Multi-source data fusion generation method for illegal fund-raising risk monitoring

Abstract

The application provides a multi-source data fusion generation method for illegal fund-raising risk monitoring. By fusing multi-source data, rule-based generation strategy and noise injection technology, high-quality and close-to-real illegal fund-raising behavior data can be generated under the premise of ensuring data privacy and security, effectively avoiding the privacy and compliance problems of real financial data acquisition, and providing researchers with rich and safe data sources. Moreover, 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 content and structure, and through the fusion of multi-source data and the injection of noise, the complex fund flow and associated account characteristics are simulated, so that the generated data has high representativeness and authenticity.

Description

Technical Field

[0001] This invention relates to the field of multi-source data processing technology, specifically to a method for generating multi-source data fusion for monitoring the risks of illegal fundraising. Background Technology

[0002] New types of illegal fundraising activities pose significant challenges to financial regulation due to their high degree of concealment, complex behavior, and cross-regional nature. Currently, researchers face considerable difficulties in risk monitoring and behavioral identification, primarily due to the following reasons: First, limited data access: Real financial data cannot be publicly disclosed due to privacy protection and legal compliance requirements, and competition among financial institutions restricts data sharing. Second, incomplete labeling: Even when partial data is obtained, many illegal fundraising activities are not promptly identified and labeled, making it difficult to provide effective monitoring signals for models. Third, complex transaction structures: Diverse cross-account and cross-regional fund flow patterns increase the difficulty of behavioral analysis and identification.

[0003] Compared to the fields of anti-money laundering and anti-fraud, datasets for emerging illegal fundraising are almost nonexistent. While anti-money laundering projects such as AMLSim and AMLworld offer some insights into data generation, their scenarios are limited to money laundering activities and cannot be directly applied to the complex patterns of illegal fundraising, thus failing to fully reflect the complexity and diversity of illegal fundraising activities.

[0004] Existing technologies for monitoring illicit financial activities rely on real transaction data and traditional machine learning methods for 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, severely impacting identification and predictive performance. To compensate for the lack of real data, some research has attempted data generation methods based on proxy simulation, such as the AMLworld project in the anti-money laundering field. While these methods have made some progress in enhancing data diversity, they lack in-depth modeling of the unique patterns of illicit fundraising activities, resulting in significant deficiencies in characterizing key features such as fund flow paths, cross-border transaction frequency, and account correlations. Furthermore, existing generation methods are highly random and often lack pre-defined rules, leading to a disconnect between the generated data and real financial behavior patterns, making it difficult to provide models with meaningful risk feature extraction capabilities. These issues limit the applicability of existing technologies in illicit fundraising risk monitoring and fail to meet the field's needs for highly complex and diverse pattern analysis.

[0005] As mentioned above, existing financial data is difficult to obtain due to restrictions such as privacy protection and legal compliance. Furthermore, incomplete labeling and information isolation result in a severe lack of data support for researchers in monitoring the risks of illegal fundraising, particularly in the area of ​​emerging illegal fundraising schemes. Traditional datasets are insufficient to fully reflect the complexity and diversity of illegal fundraising activities, and relying solely on a small amount of real data is inadequate to cover the ever-changing patterns of illegal fundraising. Therefore, there is an urgent need to design a data generation method that incorporates the characteristics of illegal fundraising activities and construct a high-quality dataset suitable for this field. Summary of the Invention

[0006] This invention is made to solve the above-mentioned problems, and its purpose is to provide a method for generating high-quality datasets related to illegal fundraising. The invention adopts the following technical solution:

[0007] This invention provides a multi-source data fusion generation method for monitoring the risks of illegal fundraising. The method includes the following steps: Step S1, constructing a rule set containing multiple rules, each rule representing a typical pattern of illegal fundraising behavior, and each rule including fund flow characteristics, related account characteristics, and cross-border transfer characteristics; Step S2, constructing multiple simulated accounts; Step S3, generating simulated transaction data based on the multiple simulated accounts and the rule set; Step S4, injecting noise into the simulated transaction data; Step S5, setting corresponding labels for the noise-injected simulated transaction data to identify data that conforms to the characteristics of illegal fundraising, thereby constituting an illegal fundraising-related dataset.

[0008] The multi-source data fusion generation method for monitoring the risk of illegal fundraising provided by the present invention may also have the following technical features: the fund flow feature is used to define the transfer frequency and transfer amount 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; and the cross-border transfer feature is used to define the fund flow rules between different countries or regions.

[0009] The multi-source data fusion generation method for monitoring the risk of illegal fundraising provided by this invention may also have the following technical feature, wherein the rule set R is represented as:

[0010] R = {f1, f2, ..., f n}

[0011] In the formula, f i Each rule represents a feature rule, and the feature rules in each rule set R collectively describe a type of illegal fundraising behavior pattern.

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

[0013] The multi-source data fusion generation method for monitoring the risk of illegal fundraising provided by the present invention may also have the following technical features: the initial balance of the multiple simulated accounts is distributed normally, 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 monitoring the risk of illegal fundraising provided by this invention may also have the following technical features, wherein step S3 includes the following sub-steps: Step S3-1, simulating transaction behavior using the corresponding simulated account according to the rule set 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 Steps S3-4: According to the selected rule R i The feature definition modifies the transaction behavior between the simulated accounts, including setting the direction of fund flow, transfer frequency, and transfer amount; step S3-5, determines whether the current time step is the last time step. If the determination is no, proceed to the next time step and repeat steps S3-3 to S3-4; if the determination is yes, obtain the simulated transaction data D containing the characteristics of illegal fundraising behavior. ′ .

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

[0016] The multi-source data fusion generation method for monitoring the risk of illegal fundraising provided by this invention may also have the following technical features: the noise injection method includes: a random injection method, randomly generating several normal transaction behaviors in the generated simulated transaction data; and an anomaly point generation method, adding abnormal transactions to a specific simulated account, the abnormal transactions including large transfers and frequent transactions. The simulated transaction data after noise injection is represented as follows:

[0017] D″=D ′ +∈

[0018] In the formula, D ′ Let be the simulated transaction data, and ∈ be the noise vector whose value is generated according to the normal distribution N(0,σ).

[0019] The multi-source data fusion generation method for monitoring the risk of illegal fundraising provided by this invention may also have the following technical feature, wherein, in step S5, the tag is represented as:

[0020]

[0021] In the formula, a label of 1 indicates that the sample contains characteristics of illegal fundraising, while a label of 0 indicates that the sample does not contain characteristics of illegal fundraising and belongs to normal transaction behavior.

[0022] The role and effect of invention

[0023] The multi-source data fusion generation method for monitoring illegal fundraising risks provided by this invention 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, noise injection, and label setting. By fusing multi-source data, employing a rule-based generation strategy, and using noise injection technology, high-quality, near-realistic data on illegal fundraising activities can be generated while ensuring data privacy and security. This effectively avoids privacy and compliance issues associated with acquiring real financial data, providing researchers with a rich and secure data source. Furthermore, since each rule in the rule set represents a typical pattern of illegal fundraising activity and includes characteristics of fund flow, related accounts, and cross-border transfers, the generated dataset closely resembles the characteristics of real illegal fundraising activities in terms of content and structure. The fusion of multi-source data and noise injection simulate complex fund flows and related account characteristics, resulting in highly representative and realistic data.

[0024] The method of this invention can provide more accurate training data for illegal fundraising risk monitoring models, which can significantly improve the accuracy of downstream models in abnormal behavior detection, risk identification and other aspects. Attached Figure Description

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

[0026] Figure 2 This is a flowchart of a multi-source data fusion generation method for monitoring the risk of illegal fundraising in this embodiment of the invention;

[0027] Figure 3 This is a flowchart of step S3 in an embodiment of the present invention. Detailed Implementation

[0028] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following describes in detail the multi-source data fusion generation method for monitoring the risk of illegal fundraising.

[0029] Figure 1 This is a schematic diagram illustrating the key steps of the multi-source data fusion generation method for monitoring the risk of illegal fundraising in this embodiment. Figure 2 This is a flowchart of the multi-source data fusion generation method for monitoring the risk of illegal fundraising in this embodiment.

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

[0031] Step S1: Construct a rule set, which contains multiple rules, each representing a typical pattern of illegal fundraising behavior.

[0032] Step S2: Create multiple demo accounts.

[0033] Step S3: Generate simulated trading data based on multiple demo accounts and a set of rules.

[0034] Step S4: Inject noise into the simulated transaction data.

[0035] Step S5: Set corresponding labels for the simulated transaction data after injecting noise, identify the data that meets the characteristics of illegal fundraising, and thus form a dataset related to illegal fundraising.

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

[0037] Step S1: Construct a rule set, which contains multiple rules, each representing a typical pattern of illegal fundraising behavior.

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

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

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

[0041] R = {f1, f2, ..., f n}

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

[0043] Step S2: Construct multiple demo accounts and use these demo accounts to conduct simulated trading, thereby generating corresponding initial simulated trading data.

[0044] In step S2, basic parameters and settings are defined for the data generation process. By setting the number of demo accounts, as well as the basic information, initial balance, and trading frequency of each demo account, including account name, account number, account type, and country of origin, a foundation is laid for generating demo trading data that conforms to the rule set.

[0045] Specifically, in step S2, the total number of simulated accounts is first set, which can be based on the scale of the actual illegal fundraising activity. Then, an initial balance is allocated to each simulated account, using either a random distribution or a normal distribution to simulate the diversity of funds in different accounts. Next, the transaction frequency for each simulated account is defined, which can be adjusted according to account type or risk level. For example, the total number of simulated accounts is set to 1000, the initial balance of each simulated account is distributed according to a normal distribution N(μ,σ), and the transaction frequency includes three levels: high, medium, and low. Each simulated account is randomly assigned to one of these transaction frequency levels.

[0046] Step S3: Generate simulated trading data based on multiple demo accounts and a set of rules.

[0047] In step S3, a predefined set of rules is applied to simulate and transform account data to generate transaction data that conforms to the characteristics of illegal fundraising. This step ensures that the generated data samples contain illegal fundraising behavior patterns, facilitating subsequent model training and testing.

[0048] Specifically, firstly, based on the rule set, simulated transactions are conducted using corresponding demo accounts to generate initial simulated transaction data. Then, rules are selected from the rule set, and the initial simulated transaction data is transformed based on the selected rules to obtain simulated transaction data that conforms to the characteristics of illegal fundraising.

[0049] Figure 3 This is a flowchart of step S3 in this embodiment.

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

[0051] Step S3-1: Based on the rule set, simulate trading behavior using the corresponding demo account to generate initial simulated trading data.

[0052] That is, based on the characteristics of related accounts and cross-border transfers in the rules, multiple sets of simulated accounts that can be traded are found, and one or more simulated transaction data are generated between each set of simulated accounts according to the corresponding fund flow characteristics. Each data includes information such as the direction of fund flow (sender and recipient), transfer time, and transfer amount.

[0053] Step S3-2: Set 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] Steps S3-4, according to the selected rule R i The feature definition modifies the transaction behavior between simulated accounts, including setting the direction of fund flow, transfer frequency, and transfer amount, thereby obtaining the transformed data D. ′ The transformed data D ′ It contains characteristics of illegal fundraising activities.

[0056] Step S3-5: Determine whether the current time step is the last time step. If the determination is no, proceed to the next time step and repeat steps S3-3 to S3-4. If the determination is yes, enter the end state and obtain simulated transaction data containing the characteristics of illegal fundraising behavior.

[0057] Step S4: Inject noise into the simulated transaction data.

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

[0059] Specifically, firstly, the proportion of noisy data in the dataset is set, for example, 5% or 10%. Then, according to this proportion, a predetermined noise injection method is used to inject noisy data into the generated simulated transaction data. In this embodiment, a random injection method and anomaly generation method are used. The random injection method specifically refers to randomly generating some normal transaction behaviors in the generated simulated transaction data to obscure the characteristics of illegal fundraising. The anomaly generation method specifically refers to adding abnormal transactions, such as large transfers or frequent transactions, to specific simulated accounts to simulate false alarms and normal data fluctuations.

[0060] The formula for noise injection is expressed as follows:

[0061] D″=D ′ +∈

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

[0063] Step S5: Set corresponding labels for the simulated transaction data after injecting noise, identify data that meets the characteristics of illegal fundraising, and thus form a dataset related to illegal fundraising.

[0064] In step S5, labels are added to the generated dataset to identify which data matches the characteristics of illegal fundraising. These labels are used for subsequent model training and evaluation, helping the model identify illegal fundraising patterns and risk categories, and can be represented as:

[0065]

[0066] In the formula, a label of 1 indicates that the sample contains characteristics of illegal fundraising, while a label of 0 indicates that the sample does not contain characteristics of illegal fundraising and belongs to normal transaction behavior.

[0067] The specific label generation process can involve automatic labeling based on application rule judgment conditions, or assisted labeling through expert review. The final generated label dataset D... L ={D i Label(D i This is used for subsequent model training.

[0068] The role and effect of the embodiments

[0069] The multi-source data fusion generation method for monitoring illegal fundraising risks provided in this embodiment 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, noise injection, and label setting. By fusing multi-source data, employing a rule-based generation strategy, and using noise injection technology, high-quality, near-realistic illegal fundraising behavior data can be generated while ensuring data privacy and security. This effectively avoids privacy and compliance issues associated with acquiring real financial data, providing researchers with a rich and secure data source. Furthermore, since each rule in the rule set represents a typical pattern of illegal fundraising behavior and includes characteristics of fund flow, related accounts, and cross-border transfers, the generated dataset closely resembles the characteristics of real illegal fundraising behavior in terms of content and structure. The fusion of multi-source data and noise injection simulate complex fund flows and related account characteristics, resulting in highly representative and realistic data. This method provides more accurate training data for illegal fundraising risk monitoring models, significantly improving the accuracy of downstream models in abnormal behavior detection and risk assessment.

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

[0071] Furthermore, the noise injection employs two methods: random injection and outlier generation. This not only incorporates normal transaction data but also common abnormal data such as large-amount transactions and frequent transactions. Therefore, it not only blurs the illegal fundraising characteristics in the dataset but also enriches the data, making the generated simulated transaction data closer to real financial transaction data. This allows the model trained using this dataset to also achieve good predictive performance for real transaction data.

[0072] The above embodiments are merely illustrative of specific implementations of the present invention, and the present invention is not limited to the scope of the description of the above embodiments. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are only for illustrating the principles of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating multi-source data fusion for monitoring the risk of illegal fundraising, characterized in that, Includes the following steps: Step S1: Construct a rule set, which contains multiple rules. Each rule represents a typical pattern of illegal fundraising behavior. Each rule includes characteristics of fund flow, characteristics of related accounts, and characteristics of cross-border transfers. Step S2: Create multiple demo accounts; Step S3: Generate simulated trading data based on multiple demo accounts and a set of rules; Step S4: Inject noise into the simulated transaction data; Step S5: Label the noise-injected simulated transaction data accordingly, identifying data that matches the characteristics of illegal fundraising, thus forming a dataset related to illegal fundraising. Wherein, the rule set Represented as: ， In the formula, Each rule set represents a feature rule. The characteristic rules in the text collectively describe a pattern of illegal fundraising activities. Step S3 includes the following sub-steps: Step S3-1: Based on the rule set, simulate transaction behavior using the corresponding demo account to generate initial simulated transaction data; Step S3-2: Set the total number of time steps; Step S3-3: In the t-th time step, randomly select the applicable rule from the rule set. ; Step S3-4, according to the selected rule The feature definition modifies the transaction behavior between the simulated accounts, including setting the direction of fund flow, transfer frequency, and transfer amount; Step S3-5: Determine whether the current time step is the last time step. If the determination is no, proceed to the next time step and repeat steps S3-3 to S3-4. If the determination is yes, obtain the simulated transaction data containing characteristics of illegal fundraising activities. .

2. The multi-source data fusion generation method for monitoring the risk of illegal fundraising according to claim 1, characterized in that: in, The fund flow characteristics are used to define the transfer frequency and transfer amount 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 features are used to define the rules for fund flows between different countries or regions.

3. The multi-source data fusion generation method for monitoring the risk of illegal fundraising according to claim 2, characterized in that: in, In step S2, the number of the demo accounts is set, and the parameters of each demo account are set, including basic information, initial balance, and transaction frequency.

4. The multi-source data fusion generation method for monitoring the risk of illegal fundraising according to claim 3, characterized in that: in, The initial balances of the multiple simulated accounts are distributed according to a normal distribution. The trading frequency includes three levels: high, medium, and low, and each of the simulated accounts is randomly assigned to one of these levels.

5. The multi-source data fusion generation method for monitoring the risk of illegal fundraising according to claim 1, characterized in that: in, In step S4, the proportion of noise data in the dataset is set, and noise data is injected into the generated simulated transaction data according to the proportion using a predetermined noise injection method to simulate abnormal situations or randomness in real financial transactions.

6. The multi-source data fusion generation method for monitoring the risk of illegal fundraising as described in claim 5, Its features are: The noise injection method includes: The random injection method randomly generates several normal trading behaviors in the generated simulated trading data; and The method for generating anomalies involves adding abnormal transactions to the demo account. These abnormal transactions include large transfers and frequent transactions. The simulated transaction data after noise injection is represented as follows: ， In the formula, The simulated transaction data, This is a noise vector whose values ​​follow a normal distribution. generate.

7. The multi-source data fusion generation method for monitoring the risk of illegal fundraising according to claim 1, characterized in that: in, In step S5, the label is represented as: ， In the formula, a label of 1 indicates that the sample contains characteristics of illegal fundraising behavior, while a label of 0 indicates that the sample does not contain characteristics of illegal fundraising behavior and belongs to normal transaction behavior.

Images