CN112651573B - Risk prediction method and device based on deep learning - Google Patents

Risk prediction method and device based on deep learning Download PDF

Info

Publication number
CN112651573B
CN112651573B CN202011639556.7A CN202011639556A CN112651573B CN 112651573 B CN112651573 B CN 112651573B CN 202011639556 A CN202011639556 A CN 202011639556A CN 112651573 B CN112651573 B CN 112651573B
Authority
CN
China
Prior art keywords
behavior
prediction
behaviors
association
topology
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.)
Active
Application number
CN202011639556.7A
Other languages
Chinese (zh)
Other versions
CN112651573A (en
Inventor
王渊
闫森亮
杜林�
刘茜
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jingdong Technology Co Ltd
Original Assignee
Shanghai Jingdong Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Jingdong Technology Co Ltd filed Critical Shanghai Jingdong Technology Co Ltd
Priority to CN202011639556.7A priority Critical patent/CN112651573B/en
Publication of CN112651573A publication Critical patent/CN112651573A/en
Application granted granted Critical
Publication of CN112651573B publication Critical patent/CN112651573B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/067—Enterprise or organisation modelling
    • 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
    • G06Q20/00—Payment architectures, schemes or protocols
    • G06Q20/38—Payment protocols; Details thereof
    • G06Q20/40—Authorisation, e.g. identification of payer or payee, verification of customer or shop credentials; Review and approval of payers, e.g. check credit lines or negative lists
    • G06Q20/401—Transaction verification
    • G06Q20/4016—Transaction verification involving fraud or risk level assessment in transaction processing
    • 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

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Accounting & Taxation (AREA)
  • Quality & Reliability (AREA)
  • Software Systems (AREA)
  • Game Theory and Decision Science (AREA)
  • Operations Research (AREA)
  • Finance (AREA)
  • Tourism & Hospitality (AREA)
  • Mathematical Physics (AREA)
  • Educational Administration (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Computer Security & Cryptography (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Technology Law (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a risk prediction method and a risk prediction device based on deep learning, which organize the relationship of behaviors into association topology, and further realize the conduction prediction of the behaviors based on the association topology, namely, the conduction prediction of next-stage behaviors is realized by utilizing the conduction of a series of behaviors which are associated with each other, so as to form a behavior track prediction vector; furthermore, based on a deep learning mechanism, a risk evaluation model is trained to establish the relationship between the behavior track prediction vector and the risk rating, so that the risk rating prediction of the next stage can be realized according to the behavior track prediction vector of the next stage. Therefore, the method has more comprehensive analysis and risk prediction mechanism aiming at the behavior of the main body, improves the applicability of the prediction mechanism, and establishes high-accuracy prediction on the risk by applying the deep learning mechanism.

Description

Risk prediction method and device based on deep learning
Technical Field
The invention relates to the technical field of behavior prediction, in particular to a risk prediction method and device based on deep learning.
Background
In any application scenario, the perception, prediction and prevention of risks are important aspects. In terms of risk handling, many application scenarios perform risk prediction based on behavior tracks of subjects. For example, in e-commerce and financial evaluations, the risk of default for a current transaction or credit is analyzed based on the transaction history and default records of the subject, subsequent traffic progress security is analyzed based on the subject's behavioral track in transit travel, and so forth.
At present, how to improve the accuracy of risk prediction is a difficult problem. In the prior art, risk prediction is carried out through probability statistics of historical data of a behavior track of a subject, so that the future prediction is actually only carried out by summarizing historical results, and the influence exerted by real-time change of subjective and objective factors of the subject cannot be fully reflected. In addition, the influence degree of each behavior on the risk in the next stage is also different according to the occurrence sequence of each behavior in the behavior trajectory, and the influence degree of each behavior on the risk relationship in the next stage is difficult to take into account in the prior art.
Therefore, how to accurately predict the risk in the next stage based on the behavior trajectory of the subject, reduce the complexity of the prediction mechanism algorithm, and improve the applicability of the prediction mechanism is a problem to be solved by those skilled in the art.
Disclosure of Invention
In view of the above problems, the present invention aims to solve the technical problems of the prior art that the risk prediction accuracy is not high, simple statistics is relied on, and the complex relationship between the subject timing factor and the risk is difficult to effectively reflect and analyze.
The embodiment of the invention provides a risk prediction method based on deep learning, which comprises the following steps:
s101, constructing a behavior track set of a main body, establishing an association topology according to behaviors in the behavior track set, and generating conduction prediction of the behaviors by using behavior states of the behaviors and the association topology;
s102, generating a behavior track prediction vector according to the behavior state and the conduction prediction of the associated topology;
and S103, substituting the behavior track prediction vector into a risk assessment model based on deep learning to realize behavior risk assessment.
Preferably, the constructing a behavior trace set of the subject in step S101, establishing an association topology according to behaviors in the behavior trace set, and generating a conduction prediction of a behavior by using a behavior state of the behavior and the association topology, includes:
s1011, constructing a behavior track set based on behaviors, and constructing a correlation topology facing the behavior track set according to the correlation of mutual influence among the behaviors in the behavior track set and the weight of the correlation influence among the behaviors;
and S1012, acquiring the behavior state in the behavior track set of the user at the previous stage, and according to the behavior state, realizing conduction prediction of the behavior of the user at the next stage through the associated topology.
Preferably, in step S1011, a behavior trace set is formed by a series of behaviors of the user in a time sequence order according to a certain standard condition; denote each behavior of a user as piWhere i 1,2, M, then the set of behavior traces is V { p ═ p1,p2,...,pi,...pM}; constructing an association topology of G ═ { V, E, W } according to the association of the mutual influence between the behaviors in the behavior track set and the weight of the association influence between the behaviors, wherein each behavior p in the behavior track set V isiRepresenting one topology node in the associated topology. In the expression of the associative topology, E ═ E11,e12,...,eij,...eMMWhere i 1, 2.., M, j 1, 2.., M; e.g. of the typeijRepresents a behavior piSubject action pjIs related or not (i.e. action p)iAnd behavior pjWhether or not to have a structure represented by pjTo piIn a unidirectional topology association) of (a), wherein e is the above-mentioned association if presentij1, otherwise, e does not have the above associationij＝0；W＝{w11,w12,...,wij,...wMMWhere i 1, 2.., M, j 1, 2.., M; w is aijRepresenting a behavior p in a behavior trace set ViSubject action pjAssociated weight size, wijW is not less than 0 within a preset value rangeijTaking the internal value of K or less, if piSubject action pjThe greater the relevance of the association, wijThe larger the value of (a).
Preferably, in step S1012, each action piIs denoted as xi，xie.R, R is a set of all behavioral states each behavior may be in; for each action piDetermining the behavior piOne neighborhood of (i.e. from e)ijJ 1,2, M, e is selected fromijBehavior p not equal to 0jCorresponding to the reference j, constitutes a neighborhood denoted NiI.e. the behavior piFor behavior piBehavior state x at the next time periodi(t +1), the specific expression formula is as follows:
Figure BDA0002877949780000031
wherein the content of the first and second substances,
Figure BDA0002877949780000032
representing the current time phase t, behavior piNeighborhood N ofiInner behavior state xj(t) vs. behavior piBehavior state x of itselfiConduction of (t +1) by which p actsiBehavior state x of the next time periodi(t +1) generating a conduction gain, β representing a gain factor, the gain factor β varying according to the type of activity, xi(t +1) relative to xi(t) sending a change in value to transition from one behavior state to another behavior state in the set R.
Preferably, p is selected for all behaviors in the behavior trace set V1,p2,...,pi,...pMThe behavior state x ofiI 1, 2.. M, which in turn forms a behavior trajectory prediction vector X { X ═ X1,x2,...,xi,...xMAnd evaluating a calculation formula of the differentiated state of the behavior track prediction vector, wherein the calculation formula comprises the following steps:
Figure BDA0002877949780000033
d represents a differentiation coefficient. And only if the differentiation coefficient D of the differentiation state accords with a preset value range, the behavior track prediction vector is used for subsequent risk assessment.
Furthermore, the present invention also provides a risk prediction device based on deep learning, including:
the behavior state prediction module is used for constructing a behavior track set of a main body, establishing an association topology according to behaviors in the behavior track set, and generating conduction prediction of the behaviors by using the behavior states of the behaviors and the association topology;
the prediction vector generation module is used for generating a behavior track prediction vector according to the behavior state and the conduction prediction of the associated topology;
and the risk evaluation module substitutes the behavior track prediction vector into a risk evaluation model based on deep learning to realize behavior risk evaluation.
Preferably, the behavior state prediction module includes:
the association topology construction unit is used for constructing a behavior track set based on behaviors, and constructing an association topology facing the behavior track set according to the association of mutual influence among the behaviors in the behavior track set and the weight of the association influence among the behaviors;
and the conduction prediction unit is used for acquiring the behavior state in the behavior track set of the user at the previous stage and realizing the conduction prediction of the behavior of the user at the next stage through the associated topology according to the behavior state.
Preferably, the association topology construction unit constructs a behavior track set by a series of behaviors of the user according to a time sequence order according to a certain standard condition; denote each behavior of a user as piWhere i is 1,2, …, M, then the behavior trace set is V { p ═ p1,p2,…,pi,…pM}; constructing an association topology of G ═ { V, E, W } according to the association of the mutual influence between the behaviors in the behavior track set and the weight of the association influence between the behaviors, wherein each behavior p in the behavior track set V isiRepresenting one topology node in the associated topology. In the expression of the associative topology, E ═ E11,e12,…,eij,…eMMWhere i is 1,2, …, M, j is 1,2, …, M; e.g. of the typeijRepresents a behavior piSubject action pjIs related or not (i.e. action p)iAnd behavior pjWhether or not to have a structure represented by pjTo piIn a unidirectional topology association) of (a), wherein e is the above-mentioned association if presentij1, otherwise, e does not have the above associationij＝0；W＝{w11,w12,...,wij,...wMMWhere i 1, 2.., M, j 1, 2.., M; w is aijRepresenting a behavior p in a behavior trace set ViSubject action pjAssociated weight size, wijW is not less than 0 within a preset value rangeijTaking the internal value of K or less, if piSubject action pjThe greater the relevance of the association, wijThe larger the value of (a).
Preferably, the conduction prediction unit is operative to predict the conduction for each behavior piIs denoted as xi，xie.R, R is a set of all behavioral states that each behavior may be in for each behavior piDetermining the behavior piOne neighborhood of (i.e. from e)ijJ 1,2, M, e is selected fromijBehavior p not equal to 0jCorresponding to the reference j, constitutes a neighborhood denoted NiI.e. the behavior piFor behavior piBehavior state x at the next time periodi(t +1), the specific expression formula is as follows:
Figure BDA0002877949780000051
wherein the content of the first and second substances,
Figure BDA0002877949780000052
representing the current time phase t, behavior piNeighborhood N ofiInner behavior state xj(t) vs. behavior piBehavior state x of itselfiConduction of (t +1) by which p actsiBehavior state x of the next time periodi(t +1) generating a conduction gain, β representing a gain factor, the gain factor β varying according to the type of activity, xi(t +1) relative to xi(t) sending a change in value to transition from one behavior state to another behavior state in the set R.
Preferably, the prediction vector generation module generates a behavior trajectory prediction vector of the set for the behavior state of each behavior in the behavior trajectory set V; for all behaviors in behavior trace set V { p }1,p2,...,pi,…pMThe behavior state x ofiI 1, 2.. M, which in turn forms a behavior trajectory prediction vector X { X ═ X1,x2,…,xi,...xMAnd the calculation formula for evaluating the differentiated states of the behavior trace prediction vectors is as follows:
Figure BDA0002877949780000053
d represents a differentiation coefficient, and the differentiation coefficient D in a differentiation state can be used for subsequent risk assessment if the differentiation coefficient D conforms to a preset value range.
In this embodiment, the mutual relationship of behaviors in the behavior trajectory of the subject is concerned, the relationship of the behaviors is organized into an association topology, and then the conduction prediction of the behaviors is realized based on the association topology, that is, the next-stage behavior is predicted by utilizing the conduction of a series of behaviors associated with each other, so as to form a behavior trajectory prediction vector; furthermore, based on a deep learning mechanism, a risk evaluation model is trained to establish the relationship between the behavior track prediction vector and the risk rating, so that the risk rating prediction of the next stage can be realized according to the behavior track prediction vector of the next stage. Therefore, the method has more comprehensive analysis and risk prediction mechanism aiming at the behavior of the main body, improves the applicability of the prediction mechanism, and establishes high-accuracy prediction on the risk by applying the deep learning mechanism.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
fig. 1 is a flowchart of a risk prediction method based on deep learning according to an embodiment of the present invention;
fig. 2 is a block diagram of a risk prediction apparatus based on deep learning according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
Referring to fig. 1, an embodiment of the present invention provides a risk prediction method based on deep learning, where the method includes:
s101, constructing a behavior track set of a main body, establishing an associated topology according to behaviors in the behavior track set, and generating conduction prediction of the behaviors by using behavior states of the behaviors and the associated topology.
Specifically, an association topology is constructed according to the relationship of interaction and mutual influence of behaviors in the behavior track set, so that conduction prediction of behavior states can be realized based on the association topology.
And S102, generating a behavior track prediction vector according to the behavior state and the conduction prediction of the associated topology.
And S103, substituting the behavior track prediction vector into a risk assessment model based on deep learning to realize behavior risk assessment.
In this embodiment, the mutual relationship of behaviors in the behavior trajectory of the subject is concerned, the relationship of the behaviors is organized into an association topology, and then the conduction prediction of the behaviors is realized based on the association topology, that is, the next-stage behavior is predicted by utilizing the conduction of a series of behaviors associated with each other, so as to form a behavior trajectory prediction vector; furthermore, based on a deep learning mechanism, a risk evaluation model is trained to establish the relationship between the behavior track prediction vector and the risk rating, so that the risk rating prediction of the next stage can be realized according to the behavior track prediction vector of the next stage. Therefore, the method has more comprehensive analysis and risk prediction mechanism aiming at the behavior of the main body, improves the applicability of the prediction mechanism, and establishes high-accuracy prediction on the risk by applying the deep learning mechanism.
In an embodiment, the constructing a behavior trace set of the subject in step S101, establishing an association topology according to behaviors in the behavior trace set, and generating a conduction prediction of the behavior by using a behavior state of the behavior and the association topology includes:
s1011, constructing a behavior track set based on the behaviors, and constructing a correlation topology facing the behavior track set according to the correlation of mutual influence among the behaviors in the behavior track set and the weight of the correlation influence among the behaviors.
Specifically, a series of behaviors of the user are formed into a behavior track set according to a time sequence order according to a certain standard condition. Denote each behavior of a user as piWhere i 1,2, M, then the set of behavior traces is V { p ═ p1,p2,...,pi,...pM}。
Further, according to the relevance of mutual influence among behaviors in the behavior track set and the weight of the relevance influence among the behaviors, constructing a relevance topology of G ═ { V, E, W }, wherein each behavior p in the behavior track set V is piRepresenting one topology node in the associated topology. In the expression of the associative topology, E ═ E11,e12,...,eij,...eMMWhere i 1, 2.., M, j 1, 2.., M; e.g. of the typeijRepresents a behavior piSubject action pjIs related or not (i.e. action p)iAnd behavior pjWhether or not to have a structure represented by pjTo piIn a unidirectional topology association) of (a), wherein e is the above-mentioned association if presentij1, otherwise, e does not have the above associationij0. For i ═ j, i.e. e11，e22…eMMAre all defined as 0. W ═ W11,w12,...,wij,...wMMWhere i 1,21,2,...,M；wijRepresenting a behavior p in a behavior trace set ViSubject action pjThe magnitude of the associated weight (i.e., by p)jTo piWeight size of the unidirectional topological association of), wijW is not less than 0 within a preset value rangeijTaking the internal value of K or less, if piSubject action pjThe greater the relevance of the association, wijThe larger the value of (A) is; for i ═ j, i.e. w11，w22…wMMAre all defined as 0. W ═ W11,w12,...,wij,...wMMThe value in the (w) can be determined according to the specific situation of the application groupijMay be a fixed value, wijOr it may be time-varying (i.e., regular over time), wijOr may be a random value.
And S1012, acquiring the behavior state in the behavior track set of the user at the previous stage, and according to the behavior state, realizing conduction prediction of the behavior of the user at the next stage through the associated topology.
In particular, each action piIs denoted as xi，xiE R, R is a set of all behavioral states that each behavior may be in.
Further, for each action piDetermining the behavior piOne neighborhood of (i.e. from e)ijJ 1,2, M, e is selected fromijBehavior p not equal to 0jCorresponding to the reference j, constitutes a neighborhood denoted NiI.e. the behavior piFor behavior piBehavior state x at the next time periodi(t +1), the specific expression formula is as follows:
Figure BDA0002877949780000081
wherein the content of the first and second substances,
Figure BDA0002877949780000082
representing the current time phase t, behavior piNeighborhood N ofiInside ofBehavioral state xj(t) vs. behavior piBehavior state x of itselfiConduction of (t +1) by which p actsiBehavior state x of the next time periodi(t +1) generating a conduction gain, β representing a gain factor, the gain factor β varying according to the type of activity, xi(t +1) relative to xi(t) sending a change in value, it is possible to transition from one behavior state to another behavior state in the set R.
S102, aiming at the behavior state of each behavior in the behavior trajectory set V, generating a behavior trajectory prediction vector of the set.
Specifically, the behavior trace set V is aimed at all behaviors { p }in the behavior trace set V1,p2,...,pi,...pMThe behavior state x ofiI 1, 2.. M, which in turn forms a behavior trajectory prediction vector X { X ═ X1,x2,...,xi,...xM}. Furthermore, the calculation formula for evaluating the differentiation state of the behavior trajectory prediction vector is as follows:
Figure BDA0002877949780000083
d represents a differentiation coefficient. Only if the differentiation coefficient D of the differentiation state accords with the preset value range, the behavior track prediction vector X is shown as { X ═ X1,x2,...,xi,…xMWith sufficient characterization of the user, it can be used for subsequent risk assessment.
And S103, substituting the behavior track prediction vector into a risk assessment model based on deep learning to realize behavior risk assessment. The risk assessment model is based on a deep learning mechanism, the risk assessment model can be trained according to the behavior track sample vector of the sample user and the sample risk level of the sample user, so that the risk assessment model can be trained to predict the vector based on the input behavior track, the risk assessment value of the user is output, and the risk assessment model can adopt a BP neural network, an SVM support vector machine and the like. After training, the behavior trajectory prediction vector X of step S102 is set to { X ═ X1,x2,...,xi,...xMSubstituting into theAnd the model is used for obtaining a risk evaluation value for the user.
Further, a risk prediction apparatus based on deep learning according to an embodiment of the present invention, as shown in fig. 2, includes:
and the behavior state prediction module is used for constructing a behavior track set of the main body, establishing an associated topology according to the behaviors in the behavior track set, and generating conduction prediction of the behaviors by using the behavior states of the behaviors and the associated topology. Specifically, the behavior state prediction module constructs an association topology from the relationship of interaction and interaction of behaviors in the behavior trajectory set, so that the conduction prediction of the behavior state can be realized based on the association topology.
And the prediction vector generation module is used for generating a behavior track prediction vector according to the behavior state and the conduction prediction of the associated topology.
And the risk evaluation module substitutes the behavior track prediction vector into a risk evaluation model based on deep learning to realize behavior risk evaluation.
In this embodiment, the risk prediction apparatus focuses on the mutual relationship of behaviors in the behavior trajectory of the subject, organizes the relationship of the behaviors into an association topology, and further implements conduction prediction of the behaviors based on the association topology, that is, implements prediction of next-stage behaviors by conduction of a series of behaviors that are associated with each other, to form a behavior trajectory prediction vector; furthermore, based on a deep learning mechanism, a risk evaluation model is trained to establish the relationship between the behavior track prediction vector and the risk rating, so that the risk rating prediction of the next stage can be realized according to the behavior track prediction vector of the next stage. Therefore, the method has more comprehensive analysis and risk prediction mechanism aiming at the behavior of the main body, improves the applicability of the prediction mechanism, and establishes high-accuracy prediction on the risk by applying the deep learning mechanism.
In one embodiment, the behavioral state prediction module includes:
and the association topology construction unit constructs a behavior track set based on the behaviors, and constructs an association topology facing the behavior track set according to the association of mutual influence among the behaviors in the behavior track set and the weight of the association influence among the behaviors.
Specifically, the association topology construction unit constructs a behavior track set by a series of behaviors of the user according to a time sequence order according to a certain standard condition. Denote each behavior of a user as piWhere i 1,2, M, then the set of behavior traces is V { p ═ p1,p2,...,pi,...pM}. Further, according to the relevance of mutual influence among behaviors in the behavior track set and the weight of the relevance influence among the behaviors, constructing a relevance topology of G ═ { V, E, W }, wherein each behavior p in the behavior track set V is piRepresenting one topology node in the associated topology. In the expression of the associative topology, E ═ E11,e12,...,eij,...eMMWhere i 1, 2.., M, j 1, 2.., M; e.g. of the typeijRepresents a behavior piSubject action pjIs related or not (i.e. action p)iAnd behavior pjWhether or not to have a structure represented by pjTo piIn a unidirectional topology association) of (a), wherein e is the above-mentioned association if presentij1, otherwise, e does not have the above associationij0. For i ═ j, i.e. e11，e22…eMMAre all defined as 0. W ═ W11,w12,...,wij,...wMMWhere i 1, 2.., M, j 1, 2.., M; w is aijRepresenting a behavior p in a behavior trace set ViSubject action pjThe magnitude of the associated weight (i.e., by p)jTo piWeight size of the unidirectional topological association of), wijW is not less than 0 within a preset value rangeijTaking the internal value of K or less, if piSubject action pjThe greater the relevance of the association, wijThe larger the value of (A) is; for i ═ j, i.e. w11，w22…wMMAre all defined as 0. W ═ W11,w12,...,wij,...wMMThe value in the (w) can be determined according to the specific situation of the application groupijMay be a fixed value, wijOr it may be time-varying (i.e., regular over time), wijOr may be a random value.
And the conduction prediction unit is used for acquiring the behavior state in the behavior track set of the user at the previous stage and realizing the conduction prediction of the behavior of the user at the next stage through the associated topology according to the behavior state.
In particular, each action piIs denoted as xi，xiE R, R is a set of all behavioral states that each behavior may be in.
Further, for each action piDetermining the behavior piOne neighborhood of (i.e. from e)ijJ 1,2, M, e is selected fromijBehavior p not equal to 0jCorresponding to the reference j, constitutes a neighborhood denoted NiI.e. the behavior piFor behavior piBehavior state x at the next time periodi(t +1), the specific expression formula is as follows:
Figure BDA0002877949780000111
wherein the content of the first and second substances,
Figure BDA0002877949780000112
representing the current time phase t, behavior piNeighborhood N ofiInner behavior state xj(t) vs. behavior piBehavior state x of itselfiConduction of (t +1) by which p actsiBehavior state x of the next time periodi(t +1) generating a conduction gain, β representing a gain factor, the gain factor β varying according to the type of activity, xi(t +1) relative to xi(t) sending a change in value, it is possible to transition from one behavior state to another behavior state in the set R.
And the prediction vector generation module is used for generating a behavior track prediction vector of the set aiming at the behavior state of each behavior in the behavior track set V.
Specifically, the behavior trace set V is aimed at all behaviors { p }in the behavior trace set V1,p2,…,pi,…pMThe behavioral state ofxiI is 1,2, …, M, and then a behavior trajectory prediction vector X is formed { X ═ X }1,x2,…,xi,…xM}. Furthermore, the calculation formula for evaluating the differentiation state of the behavior trajectory prediction vector is as follows:
Figure BDA0002877949780000113
d represents a differentiation coefficient. Only if the differentiation coefficient D of the differentiation state accords with the preset value range, the behavior track prediction vector X is shown as { X ═ X1,x2,...,xi,...xMWith sufficient characterization of the user, it can be used for subsequent risk assessment.
And the risk evaluation module substitutes the behavior track prediction vector into a risk evaluation model based on deep learning to realize behavior risk evaluation. The risk assessment model is based on a deep learning mechanism, the risk assessment model can be trained according to the behavior track sample vector of the sample user and the sample risk level of the sample user, so that the risk assessment model can be trained to predict the vector based on the input behavior track, the risk assessment value of the user is output, and the risk assessment model can adopt a BP neural network, an SVM support vector machine and the like. After training, the behavior trajectory prediction vector X of step S102 is set to { X ═ X1,x2,…,xi,...xMSubstituting the model to obtain a risk rating value for the user.
In this embodiment, the mutual relationship of behaviors in the behavior trajectory of the subject is concerned, the relationship of the behaviors is organized into an association topology, and then the conduction prediction of the behaviors is realized based on the association topology, that is, the next-stage behavior is predicted by utilizing the conduction of a series of behaviors associated with each other, so as to form a behavior trajectory prediction vector; furthermore, based on a deep learning mechanism, a risk evaluation model is trained to establish the relationship between the behavior track prediction vector and the risk rating, so that the risk rating prediction of the next stage can be realized according to the behavior track prediction vector of the next stage. Therefore, the method has more comprehensive analysis and risk prediction mechanism aiming at the behavior of the main body, improves the applicability of the prediction mechanism, and establishes high-accuracy prediction on the risk by applying the deep learning mechanism.

Claims (6)

1. A risk prediction method based on deep learning is characterized by comprising the following steps:
s101, constructing a behavior track set of a main body, establishing an association topology according to behaviors in the behavior track set, and generating conduction prediction of the behaviors by using behavior states of the behaviors and the association topology;
s102, generating a behavior track prediction vector according to the behavior state and the conduction prediction of the associated topology;
s103, substituting the behavior track prediction vector into a risk assessment model based on deep learning to realize behavior risk assessment;
in step S101, the constructing a behavior trace set of the subject, establishing an association topology according to behaviors in the behavior trace set, and generating a conduction prediction of a behavior by using a behavior state of the behavior and the association topology, includes:
s1011, constructing a behavior track set based on behaviors, and constructing a correlation topology facing the behavior track set according to the correlation of mutual influence among the behaviors in the behavior track set and the weight of the correlation influence among the behaviors;
s1012, acquiring behavior states in the behavior track set of the user at the previous stage, and according to the behavior states, realizing conduction prediction of the user at the next stage through the associated topology;
in step S1011, a series of behaviors of the user are formed into a behavior trace set according to a time sequence order according to a certain standard condition; denote each behavior of a user as piWhere i 1,2, M, then the set of behavior traces is V { p ═ p1,p2,...,pi,...pM}; constructing an association topology of G ═ { V, E, W } according to the association of the mutual influence between the behaviors in the behavior track set and the weight of the association influence between the behaviors, wherein each behavior p in the behavior track set V isiRepresenting one topology node in the associated topology; in the expression of the associative topology, E ═ E11,e12,...,eij,...eMMWhere i 1, 2.., M, j 1, 2.., M; e.g. of the typeijRepresents a behavior piSubject action pjIs related or not (i.e. action p)iAnd behavior pjWhether or not to have a structure represented by pjTo piIn a unidirectional topology association) of (a), wherein e is the above-mentioned association if presentij1, otherwise, e does not have the above associationij＝0；W＝{w11,w12,...,wij,...wMMWhere i 1, 2.., M, j 1, 2.., M; w is aijRepresenting a behavior p in a behavior trace set ViSubject action pjAssociated weight size, wijW is not less than 0 within a preset value rangeijTaking the internal value of K or less, if piSubject action pjThe greater the relevance of the association, wijThe larger the value of (a).
2. The deep learning-based risk prediction method according to claim 1, wherein in step S1012, each behavior piIs denoted as xi，xie.R, R is a set of all behavioral states each behavior may be in; for each action piDetermining the behavior piOne neighborhood of (i.e. from e)ijJ 1,2, M, e is selected fromijBehavior p not equal to 0jCorresponding to the reference j, constitutes a neighborhood denoted NiI.e. the behavior piFor behavior piBehavior state x at the next time periodi(t +1), the specific expression formula is as follows:
Figure FDA0003073471120000021
wherein the content of the first and second substances,
Figure FDA0003073471120000022
representing the current time phase t, behavior piNeighborhood N ofiInner behavior state xj(t) vs. behavior piBehavior of itselfState xiConduction of (t +1) by which p actsiBehavior state x of the next time periodi(t +1) generating a conduction gain, β representing a gain factor, the gain factor β varying according to the type of activity, xi(t +1) relative to xi(t) sending a change in value to transition from one behavior state to another behavior state in the set R.
3. The deep learning based risk prediction method according to claim 2, characterized by that for all behaviors { p ] in the behavior trajectory set V1,p2,...,pi,...pMThe behavior state x ofiI 1, 2.. M, which in turn forms a behavior trajectory prediction vector X { X ═ X1,x2,...,xi,...xMAnd evaluating a calculation formula of the differentiated state of the behavior track prediction vector, wherein the calculation formula comprises the following steps:
Figure FDA0003073471120000023
d represents a differentiation coefficient; and only if the differentiation coefficient D of the differentiation state accords with a preset value range, the behavior track prediction vector is used for subsequent risk assessment.
4. A risk prediction device based on deep learning, comprising:
the behavior state prediction module is used for constructing a behavior track set of a main body, establishing an association topology according to behaviors in the behavior track set, and generating conduction prediction of the behaviors by using the behavior states of the behaviors and the association topology;
the prediction vector generation module is used for generating a behavior track prediction vector according to the behavior state and the conduction prediction of the associated topology;
the risk evaluation module substitutes the behavior track prediction vector into a risk evaluation model based on deep learning to realize behavior risk evaluation;
the behavior state prediction module comprises:
the association topology construction unit is used for constructing a behavior track set based on behaviors, and constructing an association topology facing the behavior track set according to the association of mutual influence among the behaviors in the behavior track set and the weight of the association influence among the behaviors;
the conduction prediction unit is used for acquiring the behavior state in the behavior track set of the user at the previous stage and realizing the conduction prediction of the behavior of the user at the next stage through the associated topology according to the behavior state;
the association topology construction unit constructs a series of behaviors of the user according to a time sequence order to form a behavior track set according to a certain standard condition; denote each behavior of a user as piWhere i 1,2, M, then the set of behavior traces is V { p ═ p1,p2,...,pi,...pM}; constructing an association topology of G ═ { V, E, W } according to the association of the mutual influence between the behaviors in the behavior track set and the weight of the association influence between the behaviors, wherein each behavior p in the behavior track set V isiRepresenting one topology node in the associated topology; in the expression of the associative topology, E ═ E11,e12,...,eij,...eMMWhere i 1, 2.., M, j 1, 2.., M; e.g. of the typeijRepresents a behavior piSubject action pjIs related or not (i.e. action p)iAnd behavior pjWhether or not to have a structure represented by pjTo piIn a unidirectional topology association) of (a), wherein e is the above-mentioned association if presentij1, otherwise, e does not have the above associationij＝0；W＝{w11,w12,…,wij,…wMMWhere i 1, 2.., M, j 1, 2.., M; w is aijRepresenting a behavior p in a behavior trace set ViSubject action pjAssociated weight size, wijW is not less than 0 within a preset value rangeijTaking the internal value of K or less, if piSubject action pjThe greater the relevance of the association, wijThe larger the value of (a).
5. The deep learning based risk prediction device of claim 4, wherein conductance is basedPrediction unit for each behavior piIs denoted as xi，xie.R, R is a set of all behavioral states that each behavior may be in for each behavior piDetermining the behavior piOne neighborhood of (i.e. from e)ijJ 1,2, M, e is selected fromijBehavior p not equal to 0jCorresponding to the reference j, constitutes a neighborhood denoted NiI.e. the behavior piFor behavior piBehavior state x at the next time periodi(t +1), the specific expression formula is as follows:
Figure FDA0003073471120000041
wherein the content of the first and second substances,
Figure FDA0003073471120000042
representing the current time phase t, behavior piNeighborhood N ofiInner behavior state xj(t) vs. behavior piBehavior state x of itselfiConduction of (t +1) by which p actsiBehavior state x of the next time periodi(t +1) generating a conduction gain, β representing a gain factor, the gain factor β varying according to the type of activity, xi(t +1) relative to xi(t) sending a change in value to transition from one behavior state to another behavior state in the set R.
6. The deep learning based risk prediction device of claim 5, wherein the prediction vector generation module generates a behavior trace prediction vector for each behavior state of a behavior trace set V; for all behaviors in behavior trace set V { p }1,p2,...,pi,...pMThe behavior state x ofiI 1, 2.. M, which in turn forms a behavior trajectory prediction vector X { X ═ X1,x2,...,xi,...xMAnd evaluating the differentiated state of the behavior trace prediction vectorThe calculation formula of the state is as follows:
Figure FDA0003073471120000043
d represents a differentiation coefficient, and the differentiation coefficient D in a differentiation state can be used for subsequent risk assessment if the differentiation coefficient D conforms to a preset value range.
CN202011639556.7A 2020-12-31 2020-12-31 Risk prediction method and device based on deep learning Active CN112651573B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011639556.7A CN112651573B (en) 2020-12-31 2020-12-31 Risk prediction method and device based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011639556.7A CN112651573B (en) 2020-12-31 2020-12-31 Risk prediction method and device based on deep learning

Publications (2)

Publication Number Publication Date
CN112651573A CN112651573A (en) 2021-04-13
CN112651573B true CN112651573B (en) 2021-07-06

Family

ID=75366963

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011639556.7A Active CN112651573B (en) 2020-12-31 2020-12-31 Risk prediction method and device based on deep learning

Country Status (1)

Country Link
CN (1) CN112651573B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109359673A (en) * 2018-09-25 2019-02-19 佛山科学技术学院 A kind of intelligence manufacture failure prediction method and device based on on-line study
CN109543925A (en) * 2019-01-07 2019-03-29 平安科技（深圳）有限公司 Risk Forecast Method, device, computer equipment and storage medium based on machine learning
CN109816245A (en) * 2019-01-25 2019-05-28 北京海致星图科技有限公司 For conducting assessment system and method to the risk of public credit customer risk early warning
CN110378786A (en) * 2019-07-29 2019-10-25 中国工商银行股份有限公司 Model training method, promise breaking conduction Risk Identification Method, device and storage medium
CN110717824A (en) * 2019-10-17 2020-01-21 北京明略软件***有限公司 Method and device for conducting and calculating risk of public and guest groups by bank based on knowledge graph

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4495669B2 (en) * 2005-12-06 2010-07-07 株式会社日立製作所 Business process design support method and system based on role relationship modeling
US10699335B2 (en) * 2013-05-02 2020-06-30 The Dun & Bradstreet Corporation Apparatus and method for total loss prediction
CN109657918B (en) * 2018-11-19 2023-07-18 平安科技（深圳）有限公司 Risk early warning method and device for associated evaluation object and computer equipment
CN110674970A (en) * 2019-08-19 2020-01-10 广州荔支网络技术有限公司 Enterprise legal risk early warning method, device, equipment and readable storage medium
CN111105311A (en) * 2019-12-19 2020-05-05 第一创业证券股份有限公司 Dynamic credit rating method and device for bond body

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109359673A (en) * 2018-09-25 2019-02-19 佛山科学技术学院 A kind of intelligence manufacture failure prediction method and device based on on-line study
CN109543925A (en) * 2019-01-07 2019-03-29 平安科技（深圳）有限公司 Risk Forecast Method, device, computer equipment and storage medium based on machine learning
CN109816245A (en) * 2019-01-25 2019-05-28 北京海致星图科技有限公司 For conducting assessment system and method to the risk of public credit customer risk early warning
CN110378786A (en) * 2019-07-29 2019-10-25 中国工商银行股份有限公司 Model training method, promise breaking conduction Risk Identification Method, device and storage medium
CN110717824A (en) * 2019-10-17 2020-01-21 北京明略软件***有限公司 Method and device for conducting and calculating risk of public and guest groups by bank based on knowledge graph

Also Published As

Publication number Publication date
CN112651573A (en) 2021-04-13

Similar Documents

Publication Publication Date Title
WO2020037918A1 (en) Risk control strategy determining method based on predictive model, and related device
CN109117380B (en) Software quality evaluation method, device, equipment and readable storage medium
US20210311751A1 (en) Machine-learning models applied to interaction data for determining interaction goals and facilitating experience-based modifications to interface elements in online environments
CN106982196A (en) A kind of abnormal access detection method and equipment
US20200234144A1 (en) Generating training datasets for training neural networks
US11663497B2 (en) Facilitating changes to online computing environment by assessing impacts of actions using a knowledge base representation
JP7059458B2 (en) Generating hostile neuropil-based classification systems and methods
KR102079745B1 (en) Method for training artificial agent, method for recommending user action based thereon, and apparatuses using the same
Wang et al. Using knowledge concept aggregation towards accurate cognitive diagnosis
Cheng et al. Long-term effect estimation with surrogate representation
Song Web service reliability prediction based on machine learning
CN115631008B (en) Commodity recommendation method, device, equipment and medium
CN112651573B (en) Risk prediction method and device based on deep learning
CN111191722A (en) Method and device for training prediction model through computer
CN117632905B (en) Database management method and system based on cloud use records
Mantik et al. A preference elicitation framework for automated planning
CN109993660A (en) The method and system of quantitative analysis and push based on magnanimity financial investment data
CN110399279B (en) Intelligent measurement method for non-human intelligent agent
MacDonell et al. Alternatives to regression models for estimating software projects
CN117952122A (en) Training method of user portrait model, dialogue processing method and related equipment
CN114444698A (en) Information recommendation model training method and device, computer equipment and storage medium
CN114912623A (en) Method and device for model interpretation
Ferrucci et al. Estimating web application development effort using web-cobra and COSMIC: an empirical study
Lian [Retracted] Optimization of Music Teaching Management System for College Students Based on Similarity Distribution Method
Branchi et al. Recommending the optimal policy by learning to act from temporal data

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
GR01 Patent grant
GR01 Patent grant