Intelligent transaction engine risk early warning system

Abstract

The invention discloses an intelligent transaction engine risk early warning system, which relates to the technical field of transaction risks, and comprises the steps of classifying all user groups on a transaction engine according to transaction records of users, accurately classifying the user groups according to transaction habit conditions of the users, and carrying out risk early warning on the transaction engine. And a basis is provided for subsequent multi-layer user risk contour structural body and model training. And aiming at the transaction record of each type of user group, generating a multi-layer user risk contour structure body corresponding to each type of user group to help the training of a subsequent model. And determining the risk difficulty of each type of user group through the multi-layer user risk contour structural body of each type of user group, defining model parameters through the risk difficulty of each type of user group, and training a safety risk early warning model of each type of user group through the model parameters and the multi-layer user risk contour structural body. The accuracy and adaptability of risk early warning are improved, and safe transaction on the transaction platform is ensured.

Description

Technical Field

[0001] The present invention relates to the field of transaction risk technology, and in particular to an intelligent transaction engine risk early warning system. Background Art

[0002] The trading engine risk warning solution is proposed in the context of the rapid development of financial technology, targeting the deficiencies of traditional risk warning systems in data acquisition, analysis efficiency and accuracy. With the widespread application of technologies such as big data, cloud computing, and artificial intelligence, the transaction data in the financial market has shown explosive growth, and traditional risk warning methods are gradually unable to meet the needs of real-time and accurate risk management. The intelligent trading engine risk warning solution integrates internal and external data resources and uses advanced machine learning and artificial intelligence technologies to achieve real-time monitoring, dynamic analysis and accurate warning of market risks.

[0003] In the prior art, there are a large number of users circulating on trading platforms or engines, and different users have different trading habits. However, the platform uses a unified risk warning algorithm for risk detection or prediction, resulting in poor accuracy and adaptability of risk warnings, and the security of transactions cannot be guaranteed.

[0004] Therefore, how to improve the accuracy and adaptability of risk warning is a technical problem that needs to be solved. Summary of the invention

[0005] The purpose of the present invention is to solve the problem of poor accuracy and adaptability of risk warning in the prior art, and to propose an intelligent trading engine risk warning system, which includes a classification module, a generation module, a determination module, a training module and a warning module.

[0006] The classification module is used to obtain the transaction records of users on the transaction engine and classify all user groups on the transaction engine according to the transaction records of the users;

[0007] The generating module is used to generate a multi-layer user risk profile structure corresponding to each type of user group according to the transaction records of each type of user group;

[0008] The determination module is used to determine the risk difficulty of each type of user group through a multi-layer user risk profile structure of each type of user group;

[0009] The training module is used to define model parameters according to the risk difficulty of each user group, and train the security risk warning model through the model parameters and the multi-layer user risk profile structure;

[0010] The early warning module is used to implement risk identification and early warning functions on the intelligent transaction engine according to the security risk early warning model.

[0011] In some embodiments of the present application, the classification module is used to classify all user groups on the transaction engine according to the user's transaction records, which is specifically:

[0012] The user's transaction record includes transaction volume information, transaction behavior information and transaction business type, and the transaction volume features and transaction behavior features are extracted from the transaction volume information and transaction behavior information;

[0013] Construct each user's transaction business type set through transaction business types, and compare the intersection of multiple users' transaction business type sets to perform the first classification of users on the transaction engine;

[0014] Normalize the transaction behavior features. The transaction volume features include the average transaction volume and transaction volume volatility. The user group after the first classification is recorded as a user group. For each user in the same user group, calculate the similarity of each transaction behavior feature between two users.

[0015] Determine the similarity between two users in the same user group based on the similarity of transaction behavior characteristics and transaction volume characteristics;

[0016]

[0017] in, is the similarity between the i2th user and the i3th user in the i1th user group, τ is the conversion coefficient, n is the number of transaction behavior features, is the similarity weight of the i4th transaction behavior feature, is the similarity of the trading behavior characteristics of the i4th user and the i3th user in the i1th user group. β1 and β2 are the combined weights of the average trading volume and the trading volume volatility, respectively. are the average transaction volumes of the i2th user and the i3th user, respectively. are the transaction volume volatility of the i2th user and the i3th user respectively, is the constant corresponding to the i2th user and the i3th user;

[0018] The user groups are classified a second time by measuring the similarity between two users in the same user group.

[0019] In some embodiments of the present application, the generation module includes a first unit, a second unit, a third unit and a fourth unit, the multi-layer user risk profile structure includes a basic information layer, a common feature layer, a risk feature layer and an association layer, and the first unit is used to:

[0020] Filter out the transaction records of each user group from the user's transaction records, obtain transaction-related information of each user group from the transaction records of each user group, perform data preprocessing on the transaction-related information of each user group, and construct a basic information layer through the preprocessed transaction-related information.

[0021] In some embodiments of the present application, the second unit is used to:

[0022] The pre-processed transaction related information is divided into two parts: numerical category information and non-numerical category information;

[0023] For non-numerical category information, the same or similar information of multiple users in the same user group is screened out to obtain the common features of the non-numerical category;

[0024] For the numerical category information, calculate the characteristic value domain range of each numerical category information of multiple users in the same user group, determine the first intersection degree range by the business security involved in each numerical category information, determine the second intersection degree range by the characteristic value domain range of each numerical category information, and screen out the common characteristic types and value intervals of the numerical categories of multiple users in the same user group by the intersection of the first intersection degree range and the second intersection degree range, and use them as the common characteristics of the numerical categories;

[0025] A common feature layer is constructed based on the common features of non-numerical categories and the common features of numerical categories.

[0026] In some embodiments of the present application, the third unit is used to:

[0027] Extract all risk features in the basic information layer and record them as the first risk features, and extract all risk features in the common feature layer and record them as the second risk features;

[0028] The first risk feature and the second risk feature are compared to determine the type of risk feature, the types of risk features include common risk features and non-common risk features, different weights are assigned to the two types of common risk features and non-common risk features, and a risk feature layer is constructed.

[0029] In some embodiments of the present application, the fourth unit is used to:

[0030] Perform correlation analysis among the basic information layer, common feature layer, and risk feature layer, and quantify the degree of correlation. Use a network diagram to display the correlation relationship among the basic information layer, common feature layer, and risk feature layer, and construct a correlation layer.

[0031] In some embodiments of the present application, the generating module further includes a fifth unit, wherein the fifth unit is configured to:

[0032] The multi-layer user risk profile structure also includes a balancing information layer;

[0033] Calculate the data volume of the common feature layer and the risk feature layer in the multi-layer user risk profile structure of the target user group, and calculate the similarity between the multi-layer user risk profile structure of the target user group and the multi-layer user risk profile structure of other user groups. Extract data from the common feature layer and the risk feature layer of the multi-layer user risk profile structure of other user groups through the data volume and similarity of the common feature layer and the risk feature layer of the target user group, and add these data to the balanced information layer in the multi-layer user risk profile structure of the target user group, so as to construct the balanced information layer of the multi-layer user risk profile structure of all user groups.

[0034] In some embodiments of the present application, the determination module is used to determine the risk difficulty of each type of user group through the multi-layer user risk profile structure of each type of user group, which is specifically:

[0035] Calculate the risk difficulty level corresponding to each user group based on the common risk features and non-common risk features in the risk feature layer within the multi-layer user risk profile structure of each user group;

[0036]

[0037] in, is the risk difficulty level corresponding to the j1-th user group, m1 and m2 are the numbers of shared risk features and non-shared risk features in the risk feature layer of the multi-layer user risk profile structure corresponding to the j1-th user group, are the weight of the j1th shared risk feature and the weight of the j2th non-shared risk feature, respectively. are the size of the j1th shared risk feature and the size of the j2th non-shared risk feature, respectively, and max(γ3D3) is and The maximum value in are the first constant and the second constant respectively, and [] is the rounding symbol.

[0038] In some embodiments of the present application, the training module is used to define model parameters according to the risk difficulty of each user group, and train the security risk warning model through the model parameters and the multi-layer user risk profile structure, which is specifically:

[0039] For the same type of user group, the model parameter combination is determined according to the risk difficulty level, and the multi-layer user risk profile structure of this type of user group is converted into samples. The security risk warning model for this type of user group is trained through the combination of samples and model parameters, thereby training a security risk warning model corresponding to each type of user group.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. Classify all user groups on the trading engine according to their transaction records, and accurately classify user groups according to their transaction habits, providing a basis for subsequent multi-layer user risk profile structures and model training. Generate a multi-layer user risk profile structure corresponding to each user group based on the transaction records of each user group. The multi-layer user risk profile structure describes the transaction risk of each user group and the correlation between multi-layer information, which helps the subsequent model training.

[0042] 2. The risk difficulty of each user group is determined through the multi-layer user risk profile structure of each user group, the model parameters are defined through the risk difficulty of each user group, and the security risk warning model of each user group is trained through the model parameters and the multi-layer user risk profile structure. The accuracy and adaptability of risk warning are improved, and the safe conduct of transactions on the trading platform is ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a schematic diagram of the structure of an intelligent trading engine risk early warning system proposed by the present invention;

[0044] Figure 2 This is a structural schematic diagram of a generation module in an intelligent trading engine risk warning system proposed by the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0046] Reference Figure 1 , an intelligent trading engine risk warning system, including the following modules,

[0047] It includes classification module, generation module, determination module, training module and early warning module.

[0048] The classification module is used to obtain transaction records of users on the transaction engine and classify all user groups on the transaction engine according to the transaction records of the users.

[0049] In this embodiment, the system of the transaction engine is connected to ensure that the user's transaction record data can be obtained in real time or regularly. This usually involves calling the API interface, connecting to the database or transmitting data files. The transaction record includes relevant information such as transaction volume information, transaction behavior information and transaction business type.

[0050] In some embodiments of the present application, the classification module is used to classify all user groups on the transaction engine according to the user's transaction records, which is specifically:

[0051] The user's transaction record includes transaction volume information, transaction behavior information and transaction business type, and the transaction volume features and transaction behavior features are extracted from the transaction volume information and transaction behavior information;

[0052] Construct each user's transaction business type set through transaction business types, and compare the intersection of multiple users' transaction business type sets to perform the first classification of users on the transaction engine;

[0053] Normalize the transaction behavior features. The transaction volume features include the average transaction volume and transaction volume volatility. The user group after the first classification is recorded as a user group. For each user in the same user group, calculate the similarity of each transaction behavior feature between two users.

[0054] Determine the similarity between two users in the same user group based on the similarity of transaction behavior characteristics and transaction volume characteristics;

[0055]

[0056] in, is the similarity between the i2th user and the i3th user in the i1th user group, τ is the conversion coefficient, n is the number of transaction behavior features, is the similarity weight of the i4th transaction behavior feature, is the similarity of the trading behavior characteristics of the i4th user and the i3th user in the i1th user group. β1 and β2 are the combined weights of the average trading volume and the trading volume volatility, respectively. are the average transaction volumes of the i2th user and the i3th user, respectively. are the transaction volume volatility of the i2th user and the i3th user respectively, is the constant corresponding to the i2th user and the i3th user;

[0057] The user groups are classified a second time by measuring the similarity between two users in the same user group.

[0058] In this embodiment, the transaction business types include stock trading, fund subscription, etc., and the transaction behavior characteristics include transaction frequency (number of buy, sell, and hold), transaction duration, etc. The similarity of each transaction behavior feature between two users is calculated, and the similarity can be calculated by Euclidean distance, cosine similarity, etc. It represents the modification of the comprehensive characteristics of trading behavior by similar situations of trading volume.

[0059] The generating module is used to generate a multi-layer user risk profile structure corresponding to each type of user group according to the transaction records of each type of user group.

[0060] In this embodiment, a multi-layer user risk profile structure is constructed for each type of user group, and the risk characteristics of the user group and the association between the multi-layer information are described by the multi-layer user risk profile structure.

[0061] In some embodiments of the present application, Figure 2 As shown, the generation module includes a first unit, a second unit, a third unit and a fourth unit, the multi-layer user risk profile structure includes a basic information layer, a common feature layer, a risk feature layer and an association layer, and the first unit is used to:

[0062] Filter out the transaction records of each user group from the user's transaction records, obtain transaction-related information of each user group from the transaction records of each user group, perform data preprocessing on the transaction-related information of each user group, and construct a basic information layer through the preprocessed transaction-related information.

[0063] In this embodiment, the multi-layer user risk profile structure includes a basic information layer, a common feature layer, a risk feature layer, an association layer, and a balance information layer. The basic information layer is the most original data of user transactions, the common feature layer is some common features of this group of users, the risk feature layer is the risk feature situation involved in this group of users, the association layer describes the association relationship between the three layers of the basic information layer, the common feature layer, and the risk feature layer, and the balance information layer is data extracted from other user groups to ensure data diversity.

[0064] In some embodiments of the present application, the second unit is used to:

[0065] The pre-processed transaction related information is divided into two parts: numerical category information and non-numerical category information;

[0066] For non-numerical category information, the same or similar information of multiple users in the same user group is screened out to obtain the common features of the non-numerical category;

[0067] For the numerical category information, calculate the characteristic value domain range of each numerical category information of multiple users in the same user group, determine the first intersection degree range by the business security involved in each numerical category information, determine the second intersection degree range by the characteristic value domain range of each numerical category information, and screen out the common characteristic types and value intervals of the numerical categories of multiple users in the same user group by the intersection of the first intersection degree range and the second intersection degree range, and use them as the common characteristics of the numerical categories;

[0068] A common feature layer is constructed based on the common features of non-numerical categories and the common features of numerical categories.

[0069] In this embodiment, the shared features can be numerical (such as transaction volume, transaction frequency, etc.) or non-numerical category (such as user occupation, region, etc.). The cross-degree range is the degree of overlap of the same numerical category information between different users. Different numerical category information involves different business security situations, and the corresponding first cross-degree ranges are also different. The feature value range corresponds to different first cross-degree ranges. The intersection of the first cross-degree range and the second cross-degree range is used to screen out the common feature types and numerical ranges of the numerical categories of multiple users in the same user group.

[0070] In some embodiments of the present application, the third unit is used to:

[0071] Extract all risk features in the basic information layer and record them as the first risk features, and extract all risk features in the common feature layer and record them as the second risk features;

[0072] The first risk feature and the second risk feature are compared to determine the type of risk feature, the types of risk features include common risk features and non-common risk features, different weights are assigned to the two types of common risk features and non-common risk features, and a risk feature layer is constructed.

[0073] In this embodiment, the shared risk features are risk features that are shared by multiple users in this user group, and the remaining risk features are non-shared risk features. Generally speaking, the weight of the shared risk features is greater than the weight of the non-shared risk features.

[0074] In some embodiments of the present application, the fourth unit is used to:

[0075] Perform correlation analysis among the basic information layer, common feature layer, and risk feature layer, and quantify the degree of correlation. Use a network diagram to display the correlation relationship among the basic information layer, common feature layer, and risk feature layer, and construct a correlation layer.

[0076] In this embodiment, for example, the basic information layer and the risk feature layer can use statistical methods (such as correlation analysis, chi-square test, etc.) to quantify these associations. A series of numerical correlation degree indicators, such as correlation coefficients, feature importance scores, etc., can be obtained. According to the complexity and data type of the correlation relationship, select the appropriate chart type to display the correlation situation. Use a network diagram to display the correlation relationship network between features, and use Python's matplotlib, seaborn, networkx and other libraries, or R language ggplot2, igraph and other packages to draw charts. These tools provide rich visualization options and highly customized capabilities. Provide a brief explanation and interpretation next to the chart to help users understand the meaning of the correlation relationship and the significance of the quantitative results.

[0077] In this embodiment, the associations between multiple layers of information provide richer contextual information for the security risk warning model. By incorporating these associations into model training, the model can better understand user behavior, transaction patterns, and potential risk points, thereby improving the model's ability to identify security risks. The associations reveal the path and process of risk transmission, which enables the model to more accurately capture risk signals and issue warnings before risks occur. By training a model that includes associations, the accuracy and timeliness of security risk warnings can be significantly improved.

[0078] In some embodiments of the present application, the generating module further includes a fifth unit, wherein the fifth unit is configured to:

[0079] The multi-layer user risk profile structure also includes a balancing information layer;

[0080] Calculate the data volume of the common feature layer and the risk feature layer in the multi-layer user risk profile structure of the target user group, and calculate the similarity between the multi-layer user risk profile structure of the target user group and the multi-layer user risk profile structure of other user groups. Extract data from the common feature layer and the risk feature layer of the multi-layer user risk profile structure of other user groups through the data volume and similarity of the common feature layer and the risk feature layer of the target user group, and add these data to the balanced information layer in the multi-layer user risk profile structure of the target user group, so as to construct the balanced information layer of the multi-layer user risk profile structure of all user groups.

[0081] In this embodiment, the information in the multi-layer user risk profile structure is all information under the same type of user group. There may be overfitting problems in subsequent model training. The balanced information layer is designed to solve this problem, and the information of other categories of user groups is added to it. Increase sample diversity: If the original data set only contains information about one type of user group, adding information about other types of user groups can significantly increase the diversity of samples and help the model learn a wider data distribution. Avoid data imbalance: When adding information about other types of user groups, pay attention to maintaining data balance between categories to avoid the model being biased towards users of that type due to excessive data from a certain type of user group.

[0082] In this embodiment, data is extracted from the common feature layer and the risk feature layer of other types of user groups based on the data volume and similarity of the common feature layer and the risk feature layer, and data is extracted from both similar groups and dissimilar groups.

[0083] The determination module is used to determine the risk difficulty of each type of user group through a multi-layer user risk profile structure of each type of user group.

[0084] In this embodiment, risk difficulty refers to the difficulty of risk detection, assessment, and early warning. The higher the difficulty, the higher the complexity requirement for the model.

[0085] In some embodiments of the present application, the determination module is used to determine the risk difficulty of each type of user group through the multi-layer user risk profile structure of each type of user group, which is specifically:

[0086] Calculate the risk difficulty level corresponding to each user group based on the common risk features and non-common risk features in the risk feature layer within the multi-layer user risk profile structure of each user group;

[0087]

[0088] in, is the risk difficulty level corresponding to the j1-th user group, m1 and m2 are the numbers of shared risk features and non-shared risk features in the risk feature layer of the multi-layer user risk profile structure corresponding to the j1-th user group, are the weight of the j1th shared risk feature and the weight of the j2th non-shared risk feature, respectively. are the size of the j1th shared risk feature and the size of the j2th non-shared risk feature, respectively, and max(γ3D3) is and The maximum value in are the first constant and the second constant respectively, and [] is the rounding symbol.

[0089] In this embodiment, max(γ3D3) is and The maximum value in and Pick out the maximum value from each of them and compare which one is bigger between the two maximum values. It represents the correction of the maximum risk characteristic impact to the combination of shared risk characteristics and non-shared risk characteristics.

[0090] The training module is used to define model parameters according to the risk difficulty of each type of user group, and to train the security risk warning model through the model parameters and the multi-layer user risk profile structure.

[0091] In this embodiment, if a security risk warning model is to be able to provide targeted risk warnings for all categories of user groups, it must have a high degree of plasticity and adaptability. This means that the model needs to be able to automatically adjust its parameters and strategies according to the characteristics of different user groups to adapt to different risk environments and detection needs. However, in practical applications, this high degree of plasticity and adaptability is often difficult to achieve. Therefore, a corresponding security risk warning model is trained for each type of user group. The data is highly targeted: the training data is more focused on the behavior patterns and risk characteristics of a specific user group, which improves the model's prediction accuracy for the user group. The features are more accurate: it is possible to select more relevant and significant risk features for a specific user group, thereby improving the sensitivity and specificity of the model. Model structure optimization: the structure and complexity of the model can be adjusted according to the risk patterns of specific user groups to make it more adaptable to the actual risk environment.

[0092] In some embodiments of the present application, the training module is used to define model parameters according to the risk difficulty of each user group, and train the security risk warning model through the model parameters and the multi-layer user risk profile structure, which is specifically:

[0093] For the same type of user group, the model parameter combination is determined according to the risk difficulty level, and the multi-layer user risk profile structure of this type of user group is converted into samples. The security risk warning model for this type of user group is trained through the combination of samples and model parameters, thereby training a security risk warning model corresponding to each type of user group.

[0094] In this embodiment, the model parameter combination (model architecture parameters, such as the number of layers, number of neurons and other model architecture parameters) is determined according to the risk difficulty level. The model parameter combination is a combination of different model architecture parameter sizes. The higher the risk difficulty level, the more complex the corresponding model and the more diverse the architecture.

[0095] In this embodiment, the revealed association relationship is converted into features that can be understood by the model. This may include encoding the association relationship into numerical features (such as association strength, association frequency, etc.), or converting the association relationship into graph structure data (such as user-transaction-risk feature graph). Select a suitable security risk warning model (such as a machine learning model, a deep learning model, etc.), and use the converted association relationship features as input data for model training.

[0096] The early warning module is used to implement risk identification and early warning functions on the intelligent transaction engine according to the security risk early warning model.

[0097] Compared with the prior art, the present invention has the following beneficial effects:

[0098] 1. Classify all user groups on the trading engine according to their transaction records, and accurately classify user groups according to their transaction habits, providing a basis for subsequent multi-layer user risk profile structures and model training. Generate a multi-layer user risk profile structure corresponding to each user group based on the transaction records of each user group. The multi-layer user risk profile structure describes the transaction risk of each user group and the correlation between multi-layer information, which helps the subsequent model training.

[0099] 2. The risk difficulty of each user group is determined through the multi-layer user risk profile structure of each user group, the model parameters are defined through the risk difficulty of each user group, and the security risk warning model of each user group is trained through the model parameters and the multi-layer user risk profile structure. The accuracy and adaptability of risk warning are improved, and the safe conduct of transactions on the trading platform is ensured.

[0100] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.

[0101] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present invention.

[0102] Those skilled in the art will appreciate that the modules in the system in the implementation scenario can be distributed in the system of the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more systems different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple submodules.

[0103] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent trading engine risk warning system, characterized in that: It includes classification module, generation module, determination module, training module and warning module. The classification module is used to obtain the transaction records of users on the transaction engine and classify all user groups on the transaction engine according to the transaction records of the users; The generating module is used to generate a multi-layer user risk profile structure corresponding to each user group according to the transaction records of each user group; The determination module is used to determine the risk difficulty of each type of user group through a multi-layer user risk profile structure of each type of user group; The training module is used to define model parameters according to the risk difficulty of each user group, and train the security risk warning model through the model parameters and the multi-layer user risk profile structure; The early warning module is used to implement risk identification and early warning functions on the intelligent transaction engine according to the security risk early warning model.

2. The intelligent trading engine risk early warning system according to claim 1 is characterized in that: The classification module is used to classify all user groups on the transaction engine according to the user's transaction records, which is specifically: The user's transaction record includes transaction volume information, transaction behavior information and transaction business type, and the transaction volume features and transaction behavior features are extracted from the transaction volume information and transaction behavior information; Construct each user's transaction business type set through transaction business types, and compare the intersection of multiple users' transaction business type sets to perform the first classification of users on the transaction engine; Normalize the transaction behavior features. The transaction volume features include the average transaction volume and transaction volume volatility. The user group after the first classification is recorded as a user group. For each user in the same user group, calculate the similarity of each transaction behavior feature between two users. Determine the similarity between two users in the same user group based on the similarity of transaction behavior characteristics and transaction volume characteristics; in, is the similarity between the i2th user and the i3th user in the i1th user group, τ is the conversion coefficient, n is the number of transaction behavior features, is the similarity weight of the i4th transaction behavior feature, is the similarity of the trading behavior characteristics of the i4th user and the i3th user in the i1th user group. β1 and β2 are the combined weights of the average trading volume and the trading volume volatility, respectively. are the average transaction volumes of the i2th user and the i3th user, respectively. are the transaction volume volatility of the i2th user and the i3th user respectively, is the constant corresponding to the i2th user and the i3th user; The user groups are classified a second time by measuring the similarity between two users in the same user group.

3. The intelligent trading engine risk warning system according to claim 2 is characterized in that: The generation module includes a first unit, a second unit, a third unit and a fourth unit. The multi-layer user risk profile structure includes a basic information layer, a common feature layer, a risk feature layer and an association layer. The first unit is used to: Filter out the transaction records of each user group from the user's transaction records, obtain transaction-related information of each user group from the transaction records of each user group, perform data preprocessing on the transaction-related information of each user group, and construct a basic information layer through the preprocessed transaction-related information.

4. The intelligent trading engine risk warning system according to claim 3 is characterized in that: The second unit is used for: The pre-processed transaction related information is divided into two parts: numerical category information and non-numerical category information; For non-numerical category information, the same or similar information of multiple users in the same user group is screened out to obtain the common features of the non-numerical category; For the numerical category information, calculate the characteristic value domain range of each numerical category information of multiple users in the same user group, determine the first intersection degree range by the business security involved in each numerical category information, determine the second intersection degree range by the characteristic value domain range of each numerical category information, and screen out the common characteristic types and value intervals of the numerical categories of multiple users in the same user group by the intersection of the first intersection degree range and the second intersection degree range, and use them as the common characteristics of the numerical categories; A common feature layer is constructed based on the common features of non-numerical categories and the common features of numerical categories.

5. The intelligent trading engine risk early warning system according to claim 4 is characterized in that: The third unit is used for: Extract all risk features in the basic information layer and record them as the first risk features, and extract all risk features in the common feature layer and record them as the second risk features; The first risk feature and the second risk feature are compared to determine the type of risk feature, the types of risk features include common risk features and non-common risk features, different weights are assigned to the two types of common risk features and non-common risk features, and a risk feature layer is constructed.

6. The intelligent trading engine risk early warning system according to claim 3 is characterized in that: The fourth unit is used for: Perform correlation analysis among the basic information layer, common feature layer, and risk feature layer, and quantify the degree of correlation. Use a network diagram to display the correlation relationship among the basic information layer, common feature layer, and risk feature layer, and construct a correlation layer.

7. The intelligent trading engine risk early warning system according to claim 1, characterized in that: The generating module further comprises a fifth unit, wherein the fifth unit is configured to: The multi-layer user risk profile structure also includes a balancing information layer; Calculate the data volume of the common feature layer and the risk feature layer in the multi-layer user risk profile structure of the target user group, and calculate the similarity between the multi-layer user risk profile structure of the target user group and the multi-layer user risk profile structure of other user groups. Extract data from the common feature layer and the risk feature layer of the multi-layer user risk profile structure of other user groups through the data volume and similarity of the common feature layer and the risk feature layer of the target user group, and add these data to the balanced information layer in the multi-layer user risk profile structure of the target user group, so as to construct the balanced information layer of the multi-layer user risk profile structure of all user groups.

8. The intelligent trading engine risk warning system according to claim 5, characterized in that: The determination module is used to determine the risk difficulty of each user group through the multi-layer user risk profile structure of each user group, which is specifically: Calculate the risk difficulty level corresponding to each user group based on the common risk features and non-common risk features in the risk feature layer within the multi-layer user risk profile structure of each user group; in, is the risk difficulty level corresponding to the j1-th user group, m1 and m2 are the numbers of shared risk features and non-shared risk features in the risk feature layer of the multi-layer user risk profile structure corresponding to the j1-th user group, are the weight of the j1th shared risk feature and the weight of the j2th non-shared risk feature, respectively. are the size of the j1th shared risk feature and the size of the j2th non-shared risk feature, respectively, and max(γ3D3) is and The maximum value in are the first constant and the second constant respectively, and [] is the rounding symbol.

9. The intelligent trading engine risk early warning system according to claim 8, characterized in that: The training module is used to define model parameters according to the risk difficulty of each user group, and train the security risk warning model through the model parameters and the multi-layer user risk profile structure, which is specifically: For the same type of user group, the model parameter combination is determined according to the risk difficulty level, and the multi-layer user risk profile structure of this type of user group is converted into samples. The security risk warning model for this type of user group is trained through the combination of samples and model parameters, thereby training a security risk warning model corresponding to each type of user group.

Images