Multi-source data fusion generation method for illegal capital risk monitoring

Abstract

The invention provides a multi-source data fusion generation method for illegal capital risk monitoring, which can generate high-quality and approximately real illegal capital behavior data on the premise of ensuring data privacy security through fusion of multi-source data, a rule-based generation strategy and a noise injection technology. The privacy and compliance problems of real financial data acquisition are effectively avoided, and a rich and safe data source is provided for researchers. Moreover, each rule in the rule set represents a typical mode of an illegal capital collection behavior, and comprises capital flow features, associated account features and cross-border transfer features, so that the generated data set can be close to the features of the real illegal capital collection behavior in content and structure, and the data set is more accurate to use. And the characteristics of complex fund flow, associated accounts and the like are simulated through fusion of multi-source data and injection of noise, so that the generated data has relatively high representativeness and authenticity.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source data processing, and in particular to a multi-source data fusion generation method for illegal fund-raising risk monitoring. Background Art

[0002] New types of illegal fundraising activities pose great challenges to financial supervision due to their strong concealment, complex behavior and cross-regional characteristics. At present, relevant researchers are facing significant difficulties in the field of risk monitoring and behavior identification. The main reasons include the following: First, data acquisition is limited: real financial data cannot be made public due to privacy protection and legal compliance, and competition factors limit data sharing among financial institutions. Second, incomplete labels: Even if some data is obtained, many illegal fundraising behaviors in the data are not identified and labeled in time, making it difficult to provide effective supervision signals for the model. Third, the transaction structure is complex: the cross-account and cross-regional fund flow patterns are diverse, which increases the difficulty of behavior analysis and identification.

[0003] Compared with the fields of anti-money laundering and anti-fraud, the data set in the new illegal fund-raising field is almost blank. Although anti-money laundering projects such as AMLSim and AMLworld provide some references in generating data, their scenarios are limited to money laundering behaviors and cannot be directly applied to the complex models of illegal fund-raising, and it is difficult to fully reflect the complexity and diversity of illegal fund-raising behaviors.

[0004] Existing illegal financial activity supervision technologies rely on real transaction data and traditional machine learning methods in risk monitoring. However, due to privacy, compliance and competition restrictions, data sharing is hindered, labels are incomplete, and models lack accurate supervisory signals during training, which seriously affects recognition and prediction performance. In order to make up for the lack of real data, some studies have tried data generation methods based on agent simulation, such as the AMLworld project in the field of anti-money laundering. Although these methods have made some progress in enhancing data diversity, they lack in-depth modeling of the unique patterns of illegal fundraising behavior, and have obvious deficiencies in characterizing key features such as capital flow paths, cross-border transaction frequencies, and account correlations. In addition, the existing generation methods are highly random and usually lack the guidance of preset rules, resulting in the generated data being out of touch with the real financial behavior patterns, making it difficult to provide the model with meaningful risk feature extraction capabilities. These problems limit the applicability of existing technologies in illegal fundraising risk monitoring and cannot meet the needs of this field for high-complexity and diversified pattern analysis.

[0005] As mentioned above, existing financial data is difficult to obtain due to restrictions such as privacy protection and legal compliance. At the same time, problems such as incomplete labels and isolated information have led to a serious lack of data support for researchers in the monitoring of illegal fundraising risks, especially in the field of new types of illegal fundraising. Traditional data sets are difficult to fully reflect the complexity and diversity of illegal fundraising behaviors, and relying only on a small amount of real data is not enough to cover the changing illegal fundraising models. Therefore, it is urgent to design a data generation method based on the characteristics of illegal fundraising behaviors and build a high-quality data set suitable for this field. Summary of the invention

[0006] The present invention is made to solve the above problems, and aims to provide a method for generating a high-quality data set related to illegal fund-raising. The present invention adopts the following technical solutions:

[0007] The present invention provides a multi-source data fusion generation method for illegal fund-raising risk monitoring, which has the following technical features and includes the following steps: step S1, constructing a rule set, which includes multiple rules, each of which represents a typical pattern of illegal fund-raising behavior, and each of which includes capital flow characteristics, associated account characteristics and cross-border transfer characteristics; step S2, constructing multiple simulation accounts; step S3, generating simulated transaction data based on multiple simulation accounts and the rule set; step S4, injecting noise into the simulated transaction data; step S5, setting corresponding labels for the simulated transaction data after the noise is injected, identifying data that meets the characteristics of illegal fund-raising, thereby forming an illegal fund-raising related data set.

[0008] The multi-source data fusion generation method for illegal fundraising risk monitoring provided by the present invention may also have such technical characteristics, wherein the capital flow characteristics are used to define the transfer frequency and transfer amount between the simulated accounts, the associated account characteristics are used to define the number of other simulated accounts associated with the same simulated account and their transaction patterns, and the cross-border transfer characteristics are used to define the capital flow rules between different countries or regions.

[0009] The multi-source data fusion generation method for illegal fund-raising risk monitoring provided by the present invention may also have such a technical feature, wherein the rule set R is expressed as:

[0010] R = {f 1 ,f 2 ,…,f n}

[0011] In the formula, f i Represents a characteristic rule, and the characteristic rules in each rule set R jointly describe a type of illegal fundraising behavior pattern.

[0012] The multi-source data fusion generation method for illegal fundraising risk monitoring provided by the present invention may also have such a technical feature, wherein, in step S2, the number of the simulated accounts is set, and the parameters of each of the simulated accounts are set, and the parameters include basic information, initial balance, and transaction frequency.

[0013] The multi-source data fusion generation method for illegal fundraising risk monitoring provided by the present invention may also have such a technical feature, wherein the distribution of the initial balances of multiple simulated accounts is a normal distribution, and the transaction frequency includes three levels: high, medium, and low, and each simulated account is randomly set to one of the levels.

[0014] The multi-source data fusion generation method for illegal fund-raising risk monitoring provided by the present invention may also have such a technical feature, wherein step S3 includes the following sub-steps: step S3-1, according to the rule set, using the corresponding simulated account to simulate the transaction behavior to generate initial simulated transaction data; step S3-2, setting the total number of time steps; step S3-3, in the nth time step, randomly selecting the applicable rule R from the rule set i ; Step S3-4, according to the selected rule R i The characteristic definition of the simulated account is modified, including setting the direction of fund flow, transfer frequency, and transfer amount; step S3-5, judging whether the current time step is the last time step, and when the judgment is no, entering the next time step, repeating steps S3-3 to S3-4; when the judgment is yes, obtaining the simulated transaction data D containing the characteristics of illegal fund-raising behavior ′ .

[0015] The multi-source data fusion generation method for illegal fundraising risk monitoring provided by the present invention may also have such a technical feature, wherein, in step S4, the proportion of noise data in the data set is set, and a predetermined noise injection method is used to inject noise data into the generated simulated transaction data according to the proportion, so as to simulate abnormal situations or randomness in real financial transactions.

[0016] The multi-source data fusion generation method for illegal fund-raising risk monitoring provided by the present invention may also have such a technical feature, wherein the noise injection method includes: a random injection method, randomly generating a number of normal transaction behaviors in the generated simulated transaction data; and an abnormal point generation method, adding abnormal transactions to a specific simulated account, wherein the abnormal transactions include large transfers and frequent transactions, and the simulated transaction data after noise injection is expressed as:

[0017] D″=D ′ +∈

[0018] Where D ′ is the simulated transaction data, ∈ is a noise vector, and its value is generated according to the normal distribution N(0,σ).

[0019] The multi-source data fusion generation method for illegal fund-raising risk monitoring provided by the present invention may also have such a technical feature, wherein, in step S5, the label is represented as:

[0020]

[0021] In the formula, the label 1 represents that the sample contains the characteristics of illegal fund-raising behavior, and the label 0 represents that the sample does not contain the characteristics of illegal fund-raising behavior and belongs to normal trading behavior.

[0022] Functions and Effects of the Invention

[0023] According to the multi-source data fusion generation method for illegal fund-raising risk monitoring provided by the present invention, the method includes the steps of constructing a rule set, constructing a simulated account, using the simulated account to generate simulated transaction data according to the rule set, the step of noise injection, and the step of label setting. Among them, by fusing multi-source data, rule-based generation strategy and noise injection technology, high-quality, close to real illegal fund-raising behavior data can be generated under the premise of ensuring data privacy security, effectively avoiding the privacy and compliance issues of real financial data acquisition, and providing researchers with a rich and secure data source. In addition, since each rule in the rule set represents a typical pattern of illegal fund-raising behavior, and contains fund flow characteristics, associated account characteristics and cross-border transfer characteristics, the generated data set can be close to the characteristics of real illegal fund-raising behavior in terms of content and structure, and the complex fund flow and associated account characteristics are simulated through the fusion of multi-source data and the injection of noise, so that the generated data has high representativeness and authenticity.

[0024] The method of the present invention can provide more accurate training data for the illegal fund-raising risk monitoring model, so that the accuracy of the downstream model in abnormal behavior detection, risk identification, etc. is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the key steps of the multi-source data fusion generation method for illegal fund-raising risk monitoring in an embodiment of the present invention;

[0026] Figure 2 is a flow chart of a multi-source data fusion generation method for illegal fund-raising risk monitoring in an embodiment of the present invention;

[0027] Figure 3 is a flow chart of step S3 in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the multi-source data fusion generation method for illegal fund-raising risk monitoring of the present invention is specifically described below in combination with embodiments and drawings.

[0029] Figure 1 is a schematic diagram of the key steps of the multi-source data fusion generation method for illegal fund-raising risk monitoring in this embodiment, Figure 2 It is a flow chart of the multi-source data fusion generation method for illegal fund-raising risk monitoring in this embodiment.

[0030] like Figure 1 and Figure 2 As shown, the method comprises the following steps:

[0031] Step S1, constructing a rule set, which includes multiple rules, each rule represents a typical pattern of illegal fund-raising behavior.

[0032] Step S2, construct multiple simulation accounts.

[0033] Step S3, generating simulated transaction data based on multiple simulated accounts and rule sets.

[0034] Step S4, injecting noise into the simulated transaction data.

[0035] Step S5, setting corresponding labels for the simulated transaction data after the noise is injected, identifying the data that meets the characteristics of illegal fund-raising, thereby forming a data set related to illegal fund-raising.

[0036] The above steps will be described in detail below.

[0037] Step S1, constructing a rule set, which includes multiple rules, each rule represents a typical pattern of illegal fund-raising behavior.

[0038] In step S1, a set of rules containing illegal fund-raising features is constructed to provide a structured pattern for generating data. Each rule represents a typical pattern of illegal fund-raising behavior, including fund flow features, account association features, and cross-border transfer features. This step ensures that the generated data has the characteristics of real illegal fund-raising behavior, so that the model can capture the patterns in real illegal fund-raising activities during training.

[0039] Specifically, a rule set R is constructed, in which each rule contains the following elements: capital flow characteristics, which define the transfer frequency and transfer amount between simulated accounts; associated account characteristics, which define the number and transaction pattern of other simulated accounts associated with the same simulated account; cross-border transfer characteristics, which set capital flow rules between different countries or regions.

[0040] The definition of the rule set R can be designed based on existing illegal fundraising behavior patterns, expert knowledge or existing anti-money laundering rules, or it can be derived based on observations of real data. These rules are expressed by the formula:

[0041] R = {f 1 ,f 2 ,…,f n}

[0042] In the formula, f i Represents a feature rule, and the features in each rule set R together describe a type of illegal fundraising behavior pattern.

[0043] Step S2, constructing multiple simulated accounts, and using the multiple simulated accounts to perform simulated transactions, thereby generating corresponding initial simulated transaction data.

[0044] In step S2, basic parameters and settings are defined for the data generation process. By setting the number of simulated accounts, as well as basic information, initial balance, transaction frequency and other parameters of each simulated account, the basic information includes account name, account number, account type, country, etc., to lay the foundation for the subsequent generation of simulated transaction data that conforms to the rule set.

[0045] Specifically, in step S2, the total number of simulated accounts is first set, which can be set according to the scale of the actual illegal fund-raising activities. Then, an initial balance is assigned to each simulated account, and a random distribution or a normal distribution can be used to simulate the diversity of fund sizes of different accounts. After that, the transaction frequency of each simulated account is defined, which can be adjusted according to the account type or risk level. For example, the total number of simulated accounts is set to 1,000, the distribution of the initial balance of each simulated account is a normal distribution N(μ,σ), the transaction frequency includes three levels: high, medium, and low, and each simulated account is randomly set to one of the transaction frequency levels.

[0046] Step S3, generating simulated transaction data based on multiple simulated accounts and rule sets.

[0047] In step S3, a predefined set of rules is applied to simulate and transform the account data to generate transaction data that meets the characteristics of illegal fund-raising. This step ensures that the generated data sample contains illegal fund-raising behavior patterns, which is convenient for subsequent model training and testing.

[0048] Specifically, firstly, simulated transactions are performed using corresponding simulated accounts according to the rule set to generate initial simulated transaction data. Then, rules are selected from the rule set, and the initial simulated transaction data are transformed based on the selected rules to obtain simulated transaction data that meets the characteristics of illegal fund-raising.

[0049] Figure 3 is a flow chart of step S3 in this embodiment.

[0050] like Figure 3 As shown, step S3 specifically includes the following sub-steps:

[0051] Step S3-1, according to the rule set, use the corresponding simulation account to simulate the transaction behavior to generate initial simulation transaction data.

[0052] That is, based on the relationship account characteristics and cross-border transfer characteristics in the rules, multiple groups of simulated accounts that can be traded are found, and one or more simulated transaction data are generated between each group of simulated accounts based on the corresponding capital flow characteristics. Each data includes information such as the direction of capital flow (payer and payee), transfer time, and transfer amount.

[0053] Step S3-2, setting the total number of time steps.

[0054] Step S3-3, in the nth time step, randomly select an applicable rule R from the rule set i .

[0055] Step S3-4, according to the selected rule R i The feature definition of , modify the transaction behavior between the simulated accounts, including setting the direction of capital flow, transfer frequency, transfer amount, etc., so as to obtain the transformed data D ′ , the transformed data D ′ It contains the characteristics of illegal fund-raising activities.

[0056] Step S3-5, determine whether the current time step is the last time step. If it is not, enter the next time step and repeat steps S3-3 to S3-4. If it is yes, enter the end state and obtain simulated transaction data containing the characteristics of illegal fund-raising behavior.

[0057] Step S4, injecting noise into the simulated transaction data.

[0058] In step S4, in order to make the data closer to the real environment, a certain proportion of noise data is added in the generation process to simulate the abnormal situations or randomness in real financial transactions and increase the diversity and robustness of the data set.

[0059] Specifically, first, the proportion of noise data in the data set is set, such as 5% or 10%. Then, a predetermined noise injection method is used to inject noise data into the generated simulated transaction data according to the proportion. In this embodiment, a random injection method and an abnormal point generation method are used. The random injection method specifically refers to randomly generating some normal transaction behaviors in the generated simulated transaction data to blur the illegal fund-raising characteristics. The abnormal point generation method specifically refers to adding abnormal transactions, such as large transfers or frequent transactions, to a specific simulated account to simulate false positives and normal data fluctuations.

[0060] The noise injection formula is as follows:

[0061] D″=D ′ +∈

[0062] Where ∈ is a noise vector, whose value is generated according to the normal distribution N(0,σ) to control the intensity of the noise. The generated data is denoted as D″, which includes the noise and the data after the rule transformation.

[0063] Step S5, setting corresponding labels for the simulated transaction data after the noise is injected, identifying the data that meets the characteristics of illegal fund-raising, thereby forming a data set related to illegal fund-raising.

[0064] In step S5, labels are added to the generated data set to identify which data meet the characteristics of illegal fund-raising. These labels are used for subsequent model training and evaluation to help the model identify illegal fund-raising patterns and risk categories, which can be expressed as:

[0065]

[0066] In the formula, the label 1 represents that the sample contains the characteristics of illegal fund-raising behavior, and the label 0 represents that the sample does not contain the characteristics of illegal fund-raising behavior and belongs to normal trading behavior.

[0067] The specific label generation process can be combined with the judgment conditions of the application rules to automatically label, or through expert review for auxiliary labeling. The final generated label dataset D L ={D i ,Label(D i )} for subsequent model training.

[0068] Functions and Effects of the Embodiments

[0069] According to the multi-source data fusion generation method for illegal fund-raising risk monitoring provided by this embodiment, the method includes the steps of constructing a rule set, constructing a simulated account, generating simulated transaction data using the simulated account according to the rule set, injecting noise, and setting labels. Among them, by fusing multi-source data, rule-based generation strategy, and noise injection technology, high-quality, close-to-real illegal fund-raising behavior data can be generated under the premise of ensuring data privacy security, effectively avoiding the privacy and compliance issues of real financial data acquisition, and providing researchers with a rich and secure data source. In addition, since each rule in the rule set represents a typical pattern of illegal fund-raising behavior, and contains fund flow characteristics, associated account characteristics, and cross-border transfer characteristics, the generated data set can be close to the characteristics of real illegal fund-raising behavior in terms of content and structure, and the complex fund flow and associated account characteristics are simulated by the fusion of multi-source data and the injection of noise, so that the generated data has high representativeness and authenticity. The method of this embodiment can provide more accurate training data for the illegal fund-raising risk monitoring model, so that the accuracy of the downstream model in abnormal behavior detection, risk discrimination, etc. is significantly improved.

[0070] In the embodiment, the transaction behavior between simulated accounts is modified according to the rules in the rule set, including corresponding modifications to the direction of capital flow, transaction frequency, transaction financial size, etc. Therefore, the simulated transaction data after such conversion not only has the characteristics of illegal fund-raising behavior, but also has good data richness, which is also conducive to model training.

[0071] Furthermore, random injection and anomaly point generation are used in noise injection, which not only adds normal transaction data, but also adds conventional abnormal data such as large transactions and frequent transactions. Therefore, it not only blurs the illegal fundraising characteristics in the data set, but also further enriches the data, making the generated simulated transaction data closer to the real financial transaction data, so that the model trained using this data set can also have a good predictive effect on the real transaction data.

[0072] The above embodiments are only used to illustrate the specific implementation of the present invention, and the present invention is not limited to the description scope of the above embodiments. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only to illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multi-source data fusion generation method for illegal fund-raising risk monitoring, characterized in that: The following steps are involved: Step S1, constructing a rule set, which includes multiple rules, each of which represents a typical pattern of illegal fund-raising behavior, and each of which includes fund flow characteristics, associated account characteristics, and cross-border transfer characteristics; Step S2, construct multiple simulation accounts; Step S3, generating simulated transaction data based on multiple simulated accounts and rule sets; Step S4, injecting noise into the simulated transaction data; Step S5, setting corresponding labels for the simulated transaction data after the noise is injected, identifying the data that meets the characteristics of illegal fund-raising, thereby forming a data set related to illegal fund-raising.

2. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 1 is characterized by: in, The fund flow characteristics are used to define the frequency and amount of transfers between the simulated accounts. The associated account feature is used to define the number of other simulated accounts associated with the same simulated account and their transaction patterns. The cross-border transfer feature is used to define the rules for the flow of funds between different countries or regions.

3. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 2 is characterized by: in, The rule set R is expressed as: R={f1,f2,…,f n } In the formula, f i Represents a characteristic rule, and the characteristic rules in each rule set R jointly describe a type of illegal fundraising behavior pattern.

4. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 2 is characterized by: in, In step S2, the number of the simulated accounts is set, and the parameters of each of the simulated accounts are set, wherein the parameters include basic information, initial balance, and transaction frequency.

5. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 4 is characterized by: in, The distribution of the initial balances of the plurality of simulated accounts is a normal distribution, The transaction frequency includes three levels: high, medium and low, and is randomly set to one of the levels for each of the simulated accounts.

6. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 4 is characterized by: in, Step S3 includes the following sub-steps: Step S3-1, according to the rule set, using the corresponding simulated account to simulate the transaction behavior to generate initial simulated transaction data; Step S3-2, setting the total number of time steps; Step S3-3, in the nth time step, randomly select the applicable rule R from the rule set i ; Step S3-4, according to the selected rule R i feature definition, modify the transaction behavior between the simulated accounts, including setting the direction of fund flow, transfer frequency, and transfer amount; Step S3-5, judging whether the current time step is the last time step, if the judgment is no, entering the next time step, repeating steps S3-3 to S3-4; if the judgment is yes, obtaining the simulated transaction data D containing the characteristics of illegal fund-raising behavior ′ .

7. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 1 is characterized by: in, In step S4, the proportion of noise data in the data set is set, and according to the proportion, a predetermined noise injection method is used to inject noise data into the generated simulated transaction data to simulate abnormal situations or randomness in real financial transactions.

8. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 7, Features: Wherein, the noise injection method comprises: A random injection method is used to randomly generate a number of normal transaction behaviors in the generated simulated transaction data; and The abnormal point generation method is to add abnormal transactions to the specific simulated account, and the abnormal transactions include large transfers and frequent transactions. The simulated transaction data after noise injection is expressed as: D″=D ′ +∈ Where D ′ is the simulated transaction data, ∈ is a noise vector, and its value is generated according to the normal distribution N(0,σ).

9. The multi-source data fusion generation method for illegal fund-raising risk monitoring according to claim 1 is characterized by: in, In step S5, the tag is represented as: In the formula, the label 1 represents that the sample contains the characteristics of illegal fund-raising behavior, and the label 0 represents that the sample does not contain the characteristics of illegal fund-raising behavior and belongs to normal trading behavior.

Images