Risk early warning method and system applied to network transaction

Abstract

The invention provides a risk early warning method and system applied to network transaction, and the method comprises the steps: obtaining a basic transaction behavior connection array corresponding to a plurality of behavior relation links in a sample network transaction behavior graph, and comprehensively capturing an actual behavior mode in the network transaction; a deep relation network unit is utilized to perform potential interaction relation extraction on the basic transaction behavior connected array, a derivative transaction behavior connected array is generated, the hidden and complex interaction relation between transaction subjects is further revealed, then according to the derivative transaction behavior connected array, an example network transaction behavior graph is subjected to embedded representation, and the network transaction behavior graph is obtained. A graph is generated to be embedded into a representation vector sequence, so that a subsequent risk early warning model can be conveniently learned and analyzed. And finally, performing model parameter optimization on the network transaction risk early warning model according to the graph embedding representation vector sequence, and performing risk label estimation on the target network transaction behavior graph of any user, thereby realizing accurate early warning of potential transaction risks and timely pushing of risk early warning information.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a risk warning method and system for network transactions. Background Art

[0002] With the rapid development of the digital economy and the increasing popularity of online transactions, the scale of financial transactions has continued to expand and the means of transactions have become increasingly diverse. However, this has also brought about the problem of increased transaction risks, especially the increasingly rampant criminal activities such as illegal fund transfers and telecommunications fraud, which have posed a serious threat to financial security. Traditional risk warning methods mainly rely on manual review and simple rule matching, which are difficult to cope with the increasingly complex online transaction environment. Therefore, a more efficient and accurate risk warning method is urgently needed.

[0003] Existing network transaction risk warning methods are mostly based on a single data source and analysis model, which makes it difficult to fully capture the complex behavior patterns in network transactions. Especially when dealing with massive amounts of network transaction data, traditional analysis methods often face problems such as high data dimensions and sparse features, resulting in unsatisfactory warning effects. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides a risk warning method applied to network transactions, the method comprising: Obtain a connected array of basic transaction behaviors corresponding to multiple behavior relationship links in the sample network transaction behavior graph; Extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array; Embedding the sample network transaction behavior graph according to the derived transaction behavior connectivity array to generate a graph embedding representation vector sequence of the sample network transaction behavior graph; The model parameters of the network transaction risk warning model are optimized according to the graph embedding representation vector sequence of the sample network transaction behavior graph, and the network transaction risk label of the target network transaction behavior graph of any user is estimated based on the network transaction risk warning model after the model parameter optimization is completed. After the corresponding network transaction risk label is generated, the risk warning information is pushed to the user.

[0005] In a possible implementation of the first aspect, the network transaction risk warning model includes a plurality of deep relationship network units functionally connected according to feature orientation; The step of extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array includes: Using each of the deep relationship network units to extract potential interactive relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links, and generating a derived transaction behavior connectivity array corresponding to each of the deep relationship network units; The embedding representation of the sample network transaction behavior graph according to the derived transaction behavior connectivity array to generate a graph embedding representation vector sequence of the sample network transaction behavior graph includes: The sample network transaction behavior graph is embedded according to feature-pointing functional connections based on the derived transaction behavior connected array corresponding to the multiple deep relationship network units that are functionally connected according to feature pointing, and a graph embedding representation vector sequence of the sample network transaction behavior graph is generated.

[0006] In a possible implementation of the first aspect, using each of the deep relationship network units to extract potential interactive relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array corresponding to each of the deep relationship network units includes: Using any of the deep relationship network units to compare the characteristics of the transaction behavior trajectory vectors of the two network transaction behavior subjects corresponding to any of the behavior relationship links, and generating the correlation between the two network transaction behavior subjects corresponding to any of the behavior relationship links; Accumulating the correlations between two network transaction behavior subjects corresponding to a plurality of the behavior relationship links to generate a cumulative correlation, and taking the ratio of the correlation between two network transaction behavior subjects corresponding to any one of the behavior relationship links to the cumulative correlation as the mutual influence coefficient corresponding to any one of the behavior relationship links; According to the mutual influence coefficient corresponding to any one of the behavior relationship links, the link knowledge vector corresponding to any one of the behavior relationship links is blended to generate a blended link knowledge vector corresponding to any one of the behavior relationship links; According to a sequence of intermingling link knowledge vectors corresponding to a plurality of said behavior relationship links, a sequence of intermingling link knowledge vectors corresponding to each of said deep relationship network units is formed; The potential interaction relationship of the basic transaction behavior connectivity array is extracted according to the fusion link knowledge vector sequence corresponding to each of the deep relationship network units, and a derivative transaction behavior connectivity array corresponding to each of the deep relationship network units is generated.

[0007] In a possible implementation of the first aspect, using each of the deep relationship network units to extract potential interactive relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array corresponding to each of the deep relationship network units includes: Using the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of the plurality of deep relationship network units, a progressive potential interaction relationship extraction is performed on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit to generate a connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit; Among them, x increases successively and 2≤x≤M, M is the number of the multiple deep relationship network units functionally connected according to the feature pointing, and the derived transaction behavior connectivity array corresponding to the first deep relationship network unit is generated by extracting potential interactive relationships from the basic transaction behavior connectivity array based on the link knowledge vectors corresponding to the multiple behavior relationship links.

[0008] In a possible implementation of the first aspect, the method of utilizing the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of the plurality of deep relationship network units to perform progressive potential interaction relationship extraction on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit to generate the connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit includes: The basic transaction behavior connectivity array is distinguished and compared with the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit to generate a first distinguished transaction behavior connectivity array, wherein the first distinguished transaction behavior connectivity array includes the behavior relationship links that are not extracted from the x-1th deep relationship network unit; Extracting potential interaction relationships from the first distinguishing transaction behavior connected array according to the link knowledge vectors respectively corresponding to the plurality of behavior relationship links, and generating a second distinguishing transaction behavior connected array; The derivative transaction behavior connected array corresponding to the x-1th deep relationship network unit is merged with the second distinguished transaction behavior connected array to generate a derivative transaction behavior connected array corresponding to the xth deep relationship network unit.

[0009] In a possible implementation of the first aspect, extracting potential interaction relationships from the first distinguishable transaction behavior connectivity array according to the link knowledge vectors respectively corresponding to the plurality of behavior relationship links to generate a second distinguishable transaction behavior connectivity array includes: fusing the link knowledge vectors respectively corresponding to the plurality of behavior relationship links to generate a fusing link knowledge vector sequence; The potential interaction relationship of the first differentiated transaction behavior connected array is extracted according to the fusion link knowledge vector sequence to generate a second differentiated transaction behavior connected array.

[0010] In a possible implementation of the first aspect, the method of utilizing the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of the plurality of deep relationship network units to perform progressive potential interaction relationship extraction on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit to generate the connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit includes: Extracting potential interaction relationships from the derivative transaction behavior connectivity array corresponding to the x-1th deep relationship network unit according to the link knowledge vectors corresponding to the multiple behavior relationship links, and generating a reference transaction behavior connectivity array; The derivative transaction behavior connectivity array corresponding to the x-1th deep relationship network unit is distinguished and compared with the reference transaction behavior connectivity array, and the generated third distinguished transaction behavior connectivity array is used as the derivative transaction behavior connectivity array corresponding to the xth deep relationship network unit.

[0011] In a possible implementation of the first aspect, the embedding representation of the sample network transaction behavior graph according to the feature-pointing functional connections based on the derived transaction behavior connectivity array corresponding to the multiple deep relationship network units that are functionally connected according to the feature-pointing functions is generated to generate a graph embedding representation vector sequence of the sample network transaction behavior graph, including: Embedding the sample network transaction behavior graph by using a first deep relationship network unit of the plurality of deep relationship network units functionally connected according to feature pointing and a connected array of derived transaction behaviors corresponding to the first deep relationship network unit; Loading the embedding representation result of the first deep relational network unit into the backward connected deep relational network unit, so that the backward connected deep relational network unit continues to perform embedding representation and output until it is loaded into the last deep relational network unit; Using the embedding representation result generated by the last deep relationship network unit as the graph embedding representation vector sequence of the sample network transaction behavior graph; In a possible implementation of the first aspect, the step of loading the embedding representation result of the first deep relationship network unit into a backward-connected deep relationship network unit so that the backward-connected deep relationship network unit continues to perform embedding representation and output until the last deep relationship network unit is loaded includes: Utilizing the yth deep relationship network unit of multiple deep relationship network units, dot product operations are performed on the connected array of derivative transaction behaviors corresponding to the yth deep relationship network unit, the embedded representation result of the y-1th deep relationship network unit, and the trainable weight array of the yth deep relationship network unit to generate the embedded representation result of the yth deep relationship network unit, and the embedded representation result of the yth deep relationship network unit is loaded into the y+1th deep relationship network unit; wherein y increases successively and 2≤y≤M-1, and M is the number of the multiple deep relationship network units functionally connected according to feature pointing.

[0012] In a possible implementation of the first aspect, the optimizing model parameters of the network transaction risk warning model according to the graph embedding representation vector sequence of the sample network transaction behavior graph includes: Using the network transaction risk early warning model, estimating the network transaction risk label of the graph embedding representation vector sequence of the sample network transaction behavior graph, and generating an estimated network transaction risk label of the sample network transaction behavior graph; Calculating the training cost of the network transaction risk warning model based on the estimated network transaction risk label of the sample network transaction behavior graph and the training supervision data of the sample network transaction behavior graph; The network transaction risk warning model is optimized based on the training cost value until the training cost value no longer continues to decrease, and the network transaction risk warning model is output as a target network transaction risk warning model.

[0013] On the other hand, an embodiment of the present application also provides a risk warning system for online transactions, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method. Based on the above aspects, the embodiment of the present application can fully capture the actual behavior patterns in network transactions by obtaining the basic transaction behavior connectivity array corresponding to multiple behavior relationship links in the sample network transaction behavior graph. Secondly, the deep relationship network unit is used to extract the potential interactive relationship of the basic transaction behavior connectivity array, and the derived transaction behavior connectivity array is generated, which effectively reveals the hidden and complex interactive relationship between the transaction subjects and improves the accuracy and comprehensiveness of risk warning. Further, the sample network transaction behavior graph is embedded based on the derivative transaction behavior connectivity array to generate a graph embedding representation vector sequence, and the complex graph structure is converted into a vector form that is easy to handle mathematically, which is convenient for subsequent risk warning models to learn and analyze. Finally, the model parameters of the network transaction risk warning model are optimized according to the graph embedding representation vector sequence, and the risk label of the target network transaction behavior graph of any user is estimated based on the optimized model, which realizes the accurate early warning of potential transaction risks and timely push of risk warning information, thereby effectively protecting the security of network transactions and the rights and interests of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic diagram of the execution flow of the risk warning method applied to network transactions provided in an embodiment of the present application.

[0015] Figure 2 It is a schematic diagram of the hardware architecture of a risk warning system for network transactions provided in an embodiment of the present application. DETAILED DESCRIPTION

[0016] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a risk warning method applied to network transactions provided by an embodiment of the present application. The risk warning method applied to network transactions is introduced in detail below.

[0017] Step S110, obtaining a connected array of basic transaction behaviors corresponding to multiple behavior relationship links in the sample network transaction behavior graph.

[0018] In detail, the sample network transaction behavior graph is a complex network graph representing network transaction behavior subjects and the transaction behavior relationships between them, including multiple network transaction behavior subjects (such as buyers, sellers, payment platforms, etc.) and the connection relationships formed between these network transaction behavior subjects through actual transaction behaviors (such as transfers, purchases, evaluations, etc.). Each node in the sample network transaction behavior graph represents a transaction behavior subject, and each edge represents a transaction behavior relationship between two subjects. For example, suppose there is an online shopping platform, buyer A purchases the goods of seller B, and completes the payment through payment platform C. In the sample network transaction behavior graph, A, B, and C exist as nodes respectively, and the edge from A to B represents the purchase behavior, the edge from A to C represents the payment behavior, and the edge from B to C may represent the seller's behavior of receiving the payment. These nodes and edges together constitute the sample network transaction behavior graph.

[0019] The behavior relationship link refers to the actual transaction behavior relationship between two or more transaction behavior subjects in the sample network transaction behavior graph, representing the specific transaction activities between the transaction behavior subjects, such as the purchase behavior between the buyer and the seller. Continuing with the above example, the purchase behavior between buyer A and seller B constitutes a behavior relationship link. This behavior relationship link is represented as an edge connecting A and B in the sample network transaction behavior graph, recording one or more purchase behaviors between A and B.

[0020] The basic transaction behavior connectivity array is a data structure (such as a two-dimensional array or matrix) used to represent the basic connection relationship between each transaction behavior subject in the sample network transaction behavior map and the frequency, amount and other information of the transaction activities. It is a mathematical representation of the behavior relationship link in the sample network transaction behavior map, which facilitates the server to perform subsequent data processing and analysis.

[0021] That is, in this embodiment, the server will first retrieve the sample network transaction behavior graph from the database, which is a complex network composed of multiple network transaction behavior subjects (such as buyers, sellers, payment platforms, etc.) and the transaction behavior relationships between multiple network transaction behavior subjects. Among them, the server pays special attention to the behavior relationship links in the sample network transaction behavior graph. These behavior relationship links represent the actual transaction behaviors between different transaction subjects, such as transfers, purchases, and evaluations. The server extracts the basic transaction behavior connected arrays corresponding to these behavior relationship links through a specific algorithm. These basic transaction behavior connected arrays can be expressed as a two-dimensional array or matrix in the data structure, which records the connection relationship between each transaction behavior subject and the frequency, amount, and other information of the transaction activities.

[0022] Step S120, extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array.

[0023] In detail, potential interaction relationship extraction refers to mining potential transaction behavior relationships that may exist but are not directly represented in the sample network transaction behavior graph by analyzing the data in the basic transaction behavior connectivity array. These potential relationships may be indirect and hidden, but they are of great significance for understanding the entire transaction network. For example, in the above example, although there is no direct transaction record between A and C, there are transactions between A and B, and between B and C. By analyzing these data, the server may find that there is a potential interaction relationship between A and C, such as A may be a potential buyer of C, or C is a potential payment platform for A.

[0024] The derived transaction behavior connectivity array is a new connectivity array generated by extracting potential interaction relationships based on the basic transaction behavior connectivity array, which includes direct transaction relationships in the basic array and potential transaction relationships mined by the algorithm. Based on the above example, the server may generate a derived transaction behavior connectivity array after extracting potential interaction relationships, which includes not only direct transaction relationships between A and B, B and C, but also potential transaction relationships between A and C. The derived transaction behavior connectivity array reflects the complex relationships in the transaction network more comprehensively than the basic transaction behavior connectivity array.

[0025] That is, in this embodiment, after obtaining the basic transaction behavior connectivity array, the server will use the deep relationship network unit to extract the potential interactive relationship of these basic transaction behavior connectivity arrays. In this process, the server will analyze the transaction behavior trajectory in each behavior relationship link, such as multiple transaction records between buyer A and seller B, and calculate the transaction correlation between the two by comparing the characteristics of these data. The server will further process these data to generate a value representing the mutual influence coefficient between transaction subjects, which reflects the degree of mutual influence between different transaction subjects. Based on these mutual influence coefficients, the server performs information fusion on the basic transaction behavior connectivity array, excavates potential and deeper transaction behavior relationships, and thus generates a derivative transaction behavior connectivity array.

[0026] Step S130, embedding the sample network transaction behavior graph according to the derived transaction behavior connected array to generate a graph embedding representation vector sequence of the sample network transaction behavior graph.

[0027] In detail, the graph embedding representation vector sequence is a sequence generated by mapping the nodes and edges in the sample network transaction behavior graph to points in a high-dimensional vector space. Specifically, the structure and feature information in the graph can be represented in the form of mathematical vectors, which is convenient for the server to perform subsequent machine learning or deep learning processing. For example, assuming that there are N nodes in the sample network transaction behavior graph, the server maps each node to a D-dimensional vector through an embedding representation algorithm. In this way, all nodes in the sample network transaction behavior graph are represented as an NxD matrix or vector sequence, which contains the structural information and transaction behavior characteristics of the graph, and can be used to train and optimize the network transaction risk warning model.

[0028] Therefore, in this embodiment, the server will then embed the sample network transaction behavior graph according to the derived transaction behavior connectivity array. The embedding process is to convert the complex graph structure into a mathematically operable vector form. Specifically, the server will use multiple deep relationship network units to point to functional connections according to the characteristics between them, and gradually map each node (i.e., transaction behavior subject) and edge (i.e., behavior relationship link) in the sample network transaction behavior graph to a point in the high-dimensional vector space. The mapping process retains the structural information and transaction behavior characteristics in the original sample network transaction behavior graph. In the end, the server will obtain a graph embedding representation vector sequence, which can comprehensively and compactly describe the key features of the sample network transaction behavior graph.

[0029] Step S140, optimizing the model parameters of the network transaction risk warning model according to the graph embedding representation vector sequence of the sample network transaction behavior graph, and estimating the network transaction risk label of the target network transaction behavior graph of any user based on the network transaction risk warning model after completing the model parameter optimization, and after generating the corresponding network transaction risk label, pushing the risk warning information to the user.

[0030] In this embodiment, after obtaining the graph embedding representation vector sequence, the server will use these graph embedding representation vector sequences to train and optimize the network transaction risk warning model. The network transaction risk warning model can be a deep learning model, such as a neural network, which predicts the potential risks in new transaction behaviors by learning the complex mapping relationship between network transaction risk warnings and network transaction risk labels. In detail, the server can use the graph embedding representation vector sequence as the model input, and provide corresponding training supervision data (such as known risk transaction labels), and adjust the model parameters through the back propagation algorithm, so that the network transaction risk warning model can more accurately identify transaction risks. Once the network transaction risk warning model is trained and achieves satisfactory performance, the server can be used to evaluate the target network transaction behavior graph of any user and generate corresponding network transaction risk labels. If a high-risk transaction behavior is detected, the server will immediately push risk warning information to the user to help the user take preventive measures in time.

[0031] Based on the above steps, the embodiment of the present application can fully capture the actual behavior patterns in network transactions by obtaining the basic transaction behavior connectivity array corresponding to multiple behavior relationship links in the sample network transaction behavior graph. Secondly, the deep relationship network unit is used to extract the potential interactive relationship of the basic transaction behavior connectivity array, and the derived transaction behavior connectivity array is generated, which effectively reveals the hidden and complex interactive relationship between the transaction subjects and improves the accuracy and comprehensiveness of risk warning. Furthermore, the sample network transaction behavior graph is embedded based on the derivative transaction behavior connectivity array to generate a graph embedding representation vector sequence, and the complex graph structure is converted into a vector form that is easy to handle mathematically, which is convenient for subsequent risk warning models to learn and analyze. Finally, the model parameters of the network transaction risk warning model are optimized according to the graph embedding representation vector sequence, and the risk label of the target network transaction behavior graph of any user is estimated based on the optimized model, which realizes the accurate early warning of potential transaction risks and timely push of risk warning information, thereby effectively protecting the security of network transactions and the rights and interests of users.

[0032] In a possible implementation, the network transaction risk warning model includes a plurality of deep relationship network units that are functionally connected according to feature orientations.

[0033] Step S120 includes: using each of the deep relationship network units to extract potential interactive relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links, and generating a derived transaction behavior connectivity array corresponding to each of the deep relationship network units.

[0034] Step S130 includes: embedding the sample network transaction behavior graph according to feature-pointing functional connections based on the derived transaction behavior connectivity array corresponding to the multiple deep relationship network units that are functionally connected according to feature pointing, and generating a graph embedding representation vector sequence of the sample network transaction behavior graph.

[0035] In this embodiment, the network transaction risk warning model is composed of a plurality of deep relationship network units, and these deep relationship network units are functionally connected according to specific feature orientations.

[0036] For example, the server built a complex network transaction risk warning model, which is not a single neural network, but a series of multiple deep relationship network units. Each deep relationship network unit is specially designed to capture and analyze specific patterns and relationships in network transaction behavior. The connections between these deep relationship network units are not random, but are precisely designed according to the types of features they can process and identify.

[0037] Therefore, the server uses each deep relationship network unit to extract potential interactive relationships from the basic transaction behavior connectivity array and generates a corresponding derivative transaction behavior connectivity array.

[0038] In detail, the basic transaction behavior connectivity array reflects the basic connection relationship and activity status between each transaction subject in the network transaction behavior map. The server inputs this basic transaction behavior connectivity array into the first deep relationship network unit. The deep relationship network unit mines the potential interaction relationship between transaction subjects by analyzing the transaction frequency, amount change and other data in the array. For example, which subjects may have frequent indirect transactions through other subjects although there are not many direct transactions between them. This process will generate a new derivative transaction behavior connectivity array, which contains more in-depth transaction relationship information.

[0039] After obtaining multiple connected arrays of derivative transaction behaviors, the server begins to embed these connected arrays of derivative transaction behaviors. This process can be understood as converting a complex graph structure into a format that is easier to manipulate mathematically. Each deep relational network unit maps the nodes and edges in the graph to points in the high-dimensional vector space according to the type of features it specializes in. These points not only retain the structural information of the original graph, but also reflect the deep-level characteristics of the transaction behavior. Finally, the server obtains a graph embedding representation vector sequence that can describe the key features of network transaction behavior in a compact and comprehensive way.

[0040] In a possible implementation, the step of extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links using each of the deep relationship network units to generate a derived transaction behavior connectivity array corresponding to each of the deep relationship network units includes: Step S121, using any of the deep relationship network units to compare the transaction behavior trajectory vectors of the two network transaction behavior subjects corresponding to any of the behavior relationship links, to generate the correlation between the two network transaction behavior subjects corresponding to any of the behavior relationship links.

[0041] In this embodiment, the server first selects a deep relationship network unit and a behavior relationship link. Assume that this behavior relationship link connects two network transaction behavior subjects A and B. The server obtains the transaction behavior trajectory vectors of A and B, which may include multiple features such as transaction time, transaction amount, and transaction frequency. The server then compares the transaction behavior trajectory vectors of A and B, for example, by calculating the overlap degree of their transaction time, the similarity of their transaction amounts, etc., to evaluate the transaction behavior similarity between A and B, thereby generating the correlation between A and B.

[0042] Step S122, summing up the correlations between two network transaction behavior subjects corresponding to a plurality of the behavior relationship links to generate a cumulative correlation, and taking the ratio of the correlation between two network transaction behavior subjects corresponding to any one of the behavior relationship links to the cumulative correlation as the mutual influence coefficient corresponding to any one of the behavior relationship links.

[0043] In this embodiment, the server continues to perform similar calculations on the correlation between the two network transaction behavior subjects corresponding to each behavior relationship link. After completing the correlation calculations for all behavior relationship links, the server accumulates these correlation values ​​to obtain a total cumulative correlation. Next, the server calculates the ratio of the correlation between the two network transaction behavior subjects corresponding to each behavior relationship link to the cumulative correlation. This ratio is called the mutual influence coefficient, which reflects the relative importance of the behavior relationship link in the entire network.

[0044] Step S123 , blending the link knowledge vector corresponding to any one of the behavior relationship links according to the mutual influence coefficient corresponding to any one of the behavior relationship links, to generate a blended link knowledge vector corresponding to any one of the behavior relationship links.

[0045] Next, the server will blend the link knowledge vector according to the mutual influence coefficient of each behavior relationship link. The link knowledge vector may contain information such as the transaction characteristics and behavior patterns of the link. The blending process can be understood as weighted mixing of this information according to the mutual influence coefficient, thereby generating a new link knowledge vector that integrates the overall influence of the network, namely the blended link knowledge vector.

[0046] Step S124: construct a sequence of intermingled link knowledge vectors corresponding to each of the deep relationship network units according to a sequence of intermingled link knowledge vectors corresponding to a plurality of the behavior relationship links.

[0047] After the server has completed the generation of the interlinked link knowledge vectors for all behavioral relationship links, it will organize these interlinked link knowledge vectors into a sequence in a certain order (such as time sequence, link importance order, etc.), and output them as an interlinked link knowledge vector sequence, which reflects the comprehensive characteristics of the transaction behavior in the entire deep relationship network.

[0048] Step S125 , extracting potential interactive relationships from the basic transaction behavior connectivity array according to the fusion link knowledge vector sequence corresponding to each of the deep relationship network units, and generating a derived transaction behavior connectivity array corresponding to each of the deep relationship network units.

[0049] Finally, the server will use this fusion link knowledge vector sequence to extract potential interaction relationships from the basic transaction behavior connectivity array. This process involves complex algorithms and models to identify and predict potential, not directly observed transaction behavior relationships in the network. Based on these extracted potential interaction relationships, the server finally generates a derivative transaction behavior connectivity array corresponding to each deep relationship network unit. The derived transaction behavior connectivity array not only contains the original transaction behavior relationship, but also incorporates the potential interaction relationship obtained through network analysis, thereby providing richer and deeper insights into network transaction behavior.

[0050] In a possible implementation, the step of extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links using each of the deep relationship network units to generate a derived transaction behavior connectivity array corresponding to each of the deep relationship network units includes: Step A110, utilizing the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of multiple deep relationship network units, performs progressive potential interaction relationship extraction on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit, and generates a connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit.

[0051] Among them, x increases successively and 2≤x≤M, M is the number of the multiple deep relationship network units functionally connected according to the feature pointing, and the derived transaction behavior connectivity array corresponding to the first deep relationship network unit is generated by extracting potential interactive relationships from the basic transaction behavior connectivity array based on the link knowledge vectors corresponding to the multiple behavior relationship links.

[0052] In this embodiment, the server will first process the first deep relationship network unit, which is special because it is the starting point of the entire progressive relationship. The server will extract potential interactive relationships from the basic transaction behavior connectivity array based on the link knowledge vectors corresponding to multiple behavior relationship links, which involves analyzing transaction behavior patterns, frequencies, amounts, etc., to identify those behavior subjects that are not directly connected but show potential associations in actual transactions.

[0053] After completing the processing of the first deep relationship network unit, the server will obtain the derived transaction behavior connectivity array corresponding to this deep relationship network unit. The derived transaction behavior connectivity array not only contains the direct behavior relationship, but also reveals the potential interaction relationship.

[0054] Next, the server processes the second deep relationship network unit. At this point, the server uses the inter-unit association logic between the first and second deep relationship network units, which may include specific algorithms, models, or rules, to further explore the potential connections between transaction behaviors.

[0055] Based on this association logic, the server will perform progressive potential interaction relationship extraction on the connected array of derivative transaction behaviors corresponding to the first deep relationship network unit, which means that the server will further identify and analyze the complex relationships between transaction behaviors on this basis, such as hidden patterns, periodic transactions, abnormal transactions, etc.

[0056] By progressively extracting potential interaction relationships, the server can identify more subtle and complex transaction behavior relationships. These newly discovered relationships will be integrated into the derived transaction behavior connectivity array corresponding to the second deep relationship network unit. The newly generated array not only inherits the information of the previous array, but also contains deeper interaction relationships.

[0057] The server will continue to process subsequent deep relationship network units according to the same logic. For each new unit, the server will use the output of the previous deep relationship network unit (i.e., the connected array of derived transaction behaviors) and the specific association logic of the current unit to perform progressive potential interaction relationship extraction.

[0058] When the server processes all M deep relationship network units, it will get a series of connected arrays of derivative transaction behaviors. Each array contains more and deeper potential interaction relationship information than the previous one. Ultimately, these arrays can provide strong data support for financial analysis, risk assessment, trading strategy formulation, etc.

[0059] Through this progressive processing method, the server can reveal the complex relationships between transaction behaviors layer by layer, providing financial institutions with more comprehensive and accurate transaction behavior analysis.

[0060] In a possible implementation, step A110 includes: Step A111, compare and distinguish the basic transaction behavior connectivity array with the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit to generate a first differentiated transaction behavior connectivity array, wherein the first differentiated transaction behavior connectivity array includes the behavior relationship links that are not extracted from the x-1th deep relationship network unit.

[0061] In this embodiment, the server first obtains the basic transaction behavior connectivity array and the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit. The basic transaction behavior connectivity array contains all the original transaction behavior data, while the derived transaction behavior connectivity array of the x-1th deep relationship network unit is the result after the previous round of potential interaction relationship extraction.

[0062] Next, the server compares the two arrays. The comparison process is mainly to find out the behavioral relationship links that are not extracted in the derivative transaction behavior connectivity array of the x-1th deep relationship network unit. These unextracted behavioral relationship links may be due to the fact that some complex or hidden relationships were not fully captured in the previous extraction process.

[0063] By comparison, the server generates a first distinguishing transaction behavior connectivity array, which specifically includes those behavioral relationship links that have not been extracted in the x-1th deep relationship network unit, providing a new data source for subsequent potential interaction relationship extraction.

[0064] Step A112: extracting potential interaction relationships from the first differentiated transaction behavior connected array according to the link knowledge vectors corresponding to the plurality of behavior relationship links, and generating a second differentiated transaction behavior connected array.

[0065] The server then uses the link knowledge vectors corresponding to the multiple behavior relationship links to extract potential interaction relationships from the first distinguished transaction behavior connectivity array. The link knowledge vector may include various feature information about the transaction behavior, such as transaction frequency, amount, historical behavior patterns of both parties to the transaction, etc.

[0066] During the extraction process, the server analyzes these link knowledge vectors to identify hidden, indirect or complex interaction relationships in the first differentiated transaction behavior connectivity array. These newly discovered potential interaction relationships are integrated into the second differentiated transaction behavior connectivity array, which represents the potential interaction relationships further extracted from the first differentiated transaction behavior connectivity array.

[0067] Step A113, merging the derived transaction behavior connected array corresponding to the x-1th deep relationship network unit with the second distinguished transaction behavior connected array to generate a derived transaction behavior connected array corresponding to the xth deep relationship network unit.

[0068] Finally, the server merges the derivative transaction behavior connected array corresponding to the x-1th deep relationship network unit with the second distinguished transaction behavior connected array. During the fusion process, the server will integrate the information in the two arrays to form a more comprehensive and richer derivative transaction behavior connected array.

[0069] This newly generated derivative transaction behavior connectivity array not only contains the interaction relationships extracted from the x-1th deep relationship network unit, but also incorporates the potential interaction relationships newly extracted from the first differentiated transaction behavior connectivity array. Therefore, it provides a deeper analysis of transaction behavior and provides stronger data support for subsequent financial decision-making, risk assessment, etc.

[0070] Through the above steps, the server can progressively reveal the complex relationships in transaction behaviors and continuously improve its understanding and analysis capabilities of the transaction network.

[0071] In a possible implementation, step A112 may include: Step A1121: blend the link knowledge vectors corresponding to the plurality of behavior relationship links to generate a blended link knowledge vector sequence.

[0072] Step A1122: extracting potential interaction relationships from the first distinct transaction behavior connected array according to the fusion link knowledge vector sequence to generate a second distinct transaction behavior connected array.

[0073] In this embodiment, the server first obtains link knowledge vectors corresponding to multiple behavior relationship links, each of which contains information such as transaction behavior characteristics, historical data, behavior patterns, etc. of a specific link. In order to comprehensively utilize this information, the server integrates the link knowledge vectors.

[0074] During the blending process, the server considers the uniqueness of each link knowledge vector while exploring the commonalities and complementarities between them. For example, some links may show significant characteristics in terms of transaction amount, while other links may be more prominent in terms of transaction frequency or transaction time. Through blending, the server is able to generate a new sequence of blended link knowledge vectors that incorporates the knowledge of multiple links, providing a more comprehensive data foundation for the subsequent extraction of potential interaction relationships.

[0075] After generating the intermingling link knowledge vector sequence, the server begins to use the intermingling link knowledge vector sequence to extract potential interaction relationships from the first differentiated transaction behavior connectivity array. The first differentiated transaction behavior connectivity array contains behavior relationship links that are not extracted in the x-1th deep relationship network unit, and these links may hide complex interaction relationships.

[0076] The server compares the fusion link knowledge vector sequence with the first distinguished transaction behavior connectivity array to find potential connections between them. For example, the server may discover some transaction behavior patterns that have not been noticed before, or identify indirect but important interactions between some transaction entities.

[0077] Through this process, the server is able to generate a second connected array of differentiated transaction behaviors, which not only contains the original information in the first connected array of differentiated transaction behaviors, but also adds potential interaction relationships extracted through the fusion link knowledge vector sequence. These newly discovered potential interaction relationships may provide financial institutions with deeper market insights and risk management strategies.

[0078] Therefore, by integrating link knowledge vectors and extracting potential interactive relationships, the server can more effectively reveal the complex relationships in the transaction network and provide more accurate data support for financial decision-making.

[0079] In a possible implementation, step A110 further includes: Step A114, extracting potential interactive relationships from the derived transaction behavior connected array corresponding to the x-1th deep relationship network unit for the link knowledge vectors corresponding to the relationship links, and generating a reference transaction behavior connected array.

[0080] In this embodiment, the server first obtains the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit. This derived transaction behavior connectivity array is the result obtained after the previous round of potential interaction relationship extraction, and includes transaction behavior relationships that have been identified and analyzed.

[0081] Next, the server will use the link knowledge vectors corresponding to multiple behavior relationship links to further extract potential interaction relationships from this derived transaction behavior connected array. The link knowledge vector contains rich transaction behavior features and patterns, which can help the server deeply explore the complex relationships that may be hidden in the array.

[0082] During the extraction process, the server analyzes the correlation between the link knowledge vector and the derived transaction behavior connectivity array to identify those behavioral entities that are not directly connected but show potential correlation in actual transactions, as well as the indirect transaction relationships between them. These newly discovered potential interaction relationships are integrated into a new array, namely the reference transaction behavior connectivity array, which supplements and improves the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit, and provides a reference for subsequent distinction and comparison.

[0083] Step A115, compare and contrast the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit with the reference transaction behavior connectivity array, and use the generated third differentiated transaction behavior connectivity array as the derived transaction behavior connectivity array corresponding to the xth deep relationship network unit.

[0084] In this embodiment, after generating the reference transaction behavior connectivity array, the server will compare it with the derivative transaction behavior connectivity array corresponding to the x-1th deep relationship network unit. The purpose of the comparison is to find the differences between the two, that is, those behavior relationship links that appear in the reference transaction behavior connectivity array but not in the derivative transaction behavior connectivity array of the x-1th deep relationship network unit.

[0085] By comparison, the server generates a third differentiated transaction behavior connectivity array, which specifically includes those behavioral relationship links that were not extracted in the derived transaction behavior connectivity array of the x-1th deep relationship network unit but were newly discovered in the reference transaction behavior connectivity array. These behavioral relationship links represent deeper potential interaction relationships and may be complex transaction patterns that were previously ignored or not fully captured.

[0086] Finally, the server uses the third difference transaction behavior connectivity array as the derivative transaction behavior connectivity array corresponding to the xth deep relationship network unit. The derivative transaction behavior connectivity array not only inherits the information of the previous deep relationship network unit, but also incorporates the newly discovered behavior relationship links extracted through potential interactive relationships. Therefore, it provides a deeper transaction behavior analysis and provides more accurate data support for subsequent financial decision-making, risk assessment, etc.

[0087] Through the above steps, the server can gradually reveal the complex relationships in transaction behaviors and continuously improve the understanding and analysis capabilities of the transaction network. At the same time, this progressive potential interaction relationship extraction method also helps to discover hidden transaction behavior patterns that are difficult to detect in traditional analysis methods.

[0088] In a possible implementation, the embedding representation of the sample network transaction behavior graph according to the feature-directed functional connections based on the derived transaction behavior connectivity array corresponding to the multiple deep relationship network units functionally connected according to the feature-directed functional connections is generated to generate a graph embedding representation vector sequence of the sample network transaction behavior graph, including: Step S131, using the first deep relationship network unit of the plurality of deep relationship network units functionally connected according to feature pointing in combination with the connected array of derived transaction behaviors corresponding to the first deep relationship network unit, to embed the sample network transaction behavior graph.

[0089] In this embodiment, the server first obtains multiple deep relationship network units connected by feature-oriented functionalities, and selects the first deep relationship network unit, which is the starting point of the embedding representation process. At the same time, the server also obtains the derived transaction behavior connectivity array corresponding to the first deep relationship network unit, which contains the specific transaction behavior patterns and relationships of the deep relationship network unit.

[0090] Next, the server uses the first deep relational network unit and its corresponding derivative transaction behavior connected array to embed the sample network transaction behavior graph. In this process, the server matches the nodes and edges in the graph with the elements in the deep relational network unit, and maps the structure and information of the graph to the deep relational network unit. In this way, the server can capture the features and functional connections in the graph and convert them into embedded representations in numerical form.

[0091] Step S132, loading the embedding representation result of the first deep relationship network unit into the backward connected deep relationship network unit, so that the backward connected deep relationship network unit continues to perform embedding representation and output until it is loaded into the last deep relationship network unit.

[0092] After completing the embedding representation of the first deep relational network unit, the server will load this embedding representation result into the backward connected deep relational network unit. This process is progressive. Each backward connected deep relational network unit will receive the embedding representation result of the previous deep relational network unit and continue to embed the representation on this basis.

[0093] In each backward-connected deep relational network unit, the server will further process and analyze the input embedding representation in combination with the unit's derived transaction behavior connected array. During this process, the server will deeply explore the potential relationships and patterns in the sample network transaction behavior graph, and continuously update and improve the embedding representation.

[0094] When the embedding representation process progresses to the last deep relational network unit, the server will output the embedding representation result of this deep relational network unit. The embedding representation result is a comprehensive and in-depth representation of the sample network transaction behavior graph, capturing all the features and functional connections in the sample network transaction behavior graph.

[0095] Step S133, using the embedding representation result generated by the last deep relationship network unit as the graph embedding representation vector sequence of the sample network transaction behavior graph.

[0096] Finally, the server uses this embedding representation result as the graph embedding representation vector sequence of the sample network transaction behavior graph. The graph embedding representation vector sequence is a numerical representation of the graph structure and information, which can be used for subsequent data analysis, pattern recognition and other tasks.

[0097] Through the above steps, the server can embed the sample network transaction behavior graph in a way that features point to functional connections, generating a graph embedding representation vector sequence that is rich in information and highly representative. This process helps the server to have a deeper understanding of transaction behavior patterns and network structures, and provide strong support for decision-making and analysis in the financial field.

[0098] In a possible implementation, step S132 includes: using the yth deep relationship network unit of the plurality of deep relationship network units, performing dot product operations on the connected array of derivative transaction behaviors corresponding to the yth deep relationship network unit, the embedded representation result of the y-1th deep relationship network unit, and the trainable weight array of the yth deep relationship network unit, generating the embedded representation result of the yth deep relationship network unit, and loading the embedded representation result of the yth deep relationship network unit into the y+1th deep relationship network unit. Wherein, y increases successively and 2≤y≤M-1, and M is the number of the plurality of deep relationship network units functionally connected according to feature pointing.

[0099] In this embodiment, the server starts to process the continuous embedding representation process of multiple deep relationship network units. The process is progressive layer by layer, starting from the first deep relationship network unit and ending with the last deep relationship network unit.

[0100] The server first ensures that all deep relational network units are initialized correctly, and each deep relational network unit has its corresponding derived transaction behavior connectivity array and trainable weight array, which are obtained through training and learning in the previous steps and contain rich transaction behavior features and pattern information.

[0101] In the previous step, the server has embedded the sample network transaction behavior graph using the first deep relational network unit and its corresponding connected array of derivative transaction behaviors, and obtained the embedding representation result of the first deep relational network unit.

[0102] Now, the server starts to load the embedding representation result of the first deep relational network unit into the backward connected deep relational network unit. This process is progressive, starting from the second deep relational network unit (ie y=2) to the M-1th deep relational network unit.

[0103] For the y-th deep relational network unit (2≤y≤M-1), the server performs the following operations: a. Obtain necessary data: The server obtains the connected array of derivative transaction behaviors corresponding to the y-th deep relational network unit, the embedding representation result of the y-1-th deep relational network unit, and the trainable weight array of the y-th deep relational network unit.

[0104] b. Dot product operation processing: The server performs dot product operation on the three data obtained above. This operation can comprehensively consider the behavior pattern in the connected array of derived transaction behaviors, the feature information in the embedding representation result of the previous deep relationship network unit, and the weight distribution in the trainable weight array of the current unit, thereby generating the embedding representation result of the yth deep relationship network unit.

[0105] c. Loading to the next deep relational network unit: After completing the dot product operation, the server loads the embedding representation result of the yth deep relational network unit into the y+1th deep relational network unit, so that the next deep relational network unit can continue to perform embedding representation on this basis.

[0106] The server repeats the above progressive embedding representation process until it loads the last deep relational network unit (i.e., the Mth deep relational network unit). In this process, each deep relational network unit performs more in-depth feature extraction and representation learning based on the previous deep relational network unit.

[0107] Through this progressive embedding representation method, the server can gradually and deeply explore the potential features and functional connections in the sample network transaction behavior graph, and generate a highly representative graph embedding representation vector sequence, which is of great significance for subsequent financial data analysis, risk prediction and other tasks.

[0108] In a possible implementation, step S140 includes: Step S141, using the network transaction risk warning model to estimate the network transaction risk label of the graph embedding representation vector sequence of the sample network transaction behavior graph, to generate an estimated network transaction risk label of the sample network transaction behavior graph.

[0109] Step S142, calculating the training cost of the network transaction risk warning model based on the estimated network transaction risk label of the sample network transaction behavior graph and the training supervision data of the sample network transaction behavior graph.

[0110] Step S143, optimizing the network transaction risk warning model based on the training cost value until the training cost value no longer continues to decrease, and outputting the network transaction risk warning model as a target network transaction risk warning model.

[0111] In this embodiment, the server embeds these graphs into the representation vector sequence and inputs it into the network transaction risk warning model. The network transaction risk warning model is specially used to predict network transaction risks and can estimate the possible risks of transaction behaviors by analyzing the input data.

[0112] The server uses the prediction function of the network transaction risk warning model to estimate the network transaction risk label of the graph embedding representation vector sequence. This process is actually the network transaction risk warning model judging the possible corresponding network transaction risk category or level based on the characteristics of the input data.

[0113] Finally, the server generates estimated network transaction risk labels for the sample network transaction behavior graphs. These estimated network transaction risk labels are prediction results of the possible risks of each transaction behavior graph.

[0114] After obtaining the estimated network transaction risk labels, the server needs to evaluate the prediction accuracy of the network transaction risk warning model. To this end, the server obtains training supervision data of sample network transaction behavior graphs, which are known and verified real network transaction risk labels.

[0115] The server compares the estimated network transaction risk label with the training supervision data and calculates the difference between the two through a specific loss function (such as the cross entropy loss function). The difference value reflects the degree of inconsistency between the model prediction result and the actual risk label, which is called the training cost.

[0116] The size of the training cost value directly reflects the accuracy of the prediction of the network transaction risk warning model: the smaller the cost value, the closer the prediction result of the network transaction risk warning model is to the actual situation, that is, the better the performance of the network transaction risk warning model.

[0117] Based on the calculated training cost value, the server starts to optimize the parameters of the network transaction risk warning model. The goal of the optimization is to reduce the training cost value and improve the prediction accuracy of the network transaction risk warning model.

[0118] The server uses optimization algorithms such as gradient descent to adjust the parameters of the model according to the training cost value. The process is iterative. The server will perform forward propagation (calculating the predicted value and cost value) and back propagation (adjusting the model parameters according to the cost value) operations multiple times until the training cost value no longer continues to decrease. When the training cost value reaches a relatively stable and low level, the server believes that the network transaction risk warning model has been fully optimized. At this time, the server outputs the optimized network transaction risk warning model as the target network transaction risk warning model.

[0119] Through the above steps, the server successfully optimized the model parameters of the network transaction risk warning model based on the graph embedding representation vector sequence of the sample network transaction behavior graph, and obtained a target network transaction risk warning model with better performance and more accurate prediction.

[0120] Figure 2 The hardware structure of the risk warning system 100 applied to network transactions for implementing the risk warning method applied to network transactions provided in the embodiment of the present application is shown. Figure 2 As shown, the risk warning system 100 applied to network transactions may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0121] In a possible design, the risk warning system 100 applied to network transactions can be a single server or a server group. The server group can be centralized or distributed (for example, the risk warning system 100 applied to network transactions can be a distributed system). In some embodiments, the risk warning system 100 applied to network transactions can be local or remote. For example, the risk warning system 100 applied to network transactions can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the risk warning system 100 applied to network transactions can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the risk warning system 100 applied to network transactions can be implemented on the risk warning system applied to network transactions. As an example only, the risk warning system applied to network transactions can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0122] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the risk warning system 100 for network transactions uses to execute or use to complete the exemplary method described in this application.

[0123] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the risk warning method applied to network transactions in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0124] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the risk warning system 100 applied to network transactions mentioned above. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0125] In addition, an embodiment of the present application further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the risk warning method applied to network transactions as above is implemented.

[0126] It should be noted that in order to simplify the description disclosed in the present application and thereby facilitate the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, various features are sometimes grouped into one embodiment, drawing or description thereof.

Claims

1. A risk warning method for network transactions, characterized in that: The method comprises: Obtain a connected array of basic transaction behaviors corresponding to multiple behavior relationship links in the sample network transaction behavior graph; Extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array; Embedding the sample network transaction behavior graph according to the derived transaction behavior connectivity array to generate a graph embedding representation vector sequence of the sample network transaction behavior graph; The model parameters of the network transaction risk warning model are optimized according to the graph embedding representation vector sequence of the sample network transaction behavior graph, and the network transaction risk label of the target network transaction behavior graph of any user is estimated based on the network transaction risk warning model after the model parameter optimization is completed. After the corresponding network transaction risk label is generated, the risk warning information is pushed to the user.

2. The risk early warning method for network transactions according to claim 1 is characterized in that: The network transaction risk early warning model includes a plurality of deep relationship network units connected functionally according to feature orientation; The extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links to generate a derived transaction behavior connectivity array includes: Using each of the deep relationship network units to extract potential interactive relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links, and generating a derived transaction behavior connectivity array corresponding to each of the deep relationship network units; The embedding representation of the sample network transaction behavior graph according to the derived transaction behavior connectivity array to generate a graph embedding representation vector sequence of the sample network transaction behavior graph includes: The sample network transaction behavior graph is embedded according to feature-pointing functional connections based on the derived transaction behavior connected array corresponding to the multiple deep relationship network units that are functionally connected according to feature pointing, and a graph embedding representation vector sequence of the sample network transaction behavior graph is generated.

3. The risk warning method for network transactions according to claim 2 is characterized in that: The step of extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links using each of the deep relationship network units to generate a derived transaction behavior connectivity array corresponding to each of the deep relationship network units includes: Using any of the deep relationship network units to compare the characteristics of the transaction behavior trajectory vectors of the two network transaction behavior subjects corresponding to any of the behavior relationship links, and generating the correlation between the two network transaction behavior subjects corresponding to any of the behavior relationship links; Accumulating the correlations between two network transaction behavior subjects corresponding to a plurality of the behavior relationship links to generate a cumulative correlation, and taking the ratio of the correlation between two network transaction behavior subjects corresponding to any one of the behavior relationship links to the cumulative correlation as the mutual influence coefficient corresponding to any one of the behavior relationship links; According to the mutual influence coefficient corresponding to any one of the behavior relationship links, the link knowledge vector corresponding to any one of the behavior relationship links is blended to generate a blended link knowledge vector corresponding to any one of the behavior relationship links; According to a sequence of intermingling link knowledge vectors corresponding to a plurality of said behavior relationship links, a sequence of intermingling link knowledge vectors corresponding to each of said deep relationship network units is formed; The potential interaction relationship of the basic transaction behavior connectivity array is extracted according to the fusion link knowledge vector sequence corresponding to each of the deep relationship network units, and a derivative transaction behavior connectivity array corresponding to each of the deep relationship network units is generated.

4. The risk warning method for network transactions according to claim 2 is characterized in that: The step of extracting potential interaction relationships from the basic transaction behavior connectivity arrays corresponding to the multiple behavior relationship links using each of the deep relationship network units to generate a derived transaction behavior connectivity array corresponding to each of the deep relationship network units includes: Using the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of the plurality of deep relationship network units, a progressive potential interaction relationship extraction is performed on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit to generate a connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit; Among them, x increases successively and 2≤x≤M, M is the number of the multiple deep relationship network units functionally connected according to the feature pointing, and the derived transaction behavior connectivity array corresponding to the first deep relationship network unit is generated by extracting potential interactive relationships from the basic transaction behavior connectivity array based on the link knowledge vectors corresponding to the multiple behavior relationship links.

5. The risk warning method for network transactions according to claim 4 is characterized in that: The method utilizes the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of the plurality of deep relationship network units to perform progressive potential interaction relationship extraction on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit to generate the connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit, including: The basic transaction behavior connectivity array is distinguished and compared with the derived transaction behavior connectivity array corresponding to the x-1th deep relationship network unit to generate a first distinguished transaction behavior connectivity array, wherein the first distinguished transaction behavior connectivity array includes the behavior relationship links that are not extracted from the x-1th deep relationship network unit; Extracting potential interaction relationships from the first distinguishing transaction behavior connected array according to the link knowledge vectors respectively corresponding to the plurality of behavior relationship links, and generating a second distinguishing transaction behavior connected array; The derivative transaction behavior connected array corresponding to the x-1th deep relationship network unit is merged with the second distinguished transaction behavior connected array to generate a derivative transaction behavior connected array corresponding to the xth deep relationship network unit.

6. The risk early warning method for network transactions according to claim 5 is characterized in that: The step of extracting potential interaction relationships from the first distinguishing transaction behavior connectivity array based on the link knowledge vectors respectively corresponding to the plurality of behavior relationship links to generate a second distinguishing transaction behavior connectivity array includes: blending the link knowledge vectors respectively corresponding to the plurality of behavior relationship links to generate a blended link knowledge vector sequence; The potential interaction relationship of the first differentiated transaction behavior connected array is extracted according to the fusion link knowledge vector sequence to generate a second differentiated transaction behavior connected array.

7. The risk warning method for network transactions according to claim 4 is characterized in that: The method utilizes the inter-unit association logic between the x-1th deep relationship network unit and the xth deep relationship network unit of the plurality of deep relationship network units to perform progressive potential interaction relationship extraction on the connected array of derivative transaction behaviors corresponding to the x-1th deep relationship network unit to generate the connected array of derivative transaction behaviors corresponding to the xth deep relationship network unit, including: Extracting potential interaction relationships from the derivative transaction behavior connectivity array corresponding to the x-1th deep relationship network unit according to the link knowledge vectors corresponding to the multiple behavior relationship links, and generating a reference transaction behavior connectivity array; The derivative transaction behavior connectivity array corresponding to the x-1th deep relationship network unit is distinguished and compared with the reference transaction behavior connectivity array, and the generated third distinguished transaction behavior connectivity array is used as the derivative transaction behavior connectivity array corresponding to the xth deep relationship network unit.

8. The risk early warning method for network transactions according to claim 2 is characterized in that: The method of embedding the sample network transaction behavior graph according to the feature-pointing functional connections based on the derived transaction behavior connected array corresponding to the multiple deep relationship network units according to the feature-pointing functional connections to generate a graph embedding representation vector sequence of the sample network transaction behavior graph includes: Embedding the sample network transaction behavior graph by using a first deep relationship network unit of the plurality of deep relationship network units functionally connected according to feature pointing and a connected array of derived transaction behaviors corresponding to the first deep relationship network unit; Loading the embedding representation result of the first deep relational network unit into the backward connected deep relational network unit, so that the backward connected deep relational network unit continues to perform embedding representation and output until it is loaded into the last deep relational network unit; Using the embedding representation result generated by the last deep relationship network unit as the graph embedding representation vector sequence of the sample network transaction behavior graph; The step of loading the embedding representation result of the first deep relational network unit into the backward connected deep relational network unit so that the backward connected deep relational network unit continues to output the embedded representation until the last deep relational network unit is loaded includes: Utilizing the yth deep relationship network unit of multiple deep relationship network units, dot product operations are performed on the connected array of derivative transaction behaviors corresponding to the yth deep relationship network unit, the embedded representation result of the y-1th deep relationship network unit, and the trainable weight array of the yth deep relationship network unit to generate the embedded representation result of the yth deep relationship network unit, and the embedded representation result of the yth deep relationship network unit is loaded into the y+1th deep relationship network unit; wherein y increases successively and 2≤y≤M-1, and M is the number of the multiple deep relationship network units functionally connected according to feature pointing.

9. The risk warning method for network transactions according to claim 1, characterized in that: The optimizing of model parameters of the network transaction risk warning model based on the graph embedding representation vector sequence of the sample network transaction behavior graph includes: Using the network transaction risk early warning model, estimating the network transaction risk label of the graph embedding representation vector sequence of the sample network transaction behavior graph, and generating an estimated network transaction risk label of the sample network transaction behavior graph; Calculating the training cost of the network transaction risk warning model based on the estimated network transaction risk label of the sample network transaction behavior graph and the training supervision data of the sample network transaction behavior graph; The network transaction risk warning model is optimized based on the training cost value until the training cost value no longer continues to decrease, and the network transaction risk warning model is output as a target network transaction risk warning model.

10. A risk warning system applied to network transactions, characterized in that: The risk warning system applied to network transactions includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the risk warning method applied to network transactions as described in any one of claims 1 to 9 above.

Images