An artificial intelligence-based multi-factor fraud link tracing method and system

Abstract

This invention discloses a multi-factor fraud link tracing method and system based on artificial intelligence, relating to the field of data security tracking. The method includes: constructing a multimodal behavioral perturbation structure model; extracting perturbation responses from multi-dimensional behavioral source signals and calculating perturbation absorption intensity; constructing an energy flow coupling relationship graph between entities based on the perturbation absorption intensity; performing structure propagation and embedding representation updates according to the energy flow coupling relationship graph, iteratively generating entity structure embedding representations using an edge-weight-driven nonlinear propagation mechanism; calculating link credibility scores for path sequences between arbitrary entities to construct a fraud link set; and combining link credibility scores and entity structure embedding representations to calculate the inverse sensitivity of entities in fraud links, generating response control strategies to achieve source tracing control. Through multimodal perturbation modeling, energy flow coupling structure construction, and multi-factor path evaluation mechanisms, the accuracy of fraud behavior identification, collaborative chain modeling capability, and link screening robustness are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of data security tracking, specifically to a multi-factor fraud tracking method and system based on artificial intelligence. Background Technology

[0002] With the development of information technology and the continuous expansion of digital transaction scenarios, multi-source, heterogeneous, high-frequency, and dynamic forms of fraud have emerged in the network environment. Traditional fraud detection methods based on rule matching, feature clustering, or static graph mining exhibit insufficient identification capabilities, slow response times, and difficulties in tracing the source when faced with complex collaborative fraud patterns. In applications such as financial transactions, e-commerce activities, social networks, and the Internet of Things, fraud perpetrators often evade the discrimination mechanisms of conventional detection algorithms by constructing concealed collaborative links, manipulating behavioral disturbance trajectories, and avoiding feature homogeneity, creating systemic risk control blind spots.

[0003] In existing technologies, some methods attempt to introduce algorithms such as graph neural networks and graph embedding learning to enhance structural modeling and improve the accuracy of identifying fraudulent entities. However, these methods mostly rely on static adjacency relationships and fixed propagation rules, making it difficult to dynamically characterize the evolution process of behavioral perturbations and cross-modal cooperative behavioral structures. Furthermore, they lack systematic modeling mechanisms in areas such as reliable path link modeling, perturbation consistency constraints, and characterization of behavioral complexity.

[0004] Furthermore, current fraud tracking solutions generally overlook the differences in entities' ability to absorb external disturbances and the dynamic evolution of coupling response strength in multimodal behavioral scenarios. This leads to insufficient accuracy in identifying potential collaborative relationships between entities and in constructing link credibility. Simultaneously, the lack of a mechanism for unified modeling of structural collaboration and disturbance response within the path makes it difficult to develop an interpretable and controllable fraud link assessment method, thus limiting the realization of global perception capabilities and reverse tracing capabilities at the fraud chain level.

[0005] Therefore, there is an urgent need to propose a novel fraud link tracing technology that integrates multimodal disturbance analysis, energy flow coupling modeling, structural propagation mechanism and path credibility assessment to achieve highly robust identification of complex fraud behaviors, linkage modeling and refined source control. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a multi-factor fraud link tracing method and system based on artificial intelligence to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a multi-factor fraud link tracing method based on artificial intelligence, comprising:

[0008] A multimodal behavioral perturbation structure model is constructed, and the perturbation response is extracted by the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and the perturbation absorption intensity is calculated.

[0009] Based on the disturbance absorption strength and the nonlinear interaction relationship between behaviors, an energy flow coupling relationship graph between entities is constructed, and entity pairs with potential fraudulent cooperative behavior are identified as high-strength connection edges.

[0010] The structure propagation and embedding representation update are performed based on the energy flow coupling graph, and the entity structure embedding representation is generated iteratively using a weight-driven nonlinear propagation mechanism.

[0011] The credibility of path sequences between any entities is evaluated, the link credibility score is calculated, and a set of fraudulent links with perturbation consistency, path coordination and low behavioral complexity difference is constructed.

[0012] By combining link credibility scoring and entity structure embedding representation, a path reverse contribution function is constructed to calculate the reverse sensitivity of an entity in a fraud link and generate a response control strategy to achieve source tracing control.

[0013] The present invention is further configured such that the step of constructing the multimodal behavioral perturbation structure model includes:

[0014] The perturbation response model is performed on the time evolution process of multidimensional behavioral source signals in different modal domains, and the dynamic perturbation trajectory with nonlinear variation characteristics is extracted.

[0015] Based on the dynamic perturbation trajectory of behavior, the perturbation energy consumption is constructed to obtain the perturbation absorbed energy under the mode;

[0016] By integrating the absorbed energy across all modes, the intermodal perturbation absorption intensity is calculated, thus measuring the response capability and absorption characteristics of a behavioral entity to external perturbations in different modes.

[0017] The present invention is further configured such that the process of constructing the energy flow coupling relationship graph between entities includes:

[0018] Generate the behavioral coupling response degree between entity pairs based on the temporal evolution results of the multimodal behavioral interaction function;

[0019] Energy flow sensing weights between entity pairs are generated based on the perturbation absorption intensity and the behavioral coupling responsivity.

[0020] Based on a preset coupling strength threshold, entities with high-weight connections in the graph structure are determined as edge connection nodes, forming an energy flow coupling relationship graph between entities.

[0021] The present invention is further configured such that the generation of the entity structure embedding representation includes:

[0022] The initial entity structure embedding representation is assigned based on the perturbation absorption intensity of each entity;

[0023] Based on the constructed energy flow coupling graph, and based on the adjacency relationship of each entity, the diffusion contribution of the adjacent entity to the current round entity structure embedding representation is calculated through the edge weight-driven nonlinear propagation mechanism.

[0024] The entity structure embedding representation of the previous round is nonlinearly combined with the diffusion contribution of the current round to form the entity structure embedding representation of the next round.

[0025] Multiple propagation and update rounds are performed until the entity structure embedding representation converges or reaches the preset iteration limit, ultimately obtaining the stable entity structure embedding representation for each entity.

[0026] The present invention is further configured such that constructing the fraud link set includes:

[0027] The path sequence between any pair of entities is extracted, and a set of path factors is constructed based on the perturbation absorption intensity of each node in the path, the entity structure embedding representation, and the complex coding features of behavior.

[0028] A perturbation consistency factor is constructed based on the perturbation absorption intensity in the path;

[0029] Construct a path synergy factor based on the entity structure embedding representation in the path;

[0030] Construct a behavior complexity difference factor based on the behavior complexity encoding in the path;

[0031] Calculate the link credibility score corresponding to the path according to the preset factor fusion rules;

[0032] By combining the link credibility scoring threshold, path sequences with high credibility characteristics are selected from all paths to construct a fraud link set.

[0033] The present invention is further configured such that the construction logic of the perturbation consistency factor is as follows:

[0034] Based on the path factor set, the disturbance absorption intensity of each entity node is extracted;

[0035] The consistency of the in-path disturbance response is measured based on the nonlinear variation characteristics of the disturbance absorption intensity between adjacent nodes.

[0036] By setting the disturbance absorption offset and the power-law factor, the overall disturbance consistency factor is constructed by multiplying the pairs of disturbance differences within the path.

[0037] The present invention is further configured such that the construction logic of the path synergy factor is as follows:

[0038] Based on any pair of adjacent entity nodes in any path, obtain their entity structure embedding representation, apply idempotent operations to each pair, and then perform element-wise mapping and fusion.

[0039] A chain-nested product approach is used to compress the fully embedded coupling sequence, construct a nonlinear feature mapping function, extract the stable kernel response of the multi-level embedded product features, and generate a path structure synergy factor.

[0040] The present invention is further configured such that the calculation of the behavioral complexity encoding includes:

[0041] Obtain the original sequence of behavioral events of the target entity in multiple behavioral modal domains, including interactive behavior, triggering behavior, response behavior, and modal transition behavior;

[0042] The original behavioral event sequence is subjected to structured discretization processing to extract a multidimensional set of behavioral features;

[0043] Based on a multidimensional set of behavioral features, an original behavioral complexity factor is constructed.

[0044] The original behavioral complexity factor is perturbed and enhanced by nonlinear interaction to generate a behavioral complexity code.

[0045] The present invention is further configured such that the generation response control strategy includes:

[0046] Based on link credibility scoring and entity structure embedding representation, a path reverse contribution function is constructed to model the structural offset of terminal nodes relative to path source nodes in fraudulent links;

[0047] Calculate the reverse sensitivity for each entity in the fraud path it participates in;

[0048] Inductive modeling is performed on all inverse sensitivities. By combining the structural embedding distance between entities and the interference modulation factor, a set of response control strategies is constructed to indicate the distribution of control levels of different entities in the traceability chain.

[0049] This invention also provides an artificial intelligence-based multi-factor fraud tracking system, the system comprising:

[0050] Multimodal perturbation extraction module: Constructs a multimodal behavioral perturbation structure model, extracts the perturbation response through the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and calculates the perturbation absorption intensity;

[0051] Energy flow coupling relationship construction module: Based on the disturbance absorption intensity and the nonlinear interaction relationship between behaviors, construct the energy flow coupling relationship graph between entities, and identify entity pairs with potential fraudulent cooperative behavior as high-strength connection edges;

[0052] Structure propagation and embedding update module: Based on the energy flow coupling graph, structure propagation and embedding representation update are performed, and the entity structure embedding representation is generated iteratively using a weight-driven nonlinear propagation mechanism.

[0053] Fraud link credibility assessment module: evaluates the credibility of path sequences between any entities, calculates link credibility scores, and constructs a set of fraud links with perturbation consistency, path coordination and low behavioral complexity differences;

[0054] Reverse tracing control decision module: Combining link credibility scoring and entity structure embedding representation, constructing a path reverse contribution function, calculating the reverse sensitivity of an entity in a fraud link, and generating a response control strategy to achieve tracing control.

[0055] This invention provides a multi-factor fraud link tracing method and system based on artificial intelligence. The method constructs a multimodal behavioral perturbation structure model, extracts the perturbation response through the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and calculates the perturbation absorption intensity. Based on the perturbation absorption intensity and the nonlinear interaction relationship between behaviors, it constructs an energy flow coupling relationship graph between entities, identifying entities with potential fraudulent collaborative behavior as high-strength connection edges. It performs structure propagation and embedding representation updates based on the energy flow coupling relationship graph, iteratively generating entity structure embedding representations using an edge-weight-driven nonlinear propagation mechanism. It evaluates the credibility of path sequences between arbitrary entities, calculates link credibility scores, and constructs a fraud link set with perturbation consistency, path collaboration, and low behavioral complexity differences. Combining the link credibility scores and entity structure embedding representations, it constructs a path back contribution function, calculates the back sensitivity of entities in fraudulent links, and generates a response control strategy to achieve source tracing control. The beneficial effects include:

[0056] 1. Improve the accuracy of disturbance response modeling: By constructing a multimodal behavioral disturbance structure model, the nonlinear disturbance absorption characteristics of multi-source behavioral signals in each modal domain are systematically extracted to characterize the dynamic response capability of entities to external disturbances and enhance the fine-grained expressive capability of fraud behavior modeling;

[0057] 2. Enhance entity collaboration identification capability: By constructing an energy flow coupling relationship graph, a coupling mechanism between disturbance absorption intensity and behavioral interaction response is introduced to realize structural modeling of potential collaborative behavior chains between entities, thereby improving the accuracy of entity correlation identification in complex fraud behavior networks;

[0058] 3. Construct a multi-factor path credibility scoring mechanism: By introducing multi-dimensional path factors such as perturbation consistency factor, path synergy factor, and behavioral complexity difference, a fine-grained link credibility scoring model is constructed, which significantly improves the identification accuracy and screening robustness of fraudulent paths.

[0059] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0061] Figure 1 A flowchart illustrating an artificial intelligence-based multi-factor fraud tracing method is shown as an exemplary embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram illustrating the structure of an artificial intelligence-based multi-factor fraud tracing system, which is an exemplary embodiment of the present invention. Detailed Implementation

[0063] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0064] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0065] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0066] Example 1:

[0067] A multi-factor fraud tracing method based on artificial intelligence, such as Figure 1 As shown, it includes:

[0068] A multimodal behavioral perturbation structure model is constructed, and the perturbation response is extracted by the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and the perturbation absorption intensity is calculated.

[0069] Based on the disturbance absorption strength and the nonlinear interaction relationship between behaviors, an energy flow coupling relationship graph between entities is constructed, and entity pairs with potential fraudulent cooperative behavior are identified as high-strength connection edges.

[0070] The structure propagation and embedding representation update are performed based on the energy flow coupling graph, and the entity structure embedding representation is generated iteratively using a weight-driven nonlinear propagation mechanism.

[0071] The credibility of path sequences between any entities is evaluated, the link credibility score is calculated, and a set of fraudulent links with perturbation consistency, path coordination and low behavioral complexity difference is constructed.

[0072] By combining link credibility scoring and entity structure embedding representation, a path reverse contribution function is constructed to calculate the reverse sensitivity of an entity in a fraud link and generate a response control strategy to achieve source tracing control.

[0073] The present invention is further configured such that the step of constructing the multimodal behavioral perturbation structure model includes:

[0074] The perturbation response model is performed on the time evolution of multidimensional behavioral source signals in different modal domains. Dynamic perturbation trajectories with nonlinear variation characteristics are extracted. Specifically, an intramodal perturbation response structure is constructed. Based on the nonlinear rate of change of the behavioral signal along the time axis and the abrupt change sensitivity factor, the perturbation response basis is formed, as shown in the following formula: ,in, For behavioral entities exist Time-mode The following behavioral timing signals, As a component of behavioral acceleration, it is used to detect mutation trends. The nonlinear acceleration response coefficients of the modal are... This is a power-law amplification of the first-order growth rate of the mode. To activate the bias for the perturbation, Let be the time-growth tension function, where , For the disturbance response mapping function, ;

[0075] The disturbance energy consumption is constructed based on the dynamic disturbance trajectory, the disturbance absorbed energy under different modes is obtained, the disturbance absorbed power density function is defined, and the disturbance energy consumption is constructed by combining the response rate of change and the mode amplification function. The formula is as follows: ,in, For the first Mode disturbances absorb energy. The intensity of the change in the disturbance response. For modal absorption converters, a tradeoff is made between behavioral response and time-varying factors;

[0076] The intermodal perturbation absorption intensity is calculated by integrating the absorbed energy across all modes, measuring the behavioral entity's response to external perturbations and its absorption characteristics in different modes. The formula for calculating the perturbation absorption intensity is as follows: ,in, The perturbation absorption strength reflects the degree of instability of an entity in multimodal behavior and is used for fraud potential modeling. The modal disequilibrium amplification index, For the number of modes, These are the full-modal power reduction coefficients, which control the convergence rate of modal fusion.

[0077] The present invention is further configured such that the process of constructing the energy flow coupling relationship graph between entities includes:

[0078] The behavioral coupling response degree between entity pairs is generated based on the temporal evolution results of the multimodal behavioral interaction function. Specifically, the entity set is defined. Family of intermodal interaction functions , representing entities and In modality , The calculation logic for the behavioral coupling response degree of the cross-behavioral response curve is as follows: ,in, For entity pairs , Behavioral coupling responsivity, This is the first-order time derivative of the interaction function between behavioral modalities, used to characterize the interaction rate. It is a time-sensitive modulation function. As a regulating factor;

[0079] Energy flow sensing weights between entity pairs are generated based on the perturbation absorption strength and behavioral coupling responsivity. These weights are then jointly and nonlinearly reconstructed using the perturbation absorption strength and behavioral coupling responsivity of the two entities to construct the energy flow sensing weights. ,in, , For entities , The intensity of disturbance absorption, , To absorb the weighting index, The behavior coupling response amplification index, As a regulating factor, harmonic convergence scalar control, As a perturbation isomerism enhancement factor, Assign weights to the energy flow between entity pairs;

[0080] Based on a preset coupling strength threshold, entity pairs with high-weight connections in the graph structure are identified as edge nodes, forming a graph of energy flow coupling relationships between entities, thus defining the graph. The edge weight function is derived from Decision, and setting of coupling strength threshold. To determine high-risk connecting edges, the edge set is defined by the following formula: ,like Then and A connected edge indicates the existence of potential fraudulent collaborative behavior.

[0081] The present invention is further configured such that the generation of the entity structure embedding representation includes:

[0082] The initial entity structure embedding representation is assigned based on the perturbation absorption strength of each entity. Specifically, for each entity... Assigning initial entity structure representation Its elements and the perturbation absorption intensity are nonlinearly scaled: , ,in, For dimension The initial entity structure embedding representation, For embedded dimensions, It is a non-linear weighted index. Initialize the noise disturbance for the disturbance;

[0083] Based on the constructed energy flow coupling graph, and considering the adjacency relationships of each entity, the diffusion contribution of adjacent entities to the current round entity structure embedding representation is calculated through an edge-weighted nonlinear propagation mechanism. The above defines a propagation operation with edge-weighted activation functions, which couples and diffuses all entity structure representations of each entity through their adjacent edges. The propagation formula is as follows: ,in, Contribution to diffusion For edge weight activation index, For entities The set of adjacent entities, This represents the embedded representation of adjacent nodes in the entity structure of the previous round. This is an adaptive perturbation weight vector driven by edge weights. Element-wise multiplication It is a nonlinear mapping kernel function;

[0084] The entity structure embedding representation of the previous round is nonlinearly combined with the diffusion contribution of the current round to form the entity structure embedding representation of the next round, as shown in the formula: , , To update the scaling factor, This indicates a splicing operation. This is an embedding representation of the entity structure from the previous round. For the perturbation gating function, Indicates bitwise exponentiation;

[0085] Multiple propagation and update rounds are performed until the entity structure embedding representation converges or reaches a preset iteration limit, ultimately yielding a stable entity structure embedding representation for each entity. , For the number of times it is spread.

[0086] The present invention is further configured such that constructing the fraud link set includes:

[0087] For any pair of entities, the path sequence is extracted, and a path factor set and a set of all reachable path sequences are constructed based on the perturbation absorption strength of each node in the path, the entity structure embedding representation, and the behavioral complexity coding features. Each path Represented as an ordered sequence of entities ;

[0088] A perturbation consistency factor is constructed based on the perturbation absorption intensity in the path. ;

[0089] Constructing a path collaboration factor based on the entity structure embedding representation in the path. ;

[0090] Complex encoding based on path behavior Constructing behavioral complexity difference factor , specifically ,in, For the difference regularization term, A smaller value indicates higher stability of the path behavior;

[0091] According to the preset factor fusion rules, the link credibility score corresponding to the path is calculated using the following formula: ,in, , , To balance the regulatory factors, For overall compression modulation parameters, Score the reliability of the link;

[0092] By combining link credibility scoring thresholds to filter path sequences with high credibility characteristics from all paths, a fraud link set is constructed, and a preset link credibility scoring threshold is set. Then the fraud link set .

[0093] The present invention is further configured such that the construction logic of the perturbation consistency factor is as follows:

[0094] Based on the path factor set, the disturbance absorption intensity of each entity node is extracted;

[0095] The consistency of the in-path disturbance response is measured based on the nonlinear variation characteristics of the disturbance absorption intensity between adjacent nodes.

[0096] By setting a disturbance absorption offset and an power-law factor, and multiplying the differences between disturbance pairs within the path, an overall disturbance consistency factor is constructed. Specifically, for the path... Extract the perturbation absorption encoded value sequence of all path nodes. Constructing a perturbation consistency factor The calculation formula is: ,in, For the first The intensity of disturbance absorption at each node For perturbation power factor, For disturbance absorption offset; This is the perturbation consistency factor; the closer the value is to 1, the more consistent the perturbation response within the path. This represents the number of nodes in the path.

[0097] The present invention is further configured such that the construction logic of the path synergy factor is as follows:

[0098] Based on any pair of adjacent entity nodes in any path, obtain their entity structure embedding representation, apply idempotent operations to each pair, and then perform element-wise mapping and fusion.

[0099] A chain-nested multiplication approach is used to compress the fully embedded coupled sequence, construct a nonlinear feature mapping function, extract stable kernel responses of multi-level embedded multiplication features, and generate path coherence factors. Specifically, the path... Each pair of continuous entities , The structural embedding representations are respectively , Define path synergy factor The chained kernel embedding compression result for the product of all embedding mapping vectors along the path: ,in, For embedding exponentiation coefficients in the structure, For element-wise multiplication, This is a chained nested product operator. The mapping compression function is defined as follows: , , A larger value indicates stronger path structure coordination.

[0100] The present invention is further configured such that the calculation of the behavioral complexity encoding includes:

[0101] Obtain the original sequence of behavioral events of the target entity in multiple behavioral modal domains, including interactive behavior, triggering behavior, response behavior, and modal transition behavior;

[0102] The original behavioral event sequence is subjected to structured discretization processing to extract a multidimensional set of behavioral features;

[0103] Based on a multidimensional set of behavioral features, an original behavioral complexity factor is constructed.

[0104] The original behavioral complexity factors are perturbed and enhanced with nonlinear interactions to generate behavioral complexity codes. Specifically, each entity's code is obtained. Observational behavior feature set: :A set of interactive trigger sequences (such as login, payment, transfer, etc.); Temporal density feature sequence (frequency of behavior occurring per unit time); : Behavior response type mapping (e.g., multi-turn interactions, request chain distribution); : Heterogeneous modal transition feature sequence (number of behavioral modal jumps and path);

[0105] Constructing the original behavioral complexity factor: ,in This is a structural compression coding function that outputs a behavioral complexity factor.

[0106] The original behavioral complexity factor is perturbed and enhanced by nonlinear interaction to generate a behavioral complexity code: ,in, Encoding complex behaviors For the first Each behavioral dimension feature Let be the dimension of the behavioral features. As a local suppression modulation factor, This is the singular response correction constant. To amplify the perturbation complexity parameters, It is a non-linear aggregate power exponent.

[0107] The present invention is further configured such that the generation response control strategy includes:

[0108] Based on entity structure embedding representation, a path reverse contribution function is constructed to model the structural offset of terminal nodes relative to path source nodes in a fraudulent link. The formula for constructing the path reverse contribution function is as follows: ,in, For nodes In the 3D structure embedded components, The source node suppressor modulation factor, For embedded dimensions, For the coordinated offset damping coefficient, The nonlinear embedded perturbation index. As a metric for balancing path structure offset and reverse contribution, The starting node of each path , End node ;

[0109] Based on link credibility scoring and path reverse contribution function, the reverse sensitivity of each entity in the fraud path it participates in is calculated using the following formula: ,in, For containing nodes A subset of fraud paths Score the path credibility. Contribute the function in the reverse direction of the path. Nodes in the path The normalized positional influence factor to the endpoint node is defined as follows: ,in This represents the position of the node in the path. As multiple trade-off factors, For nodes Reverse sensitivity to terminal fraud;

[0110] Inductive modeling is performed on all inverse sensitivities. Combining the inter-entity structural embedding distance and interference modulation factor, a set of response control strategies is constructed to indicate the control level distribution of different entities in the tracing chain, thus defining the control strategy set. Generate control levels for each sensitive node. The format is as follows: ,in, As a basic control factor, To control activation weights, As the sensitivity index compression factor, For entities Local interference factors, An inter-entity structural distance metric is embedded, defined as: , For distance suppression modulation parameters, For entities The final response control strategy strength.

[0111] Example 2:

[0112] Please see Figure 2 This exemplary AI-based multi-factor fraud tracing system includes:

[0113] Multimodal perturbation extraction module: Constructs a multimodal behavioral perturbation structure model, extracts the perturbation response through the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and calculates the perturbation absorption intensity;

[0114] Energy flow coupling relationship construction module: Based on the disturbance absorption intensity and the nonlinear interaction relationship between behaviors, construct the energy flow coupling relationship graph between entities, and identify entity pairs with potential fraudulent cooperative behavior as high-strength connection edges;

[0115] Structure propagation and embedding update module: Based on the energy flow coupling graph, structure propagation and embedding representation update are performed, and the entity structure embedding representation is generated iteratively using a weight-driven nonlinear propagation mechanism.

[0116] Fraud link credibility assessment module: evaluates the credibility of path sequences between any entities, calculates link credibility scores, and constructs a set of fraud links with perturbation consistency, path coordination and low behavioral complexity differences;

[0117] Reverse tracing control decision module: Combining link credibility scoring and entity structure embedding representation, constructing a path reverse contribution function, calculating the reverse sensitivity of an entity in a fraud link, and generating a response control strategy to achieve tracing control.

[0118] It should be noted that the AI-based multi-factor fraud tracing system and the AI-based multi-factor fraud tracing method provided in the above embodiments belong to the same concept. The specific methods by which each module and unit performs its operations have been described in detail in the method embodiments and will not be repeated here. In practical applications, the AI-based multi-factor fraud tracing system provided in the above embodiments can be configured to have different functional modules assigned to it as needed. That is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above, and this is not a limitation here.

[0119] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0120] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0121] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0122] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0123] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0125] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0126] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0127] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0128] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A multi-factor fraud tracing method based on artificial intelligence, characterized in that, include: A multimodal behavioral perturbation structure model is constructed, and the perturbation response is extracted by the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and the perturbation absorption intensity is calculated. Based on the perturbation absorption strength and the nonlinear interaction relationship between behaviors, an energy flow coupling relationship graph between entities is constructed, and entity pairs with potential fraudulent cooperative behavior are identified as high-strength connection edges. The structure propagation and embedding representation update are performed based on the energy flow coupling graph, and the entity structure embedding representation is generated iteratively using a weight-driven nonlinear propagation mechanism. The credibility of path sequences between arbitrary entities is evaluated, link credibility scores are calculated, and a set of fraudulent links with perturbation consistency, path synergy, and low behavioral complexity difference is constructed. The construction of the fraudulent link set includes: extracting path sequences between arbitrary entity pairs; constructing a path factor set based on the perturbation absorption strength, entity structure embedding representation, and behavioral complexity coding features of each node in the path; constructing a perturbation consistency factor based on the perturbation absorption strength in the path; constructing a path synergy factor based on the entity structure embedding representation in the path; constructing a behavioral complexity difference factor based on the behavioral complexity coding in the path; calculating the link credibility score corresponding to the path according to a preset factor fusion rule; and selecting path sequences with high credibility features from all paths based on the link credibility score threshold to construct the fraudulent link set. The construction logic of the perturbation consistency factor is as follows: based on the path factor set, the perturbation absorption strength of each entity node is extracted; and the perturbation within the path is calculated based on the nonlinear variation characteristics of the perturbation absorption strength between adjacent nodes. Dynamic response consistency measurement; setting disturbance absorption offset and power-law factor, multiplying and accumulating the disturbance difference pairs within the path to construct the overall disturbance consistency factor; the construction logic of the path synergy factor is as follows: based on any pair of adjacent entity nodes in any path, obtain their entity structure embedding representation, apply power-law operation respectively, and then perform element-wise mapping fusion; use a chain-like nested multiplication method to compress the full path embedding coupling sequence, construct a nonlinear feature mapping function, extract the stable kernel response of multi-level embedding product features, and generate the path structure synergy factor; the calculation of behavioral complexity coding includes: obtaining the original behavioral event sequence of the target entity in multiple behavioral modal domains, including interactive behavior, triggering behavior, response behavior, and modal jump behavior; performing structured discretization processing on the original behavioral event sequence to extract a multi-dimensional behavioral feature set; constructing the original behavioral complexity factor based on the multi-dimensional behavioral feature set; performing disturbance enhancement and nonlinear interactive fusion on the original behavioral complexity factor to generate behavioral complexity coding; By combining link credibility scoring and entity structure embedding representation, a path reverse contribution function is constructed to calculate the reverse sensitivity of an entity in a fraud link and generate a response control strategy to achieve source tracing control.

2. The multi-factor fraud link tracing method based on artificial intelligence according to claim 1, characterized in that, The steps for constructing a multimodal behavioral perturbation structure model include: The perturbation response model is performed on the time evolution process of multidimensional behavioral source signals in different modal domains, and the dynamic perturbation trajectory with nonlinear variation characteristics is extracted. Based on the dynamic perturbation trajectory of behavior, the perturbation energy consumption is constructed to obtain the perturbation absorbed energy under the mode; By integrating the absorbed energy across all modes, the intermodal perturbation absorption intensity is calculated, thus measuring the response capability and absorption characteristics of a behavioral entity to external perturbations in different modes.

3. The multi-factor fraud link tracing method based on artificial intelligence according to claim 2, characterized in that, The process of constructing the energy flow coupling relationship diagram between entities includes: Generate the behavioral coupling response degree between entity pairs based on the temporal evolution results of the multimodal behavioral interaction function; Energy flow sensing weights between entity pairs are generated based on the perturbation absorption intensity and the behavioral coupling responsivity. Based on a preset coupling strength threshold, entities with high-weight connections in the graph structure are determined as edge connection nodes, forming an energy flow coupling relationship graph between entities.

4. A multi-factor fraud tracing method based on artificial intelligence according to claim 2 or 3, characterized in that, The generation of entity structure embedding representations includes: The initial entity structure embedding representation is assigned based on the perturbation absorption intensity of each entity; Based on the constructed energy flow coupling graph, and based on the adjacency relationship of each entity, the diffusion contribution of adjacent entities to the current round entity structure embedding representation is calculated through an edge weight-driven nonlinear propagation mechanism. The entity structure embedding representation of the previous round is nonlinearly combined with the diffusion contribution of the current round to form the entity structure embedding representation of the next round. Multiple propagation and update rounds are performed until the entity structure embedding representation converges or reaches the preset iteration limit, ultimately obtaining the stable entity structure embedding representation for each entity.

5. The multi-factor fraud link tracing method based on artificial intelligence according to claim 4, characterized in that, The generation response control strategy includes: Based on link credibility scoring and entity structure embedding representation, a path reverse contribution function is constructed to model the structural offset of terminal nodes relative to path source nodes in fraud links; Calculate the reverse sensitivity for each entity in the fraud path it participates in; Inductive modeling is performed on all inverse sensitivities. By combining the structural embedding distance between entities and the interference modulation factor, a set of response control strategies is constructed to indicate the distribution of control levels of different entities in the traceability chain.

6. A multi-factor fraud tracing system based on artificial intelligence, used to implement the multi-factor fraud tracing method based on artificial intelligence as described in any one of claims 1-5, characterized in that, include: Multimodal perturbation extraction module: Constructs a multimodal behavioral perturbation structure model, extracts the perturbation response through the temporal evolution of multidimensional behavioral source signals in multiple modal domains, and calculates the perturbation absorption intensity; Energy flow coupling relationship construction module: Based on the disturbance absorption intensity and the nonlinear interaction relationship between behaviors, construct the energy flow coupling relationship graph between entities, and identify entity pairs with potential fraudulent cooperative behavior as high-strength connection edges; Structure propagation and embedding update module: Based on the energy flow coupling graph, structure propagation and embedding representation update are performed, and the entity structure embedding representation is generated iteratively using a weight-driven nonlinear propagation mechanism. Fraud link credibility assessment module: evaluates the credibility of path sequences between any entities, calculates link credibility scores, and constructs a set of fraud links with perturbation consistency, path coordination and low behavioral complexity differences; Reverse tracing control decision module: Combining link credibility scoring and entity structure embedding representation, constructing a path reverse contribution function, calculating the reverse sensitivity of an entity in a fraud link, and generating a response control strategy to achieve tracing control.

Images