CN108961032A - Borrow or lend money processing method, device and server - Google Patents
Borrow or lend money processing method, device and server Download PDFInfo
- Publication number
- CN108961032A CN108961032A CN201710380235.1A CN201710380235A CN108961032A CN 108961032 A CN108961032 A CN 108961032A CN 201710380235 A CN201710380235 A CN 201710380235A CN 108961032 A CN108961032 A CN 108961032A
- Authority
- CN
- China
- Prior art keywords
- user
- risk
- bull
- lend
- group
- 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/03—Credit; Loans; Processing thereof
Landscapes
- Business, Economics & Management (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
This application discloses a kind of debt-credit processing method, device and servers, this method comprises: receiving the loan application that user submits to network finance platform;In response to the loan application, access behavioral data of the user in network finance platform is obtained;Based on preset risk behavior data, analyzes access behavioral data and characterize the user there are when the risk of bull lend-borrow action, obtain the multiple user characteristics for indicating a variety of different attributes of the user;According in the risk score model that training obtains in advance, there are the influence degrees of bull lend-borrow action to characterization for the different user feature that every attribute is included, and using the risk score model are converted to multiple user characteristics of the user and to be used to characterize the user there are the risk scores of bull lend-borrow action;According to the risk score, the loan application submitted to user is handled.The program can be reduced to the case where providing loaning bill there are the borrower of financial risks, improve the financing security of internet financial platform.
Description
Technical field
This application involves technical field of image processing more particularly to a kind of debt-credit processing methods, device and server.
Background technique
Network loan refers to the lend-borrow action realized in internet financial platform, it is with the development of internet and the people
Between the rise borrowed or lent money, and financial models that one kind for growing up is new.
After network loan person borrows or lends money to internet financial platform application, internet financial platform can be according to network loan
Whether there are loaning bill behaviors in the internet financial platform by person, to determine whether agreeing to the debt-credit Shen of the network loan person
Please.However, the lend-borrow action based on network loan person in the internet financial platform, not can accurately reflect the network loan
The current financial condition of person, this this may result in internet financial platform, and there are the network loan person of financial risks offers to some
It borrows money, to increase the risk that network loan person can not can not even repay in time, leads to the fund wind of internet financial platform
Danger is higher.
Summary of the invention
In view of this, this application provides a kind of debt-credit processing method, device and server, to reduce to there are funds
The borrower of risk provides the case where loaning bill, improves the financing security of internet financial platform.
To achieve the above object, on the one hand, this application provides a kind of debt-credit processing methods, comprising:
It receives user and passes through the loan application that terminal is submitted to network finance platform;
In response to the loan application, access behavioral data of the user in the network finance platform is obtained;
Based on preset for characterizing there are the risk behavior data of bull debt-credit risk, the access row of the user is analyzed
Whether characterizing the user for data, there are the risks of bull lend-borrow action;
Analyzing the user, there are when the risk of bull lend-borrow action, obtain to indicate that a variety of of user do not belong to
Multiple user characteristics of property;
According in the risk score model that training obtains in advance, the different user feature that every attribute is included deposits characterization
Multiple user characteristics of the user are converted in the influence degree of bull lend-borrow action, and using the risk score model
For characterizing the user, there are the risk scores of bull lend-borrow action
According to the risk score, the loan application submitted to the user is handled.
Another aspect, present invention also provides a kind of debt-credit processing units, comprising:
Apply for receiving unit, passes through the loan application that terminal is submitted to network finance platform for receiving user;
Behavior acquiring unit, for obtaining visit of the user in the network finance platform in response to the loan application
Ask behavioral data;
Risk assessment unit, for, for characterizing there are the risk behavior data of bull debt-credit risk, being divided based on preset
Whether the access behavioral data for analysing the user characterizes the user there are the risks of bull lend-borrow action;
Feature acquiring unit, for analyzing the user there are when the risk of bull lend-borrow action, obtaining indicates institute
State multiple user characteristics of a variety of different attributes of user;
Score determination unit, for according in the risk score model that training obtains in advance, every attribute to be included not
With user characteristics, to characterization, there are the influence degrees of bull lend-borrow action, and using the risk score model by the user's
Multiple user characteristics are converted to that there are the risk scores of bull lend-borrow action for characterizing the user;
Apply for processing unit, for according to the risk score, the loan application submitted to the user to be handled.
Another aspect, present invention also provides a kind of servers, comprising:
Processor, memory and communication interface, wherein the processor, memory and communication interface are total by communication
Line is connected;
The communication interface passes through the loan application that terminal is submitted to network finance platform for receiving user;
The processor, for calling the program stored in the memory;
For storing program, described program is used for the memory:
In response to the loan application, access behavioral data of the user in the network finance platform is obtained;
Based on preset for characterizing there are the risk behavior data of bull debt-credit risk, the access row of the user is analyzed
Whether characterizing the user for data, there are the risks of bull lend-borrow action;
Analyzing the user, there are when the risk of bull lend-borrow action, obtain to indicate that a variety of of user do not belong to
Multiple user characteristics of property;
According in the risk score model that training obtains in advance, the different user feature that every attribute is included deposits characterization
Multiple user characteristics of the user are converted in the influence degree of bull lend-borrow action, and using the risk score model
For characterizing the user, there are the risk scores of bull lend-borrow action;
According to the risk score, the loan application submitted to the user is handled.
It can be seen via above technical scheme that after network finance platform receives the loan application of user, if root
According to access behavioral data of the user in the network finance platform and preset characterization user, there are bull lend-borrow actions
Risk behavior data, analyzing the user, there are the risks of bull lend-borrow action, then can further obtain the user it is a variety of not
With multiple user characteristics of attribute, and according to the difference that in the preparatory risk score model trained and obtained, every attribute is included
User characteristics are to characterization there are the influence degree of bull lend-borrow action, and determining the user, there are the risks of bull lend-borrow action to comment
Point, can identifying the user by the risk score, there are the degrees of risk of bull lend-borrow action, since there are bulls to borrow by user
Loan behavior then illustrates that very big problem occurs in the fund of user, it is likely that can not pay off a loan, therefore, comment according to the risk in time
Divide and loan application is handled, can reduce to there are the users of lend-borrow action to provide the risk borrowed money, and then reduce malice
The case where debt, improves the financing security of internet financial platform.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 shows a kind of a kind of composed structure schematic diagram of application scenarios of debt-credit processing method of the application;
Fig. 2 shows a kind of composed structures of server used in a kind of debt-credit processing method suitable for the application to show
It is intended to;
Fig. 3 shows a kind of flow diagram for borrowing or lending money processing method one embodiment of the application;
Fig. 4 shows a kind of flow diagram that the risk score of user is determined in a kind of debt-credit processing method of the application;
Fig. 5 shows the Rnn model trained in a kind of debt-credit processing method of the application for by group's name translation being vector
Flow diagram;
Fig. 6 shows a kind of flow diagram for borrowing or lending money another embodiment of processing method of the application;
Fig. 7 shows a kind of composed structure schematic diagram for borrowing or lending money processing unit one embodiment of the application.
Specific embodiment
The debt-credit processing method of the application can be applied to the loaning bill Shen that network finance platform submits user by terminal
Row processing that come in reduces to the case where loaning bill there are the user of bull lend-borrow action to realize more reasonable lending, improves net
The safety of fund in network financial platform.
Wherein, bull debt-credit refers to that the same borrower proposes the behavior of credit requirements to more financial institutions simultaneously.Such as,
User proposes loaning bill request to financial loan platform 2 to after financial 1 loan application of loan platform, then the user just belongs to
User with bull debt-credit.
If user there are bull lend-borrow action, illustrates the fund of the user, there are wretched insufficiency or the users
In the presence of the risk etc. that malice is borrowed money, in that case, may result in fund to user loaning bill can not be recycled in time even
Irretrievable situation leads to the financial risks such as resources loss.
The scheme of the application in order to facilitate understanding, the one kind being first applicable in the debt-credit processing method of the embodiment of the present application are answered
It is introduced with scene.Such as Fig. 1, may include: in application scenarios shown in Fig. 1
At least one in terminal 101 and network finance platform is for providing the server 102 of network loan service.
Wherein, terminal 101 can be mobile phone, tablet computer, desktop computer etc..
In the embodiment of the present application, for accessing the server 102 in the terminal 101, to send loaning bill Shen to server
Please or other borrow or lend money relevant application.
Such as, user is relevant to the network finance platform using journey by installing in browser in terminal or terminal
Sequence logs in the server, and submits loan application and loaning bill amount etc. information in the related pages that server returns.
Server 102, loan application or other relevant applications of debt-credit for being sent according to terminal, to the terminal
User's qualification is audited, to analyze the condition whether user has loaning bill, and amount for allowing the user to borrow money etc.,
And the processing result to related application is returned to terminal.
Certainly, the loaning bill of other users into the network finance platform which can also submit the user of the terminal
Application is sent to server;Correspondingly, the loan application that server can submit the terminal is transmitted to the terminal of other users,
To complete the debt-credit operation in network finance platform between different user.
Optionally, in the embodiment of the present application, it is additionally provided in the network finance platform: with 102 conjoint data of server
Library 103, for storing the characteristic of user in network finance platform, e.g., with the attribute of the user of logged and
Behavioral data etc..For example, the attribute of user may include: the gender of user, age, occupation, marital status, family status etc.
Information;For another example, behavioral data of the user in network finance platform may include: good friend's distribution, good friend's quantity and good friend
Attribute, user's loan information etc..
Further, it can more true, comprehensively reflect user in the network finance platform in order to subsequent
Feature can also store reference situation, behavioral data of the user got from other network platforms etc. in the database
Etc. user characteristic datas.
In the embodiment of the present application, which is also used to determine the use according to the user characteristics stored in database
There are the risk scores of bull lend-borrow action at family.
Such as, referring to fig. 2, a kind of composition for the server being applicable in it illustrates the debt-credit processing method of the embodiment of the present application
Structural schematic diagram.In Fig. 2, which may include: processor 201, memory 202, communication interface 203, input list
Member 204 and display 205 and communication bus 206.
Processing module 201, communication interface 203, input unit 204, display 205, passes through communication always at memory 202
Line 206 completes mutual communication.
In the embodiment of the present application, the processor 201 can be central processing unit (Central Processing
Unit, CPU), application-specific integrated circuit (application-specific integrated circuit, ASIC), number
Signal processor (DSP), specific integrated circuit (ASIC), ready-made programmable gate array (FPGA) or other programmable logic devices
Part etc..
The processor can call the program stored in memory 202, specifically, can handle device can execute with the following figure
3 operations performed by server side into Fig. 7.
The communication interface 203 can be the interface of communication module, pass through terminal to network finance platform for receiving user
The loan application of submission.
For storing one or more than one program in memory 202, program may include program code, described program
Code includes computer operation instruction, in the embodiment of the present application, is at least stored in the memory for realizing following functions
Program:
In response to the loan application that communication interface receives, access of the user in the network finance platform is obtained
Behavioral data;
Based on preset for characterizing there are the risk behavior data of bull debt-credit risk, the access row of the user is analyzed
Whether characterizing the user for data, there are the risks of bull lend-borrow action;
Analyzing the user, there are when the risk of bull lend-borrow action, obtain to indicate that a variety of of user do not belong to
Multiple user characteristics of property;
According in the risk score model that training obtains in advance, the different user feature that every attribute is included deposits characterization
Multiple user characteristics of the user are converted in the influence degree of bull lend-borrow action, and using the risk score model
For characterizing the user, there are the risk scores of bull lend-borrow action;
According to the risk score, the loan application submitted to the user is handled.
In one possible implementation, which may include storing program area and storage data area, wherein
Storing program area can storage program area, above mentioned program and at least one function (such as sound-playing function,
Image player function etc.) needed for application program etc.;Storage data area can be stored to be created in the use process according to server
Data, for example, audio data, log recording etc..
In addition, memory 202 may include high-speed random access memory, it can also include nonvolatile memory, example
Such as at least one disk memory, flush memory device or other volatile solid-state parts.
The server can also include input unit 205, e.g., keyboard etc. in the application.
The display 204 includes display panel.In the case where a kind of possible, liquid crystal display (Liquid can be used
Crystal Display, LCD), the forms such as Organic Light Emitting Diode (Organic Light-Emitting Diode, OLED)
To configure display panel.
Certainly, server architecture shown in Fig. 2 does not constitute the restriction to server in the embodiment of the present application, is actually answering
It may include than more or fewer components shown in Fig. 2, or the certain components of combination with middle server.
In order to reduce the data processing amount of server, alternatively, can also be wrapped in the network finance platform
Analysis server 104 is included, which can determine that there are the wind of bull lend-borrow action by user in the network finance platform
Danger scoring.Correspondingly, server 102 can obtain different user from Analysis server 104, there are the wind of bull lend-borrow action
Danger scoring.
The composed structure of the Analysis server may refer to the composed structure of server shown in Fig. 2, and details are not described herein.
Below with reference to the above general character of the application, describe in detail to the debt-credit processing method of the embodiment of the present application.
Referring to Fig. 3, it illustrates a kind of flow diagram for borrowing or lending money processing method one embodiment of the application, the present embodiment
Method may include:
S301, terminal send the loan application that user submits to the server of network finance platform.
Wherein, loan application is used for network finance platform request for funds.
User to network finance platform submit the loan application form can there are many, e.g., which can be
Apply making loans or transferring accounts to destiny account, is also possible to advance payment application submitted in user's shopping process etc..
S302, server obtain access behavioral data of the user in the network finance platform in response to the loan application.
S303, server, for characterizing there are the risk behavior data of bull debt-credit risk, analyze the use based on preset
Whether the access behavioral data at family characterizes the user, and there are the risks of bull lend-borrow action.
Wherein, which can characterize the behavioural characteristic that user accesses the network finance platform.And risk row
It is preset there are some behavioral datas that bull borrows or lends money risk for data, which can according to need setting;
It is also possible to by determining to the access behavioural analysis there are the user of bull lend-borrow action in the network finance platform.
Such as, which may include one or more of behavioral data:
The document that user is accessed in the network finance platform, e.g., the article etc. read or downloaded;
The information for the groups of users that user is added in the network finance platform, e.g., the groups of users that user is added
Group name and type etc.;
Lend-borrow action data of the user in the presence of network finance platform, in the user and the network finance platform
Other users between debtor-creditor relationship, borrowing balance, refund situation of loaning bill etc..
For the first behavioral data as above, it is to be understood that the document in network finance platform, which has, much includes
About the article of bull debt-credit, if user, which read or largely downloaded or read, largely borrows or lends money related text with bull
Chapter, then illustrating the user, there are the risks that bull is borrowed or lent money.Such as, current other nets except the network finance platform of user
Network platform application is borrowed money, but the user do not complete refund in the case where, it is desirable to borrowed money, then may by network implementations
Bull can more be paid close attention to and borrow or lend money relevant document.
Correspondingly, preset risk behavior data may include: the document for accessing at least one specified title;It accessed
Document etc. at least one designated key word.Wherein, if there is document the specified title can consider the document category
Relevant document is borrowed or lent money in bull;Correspondingly, the designated key word is keyword relevant to bull debt-credit, if had in document
There is at least one designated key word, then illustrates that the content of the document is related to bull debt-credit.
In the case where the information for the groups of users that the behavioral data of user is added by user, preset risk behavior number
According to can be with are as follows: the group name of preset multiple risk groups.Wherein, risk group is the use for discussing bull lend-borrow action
Family group.It is understood that if the groups of users that user is added largely is for discussing bull lend-borrow action user group
Group, then there is also the risks that bull is borrowed or lent money by the user, therefore, and preset by the groups of users that the user is added
The group name of multiple risk groups is compared, and can detecte at least one added groups of users of the user, if
In the presence of the groups of users for belonging to exchange bull lend-borrow action.
In the case where the user behavior data of user is lend-borrow action data, which be can wrap
Include: there are the debtor-creditor relationships of bull debt-credit risk, debt-credit state etc..Such as, debtor-creditor relationship can be to the network finance platform
In at least one user applied borrow money;Debt-credit state is in the presence of the loaning bill not yet paid off;Or it is specified before current time
To the number of the other users loan application of network finance platform more than preset times in duration, and there is the loaning bill not yet paid off
Etc..It is understood that if the user other users application multiple loaning bill or the user into network finance platform
The loaning bill etc. for currently not yet paying off other users, if the user illustrates that user deposits to the network finance platform loan application
In the risk of bull debt-credit.
Certainly, above to be only introduced by taking several possible situations of preset risk behavior data as an example, but can
With understanding, according to actual needs, which can also have other possibility, correspondingly, the use obtained
Family behavioral data can also have other possible, and details are not described herein.
S304 is analyzing the user there are when the risk of bull lend-borrow action, is obtaining a variety of differences for indicating the user
Multiple user characteristics of attribute.
In the embodiment of the present application, in the case where determining that user has the risk of bull debt-credit, it is also necessary to further divide
Analysing user, there are the degrees of risk that bull is borrowed or lent money, therefore, it is also desirable to further obtain user of the user in a variety of different attributes
Feature.Wherein, different attributes can reflect user in the feature of different aspect, in this application needed for get the number of attribute
Amount and type can be set as needed, wherein with multiple attribute be able to reflect out user there are bull debt-credit risk be
It is quasi-.
The feature performance of user characteristics of the user in some attribute, the actually user in the attribute.Such as, belong to
Property may include: the multiple classifications of social activity, habit, interest, personal information etc.;User of the different user in the same attribute is special
Sign may be different, for example, the educational background of user A can be senior middle school;And the educational background of user B can be undergraduate education.
In one possible implementation, the user characteristics of acquisition can be at least two categories in following a variety of attributes
The user characteristics of property.
Characterize the feature of user's stability: such as, whether the region in residence is stable, stability in use of phone number etc.;
The feature of the essential information of user: such as, the age of user, gender, educational background, occupation, marital status, fertility condition
(such as, if having child, there is several children etc.), house property possess situation etc.;
The feature of the social attribute of user: such as, interaction times, user comment good friend or the network of user and user good friend
The comment number of the message of middle user's publication and comment feature etc.;
The feature of the interest attribute of user: such as, article, movement and song that user likes etc.;
The feature of the habit of user: such as, the behavior patterns of various fixations, for example, have breakfast set time, drink coffee
Set time point etc..
S305, according to the different user feature pair that in the risk score model that training obtains in advance, every attribute is included
There are the influence degrees of bull lend-borrow action for characterization, and are converted to multiple user characteristics of user using the risk score model
For characterizing user, there are the risk scores of bull lend-borrow action.
Wherein, there are the degrees of risk of bull lend-borrow action by the risk score characterization user of user.
Wherein, the concrete form of the risk score can there are many, e.g., the risk score can be characterization user exist it is more
The probability value of head lend-borrow action, probability value is higher, and the user is bigger a possibility that there are bull lend-borrow actions.For another example, the risk
Scoring is also possible to characterize user there are the risk class of bull lend-borrow action, for example, high-risk grade, medium risk grade,
Low risk level etc..For another example, which can also be specific score, and e.g., risk score can be for from 0 to 100 point
Scoring, risk score is higher, then there are the risk of bull lend-borrow action is higher by the user.
Wherein, there is characterization more in the different user feature that training has obtained that every attribute is included in risk score model
The influence degree of head lend-borrow action, therefore, by the risk score model comprehensive analysis, the user is specific in different attribute
User characteristics can analyze out the use of characterization corresponding to multiple user characteristics of the user to the influence degree of bull lend-borrow action
There are the risk scores of bull debt-credit degree of risk at family.
Such as, in a kind of possible situation, in advance in trained risk score model, the difference that every attribute is included is used
There are the influence degrees of bull lend-borrow action can be to characterization for family feature, distributes weight for different attribute, and be directed to every kind
Property different user feature distribute scoring.Correspondingly, after the multiple user characteristics for a variety of different attributes for obtaining user, base
In the scoring of the user characteristics of a variety of respective weights of attribute and every attribute, the corresponding scoring of the user can be calculated and added
Quan He, so as to obtain reflecting the user, there are the risk scores of bull lend-borrow action.Such as, it is assumed that the weight of attribute A is
0.6, and the scoring of user characteristics a1 is 10 points, the scoring of user characteristics a2 is 6 points in attribute A;The weight of attribute B is 0.4, and
The scoring of user characteristics b1 in attribute B is 8 points, and the scoring of user characteristics b2 is 4 points, if the user characteristics packet got
Include: user characteristics a2 and user characteristics b1, then the scoring of the user can be 6*0.6+8*0.4=8.4.
Certainly, in practical applications, training has obtained the different user spy that every attribute is included in risk score model
Sign to characterization there are the specific manifestation form of the influence degree of bull lend-borrow action, may than example from above possibility situation more
Therefore the form of expression of the different user feature in risk model to the influence degree of bull lend-borrow action is not subject to for complexity
Limitation.
In the embodiment of the present application, the training risk score model can use that multiple there are the positive samples of bull lend-borrow action
This user and multiple train there is no the respective user characteristics of the negative sample user of bull lend-borrow action obtain.Wherein, risk
The model of Rating Model constitute it is different, training process can difference, using risk score model as deep neural network model
For (Deep Neural Network, DNN), determine it is multiple there are the positive sample user of bull lend-borrow action and more
It is a there is no after the negative sample user of bull lend-borrow action, the process of the training risk score model can be such that
Multiple positive sample user characteristics of each positive sample user are obtained respectively and the multiple of each negative sample user bear
Sample of users feature, wherein multiple positive sample user characteristics indicate multiple user characteristics of the different attribute of positive sample user,
Multiple negative sample user characteristics indicate multiple user characteristics of the different attribute of negative sample user;
The first score value is set by the risk score of positive sample user, and sets for the risk score of negative sample user
Two score values, first score value are greater than second score value;
Utilize multiple positive sample user characteristics of positive sample user and the first score value and multiple negative samples of negative sample user
This user characteristics and the second score value, training deep neural network model, using the deep neural network model trained as
Risk score model, wherein the deep neural network model trained can reflect out the different user that every attribute is included
There are the influence degrees of bull lend-borrow action to characterization for feature.
For example, can setting positive sample user, there are the probability of bull lend-borrow action so that risk score is probability value as an example
It is 1, and negative sample user is 0 there are the probability of bull lend-borrow action, and respectively using positive sample user and negative sample user
Multiple user characteristics training the DNN model so that the DNN model can the user based on input in a variety of different attributes
In multiple user characteristics, exporting the user, there are the probability values of bull lend-borrow action.
Wherein, for the ease of distinguishing, in the application training risk score model and subsequent other models being previously mentioned
In the process, the user that bull lend-borrow action will be present is known as positive sample user, and there will be no the users of bull lend-borrow action to claim
Be negative sample of users.
Certainly, be above be introduced so that risk score model is DNN model as an example, but it is understood that,
In the case that risk score model is other models, the process of the training risk score model is similar to the above training process,
This is repeated no more.
It in the embodiment of the present application, is to calculate the wind of user in real time with server after receiving the loan application
It is introduced for the scoring of danger, but it is understood that, the risk score of user is also possible to by the server or analysis
Server is precalculated and is stored, in this way, after receiving the loan application of the user, it can from multiple users of storage
Risk score in, obtain the risk score of the user.
Whether S306, the risk score that server detects the user are greater than preset risk threshold value.
Wherein, which can be set as needed, e.g., then can be with when risk score is probability value
Risk threshold value is set as 60 percent;For another example, the highest score of risk score is 100 timesharing, and risk threshold value can be 70 points.
S307 is then returned to the terminal when server determines the risk score of the user greater than preset risk threshold value
Refuse the prompt information of the loan application;
If the risk score of the user is greater than the risk threshold value, illustrate that the user belongs to that there are bull lend-borrow actions
The risk of user is larger, and in that case, in order to avoid occurring that the risks such as fund can not be recycled, server can refuse this and borrow
Money application, is not handled loan application, while can return to corresponding prompt information to terminal.
S308, when server determines the risk score of the user less than or equal to preset risk threshold value, then foundation should
Loan application returns to loaning bill response page to the terminal.
Show user there are in the lower situation of the risk of bull lend-borrow action in the risk score of user, server can be with
The loan application is responded, returns to corresponding response page for user, e.g., which can be the prompt of agreement loan application
The page is borrowed money so that user finally determines.For another example, which can also be that the loaning bill detail of loan application fills in page etc.
Deng.
It is understood that step S303 to S305 is only risk score of the server according to user, which is mentioned
A kind of implementation that the loan application of friendship is handled can also have other possible implementations in practical applications.Such as,
Loaning bill amount corresponding to different risk class can be set, in this way, according to the amount of the loan application institute loan application, with
And the risk score of the user, it is comprehensive to determine whether the applied loaning bill amount of user meets the requirements, if do not met, refuse
The loaning bill of the user is authorized.
Certainly, in practical applications, risk of server during handling loan application, in addition to considering the user
Grade can also integrate the reference situation of the user, the loaning bill amount of the user etc. much information, come whether Comprehensive Assessment is refused
The loan application of the exhausted user, it is without restriction herein.
It is understood that can also include: to be deposited based on preset for characterizing in server in the embodiment of the present application
Bull debt-credit risk risk behavior data, analyze the user do not have there are the risks of bull lend-borrow action, then can ring
Loaning bill response page should be returned to terminal in the loan application.
As it can be seen that in the embodiment of the present application, after the server of network finance platform receives the loan application of user, such as
There are bull debt-credits to go according to access behavioral data of the user in the network finance platform and preset characterization user for fruit
For risk behavior data, analyzing the user, there are the risks of bull lend-borrow action, then can further obtain the more of the user
Multiple user characteristics of kind different attribute, and in the risk score model obtained according to preparatory training, every attribute is included
There are the influence degrees of bull lend-borrow action to characterization for different user feature, and determining user, there are the risks of bull lend-borrow action to comment
Point, can identifying the user by the risk score, there are the risks of bull lend-borrow action, since there are bull debt-credits to go by user
Then to illustrate that very big problem occurs in the fund of user, it is likely that can not pay off a loan in time, therefore, according to the risk score pair
Loan application is handled, and due to reducing to there are the users of lend-borrow action to provide the risk borrowed money, and then reduces malice debt
The case where, improve the financing security of internet financial platform.
In order to make it easy to understand, the scheme of the application, the access with the user that gets in the network finance platform below
Behavioral data is the user in the network finance platform for the group name of at least one added groups of users, to this
Application embodiment determines that the process of risk score is introduced, and referring to fig. 4, it illustrates a kind of risks of determining user of the application
The flow diagram of scoring, the present embodiment can be applied to server, comprising:
S401 obtains the group name of user's at least one added groups of users in network finance platform;
Wherein, groups of users can be the social group or multiple use that user is added in the network finance platform
Any exchange group etc. of family composition.
S402, according to the group name of preset multiple risk groups and at least one added user group of the user
The group name of group, is detected at least one added groups of users of the user, if is existed and is belonged to exchange bull debt-credit row
For target user group, if so, S404 is thened follow the steps, if not, thening follow the steps S403;
For the ease of distinguishing, in the embodiment of the present application, exchange bull debt-credit in the added groups of users of user, will be belonged to
The groups of users of behavior is known as target user group.
Wherein, preset risk group is the group for exchanging bull lend-borrow action.
It is understood that for there is certain general character between the group name for the group for exchanging bull lend-borrow action,
Therefore, after the group name of preset multiple risk groups for belonging to exchange bull lend-borrow action, by the way that the user has been added
The group name of groups of users be compared with the group name of the risk group, can analyze out the added use of the user
With the presence or absence of the groups of users for belonging to exchange bull lend-borrow action in the group of family.
Wherein, which can be the common name of the group of some exchange bull lend-borrow actions, specifically
Rule of thumb, or it can be analyzed and be determined by the risk group to exchange bull lend-borrow action, e.g., risk group
Group name may include: profession net loan group, net borrows exchange group, net loan strategy promotes exchange group, P2P circle-exchange of technology, net are borrowed
Manage money matters group etc..
In one implementation, each groups of users added for user, can be by the group of the groups of users
The group name of title and each risk group carries out similarity system design, if the group name of the groups of users and some risk
Similarity is more than preset threshold between the group name of group, then illustrates that the groups of users belongs to the group of exchange bull lend-borrow action
Group.If being more than default there are the similarity of at least one and preset risk group in the added groups of users of the user
The group of threshold value then illustrates that the user joined the target user group that user exchanges bull lend-borrow action.
It wherein, can be in order to calculate the similarity between the group name of groups of users and the group name of risk group
The group name of the groups of users is converted into primary vector, and the group name of risk group is converted into secondary vector, and
The distance between primary vector and secondary vector is calculated to be judged between the two vectors according to the distance between the two vectors
Similarity.Such as, the COS distance between primary vector and secondary vector can be calculated, if the COS distance be greater than it is default away from
From value, then illustrate that a possibility that similarity of primary vector and secondary vector is high, then the groups of users belongs to risk group is high.
Wherein, primary vector and secondary vector are used for the purpose of the vector for converting out the group name of groups of users,
The vector converted out with the group name of risk group distinguishes.
The mode that group name is converted to vector can there are many, as an alternative embodiment, can adopt
The Recognition with Recurrent Neural Network model (Recurrent Neural Networks, RNN) obtained with preparatory training, by risk group
The group name for the groups of users that group name and user are added is converted to corresponding vector.
Such as, referring to Fig. 5, it illustrates the flow diagram trained for group name to be converted to the RNN model of vector,
Its process may include:
S501 obtains the group name of multiple sample groups;
Wherein, sample group can be the groups of users chosen in advance, which can be set as needed, generally
In the case of, there is otherness between the group name of the multiple sample groups selected.
S502, for each sample group, using text depth transformation model word2vec by group's name of sample group
Each character in title is respectively converted into word vector;
Wherein, for the ease of distinguishing, the vector that character (words of such as Chinese character) is converted out is known as word vector.Wherein, word
The dimension for according with the vector converted out can be set as needed, and e.g., dimension can be 100 dimensions etc..
It is to be introduced for character is converted to vector using word2vec herein, but it is understood that utilizing
Words can be converted to the tool of vector by other, and it is also the same that each character in conversion group name is respectively converted into word vector
Suitable for the application, details are not described herein.
The vector that each character is converted out in the group name of each sample group is successively input to RNN model by S503
In being trained, the group name until all sample groups is corresponded to until word vector is input to and is trained in RNN model.
After each character is converted to word vector in the group name of each sample group, it is required to for word vector being input to
It is trained in RNN model, in this way, having carried out multiple training to RNN model by the corresponding word vector of multiple sample group.
Particularly, in practical applications, S502 and S503 can also be repeated, so as to multiple circuit training RNN model,
In, the number of circulation can be set as needed.
After training obtains RNN model, in step S402, any one user group added for user
Group, after the group name of the groups of users is input to the RNN model, the group name that can export the groups of users turns
The primary vector to swap out, meanwhile, the group name of multiple risk groups is input in RNN model respectively, available this is more
Multiple secondary vectors that the group name of a risk group is converted out, calculate separately the primary vector and each secondary vector away from
From (such as COS distance), may determine that according to this distance groups of users group name and each risk group group name it
Between similarity, when there are the risk groups that the similarity of group name and the group name of the groups of users is more than preset threshold
Group, then it is assumed that the group name of the groups of users also belongs to the group name of the group in the presence of exchange bull lend-borrow action, that is, should
Groups of users belongs to risk group.
S403, if at least one added groups of users of user is not admitted to the wind for exchanging bull lend-borrow action
Dangerous group, it is determined that the risk score of user is the scoring of preset low-risk.
If the risk group for belonging to exchange bull lend-borrow action is not added by user, it can be assumed that the user is not belonging to deposit
The use is analyzed without multiple user characteristics of the different attribute further according to user in the risk subscribers of bull lend-borrow action
There are the risk scores such as the probability of bull lend-borrow action or value-at-risk at family, and directly distribute a preset low wind for the user
Danger scoring.
Wherein, the low-risk grade form requisition family there is no bull lend-borrow action risk or the user there are bulls
The risk of debt-credit is lower.The preset low-risk scoring can be set as needed, and e.g., risk score is that there are bull debt-credits to go
For probability when, then low-risk can be scored and be set as probability value less than 10;For another example, risk score is that characterization is deposited
In the risk class of bull lend-borrow action, then the preset low-risk scoring can be low risk level.
S404, in the case where at least one target user group has been added in user, obtain characterize the user it is a variety of not
With multiple user characteristics of attribute;
If it is judged that the user joined the risk group for exchanging bull lend-borrow action, then illustrate that the user exists
The risk of bull lend-borrow action is larger, and therefore, it is necessary to obtain user in multiple user characteristics of different attribute, so as to subsequent into one
Walking the determining user, there are the degrees of risk of bull lend-borrow action.
It is understood that being determined for other access behavioral datas according to user in the network finance platform
User belongs to there are the mode of the risk subscribers of bull lend-borrow action, is applied equally to the embodiment of the present application, no longer superfluous herein
It states.
S405 determines to belong at least one attribute of specified type at least from multiple user characteristics of the user
One user characteristics;
Wherein, the attribute of the specified type include: characteristic value be numeric form attribute or characteristic value can characterize life
At the attribute of sequence or upper-lower hierarchy.Such as, attribute is the interaction times of good friend in user and network finance platform, due to interaction time
Number can be specific numerical value, then " interaction times of user and good friend in network finance platform " this user property just belongs to finger
Determine the attribute of type.
For another example, attribute is the age, then the age can compare age size, and therefore, the age may belong to the specified type
Attribute.
Certainly, the attribute of the specified type can be preset, and after obtaining multiple user characteristics, can compare this
The attribute of multiple corresponding attributes of user characteristics and the preset specified type, so that it is determined which user characteristics gone out
Belong to user characteristics corresponding to the attribute of preset specified type.
S406 is subordinated at least one use of at least one attribute of specified type according to preset effective attribute set
In the feature of family, at least one target user's feature of the effective attribute belonged in effective attribute set is extracted;
Wherein, effective attribute set includes multiple relatively large with the presence or absence of the influence of bull lend-borrow action on user is judged
Multiple user properties.For the ease of distinguishing, attribute included by validity feature set is known as effective attribute, and will be according to effective
Attribute set, the user characteristics for belonging to effective attribute extracted are known as target user's feature.
It is understood that analysis user characteristics can be analyzed from a variety of different attributes, and the attribute of user can be with
There is multiclass, and in these a large amount of attributes, there are some pairs to judge user with the presence or absence of the influence of the risk of bull lend-borrow action
It is larger.In order to reduce data processing amount, and in order to it is subsequent can more accurate judge go out the user there are bull lend-borrow actions
Risk, the embodiment of the present application carried out further extraction to the user characteristics of the attribute of specified type, belonged to extracting
At least one user characteristics of effective attribute in validity feature set.
It is understood that effectively attribute can be by existing to multiple users there are bull lend-borrow action in attribute set
The user characteristics of a variety of different attributes are analyzed to determine.
It is optionally, available that multiple there are the first kind user of bull lend-borrow action (i.e. positive sample user) each leisure is more
Multiple positive sample user characteristics in kind of different attribute and multiple there is no the second class users of bull lend-borrow action (that is, bearing
Sample of users) multiple negative sample user characteristics in each a variety of different attributes of leisure, and based on the multiple of each first kind user
User characteristics and multiple user characteristics of each second class user, training are promoted for extracting the gradient of validity feature set
Decision tree (Gradient Boosting Decision Tree, GBDT) model.
Such as, positive and negative sample is utilized using the second class user as negative sample user using first kind user as positive sample user
This respective multiple user characteristics, the training GBDT model, can excavate in the training process in multiple positive sample user
The general character of user characteristics, to obtain there are some user characteristics common to the user of bull lend-borrow action, these users are special
Sign is exactly validated user feature.
Wherein, validity feature set is contained in the GBDT model trained, correspondingly, by multiple=user of the user
Feature is input in the trained GBDT model, can be gone out the user by the GBDT model extraction and be belonged to validity feature collection
The user characteristics of conjunction.
Optionally, after the user characteristics for extracting the validity feature set, can be indicated by way of vector should
User belongs to the user characteristics of the validity feature set.Correspondingly, it is respective positive and negative samples can be constructed in training GBDT
Vector corresponding to user characteristics, and using the training of vector corresponding to the user characteristics of the positive and negative samples GBDT model, so that
Validated user feature can be extracted by obtaining the GBDT model, and be vector form by validated user Feature Conversion.
Certainly, training GBDT model is only a kind of implementation as the extraction model for extracting validity feature set, is led to
It crosses and trains other models to extract the mode of the validated user feature in validity feature set and be also applied for the application.
It should be noted that step S405 and S406 can be used as a kind of optional step, purpose is merely by extraction
Can more effectively it reflect there are the user characteristics of bull lend-borrow action, to improve the precision of subsequent determining risk score, and
Reduce data processing amount.But it is understood that in order not to consider data processing amount, and the situation not high to required precision
Under, step S405 and S406 can not also be executed, and the user characteristics of user are directly subjected to subsequent feature combination, and hold
The subsequent step S408 of row.
S407 will be not belonging to the specified class according to preset at least one combinations of attributes relationship in multiple user characteristics
In at least one user characteristics of at least one attribute of type and at least one target user's feature, meet combinations of attributes pass
Corresponding at least two user characteristics of two or more attributes of system are combined, the user characteristics after obtaining at least one set of combination
Group;
Wherein, which includes the syntagmatic of at least two attributes.It is understood that certain attributes
User characteristics can be more accurate after being combined reflection there are user's features of bull lend-borrow action, relative to without group
The user characteristics for meeting at least two attributes of the combinations of attributes relationship are combined by the user characteristics that the attribute of conjunction is included
Acquired user characteristics group, can more characterize that there are the features of bull lend-borrow action.Such as, there are multiple users of bull lend-borrow action
There are some general character in the value of the two features of user characteristics A and user characteristics B, therefore, user characteristics A and user characteristics
Feature combination after B combination advantageously can whether there is bull lend-borrow action in accurate judgement user.
Wherein, preset combinations of attributes relationship can be one or more syntagmatics, wrap in every attribute syntagmatic
It includes: the syntagmatic between two attribute that can be combined.Such as, combinations of attributes relationship may include:
The user characteristics of attribute 1 and the user characteristics of attribute 4;
The user characteristics of attribute 2 and the user characteristics of attribute 3.
Then the user characteristics of the user characteristics 1 of attribute 1 and attribute 4 are the spy of the user after forming combination that can be grouped together
Sign group, and the user characteristics of the user characteristics of attribute 2 and attribute 3 can be grouped together to be formed combination after user characteristics group.
It can be combined between user characteristics in order to determine which attribute, it can be to there are bull lend-borrow actions
The user characteristics of multiple first kind users, and, there is no the user characteristics of the second class user of bull lend-borrow action to divide
Analysis, so that it is determined that going out to be suitble to the syntagmatic of combined attribute.
Optionally, the user characteristics of first kind user and the user characteristics of the second class user, training place sense be can use
Disassembler FMM model is known, to learn effectively identify that there are the risk subscribers of bull lend-borrow action out automatically by FMM model
Different attribute between syntagmatic.User characteristics group corresponding to attribute after the combination come out by the FMM model learning
It is a kind of high-order feature combination, more acurrate can reflects the more essential characteristic of user.Such as, by taking the personality of people as an example, strictly
People may have a kind of shallow-layer data behavior;And arbitrarily people may be another data behavior, and should by what is trained
FMM model then can more analyse in depth out the feature combination for the personality essence for being able to reflect people.
It should be noted that if in the case where not executing above step S405 and S406, it can also be directly by the user
Multiple user characteristics of multiple and different attributes be directly inputted to the FMM model.
Certainly, after the user characteristics group after multiple groups combination being determined using the FMM model, may be used also by the FMM model
To export the corresponding vector of user characteristics group after multiple groups combination.
Optionally, in the case where step S406 exports the vector of at least one target user's feature, the application can be with
From the user characteristics (being referred to as discrete features) for the attribute for being not belonging to the specified type in multiple user characteristics, analysis
The biggish two or more user characteristics of correlation out, e.g., the movement of user preferences are to play badminton, the movement with user preferences
It is the biggish user characteristics of correlation for the two user characteristics of playing table tennis, then can be merged into a dimension table in vector
Show, and the more user characteristics of correlation are indicated using the same dimension in identical vector, is not belonging to obtain this
Vector corresponding to the user characteristics of the attribute of specified type is properly termed as dense vector to distinguish.Such as, using embedded
The user characteristics for practising the attribute that embeding model will not belong to specified type are converted to vector, so that in the process of converting vector
In, it is indicated using the user characteristics with correlation as with a dimension, finally obtains the use for being not belonging to the attribute of specified type
Vector corresponding to the feature of family.Wherein, the method for embeding is discrete features to be put into a self-encoding encoder, automatic training
Encoding model out, and self-encoding encoder Autoencoders, are a kind of methods of unsupervised learning, after input feature vector, by feature
Characteristic value as learning objective, one neural network of training, it is ensured that the output valve and input value gap of neural network are minimum, this
The middle layer value of sample neural network, so that it may as the corresponding dense vector of discrete features.
Correspondingly, can will belong to the vector of at least one target user's feature of validity feature in validity feature set,
And vector corresponding to the user characteristics of the attribute for being not belonging to specified type is input to together in FMM model, eventually by
The FMM model exports the corresponding vector of user characteristics group after multiple groups combination.
It should be noted that step S407 is only a kind of optional step, its purpose is to by by user characteristics
It is combined, to obtain more accurate to reflect that there are the substantive characteristics of the user of bull lend-borrow action, but if to essence
Spend it is of less demanding in the case where, step S407 can not also be executed, then execute using user multiple user characteristics directly into
The operation of the subsequent step S408 of row.
S408, according in the risk score model that training obtains in advance, every attribute and combinations of attributes are included not
With user characteristics, to characterization, there are the influence degrees of bull lend-borrow action, and utilize the risk score model by at least one set
It user characteristics group, at least one target user's feature and is not belonging to the user characteristics of attribute of specified type and is converted to for table
Levying the user, there are the risk scores of bull lend-borrow action.
Such as, in the obtained risk score model of training can with preset every attribute and the respective weight of combinations of attributes, with
And the scoring of every attribute different user feature for being included, the scoring for the different user feature group that every attribute combination includes,
Then it at least one set of user characteristics group for having according to the user, at least one the target user's feature extracted and is not belonging to
The user characteristics of the attribute of specified type calculate scoring weighted sum, to obtain the risk score of the user.
It is understood that then can use risk score model will in the case where not executing step S405 and S406
Multiple user characteristics of user and at least one set of user characteristics group are converted to that there are bull lend-borrow actions for characterizing the user
Risk score.As it can be seen that the essence for combining reflection user is special during the risk score of determining user shown in Fig. 4
User characteristics after the combination of property, original user characteristics, and the user characteristics for belonging to validity feature set extracted, and
Using DNN model is trained, it can more accurately learn the depth abstract characteristics to user, so as to more acurrate assessment user presence
The risk probability of bull behavior of lending.
Certainly, above to be introduced so that risk score model is DNN model as an example, but it is understood that, train DNN
Model is only a kind of implementation come the risk score for analyzing user, and risk score model can also have other a variety of possibility,
Such as, in practical applications, the user characteristics of user can also be input in trained classifier, and is determined using classifier
There are the risk scores of bull lend-borrow action by user out.
It is understood that the process of determining risk score shown in Fig. 4, which is also possible to server, is receiving user's
It is executed after loan application.
The scheme of the embodiment of the present application in order to facilitate understanding is introduced below with reference to an application example, is commented with risk
Being divided into user, there are the probability of bull lend-borrow action, and determine user in the presence of more come final using preparatory trained a variety of models
It is introduced for the probability of head lend-borrow action.Such as, referring to Fig. 6, it illustrates a kind of debt-credit processing methods of the application in one kind
The method of flow diagram in application example, the present embodiment may include:
S601, terminal send the loan application that user submits to the server of network finance platform;
S602, server obtain the user added at least one in network finance platform in response to the loan application
The group name of a groups of users;
S603, the RNN model for being used to be converted to group name vector that server by utilizing trains in advance, by the user
The group name of added each groups of users is respectively converted into primary vector, and utilizes RNN model by preset multiple wind
The group name of dangerous group is respectively converted into secondary vector;
S604, any one groups of users added for user, server calculate separately the group of the groups of users
COS distance between the secondary vector corresponding with the group name of each risk group of primary vector corresponding to title;
S605, if there is no the COS distance for being greater than preset threshold, then server determines the added user group of user
There is no the risk groups for exchanging bull lend-borrow action in group, it is determined that the probability that user belongs to bull debt-credit user is 0;
For any one groups of users added for user, if the first of the group name of the groups of users to
COS distance between amount and each secondary vector is respectively less than or is equal to preset threshold, then illustrates that the groups of users is not intended to hand over
Flow the risk group of bull lend-borrow action.
S606, if there is the COS distance for being greater than preset threshold, then server determines the added groups of users of user
The middle risk group existed for exchanging bull lend-borrow action, then obtaining indicates that multiple users of the different attribute of the user are special
Sign;
S607, server determine at least one attribute for belonging to specified type from multiple user characteristics of the user
At least one user characteristics;
Step S606 and S607 may refer to the related introduction of preceding embodiment, and details are not described herein.
At least one user characteristics for belonging at least one attribute of the specified type are input in advance by S608, server
In trained GBDT model, with special by least one user of the GBDT model from least one attribute of the specified type
In sign, at least one the target user's feature for belonging to effective attribute in validity feature set is extracted, and respectively by each target
User characteristics are converted to first eigenvector;
The discrete user of S609, the attribute which is not belonging to the specified type by server by utilizing embeding method are special
Sign is converted to second feature vector;
Wherein, for ease of description and distinguish, the user characteristics of the attribute of specified type will be not belonging in user characteristics
Referred to as discrete user characteristics.
S610, server are special by the second of the first eigenvector of obtained target user's feature and discrete user characteristics
Sign vector is input in preparatory trained FMM model, with by the FMM model that target user's feature and discrete user is special
The user characteristics group after two or more multiple groups combinations being combined in sign, and convert out the multiple groups user characteristics group
Corresponding third feature vector;
For the ease of distinguishing, the vector converted out by user characteristics is known as feature vector by the present embodiment, and will belong to and have
The vector that user characteristics in effect characteristic set are converted out is known as first eigenvector;By the user by being not belonging to specified type
The vector that feature is converted out is known as second feature vector;The vector that user characteristics after combination are converted out is known as third spy
Levy vector.
Obtained first eigenvector, second feature vector and third feature vector are input to preparatory training by S611
In DNN model out, to export the user by the DNN model, there are the probability of bull lend-borrow action.
S612, server handle the loan application according to for characterizing the user there are the probability of bull lend-borrow action.
Wherein, processing loan application can there are the probability of bull lend-borrow action, the letter of credit of the user in conjunction with the user
Condition determines whether to agree to the loan application of the user to integrate, to advantageously reduce to there are the users of bull lend-borrow action to mention
The case where for borrowing money, improves the safety of fund in platform.
A kind of debt-credit processing unit provided in an embodiment of the present invention is introduced below, at a kind of debt-credit described below
Reason device can correspond to each other reference with a kind of above-described debt-credit processing method.
Such as, referring to Fig. 7, it illustrates a kind of composed structure schematic diagram for borrowing or lending money processing unit one embodiment of the application,
The device of the present embodiment may include:
Apply for receiving unit 701, passes through the loan application that terminal is submitted to network finance platform for receiving user;
Behavior acquiring unit 702, for obtaining user in the network finance platform in response to the loan application
Access behavioral data;
Risk assessment unit 703, for being used to characterize based on preset there are the risk behavior data of bull debt-credit risk,
Whether the access behavioral data for analyzing the user characterizes the user there are the risks of bull lend-borrow action;
Feature acquiring unit 704, for analyzing the user there are when the risk of bull lend-borrow action, obtaining is indicated
Multiple user characteristics of a variety of different attributes of the user;
Score determination unit 705, for according in the risk score model that training obtains in advance, every attribute to be included
There are the influence degrees of bull lend-borrow action to characterization for different user feature, and utilize the risk score model by the user
Multiple user characteristics be converted to that there are the risk scores of bull lend-borrow action for characterizing the user;
Apply for processing unit 706, at the loan application for being used to, according to the risk score, submit the user
Reason.
In one implementation, the behavior acquiring unit, comprising:
Group's information acquisition unit, for obtaining the user in the network finance platform in response to the loan application
In at least one added groups of users group name;
The risk assessment unit, comprising:
Group's risk assessment unit, for the group name and at least one described use according to preset multiple risk groups
The respective group name of family group detects at least one added groups of users of the user, if exist and belong to exchange
The target user group of bull lend-borrow action, the risk group are the group for exchanging bull lend-borrow action;And when described
There are the target user groups at least one added described groups of users of user, it is determined that there are bulls by the user
The risk of lend-borrow action.
In the case where a kind of possible, group's risk assessment unit may include:
Group's similarity calculated, for detecting at least one described groups of users, if there are group name and extremely
The similarity of the group name of a few risk group is more than the target user group of preset threshold.
In one implementation, the scoring determination unit, comprising:
First, which scores, determines subelement, in the risk score model for being obtained according to preparatory training, the power of every attribute
Scoring corresponding to the different user feature that weight and every attribute are included, and by the risk score model by the use
The corresponding multiple user characteristics for indicating different attribute in family are converted to that there are the wind of bull lend-borrow action for characterizing the user
Danger scoring.
In one implementation, described device further include:
Rating Model training unit obtains the risk score model for training in the following way:
It determines multiple there are the positive sample user of bull lend-borrow action and multiple there is no the negative samples of bull lend-borrow action
This user;
The multiple positive sample user characteristics for indicating the different attribute of the positive sample user are obtained respectively, and described in expression
Multiple negative sample user characteristics of the different attribute of negative sample user;
The first score value is set by the risk score of the positive sample user, and by the risk score of the negative sample user
It is set as the second score value, first score value is greater than second score value;
It is used using multiple positive sample user characteristics of the positive sample user and first score value and the negative sample
Multiple negative sample user characteristics and second score value at family, training deep neural network model, the depth that will be trained
Neural network model is as the risk score model, wherein the deep neural network model trained can reflect out every kind
There are the influence degrees of bull lend-borrow action to characterization for the different user feature that attribute is included.
In one implementation, described device can also include:
Feature assembled unit, for after the feature acquiring unit gets multiple user characteristics, according to preset
At least one combinations of attributes relationship meets at least the two of the combinations of attributes relationship in multiple user characteristics of the user
Corresponding at least two user characteristics of attribute are combined, and obtain at least one set of user characteristics group, wherein the combinations of attributes
Relationship includes the syntagmatic of at least two attributes, relative to the user characteristics that not combined attribute is included, by meeting
The user characteristics for stating at least two attributes of combinations of attributes relationship are combined acquired user characteristics group, can more characterize in the presence of more
The feature of head lend-borrow action;
Correspondingly, the scoring determination unit, specifically, for being trained in obtained risk score model according to preparatory,
There are the influence degrees of bull lend-borrow action, and benefit to characterizing for the different user feature that every attribute and combinations of attributes are included
Multiple user characteristics of the user and at least one set of user characteristics group are converted into use with the risk score model
In characterizing the user, there are the risk scores of bull lend-borrow action.
It should be noted that all the embodiments in this specification are described in a progressive manner, each embodiment weight
Point explanation is the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other.
For device class embodiment, since it is basically similar to the method embodiment, so being described relatively simple, related place ginseng
See the part explanation of embodiment of the method.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or equipment for including element.
The foregoing description of the disclosed embodiments can be realized those skilled in the art or using the present invention.To this
A variety of modifications of a little embodiments will be apparent for a person skilled in the art, and the general principles defined herein can
Without departing from the spirit or scope of the present invention, to realize in other embodiments.Therefore, the present invention will not be limited
It is formed on the embodiments shown herein, and is to fit to consistent with the principles and novel features disclosed in this article widest
Range.
The above is only the preferred embodiment of the present invention, it is noted that those skilled in the art are come
It says, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications also should be regarded as
Protection scope of the present invention.
Claims (13)
1. a kind of debt-credit processing method characterized by comprising
It receives user and passes through the loan application that terminal is submitted to network finance platform;
In response to the loan application, access behavioral data of the user in the network finance platform is obtained;
Based on preset for characterizing there are the risk behavior data of bull debt-credit risk, the access behavior number of the user is analyzed
According to whether characterizing the user there are the risks of bull lend-borrow action;
The user is being analyzed there are when the risk of bull lend-borrow action, is obtaining a variety of different attributes for indicating the user
Multiple user characteristics;
Exist according to the different user feature that in the risk score model that training obtains in advance, every attribute is included to characterization more
The influence degree of head lend-borrow action, and multiple user characteristics of the user are converted to using the risk score model and are used for
Characterizing the user, there are the risk scores of bull lend-borrow action;
According to the risk score, the loan application submitted to the user is handled.
2. debt-credit processing method according to claim 1, which is characterized in that the acquisition user is flat in the network finance
Access behavioral data in platform, comprising:
Obtain the group name of the user at least one added groups of users in the network finance platform;
It is described to be used to characterize there are the risk behavior data of bull debt-credit risk based on preset, analyze the access row of the user
Whether characterizing the user for data, there are the risks of bull lend-borrow action, comprising:
According to the group name and the respective group name of at least one groups of users of preset multiple risk groups, inspection
It surveys at least one added groups of users of the user, if there is the potential user group for belonging to exchange bull lend-borrow action
Group, the risk group are the group for exchanging bull lend-borrow action;
When there are the target user groups at least one added described groups of users of the user, it is determined that the use
There are the risks of bull lend-borrow action at family.
3. debt-credit processing method according to claim 2, which is characterized in that the detection user is added at least
In one groups of users, if there is the target user group for belonging to exchange bull lend-borrow action, comprising:
It detects at least one described groups of users, if there are the group of group name and at least one risk group names
The similarity of title is more than the target user group of preset threshold.
4. debt-credit processing method according to claim 1, which is characterized in that the foundation risk that training obtains in advance is commented
In sub-model, there are the influence degrees of bull lend-borrow action to characterization for the different user feature that every attribute is included, and utilize
The risk score model multiple user characteristics of the user are converted to be used to characterize the user there are bull debt-credit go
For risk score, comprising:
According to the different use that in the risk score model that training obtains in advance, the weight of every attribute and every attribute are included
Scoring corresponding to the feature of family, and by the risk score model by the user it is corresponding indicate different attribute multiple use
Family Feature Conversion is that there are the risk scores of bull lend-borrow action for characterizing the user.
5. debt-credit processing method according to claim 1, which is characterized in that the risk score model is in the following way
Training obtains:
It determines multiple there are the positive sample user of bull lend-borrow action and multiple there is no the negative sample of bull lend-borrow action use
Family;
The multiple positive sample user characteristics for indicating the different attribute of the positive sample user are obtained respectively, and indicate the negative sample
Multiple negative sample user characteristics of the different attribute of this user;
The first score value is set by the risk score of the positive sample user, and the risk score of the negative sample user is arranged
For the second score value, first score value is greater than second score value;
Utilize multiple positive sample user characteristics of the positive sample user and first score value and the negative sample user
Multiple negative sample user characteristics and second score value, training deep neural network model, by the depth trained nerve
Network model is as the risk score model, wherein the deep neural network model trained can reflect out every attribute
There are the influence degrees of bull lend-borrow action to characterization for the different user feature for being included.
6. debt-credit processing method according to any one of claims 1 to 5, which is characterized in that described in acquisition expression
After multiple user characteristics of a variety of different attributes of user, further includes:
Meet the set of properties in multiple user characteristics of the user according to preset at least one combinations of attributes relationship
Corresponding at least two user characteristics of at least two attributes of conjunction relationship are combined, and obtain at least one set of user characteristics group,
In, the combinations of attributes relationship includes the syntagmatic of at least two attributes, the use for being included relative to not combined attribute
The user characteristics for meeting at least two attributes of the combinations of attributes relationship are combined acquired user characteristics by family feature
Group, can more characterize that there are the features of bull lend-borrow action;
In the foundation risk score model that training obtains in advance, the different user feature that every attribute is included deposits characterization
Multiple user characteristics of the user are converted in the influence degree of bull lend-borrow action, and using the risk score model
For characterizing the user, there are the risk scores of bull lend-borrow action, comprising:
According to the different user feature that in the risk score model that training obtains in advance, every attribute and combinations of attributes are included
To characterization, there are the influence degrees of bull lend-borrow action, and using the risk score model that multiple users of the user are special
Sign and at least one set of user characteristics group are converted to that there are the risk scores of bull lend-borrow action for characterizing the user.
7. a kind of debt-credit processing unit characterized by comprising
Apply for receiving unit, passes through the loan application that terminal is submitted to network finance platform for receiving user;
Behavior acquiring unit, for obtaining access row of the user in the network finance platform in response to the loan application
For data;
Risk assessment unit, for, for characterizing there are the risk behavior data of bull debt-credit risk, analyzing institute based on preset
Whether the access behavioral data for stating user characterizes the user there are the risks of bull lend-borrow action;
Feature acquiring unit, for analyzing the user there are when the risk of bull lend-borrow action, obtaining indicates the use
Multiple user characteristics of a variety of different attributes at family;
Score determination unit, for according in the risk score model that training obtains in advance, the difference that every attribute is included to be used
There are the influence degrees of bull lend-borrow action to characterization for family feature, and utilize the risk score model by the multiple of the user
User characteristics are converted to that there are the risk scores of bull lend-borrow action for characterizing the user;
Apply for processing unit, for according to the risk score, the loan application submitted to the user to be handled.
8. debt-credit processing unit according to claim 7, which is characterized in that the behavior acquiring unit, comprising:
Group's information acquisition unit obtains the user in the network finance platform in response to the loan application
The group name at least one groups of users being added;
The risk assessment unit, comprising:
Group's risk assessment unit, for the group name and at least one described user group according to preset multiple risk groups
The respective group name of group, is detected at least one added groups of users of the user, if is existed and is belonged to exchange bull
The target user group of lend-borrow action, the risk group are the group for exchanging bull lend-borrow action;And work as the user
There are the target user groups at least one added described groups of users, it is determined that there are bull debt-credits by the user
The risk of behavior.
9. debt-credit processing unit according to claim 8, which is characterized in that group's risk assessment unit, comprising:
Group's similarity calculated, for detecting at least one described groups of users, if there are group names and at least one
The similarity of the group name of a risk group is more than the target user group of preset threshold.
10. debt-credit processing unit according to claim 7, which is characterized in that the scoring determination unit, comprising:
First, which scores, determines subelement, for according in the obtained risk score model of training in advance, the weight of every attribute with
And every attribute included different user feature corresponding to scoring, and by the risk score model by the user couple
The multiple user characteristics for the expression different attribute answered are converted to that there are the risks of bull lend-borrow action to comment for characterizing the user
Point.
11. debt-credit processing unit according to claim 7, which is characterized in that described device further include:
Rating Model training unit obtains the risk score model for training in the following way:
It determines multiple there are the positive sample user of bull lend-borrow action and multiple there is no the negative sample of bull lend-borrow action use
Family;
The multiple positive sample user characteristics for indicating the different attribute of the positive sample user are obtained respectively, and indicate the negative sample
Multiple negative sample user characteristics of the different attribute of this user;
The first score value is set by the risk score of the positive sample user, and the risk score of the negative sample user is arranged
For the second score value, first score value is greater than second score value;
Utilize multiple positive sample user characteristics of the positive sample user and first score value and the negative sample user
Multiple negative sample user characteristics and second score value, training deep neural network model, by the depth trained nerve
Network model is as the risk score model, wherein the deep neural network model trained can reflect out every attribute
There are the influence degrees of bull lend-borrow action to characterization for the different user feature for being included.
12. according to the described in any item debt-credit processing units of claim 7 to 11, which is characterized in that further include:
Feature assembled unit, for after the feature acquiring unit gets multiple user characteristics, according to it is preset at least
A kind of combinations of attributes relationship meets at least two categories of the combinations of attributes relationship in multiple user characteristics of the user
Corresponding at least two user characteristics of property are combined, and obtain at least one set of user characteristics group, wherein the combinations of attributes relationship
Syntagmatic including at least two attributes will meet the category relative to the user characteristics that not combined attribute is included
Property syntagmatic at least two attributes user characteristics be combined obtained by user characteristics groups, can more characterize that there are bulls to borrow
The feature of loan behavior;
The scoring determination unit, specifically, for according in the obtained risk score model of training in advance, every attribute and
There are the influence degrees of bull lend-borrow action to characterization for the different user feature that combinations of attributes is included, and are commented using the risk
Multiple user characteristics of the user and at least one set of user characteristics group are converted to and are used to characterize the use by sub-model
There are the risk scores of bull lend-borrow action at family.
13. a kind of server characterized by comprising
Processor, memory and communication interface, wherein the processor, memory and communication interface pass through communication bus phase
Even;
The communication interface passes through the loan application that terminal is submitted to network finance platform for receiving user;
The processor, for calling the program stored in the memory;
For storing program, described program is used for the memory:
In response to the loan application, access behavioral data of the user in the network finance platform is obtained;
Based on preset for characterizing there are the risk behavior data of bull debt-credit risk, the access behavior number of the user is analyzed
According to whether characterizing the user there are the risks of bull lend-borrow action;
The user is being analyzed there are when the risk of bull lend-borrow action, is obtaining a variety of different attributes for indicating the user
Multiple user characteristics;
Exist according to the different user feature that in the risk score model that training obtains in advance, every attribute is included to characterization more
The influence degree of head lend-borrow action, and multiple user characteristics of the user are converted to using the risk score model and are used for
Characterizing the user, there are the risk scores of bull lend-borrow action;
According to the risk score, the loan application submitted to the user is handled.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710380235.1A CN108961032A (en) | 2017-05-25 | 2017-05-25 | Borrow or lend money processing method, device and server |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710380235.1A CN108961032A (en) | 2017-05-25 | 2017-05-25 | Borrow or lend money processing method, device and server |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108961032A true CN108961032A (en) | 2018-12-07 |
Family
ID=64494440
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710380235.1A Pending CN108961032A (en) | 2017-05-25 | 2017-05-25 | Borrow or lend money processing method, device and server |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108961032A (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919414A (en) * | 2019-01-16 | 2019-06-21 | 国家计算机网络与信息安全管理中心 | P2P network loan platform risk analysis method, device and storage medium |
CN110334909A (en) * | 2019-06-04 | 2019-10-15 | 阿里巴巴集团控股有限公司 | A kind of risk management and control method, device and equipment |
CN110334936A (en) * | 2019-06-28 | 2019-10-15 | 阿里巴巴集团控股有限公司 | A kind of construction method, device and the equipment of credit qualification Rating Model |
CN110349009A (en) * | 2019-07-02 | 2019-10-18 | 北京淇瑀信息科技有限公司 | A kind of bull debt-credit violation correction method, apparatus and electronic equipment |
CN110363655A (en) * | 2019-07-02 | 2019-10-22 | 北京淇瑀信息科技有限公司 | Target user's recognition methods, device and electronic equipment based on temporal characteristics |
CN110458686A (en) * | 2019-07-02 | 2019-11-15 | 阿里巴巴集团控股有限公司 | For determining the method and device of debt-credit risk |
CN110826621A (en) * | 2019-11-01 | 2020-02-21 | 北京芯盾时代科技有限公司 | Risk event processing method and device |
CN111401915A (en) * | 2020-04-14 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Data processing method and device |
CN112184427A (en) * | 2020-10-16 | 2021-01-05 | 上海印闪网络科技有限公司 | Method for analyzing loan risk based on operation behavior of user on-line loan recommendation application |
CN112232950A (en) * | 2020-12-10 | 2021-01-15 | 银联商务股份有限公司 | Loan risk assessment method and device, equipment and computer-readable storage medium |
CN113362156A (en) * | 2021-05-26 | 2021-09-07 | 哈尔滨工业大学重庆研究院 | Financial fraud detection and identification system based on Internet of things |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101599164A (en) * | 2009-06-25 | 2009-12-09 | 阿里巴巴集团控股有限公司 | The method and system that the potential network client is passed judgment on |
US20110173116A1 (en) * | 2010-01-13 | 2011-07-14 | First American Corelogic, Inc. | System and method of detecting and assessing multiple types of risks related to mortgage lending |
CN104616198A (en) * | 2015-02-12 | 2015-05-13 | 哈尔滨工业大学 | P2P (peer-to-peer) network lending risk prediction system based on text analysis |
CN106469376A (en) * | 2015-08-20 | 2017-03-01 | 阿里巴巴集团控股有限公司 | A kind of risk control method and equipment |
CN106651190A (en) * | 2016-12-28 | 2017-05-10 | 深圳微众税银信息服务有限公司 | Enterprise risk level assessment method and system |
-
2017
- 2017-05-25 CN CN201710380235.1A patent/CN108961032A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101599164A (en) * | 2009-06-25 | 2009-12-09 | 阿里巴巴集团控股有限公司 | The method and system that the potential network client is passed judgment on |
US20110173116A1 (en) * | 2010-01-13 | 2011-07-14 | First American Corelogic, Inc. | System and method of detecting and assessing multiple types of risks related to mortgage lending |
CN104616198A (en) * | 2015-02-12 | 2015-05-13 | 哈尔滨工业大学 | P2P (peer-to-peer) network lending risk prediction system based on text analysis |
CN106469376A (en) * | 2015-08-20 | 2017-03-01 | 阿里巴巴集团控股有限公司 | A kind of risk control method and equipment |
CN106651190A (en) * | 2016-12-28 | 2017-05-10 | 深圳微众税银信息服务有限公司 | Enterprise risk level assessment method and system |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919414A (en) * | 2019-01-16 | 2019-06-21 | 国家计算机网络与信息安全管理中心 | P2P network loan platform risk analysis method, device and storage medium |
CN110334909A (en) * | 2019-06-04 | 2019-10-15 | 阿里巴巴集团控股有限公司 | A kind of risk management and control method, device and equipment |
CN110334909B (en) * | 2019-06-04 | 2023-11-21 | 阿里巴巴集团控股有限公司 | Risk management and control method, device and equipment |
CN110334936B (en) * | 2019-06-28 | 2023-09-29 | 创新先进技术有限公司 | Method, device and equipment for constructing credit qualification scoring model |
CN110334936A (en) * | 2019-06-28 | 2019-10-15 | 阿里巴巴集团控股有限公司 | A kind of construction method, device and the equipment of credit qualification Rating Model |
CN110363655A (en) * | 2019-07-02 | 2019-10-22 | 北京淇瑀信息科技有限公司 | Target user's recognition methods, device and electronic equipment based on temporal characteristics |
CN110458686B (en) * | 2019-07-02 | 2023-05-02 | 蚂蚁金服(杭州)网络技术有限公司 | Method and device for determining loan risk |
CN110458686A (en) * | 2019-07-02 | 2019-11-15 | 阿里巴巴集团控股有限公司 | For determining the method and device of debt-credit risk |
CN110349009A (en) * | 2019-07-02 | 2019-10-18 | 北京淇瑀信息科技有限公司 | A kind of bull debt-credit violation correction method, apparatus and electronic equipment |
CN110349009B (en) * | 2019-07-02 | 2024-01-26 | 北京淇瑀信息科技有限公司 | Multi-head lending default prediction method and device and electronic equipment |
CN110826621A (en) * | 2019-11-01 | 2020-02-21 | 北京芯盾时代科技有限公司 | Risk event processing method and device |
CN111401915A (en) * | 2020-04-14 | 2020-07-10 | 支付宝(杭州)信息技术有限公司 | Data processing method and device |
CN112184427A (en) * | 2020-10-16 | 2021-01-05 | 上海印闪网络科技有限公司 | Method for analyzing loan risk based on operation behavior of user on-line loan recommendation application |
CN112232950A (en) * | 2020-12-10 | 2021-01-15 | 银联商务股份有限公司 | Loan risk assessment method and device, equipment and computer-readable storage medium |
CN113362156A (en) * | 2021-05-26 | 2021-09-07 | 哈尔滨工业大学重庆研究院 | Financial fraud detection and identification system based on Internet of things |
CN113362156B (en) * | 2021-05-26 | 2023-10-17 | 哈尔滨工业大学重庆研究院 | Financial fraud detection and identification system based on Internet of Things |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108961032A (en) | Borrow or lend money processing method, device and server | |
TWI712981B (en) | Risk identification model training method, device and server | |
CN104915879B (en) | The method and device that social relationships based on finance data are excavated | |
Liu et al. | " I loan because..." understanding motivations for pro-social lending | |
TW201944305A (en) | Method and apparatus for determining risk probability of service request event | |
CN110717816A (en) | Artificial intelligence technology-based global financial risk knowledge graph construction method | |
CN106611375A (en) | Text analysis-based credit risk assessment method and apparatus | |
CN108182634A (en) | A kind of training method for borrowing or lending money prediction model, debt-credit Forecasting Methodology and device | |
TWI752349B (en) | Risk identification method and device | |
CN113743111B (en) | Financial risk prediction method and device based on text pre-training and multi-task learning | |
CN108009911A (en) | A kind of method of identification P2P network loan borrower's default risks | |
Zhigang et al. | The trend and mechanism of intergenerational income mobility in China: An analysis from the perspective of human capital, social capital and wealth | |
Liu et al. | A graph learning based approach for identity inference in dapp platform blockchain | |
US20230134118A1 (en) | Decentralized social news network website application (dapplication) on a blockchain including a newsfeed, nft marketplace, and a content moderation process for vetted content providers | |
Wang et al. | A conceptual peer review model for arXiv and other preprint databases | |
CN109670927A (en) | The method of adjustment and its device of credit line, equipment, storage medium | |
CN113011884A (en) | Account feature extraction method, device and equipment and readable storage medium | |
Wu et al. | Application analysis of credit scoring of financial institutions based on machine learning model | |
CN115238688A (en) | Electronic information data association relation analysis method, device, equipment and storage medium | |
Yen et al. | Unanswerable question correction in question answering over personal knowledge base | |
WO2019242453A1 (en) | Information processing method and device, storage medium, and electronic device | |
Liu et al. | An innovative model fusion algorithm to improve the recall rate of peer-to-peer lending default customers | |
CN114169439A (en) | Abnormal communication number identification method and device, electronic equipment and readable medium | |
Yangyudongnanxin | Financial credit risk control strategy based on weighted random forest algorithm | |
CN110717817A (en) | Pre-loan approval method and device, electronic equipment and computer-readable storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181207 |
|
RJ01 | Rejection of invention patent application after publication |