CN117575596A - Fraud analysis method based on artificial intelligence and digital financial big data system - Google Patents

Fraud analysis method based on artificial intelligence and digital financial big data system Download PDF

Info

Publication number
CN117575596A
CN117575596A CN202311141405.2A CN202311141405A CN117575596A CN 117575596 A CN117575596 A CN 117575596A CN 202311141405 A CN202311141405 A CN 202311141405A CN 117575596 A CN117575596 A CN 117575596A
Authority
CN
China
Prior art keywords
fraud
user behavior
vector
data
kth
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
Application number
CN202311141405.2A
Other languages
Chinese (zh)
Inventor
李建军
陈伟航
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.)
Li Jianjun
Original Assignee
Linyi Wanding Network 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 Linyi Wanding Network Technology Co ltd filed Critical Linyi Wanding Network Technology Co ltd
Priority to CN202311141405.2A priority Critical patent/CN117575596A/en
Publication of CN117575596A publication Critical patent/CN117575596A/en
Pending legal-status Critical Current

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
    • 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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367—Ontology
    • 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
    • G06Q30/00—Commerce
    • G06Q30/018—Certifying business or products
    • G06Q30/0185—Product, service or business identity fraud

Abstract

The embodiment of the application provides an artificial intelligence-based fraud analysis method and a digital financial big data system, which are characterized in that a first fraud tag value estimation strategy is carried out on user behavior directed graph data of a target financial service user to obtain a first fraud tag value of the target financial service user, and the estimated performance value of the first fraud tag value estimation strategy is higher than that of a second fraud tag value estimation strategy, so that the estimated performance value of the fraud tag value estimation strategy is improved, fraud tag deviation value estimation is carried out on the user behavior directed graph data of the target financial service user through a fraud analysis neural network to obtain a fraud tag deviation value of the target financial service user, and deviation value optimization is carried out on the determined first fraud tag value according to the fraud tag deviation value to obtain a second fraud tag value of the target financial service user, namely, the accuracy and the speed of predicting fraud tag values of related users can be considered by adopting the technical scheme of the application.

Description

Fraud analysis method based on artificial intelligence and digital financial big data system
Technical Field
The application relates to the technical field of artificial intelligence, in particular to an artificial intelligence-based fraud analysis method and a digital financial big data system.
Background
Digital finance refers to innovative financial products, business modes, technical application and business processes by utilizing technological methods such as a large-technology platform, large data and cloud computing, and is in face of high-speed development of financial technology, and the customer demands are changeable and the product iteration upgrade becomes a new normal state. According to the needs of the times development, each digital financial service platform actively promotes the innovation of financial science and technology, accelerates the application of technologies such as big data, cloud computing and the like, accelerates the digital transformation, and maximizes the advantages of digital financial services in the aspects of asset management, credit management, payment exchange and the like. However, in the process of providing the digital financial and technical service, there are also many fraudulent users to implement financial fraud through various fraudulent activities, and how to identify these fraudulent users is a technical problem to be solved by the digital financial and technical service platform. For example, in the related art, the user behavior data of the related user is analyzed by combining with the neural network model based on artificial intelligence, so as to predict the fraud tag value (which can also be understood as a fraud risk value) of the related user, however, the accuracy and the speed of predicting the fraud tag value of the related user cannot be considered in the related art.
Disclosure of Invention
In order to at least overcome the above-mentioned shortcomings in the prior art, an object of an embodiment of the present application is to provide an artificial intelligence-based fraud analysis method and a digital financial big data system.
In a first aspect, an embodiment of the present application provides an artificial intelligence based fraud analysis method applied to a digital financial big data system, the method including:
acquiring user behavior directed graph data of a target financial service user;
based on the fraud analysis neural network, fraud label deviation value estimation is carried out on the user behavior directed graph data of the target financial service user, so as to obtain fraud label deviation values of the target financial service user; the fraud analysis neural network is generated by updating network weight parameters by combining fraud tag deviation values of template user behavior directed graph data, wherein the fraud tag deviation values of the template user behavior directed graph data are used for representing fraud tag loss values obtained when fraud tag values of the template user behavior directed graph data are calculated according to two fraud tag value estimation strategies respectively, the two fraud tag value estimation strategies comprise a first fraud tag value estimation strategy and a second fraud tag value estimation strategy, the accuracy score of the first fraud tag value estimation strategy is smaller than the accuracy score of the second fraud tag value estimation strategy, and the estimated performance value of the first fraud tag value estimation strategy is larger than the estimated performance value of the second fraud tag value estimation strategy;
Performing the first fraud tag value estimation strategy on the user behavior directed graph data of the target financial service user to obtain a first fraud tag value of the target financial service user;
and optimizing the deviation value of the first fraud tag value of the target financial service user according to the fraud tag deviation value of the target financial service user to obtain a second fraud tag value of the target financial service user.
In a possible implementation manner of the first aspect, the fraud tag bias value estimating, by the fraud analysis neural network, the user behavior directed graph data of the target financial service user, to obtain a fraud tag bias value of the target financial service user includes:
performing fraud vector description on the user behavior directed graph data based on the fraud analysis neural network to obtain fraud label deviation vectors of the target financial service users;
and estimating the fraud tag deviation value of the fraud tag deviation vector based on the fraud analysis neural network to obtain the fraud tag deviation value of the target financial service user.
In a possible implementation manner of the first aspect, the fraud analysis neural network includes K fraud knowledge description units connected in sequence, and the fraud vector description is performed on the user behavior directed graph data based on the fraud analysis neural network to obtain a fraud tag deviation vector of the target financial service user, including:
Performing basic fraud knowledge vector description on each user behavior data in the user behavior directed graph data to obtain basic fraud knowledge vectors of each user behavior data;
responding K is 1-K-2, carrying out kth fraud vector description on the loading data of kth fraud knowledge description units based on kth fraud knowledge description units in K fraud knowledge description units which are connected in sequence, obtaining kth fraud knowledge vectors of the user behavior data, and loading the kth fraud knowledge vectors to the kth+1 fraud knowledge description units to continue the kth+1 fraud vector description;
responding K to K-1, carrying out fraud time domain vector extraction on the kth fraud knowledge vector of each user behavior data based on the kth+1 fraud knowledge description unit to obtain the kth+1 fraud time domain vector of each user behavior data, carrying out fraud space vector extraction on the kth fraud knowledge vector of each user behavior data based on the kth+1 fraud knowledge description unit to obtain the kth+1 fraud space vector of each user behavior data, and forming the kth+1 fraud time domain vector and the kth+1 fraud space vector of each user behavior data into the fraud tag deviation vector;
Wherein K is more than or equal to 2 and is a positive integer which is gradually increased by 1, and K is more than or equal to 1 and is less than or equal to K-1; and when K is 1, the loading data of the kth fraud knowledge description unit is a basic fraud knowledge vector of the behavior data of each user, and when K is 2-K-1, the loading data of the kth fraud knowledge description unit is a kth-1 fraud knowledge vector of the behavior data of each user output by the kth-1 fraud knowledge description unit.
In a possible implementation manner of the first aspect, the performing a basic fraud knowledge vector description on each user behavior data in the user behavior directed graph data to obtain a basic fraud knowledge vector of each user behavior data includes:
acquiring basic fraud time domain vectors of all user behavior data in the user behavior directed graph data, and acquiring basic fraud airspace vectors of all user behavior data in the user behavior directed graph data; the basic fraud time domain vector is used for reflecting a time sequence vector of the user behavior data, and the basic fraud airspace vector is used for reflecting a space sequence vector of the user behavior data;
the following steps are executed for each piece of user behavior data in the user behavior directed graph data:
Acquiring at least one other user behavior data except the user behavior data in the user behavior directed graph data, and acquiring an initial directed path relation between the user behavior data and each other user behavior data, wherein the initial directed path relation is used for reflecting the behavior flow direction relation between the user behavior data and the other user behavior data;
and forming basic fraud knowledge vectors of the user behavior data by the basic fraud time domain vectors of the user behavior data, the basic fraud airspace vectors of the user behavior data and the initial directed path relation of the user behavior data.
In a possible implementation manner of the first aspect, the K-th fraud knowledge description unit according to the K-th fraud knowledge description units connected in sequence performs a K-th fraud vector description on the loading data of the K-th fraud knowledge description unit to obtain a K-th fraud knowledge vector of each user behavior data, including:
performing the following steps for each of the user behavior data based on the kth fraud knowledge description unit: acquiring other user behavior data except the user behavior data in the user behavior directed graph data;
Performing a first vector mapping on a kth-1 fraudulent knowledge vector of the user behavior data and a kth-1 fraudulent knowledge vector of each other user behavior data to obtain a kth directed relation vector of each other user behavior data corresponding to the user behavior data;
and performing second vector mapping on the kth-1 fraudulent knowledge vector of the user behavior data and the kth directed relation vector of the user behavior data corresponding to each other user behavior data to obtain the kth fraudulent knowledge vector of the user behavior data.
In a possible implementation manner of the first aspect, the performing a first vector mapping on the kth-1 fraudulent knowledge vector of the user behavior data and the kth-1 fraudulent knowledge vector of each of the other user behavior data to obtain a kth directional relation vector of the user behavior data corresponding to each of the other user behavior data includes:
extracting a kth-1 fraudulent airspace vector of the user behavior data, a kth-1 fraudulent time domain vector of the user behavior data, and a kth-1 directed path relationship of the user behavior data from a kth-1 fraudulent knowledge vector of the user behavior data;
Extracting a kth-1 fraudulent spatial vector of each of the other user behavior data and a kth-1 fraudulent time domain vector of each of the other user behavior data from a kth-1 fraudulent knowledge vector of each of the other user behavior data;
the following steps are performed for each of the other user behavior data:
extracting a kth-1 directed path relation of the user behavior data for the other user behavior data from the kth-1 directed path relation of the user behavior data;
acquiring a first vector distance between a kth-1 fraudulent airspace vector of the user behavior data and a kth-1 fraudulent airspace vector of the other user behavior data;
and performing first vector fusion on the square of the first vector distance, the k-1 fraud time domain vector of the user behavior data, the k-1 fraud time domain vector of the other user behavior data and the k-1 directed path relation of the user behavior data to the other user behavior data to obtain the k directed relation vector of the user behavior data corresponding to the other user behavior data.
In a possible implementation manner of the first aspect, the performing the second vector mapping on the kth-1 fraudulent knowledge vector of the user behavior data and the kth directional relation vector of the user behavior data corresponding to each of the other user behavior data to obtain the kth fraudulent knowledge vector of the user behavior data includes:
Fusing the kth directed relation vector of the user behavior data corresponding to a plurality of other user behavior data to obtain the kth directed relation vector of the user behavior data;
performing second vector fusion on the kth-1 fraudulent time domain vector of the user behavior data and the kth directed relation vector of the user behavior data to obtain the kth fraudulent time domain vector of the user behavior data;
acquiring a first vector distance between a kth-1 fraudulent airspace vector of the user behavior data and a kth-1 fraudulent airspace vector of each other user behavior data;
vector mapping is carried out on the kth directed relation vector of the user behavior data corresponding to each other user behavior data, and behavior influence coefficients of each other user behavior data are obtained;
according to the behavior influence coefficients of the other user behavior data, performing weight fusion calculation on the first vector distances of the plurality of other user behavior data to obtain fusion calculation data corresponding to the user behavior data;
fusing the fused calculation data of the user behavior data with the k-1 fraudulent airspace vector of the user behavior data to obtain the k fraudulent airspace vector of the user behavior data;
And taking the initial directed path relation of the user behavior data as a kth directed path relation of the user behavior data, and forming a kth fraud knowledge vector of the user behavior data by the kth directed path relation of the user behavior data, a kth fraud time domain vector of the user behavior data and a kth fraud space vector of the user behavior data.
In a possible implementation manner of the first aspect, the performing, by the k+1-th fraud knowledge description unit, fraud time domain vector extraction on the k-th fraud knowledge vector of each of the user behavior data to obtain a k+1-th fraud time domain vector of each of the user behavior data includes:
extracting a kth fraud airspace vector of the user behavior data, a kth fraud time domain vector of the user behavior data and a kth directed path relation of the user behavior data from kth fraud knowledge vectors of the user behavior data;
for each piece of user behavior data, acquiring other pieces of user behavior data except the user behavior data in the user behavior directed graph data, and executing the following steps for each piece of other pieces of user behavior data:
Extracting a kth directed path relation of the user behavior data aiming at the other user behavior data from the kth directed path relation of the user behavior data, and acquiring a second vector distance between a kth fraudulent airspace vector of the user behavior data and a kth fraudulent airspace vector of the other user behavior data;
performing a first vector fusion on the square of the second vector distance, a kth fraud time domain vector of the user behavior data, a kth fraud time domain vector of the other user behavior data, and a kth directed path relation of the user behavior data with respect to the other user behavior data to obtain a kth+1 directed relation vector of the user behavior data corresponding to the other user behavior data;
fusing the k+1th directional relation vector of the user behavior data corresponding to a plurality of other user behavior data to obtain the k+1th directional relation vector of the user behavior data;
and carrying out second vector fusion on the kth fraud time domain vector of the user behavior data and the kth+1 directional relation vector of the user behavior data to obtain the kth+1 fraud time domain vector of the user behavior data.
In a possible implementation manner of the first aspect, before performing fraud tag bias value estimation on the user behavior directed graph data of the target financial service user based on the fraud analysis neural network to obtain a fraud tag bias value of the target financial service user, the method further includes:
acquiring template user behavior directed graph data, and performing feature derivation on the template user behavior directed graph data to acquire a plurality of template derived behavior directed graph data;
acquiring fraud tag deviation labeling values of the template derived behavior directed graph data;
forward transmitting each template-derived behavior directed graph data in an initial fraud analysis neural network to obtain fraud label deviation prediction values of each template-derived behavior directed graph data;
determining global training cost values according to the fraud tag deviation labeling values of the template-derived behavior directed graph data and the fraud tag deviation predicting values of the template-derived behavior directed graph data;
performing back propagation processing on the global training cost value in the initial fraud analysis neural network to obtain network weight updating information of the initial fraud analysis neural network when the global training cost value converges, and updating the network weight information of the initial fraud analysis neural network according to the network weight updating information;
The obtaining the fraud tag deviation labeling value of each template derived behavior directed graph data comprises the following steps:
the following steps are executed for each template derived behavior directed graph data:
performing a first fraud tag value estimation strategy on the template-derived behavior directed graph data to obtain a first fraud tag value of the template-derived behavior directed graph data; performing a second fraud tag value estimation strategy on the template-derived behavior directed graph data to obtain a second fraud tag value of the template-derived behavior directed graph data;
acquiring a first loss value between a second fraud tag value of the template-derived behavior directed graph data and a first fraud tag value of the template-derived behavior directed graph data as a fraud tag deviation labeling value of the template-derived behavior directed graph data;
the determining the global training cost value according to the fraud tag deviation labeling value of each template derived behavior directed graph data and the fraud tag deviation predicting value of each template derived behavior directed graph data comprises the following steps:
the following steps are executed for each template derived behavior directed graph data:
determining a first root mean square loss value of the template-derived behavior directed graph data according to the fraud tag deviation labeling value of the template-derived behavior directed graph data and the fraud tag deviation predicting value of the template-derived behavior directed graph data;
Acquiring other template-derived behavior directed graph data except for the template-derived behavior directed graph data of the template user behavior directed graph data;
the following steps are performed for each of the other template-derived behavioral directed graph data: determining a second loss value between the fraud tag deviation labeling value of the template-derived directed graph data and the fraud tag deviation labeling value of the other template-derived directed graph data, and determining a third loss value between the fraud tag deviation prediction value of the template-derived directed graph data and the fraud tag deviation prediction value of the other template-derived directed graph data;
performing root mean square calculation on the second loss value and the third loss value to obtain a second root mean square loss value of the other template-derived behavior directed graph data corresponding to the template-derived behavior directed graph data;
fusing the second root mean square loss values of the template-derived behavior directed graph data corresponding to a plurality of other template-derived behavior directed graph data to obtain a third root mean square loss value of the template-derived behavior directed graph data;
and carrying out third vector fusion on the first root mean square loss values of the plurality of template-derived behavior directed graph data and the third root mean square loss values of the plurality of template-derived behavior directed graph data to obtain global training cost values corresponding to the fraud analysis neural network.
In a second aspect, an embodiment of the present application further provides an artificial intelligence-based fraud analysis system, where the artificial intelligence-based fraud analysis system includes a digital financial big data system and a plurality of digital financial service terminals communicatively connected to the digital financial big data system;
the digital financial big data system is used for:
acquiring user behavior directed graph data of a target financial service user;
based on the fraud analysis neural network, fraud label deviation value estimation is carried out on the user behavior directed graph data of the target financial service user, so as to obtain fraud label deviation values of the target financial service user; the fraud analysis neural network is generated by updating network weight parameters by combining fraud tag deviation values of template user behavior directed graph data, wherein the fraud tag deviation values of the template user behavior directed graph data are used for representing fraud tag loss values obtained when fraud tag values of the template user behavior directed graph data are calculated according to two fraud tag value estimation strategies respectively, the two fraud tag value estimation strategies comprise a first fraud tag value estimation strategy and a second fraud tag value estimation strategy, the accuracy score of the first fraud tag value estimation strategy is smaller than the accuracy score of the second fraud tag value estimation strategy, and the estimated performance value of the first fraud tag value estimation strategy is larger than the estimated performance value of the second fraud tag value estimation strategy;
Performing the first fraud tag value estimation strategy on the user behavior directed graph data of the target financial service user to obtain a first fraud tag value of the target financial service user;
and optimizing the deviation value of the first fraud tag value of the target financial service user according to the fraud tag deviation value of the target financial service user to obtain a second fraud tag value of the target financial service user.
In a third aspect, embodiments of the present application also provide a digital financial big data system comprising a processor and a machine-readable storage medium having stored therein a computer program loaded and executed in conjunction with the processor to implement the artificial intelligence based fraud analysis method of the first aspect above.
In a fourth aspect, embodiments of the present application provide a computer-readable storage medium storing computer-executable instructions for execution by a processor to implement the artificial intelligence-based fraud analysis method of the first aspect above.
In a fifth aspect, embodiments of the present application provide a computer program product comprising a computer program or computer executable instructions which, when executed by a processor, implement the artificial intelligence based fraud analysis method of the first aspect above.
The embodiment of the application has at least the following beneficial effects:
the method comprises the steps of carrying out a first fraud tag value estimation strategy on user behavior directed graph data of a target financial service user to obtain a first fraud tag value of the target financial service user, wherein the estimated performance value of the first fraud tag value estimation strategy is higher than that of the second fraud tag value estimation strategy, so that the estimated performance value of the fraud tag value estimation strategy is improved, carrying out fraud tag deviation value estimation on the user behavior directed graph data of the target financial service user through a fraud behavior analysis neural network to obtain a fraud tag deviation value of the target financial service user, carrying out deviation value optimization on the determined first fraud tag value according to the fraud tag deviation value to obtain a second fraud tag value of the target financial service user, and optimizing the determined first fraud tag value according to the AI estimated fraud tag deviation value because the fraud tag deviation value can be used for reflecting the estimated deviation between the second fraud tag value estimation strategy with high accuracy and the first fraud tag value estimation strategy with low accuracy score.
Drawings
Fig. 1 is a schematic flow chart of an artificial intelligence-based fraud analysis method according to an embodiment of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the accompanying drawings, and the described embodiments should not be construed as limiting the present application, and all other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present application.
Unless defined otherwise, all technical and scientific terms used in the examples of this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the embodiments of the application is for the purpose of describing the embodiments of the application only and is not intended to be limiting of the application.
It should be noted that, all kinds of data obtained in the following embodiments are obtained on the basis of obtaining the authorized license of the user.
See fig. 1:
in step S101, user behavior directed graph data of a target financial service user is acquired.
In an alternative embodiment, the target financial service user includes at least one user behavior data, and for the target financial service user having a plurality of user behavior data, the user behavior directed graph data is a knowledge graph structure formed by the plurality of user behavior data and at least one user behavior directed relationship (a behavior a trigger relationship, a behavior linkage relationship, etc.).
In step S102, the fraud tag deviation value estimation is performed on the user behavior directed graph data of the target financial service user based on the fraud behavior analysis neural network, and the fraud tag deviation value of the target financial service user is obtained.
In an alternative embodiment, the fraud vector description is performed on the user behavior directed graph data of the target financial service user by means of a fraud analysis neural network, a fraud tag deviation vector is obtained, and the fraud tag deviation vector is mapped to a fraud tag loss value obtained when the fraud tag value calculation is performed on the target financial service user by means of two preset fraud tag value estimation strategies, the fraud analysis neural network is based on the vector mapping, and the fraud tag deviation value is obtained in combination, the fraud tag value calculation is not actually performed on the target financial service user by means of two given fraud tag value estimation strategies, because the functions that can be implemented by the fraud analysis neural network depend on the network learning mode, the fraud analysis neural network is generated by performing the network weight parameter update in combination with the fraud tag deviation value of the fraud template user behavior directed graph data, the fraud tag deviation value of the template user behavior directed graph data refers to a fraud tag loss value obtained when fraud tag values of the template user behavior directed graph data are calculated according to two fraud tag value estimation strategies respectively, the fraud tag value estimation strategies are used for reflecting a mode of carrying out fraud tag value calculation on given user behavior directed graph data, the two fraud tag value estimation strategies comprise a first fraud tag value estimation strategy and a second fraud tag value estimation strategy, the accuracy score of the first fraud tag value estimation strategy is smaller than that of the second fraud tag value estimation strategy, and the estimated performance value of the first fraud tag value estimation strategy is larger than that of the second fraud tag value estimation strategy.
In step S103, a first fraud tag value estimation policy is performed on the user behavior directed graph data of the target financial service user, to obtain a first fraud tag value of the target financial service user.
In step S104, the bias value optimization is performed on the first bias value of the target financial service user according to the bias value of the bias label of the target financial service user, so as to obtain the second bias value of the bias label of the target financial service user.
In an alternative embodiment, the bias value optimization refers to adding or subtracting the first fraud tag value and the fraud tag bias value, when the fraud tag bias value is used to reflect a fraud tag loss value between the first fraud tag value estimation policy and the second fraud tag value estimation policy, the bias value optimization is to subtract the first fraud tag value and the fraud tag bias value to obtain a second fraud tag value for the target financial service user, and when the fraud tag bias value is used to reflect a fraud tag loss value between the second fraud tag value estimation policy and the first fraud tag value estimation policy, the bias value optimization is to add the first fraud tag value and the fraud tag bias value to obtain a second fraud tag value for the target financial service user.
In an alternative embodiment, the fraud tag deviation value estimation performed on the user behavior directed graph data of the target financial service user based on the fraud analysis neural network in step S102, and obtaining the fraud tag deviation value of the target financial service user may be implemented in steps S1021 to S1022 described below.
In step S1021, the user behavior directed graph data is subjected to fraud vector description through the fraud analysis neural network, and a fraud tag deviation vector of the target financial service user is obtained.
In step S1022, the fraud tag bias vector is subjected to fraud tag bias value estimation by the fraud analysis neural network, and a fraud tag bias value of the target financial service user is obtained.
The fraud tag deviation vector of the user behavior directed graph data is extracted, and the fraud tag deviation value of the target financial service user is predicted according to the fraud tag deviation vector, so that the fraud tag loss value of the fraud tag value of the user behavior directed graph data of the target financial service user can be calculated by combining two fraud tag value estimation strategies.
In an alternative embodiment, the fraud analysis neural network includes K fraud knowledge description units connected in sequence, and in step S1021, fraud vector description is performed on the user behavior directed graph data by the fraud analysis neural network, so as to obtain a fraud tag deviation vector of the target financial service user, which may be implemented based on the following steps a to C.
In the step A, basic fraud knowledge vector description is carried out on each user behavior data in the user behavior directed graph data, and basic fraud knowledge vectors of each user behavior data are obtained.
In an alternative embodiment, when the target financial service user E includes three user behavior data (user behavior data a, user behavior data B, and user behavior data C) and one behavior directed relationship (the user behavior data a and the user behavior data B are connected by the behavior directed relationship), basic fraud knowledge vector description is performed for each user behavior data, a basic fraud time domain vector of each user behavior data in the user behavior directed graph data is obtained, the basic fraud time domain vector is used for reflecting a time sequence vector of the user behavior data, and a basic fraud airspace vector of each user behavior data in the user behavior directed graph data is obtained, and the basic fraud airspace vector is used for reflecting a spatial sequence vector of the user behavior data. The following steps are performed for each user behavior data in the user behavior directed graph data, for example, the following steps are performed for the user behavior data a: obtaining at least one other user behavior data (namely user behavior data B and user behavior data C) except the user behavior data A in the user behavior directed graph data, obtaining an initial directed path relation between the user behavior data and each other user behavior data, wherein the initial directed path relation is used for reflecting a behavior flow direction relation between the user behavior data and the other user behavior data, the initial directed path relation is used for reflecting a behavior flow direction relation between the user behavior data A and the user behavior data B and a behavior flow direction relation between the user behavior data A and the user behavior data C, two behavior flow direction relations exist, and a behavior flow direction relation and a no behavior flow direction relation, if the two user behavior data are connected through the behavior directed relation, the two user behavior data have the behavior flow direction relation, and if the two user behavior data are not connected through the behavior directed relation, the basic fraud time domain vector of each user behavior data, the basic fraud vector of each user behavior data and the initial directed path relation of each user behavior data form the basic fraud time domain vector of each user behavior data.
The basic fraud time domain vector, the initial directed path relationship and the basic fraud airspace vector are formed into a basic fraud knowledge vector, which not only can be used for reflecting the behavior category of the user behavior data, but also can be used for reflecting the relative behavior node of the user behavior data and the connection condition of the directed relationship through behavior, thereby effectively improving the characteristic expression capability of the basic fraud knowledge vector.
In the step B, when K is 1-K-2, carrying out kth fraud vector description on the loading data of the kth fraud knowledge description unit based on the kth fraud knowledge description unit in the K fraud knowledge description units which are connected in sequence to obtain kth fraud knowledge vectors of all user behavior data, and loading the kth fraud knowledge vectors to the kth+1 fraud knowledge description unit to continue carrying out kth+1 fraud vector description.
In an alternative embodiment, K is 2.ltoreq.K, K is a positive integer with a value gradually increasing from 1, and K is 1.ltoreq.k-1; and when K is 1, the loading data of the kth fraud knowledge description unit is a basic fraud knowledge vector of the behavior data of each user, and when K is 2-K-1, the loading data of the kth fraud knowledge description unit is a kth-1 fraud knowledge vector of the behavior data of each user output by the kth-1 fraud knowledge description unit.
In an alternative embodiment, assuming that K has a value of 3, the following steps are still performed taking the user behavior data a as an example: and carrying out 1 st fraud vector description on the basic fraud knowledge vector of the user behavior data A through a 1 st fraud knowledge description unit to obtain a 1 st fraud knowledge vector of the user behavior data A, loading the 1 st fraud knowledge vector into a 2 nd fraud knowledge description unit to continue carrying out 2 nd fraud vector description, and carrying out 2 nd fraud vector description on the 1 st fraud knowledge vector of the user behavior data A through the 2 nd fraud knowledge description unit to obtain a 2 nd vector of the user behavior data A.
In step C, in response to K being K-1, the k+1-th fraud knowledge description unit is used for carrying out fraud time domain vector extraction on the K-th fraud knowledge vector of each user behavior data to obtain the k+1-th fraud time domain vector of each user behavior data, the k+1-th fraud knowledge vector of each user behavior data is used for carrying out fraud space domain vector extraction on the K-th fraud knowledge vector of each user behavior data to obtain the k+1-th fraud space domain vector of each user behavior data, and the k+1-th fraud time domain vector and the k+1-th fraud space domain vector of each user behavior data are combined into a fraud tag deviation vector.
In an alternative embodiment, assuming that K has a value of 3, the following steps are still performed taking the user behavior data a as an example: and 3 rd fraudulent time domain vector extraction is carried out on the 2 nd vector of the user behavior data A through a 3 rd fraudulent knowledge description unit, the 3 rd fraudulent time domain vector of the user behavior data A is obtained, 3 rd fraudulent space vector extraction is carried out on the 2 nd vector of the user behavior data A through the 3 rd fraudulent knowledge description unit, the 3 rd fraudulent space vector of the user behavior data A is obtained, and the 3 rd fraudulent time domain vector of the user behavior data A and the 3 rd fraudulent space vector of the user behavior data A are used as fraud tag deviation vectors.
The fraudulent airspace vector of each user behavior data and the fraudulent time domain vector of each user behavior data can be obtained through iteration steps, and then the fraudulent airspace vectors of a plurality of user behavior data and the fraudulent time domain vectors of a plurality of user behavior data are formed into a fraudulent label deviation vector, so that the vector expression capacity of the fraudulent label deviation vector is improved, and the accuracy score of the follow-up fraudulent label deviation predicted value is improved.
In an alternative embodiment, the kth fraud knowledge description unit in the kth fraud knowledge description unit is used for carrying out kth fraud vector description on the loading data of the kth fraud knowledge description unit in the step B based on the kth fraud knowledge description units in the K fraud knowledge description units, so as to obtain the kth fraud knowledge vector of each user behavior data, and the following steps can be executed on each user behavior data (taking the user behavior data a in the target financial service user E as an example for illustration) through the kth fraud knowledge description unit.
In step B1, other user behavior data than the user behavior data in the user behavior directed graph data, still taking the target financial service user E as an example, are obtained, and the other user behavior data are other user behavior data B than the user behavior data a and other user behavior data C.
In step B2, a first vector mapping is performed on the kth-1 fraudulent knowledge vector of the user behavior data and the kth-1 fraudulent knowledge vector of each other user behavior data to obtain a kth directed relation vector of the user behavior data corresponding to each other user behavior data.
Extracting a kth-1 fraudulent airspace vector of the user behavior data from a kth-1 fraudulent knowledge vector of the user behavior data, and extracting a kth-1 fraudulent airspace vector of each other user behavior data from a kth-1 fraudulent knowledge vector of each other user behavior data; extracting a kth-1 fraudulent time domain vector of the user behavior data from the kth-1 fraudulent knowledge vector of the user behavior data, and extracting a kth-1 fraudulent time domain vector of each other user behavior data from the kth-1 fraudulent knowledge vector of each other user behavior data; extracting a kth-1 directed path relation of the user behavior data from a kth-1 fraud knowledge vector of the user behavior data; the following steps are performed for each other user behavior data (hereinafter, user behavior data B is taken as an example): extracting the kth-1 directed path relation of the user behavior data aiming at other user behavior data from the kth-1 directed path relation of the user behavior data; acquiring a first vector distance between a kth-1 fraudulent airspace vector of user behavior data and a kth-1 fraudulent airspace vector of other user behavior data; and performing first vector fusion on the square of the first vector distance, the k-1 fraud time domain vector of the user behavior data, the k-1 fraud time domain vector of other user behavior data and the k-1 directed path relation of the user behavior data aiming at the other user behavior data to obtain a k directed relation vector of the user behavior data corresponding to the other user behavior data.
The vector distance between any two user behavior data and the behavior flow direction relation of any two user behavior data are combined into the directed relation vector, so that the directed relation vector can learn the global information of the user behavior directed graph data, and the characteristic learning performance of the fraudulent analysis neural network can be improved.
In an alternative embodiment, assuming K is 3, still taking user behavior data a and other user behavior data B as an example, the 2 nd fraudulent airspace vector of user behavior data a is extracted from the 2 nd vector of user behavior data a, the 2 nd fraudulent airspace vector of user behavior data B is extracted from the 2 nd vector of user behavior data B, the 2 nd fraudulent time domain vector of user behavior data is extracted from the 2 nd vector of user behavior data a, and the 2 nd fraudulent time domain vector of user behavior data B is extracted from the 2 nd vector of user behavior data B. The kth-1 directional path relationship is an initial directional path relationship, that is, the directional path relationship is not changed due to vector iteration, the 2 nd directional path relationship of the user behavior data a (the initial directional path relationship of the user behavior data a) is extracted from the 2 nd vector of the user behavior data a, and the 2 nd directional path relationship of the user behavior data (the initial directional path relationship of the user behavior data B) is extracted from the 2 nd vector of the user behavior data B.
In step B3, a second vector mapping is performed on the kth-1 fraudulent knowledge vector of the user behavior data and the kth directed relation vector of the user behavior data corresponding to each other user behavior data, so as to obtain the kth fraudulent knowledge vector of the user behavior data.
Fusing the kth directed relation vector of the user behavior data corresponding to a plurality of other user behavior data to obtain the kth directed relation vector of the user behavior data; performing second vector fusion on the kth-1 fraudulent time domain vector of the user behavior data and the kth directed relation vector of the user behavior data to obtain the kth fraudulent time domain vector of the user behavior data; acquiring a first vector distance between a kth-1 fraudulent airspace vector of user behavior data and a kth-1 fraudulent airspace vector of each other user behavior data; vector mapping is carried out on the kth directed relation vector of the user behavior data corresponding to each other user behavior data, and behavior influence coefficients of each other user behavior data are obtained; according to the behavior influence coefficient of each other user behavior data, performing weight fusion calculation on the first vector distances of the plurality of other user behavior data to obtain fusion calculation data of the corresponding user behavior data; fusing the fused calculation data of the user behavior data with the k-1 fraudulent airspace vector of the user behavior data to obtain the k fraudulent airspace vector of the user behavior data; taking the initial directed path relation of the user behavior data as the kth directed path relation of the user behavior data; and forming a kth fraud knowledge vector of the user behavior data by the kth directed path relation of the user behavior data, the kth fraud time domain vector of the user behavior data and the kth fraud space domain vector of the user behavior data.
By combining the vector distance between any two user behavior data and the behavior flow direction relation of any two user behavior data into the kth fraud knowledge vector, the kth fraud knowledge vector can learn global information of the user behavior directed graph data, and the characteristic learning performance of the fraud behavior analysis neural network can be improved.
In an alternative embodiment, a first vector distance between a kth-1 fraudulent airspace vector of user behavior data A and a kth-1 fraudulent airspace vector of other user behavior data (other user behavior data B and other user behavior data C) is obtained; taking the kth directional relation vector of the user behavior data A corresponding to each other user behavior data (other user behavior data B and other user behavior data C) as a weight, and carrying out weight fusion calculation on first vector distances of a plurality of other user behavior data (other user behavior data B and other user behavior data C) to obtain fusion calculation data corresponding to the user behavior data A; and fusing the fused calculation data of the user behavior data A with the k-1 fraudulent airspace vector of the user behavior data A to obtain the k fraudulent airspace vector of the user behavior data A.
In an alternative embodiment, the initial directed path relation of the user behavior data a is taken as the kth directed path relation of the user behavior data a; and forming a kth fraud knowledge vector of the user behavior data by using the kth directed path relation aij of the user behavior data A corresponding to each other user behavior data, the kth fraud time domain vector of the user behavior data A and the kth fraud space vector of the user behavior data A.
In an alternative embodiment, in step C, the kth fraud knowledge vector of each user behavior data is extracted by the kth+1 fraud knowledge description unit to obtain the kth+1 fraud time domain vector of each user behavior data, which may specifically be: extracting a kth fraud airspace vector of the user behavior data, a kth fraud time domain vector of the user behavior data and a kth directed path relation of the user behavior data from kth fraud knowledge vectors of the user behavior data; the following steps are performed for each user behavior data: acquiring other user behavior data except the user behavior data in the user behavior directed graph data, and executing the following steps aiming at each other user behavior data: extracting a kth directed path relation of the user behavior data aiming at other user behavior data from a kth directed path relation of the user behavior data, and acquiring a second vector distance between a kth fraudulent airspace vector of the user behavior data and a kth fraudulent airspace vector of other user behavior data; performing first vector fusion on the square of the second vector distance, a kth fraud time domain vector of the user behavior data, a kth fraud time domain vector of other user behavior data and a kth directed path relation of the user behavior data aiming at other user behavior data to obtain a kth+1 directed relation vector of the user behavior data corresponding to other user behavior data; fusing the k+1th directional relation vector of the user behavior data corresponding to a plurality of other user behavior data to obtain the k+1th directional relation vector of the user behavior data; and carrying out second vector fusion on the kth fraud time domain vector of the user behavior data and the kth+1 directional relation vector of the user behavior data to obtain the kth+1 fraud time domain vector of the user behavior data.
In an alternative embodiment, the kth fraudulent spatial vector of the user behavior data and the kth directional path relation of the user behavior data are extracted from kth fraudulent knowledge vectors of the user behavior data, specifically, assuming that K has a value of 3, taking the user behavior data a, the user behavior data B and the user behavior data C as an example, the following steps are still performed, the 2 nd fraudulent spatial vector of the user behavior data is extracted from the 2 nd vector of the user behavior data a, the 2 nd fraudulent spatial vector of the user behavior data B is extracted from the 2 nd vector of the user behavior data B, the 2 nd fraudulent spatial vector of the user behavior data C is extracted from the 2 nd vector of the user behavior data C, the 2 nd fraudulent temporal vector of the user behavior data is extracted from the 2 nd vector of the user behavior data B, and the 2 nd fraudulent spatial vector of the user behavior data C is extracted from the 2 nd vector of the user behavior data C. The kth directional path relationship is an initial directional path relationship, that is, the directional path relationship is not changed due to vector iteration, the 2 nd directional path relationship of the user behavior data (the initial directional path relationship of the user behavior data a) is extracted from the 2 nd vector of the user behavior data a, the 2 nd directional path relationship of the user behavior data (the initial directional path relationship of the user behavior data B) is extracted from the 2 nd vector of the user behavior data B, and the 2 nd directional path relationship of the user behavior data (the initial directional path relationship of the user behavior data C) is extracted from the 2 nd vector of the user behavior data C.
In an alternative embodiment, before the fraud tag deviation value estimation is performed on the user behavior directed graph data of the target financial service user based on the fraud analysis neural network, the steps S105 to S109 may be further performed, so as to obtain the fraud tag deviation value of the target financial service user.
In step S105, template user behavior directed graph data is acquired, and feature derivation is performed on the template user behavior directed graph data, so as to obtain a plurality of template derived behavior directed graph data.
In step S106, a fraud tag deviation label value of each template-derived behavior directed graph data is acquired.
In an alternative embodiment, the acquiring the fraud tag deviation label value of each template derived behavior directed graph data in step S106 may specifically be: the following steps are performed for each template-derived behavioral directed graph data: performing a first fraud tag value estimation strategy on the template-derived behavior directed graph data to obtain a first fraud tag value of the template-derived behavior directed graph data; performing a second fraud tag value estimation strategy on the template-derived behavior directed graph data to obtain a second fraud tag value of the template-derived behavior directed graph data; and acquiring a second fraud tag value of the template user behavior directed graph data and a first loss value of a first fraud tag value of the template user behavior directed graph data as fraud tag deviation labeling values of the template derived behavior directed graph data.
In step S107, the template-derived behavior directed graph data is forward-propagated in the initial fraud analysis neural network, and fraud tag deviation prediction values of the template-derived behavior directed graph data are obtained.
In an alternative embodiment, the fraud vector description is performed on the user behavior directed graph data through an initial fraud analysis neural network to obtain a fraud tag deviation vector of the target financial service user; and performing fraud tag deviation value estimation on the fraud tag deviation vector through the initial fraud analysis neural network to obtain a fraud tag deviation value of the target financial service user. Specific implementation of the initial fraud analysis neural network may be found in the foregoing description.
In step S108, the global training cost value of the corresponding fraud analysis neural network is determined according to the fraud label deviation labeling value of each template-derived behaviour directed graph data and the fraud label deviation prediction value of each template-derived behaviour directed graph data.
In an alternative embodiment, in step S108, the global training cost value of the corresponding fraud analysis neural network is determined according to the fraud label deviation labeling value of each template-derived behavior directed graph data and the fraud label deviation prediction value of each template-derived behavior directed graph data, which may specifically be: the following steps are performed for each template-derived behavioral directed graph data: determining a first root mean square loss value of the template-derived directed graph data according to the fraud tag deviation labeling value of the template-derived directed graph data and the fraud tag deviation predicted value of the template-derived directed graph data; acquiring other template-derived behavior directed graph data except for the template-derived behavior directed graph data of the template user behavior directed graph data; the following steps are performed for each other template-derived behavioral directed graph data: determining a second loss value between the fraud tag deviation labeling value of the template-derived behavior directed graph data and the fraud tag deviation labeling value of the other template-derived behavior directed graph data, and determining a third loss value between the fraud tag deviation prediction value of the template-derived behavior directed graph data and the fraud tag deviation prediction value of the other template-derived behavior directed graph data; performing root mean square calculation on the second loss value and the third loss value to obtain a second root mean square loss value of the template-derived behavior directed graph data corresponding to other template-derived behavior directed graph data; fusing second root mean square loss values of the template derived behavior directed graph data corresponding to a plurality of other template derived behavior directed graph data to obtain third root mean square loss values of the template derived behavior directed graph data; and carrying out third vector fusion on the first root mean square loss value of the plurality of template derived behavior directed graph data and the third root mean square loss value of the plurality of template user behavior directed graph data to obtain the global training cost value of the corresponding fraud analysis neural network.
In step S109, the global training cost value is back-propagated in the fraud analysis neural network, so as to obtain the network weight update information of the fraud analysis neural network when the global training cost value converges, and the network weight information of the fraud analysis neural network is updated according to the network weight update information.
In one possible embodiment, a digital financial big data system, which may be a server, is provided that includes a processor, a memory, an Input/Output interface (I/O) and a communication interface. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface is connected to the system bus through the input/output interface. Wherein the processor of the digital financial big data system is configured to provide computing and control capabilities. The memory of the digital financial big data system comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the digital financial big data system is used for storing the data related to the method. The model loading data/output interface of the digital financial big data system is used for exchanging information between the processor and the external equipment. The communication interface of the digital financial big data system is used for communicating with an external terminal through network connection. The computer program, when executed by a processor, implements an artificial intelligence based fraud analysis method.
In some design considerations, a digital financial big data system is provided, which may be a terminal. The digital financial big data system comprises a processor, a memory, an input/output interface, a communication interface, a display unit and an input device. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input/output interface. Wherein the processor of the digital financial big data system is configured to provide computing and control capabilities. The memory of the digital financial big data system comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The model loading data/output interface of the digital financial big data system is used for exchanging information between the processor and the external equipment. The communication interface of the digital financial big data system is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program, when executed by a processor, implements an artificial intelligence based fraud analysis method. The display unit of the digital financial big data system is used for forming a visual picture.
In some design considerations, a digital financial big data system is provided, comprising a memory and a processor, the memory storing a computer program, the processor implementing the steps of the method embodiments described above when executing the computer program.
In some design considerations, a computer readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method embodiments described above.
In some design considerations, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of the method embodiments described above.
The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.

Claims (10)

1. A method for analysis of fraud based on artificial intelligence, the method comprising:
Acquiring user behavior directed graph data of the target financial service user;
based on the fraud analysis neural network, fraud label deviation value estimation is carried out on the user behavior directed graph data of the target financial service user, so as to obtain fraud label deviation values of the target financial service user; the fraud analysis neural network is generated by updating network weight parameters by combining fraud tag deviation values of template user behavior directed graph data, wherein the fraud tag deviation values of the template user behavior directed graph data are used for representing fraud tag loss values obtained when fraud tag values of the template user behavior directed graph data are calculated according to two fraud tag value estimation strategies respectively, the two fraud tag value estimation strategies comprise a first fraud tag value estimation strategy and a second fraud tag value estimation strategy, the accuracy score of the first fraud tag value estimation strategy is smaller than the accuracy score of the second fraud tag value estimation strategy, and the estimated performance value of the first fraud tag value estimation strategy is larger than the estimated performance value of the second fraud tag value estimation strategy;
performing the first fraud tag value estimation strategy on the user behavior directed graph data of the target financial service user to obtain a first fraud tag value of the target financial service user;
And optimizing the deviation value of the first fraud tag value of the target financial service user according to the fraud tag deviation value of the target financial service user to obtain a second fraud tag value of the target financial service user.
2. The artificial intelligence based fraud analysis method of claim 1, wherein the fraud analysis neural network based on fraud label bias estimation for the target financial service user's user behavior directed graph data, obtaining a fraud label bias for the target financial service user, comprises:
performing fraud vector description on the user behavior directed graph data based on the fraud analysis neural network to obtain fraud label deviation vectors of the target financial service users;
and estimating the fraud tag deviation value of the fraud tag deviation vector based on the fraud analysis neural network to obtain the fraud tag deviation value of the target financial service user.
3. The artificial intelligence based fraud analysis method of claim 2, wherein the fraud analysis neural network includes K fraud knowledge description units connected in sequence, and the fraud vector description is performed on the user behavior directed graph data based on the fraud analysis neural network to obtain a fraud tag deviation vector of the target financial service user, including:
Performing basic fraud knowledge vector description on each user behavior data in the user behavior directed graph data to obtain basic fraud knowledge vectors of each user behavior data;
responding K is 1-K-2, carrying out kth fraud vector description on the loading data of kth fraud knowledge description units based on kth fraud knowledge description units in K fraud knowledge description units which are connected in sequence, obtaining kth fraud knowledge vectors of the user behavior data, and loading the kth fraud knowledge vectors to the kth+1 fraud knowledge description units to continue the kth+1 fraud vector description;
responding K to K-1, carrying out fraud time domain vector extraction on the kth fraud knowledge vector of each user behavior data based on the kth+1 fraud knowledge description unit to obtain the kth+1 fraud time domain vector of each user behavior data, carrying out fraud space vector extraction on the kth fraud knowledge vector of each user behavior data based on the kth+1 fraud knowledge description unit to obtain the kth+1 fraud space vector of each user behavior data, and forming the kth+1 fraud time domain vector and the kth+1 fraud space vector of each user behavior data into the fraud tag deviation vector;
Wherein K is more than or equal to 2 and is a positive integer which is gradually increased by 1, and K is more than or equal to 1 and is less than or equal to K-1; and when K is 1, the loading data of the kth fraud knowledge description unit is a basic fraud knowledge vector of the behavior data of each user, and when K is 2-K-1, the loading data of the kth fraud knowledge description unit is a kth-1 fraud knowledge vector of the behavior data of each user output by the kth-1 fraud knowledge description unit.
4. The artificial intelligence based fraud analysis method of claim 3, wherein the performing basic fraud knowledge vector description on each user behavior data in the user behavior directed graph data to obtain a basic fraud knowledge vector of each user behavior data includes:
acquiring basic fraud time domain vectors of all user behavior data in the user behavior directed graph data, and acquiring basic fraud airspace vectors of all user behavior data in the user behavior directed graph data; the basic fraud time domain vector is used for reflecting a time sequence vector of the user behavior data, and the basic fraud airspace vector is used for reflecting a space sequence vector of the user behavior data;
the following steps are executed for each piece of user behavior data in the user behavior directed graph data:
Acquiring at least one other user behavior data except the user behavior data in the user behavior directed graph data, and acquiring an initial directed path relation between the user behavior data and each other user behavior data, wherein the initial directed path relation is used for reflecting the behavior flow direction relation between the user behavior data and the other user behavior data;
and forming basic fraud knowledge vectors of the user behavior data by the basic fraud time domain vectors of the user behavior data, the basic fraud airspace vectors of the user behavior data and the initial directed path relation of the user behavior data.
5. A fraud analysis method based on artificial intelligence according to claim 3, wherein the K-th fraud knowledge description unit of the K-th fraud knowledge description units connected in sequence performs a K-th fraud vector description on the loaded data of the K-th fraud knowledge description unit to obtain a K-th fraud knowledge vector of each of the user behavior data, comprising:
performing the following steps for each of the user behavior data based on the kth fraud knowledge description unit: acquiring other user behavior data except the user behavior data in the user behavior directed graph data;
Performing a first vector mapping on a kth-1 fraudulent knowledge vector of the user behavior data and a kth-1 fraudulent knowledge vector of each other user behavior data to obtain a kth directed relation vector of each other user behavior data corresponding to the user behavior data;
and performing second vector mapping on the kth-1 fraudulent knowledge vector of the user behavior data and the kth directed relation vector of the user behavior data corresponding to each other user behavior data to obtain the kth fraudulent knowledge vector of the user behavior data.
6. The artificial intelligence based fraud analysis method of claim 5, wherein the performing a first vector mapping of the kth-1 fraudulent knowledge vector of the user action data and the kth-1 fraudulent knowledge vector of each of the other user action data to obtain a kth directed relationship vector of the user action data for each of the other user action data includes:
extracting a kth-1 fraudulent airspace vector of the user behavior data, a kth-1 fraudulent time domain vector of the user behavior data, and a kth-1 directed path relationship of the user behavior data from a kth-1 fraudulent knowledge vector of the user behavior data;
Extracting a kth-1 fraudulent spatial vector of each of the other user behavior data and a kth-1 fraudulent time domain vector of each of the other user behavior data from a kth-1 fraudulent knowledge vector of each of the other user behavior data;
the following steps are performed for each of the other user behavior data:
extracting a kth-1 directed path relation of the user behavior data for the other user behavior data from the kth-1 directed path relation of the user behavior data;
acquiring a first vector distance between a kth-1 fraudulent airspace vector of the user behavior data and a kth-1 fraudulent airspace vector of the other user behavior data;
and performing first vector fusion on the square of the first vector distance, the k-1 fraud time domain vector of the user behavior data, the k-1 fraud time domain vector of the other user behavior data and the k-1 directed path relation of the user behavior data to the other user behavior data to obtain the k directed relation vector of the user behavior data corresponding to the other user behavior data.
7. The artificial intelligence based fraud analysis method of claim 6, wherein the performing a second vector mapping of the kth-1 fraud knowledge vector of the user action data and the kth directed relationship vector of the user action data for each of the other user action data to obtain the kth fraud knowledge vector of the user action data includes:
Fusing the kth directed relation vector of the user behavior data corresponding to a plurality of other user behavior data to obtain the kth directed relation vector of the user behavior data;
performing second vector fusion on the kth-1 fraudulent time domain vector of the user behavior data and the kth directed relation vector of the user behavior data to obtain the kth fraudulent time domain vector of the user behavior data;
acquiring a first vector distance between a kth-1 fraudulent airspace vector of the user behavior data and a kth-1 fraudulent airspace vector of each other user behavior data;
vector mapping is carried out on the kth directed relation vector of the user behavior data corresponding to each other user behavior data, and behavior influence coefficients of each other user behavior data are obtained;
according to the behavior influence coefficients of the other user behavior data, performing weight fusion calculation on the first vector distances of the plurality of other user behavior data to obtain fusion calculation data corresponding to the user behavior data;
fusing the fused calculation data of the user behavior data with the k-1 fraudulent airspace vector of the user behavior data to obtain the k fraudulent airspace vector of the user behavior data;
And taking the initial directed path relation of the user behavior data as a kth directed path relation of the user behavior data, and forming a kth fraud knowledge vector of the user behavior data by the kth directed path relation of the user behavior data, a kth fraud time domain vector of the user behavior data and a kth fraud space vector of the user behavior data.
8. The artificial intelligence based fraud analysis method according to claim 3, wherein the (k+1) -th fraud knowledge description unit extracts a kth fraud time domain vector of each of the user behavior data to obtain a kth+1-th fraud time domain vector of each of the user behavior data, including:
extracting a kth fraud airspace vector of the user behavior data, a kth fraud time domain vector of the user behavior data and a kth directed path relation of the user behavior data from kth fraud knowledge vectors of the user behavior data;
for each piece of user behavior data, acquiring other pieces of user behavior data except the user behavior data in the user behavior directed graph data, and executing the following steps for each piece of other pieces of user behavior data:
Extracting a kth directed path relation of the user behavior data aiming at the other user behavior data from the kth directed path relation of the user behavior data, and acquiring a second vector distance between a kth fraudulent airspace vector of the user behavior data and a kth fraudulent airspace vector of the other user behavior data;
performing a first vector fusion on the square of the second vector distance, a kth fraud time domain vector of the user behavior data, a kth fraud time domain vector of the other user behavior data, and a kth directed path relation of the user behavior data with respect to the other user behavior data to obtain a kth+1 directed relation vector of the user behavior data corresponding to the other user behavior data;
fusing the k+1th directional relation vector of the user behavior data corresponding to a plurality of other user behavior data to obtain the k+1th directional relation vector of the user behavior data;
and carrying out second vector fusion on the kth fraud time domain vector of the user behavior data and the kth+1 directional relation vector of the user behavior data to obtain the kth+1 fraud time domain vector of the user behavior data.
9. The artificial intelligence based fraud analysis method of claim 1, wherein prior to fraud tag bias value estimation on the targeted financial service user's user behavior directed graph data based on fraud analysis neural network, obtaining the targeted financial service user's fraud tag bias value, the method further comprises:
acquiring template user behavior directed graph data, and performing feature derivation on the template user behavior directed graph data to acquire a plurality of template derived behavior directed graph data;
acquiring fraud tag deviation labeling values of the template derived behavior directed graph data;
forward transmitting each template-derived behavior directed graph data in an initial fraud analysis neural network to obtain fraud label deviation prediction values of each template-derived behavior directed graph data;
determining global training cost values according to the fraud tag deviation labeling values of the template-derived behavior directed graph data and the fraud tag deviation predicting values of the template-derived behavior directed graph data;
performing back propagation processing on the global training cost value in the initial fraud analysis neural network to obtain network weight updating information of the initial fraud analysis neural network when the global training cost value converges, and updating the network weight information of the initial fraud analysis neural network according to the network weight updating information;
The obtaining the fraud tag deviation labeling value of each template derived behavior directed graph data comprises the following steps:
the following steps are executed for each template derived behavior directed graph data:
performing a first fraud tag value estimation strategy on the template-derived behavior directed graph data to obtain a first fraud tag value of the template-derived behavior directed graph data; performing a second fraud tag value estimation strategy on the template-derived behavior directed graph data to obtain a second fraud tag value of the template-derived behavior directed graph data;
acquiring a first loss value between a second fraud tag value of the template-derived behavior directed graph data and a first fraud tag value of the template-derived behavior directed graph data as a fraud tag deviation labeling value of the template-derived behavior directed graph data;
the determining the global training cost value according to the fraud tag deviation labeling value of each template derived behavior directed graph data and the fraud tag deviation predicting value of each template derived behavior directed graph data comprises the following steps:
the following steps are executed for each template derived behavior directed graph data:
determining a first root mean square loss value of the template-derived behavior directed graph data according to the fraud tag deviation labeling value of the template-derived behavior directed graph data and the fraud tag deviation predicting value of the template-derived behavior directed graph data;
Acquiring other template-derived behavior directed graph data except for the template-derived behavior directed graph data of the template user behavior directed graph data;
the following steps are performed for each of the other template-derived behavioral directed graph data: determining a second loss value between the fraud tag deviation labeling value of the template-derived directed graph data and the fraud tag deviation labeling value of the other template-derived directed graph data, and determining a third loss value between the fraud tag deviation prediction value of the template-derived directed graph data and the fraud tag deviation prediction value of the other template-derived directed graph data;
performing root mean square calculation on the second loss value and the third loss value to obtain a second root mean square loss value of the other template-derived behavior directed graph data corresponding to the template-derived behavior directed graph data;
fusing the second root mean square loss values of the template-derived behavior directed graph data corresponding to a plurality of other template-derived behavior directed graph data to obtain a third root mean square loss value of the template-derived behavior directed graph data;
and carrying out third vector fusion on the first root mean square loss values of the plurality of template-derived behavior directed graph data and the third root mean square loss values of the plurality of template-derived behavior directed graph data to obtain global training cost values corresponding to the fraud analysis neural network.
10. A digital financial big data system, characterized in that it comprises a processor and a memory for storing a computer program capable of running on the processor, said processor being adapted to execute the artificial intelligence based fraud analysis method according to any of claims 1-9 when said computer program is run.
CN202311141405.2A 2023-09-06 2023-09-06 Fraud analysis method based on artificial intelligence and digital financial big data system Pending CN117575596A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311141405.2A CN117575596A (en) 2023-09-06 2023-09-06 Fraud analysis method based on artificial intelligence and digital financial big data system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311141405.2A CN117575596A (en) 2023-09-06 2023-09-06 Fraud analysis method based on artificial intelligence and digital financial big data system

Publications (1)

Publication Number Publication Date
CN117575596A true CN117575596A (en) 2024-02-20

Family

ID=89888697

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311141405.2A Pending CN117575596A (en) 2023-09-06 2023-09-06 Fraud analysis method based on artificial intelligence and digital financial big data system

Country Status (1)

Country Link
CN (1) CN117575596A (en)

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109064318A (en) * 2018-08-24 2018-12-21 苏宁消费金融有限公司 A kind of internet financial risks monitoring system of knowledge based map
CN109816397A (en) * 2018-12-03 2019-05-28 北京奇艺世纪科技有限公司 A kind of fraud method of discrimination, device and storage medium
CN109871445A (en) * 2019-01-23 2019-06-11 平安科技（深圳）有限公司 Fraudulent user recognition methods, device, computer equipment and storage medium
US20190279218A1 (en) * 2014-08-08 2019-09-12 Brighterion, Inc. Behavior tracking smart agents for artificial intelligence fraud protection and management
US20200252802A1 (en) * 2019-02-06 2020-08-06 Pccw Vuclip (Singapore) Pte. Ltd. Apparatus and method for fraud detection
US20200250675A1 (en) * 2019-02-05 2020-08-06 International Business Machines Corporation Fraud Detection Based on Community Change Analysis Using a Machine Learning Model
WO2021213123A1 (en) * 2020-11-25 2021-10-28 平安科技（深圳）有限公司 User fraudulent behavior detection method, apparatus and device, and storage medium
US20210374756A1 (en) * 2020-05-29 2021-12-02 Mastercard International Incorporated Methods and systems for generating rules for unseen fraud and credit risks using artificial intelligence
CN116433345A (en) * 2023-05-05 2023-07-14 辽宁慧远科技开发有限公司 AI-based fraudulent activity analysis method and digital financial product service system
CN116542673A (en) * 2023-07-05 2023-08-04 成都乐超人科技有限公司 Fraud identification method and system applied to machine learning
CN116611916A (en) * 2023-06-25 2023-08-18 云南却蓉科技有限责任公司 Digital finance anti-fraud processing method and system based on AI model identification

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190279218A1 (en) * 2014-08-08 2019-09-12 Brighterion, Inc. Behavior tracking smart agents for artificial intelligence fraud protection and management
CN109064318A (en) * 2018-08-24 2018-12-21 苏宁消费金融有限公司 A kind of internet financial risks monitoring system of knowledge based map
CN109816397A (en) * 2018-12-03 2019-05-28 北京奇艺世纪科技有限公司 A kind of fraud method of discrimination, device and storage medium
CN109871445A (en) * 2019-01-23 2019-06-11 平安科技（深圳）有限公司 Fraudulent user recognition methods, device, computer equipment and storage medium
US20200250675A1 (en) * 2019-02-05 2020-08-06 International Business Machines Corporation Fraud Detection Based on Community Change Analysis Using a Machine Learning Model
US20200252802A1 (en) * 2019-02-06 2020-08-06 Pccw Vuclip (Singapore) Pte. Ltd. Apparatus and method for fraud detection
US20210374756A1 (en) * 2020-05-29 2021-12-02 Mastercard International Incorporated Methods and systems for generating rules for unseen fraud and credit risks using artificial intelligence
WO2021213123A1 (en) * 2020-11-25 2021-10-28 平安科技（深圳）有限公司 User fraudulent behavior detection method, apparatus and device, and storage medium
CN116433345A (en) * 2023-05-05 2023-07-14 辽宁慧远科技开发有限公司 AI-based fraudulent activity analysis method and digital financial product service system
CN116611916A (en) * 2023-06-25 2023-08-18 云南却蓉科技有限责任公司 Digital finance anti-fraud processing method and system based on AI model identification
CN116542673A (en) * 2023-07-05 2023-08-04 成都乐超人科技有限公司 Fraud identification method and system applied to machine learning

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
DAIXIN WANG ET AL: "A Semi-supervised Graph Attentive Network for Financial Fraud Detection", 《2019 IEEE INTERNATIONAL CONFERENCE ON DATA MINING》, 30 January 2020 (2020-01-30) *
EREN KURSHAN ET AL: "Financial Crime & Fraud Detection Using Graph Computing: Application Considerations & Outlook", 《2020 SECOND INTERNATIONAL CONFERENCE ON TRANSDISCIPLINARY AI》, 30 November 2020 (2020-11-30) *
TAHEREH POURHABIBI ET AL: "Fraud detection: A systematic literature review of graph-based anomaly detection approaches", 《DECISION SUPPORT SYSTEMS》, 17 April 2020 (2020-04-17) *
王威巍;: "人工智能在金融反欺诈领域的应用", 中国科技信息, no. 20, 15 October 2018 (2018-10-15) *
肖媛, 柳炳祥: "金融欺诈行为的粗集神经网络分析法", 生产力研究, no. 08, 30 August 2005 (2005-08-30) *

Similar Documents

Publication Publication Date Title
AU2020202542B2 (en) Transforming attributes for training automated modeling systems
US10917425B2 (en) Graph structure model training and junk account identification
EP3654610B1 (en) Graphical structure model-based method for prevention and control of abnormal accounts, and device
CN111738441B (en) Prediction model training method and device considering prediction precision and privacy protection
JP6782680B2 (en) Systems and methods for instructional analysis
Luna et al. Adaptive fuzzy system to forecast financial time series volatility
CN111428217B (en) Fraudulent party identification method, apparatus, electronic device and computer readable storage medium
CN113283671A (en) Method and device for predicting replenishment quantity, computer equipment and storage medium
US20230186331A1 (en) Generalized demand estimation for automated forecasting systems
CN113592593B (en) Training and application method, device, equipment and storage medium of sequence recommendation model
CN109754104B (en) Method, system, equipment and medium for optimizing enterprise supply chain by applying artificial intelligence
US20230342626A1 (en) Model processing method and related apparatus
CN114978765B (en) Big data processing method for information attack defense and AI attack defense system
CN117575596A (en) Fraud analysis method based on artificial intelligence and digital financial big data system
CN116433345A (en) AI-based fraudulent activity analysis method and digital financial product service system
CN114168966B (en) Big data analysis-based security protection upgrade mining method and information security system
CN111931870B (en) Model prediction method, model prediction device and system based on model multiplexing
CN110322339B (en) Virtual resource management method, device, computer system and readable storage medium
CN112668223A (en) Electromechanical equipment simulation method and system based on digital twin lightweight model
CN112232824A (en) Abstract syntax tree-based risk result calculation method and equipment
CN112905987B (en) Account identification method, device, server and storage medium
CN113946758B (en) Data identification method, device, equipment and readable storage medium
CN116151635B (en) Optimization method and device for decision-making of anti-risk enterprises based on multidimensional relation graph
US20230351491A1 (en) Accelerated model training for real-time prediction of future events
US20230351169A1 (en) Real-time prediction of future events using integrated input relevancy

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
TA01 Transfer of patent application right

Effective date of registration: 20240407

Address after: No. 59 Gongye South Road, Lixia District, Jinan City, Shandong Province, 250101

Applicant after: Li Jianjun

Country or region after: China

Address before: 277727 Wanding Technology, E-Huang Road Intelligent Manufacturing Industrial Park, High tech Zone, Lanshan District, Linyi City, Shandong Province

Applicant before: Linyi Wanding Network Technology Co.,Ltd.

Country or region before: China