Financial anti-fraud intelligent evaluation method fusing rule engine and graph neural network

Abstract

The invention provides a financial anti-fraud intelligent evaluation method fusing a rule engine and a graph neural network. The method comprises the following steps: extracting multiple pieces of financial service feature information from current financial service request information; performing rule matching on the multiple pieces of financial service feature information based on a rule engine, and outputting a current financial service rule; constructing a graph neural network model based on the personal information of the current financial business requester and the associated information, and performing convolution calculation to obtain current network structure feature information; and based on the current network structure feature information and the current financial service rule, performing evaluation through a preset evaluation algorithm to obtain a fraud risk value. Through the fusion rule engine and the graph neural network, whether financial service request information is fraud or not can be accurately and intelligently evaluated, anti-fraud measures are taken for fraudulent financial services, and heavy losses caused by financial fraud to financial institutions are effectively prevented.

Description

Technical Field

[0001] The present invention relates to the field of financial anti-fraud technology, and in particular to a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network. Background Art

[0002] Credit risk, also known as default risk, refers to the possibility that a borrower, securities issuer, or counterparty, for various reasons, will be unwilling or unable to fulfill the terms of a contract, resulting in default and causing losses to the bank, investors, or the counterparty. Credit risk is a major risk faced by banks. If banks fail to promptly identify lost assets, increase provisions for bad debt write-offs, and, where appropriate, cease interest income recognition, they face serious risks.

[0003] User fraud refers to the fabrication of false reasons such as the introduction of funds and projects for the purpose of illegal possession, the use of false economic contracts, certification documents, or false property rights certificates as collateral, as well as other methods of fabricating facts to conceal the truth, in order to defraud banks or other financial institutions of loans.

[0004] As the credit market in developing countries rapidly expands, the prospects for banking development are also evolving. Preventing personal credit fraud and reducing the rate of non-performing loans have become crucial research topics for commercial banks. Despite the significant growth in the domestic credit market, the quality of services provided has not significantly improved. In particular, commercial banks face various challenges, such as overdue loans and non-performing loans, which have resulted in significant losses. Summary of the Invention

[0005] In order to solve at least one of the above technical problems, the present invention proposes a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network. By integrating the rule engine and the graph neural network, it can make an accurate and intelligent assessment of whether the financial business request information is fraudulent or not, and implement anti-fraud measures for fraudulent financial business, effectively preventing financial fraud from causing significant losses to financial institutions.

[0006] The present invention proposes a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network. The method includes: Get current financial business request information; Extracting multiple financial service feature information from the current financial service request information; Based on the rule engine, it matches multiple financial business feature information according to the preset matching algorithm and outputs the corresponding current financial business rules; Extracting personal information of the current financial service requester from the current financial service request information; Based on the personal information of the current financial service requester, and combined with big data technology to obtain related information; Build a graph neural network model based on the personal information and related information of the current financial service requester; Perform convolution calculations based on the graph neural network model to obtain the current network structure feature information; Based on the current network structure feature information and the current financial business rules, an evaluation is performed through a preset evaluation algorithm to obtain a fraud risk value.

[0007] Furthermore, after obtaining the fraud risk value, the method further includes: Determining whether the fraud risk value is greater than a first preset threshold; If so, the current financial service request information is identified as fraudulent information and anti-fraud response measures are initiated.

[0008] Furthermore, based on the current network structure feature information and the current financial business rules, a fraud risk value is obtained by performing an evaluation using a preset evaluation algorithm, specifically including: Build anti-fraud assessment models; Train the anti-fraud assessment model using sample data; Inputting the current network structure feature information and the current financial business rules into an anti-fraud evaluation model; The output is the fraud risk value.

[0009] Furthermore, after outputting the fraud risk value, the method further includes: Obtain multiple historical financial business data; Filtering historical financial business data with similar characteristics to the current financial business request information from multiple historical financial business data and saving them to a similar characteristics database; Based on the historical financial business data in the similar feature library, the correction value is calculated according to the correction value calculation method; The correction value is added to the output fraud risk value to obtain the corrected fraud risk value.

[0010] Furthermore, historical financial business data with similar characteristics to the current financial business request information is screened out from multiple historical financial business data and saved in a similar characteristics database, specifically including: It is preset that each historical financial business data at least includes historical financial business request information and historical actual fraud situations; Performing feature calculation on historical financial service request information in each historical financial service data to obtain feature values ​​of the historical financial service request information; Perform feature calculation on the current financial service request information to obtain a feature value of the current financial service request information; Based on each historical financial business data, calculating the difference between the characteristic value of the historical financial business request information of the historical financial business data and the characteristic value of the current financial business request information to obtain the difference between the two; It is determined whether the difference is less than a second preset threshold value. If so, the corresponding historical financial business data is saved in a similar feature library.

[0011] Furthermore, based on the historical financial business data in the similar feature library and according to the correction value calculation method, a correction value is calculated, specifically including: Based on each historical financial business data in the similar feature library, the corresponding historical financial business rules and historical network structure feature information are calculated through the preset matching algorithm and graph neural network convolution algorithm respectively; Based on the historical network structure feature information and historical current financial business rules, and through a preset evaluation algorithm, a historical fraud risk prediction value is obtained; Based on each historical financial business data in the similar feature library, according to the historical fraud risk prediction value of the historical financial business data and the historical actual fraud situation, a correction component value of each historical financial business data is calculated by a first algorithm; The correction component values ​​calculated for each historical financial business data in the similar feature library are added together to obtain the sum of the correction component values, and the sum of the correction component values ​​is divided by the amount of historical financial business data in the similar feature library to obtain the correction value.

[0012] Furthermore, the corresponding historical financial business rules and historical network structure feature information are calculated through the preset matching algorithm and graph neural network convolution algorithm, including: Extracting a plurality of historical financial business feature information from the historical financial business request information of the historical financial business data; Based on the rule engine, it matches multiple historical financial business feature information according to the preset matching algorithm and outputs the corresponding historical financial business rules; Extracting personal information of historical financial business requesters from historical financial business request information; Based on the personal information of historical financial business requesters, and combined with big data technology to obtain related information; Construct a graph neural network model of historical financial business requesters based on their personal information and related information; Based on the graph neural network model of historical financial business requesters, convolution calculation is performed to obtain historical network structure feature information.

[0013] Furthermore, feature calculation is performed on the historical financial service request information in each historical financial service data to obtain feature values ​​of the historical financial service request information, specifically including: Obtain the age, credit score, historical financial purpose, and historical financial transaction amount of each historical financial transaction request information of the historical financial transaction requester; Use big data to calculate the average credit value of different ages at historical time points; According to the age of the historical financial service requester of each historical financial service request information, obtaining the average credit value corresponding to the age of the historical financial service requester; The attribute characteristic value of the historical financial service requester is calculated based on the credit value of the historical financial service requester and the average credit value corresponding to the age of the historical financial service requester, and the attribute characteristic value of the historical financial service requester = the credit value of the historical financial service requester / the average credit value corresponding to the age of the historical financial service requester; Use big data to calculate the average amount of various financial uses at historical points in time; According to the historical financial purpose of each historical financial business request information, obtain the average amount corresponding to the historical financial purpose; The historical financial business content characteristic value is calculated based on the historical financial business amount and the average amount corresponding to the historical financial purpose, and the historical financial business content characteristic value = historical financial business amount / average amount corresponding to the historical financial purpose.

[0014] Furthermore, feature calculation is performed on the current financial service request information to obtain feature values ​​of the current financial service request information, specifically including: Obtain the current financial service request information, including the age and credit score of the current financial service requester, as well as the current financial purpose and the current financial service amount; Use big data to calculate the average credit value of different ages at the current time point; According to the age of the current financial service requester in the current financial service request information, obtaining the average credit value corresponding to the age of the current financial service requester; The attribute characteristic value of the current financial service requester is calculated based on the credit value of the current financial service requester and the average credit value corresponding to the age of the current financial service requester, and the attribute characteristic value of the current financial service requester = the credit value of the current financial service requester / the average credit value corresponding to the age of the current financial service requester; Use big data to calculate the average amount of each financial purpose at the current time point; According to the current financial purpose of the current financial service request information, obtain the average amount corresponding to the current financial purpose; The current financial business content characteristic value is calculated based on the current financial business amount and the average amount corresponding to the current financial purpose, and the current financial business content characteristic value = the current financial business amount / the average amount corresponding to the current financial purpose.

[0015] Furthermore, based on each historical financial business data, a difference is calculated between the characteristic value of the historical financial business request information of the historical financial business data and the characteristic value of the current financial business request information to obtain the difference between the two, specifically including: The historical financial business requester attribute characteristic values ​​and historical financial business content characteristic values ​​of the historical financial business request information of the historical financial business data and the current financial business requester attribute characteristic values ​​and current financial business content characteristic values ​​of the current financial business request information are calculated according to the second algorithm to obtain the difference between the two.

[0016] The present invention proposes a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network. Through the rule engine, multiple financial business feature information is matched according to a preset matching algorithm to output the corresponding current financial business rules. The personal information of the current financial business requester is then extracted from the current financial business request information. Related information is obtained by combining big data technology. A graph neural network model is constructed based on the personal information and related information of the current financial business requester. Convolution calculation is performed based on the graph neural network model to obtain current network structure feature information. Based on the current network structure feature information and the current financial business rules, an assessment is performed using a preset assessment algorithm to finally obtain a fraud risk value. This facilitates judging the fraudulent status of the current financial business based on the fraud risk value, so as to implement anti-fraud measures for financial business requests that are judged to be fraudulent. By integrating the rule engine and the graph neural network, the present invention can make an accurate and intelligent assessment of whether the financial business request information is fraudulent or not, and implement anti-fraud measures for fraudulent financial business, effectively preventing the significant losses caused by financial fraud to financial institutions.

[0017] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a financial anti-fraud intelligent assessment method integrating a rule engine and a graph neural network is shown in the present invention; Figure 2 A flow chart showing the initiation of anti-fraud countermeasures according to the present invention is shown; Figure 3 A flow chart showing a fraud risk value assessment method according to the present invention is shown; Figure 4The flowchart of the calculation method of the current financial business rules of the present invention is shown. DETAILED DESCRIPTION

[0019] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0020] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0021] Figure 1 A flowchart of a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network is shown in the present invention.

[0022] like Figure 1 As shown, the present invention proposes a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network, the method comprising: S102, obtaining current financial service request information; S104, extracting multiple financial service feature information (such as amount, credit purpose, etc.) from the current financial service request information; S106, based on the rule engine and according to the preset matching algorithm, rule matching is performed on the plurality of financial service feature information, and the corresponding current financial service rules are output; S108, extracting personal information of the current financial service requester from the current financial service request information; S110, based on the personal information of the current financial service requester, and in combination with big data technology, obtain related information; S112, constructing a graph neural network model based on the personal information and related information of the current financial service requester; S114, performing convolution calculation based on the graph neural network model to obtain current network structure feature information; S116, based on the current network structure feature information and the current financial business rules, and through a preset evaluation algorithm, an evaluation is performed to obtain a fraud risk value.

[0023] It can be understood that the financial business includes credit business, loan business, mortgage business, trust business, insurance business, etc.

[0024] It can be understood that the related information can be the social information of the financial service requester, or the requester's previous credit business (credit institution, credit repayment status), transaction business (transaction object, transaction amount, transaction success status), etc.

[0025] As you can understand, a rule engine focuses on rule judgment and is a component embedded in an application. It separates business decisions from application code and uses predefined semantic modules to write business decisions. It accepts data input, interprets business rules, and makes business decisions based on the business rules.

[0026] As can be understood, graph neural networks are neural networks specialized for processing graph-structured data. They combine the advantages of graph computing and neural networks to abstract graph structural information into node features. The message propagation model is a popular processing method in graph neural networks. It includes two steps: neighbor aggregation and node update, which can obtain high-order neighbor information of a node. Common graph neural network models include: graph convolutional neural networks (GCN), graph attention networks (GAT), recurrent graph neural networks (GGNN), and autoencoder-based graph neural networks (SDNE). The present invention preferably uses graph convolutional neural networks (GCN), but is not limited to this.

[0027] The main principle of a graph convolutional neural network (GCN) is to learn the feature information of the neighboring nodes of the current node and update the feature information of the current node in combination with the features of the current node. Specifically, a graph convolutional neural network is a deep learning method that uses convolutional layers to learn and analyze graph-structured data to extract graph data features. Its core idea is to define each piece of data as a graph, then collect the neighbor information of each node in the graph and pass the neighbor information back to the current node. Finally, the neighbor information is collected and combined with the node's own features to update the node's own feature information. For a single-layer network, it can obtain the direct neighbor information of the current node. For a multi-layer network, the update of the feature matrix is ​​obtained by merging high-order domain information.

[0028] The present invention uses a rule engine to match multiple financial business feature information according to a preset matching algorithm and output the corresponding current financial business rules; then extract the personal information of the current financial business requester from the current financial business request information; and obtain related information in combination with big data technology; construct a graph neural network model based on the personal information and related information of the current financial business requester; perform convolution calculation based on the graph neural network model to obtain the current network structure feature information; based on the current network structure feature information and the current financial business rules, and through a preset evaluation algorithm, perform evaluation to finally obtain a fraud risk value. This facilitates judging the fraud situation of the current financial business based on the fraud risk value, so as to implement anti-fraud measures for financial business requests that are judged to be fraudulent. The present invention can make an accurate and intelligent evaluation of whether the financial business request information is fraudulent or not by integrating the rule engine and the graph neural network, and implement anti-fraud measures for fraudulent financial business, effectively preventing the significant losses caused by financial fraud to financial institutions.

[0029] According to a specific embodiment of the present invention, a graph neural network model is constructed based on the personal information and related information of the current financial service requester, specifically including: Data graphing: processing data into a graph data structure according to structural relationships and requirements, and further abstracting it into a data graph to obtain a graph matrix representing node relationships; Message passing: First, graph data is transmitted as input data to the graph neural network; then, node information is passed and aggregated, and node feature representation combined with graph structure information is obtained by continuously updating node information; Output graph information. After processing the input information, the graph neural network outputs different structural data according to the requirements of the training task.

[0030] According to a specific embodiment of the present invention, different structural data are output according to different training task requirements, specifically including: Construct a financial network graph of financial service requesters and learn requester features containing the financial service relationships of financial service requesters based on GAT; Secondly, we construct a labeled interaction bipartite graph that can express the positive and negative financial preferences of financial service requesters for points of interest. We use SB-GNN to learn the requester features and points of interest features that contain the positive and negative interaction preferences of financial service requesters. Construct a directed transition graph of points of interest and use SR-GNN to learn the points of interest features that contain the points of interest transfer preferences of financial business requesters; Then, the requester features with financial business relationships are fused with the requester features with positive and negative interaction preferences of interest points to obtain the final requester feature representation; The point of interest features containing the positive and negative interaction preferences of the financial service requester's points of interest are fused with the point of interest features containing the transfer preferences of the financial service requester's points of interest to obtain the final point of interest feature representation; Finally, the final requester feature representation is multiplied by the point of interest feature representation to calculate the predicted score of the financial service requester for each point of interest; Select the structural data with prediction scores greater than a certain threshold for output.

[0031] The present invention utilizes the personalized preferences of financial service requesters to explore new points of interest in the global spatiotemporal preference neighborhood, and then obtains information that is more in line with the requester's intention based on the points of interest, thereby facilitating subsequent assessment of fraud risk values.

[0032] As you can understand, GAT, or Graph Attention Neural Network, is a specialized neural network designed specifically for processing graph-structured data. Its core working principle is to calculate relationships between nodes using an attention mechanism. In traditional neural networks, each node's state is updated independently. In GAT, however, each node's state update takes into account the states of its neighbors. GAT calculates the attention weights between a node and its neighbors and then updates the node's state based on these weights. This weighted information update allows GAT to better capture structural information in the graph.

[0033] SR-GNN transforms serialized problems into graph problems, models all conversation sequences using directed graphs, and then uses graph neural networks to learn the latent vector representation of each item. Furthermore, it uses an attention network architecture model to capture users' short-term interests, achieving a vector representation that captures both long-term and short-term interests.

[0034] like Figure 2 As shown, after obtaining the fraud risk value, the method further includes: S202, determining whether the fraud risk value is greater than a first preset threshold; S204: If yes, the current financial service request information is identified as fraudulent information and anti-fraud measures are initiated; S206: If not, the current financial service request information is determined to be non-fraud information, and no anti-fraud countermeasures are initiated.

[0035] It can be understood that the first preset threshold of the present invention can be the minimum risk value for determining fraudulent financial transactions. When it is lower than the threshold, it is determined to be a fraudulent financial transaction. Otherwise, it is not determined to be a fraudulent financial transaction.

[0036] It can be understood that the first preset threshold can be calculated by summarizing and analyzing a large number of historical financial transactions, evaluating the corresponding historical fraud risk value of a large number of historical financial transactions, and then combining the actual fraud situation of the historical financial transactions to calculate the first preset threshold.

[0037] According to a specific embodiment of the present invention, the first preset threshold is calculated based on the actual historical fraud situation of financial transactions, specifically including: Taking each historical financial business as a benchmark historical financial business, obtain the fraud situation corresponding to the benchmark historical financial business, and compare the historical fraud risk value corresponding to the benchmark historical financial business with the historical fraud risk values ​​corresponding to other historical financial businesses one by one; If the fraud situation of the benchmark historical financial business is a fraud business, and the fraud situation of the other historical financial business is a fraud business; if the historical fraud risk value corresponding to the benchmark historical financial business is greater than the historical fraud risk value corresponding to the other historical financial business, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold plus 0; if the historical fraud risk value corresponding to the benchmark historical financial business is less than or equal to the historical fraud risk value corresponding to the other historical financial business, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold plus 1; If the fraud situation of the benchmark historical financial business is non-fraudulent business, and the fraud situation of other historical financial businesses is fraudulent business; if the historical fraud risk value corresponding to the benchmark historical financial business is greater than the historical fraud risk value corresponding to other historical financial businesses, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold minus 1; if the historical fraud risk value corresponding to the benchmark historical financial business is less than or equal to the historical fraud risk value corresponding to other historical financial businesses, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold plus 1; If the fraud situation of the benchmark historical financial business is non-fraudulent business, and the fraud situation of other historical financial businesses is non-fraudulent business; if the historical fraud risk value corresponding to the benchmark historical financial business is greater than or equal to the historical fraud risk value corresponding to other historical financial businesses, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold plus 1; if the historical fraud risk value corresponding to the benchmark historical financial business is less than the historical fraud risk value corresponding to other historical financial businesses, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold plus 0; If the fraud situation of the benchmark historical financial business is a fraudulent business, and the fraud situation of the other historical financial businesses is a non-fraudulent business; if the historical fraud risk value corresponding to the benchmark historical financial business is greater than or equal to the historical fraud risk value corresponding to the other historical financial businesses, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold plus 1; if the historical fraud risk value corresponding to the benchmark historical financial business is less than the historical fraud risk value corresponding to the other historical financial businesses, then the historical fraud risk value of the benchmark historical financial business is selected as the contribution value of the first preset threshold minus 1; By comparing each historical financial business pairwise, the historical fraud risk value of each historical financial business is calculated and the contribution cumulative value of the first preset threshold is selected; The historical fraud risk value of the historical financial business with the highest cumulative contribution value is used as the first preset threshold.

[0038] like Figure 3 As shown, based on the current network structure feature information and the current financial business rules, and through a preset evaluation algorithm, an evaluation is performed to obtain a fraud risk value, specifically including: S302, building an anti-fraud evaluation model; S304, training the anti-fraud evaluation model using sample data; S306: Inputting the current network structure feature information and the current financial business rules into an anti-fraud evaluation model; S308: Output the obtained fraud risk value.

[0039] According to an embodiment of the present invention, after outputting the obtained fraud risk value, the method further includes: Obtain multiple historical financial business data; Filtering historical financial business data with similar characteristics to the current financial business request information from multiple historical financial business data and saving them to a similar characteristics database; Based on the historical financial business data in the similar feature library, the correction value is calculated according to the correction value calculation method; The correction value is added to the output fraud risk value to obtain the corrected fraud risk value.

[0040] According to an embodiment of the present invention, historical financial business data having characteristics similar to those of the current financial business request information is screened out from a plurality of historical financial business data and saved in a similar characteristics database, specifically including: It is preset that each historical financial business data at least includes historical financial business request information and historical actual fraud situations; Performing feature calculation on historical financial service request information in each historical financial service data to obtain feature values ​​of the historical financial service request information; Perform feature calculation on the current financial service request information to obtain a feature value of the current financial service request information; Based on each historical financial business data, calculating the difference between the characteristic value of the historical financial business request information of the historical financial business data and the characteristic value of the current financial business request information to obtain the difference between the two; It is determined whether the difference is less than a second preset threshold value. If so, the corresponding historical financial business data is saved in a similar feature library.

[0041] According to an embodiment of the present invention, based on the historical financial business data in the similar feature library and according to the correction value calculation method, the correction value is calculated, specifically including: Based on each historical financial business data in the similar feature library, the corresponding historical financial business rules and historical network structure feature information are calculated through the preset matching algorithm and graph neural network convolution algorithm respectively; Based on the historical network structure feature information and historical current financial business rules, and through a preset evaluation algorithm, a historical fraud risk prediction value is obtained; Based on each historical financial business data in the similar feature library, according to the historical fraud risk prediction value of the historical financial business data and the historical actual fraud situation, a correction component value of each historical financial business data is calculated by a first algorithm; The correction component values ​​calculated for each historical financial business data in the similar feature library are added together to obtain the sum of the correction component values, and the sum of the correction component values ​​is divided by the amount of historical financial business data in the similar feature library to obtain the correction value.

[0042] Understandably, given that the assessment model is constrained by its own parameters, the predicted fraud risk value may be biased. The present invention analyzes and calculates multiple historical financial business data to derive a revised value for the assessment model. This revised value is then used to modify the output fraud risk value, resulting in a more accurate fraud risk value. This further facilitates the subsequent initiation of corresponding anti-fraud measures based on the presence of fraud.

[0043] According to an embodiment of the present invention, the corresponding historical financial business rules and historical network structure feature information are calculated by a preset matching algorithm and a graph neural network convolution algorithm, specifically including: Extracting a plurality of historical financial business feature information from the historical financial business request information of the historical financial business data; Based on the rule engine, it matches multiple historical financial business feature information according to the preset matching algorithm and outputs the corresponding historical financial business rules; Extracting personal information of historical financial business requesters from historical financial business request information; Based on the personal information of historical financial business requesters, and combined with big data technology to obtain related information; Construct a graph neural network model of historical financial business requesters based on their personal information and related information; Based on the graph neural network model of historical financial business requesters, convolution calculation is performed to obtain historical network structure feature information.

[0044] According to an embodiment of the present invention, feature calculation is performed on the historical financial service request information in each historical financial service data to obtain a feature value of the historical financial service request information, specifically including: Obtain the age, credit score, historical financial purpose, and historical financial transaction amount of each historical financial transaction request information of the historical financial transaction requester; Use big data to calculate the average credit value of different ages at historical time points; According to the age of the historical financial service requester of each historical financial service request information, obtaining the average credit value corresponding to the age of the historical financial service requester; The attribute characteristic value of the historical financial service requester is calculated based on the credit value of the historical financial service requester and the average credit value corresponding to the age of the historical financial service requester, and the attribute characteristic value of the historical financial service requester = the credit value of the historical financial service requester / the average credit value corresponding to the age of the historical financial service requester; Use big data to calculate the average amount of various financial uses at historical points in time; According to the historical financial purpose of each historical financial business request information, obtain the average amount corresponding to the historical financial purpose; The historical financial business content characteristic value is calculated based on the historical financial business amount and the average amount corresponding to the historical financial purpose, and the historical financial business content characteristic value = historical financial business amount / average amount corresponding to the historical financial purpose.

[0045] According to an embodiment of the present invention, performing feature calculation on the current financial service request information to obtain a feature value of the current financial service request information specifically includes: Obtain the current financial service request information, including the age and credit score of the current financial service requester, as well as the current financial purpose and the current financial service amount; Use big data to calculate the average credit value of different ages at the current time point; Obtaining an average credit score corresponding to the age of the current financial service requester according to the age of the current financial service requester in the current financial service request information; The attribute characteristic value of the current financial service requester is calculated based on the credit value of the current financial service requester and the average credit value corresponding to the age of the current financial service requester, and the attribute characteristic value of the current financial service requester = the credit value of the current financial service requester / the average credit value corresponding to the age of the current financial service requester; Use big data to calculate the average amount of each financial purpose at the current time point; According to the current financial purpose of the current financial service request information, obtain the average amount corresponding to the current financial purpose; The current financial business content characteristic value is calculated based on the current financial business amount and the average amount corresponding to the current financial purpose, and the current financial business content characteristic value = the current financial business amount / the average amount corresponding to the current financial purpose.

[0046] According to an embodiment of the present invention, based on each historical financial service data, a difference between a characteristic value of the historical financial service request information of the historical financial service data and a characteristic value of the current financial service request information is calculated to obtain the difference between the two, specifically including: The historical financial business requester attribute characteristic values ​​and historical financial business content characteristic values ​​of the historical financial business request information of the historical financial business data and the current financial business requester attribute characteristic values ​​and current financial business content characteristic values ​​of the current financial business request information are calculated according to the second algorithm to obtain the difference between the two.

[0047] According to a specific embodiment of the present invention, the difference between the historical financial service requester attribute feature value and the historical financial service content feature value of the historical financial service request information of the historical financial service data and the current financial service requester attribute feature value and the current financial service content feature value of the current financial service request information is calculated according to the second algorithm to obtain the difference between the two, specifically including: Based on each historical financial business data, performing a difference calculation between the attribute characteristic value of the historical financial business requester of each historical financial business data and the attribute characteristic value of the current financial business requester of the current financial business request information to obtain the attribute characteristic difference degree of the financial business requester based on each historical financial business data; Based on each historical financial business data, performing a difference calculation between the historical financial business content feature value of each historical financial business data and the current financial business content feature value to obtain a financial business content feature difference based on each historical financial business data; Preset financial service requester attributes and financial service content have different influence weights on the difference between historical financial service request information and current financial service request information; Based on the historical financial business request information of each historical financial business data, multiply the corresponding financial business requester attribute feature difference by the influence weight of the financial business requester attribute to obtain the financial business requester attribute feature weight difference, and multiply the corresponding financial business content feature difference by the influence weight of the financial business content to obtain the financial business content feature weight difference; Based on each historical financial business data, the influence weight of the corresponding financial business requester attribute and the weight difference of the corresponding financial business content feature are added together to obtain the sum of the weight differences; Add the influence weight of the financial service requester's attributes and the influence weight of the financial service content to obtain the sum of the weights; Based on each historical financial service data, the sum of the weighted differences is divided by the sum of the weights to obtain the difference between the characteristic value of the historical financial service request information of the historical financial service data and the characteristic value of the current financial service request information.

[0048] According to a specific embodiment of the present invention, based on each historical financial business data in the similar feature library, according to the historical fraud risk prediction value of the historical financial business data and the historical actual fraud situation, the correction component value of each historical financial business data is calculated by a first algorithm, specifically including: Comparing the historical fraud risk prediction value of the historical financial business data with a first preset threshold; If it is greater than or equal to the first preset threshold, and the actual historical fraud situation is fraudulent behavior, the correction component value of the historical financial business data is recorded as 0; If it is greater than or equal to the first preset threshold, and the actual historical fraud situation is not a fraudulent act, the correction component value of the historical financial business data is recorded as the difference between the first preset threshold and the historical fraud risk prediction value of the historical financial business data, and is a negative value; If it is less than the first preset threshold, and the historical actual fraud situation is fraudulent behavior, the correction component value of the historical financial business data is recorded as the difference between the first preset threshold and the historical fraud risk prediction value of the historical financial business data, and is a positive value; If it is less than the first preset threshold and the actual historical fraud situation is not a fraudulent act, the correction component value of the historical financial business data is recorded as 0.

[0049] like Figure 4 As shown, multiple financial service feature information is matched according to a preset matching algorithm, and the corresponding current financial service rules are output, including: S402, multiple financial business rules are preset; S404, based on each financial business feature information, comparing each financial business rule with the remaining other financial business rules one by one; S406, calculating the total weighted matching value of each financial service rule on all financial service feature information according to the third algorithm; S408: Select the financial business rule with the largest total weight matching value as the corresponding current financial business rule.

[0050] According to a specific embodiment of the present invention, based on each piece of financial service feature information, each financial service rule is compared with the remaining other financial service rules one by one; and the total weighted matching value of each financial service rule on all financial service feature information is calculated according to the third algorithm, specifically including: Select a piece of financial business characteristic information as the benchmark financial business characteristic information, compare each financial business rule with the remaining other financial business rules one by one, and determine which of the two is more consistent with the benchmark financial business characteristic information; If the former is more consistent with the benchmark financial business characteristic information, the consistency value of the former financial business rule on the benchmark financial business characteristic information is increased by 1; Based on the benchmark financial business feature information, after all financial business rules are compared pairwise, a total matching value of each financial business rule on the benchmark financial business feature information is cumulatively calculated; Compare and analyze other financial business feature information according to the above steps to obtain the total matching value of each financial business rule on each financial business feature information; The influence weights of each financial business feature information on the matching of financial business rules are preset to be different; Based on each financial business rule, its total matching value on each financial business feature information is multiplied by the influence weight of the corresponding financial business feature information, and multiple products are added together to calculate the total weight matching value of each financial business rule on all financial business feature information.

[0051] It should be noted that, due to the large amount of financial business characteristic information, a certain financial business rule may match the first financial business characteristic information but not the second financial business characteristic information, while another financial business rule may not match the first financial business characteristic information but match the second financial business characteristic information, and the degree of matching cannot be unified. Therefore, when selecting financial business rules in the traditional way, it is impossible to rationally select the optimal financial business rule. The present invention compares the financial business rules in pairs based on the corresponding financial business characteristic information to determine which one is more consistent with the financial business characteristic information, thereby calculating a more reasonable matching value for each financial business rule, and the matching value obtained is a relatively relative matching value obtained with all other financial business rules as a reference object, which is conducive to the subsequent selection of the best from multiple financial business rules and matching the corresponding current financial business rules. The present invention further combines the different influence weights of different financial business characteristic information on the matching of financial business rules, thereby promoting the subsequent matching of financial business rules that are more accurate and more in line with the current financial business request.

[0052] The present invention also proposes a financial anti-fraud intelligent assessment system that integrates a rule engine and a graph neural network, including a memory and a processor. The memory includes a financial anti-fraud intelligent assessment method program that integrates a rule engine and a graph neural network. When the financial anti-fraud intelligent assessment method program that integrates a rule engine and a graph neural network is executed by the processor, the steps of the above-mentioned financial anti-fraud intelligent assessment method are implemented.

[0053] The present invention proposes a financial anti-fraud intelligent assessment method that integrates a rule engine and a graph neural network. Through the rule engine, multiple financial business feature information is matched according to a preset matching algorithm to output the corresponding current financial business rules. The personal information of the current financial business requester is then extracted from the current financial business request information. Related information is obtained by combining big data technology. A graph neural network model is constructed based on the personal information and related information of the current financial business requester. Convolution calculation is performed based on the graph neural network model to obtain current network structure feature information. Based on the current network structure feature information and the current financial business rules, an assessment is performed using a preset assessment algorithm to finally obtain a fraud risk value. This facilitates judging the fraudulent status of the current financial business based on the fraud risk value, so as to implement anti-fraud measures for financial business requests that are judged to be fraudulent. By integrating the rule engine and the graph neural network, the present invention can make an accurate and intelligent assessment of whether the financial business request information is fraudulent or not, and implement anti-fraud measures for fraudulent financial business, effectively preventing the significant losses caused by financial fraud to financial institutions.

[0054] It should be understood that the above description is for illustrative purposes only and is not intended to limit the scope of the present invention. For those of ordinary skill in the art, various modifications and variations can be made in conjunction with the description of the present invention. However, these modifications and variations will not deviate from the scope of the present invention.

[0055] The basic concepts have been described above. It will be apparent to those skilled in the art after reading this application that the above disclosure is provided for illustrative purposes only and does not limit the present invention. Although not explicitly stated herein, various modifications, improvements, and revisions may be made by those skilled in the art to the present invention. Such modifications, improvements, and revisions are suggested in the present invention and remain within the spirit and scope of the exemplary embodiments of the present invention.

[0056] At the same time, the present invention uses specific terms to describe the embodiments of the present invention. For example, "one embodiment," "an embodiment," and / or "some embodiments," or "a specific embodiment" refer to a feature, structure, or characteristic associated with at least one embodiment of the present invention. Therefore, it should be emphasized and noted that the mention of "one embodiment," "an embodiment," "an alternative embodiment," or "a specific embodiment" two or more times in different places in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of the present invention may be appropriately combined.

[0057] In addition, it will be understood by those skilled in the art that various aspects of the present invention may be illustrated and described in conjunction with a plurality of patentable categories or situations, including any new and useful process, machine, product or combination of substances, or any new and useful improvement thereof. Accordingly, various aspects of the present invention may be performed entirely by hardware, entirely by software (including firmware, resident software, microcode, etc.), or by a combination of hardware and software. The above hardware or software may be referred to as a "unit," "module," or "system." In addition, various aspects disclosed in the present invention may take the form of a computer program product embodied in one or more computer-readable media, wherein computer-readable program code is contained therein.

[0058] A computer-readable signal medium may include a propagated data signal embodying computer program code, for example, in baseband or as part of a carrier wave. Such propagated signals may be in a variety of forms, including electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium other than a computer-readable storage medium that can be coupled to an instruction execution system, apparatus, or device to communicate, propagate, or transmit a program for use. The program code on a computer-readable signal medium may be transmitted in conjunction with any suitable medium, including radio, cable, fiber optic cable, RF, or similar media, or any combination thereof.

[0059] The computer program code required for the operation of the various components of the present invention may be written in any one or more programming languages, including subject-oriented programming languages ​​such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python, and the like, conventional procedural programming languages ​​such as C, Visual Basic, Fortran 2003, Perl, COBOL 2002, PHP, ABAP, active programming languages ​​such as Python, Ruby, and Groovy, or other programming languages. The program code may be executed entirely on the power operation partition computer, or as a standalone software package on the power operation partition computer, or partially on the power operation partition computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the power operation partition computer in conjunction with any network configuration, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., in conjunction with the Internet), or in a cloud computing environment, or as a service such as software as a service (SaaS).

[0060] In addition, unless expressly stated in the claims, the order of the processing elements and sequences, the use of alphanumeric characters, or the use of other names described in the present invention are not intended to limit the order of the processes and methods of the present invention. Although the above disclosure discusses some embodiments of the invention that are currently considered useful in conjunction with various examples, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that match the essence and scope of the embodiments of the present invention. For example, although the system components described above can be implemented in conjunction with hardware devices, they can also be implemented in conjunction with software solutions alone, such as installing the described system on an existing server or mobile device.

[0061] Similarly, it should be noted that in order to simplify the description of the present invention and thus facilitate understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the invention, various features may sometimes be combined into one embodiment, figure, or description thereof. Similarly, it should be noted that in order to simplify the description of the present invention and thus facilitate understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the invention, various features may sometimes be combined into one embodiment, figure, or description thereof.

Claims

1. A financial anti-fraud intelligent assessment method integrating rule engine and graph neural network, characterized in that: The method comprises: Get current financial business request information; Extracting multiple financial service feature information from the current financial service request information; Based on the rule engine, it matches multiple financial business feature information according to the preset matching algorithm and outputs the corresponding current financial business rules; Extracting personal information of the current financial service requester from the current financial service request information; Based on the personal information of the current financial service requester, and combined with big data technology to obtain related information; Build a graph neural network model based on the personal information and related information of the current financial service requester; Perform convolution calculations based on the graph neural network model to obtain the current network structure feature information; Based on the current network structure feature information and the current financial business rules, an evaluation is performed through a preset evaluation algorithm to obtain a fraud risk value.

2. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 1 is characterized in that: After obtaining the fraud risk value, the method further includes: Determining whether the fraud risk value is greater than a first preset threshold; If so, the current financial service request information is identified as fraudulent information and anti-fraud response measures are initiated.

3. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 1 is characterized in that: Based on the current network structure feature information and the current financial business rules, and through a preset evaluation algorithm, an evaluation is performed to obtain a fraud risk value, specifically including: Build anti-fraud assessment models; Train the anti-fraud assessment model using sample data; Inputting the current network structure feature information and the current financial business rules into an anti-fraud evaluation model; The output is the fraud risk value.

4. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 3 is characterized in that: After outputting the fraud risk value, the method further includes: Obtain multiple historical financial business data; Filtering historical financial business data with similar characteristics to the current financial business request information from multiple historical financial business data and saving them to a similar characteristics database; Based on the historical financial business data in the similar feature library, the correction value is calculated according to the correction value calculation method; The correction value is added to the output fraud risk value to obtain the corrected fraud risk value.

5. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 4 is characterized in that: Filter out historical financial business data with similar characteristics to the current financial business request information from multiple historical financial business data and save them to a similar characteristics database, including: It is preset that each historical financial business data at least includes historical financial business request information and historical actual fraud situations; Performing feature calculation on historical financial service request information in each historical financial service data to obtain feature values ​​of the historical financial service request information; Perform feature calculation on the current financial service request information to obtain a feature value of the current financial service request information; Based on each historical financial business data, calculating the difference between the characteristic value of the historical financial business request information of the historical financial business data and the characteristic value of the current financial business request information to obtain the difference between the two; It is determined whether the difference is less than a second preset threshold value. If so, the corresponding historical financial business data is saved in a similar feature library.

6. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 4 is characterized in that: Based on the historical financial business data in the similar feature library and according to the correction value calculation method, the correction value is calculated, specifically including: Based on each historical financial business data in the similar feature library, the corresponding historical financial business rules and historical network structure feature information are calculated through the preset matching algorithm and graph neural network convolution algorithm respectively; Based on the historical network structure feature information and historical current financial business rules, and through a preset evaluation algorithm, a historical fraud risk prediction value is obtained; Based on each historical financial business data in the similar feature library, according to the historical fraud risk prediction value of the historical financial business data and the historical actual fraud situation, a correction component value of each historical financial business data is calculated by a first algorithm; The correction component values ​​calculated for each historical financial business data in the similar feature library are added together to obtain the sum of the correction component values, and the sum of the correction component values ​​is divided by the amount of historical financial business data in the similar feature library to obtain the correction value.

7. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 6 is characterized in that: The corresponding historical financial business rules and historical network structure feature information are calculated through the preset matching algorithm and graph neural network convolution algorithm, including: Extracting a plurality of historical financial business feature information from the historical financial business request information of the historical financial business data; Based on the rule engine, it matches multiple historical financial business feature information according to the preset matching algorithm and outputs the corresponding historical financial business rules; Extracting personal information of historical financial business requesters from historical financial business request information; Based on the personal information of historical financial business requesters, and combined with big data technology to obtain related information; Construct a graph neural network model of historical financial business requesters based on their personal information and related information; Based on the graph neural network model of historical financial business requesters, convolution calculation is performed to obtain historical network structure feature information.

8. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 5 is characterized in that: Perform feature calculation on the historical financial service request information in each historical financial service data to obtain feature values ​​of the historical financial service request information, specifically including: Obtain the age, credit score, historical financial purpose, and historical financial transaction amount of each historical financial transaction request information of the historical financial transaction requester; Use big data to calculate the average credit value of different ages at historical time points; According to the age of the historical financial service requester of each historical financial service request information, obtaining the average credit value corresponding to the age of the historical financial service requester; The attribute characteristic value of the historical financial service requester is calculated based on the credit value of the historical financial service requester and the average credit value corresponding to the age of the historical financial service requester, and the attribute characteristic value of the historical financial service requester = the credit value of the historical financial service requester / the average credit value corresponding to the age of the historical financial service requester; Use big data to calculate the average amount of various financial uses at historical points in time; According to the historical financial purpose of each historical financial business request information, obtain the average amount corresponding to the historical financial purpose; The historical financial business content characteristic value is calculated based on the historical financial business amount and the average amount corresponding to the historical financial purpose, and the historical financial business content characteristic value = historical financial business amount / average amount corresponding to the historical financial purpose.

9. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 8 is characterized in that: Perform feature calculation on the current financial service request information to obtain the feature value of the current financial service request information, specifically including: Obtain the current financial service request information, including the age and credit score of the current financial service requester, as well as the current financial purpose and the current financial service amount; Use big data to calculate the average credit value of different ages at the current time point; Obtaining an average credit score corresponding to the age of the current financial service requester according to the age of the current financial service requester in the current financial service request information; The attribute characteristic value of the current financial service requester is calculated based on the credit value of the current financial service requester and the average credit value corresponding to the age of the current financial service requester, and the attribute characteristic value of the current financial service requester = the credit value of the current financial service requester / the average credit value corresponding to the age of the current financial service requester; Use big data to calculate the average amount of each financial purpose at the current time point; According to the current financial purpose of the current financial service request information, obtain the average amount corresponding to the current financial purpose; The current financial business content characteristic value is calculated based on the current financial business amount and the average amount corresponding to the current financial purpose, and the current financial business content characteristic value = the current financial business amount / the average amount corresponding to the current financial purpose.

10. The financial anti-fraud intelligent assessment method integrating rule engine and graph neural network according to claim 9 is characterized in that: Based on each historical financial business data, a difference between the characteristic value of the historical financial business request information of the historical financial business data and the characteristic value of the current financial business request information is calculated to obtain the difference between the two, specifically including: The historical financial business requester attribute characteristic values ​​and historical financial business content characteristic values ​​of the historical financial business request information of the historical financial business data and the current financial business requester attribute characteristic values ​​and current financial business content characteristic values ​​of the current financial business request information are calculated according to the second algorithm to obtain the difference between the two.

Images