it is characterised in that the filtering subelement bag Include:First filtering module, suitable for removing useless and multiple eigenvalue;Second filtering module, suitable for removing onrelevant characteristic value.
- The humanized device 28. identification according to claim 27 holds, it is characterised in that the first filtering module bag Include:First filter submodule, suitable for utilizing Pearson correlation coefficients, calculates the characteristic value and the coefficient correlation of real property, Remove the characteristic value that coefficient correlation is less than the default thresholding of coefficient correlation first;Second filter submodule, the coefficient correlation suitable for calculating any two characteristic values, when the coefficient correlation of any two characteristic values is high When default the second thresholding of coefficient correlation, any feature value for participating in calculating is removed.
- The humanized device 29. identification according to claim 27 holds, it is characterised in that the second filtering module bag Include:3rd filter submodule, suitable for utilizing Chi-square Test method, calculates each characteristic value and the relevance of real property, goes Except the characteristic value of onrelevant;4th filter submodule, suitable for utilizing machine learning algorithm, assesses multiple characteristic values and the relevance of real property, removes The characteristic value of onrelevant.
- The humanized device 30. the identification according to claim 26 or 29 holds, it is characterised in that the machine learning is calculated Method, including following at least one:Regression algorithm, SVM algorithm, decision Tree algorithms, random forests algorithm and Xgboost algorithms.
- The humanized device 31. the identification according to claim 25 or 26 holds, it is characterised in that the default dimension Including following at least one:Time, month, week, hour, industry major class, transaction channel, province and trade company.
- The humanized device 32. the identification according to claim 25 or 26 holds, it is characterised in that the default index Including following at least one:It is number of days, number of times, the amount of money, MCC numbers, the single average consumption amount of money, the daily average consumption amount of money, every Its average consumption number of times, MCC average consumptions number of times, the MCC average consumptions amount of money, MCC average consumptions number of days, city number, city Average consumption number of times, the city average consumption amount of money, city average consumption number of days, year, moon number, annual consumption number of days, Nian Ping Equal consumption number of times, annual spending amount, annual consumption moon number, monthly average spending amount, monthly average consumption number of days, monthly average Consumption number of times, monthly average consumption number of days, trade company's keyword.
- The humanized device 33. the identification according to claim 25 or 26 holds, it is characterised in that the default feature Statistical value includes following at least one:Maximum, minimum value, intermediate value, count value, summing value, average value, standard deviation, ranking value.
- 34. identification according to claim 18 holds humanized device, it is characterised in that it is described hold it is humanized including Below any one:Age bracket, sex.
- 35. a kind of computer-readable medium, is stored thereon with computer instruction, it is characterised in that the computer instruction operation When perform claim any one of 1 to 17 methods described of requirement the step of.
- The humanized system 36. a kind of identification holds, it is characterised in that including memory and processor, deposited on the memory Contain the computer instruction that can be run on the processor, the perform claim when processor runs the computer instruction It is required that the step of any one of 1 to 17 methods described.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710395643.4A CN107316205A (en) | 2017-05-27 | 2017-05-27 | Recognize humanized method, device, computer-readable medium and the system of holding |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710395643.4A CN107316205A (en) | 2017-05-27 | 2017-05-27 | Recognize humanized method, device, computer-readable medium and the system of holding |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107316205A true CN107316205A (en) | 2017-11-03 |
Family
ID=60183495
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710395643.4A Pending CN107316205A (en) | 2017-05-27 | 2017-05-27 | Recognize humanized method, device, computer-readable medium and the system of holding |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107316205A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108174296A (en) * | 2018-01-02 | 2018-06-15 | 武汉斗鱼网络科技有限公司 | Malicious user recognition methods and device |
CN108962382A (en) * | 2018-05-31 | 2018-12-07 | 电子科技大学 | A kind of layering important feature selection method based on breast cancer clinic high dimensional data |
WO2020015216A1 (en) * | 2018-07-19 | 2020-01-23 | 平安科技(深圳)有限公司 | Method and device for determining parameter threshold, and computer storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103927675A (en) * | 2014-04-18 | 2014-07-16 | 北京京东尚科信息技术有限公司 | Method and device for judging age brackets of users |
CN105589911A (en) * | 2014-12-31 | 2016-05-18 | 中国银联股份有限公司 | Customer value assessment method and system |
CN105678323A (en) * | 2015-12-31 | 2016-06-15 | 中国银联股份有限公司 | Image-based-on method and system for analysis of users |
CN105931068A (en) * | 2015-12-30 | 2016-09-07 | 中国银联股份有限公司 | Cardholder consumption figure generation method and device |
-
2017
- 2017-05-27 CN CN201710395643.4A patent/CN107316205A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103927675A (en) * | 2014-04-18 | 2014-07-16 | 北京京东尚科信息技术有限公司 | Method and device for judging age brackets of users |
CN105589911A (en) * | 2014-12-31 | 2016-05-18 | 中国银联股份有限公司 | Customer value assessment method and system |
CN105931068A (en) * | 2015-12-30 | 2016-09-07 | 中国银联股份有限公司 | Cardholder consumption figure generation method and device |
CN105678323A (en) * | 2015-12-31 | 2016-06-15 | 中国银联股份有限公司 | Image-based-on method and system for analysis of users |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108174296A (en) * | 2018-01-02 | 2018-06-15 | 武汉斗鱼网络科技有限公司 | Malicious user recognition methods and device |
CN108174296B (en) * | 2018-01-02 | 2019-09-10 | 武汉斗鱼网络科技有限公司 | Malicious user recognition methods and device |
CN108962382A (en) * | 2018-05-31 | 2018-12-07 | 电子科技大学 | A kind of layering important feature selection method based on breast cancer clinic high dimensional data |
CN108962382B (en) * | 2018-05-31 | 2022-05-03 | 电子科技大学 | Hierarchical important feature selection method based on breast cancer clinical high-dimensional data |
WO2020015216A1 (en) * | 2018-07-19 | 2020-01-23 | 平安科技(深圳)有限公司 | Method and device for determining parameter threshold, and computer storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Arunachalam et al. | Understanding big data analytics capabilities in supply chain management: Unravelling the issues, challenges and implications for practice | |
CN108197132B (en) | Graph database-based electric power asset portrait construction method and device | |
CN107818344A (en) | The method and system that user behavior is classified and predicted | |
CN109446341A (en) | The construction method and device of knowledge mapping | |
CN107908606A (en) | Method and system based on different aforementioned sources automatic report generation | |
CN107016103A (en) | A kind of method and device for building user's portrait | |
CN104866474A (en) | Personalized data searching method and device | |
CN104933622A (en) | Microblog popularity degree prediction method based on user and microblog theme and microblog popularity degree prediction system based on user and microblog theme | |
CN101493913A (en) | Method and system for assessing user credit in internet | |
CN104679743A (en) | Method and device for determining preference model of user | |
CN110310163A (en) | A kind of accurate method, equipment and readable medium for formulating marketing strategy | |
CN107316205A (en) | Recognize humanized method, device, computer-readable medium and the system of holding | |
CN109388634A (en) | Processing method, terminal device and the computer readable storage medium of address information | |
Yu et al. | Tracing the main path of interdisciplinary research considering citation preference: A case from blockchain domain | |
CN103678336A (en) | Method and device for identifying entity words | |
CN104933475A (en) | Network forwarding behavior prediction method and apparatus | |
CN113139876A (en) | Risk model training method and device, computer equipment and readable storage medium | |
CN103617146A (en) | Machine learning method and device based on hardware resource consumption | |
CN115062087A (en) | User portrait construction method, device, equipment and medium | |
CN109977131A (en) | A kind of house type matching system | |
CN113220885A (en) | Text processing method and system | |
CN107316204A (en) | Recognize humanized method, device, computer-readable medium and the system of holding | |
Dudek et al. | Integrated quality assessment of services in an adaptive expert system with a rule-based knowledge base | |
Li et al. | Automatic classification algorithm for multisearch data association rules in wireless networks | |
CN113705201B (en) | Text-based event probability prediction evaluation algorithm, electronic device and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20171103 |
|
WD01 | Invention patent application deemed withdrawn after publication |