CN116662574A

CN116662574A - Big data acquisition method and system for anti-fraud AI prediction model

Info

Publication number: CN116662574A
Application number: CN202310779192.XA
Authority: CN
Inventors: 沙波; 杨光; 王晓娟
Original assignee: Binzhou Weizuoban Network Technology Co ltd
Current assignee: Binzhou Weizuoban Network Technology Co ltd
Priority date: 2023-06-29
Filing date: 2023-06-29
Publication date: 2023-08-29

Abstract

The invention provides a big data acquisition method and a big data acquisition system for an anti-fraud AI prediction model, and relates to the technical field of artificial intelligence. In the invention, the key information characteristic representation corresponding to the member to be analyzed is mined, and the key information characteristic representation corresponding to each network interaction behavior is mined; analyzing related network interaction behaviors associated with members to be analyzed according to the key information characteristic representation; determining correlation information between related network interaction behaviors and other network interaction behaviors except the related network interaction behaviors, and determining distinguishing information between the related network interaction behaviors and other network interaction behaviors; and determining at least one network interaction behavior in the plurality of network interaction behaviors based on the related network interaction behaviors, the related information and the distinguishing information, and marking the network interaction behavior as a collected network interaction behavior. Based on the above, the reliability of data collection in fraud identification applications can be improved to some extent.

Description

Big data acquisition method and system for anti-fraud AI prediction model

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a big data acquisition method and a big data acquisition system for an anti-fraud AI prediction model.

Background

Fraud identification or fraud prediction based on artificial intelligence is an important security measure in the fields of finance, economy and the like, and particularly has higher convenience for network behaviors, but fraud accidents are easier to occur. Therefore, in the prior art, network interaction behavior is generally collected and then analyzed by using an AI prediction model (neural network), but in order to reduce the amount of data to be analyzed, network interaction behavior is generally screened first, so that fraudulent identification can be performed based on the screened (i.e., collected) network interaction behavior, but in the prior art, there is a problem that the reliability of data collection (screening) is not high.

Disclosure of Invention

In view of the above, the present invention aims to provide a big data collection method and system for an anti-fraud AI prediction model, so as to improve the reliability of data collection in fraud identification applications to a certain extent.

In order to achieve the above purpose, the embodiment of the present invention adopts the following technical scheme:

a big data collection method for an anti-fraud AI prediction model, the big data collection method comprising:

Excavating key information characteristic representations corresponding to members to be analyzed, and excavating key information characteristic representations corresponding to each network interaction behavior in a plurality of network interaction behaviors, wherein the members to be analyzed belong to network members, and the network interaction behaviors are network behaviors performed by the network members;

analyzing related network interaction behaviors associated with the members to be analyzed according to the key information feature representations corresponding to the members to be analyzed and the key information feature representations corresponding to the network interaction behaviors;

determining correlation information between the related network interaction behavior and other network interaction behaviors except the related network interaction behavior in the plurality of network interaction behaviors, and determining distinguishing information between the related network interaction behavior and the other network interaction behaviors;

and determining at least one network interaction behavior in the network interaction behaviors based on the related network interaction behavior, the correlation information between the related network interaction behavior and the other network interaction behaviors and the distinguishing information between the related network interaction behavior and the other network interaction behaviors, and marking the network interaction behavior as the acquired network interaction behavior of the member to be analyzed, wherein the acquired network interaction behavior is used as the basis for fraud identification of the member to be analyzed.

In some preferred embodiments, in the above big data collection method for an anti-fraud AI prediction model, the step of mining out a key information feature representation corresponding to a member to be analyzed, and mining out a key information feature representation corresponding to each of a plurality of network interaction behaviors includes:

taking a network member and a network interaction behavior as a knowledge graph object, taking an execution relationship between the network member and the network interaction behavior as a knowledge graph connecting line segment to form a member behavior relationship knowledge graph corresponding to a plurality of network members and the plurality of network interaction behaviors, wherein the plurality of network members comprise the members to be analyzed, in the member behavior relationship knowledge graph, the knowledge graph connecting line segment is arranged between two corresponding knowledge graph objects when any network member has executed any network interaction behavior, the knowledge graph connecting line segment is arranged between two corresponding knowledge graph objects when any network member has network interaction behavior, the knowledge graph connecting line segment is arranged between two corresponding knowledge graph objects when any two network interaction behaviors belong to a series of behaviors, and in the member behavior relationship knowledge graph, object attribute data of the knowledge graph object corresponding to the network member comprise member description text of the network member, and the object attribute data of the knowledge graph corresponding to the network interaction behavior comprise the network interaction attribute description text of the network member;

And carrying out network optimization operation on the member behavior analysis network based on the member behavior relation knowledge graph, and utilizing the member behavior analysis network formed by network optimization to mine out the key information characteristic representation corresponding to the member to be analyzed and mine out the key information characteristic representation corresponding to each network interaction behavior in the plurality of network interaction behaviors.

In some preferred embodiments, in the big data collection method for an anti-fraud AI prediction model, the member behavior analysis network includes a feature space mapping sub-network, a feature integration sub-network, and a feature analysis sub-network, the step of performing a network optimization operation on the member behavior analysis network based on the member behavior relationship knowledge graph, and using the member behavior analysis network formed by network optimization to discover a key information feature representation corresponding to the member to be analyzed, and discover a key information feature representation corresponding to each of the network interaction behaviors in the plurality of network interaction behaviors includes:

performing feature space mapping operation on each network member and each network interaction behavior by using the feature space mapping sub-network to form information mapping feature representations corresponding to each network member and form information mapping feature representations corresponding to each network interaction behavior;

Utilizing the characteristic integration sub-network, analyzing information integration characteristic representations corresponding to all knowledge graph objects in the member behavior relationship knowledge graph according to information mapping characteristic representations corresponding to all network members and information mapping characteristic representations corresponding to all network interaction behaviors, wherein the information integration characteristic representations are integration characteristic representations formed by performing characteristic integration operation on adjacent knowledge graph objects with first compactness to adjacent knowledge graph objects with second compactness based on the knowledge graph objects, the compactness between the knowledge graph objects is determined based on the number of the knowledge graph objects included in the shortest traversal path formed by traversing in the member behavior knowledge graph, and a knowledge graph connecting line segment is arranged between two adjacent knowledge graph objects in the traversal path during traversing;

according to the information integration feature representation corresponding to each knowledge graph object in the membership knowledge graph, analyzing the execution prediction probability parameters between the network members and the network interaction behaviors in the membership knowledge graph by utilizing the feature analysis sub-network;

Optimizing and adjusting the network parameters of the member behavior analysis network according to the analyzed execution prediction probability parameters to form the member behavior analysis network formed by network optimization;

and utilizing the member behavior analysis network formed by network optimization to mine out the key information characteristic representation corresponding to the member to be analyzed, and mine out the key information characteristic representation corresponding to each network interaction behavior in the network interaction behaviors.

In some preferred embodiments, in the above big data collection method for an anti-fraud AI prediction model, the step of using the feature integration sub-network to analyze information integration feature representations corresponding to each knowledge graph object in the membership behavior relationship knowledge graph according to the information mapping feature representations corresponding to each network member and the information mapping feature representations corresponding to each network interaction behavior includes:

utilizing the feature integration sub-network, and constructing a first feature representation distribution array corresponding to the member behavior relation knowledge graph according to the information mapping feature representation corresponding to the network members and the information mapping feature representation corresponding to each network interaction behavior;

Determining a map object relation distribution array corresponding to the member behavior relation knowledge map, and determining a target parameter distribution array, wherein parameters of target distribution positions in the target parameter distribution array are sum values of parameters of corresponding rows or corresponding columns in the map object relation distribution array;

analyzing an integrated feature representation distribution array corresponding to a target number of information integration operations in the membership knowledge graph according to the first feature representation distribution array, the graph object relationship distribution array and the target parameter distribution array, wherein the integrated feature representation distribution array comprises the information integration feature representations corresponding to all knowledge graph objects in the membership knowledge graph, the information integration operations in the target number of information integration operations are sequentially carried out, output data of the former information integration operation is used as input data of the latter information integration operation, input data of the first information integration operation comprises the first feature representation distribution array, and the graph object relationship distribution array and the target parameter distribution array are used as operation reference data of each information integration operation;

And the step of mining the key information feature representation corresponding to the member to be analyzed and mining the key information feature representation corresponding to each network interaction behavior in the plurality of network interaction behaviors by using the member behavior analysis network formed by network optimization comprises the following steps:

after the member behavior analysis network formed by network optimization is formed, extracting the information integration feature representation corresponding to the knowledge graph object corresponding to the member to be analyzed from the current integration feature representation distribution array, and taking the information integration feature representation as the key information feature representation corresponding to the member to be analyzed, and extracting the information integration feature representation corresponding to the knowledge graph object corresponding to each network interaction behavior in the network interaction behaviors respectively, and taking the information integration feature representation as the key information feature representation corresponding to each network interaction behavior in the network interaction behaviors respectively.

In some preferred embodiments, in the above big data collection method for an anti-fraud AI prediction model, the step of optimizing and adjusting the network parameters of the member behavior analysis network according to the analyzed execution prediction probability parameters to form the member behavior analysis network formed by network optimization includes:

For each network member, determining a first execution prediction probability parameter between the network member and a first network interaction behavior analyzed by the feature analysis sub-network, and analyzing a second execution prediction probability parameter between the network member and a second network interaction behavior, wherein the first network interaction behavior has an execution relationship with the network member, and the second network interaction behavior does not have an execution relationship with the network member;

according to the first execution prediction probability parameter and the second execution prediction probability parameter, calculating a network optimization cost index of the member behavior analysis network, and according to the network optimization cost index, carrying out optimization adjustment on the network parameters of the member behavior analysis network to form the member behavior analysis network formed by network optimization.

In some preferred embodiments, in the above big data collection method for an anti-fraud AI prediction model, the step of analyzing the related network interaction behavior associated with the member to be analyzed according to the key information feature representation corresponding to the member to be analyzed and the key information feature representation corresponding to each network interaction behavior includes:

Calculating the feature representation quantity product respectively between the key information feature representations corresponding to the members to be analyzed and the key information feature representations corresponding to the network interaction behaviors;

and determining the relation among the calculated characteristic representation quantity products, and marking the network interaction behavior corresponding to the characteristic representation quantity product with the maximum value to be the related network interaction behavior associated with the member to be analyzed.

In some preferred embodiments, in the big data collection method for an anti-fraud AI prediction model, the step of determining correlation information between the related network interaction behavior and other network interaction behaviors other than the related network interaction behavior of the plurality of network interaction behaviors, and determining distinction information between the related network interaction behavior and the other network interaction behaviors includes:

for each other network interaction behavior other than the related network interaction behavior in the plurality of network interaction behaviors, performing the following operations:

calculating a feature representation quantity product between key information feature representations corresponding to the related network interaction behaviors and key information feature representations corresponding to other network interaction behaviors;

Determining correlation information between the related network interaction behavior and the other network interaction behaviors based on the feature representation quantity product, wherein the correlation information and the feature representation quantity product have a positive correlation corresponding relation;

and determining distinguishing information between the related network interaction behavior and the other network interaction behaviors.

In some preferred embodiments, in the above big data collection method for an anti-fraud AI prediction model, the step of determining distinguishing information between the related network interaction behavior and the other network interaction behavior includes:

for each other network interaction behavior other than the related network interaction behavior in the plurality of network interaction behaviors, performing the following second operation:

calculating a deviation degree characterization parameter between the key information characteristic representation corresponding to the related network interaction behavior and the key information characteristic representation corresponding to the other network interaction behaviors;

and determining distinguishing information between the related network interaction behaviors and the other network interaction behaviors based on the deviation characterization parameters, wherein the distinguishing information and the deviation characterization parameters have a positive correlation corresponding relation.

In some preferred embodiments, in the above big data collection method for an anti-fraud AI prediction model, the step of determining at least one network interaction behavior among the plurality of network interaction behaviors and marking as the collection network interaction behavior of the member to be analyzed based on the related network interaction behavior, correlation information between the related network interaction behavior and the other network interaction behavior, and distinguishing information between the related network interaction behavior and the other network interaction behavior includes:

determining a corresponding relation to be processed according to the related network interaction behavior related to the member to be analyzed, the related information between the related network interaction behavior and other network interaction behaviors and the distinguishing information between the related network interaction behavior and other network interaction behaviors;

outputting characteristic representation relation characterization parameters of the to-be-processed corresponding relation corresponding to each other network interaction behavior according to the key information characteristic representation corresponding to the member to be analyzed and the key information characteristic representation corresponding to each network interaction behavior;

and according to the characteristic representation relation characterization parameters of the to-be-processed corresponding relation corresponding to each other network interaction behavior, determining at least one network interaction behavior in the plurality of network interaction behaviors, and marking the network interaction behavior as the acquired network interaction behavior of the member to be analyzed.

The embodiment of the invention also provides a big data acquisition system for the anti-fraud AI prediction model, which comprises a processor and a memory, wherein the memory is used for storing a computer program, and the processor is used for executing the computer program to realize the big data acquisition method for the anti-fraud AI prediction model.

The big data acquisition method and the big data acquisition system for the anti-fraud AI prediction model provided by the embodiment of the invention can be used for firstly mining the key information characteristic representation corresponding to the member to be analyzed and mining the key information characteristic representation corresponding to each network interaction behavior; analyzing related network interaction behaviors associated with members to be analyzed according to the key information characteristic representation; determining correlation information between related network interaction behaviors and other network interaction behaviors except the related network interaction behaviors, and determining distinguishing information between the related network interaction behaviors and other network interaction behaviors; and determining at least one network interaction behavior in the plurality of network interaction behaviors based on the related network interaction behaviors, the related information and the distinguishing information, and marking the network interaction behavior as a collected network interaction behavior. Based on the foregoing, the related network interaction behavior associated with the member to be analyzed is determined first, so that the determination of the collected network interaction behavior can be performed based on the related network interaction behavior, instead of arbitrarily selecting part of the network interaction behavior as the collected network interaction behavior, so that the method has higher reliability, and in addition, in the process of determining the collected network interaction behavior, the correlation information between the related network interaction behavior and other network interaction behaviors and the distinguishing information between the related network interaction behavior and other network interaction behaviors are referred to, so that the basis is more sufficient, thereby improving the data collection reliability in fraud identification application to a certain extent and improving the defects in the prior art.

In order to make the above objects, features and advantages of the present invention more comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

FIG. 1 is a block diagram of a big data acquisition system for an anti-fraud AI prediction model provided by an embodiment of the invention.

Fig. 2 is a schematic flow chart of steps included in a big data collection method for an anti-fraud AI prediction model according to an embodiment of the present invention.

Fig. 3 is a schematic diagram of each module included in the big data acquisition device for an anti-fraud AI prediction model according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

As shown in FIG. 1, an embodiment of the present invention provides a big data acquisition system for an anti-fraud AI prediction model. Wherein the big data acquisition system for the anti-fraud AI prediction model may comprise a memory and a processor.

In detail, the memory and the processor are electrically connected directly or indirectly to realize transmission or interaction of data. For example, electrical connection may be made to each other via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that may exist in the form of software or firmware. The processor may be configured to execute the executable computer program stored in the memory, thereby implementing the big data acquisition method for the anti-fraud AI prediction model provided by the embodiment of the invention.

It should be appreciated that in some possible embodiments, the Memory may be, but is not limited to, random access Memory (Random Access Memory, RAM), read Only Memory (ROM), programmable Read Only Memory (Programmable Read-Only Memory, PROM), erasable Read Only Memory (Erasable Programmable Read-Only Memory, EPROM), electrically erasable Read Only Memory (Electric Erasable Programmable Read-Only Memory, EEPROM), and the like.

It should be appreciated that in some possible embodiments, the processor may be a general purpose processor, including a central processing unit (Central Processing Unit, CPU), a network processor (Network Processor, NP), a System on Chip (SoC), etc.; but also Digital Signal Processors (DSPs), application Specific Integrated Circuits (ASICs), field Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

It should be appreciated that in some possible embodiments, the big data acquisition system for the anti-fraud AI prediction model may be a server with data processing capabilities.

With reference to fig. 2, the embodiment of the invention further provides a big data collection method for the anti-fraud AI prediction model, which can be applied to the big data collection system for the anti-fraud AI prediction model. The method steps defined by the flow related to the big data acquisition method for the anti-fraud AI prediction model can be realized by the big data acquisition system for the anti-fraud AI prediction model.

The specific flow shown in fig. 2 will be described in detail.

Step S110, excavating key information characteristic representations corresponding to members to be analyzed, and excavating key information characteristic representations corresponding to each network interaction behavior in a plurality of network interaction behaviors.

In the embodiment of the invention, the big data acquisition system for the anti-fraud AI prediction model can mine out the key information characteristic representation corresponding to the member to be analyzed and mine out the key information characteristic representation corresponding to each network interaction behavior in a plurality of network interaction behaviors. The member to be analyzed belongs to a network member, namely a network user, and the network interaction behavior refers to network behavior performed by the network member, such as voice call, instant text communication, offline text communication, release and browse of network content, mutual comment of the network content and the like. In addition, the key information feature representation may be a vector.

Step S120, according to the key information feature representation corresponding to the member to be analyzed and the key information feature representation corresponding to each network interaction behavior, analyzing the related network interaction behavior associated with the member to be analyzed.

In the embodiment of the invention, the big data acquisition system for the anti-fraud AI prediction model can analyze the related network interaction behaviors associated with the members to be analyzed according to the key information feature representations corresponding to the members to be analyzed and the key information feature representations corresponding to the network interaction behaviors, namely, determine the related network interaction behaviors according to the correlation among the key information feature representations.

Step S130, determining correlation information between the related network interaction behavior and other network interaction behaviors other than the related network interaction behavior in the plurality of network interaction behaviors, and determining distinguishing information between the related network interaction behavior and the other network interaction behaviors.

In the embodiment of the invention, the big data acquisition system for the anti-fraud AI prediction model may determine correlation information between the related network interaction behavior and other network interaction behaviors except the related network interaction behavior in the plurality of network interaction behaviors, and determine distinguishing information between the related network interaction behavior and the other network interaction behaviors. For example, the plurality of network interaction behaviors include a network interaction behavior 1, a network interaction behavior 2, a network interaction behavior 3, a network interaction behavior 4, and a network interaction behavior 5, wherein the network interaction behavior 2 serves as the relevant network interaction behavior, and thus, the other network interaction behaviors may include the network interaction behavior 1, the network interaction behavior 3, the network interaction behavior 4, and the network interaction behavior 5.

Step S140, determining at least one network interaction behavior among the plurality of network interaction behaviors based on the related network interaction behaviors, the correlation information between the related network interaction behaviors and the other network interaction behaviors, and the distinguishing information between the related network interaction behaviors and the other network interaction behaviors, and marking the network interaction behaviors as acquired network interaction behaviors of the member to be analyzed.

In the embodiment of the present invention, the big data collection system for the anti-fraud AI prediction model may determine at least one network interaction behavior among the plurality of network interaction behaviors based on the related network interaction behavior, correlation information between the related network interaction behavior and the other network interaction behaviors, and distinguishing information between the related network interaction behavior and the other network interaction behaviors, and mark the collected network interaction behavior as the member to be analyzed. The collecting network interaction behavior is used as a basis for carrying out fraud recognition on the member to be analyzed, for example, the collecting network interaction behavior is recognized through a corresponding fraud recognition neural network, so as to obtain a result of whether the collecting network interaction behavior belongs to fraud or a possibility of fraud of the member to be analyzed. In addition, when the network interaction behavior is determined, the network interaction behavior can be determined by collecting the network interaction behaviors except the network interaction behaviors performed by the related network interaction behaviors and the member to be analyzed, so that the related network interaction behaviors, the network interaction behaviors performed by the member to be analyzed and the network interaction behaviors can be taken as analysis basis together to obtain a fraud identification result when fraud identification is performed.

Based on the foregoing (i.e. the steps S110-S140 described above), the relevant network interaction behavior associated with the member to be analyzed is determined first, so that the determination of the collected network interaction behavior can be performed based on the relevant network interaction behavior, instead of arbitrarily selecting a part of the network interaction behaviors as the collected network interaction behavior, so that the method has higher reliability, and in addition, in the process of determining the collected network interaction behavior, the correlation information between the relevant network interaction behavior and other network interaction behaviors and the distinguishing information between the relevant network interaction behavior and other network interaction behaviors are referred to, so that the basis is more sufficient, therefore, the data collection reliability in the fraud identification application can be improved to a certain extent, and the defects in the prior art are improved.

It should be understood that, in some possible embodiments, step S110 in the foregoing description may further include the following specific implementation matters:

taking network members and network interaction behaviors as knowledge graph objects, taking the execution relationship between the network members and the network interaction behaviors as a knowledge graph connecting line segment to form a member behavior relationship knowledge graph corresponding to a plurality of network members and the network interaction behaviors, wherein the plurality of network members comprise the members to be analyzed, in the member behavior relationship knowledge graph, under the condition that any one network member has executed any one network interaction behavior, the corresponding two knowledge graph objects (namely, the knowledge graph objects corresponding to the network members and the knowledge graph objects corresponding to the network interaction behaviors) are provided with the knowledge graph connecting line segment, under the condition that any two network members have network interaction behaviors (for example, a video call operation is performed between a network member 1 and a network member 2), the knowledge graph connecting line segment is formed between the two knowledge objects corresponding to the two network members, under the condition that any two network interaction behaviors belong to a series of behaviors (the series of behaviors can be regarded as co-occurrence and the co-occurrence of the two network interaction behaviors, such as the preset knowledge graph objects can be regarded as the high-frequency of the actual knowledge graph, the fact that the two network interaction behaviors can be frequently correspond to the actual knowledge graph 2 in the network interaction graph, the high-frequency can be regarded as the fact that the actual knowledge graph is frequently-occurring between the two network interaction behaviors in the network interaction graph 1, the actual knowledge graph can be configured according to the preset knowledge graph 1, the fact that the actual knowledge graph is frequently-occurring between the two network interaction behavior has the actual knowledge graph is high-occurrence of the network interaction behavior 2, the object attribute data of the knowledge graph object corresponding to the network member comprises a member description text (such as identity information, age information and other behavior information on a network of the member, namely other network behaviors not belonging to interaction behaviors) of the network member, and the object attribute data of the knowledge graph object corresponding to the network interaction behavior comprises a behavior description text of the network interaction behavior;

It should be understood that, in some possible embodiments, the member behavior analysis network may include a feature space mapping sub-network, a feature integration sub-network, and a feature analysis sub-network, based on which, based on the member behavior relationship knowledge graph, the network optimization operation is performed on the member behavior analysis network, and the step of using the member behavior analysis network formed by network optimization to mine out the key information feature representation corresponding to the member to be analyzed and mine out the key information feature representation corresponding to each of the network interactions in the plurality of network interactions may further include the following specific implementation details:

performing feature space mapping operation on each network member and each network interaction behavior by using the feature space mapping sub-network to form information mapping feature representations corresponding to each network member and form information mapping feature representations corresponding to each network interaction behavior, that is, mapping object attribute data of knowledge map objects corresponding to the network member and the network interaction behavior into a feature space to respectively form corresponding information mapping feature representations;

Utilizing the feature integration sub-network, according to the information mapping feature representation corresponding to each network member and the information mapping feature representation corresponding to each network interaction behavior, analyzing the information integration feature representation corresponding to each knowledge graph object in the member behavior relation knowledge graph, wherein the information integration feature representation is an integration feature representation formed by performing feature integration operation on the basis of the adjacent knowledge graph object with a first compactness to the adjacent knowledge graph object with a second compactness of the knowledge graph object, namely integrating the information mapping feature representations corresponding to the adjacent knowledge graph objects with the compactness smaller than or equal to the first compactness and the compactness larger than or equal to the second compactness, in addition, the specific numerical values of the first compactness and the second compactness are not limited, the configuration can be performed according to actual requirements, the compactness between the knowledge graph objects is determined based on the number of the knowledge graph objects included in the shortest traversal path formed by traversing in the member behavior relation knowledge graph object, for example, the compactness has a negative correlation relationship with the number, and when traversing, the two adjacent knowledge graph objects have a connection knowledge graph in the traversal path;

It should be understood that, in some possible embodiments, the step of analyzing, by using the feature analysis sub-network, the execution prediction probability parameter between the network member and the network interaction behavior in the membership knowledge graph according to the information integration feature representation corresponding to each knowledge graph object in the membership knowledge graph may further include the following specific implementation contents described below:

For a network member and a network interaction in the membership knowledge graph, the information integration feature representation corresponding to the network member may be transposed to form a corresponding transposed information integration feature representation, and then a number product between the transposed information integration feature representation and the information integration feature representation corresponding to the network interaction may be calculated to obtain an execution prediction probability parameter between the network member and the network interaction, where the execution prediction probability parameter may represent a probability that the network interaction will be executed by the network member, or a probability that the network interaction has already been executed (in some cases, the executed network interaction may not be counted).

It should be appreciated that, in some possible embodiments, the step of analyzing, by using the feature integration sub-network, the information integration feature representation corresponding to each knowledge graph object in the membership knowledge graph according to the information mapping feature representation corresponding to each network member and the information mapping feature representation corresponding to each network interaction behavior may further include the following specific implementation matters described below:

Constructing a first feature representation distribution array corresponding to the member behavior relation knowledge graph according to the information mapping feature representation corresponding to the network member and the information mapping feature representation corresponding to each network interaction behavior by utilizing the feature integration sub-network, namely, the first feature representation distribution array comprises the information mapping feature representation corresponding to the network member and the information mapping feature representation corresponding to each network interaction behavior, and can be understood as a feature representation with larger size;

determining a graph object relation distribution array corresponding to the membership knowledge graph, that is, discarding object attribute data of the knowledge graph objects in the membership knowledge graph, and only preserving relations among the knowledge graph objects, for example, the graph object relation distribution array may be a matrix, parameters of a matrix position may represent whether a knowledge graph connecting line segment (for example, represented by two different parameter values) exists between two corresponding graph objects, and determining a target parameter distribution array, where parameters of a target distribution position in the target parameter distribution array are sum values of parameters of corresponding rows or corresponding columns in the graph object relation distribution array, for example, the target parameter distribution array may have the following features: the target parameter distribution array is a matrix, the values of the parameters on the non-diagonal lines of the target parameter distribution array are the same, for example, 0, the parameters on the diagonal lines of the target parameter distribution array are the sum of the parameters of the corresponding rows or corresponding columns in the map object relation distribution array, for example, the parameters of the nth row or nth column in the target parameter distribution array are the sum of the parameters of the nth row or nth column in the map object relation distribution array, and specifically, the sum of the parameters of the corresponding rows can be;

According to the first characteristic representation distribution array, the map object relation distribution array and the target parameter distribution array, analyzing an integrated characteristic representation distribution array corresponding to a target number of information integration operations in the membership knowledge map, wherein the integrated characteristic representation distribution array comprises information integration characteristic representations corresponding to all knowledge map objects in the membership knowledge map, the information integration operations in the target number of information integration operations are sequentially carried out, output data of the former information integration operation is used as input data of the latter information integration operation, input data of the first information integration operation comprises the first characteristic representation distribution array, and thus, the operation can be sequentially and circularly carried out, and the map object relation distribution array and the target parameter distribution array are used as operation reference data of each information integration operation.

It should be understood that, in some possible embodiments, the step of analyzing the integrated feature representation distribution array corresponding to the target number of information integration operations in the membership knowledge graph according to the first feature representation distribution array, the graph object relationship distribution array and the target parameter distribution array may further include the following specific implementation matters described below:

Calculating the target parameter distribution array to the power of 0.5 to obtain a first calculation result, and then calculating (the first calculation result is the map object relation distribution array is the first calculation result) to obtain a second calculation result;

for a first information integration operation in the target number of information integration operations, taking the first characteristic representation distribution array as output data corresponding to the information integration operation, wherein the specific numerical value of the target number is not limited and can be configured according to actual requirements;

for each information integration operation except the first information integration operation in the target number of information integration operations, obtaining output data of the previous information integration operation, and performing multiplication operation on the second calculation result and the output data to obtain output data of the current information integration operation;

and carrying out weighted superposition operation on the output data corresponding to each information integration operation to output a corresponding integration characteristic representation distribution array, wherein in the process of carrying out the weighted superposition operation, the weighting coefficient corresponding to each output data can be configured according to actual requirements, such as sequential increment or decrement.

It should be understood that, in some possible embodiments, the steps of using the member behavior analysis network formed by network optimization to mine the key information feature representation corresponding to the member to be analyzed and mine the key information feature representation corresponding to each of the network interaction behaviors in the plurality of network interaction behaviors may further include the following specific implementation matters described below:

after the member behavior analysis network formed by network optimization is formed, extracting the information integration feature representation corresponding to the knowledge graph object corresponding to the member to be analyzed from the current integration feature representation distribution array, and taking the information integration feature representation as a key information feature representation corresponding to the member to be analyzed, and extracting the information integration feature representation corresponding to the knowledge graph object corresponding to each of the network interaction behaviors respectively, and taking the information integration feature representation as a key information feature representation corresponding to each of the network interaction behaviors respectively, that is, taking the information integration feature representation as a key information feature representation.

It should be understood that, in some possible embodiments, the step of optimally adjusting the network parameters of the member behavior analysis network according to the analyzed execution prediction probability parameters to form the member behavior analysis network formed by network optimization may further include the following specific implementation matters:

For each network member, determining a first execution prediction probability parameter (the calculation mode of the execution prediction probability parameter can be referred to in the foregoing and is not described in detail herein), and analyzing a second execution prediction probability parameter between the network member and a second network interaction behavior, wherein the first network interaction behavior and the network member have an execution relationship, and the second network interaction behavior and the network member do not have an execution relationship;

according to the first execution prediction probability parameter and the second execution prediction probability parameter, calculating a network optimization cost index of the member behavior analysis network, and according to the network optimization cost index, performing optimization adjustment on network parameters of the member behavior analysis network to form the member behavior analysis network formed by network optimization, for example, performing optimization adjustment on the network parameters along the direction of reducing the network optimization cost index.

It should be understood that, in some possible embodiments, the step of calculating the network optimization cost indicator of the member behavior analysis network according to the first execution prediction probability parameter and the second execution prediction probability parameter may further include the following specific implementation matters:

Calculating a difference value between the first execution prediction probability parameter and the second execution prediction probability parameter (corresponding network members are the same) for each group of the first execution prediction probability parameter and the second execution prediction probability parameter, normalizing the difference value, performing logarithmic operation on the normalized value to form a corresponding logarithmic result, and performing summation operation on logarithmic results corresponding to each group of the first execution prediction probability parameter and the second execution prediction probability parameter to output a corresponding total logarithmic result;

and calculating the square sum of each parameter in the first characteristic representation distribution array, and then calculating a network optimization cost index of the member behavior analysis network based on the calculated square sum and the total logarithm taking result, wherein the network optimization cost index can have a positive correlation corresponding relation with the calculated square sum, and the network optimization cost index can have a negative correlation corresponding relation with the total logarithm taking result.

It should be understood that, in some possible embodiments, step S120 in the foregoing description may further include the following specific implementation matters:

and determining the relation among the calculated characteristic representation quantity products, and marking the network interaction behavior corresponding to the characteristic representation quantity product with the maximum value to be the related network interaction behavior associated with the member to be analyzed, namely the most related network interaction behavior.

It should be understood that, in some possible embodiments, step S130 in the foregoing description may further include the following specific implementation matters:

determining correlation information between the related network interaction behavior and the other network interaction behavior based on the feature representation quantity product, wherein the correlation information and the feature representation quantity product have a positive correlation corresponding relation, for example, the feature representation quantity product can be directly used as the correlation information;

And determining distinguishing information, namely different information, between the related network interaction behavior and the other network interaction behaviors.

It should be appreciated that, in some possible embodiments, the step of determining the distinguishing information between the related network interaction behavior and the other network interaction behavior may further include the following specific implementation matters:

calculating a deviation degree characterization parameter, such as a cosine distance, between key information feature representations corresponding to the related network interaction behaviors and key information feature representations corresponding to the other network interaction behaviors, or calculating a difference feature representation between the corresponding key information feature representations, then calculating a square sum of parameters in the difference feature representation, and then calculating the 0.5 th power of the square sum to obtain the deviation degree characterization parameter;

and determining distinguishing information between the related network interaction behaviors and the other network interaction behaviors based on the deviation characterization parameters, wherein the distinguishing information and the deviation characterization parameters have a positive correlation corresponding relationship, for example, the deviation characterization parameters can be directly used as the corresponding distinguishing information.

It should be understood that, in some possible embodiments, step S140 in the foregoing description may further include the following specific implementation matters:

determining a corresponding relation to be processed according to the related network interaction behavior related to the member to be analyzed, the related information between the related network interaction behavior and other network interaction behaviors and the distinguishing information between the related network interaction behavior and other network interaction behaviors, namely, determining a corresponding calculation rule;

outputting characteristic representation relation characterization parameters of the to-be-processed corresponding relation corresponding to each other network interaction behavior according to the key information characteristic representation corresponding to the member to be analyzed and the key information characteristic representation corresponding to each network interaction behavior, namely calculating according to the calculation rule based on the key information characteristic representation to obtain the characteristic representation relation characterization parameters;

It should be understood that, in some possible embodiments, the step of determining the to-be-processed correspondence according to the related network interaction behavior associated with the member to be analyzed, the correlation information between the related network interaction behavior and other network interaction behaviors, and the distinguishing information between the related network interaction behavior and other network interaction behaviors may further include the following specific implementation contents described below:

determining a local first corresponding relation corresponding to the related network interaction behavior, a local second corresponding relation corresponding to the correlation information between the related network interaction behavior and other network interaction behaviors, and a local third corresponding relation corresponding to the distinguishing information between the related network interaction behavior and other network interaction behaviors;

analyzing the fixed parameters of the local first corresponding relation, the fixed parameters of the local second corresponding relation and the fixed parameters of the local third corresponding relation, wherein the fixed parameters of the local first corresponding relation, the fixed parameters of the local second corresponding relation and the fixed parameters of the local third corresponding relation can be used as network parameters of a neural network and formed through network optimization operation;

And carrying out fusion operation, such as addition operation, on the product of the local first corresponding relation and the corresponding fixed parameter, the product of the local second corresponding relation and the corresponding fixed parameter and the product of the local third corresponding relation and the corresponding fixed parameter to form the corresponding relation to be processed.

It should be understood that, in some possible embodiments, on the basis of the above-mentioned to-be-processed correspondence, the step of outputting the feature representation relationship characterization parameter of the to-be-processed correspondence corresponding to each of the other network interaction behaviors according to the key information feature representation corresponding to the to-be-analyzed member and the key information feature representation corresponding to each of the network interaction behaviors may further include the following specific implementation contents:

based on the local first corresponding relation, calculating a quantity product of a transposed feature representation of the key information feature representation corresponding to the member to be analyzed and the key information feature representation of the related network interaction behavior to obtain a first numerical value;

calculating the key information characteristic representation corresponding to the related network interaction behavior and the key information characteristic representation corresponding to each other network interaction behavior based on the local second corresponding relation to obtain a second value corresponding to each other network interaction behavior, such as the related information;

Calculating key information characteristic representations corresponding to the related network interaction behaviors and key information characteristic representations corresponding to other network interaction behaviors based on the local third corresponding relationship, and obtaining third numerical values corresponding to other network interaction parameters, such as the distinctive information;

and carrying out weighted summation calculation on the first numerical value, the second numerical value and the third numerical value based on the fixed parameter of the local first corresponding relation, the fixed parameter of the local second corresponding relation and the fixed parameter of the local third corresponding relation so as to obtain the characteristic representation relation characterization parameter of the to-be-processed corresponding relation corresponding to each other network interaction behavior.

It should be understood that, in some possible embodiments, the step of determining at least one network interaction behavior from the plurality of network interaction behaviors and marking the network interaction behavior as the member to be analyzed according to the feature representing the relationship characterizing parameter of the to-be-processed correspondence corresponding to each of the other network interaction behaviors may further include the following specific implementation contents described below:

determining the other network interaction behaviors corresponding to the characteristic representation relation characterization parameters of the to-be-processed corresponding relation with the maximum value from the network interaction behaviors, and marking the other network interaction behaviors as one acquisition network interaction behavior of the to-be-analyzed member;

Screening out the other network interaction behaviors marked as the collecting network interaction behaviors from the plurality of network interaction behaviors, and executing the other network interaction behaviors corresponding to the characteristic representation relation characterization parameters of the to-be-processed corresponding relation with the maximum value in the plurality of network interaction behaviors (namely in the remaining network interaction behaviors) in a revolving mode based on the remaining network interaction behaviors, wherein the step of marking the other network interaction behaviors as one collecting network interaction behavior of the to-be-analyzed member is performed to obtain a preset number of collecting network interaction behaviors of the to-be-analyzed member, and the preset number can be configured according to actual requirements and is not limited in detail.

With reference to fig. 3, the embodiment of the invention further provides a big data acquisition device for the anti-fraud AI prediction model, which can be applied to the big data acquisition system for the anti-fraud AI prediction model. Wherein, the big data acquisition device for the anti-fraud AI prediction model may include:

the system comprises a key information mining module, a network interaction module and a network interaction module, wherein the key information mining module is used for mining key information characteristic representations corresponding to members to be analyzed, and mining key information characteristic representations corresponding to each network interaction behavior in a plurality of network interaction behaviors, the members to be analyzed belong to network members, and the network interaction behaviors are network behaviors performed by the network members;

The related behavior determining module is used for analyzing related network interaction behaviors associated with the member to be analyzed according to the key information characteristic representations corresponding to the member to be analyzed and the key information characteristic representations corresponding to the network interaction behaviors;

the interaction behavior relation analysis module is used for determining correlation information between the related network interaction behavior and other network interaction behaviors except the related network interaction behavior in the plurality of network interaction behaviors and determining distinguishing information between the related network interaction behavior and the other network interaction behaviors;

the interaction behavior determining module is used for determining at least one network interaction behavior from the network interaction behaviors based on the related network interaction behaviors, the correlation information between the related network interaction behaviors and the other network interaction behaviors and the distinguishing information between the related network interaction behaviors and the other network interaction behaviors, and marking the network interaction behaviors as acquired network interaction behaviors of the members to be analyzed, wherein the acquired network interaction behaviors are used as the basis for fraud identification of the members to be analyzed.

In summary, the big data acquisition method and system for the anti-fraud AI prediction model provided by the invention can be used for firstly mining the key information characteristic representation corresponding to the member to be analyzed and mining the key information characteristic representation corresponding to each network interaction behavior; analyzing related network interaction behaviors associated with members to be analyzed according to the key information characteristic representation; determining correlation information between related network interaction behaviors and other network interaction behaviors except the related network interaction behaviors, and determining distinguishing information between the related network interaction behaviors and other network interaction behaviors; and determining at least one network interaction behavior in the plurality of network interaction behaviors based on the related network interaction behaviors, the related information and the distinguishing information, and marking the network interaction behavior as a collected network interaction behavior. Based on the foregoing, the related network interaction behavior associated with the member to be analyzed is determined first, so that the determination of the collected network interaction behavior can be performed based on the related network interaction behavior, instead of arbitrarily selecting part of the network interaction behavior as the collected network interaction behavior, so that the method has higher reliability, and in addition, in the process of determining the collected network interaction behavior, the correlation information between the related network interaction behavior and other network interaction behaviors and the distinguishing information between the related network interaction behavior and other network interaction behaviors are referred to, so that the basis is more sufficient, thereby improving the data collection reliability in fraud identification application to a certain extent and improving the defects in the prior art.

The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A big data acquisition method for an anti-fraud AI prediction model, the big data acquisition method comprising:

2. The big data collection method for an anti-fraud AI prediction model according to claim 1, wherein the step of mining out a key information feature representation corresponding to a member to be analyzed and mining out a key information feature representation corresponding to each of a plurality of network interaction behaviors includes:

3. The big data collection method for an anti-fraud AI prediction model according to claim 2, wherein the member behavior analysis network includes a feature space mapping sub-network, a feature integration sub-network, and a feature analysis sub-network, the step of performing a network optimization operation on the member behavior analysis network based on the member behavior relationship knowledge graph, and using the member behavior analysis network formed by network optimization to discover key information feature representations corresponding to the members to be analyzed, and discover key information feature representations corresponding to each of the network interaction behaviors in the plurality of network interaction behaviors includes:

4. The big data collection method for an anti-fraud AI prediction model of claim 3, wherein the step of analyzing, by using the feature integration sub-network, information integration feature representations corresponding to each knowledge graph object in the membership knowledge graph according to the information mapping feature representations corresponding to each network member and the information mapping feature representations corresponding to each network interaction behavior includes:

5. The big data collection method for an anti-fraud AI prediction model of claim 3, wherein the step of optimizing and adjusting network parameters of the member behavior analysis network according to the analyzed execution prediction probability parameters to form the member behavior analysis network formed by network optimization includes:

6. The big data collection method for an anti-fraud AI prediction model according to claim 1, wherein the step of analyzing the related network interaction behavior associated with the member to be analyzed according to the key information feature representation corresponding to the member to be analyzed and the key information feature representation corresponding to each network interaction behavior includes:

7. The big data collection method for an anti-fraud AI prediction model of claim 1, wherein the step of determining correlation information between the relevant network interaction behavior and other network interaction behaviors than the relevant network interaction behavior of the plurality of network interaction behaviors, and determining distinguishing information between the relevant network interaction behavior and the other network interaction behaviors, comprises:

8. The big data collection method for an anti-fraud AI prediction model of claim 7, wherein the step of determining distinguishing information between the relevant network interaction behavior and the other network interaction behavior includes:

9. The big data collection method for an anti-fraud AI prediction model of any of claims 1-8, wherein the step of determining, among the plurality of network interactions, a network interaction behavior of at least one and marking as a collection network interaction behavior of the member to be analyzed based on the relevant network interaction behavior, correlation information between the relevant network interaction behavior and the other network interaction behavior, and distinguishing information between the relevant network interaction behavior and the other network interaction behavior, comprises:

10. A big data acquisition system for an anti-fraud AI prediction model, characterized by comprising a processor and a memory, the memory being for storing a computer program, the processor being for executing the computer program to implement the method of any of claims 1-9.