Financial data generation method oriented to illegal capital collection transaction risk scene

Abstract

The invention provides a financial data generation method for an illegal capital transaction risk scene, and the method comprises the steps: generating user basic data, generating a user relation network, and generating normal transaction data and multi-scene illegal capital transaction data between users. And on the basis, a relation network among the multiple users is constructed, so that real social and economic characteristics of the multiple users can be well simulated, diversified user basic data with strong sense of reality and a social relation network close to reality are provided, and a solid foundation is provided for simulation of transaction data. Furthermore, based on the user basic information, the user relationship network, the statistical data and the predetermined generation rule, the normal transaction data and the illegal fund collection transaction data of various illegal fund collection transaction scenes are respectively generated, i.e., the transaction data with a complex transaction mode can be automatically generated. And powerful support is provided for applications such as financial model testing and fraud detection.

Description

Technical Field

[0001] The present invention relates to the technical field of table data generation, and in particular to a financial data generation method for illegal fund-raising transaction risk scenarios. Background Art

[0002] With the increasingly stringent data privacy protection regulations, it has become extremely difficult to obtain and use the financial transaction data of real users. Especially in the financial field, user transaction data often involves sensitive personal privacy and is protected by strict legal constraints and privacy policies. Financial institutions cannot provide or disclose this data. In this context, financial institutions and researchers generally face the problem of difficulty in obtaining data when conducting data analysis, model training and fraud detection in the financial field, which greatly reduces the efficiency of risk monitoring and fraud detection. Although there are some privacy computing technologies, such as homomorphic encryption, federated learning and differential privacy, which can ensure privacy protection during data exchange and analysis to a certain extent and ensure that data is calculated without being exposed, the core problem of not being able to obtain original data has not been solved. In many cases, even if these technologies can ensure the security and privacy of data during the exchange process, financial institutions and other relevant parties are usually reluctant to disclose or share real transaction data due to concerns about privacy leakage, data abuse or compliance issues. Therefore, when conducting large-scale and complex financial data analysis, data acquisition still faces great challenges.

[0003] At present, although there are data synthesis methods, which are mainly based on generative models to synthesize synthetic data that is close to real data in features, such as generative adversarial networks, variational autoencoders, diffusion models, etc., these methods have achieved certain results in specific application scenarios, but in these methods, the generative models used still rely on a large amount of real original data for training. In addition, when generating data, the model often finds it difficult to fully capture the complexity of the actual scenario, especially in the field of financial transactions. Real data contains complex behavioral patterns, multi-level network relationships, and nonlinear associations, and these details are usually not well represented in the data synthesized by existing data synthesis methods. In the research on illegal fund-raising risk detection, due to the above-mentioned problems of the existing data synthesis methods, it is difficult for synthetic data to reflect a variety of illegal fund-raising transaction models, which affects the efficiency of the research and the accuracy of the corresponding detection methods designed.

[0004] Therefore, in response to the above problems and shortcomings, there is an urgent need for a method to synthesize illegal fundraising transaction data that does not rely on real data to generate high-quality transaction data, thereby assisting in modeling research on illegal fundraising risk detection. Summary of the invention

[0005] The present invention is made to solve the above problems, and aims to provide a financial data synthesis method that does not rely on real data and can synthesize complex behavior patterns, multi-level network relationships and non-linear associations. The present invention adopts the following technical solutions:

[0006] The present invention provides a financial data generation method for illegal fund-raising transaction risk scenarios, the method has the following technical features, which include the following steps: step S1, generating user basic data of multiple users; step S2, constructing a user relationship network of the multiple users based on the user basic data; step S3, generating normal transaction data between the users based on the user basic information, the user relationship network, statistical data and predetermined normal transaction generation rules; step S4, generating illegal fund-raising transaction data for multiple illegal fund-raising transaction scenarios between the users based on the user basic information, the user relationship network, statistical data and predetermined multiple illegal fund-raising transaction generation rules, wherein the user basic data of each user includes bank card information and user information, the user information at least includes age, income level, and occupational group, which are generated respectively according to corresponding statistical data, and the user relationship network includes multiple nodes and edges between the nodes, the nodes are the users, and the edges are the relationships between the users.

[0007] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such technical features, wherein the bank card information includes the bank card number, the bank to which it belongs, and the balance, and the user information also includes the name, registration date, gender, and suspicious object mark. In step S1, for the bank card information, the bank card number is generated by a unique identifier, and the balance is randomly set within a predetermined numerical range according to the corresponding income level. For the user information, the income level and the occupational group are allocated based on the corresponding statistical data and in proportion to the age distribution.

[0008] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such a technical feature, wherein the user relationship network includes multiple nodes and edges between the nodes, the nodes are the users, the edges are the relationships between the users, the types of relationships include parent relationships, relative relationships, and friend relationships, the friend relationships include general friend relationships and better friend relationships, and in step S2, an initial user relationship network is generated based on the user basic data and the types of the relationships and is dynamically updated.

[0009] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such a technical feature, wherein, in step S2, for the friend relationship, based on the age difference between the target user among the users and the user, the potential friend group of the target user is determined; two groups of predetermined numbers of the users are randomly selected from the potential friend group, and are respectively set as friends with general relationships and friends with good relationships with the target user, and the bidirectionality of the friend relationship is ensured; a predetermined number of the users from the friends of the friends of the target user are selected, and are also set as the friends of the target user; for the relative relationship, a predetermined number of the users are randomly selected for the target user, and are set as the relatives of the target user, and the bidirectionality of the relative relationship is ensured; for the parent relationship, based on the age difference between the target user and the user and a predetermined logical rule, the user is selected for the target user and is set as the father or mother, and the uniqueness of the parent relationship is ensured.

[0010] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such technical features, wherein, in step S3, statistical analysis is performed based on the user basic data of multiple users to obtain characteristic data; the user group to which the multiple users belong is determined based on the characteristic data; the normal transaction generation rules for the user group are generated based on the determined user group and predetermined multiple transaction data characteristic information corresponding to the user group; and the normal transaction data between the multiple users are generated based on the user basic data and the normal transaction generation rules.

[0011] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such technical features, wherein the transaction data characteristic information includes transaction object range, transaction time range, transaction amount range, transaction frequency range, and transaction intention range, which are set according to the age, income level, and occupational group of the users in the user group.

[0012] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such technical features, wherein, in step S4, based on the operation mode in the illegal fund-raising transaction scenario, multiple users with relationship edges are selected from the user relationship network and an illegal fund-raising relationship network is constructed; the illegal fund-raising transaction generation rules are generated based on the transaction data characteristic information in the illegal fund-raising transaction scenario; based on the user basic data of the users in the illegal fund-raising relationship network and the illegal fund-raising transaction generation rules, the illegal fund-raising transaction data between the multiple users are generated.

[0013] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such a technical feature, wherein the illegal fund-raising transaction scenario is a pyramid scheme-style illegal fund-raising transaction scenario. In step S4, according to the unique operating mode of the pyramid scheme-style illegal fund-raising transaction scenario, one of the users is randomly selected from the user relationship network as a top-level user, one or more are selected from the friends, relatives, and parents of the top-level user as the first-level subordinate users of the top-level user, and a superior-subordinate relationship edge is formed between the top-level user and the first-level subordinate users, one or more are selected from the friends, relatives, and parents of one or more of the first-level subordinate users as second-level subordinate users, and a superior-subordinate relationship edge is formed between the first-level subordinate users and the corresponding second-level subordinate users, and the above process is repeated to form a multi-level pyramid-style illegal fund-raising relationship network.

[0014] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such technical features, wherein the transaction data characteristic information includes a transaction object range, a transaction time range, a transaction amount range, and a transaction frequency range, and is set according to the operation mode of the pyramid scheme-style illegal fund-raising transaction scenario. The transaction object range includes the user's superior users and subordinate users, and the ratio range of the number of payment and collection transactions between the user and the superior user, and between the user and the subordinate user is set respectively; the transaction frequency range is once to several times per week or month.

[0015] The financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention may also have such technical characteristics, wherein the illegal fund-raising transaction scenarios include pyramid scheme-style illegal fund-raising transaction scenarios, virtual currency illegal fund-raising transaction scenarios, real estate illegal fund-raising transaction scenarios, pension project illegal fund-raising transaction scenarios, and Internet financial platform illegal fund-raising transaction scenarios.

[0016] Functions and Effects of the Invention

[0017] According to the financial data generation method for illegal fund-raising transaction risk scenarios provided by the present invention, the method includes the steps of generating user basic data, generating user relationship networks, and further generating normal transaction data between users and multiple modes of illegal fund-raising transaction data. In particular, since user information is generated based on statistical data, including the user's age, income level, and occupational group, and a relationship network between multiple users is constructed on this basis, it is possible to simulate the real socioeconomic characteristics of multiple users, provide diversified and realistic user basic data and a social relationship network that is closer to reality, and provide a solid foundation for the simulation of transaction data. Furthermore, based on user basic information, user relationship networks, statistical data, and predetermined generation rules, normal transaction data and illegal fund-raising transaction data for multiple illegal fund-raising transaction scenarios are generated respectively, that is, transaction data with complex transaction patterns can be automatically generated, thereby providing strong support for applications such as financial model testing and fraud detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of a method for generating financial data for illegal fund-raising transaction risk scenarios in an embodiment of the present invention;

[0019] Figure 2 is a flow chart of step S2 in an embodiment of the present invention;

[0020] Figure 3 is a schematic diagram of a user data relationship network constructed in an embodiment of the present invention;

[0021] Figure 4 is a flow chart of step S3 in an embodiment of the present invention;

[0022] Figure 5 is a flow chart of step S4 in an embodiment of the present invention;

[0023] Figure 6 It is a schematic diagram of a pyramid scheme illegal fund-raising scenario constructed in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the financial data generation method for illegal fund-raising transaction risk scenarios of the present invention is specifically described below in combination with embodiments and drawings.

[0025] <Example>

[0026] The present invention provides a financial data generation method for illegal fund-raising transaction risk scenarios, which aims to simulate transaction data of illegal fund-raising activities so that the transaction data can be used for applications such as financial risk model testing, fraud detection and consumer behavior analysis.

[0027] Figure 1 It is a flowchart of the financial data generation method for illegal fund-raising transaction risk scenarios in this embodiment.

[0028] like Figure 1 As shown, the method comprises the following steps:

[0029] Step S1, generating basic user data of multiple users.

[0030] Step S2: construct a user relationship network of multiple users based on the user basic data.

[0031] Step S3, generating normal transaction data between the multiple users based on user basic data, user relationship network, statistical data of each group and predetermined normal transaction generation rules.

[0032] Step S4, generating illegal fund-raising transaction data among the multiple users based on user basic data, user relationship network, statistical data of each group and predetermined illegal fund-raising transaction generation rules.

[0033] The above steps will be described in detail below.

[0034] Step S1, generating basic user data of multiple users.

[0035] Among them, user basic data includes bank card information and user information (cardholder information).

[0036] Bank card information includes bank card ID (i.e., bank card unique identification number), bank to which the bank card belongs, and balance. The bank card ID is generated through a unique identifier to ensure that all users' bank IDs are unique and non-repeated. The bank to which the bank card belongs is randomly set. The balance is randomly set within the corresponding numerical range based on the income level in the cardholder information.

[0037] User information includes name, age, registration date, gender, income level, occupational group, and suspicious object mark. Among them, the age is randomly set within the range of 18 to 60 years old, or is allocated and set based on the proportion of people in each age group in the corresponding statistical data. The income level and occupational group are allocated according to the proportion of age distribution based on the corresponding statistical data. For example, the occupational group of 18 to 25 years old includes students, new employees, freelancers, etc. The suspicious object mark is used to mark whether the user is a suspicious object related to illegal fundraising transactions. A mark of 0 indicates that the user is a whitelist user and is not a suspicious object; a mark of 1 indicates that the user is a suspicious object.

[0038] The user basic data generated in the above manner can simulate the real socioeconomic characteristics of multiple users, provide diversified and realistic user basic data, and provide a basis for subsequent transaction simulations.

[0039] Table 1 below shows the relevant fields of user basic information.

[0040] Table 1 User basic information field table

[0041]

[0042] Table 2 below shows a data sample based on the above field settings.

[0043] Table 2 User basic information data sample table

[0044]

[0045]

[0046] Step S2: construct a user relationship network of multiple users based on the basic user information.

[0047] Among them, relationship types include friends, relatives and parents, among which friends are further divided into friends with general relationships and friends with better relationships.

[0048] In step S2, an initial user relationship network is generated based on the user basic data and the above relationship types and is dynamically updated to ensure the authenticity of the relationships between users and the diversity of the network structure.

[0049] Figure 2 is a flow chart of step S2 in this embodiment.

[0050] like Figure 2 As shown, in step S2, generating the initial user relationship network specifically includes the following sub-steps:

[0051] Step S2-1, for friendship, determine the target user's potential friend group based on the age difference between the target user and other users. For example, for a target user, multiple other users whose age difference is less than a predetermined threshold are considered as the target user's potential friend group.

[0052] Step S2-2, randomly selecting two groups of predetermined number of users from the target user's potential friend group, setting them as the target user's average friends and good friends respectively, and ensuring the two-way nature of the friendship.

[0053] Step S2-3: For the target user, a predetermined number of users among the friends of the target user are selected and set as the target user's friends, thereby further expanding the friendships in the user relationship network and enhancing the connectivity and diversity of the user relationship network.

[0054] Step S2-4: For the kinship relationship, a predetermined number of other users are randomly selected as the target user's relatives, and the bidirectionality of the kinship relationship is ensured.

[0055] Step S2-5, for the parent relationship, according to the age difference between the target user and other multiple users and the predetermined logic rules, other users are selected as the father and / or mother for the target user, and the uniqueness of the parent relationship is ensured.

[0056] It is understandable that the above operation may be performed on each user separately, or some of the users may be selected to perform the above operation.

[0057] Figure 3 Schematic diagram of the user relationship network constructed in this embodiment.

[0058] like Figure 3 As shown, the user relationship network can be represented as a graph structure, where nodes are users and edges are relationships between users. For a user A, through the above steps, user B is selected as his father / mother, and a corresponding relationship edge is formed between A and B; user C is selected as his relative and a corresponding relationship edge is formed; users D and E are selected as his friends, and user D is set as his average friend, and user E is set as his good friend, and a corresponding relationship edge is formed. It can be understood that the above relationship objects can also be selected for users B~E in the same way, thereby forming a complex user relationship network.

[0059] The user relationship network generated in the above way conforms to the characteristics of real social networks in structure, and provides a solid foundation for the subsequent transaction data simulation of illegal fund-raising scenarios.

[0060] Step S3, generating normal transaction data between the multiple users based on user basic information, user relationship network, statistical data of each group and predetermined normal transaction generation rules.

[0061] Figure 4 is a flow chart of step S3 in this embodiment.

[0062] like Figure 4 As shown, step S3 specifically includes the following sub-steps:

[0063] Step S3-1, for the selected multiple users, statistical analysis is performed based on their user basic data to obtain characteristic data.

[0064] For example, statistical analysis is performed on the ages of multiple users to obtain their age groups.

[0065] Step S3-2: determining the user group to which the multiple users belong from multiple predetermined user groups according to the feature data.

[0066] For example, based on the age groups obtained through statistical analysis, it is determined that these multiple users are a student group.

[0067] Step S3-3: generating normal transaction generation rules for the user group based on the determined user group and predetermined multiple transaction data characteristic information corresponding to the user group.

[0068] For example, for each user group, the transaction object range, transaction time range, transaction amount range, transaction frequency range, transaction intention range, etc. corresponding to the user group are pre-stored respectively, and the transaction generation rules of the user group are generated based on these data ranges. For example, for the student group, the transaction object range is friends, the transaction time range is weekday evenings or weekends, the transaction amount range is less than 500 yuan, the transaction frequency range is 10 to 30 times per month, and the transaction intention range includes repayment, mobile payment consumption, etc. Based on these, the corresponding transaction generation rules are generated, such as randomly selecting one of the user's friends as the transaction object, randomly selecting a time point within the optional time range, and randomly selecting a value within the transaction amount range, etc.

[0069] Step S3-4, generating normal transaction data between the selected multiple users according to the generated normal transaction generation rules and user basic data.

[0070] For example, let's take users aged 18 to 25 as the target group. For this group, the overall data characteristics are as follows:

[0071] Transfer characteristics: Since users in this age group are mainly students or just entering the workplace, their income is limited and their expenses are mostly concentrated on daily consumption and social activities. Transfers between friends are mostly used to share the expenses of parties, dinners or social activities, while the funds between family members are relatively rare and are more independent financial transactions. Therefore, the characteristics of this user group are frequent small transfers, which are mainly used for daily consumption and social activities. Transfers mostly occur between friends and rarely involve family members.

[0072] Transfer time characteristics: Since students and young professionals usually socialize on weekends or evenings, such as having dinner, watching movies, or participating in group activities, transfers often occur during these time periods, coinciding with their social time. Therefore, the consumption of this group is mostly concentrated on weekends and evenings, in line with the time of social activities.

[0073] Amount characteristics: Since the income of this age group is generally low and living expenses are relatively limited, the amount required for daily expenses such as dining, online shopping and small loans is usually not large, basically within a few hundred yuan. Therefore, the single transfer amount of consumption of this group is small, usually within a few hundred yuan.

[0074] Frequency characteristics: Due to frequent social activities and daily consumption, users in this age group transfer money more frequently each month, especially AA dinners and repayments, which make transfers a part of daily life. Therefore, this group has a high consumption frequency, with multiple transfers per month on average.

[0075] Transfer intention: Young people usually use the AA system to share party and social expenses. In addition, takeout and online shopping are also common consumption methods, and small loans are usually used for urgent financial needs. Therefore, the consumption of this group is mainly used for AA party, repayment, mobile payment consumption (such as online shopping and takeout) and small loans. In addition, professional groups also include students, people who just entered the workplace, and freelancers.

[0076] Taking the student group in the 18-25 age group as an example, the transfer behavior of the student group has significant characteristics, mainly manifested in small and frequent transfers for daily consumption, study expenses and social activities. The transfer time is mainly concentrated on weekends and evenings, which is in line with the students' work and rest characteristics. The amount of a single transfer is usually small, ranging from 50 yuan to 200 yuan, but the overall frequency is high, with multiple transfers occurring each week. The main uses of these transfers include dinner parties, activity cost sharing, online shopping, takeout, and tuition payment related to study, reflecting the unique consumption needs and social habits of the student group.

[0077] In the generation of normal transaction data, the design is based on the characteristics of the user's occupation, age group and income level, and the transfer mode, time distribution, amount characteristics, transaction frequency and transaction intention of different occupational groups are combined to generate transaction data with high authenticity and representativeness. For example, for the student group, as mentioned above, their transfer behavior is mainly AA consumption. To simulate this transaction scenario, a specific time is selected, mainly concentrated on weekends and evenings, and a user is selected as a student, and some of the user's friends are selected to generate consumption behaviors related to dinners or online shopping. In this scenario, multiple friends of the student transfer the same amount to the student within a few hours, thereby simulating the characteristics of AA consumption of the student group in real scenarios.

[0078] Step S4, generating illegal fund-raising transaction data among the multiple users based on user basic information, user relationship network, statistical data of each group and predetermined illegal fund-raising transaction generation rules.

[0079] Among them, illegal fund-raising transactions include pyramid scheme-style illegal fund-raising, virtual currency illegal fund-raising, real estate illegal fund-raising, pension project illegal fund-raising, and Internet financial platform illegal fund-raising. The most representative of these is pyramid scheme-style illegal fund-raising.

[0080] Figure 5 It is a flow chart of step S4 in this embodiment.

[0081] like Figure 5 As shown, step S4 specifically includes the following sub-steps:

[0082] Step S4-1, based on the operation mode in the illegal fund-raising transaction scenario, multiple users with relationship edges are selected from the user relationship network and an illegal fund-raising relationship network is constructed.

[0083] For example, in a pyramid-style illegal fundraising transaction scenario, according to its unique operating mode, a user is randomly selected from the user relationship network as the top-level user, and one or more of the top-level user's friends, relatives, and parents are selected as the top-level user's first-level subordinate users, forming a superior-subordinate relationship edge between the top-level user and the first-level subordinate user. Then, one or more of the first-level subordinate users' friends, relatives, and parents are selected as the first-level subordinate users, that is, the second-level subordinate users of the top-level user, forming a superior-subordinate relationship edge between the first-level subordinate user and the corresponding second-level subordinate user. Repeat the above process to generate a multi-level pyramid-style illegal fundraising relationship network.

[0084] Step S4-2, generating illegal fund-raising transaction generation rules based on the transaction data characteristic information in the illegal fund-raising transaction scenario.

[0085] For example, for each illegal fund-raising transaction scenario, there are pre-stored transaction object ranges, transaction time ranges, transaction amount ranges, transaction frequency ranges, etc. corresponding to the illegal fund-raising transaction scenarios, and transaction generation rules between multiple users in the illegal fund-raising relationship network in this scenario are generated based on these data ranges.

[0086] For example, in the pyramid scheme illegal fund-raising transaction scenario, the transaction objects are the user's superior and subordinate users, and the superior user is mainly for making payments, while the subordinate user is mainly for receiving payments (for example, the ratio range of the amount of payments made / received is set). There is no limit on the transaction time range, the transaction amount range is 500 to 9,000 yuan, the transaction frequency range is once to several times a week, or once to several times a month and reflects periodicity, etc. Illegal fund-raising transaction generation rules are generated based on these predetermined data ranges.

[0087] Step S4-3, based on the user basic data of the users in the illegal fundraising relationship network and the illegal fundraising transaction generation rules, generate the illegal fundraising transaction data among the multiple users.

[0088] For example, let’s take the most representative pyramid scheme-style illegal fundraising as an example to provide a specific explanation.

[0089] This model usually uses quick and high returns as bait to achieve illegal fundraising through multi-level personnel development and capital rolling. For this type of illegal fundraising, the overall data characteristics are as follows:

[0090] Transfer characteristics: Participants in pyramid schemes will continuously transfer money to their superiors. At the same time, the transfer relationship is often more complicated, involving multiple accounts and middlemen, and intersecting paths.

[0091] Transfer time characteristics: periodic transfers on a monthly or weekly basis.

[0092] Amount characteristics: The amount usually ranges from a few hundred yuan to a few thousand yuan, and presents a pyramid structure. The upper-level users mainly have capital inflows with large amounts, while the lower-level users mainly have capital outflows with relatively small amounts.

[0093] Frequency characteristics: The upper-level users mainly have capital inflows, and the inflow quantity (number of transactions) is large, while the lower-level users mainly have capital outflows, and the outflows are frequent.

[0094] Therefore, based on the above data characteristics of pyramid-style illegal fund-raising, the top-level users are first randomly selected, and a three-layer pyramid structure is constructed based on their relationships with friends, relatives, and parents. The top-level users absorb funds from the lower-level users through the middle layer, thus forming a complete pyramid-style illegal fund-raising transaction network.

[0095] Figure 6 It is a schematic diagram of the pyramid scheme illegal fund-raising scenario constructed in this embodiment.

[0096] like Figure 6 As shown, in this embodiment, for the pyramid-style illegal fund-raising scenario, a three-level pyramid structure is constructed according to its unique operation mode, with user A as the top user. According to the user relationship network, a number of users (e.g., 2 to 5) are randomly selected from the parents, relatives, and friends of user A as the first-level lower users, and user B and user C are selected. User A is set as the top node, and user B and user C are respectively set as the first-level lower nodes, and upper-lower relationship edges are formed between A and B, and between A and C. Then, similarly, a number of users are randomly selected from the parents, relatives, and friends of user B and user C as the second-level lower users, and they are set as the second-level lower nodes, and upper-lower relationship edges are formed between user B, user C, and their corresponding lower nodes, thereby constructing a three-level pyramid structure.

[0097] After the pyramid structure is built, transaction data is generated based on the relationship between users at each level. First, the transaction frequency and amount are determined. The transfer amount of lower-level users is usually set to a small range, specifically, each transaction ranges from hundreds to thousands of yuan. The setting of small transfers is to simulate the characteristics of frequent transfers of funds from lower-level users to middle-level users. This amount range can reflect the limitations of the participants' economic capabilities and reduce the suspiciousness of a single transaction. Next, transaction records are generated. Lower-level users initiate high-frequency small transfers to middle-level users, while middle-level users make large transfers to top-level users. This transfer mode reflects the characteristics of fund concentration layer by layer. Finally, the transaction date and time are set, and the transaction time is reasonably distributed in different months to simulate the continuous flow of funds, thereby enhancing the authenticity and diversity of the data.

[0098] In addition, feature engineering was carried out and corresponding evaluation indicators were set to evaluate the effectiveness of illegal fund-raising detection based on the data generated above.

[0099] Among them, feature engineering: including the bank to which the account belongs, balance, card type, cardholder age, cardholder occupation, transaction amount statistics (mean, maximum, minimum, variance) and number of transactions. The label is whether the account is marked as blacklisted (marked as 1 for blacklisted and 0 for whitelisted), and the positive and negative sample ratio is 1:10. Training set test set division: 30% of the data is allocated to the test set, and the remaining 70% of the data is used to train the model.

[0100] The evaluation indicators are set as follows:

[0101] Precision: represents the proportion of transactions predicted as positive samples that are actually positive samples. A high precision rate indicates that the model can effectively identify illegal fundraising transactions and reduce false positives. The specific formula is:

[0102]

[0103] Recall: represents the proportion of transactions that are actually positive samples that are predicted by the model. A higher recall rate can ensure that illegal fundraising transactions are identified as much as possible and reduce missed reports. The specific formula is:

[0104]

[0105] F1-Score: It is the harmonic average of precision and recall, and is used to balance the model performance when the data is unbalanced. The specific formula is:

[0106]

[0107] KS value: It is an important indicator to measure the distinguishing ability of the binary classification model and is widely used in fields such as credit scoring and risk prediction. The KS value evaluates the model's ability to distinguish between the two types of samples by comparing the maximum difference between the cumulative distributions of positive examples (such as illegal fundraising transactions) and negative examples (such as normal transactions). The calculation formula is as follows:

[0108] KS=max|TPR-FPR|

[0109] In the formula, TPR represents the proportion of positive samples correctly predicted by the model among all actual positive samples, which measures the model's ability to identify positive samples. FPR represents the proportion of positive samples incorrectly predicted by the model among all actual negative samples. It measures the possibility that the model misclassifies negative samples as positive samples. The KS value ranges from 0 to 1, and the closer the result is to 1, the better the performance of the model.

[0110] After using the above method to generate simulated transaction data (including normal transaction data and various types of illegal fund-raising transaction data), three models in the prior art are used to detect illegal fund-raising on the generated transaction data, and the above-mentioned evaluation indicators are used to evaluate the detection effect. The results are shown in Table 3 below.

[0111] Table 3 Comparison of illegal fund-raising detection effects of different models

[0112]

[0113] It can be seen from Table 3 that the KS value of the three existing models in illegal fund-raising detection based on simulated transaction data only reaches about 60%, which shows that the generated transaction data successfully simulates the complex characteristics of illegal fund-raising transactions, making it impossible for the existing models to achieve perfect classification and recognition effects, which is conducive to promoting the development of financial fraud detection and related research.

[0114] Functions and Effects of the Embodiments

[0115] Compared with the financial data generation method in the prior art, the financial data generation method for illegal fund-raising transaction risk scenarios provided by this embodiment has the following beneficial effects:

[0116] Simulate real transaction data generation: By building a transaction data generation system based on user basic data, statistical data and relationship networks, it is possible to generate transaction data covering a variety of illegal fund-raising transaction patterns and normal user transactions, covering detailed transaction patterns of users of different occupations and age groups.

[0117] Enhanced detection capabilities: For pyramid scheme-style illegal fundraising transaction scenarios, the corresponding illegal fundraising transaction data generated has characteristics such as periodic transfers, pyramid structures, and complex network relationships. The use of such synthetic data will help improve the detection and analysis of illegal fundraising activities.

[0118] Data security and privacy protection: In the process of synthesizing data, the corresponding data is automatically generated according to the real statistical laws of each data, etc., without involving real user data at all, ensuring data privacy and security, complying with legal and privacy protection requirements, and the synthetic data has characteristics that are very close to real user data, solving the problem of lack of real user data and difficulty in conducting related research in the existing technology.

[0119] Flexibility and adaptability: This method can flexibly adjust the transaction generation rules according to the characteristics of users in different occupational groups and age groups, ensuring that the generated data has high authenticity and diversity, and can be well applied to a variety of different application scenarios.

[0120] 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 financial data generation method for illegal fund-raising transaction risk scenarios, characterized in that: The following steps are involved: Step S1, generating user basic data of multiple users; Step S2, constructing a user relationship network of the plurality of users based on the user basic data; Step S3, generating normal transaction data between the users based on the user basic information, the user relationship network, statistical data and a predetermined normal transaction generation rule; Step S4, generating illegal fund-raising transaction data of various illegal fund-raising transaction scenarios between the users based on the user basic information, the user relationship network, statistical data and predetermined multiple illegal fund-raising transaction generation rules, The basic user data of each user includes bank card information and user information, and the user information at least includes age, income level, and occupational group, which are generated according to corresponding statistical data. The user relationship network includes a plurality of nodes and edges between the nodes, the nodes are the users, and the edges are the relationships between the users.

2. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 1 is characterized by: in, The bank card information includes bank card number, bank, and balance. The user information also includes name, registration date, gender, and suspicious object mark. In step S1, for the bank card information, the bank card number is generated by a unique identifier, and the balance is randomly set within a predetermined value range according to the corresponding income level. For the user information, the income level and the occupational group are allocated according to the proportion of age distribution based on corresponding statistical data.

3. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 1 is characterized by: in, The user relationship network includes a plurality of nodes and edges between the nodes, the nodes are the users, and the edges are the relationships between the users. The types of relationships include parental relationships, relative relationships, and friend relationships. The friend relationships include general friend relationships and good friend relationships. In step S2, an initial user relationship network is generated based on the user basic data and the type of the relationship and is dynamically updated.

4. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 3 is characterized by: in, In step S2, for the friendship relationship, a potential friend group of the target user is determined according to the age difference between the target user among the users and the user; two groups of predetermined numbers of the users are randomly selected from the potential friend group and are respectively set as friends with whom the target user has a general relationship and friends with whom the target user has a good relationship, and the two-way nature of the friendship relationship is ensured; a predetermined number of the users among the friends of the friends of the target user are selected and are also set as the friends of the target user; For the relative relationship, a predetermined number of the users are randomly selected for the target user and set as the relatives of the target user, and the bidirectionality of the relative relationship is ensured. For the parent relationship, the user is selected for the target user and set as the father or mother according to the age difference between the target user and the user and a predetermined logic rule, and the uniqueness of the parent relationship is ensured.

5. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 1 is characterized by: in, In step S3, statistical analysis is performed based on the user basic data of the multiple users to obtain characteristic data; the user group to which the multiple users belong is determined based on the characteristic data; the normal transaction generation rule for the user group is generated based on the determined user group and predetermined multiple transaction data characteristic information corresponding to the user group; and the normal transaction data between the multiple users is generated based on the user basic data and the normal transaction generation rule.

6. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 5 is characterized by: in, The transaction data characteristic information includes a transaction object range, a transaction time range, a transaction amount range, a transaction frequency range, and a transaction intention range, which are set according to the age, income level, and occupational group of the users in the user group.

7. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 1, Features: Wherein, in step S4, based on the operation mode in the illegal fund-raising transaction scenario, multiple users with relationship edges are selected from the user relationship network and an illegal fund-raising relationship network is constructed; The illegal fund-raising transaction generation rules are generated based on the transaction data characteristic information in the illegal fund-raising transaction scenario; the illegal fund-raising transaction data between the multiple users are generated based on the user basic data of the users in the illegal fund-raising relationship network and the illegal fund-raising transaction generation rules.

8. The method for generating financial data for illegal fund-raising transaction risk scenarios according to claim 7, characterized in that: in, The illegal fund-raising transaction scenario is a pyramid scheme-style illegal fund-raising transaction scenario. In step S4, according to the unique operation mode of the pyramid scheme illegal fund-raising transaction scenario, one of the users is randomly selected from the user relationship network as the top-level user, one or more of the top-level user's friends, relatives, and parents are selected as the first-level subordinate users of the top-level user, and a superior-subordinate relationship edge is formed between the top-level user and the first-level subordinate users. One or more of the friends, relatives, and parents of one or more of the first-level subordinate users are selected as second-level subordinate users, and a superior-subordinate relationship edge is formed between the first-level subordinate users and the corresponding second-level subordinate users. The above process is repeated to form a multi-level pyramid network of illegal fund-raising relationships.

9. The method for generating financial data for illegal fund-raising transaction risk scenarios according to claim 8, characterized in that: in, The transaction data characteristic information includes the transaction object range, transaction time range, transaction amount range, and transaction frequency range, which are set according to the operation mode of the pyramid scheme illegal fund-raising transaction scenario. The transaction object range includes the user's superior users and subordinate users, and sets the ratio range of the transaction quantity between the user and the superior user, and between the user and the subordinate user, respectively. The transaction frequency ranges from once to several times per week or month.

10. The financial data generation method for illegal fund-raising transaction risk scenarios according to claim 1 is characterized by: in, The illegal fund-raising transaction scenarios include pyramid scheme-style illegal fund-raising transaction scenarios, virtual currency illegal fund-raising transaction scenarios, real estate illegal fund-raising transaction scenarios, pension project illegal fund-raising transaction scenarios, and Internet financial platform illegal fund-raising transaction scenarios.

Images