Account risk assessment method and device, electronic equipment and computer program product

Abstract

The invention discloses an account risk assessment method and device, electronic equipment and a computer program product, and relates to the financial field, the method comprises the following steps: obtaining first feature data and second feature data associated with a to-be-assessed account, the first feature data being used for indicating historical transaction behaviors of the account, and the second feature data being used for indicating the historical transaction behaviors of the account; the second feature data is used for indicating the account attribute of the account; inputting the first feature data into a first evaluation model to obtain third feature data output by the first evaluation model, the third feature data being used for indicating a time sequence dependency relationship of the first feature data; and inputting the second feature data and the third feature data into a second evaluation model to obtain a risk evaluation result of the account output by the second evaluation model, the risk evaluation result being used for indicating the risk type of the account. The account risk assessment method and device solve the technical problem that the accuracy of account risk assessment is low in the prior art.

Description

Technical Field

[0001] This invention relates to the financial field, and more specifically, to a method and apparatus for risk assessment of accounts, electronic devices, and computer program products. Background Technology

[0002] In the field of account risk assessment in the financial sector, existing technical solutions often suffer from isolated analysis when processing dynamic transaction behavior and account attribute information. That is, static attributes and dynamic transaction data are processed separately, lacking an effective integration mechanism, which limits the accuracy of account risk assessment.

[0003] Therefore, there is a technical problem in the related technologies that the accuracy of account risk assessment is relatively low. Summary of the Invention

[0004] This invention provides a method and apparatus for risk assessment of accounts, an electronic device, and a computer program product, to at least solve the technical problem of low accuracy in risk assessment of accounts in related technologies.

[0005] To achieve the above objectives, according to one aspect of this application, a risk assessment method for an account is provided, comprising: acquiring first feature data and second feature data associated with an account to be assessed, wherein the first feature data is used to indicate the account's historical transaction behavior, and the second feature data is used to indicate the account's account attributes; inputting the first feature data into a first assessment model to obtain third feature data output by the first assessment model, wherein the first assessment model is a pre-trained neural network model for capturing the temporal dependencies of the input data, and the third feature data is used to indicate the temporal dependencies of the first feature data; inputting the second feature data and the third feature data into a second assessment model to obtain a risk assessment result of the account output by the second assessment model, wherein the second assessment model is a pre-trained neural network model for predicting the risk type of the account based on the input data, and the risk assessment result is used to indicate the risk type of the account.

[0006] According to another aspect of the present invention, a risk assessment device for an account is also provided, comprising: an acquisition unit, configured to acquire first feature data and second feature data associated with an account to be assessed, wherein the first feature data is used to indicate the account's historical transaction behavior, and the second feature data is used to indicate the account's account attributes; a first assessment unit, configured to input the first feature data into a first assessment model to obtain third feature data output by the first assessment model, wherein the first assessment model is a pre-trained neural network model for capturing the temporal dependencies of the input data, and the third feature data is used to indicate the temporal dependencies of the first feature data; and a second assessment unit, configured to input the second feature data and the third feature data into a second assessment model to obtain a risk assessment result of the account output by the second assessment model, wherein the second assessment model is a pre-trained neural network model for predicting the risk type of the account based on the input data, and the risk assessment result is used to indicate the risk type of the account.

[0007] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the account risk assessment method described above.

[0008] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the account risk assessment method described above.

[0009] The embodiments provided in this application employ a two-stage evaluation model to deeply integrate and analyze historical transaction behavior and account attribute information, significantly improving the accuracy and timeliness of account transaction risk assessment. Specifically, by using historical transaction behavior data of the account to be evaluated as the first feature data and inputting it into a pre-trained first evaluation model, the system can capture and analyze the temporal dependencies in the transaction data, generating third feature data that reflects transaction sequence patterns and dynamic characteristics. This modeling approach enables the system to identify abnormal transaction patterns and effectively predict potential risks even when transaction behavior changes over time. The account attribute information (second feature data) is combined with the third feature data output by the LSTM model and input into the second evaluation model for comprehensive risk type prediction, ensuring good predictive stability even when processing high-dimensional data. The output of the second evaluation model, i.e., the risk assessment result, indicates the account's risk type, including but not limited to low risk, medium risk, and high risk. Through the synergistic effect of the first and second evaluation models, this embodiment achieves deep fusion of dynamic behavioral data and static attribute data. This integration not only fully utilizes the long-term dependencies in time-series data but also considers the complex impact of account attributes, thereby constructing a more comprehensive and accurate risk assessment framework. This achieves the technical effect of improving the accuracy of account risk assessment and solves the technical problem of low accuracy in account risk assessment in related technologies. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0011] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a risk assessment method for accounts is shown.

[0012] Figure 2 This is a flowchart of an optional account risk assessment method according to an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of an optional model training method based on the stacking algorithm according to an embodiment of the present invention.

[0014] Figure 4 This is a schematic diagram of a cross-validation process according to an embodiment of the present invention.

[0015] Figure 5 This is a schematic diagram of an optional account risk assessment device according to an embodiment of the present invention.

[0016] Figure 6This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0018] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0019] It should be noted that the account risk assessment method and device disclosed herein can be used in the computer field, or in any field other than the computer field. This disclosure does not limit the application field of the account risk assessment method and device.

[0020] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected in this public disclosure are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or organization.

[0021] It should be noted that in this disclosure, customer information is collected and analyzed, and users are provided with corresponding operation entry points to choose whether to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0022] The present invention will now be described in detail with reference to various embodiments.

[0023] Example 1

[0024] According to an embodiment of the present invention, an embodiment of a risk assessment method for an account is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0025] The account risk assessment method provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing an account risk assessment method is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1 The processor 102 (which may include, but is not limited to, a microprocessor MCU (Microcontroller Unit) or a programmable gate array (FPGA)) is shown as 102a, 102b, ..., 102n. It also includes a memory 104 for storing data and a transmission device 106 for communication functions. In addition, it may include: a display, an input / output interface (I / O interface), a Universal Serial Bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0026] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0027] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the account risk assessment method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned account risk assessment method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0028] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0029] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0030] Under the aforementioned operating environment, this application provides the following: Figure 2 The risk assessment method for the account shown. Figure 2 This is a flowchart of an optional account risk assessment method according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0031] S202, Obtain first feature data and second feature data associated with the account to be evaluated, wherein the first feature data is used to indicate the account's historical transaction behavior and the second feature data is used to indicate the account's account attributes;

[0032] S204, the first feature data is input into the first evaluation model to obtain the third feature data output by the first evaluation model, wherein the first evaluation model is a pre-trained neural network model used to capture the temporal dependency of the input data, and the third feature data is used to indicate the temporal dependency of the first feature data;

[0033] S206, Input the second feature data and the third feature data into the second evaluation model to obtain the risk assessment result of the account output by the second evaluation model. The second evaluation model is a pre-trained neural network model used to predict the risk type of the account based on the input data. The risk assessment result is used to indicate the risk type of the account.

[0034] Optionally, in this embodiment, the account to be evaluated refers to a financial account that requires risk assessment. The first feature data indicates the account's historical transaction behavior, mainly including dynamic data such as transaction time, amount, location, and merchant type. The second feature data indicates the account's attributes, including static information such as age, occupation, and income level, and historical credit data such as the number of overdue payments.

[0035] Optionally, in this embodiment, the first evaluation model is a pre-trained neural network model, mainly used to capture the temporal dependencies of the input data. It can be, but is not limited to, an LSTM (Long Short-Term Memory) model, capable of remembering long-term dependencies in a sequence and suitable for processing time-series data. The third feature data is output by the first evaluation model, reflecting the temporal dependencies of historical trading behavior in the first feature data, such as fluctuations in trading frequency and trading anomalies within a specific time period. The second evaluation model is a pre-trained neural network model used to predict the risk type of the account based on the input data. It can be, but is not limited to, a LightGBM model, and has high predictive performance.

[0036] Optionally, in this embodiment, the risk assessment result is output by the second assessment model, indicating the risk type of the account, such as "low risk", "medium risk" or "high risk", to help financial institutions make corresponding risk control decisions.

[0037] Optionally, in this embodiment, the first feature data (historical transaction behavior) and the second feature data (account attributes) of the account to be evaluated are obtained from the database. The data preprocessing steps include cleaning, denoising, filling missing values, and converting non-numerical features into a model-readable format, such as performing one-hot encoding.

[0038] Next, the first feature data is input into the first evaluation model (such as an LSTM model), which can capture the temporal dependencies in the trading sequence. Through training, the model learns how to identify normal and abnormal trading patterns and outputs the third feature data, namely the sequence pattern feature or temporal anomaly indicator.

[0039] Then, the second and third feature data are input together into a second assessment model (such as the LightGBM model). This model aims to comprehensively assess the static and dynamic risks of an account and output a risk assessment result. This two-stage assessment ensures that both static and dynamic characteristics are fully considered, thereby improving the accuracy and comprehensiveness of the risk assessment.

[0040] Before being input into the second evaluation model, the two datasets are fused using multimodal features and weighted by an attention mechanism to ensure that the importance of each feature can be correctly identified by the model.

[0041] The trained LightGBM model is used to classify and predict the fused feature data to obtain risk assessment results.

[0042] Finally, the contribution of each feature to the risk score is quantified through the model interpretability module, such as the SHAP or LIME algorithm, providing intuitive interpretation results.

[0043] Optionally, in this embodiment, a more accurate risk assessment system is formed by introducing a dual-model framework based on time-series and static features, namely a first evaluation model (LSTM model) and a second evaluation model (LightGBM model). First, the LSTM model is responsible for capturing and learning the time-series characteristics of account transactions, such as the fluctuation patterns of transaction amounts and the periodicity of transaction times. By fusing its output third feature data (time-series dependent features) with the second feature data (account attributes), a more comprehensive perspective on account risk assessment can be constructed. Second, the fused features are input into the second evaluation model. This model, with its efficient processing of high-dimensional data, can fully consider all feature information, including time-series features and account attributes, and perform comprehensive analysis to output accurate risk assessment results.

[0044] This approach enables real-time risk assessment and rapid response to potential risks in account transactions. For new users or accounts with limited credit history, the system employs cold-start optimization strategies, such as transfer learning or sample-weighted loss functions, to effectively mitigate data sparsity issues and ensure the model's stability and accuracy across various account scenarios. Furthermore, the real-time inference mechanism ensures that the risk assessment process is triggered immediately upon a transaction, enabling financial institutions to react promptly.

[0045] The embodiments provided in this application employ a two-stage evaluation model to deeply integrate and analyze historical transaction behavior and account attribute information, significantly improving the accuracy and timeliness of account transaction risk assessment. Specifically, by using historical transaction behavior data of the account to be evaluated as the first feature data and inputting it into a pre-trained first evaluation model (such as an LSTM neural network model), the system can capture and analyze the temporal dependencies in the transaction data, generating third feature data that reflects transaction sequence patterns and dynamic characteristics. This modeling approach enables the system to identify abnormal transaction patterns and effectively predict potential risks even when transaction behavior changes over time. The account attribute information (second feature data) is combined with the third feature data output by the LSTM model and input into a second evaluation model (such as a LightGBM model) for comprehensive risk type prediction, ensuring good predictive stability even when processing high-dimensional data. The output of the second evaluation model, i.e., the risk assessment result, indicates the account's risk type, including but not limited to low risk, medium risk, and high risk. Through the synergistic effect of the first and second evaluation models, this embodiment achieves deep fusion of dynamic behavioral data and static attribute data. This integration not only fully utilizes the long-term dependencies in time-series data but also considers the complex impact of account attributes, thereby constructing a more comprehensive and accurate risk assessment framework and achieving the technical effect of improving the accuracy of account risk assessment.

[0046] As an optional approach, before obtaining the first and second feature data of the account association to be evaluated, the method further includes:

[0047] Obtain N datasets for training the first and second evaluation models, where N is a positive integer greater than 2;

[0048] Divide the N datasets into 1 initial test dataset and N-1 initial training datasets;

[0049] One first test dataset and N-2 initial training datasets are determined sequentially from N-1 initial training datasets, resulting in N-1 first test datasets and N-1 sets of first training datasets. Among them, one set of first training datasets and one first test dataset corresponding to one set of first training datasets constitute N-1 initial training datasets.

[0050] After training the first evaluation model using each set of first training datasets, the trained first evaluation model is used to validate each first test dataset corresponding to each set of first training datasets to obtain N-1 first validation results, and to validate the initial test dataset to obtain N-1 second validation results.

[0051] The second evaluation model is trained based on N-1 first validation results and N-1 second validation results.

[0052] Optionally, in this embodiment, N datasets refer to multiple dataset samples used to train the model, where N is a positive integer greater than 2, indicating that model training needs to be based on multiple datasets.

[0053] Optionally, in this embodiment, the initial test dataset is randomly selected from N datasets as the test set, used to evaluate the model's performance on unseen data. The initial training dataset consists of the remaining N-1 datasets, used to train the first evaluation model and the second evaluation model.

[0054] Optionally, in this embodiment, the first test dataset and the first training dataset are selected from N-1 initial training datasets, with each dataset chosen as the first test dataset from the remaining data, and the remaining data forming a set of first training datasets. This process is repeated N-1 times to ensure that each dataset has a chance to be used for testing.

[0055] Optionally, in this embodiment, the first verification result is the prediction result of the first evaluation model on the first test dataset; the second verification result refers to the output result of using these models to predict the initial test dataset after all models have been trained, which is used to finally evaluate the generalization ability of the model.

[0056] Optionally, in this embodiment, firstly, N datasets are collected, which contain historical transaction records, user attribute information, etc., for model training and validation. The value of N is usually based on the diversity and scale of the data to ensure that the model can learn from sufficient samples.

[0057] One dataset is randomly selected from N datasets as the initial test dataset for independent evaluation of the final model performance. The remaining N-1 datasets are divided into multiple training and test set combinations for cross-validation during model training.

[0058] Cross-validation is performed on N-1 initial training datasets. Specifically, for each fold of the dataset, it serves as the validation set (first test dataset), while the remaining datasets serve as the training set (first training dataset). In this way, the model can be trained on one dataset and validated on another, ultimately yielding N-1 first validation results, ensuring that the model's generalization ability is fully evaluated.

[0059] The first evaluation model is trained using the first training dataset for each group. After each training iteration, it is validated using the corresponding first test dataset, and the first validation results are collected. This process is repeated N-1 times to ensure that each dataset participates in model training and validation. Then, the second evaluation model is trained using the N-1 first validation results and the validation results of an independent initial test dataset (the second validation results). This step further optimizes model performance through an ensemble learning method.

[0060] The second evaluation model is trained based on N-1 first and second validation results. This step improves the model's stability and generalization ability by validating and training on multiple datasets. The final model's performance will be evaluated based on an independent initial test dataset to ensure the model's predictive accuracy and reliability on unknown data.

[0061] The embodiments provided in this application, through multi-fold cross-validation ensemble learning, significantly improve the model's generalization ability and prediction accuracy, especially in the scenario of transaction risk assessment, enabling it to more robustly address the cold start problem for new users. Furthermore, final validation on independent test datasets ensures the model's reliability and effectiveness in practical applications, providing financial institutions with a highly accurate real-time risk control solution.

[0062] As an optional approach, the second evaluation model is trained based on N-1 first validation results and N-1 second validation results, including:

[0063] The average verification result corresponding to N-1 second verification results is determined as the second test dataset;

[0064] The set corresponding to the N-1 first verification results is determined as the second training dataset;

[0065] The second evaluation model was trained using the second training dataset.

[0066] The trained second evaluation model was used to validate the second test dataset.

[0067] Optionally, in this embodiment, firstly, the first validation results from N-1 first test datasets are aggregated to obtain a second training dataset containing the temporal feature extraction performance of the LSTM model. Simultaneously, the N-1 second validation results are averaged to eliminate bias caused by dataset differences, resulting in a second test dataset representing the average performance of the model on independent data.

[0068] The second evaluation model was trained using a second training dataset. Because the second training dataset incorporates the LSTM model's deep understanding of time-series features, the LightGBM model can be trained to more comprehensively consider the impact of dynamic trading behavior on risk assessment, thereby improving the model's overall predictive ability.

[0069] After training, the trained second evaluation model is finally validated using an independent second test dataset (i.e., the average result of N-1 second validation results). This validation process not only evaluates the model's predictive ability on the initial test dataset, but also examines the model's stability and robustness when integrating multi-source data for risk assessment.

[0070] By comparing the prediction results of the second evaluation model on the second test dataset with its true labels, the model's accuracy, recall, F1 score, and other metrics can be calculated to further evaluate the model's predictive performance, especially its generalization ability on unknown data.

[0071] Through the embodiments provided in this application, during the in-depth training process, the second training dataset contains the LSTM model's profound understanding of dynamic trading behavior, enabling the LightGBM model to learn a more comprehensive risk assessment pattern based on these time-dependent features and account attribute information. Furthermore, by calculating the average of N-1 second validation results, we can obtain a second test dataset representing the model's average performance on independent data. This dataset is then used to validate the in-depth trained second evaluation model to assess its generalization ability and predictive accuracy on unknown data. The entire validation process not only examines the model's adaptability to dynamically changing data but also ensures that the model can stably handle different types of risk features, ultimately outputting a high-quality risk assessment result. Through multi-level model training and validation, the comprehensive predictive ability of the second evaluation model is effectively enhanced, especially when dealing with complex and volatile account risk scenarios, significantly improving the model's accuracy and stability.

[0072] As an optional approach, first and second characteristic data of the account associations to be evaluated are obtained, including:

[0073] Obtain the account's historical transaction data, which includes transaction information corresponding to the transaction operations triggered by the account. The transaction information includes the operation subject information, operation time information, operation evaluation information, and operation resource information of the transaction operation.

[0074] Feature extraction is performed on historical transaction data to obtain the first feature data.

[0075] Optionally, in this embodiment, historical transaction data refers to all transaction records executed by the account in the past, including but not limited to account transactions, transfers, deposits, withdrawals, etc., which is an important basis for assessing account risk.

[0076] Optionally, in this embodiment, the operation subject information is the information of the initiator or participant of the transaction operation, such as the basic information of the account holder and the identification information of the counterparty.

[0077] Optionally, in this embodiment, the operation time information is the specific time when the transaction occurs, which is used to analyze the periodicity and time series characteristics of the transaction.

[0078] Optionally, in this embodiment, the operation evaluation information may include feedback or ratings on transactions, such as customer reviews of goods in e-commerce transactions, and credit scores of customers during the transaction process, which helps to understand the credit behavior patterns of accounts.

[0079] Optionally, in this embodiment, the operation resource information includes: details of the resources involved in the transaction, such as the transaction amount, transaction type (purchase, return, repayment, etc.), and the flow of funds involved in the transaction.

[0080] Optionally, in this embodiment, the historical transaction records of the account to be evaluated are first retrieved from the database. These records contain detailed information about all transaction operations, such as the subject of the operation, the time of the operation, the transaction amount, and the transaction type.

[0081] Historical transaction data is preprocessed, including data cleaning (removing invalid or abnormal data), data integration (ensuring all transaction information is complete and consistent), and data encoding (converting non-numerical features into a format that the model can understand).

[0082] On the preprocessed data, a series of data mining and feature engineering techniques are applied to extract key features that reflect account transaction behavior patterns. For example, the frequency of transactions within a specific time window is calculated, the distribution of transaction amounts is analyzed, and the periodic patterns of transaction times are explored to form the first feature data.

[0083] During the feature extraction stage, a large number of features may be generated. At this time, it is necessary to optimize and select the features through methods such as feature importance analysis and correlation analysis to ensure that the features input into the model are both representative and can efficiently reflect the risk status of the account.

[0084] As an optional approach, first and second characteristic data of the account associations to be evaluated are obtained, including:

[0085] Obtain account attribute data, including historical attribute data such as account subject information, account rating information, and account resource information;

[0086] Feature extraction is performed on historical attribute data to obtain second feature data.

[0087] Optionally, in this embodiment, account attribute data refers to other relevant information about the account besides transaction data, covering static or semi-static characteristics such as the account holder's personal information, credit score, and account activity. Account subject information includes the account holder's basic personal information, such as name, age, gender, occupation, and income level, which is an important basis for assessing personal credit. Account evaluation information comes from the account holder's historical credit behavior, such as credit score, repayment status of past loans, account usage limits, and delinquency rates. This information reflects the account's credit history and repayment ability. Account resource information involves the account holder's financial situation, such as deposit balance, investment status, and real estate holdings. This information helps financial institutions assess the account's debt repayment ability and asset status.

[0088] Optionally, in this embodiment, account attribute data is extracted from the account management system, including account subject information (such as personal basic information), account evaluation information (such as credit score and historical repayment records), and account resource information (such as asset status). This data provides static feature input for the second evaluation model.

[0089] The acquired account attribute data is preprocessed, including data cleaning, feature encoding, and standardization, to ensure that the data quality meets the model input requirements.

[0090] By applying feature engineering techniques to conduct in-depth analysis of account attribute data, features highly correlated with account risk assessment, such as credit score, income stability, and asset ownership, are identified to form secondary feature data.

[0091] By analyzing feature correlation and calculating feature importance, the second feature data is screened and optimized to ensure that the model training only includes the features with the most predictive value, thus avoiding overfitting and interference from redundant information.

[0092] The embodiments provided in this application, based on the acquisition of first feature data (historical transaction behavior), further focus on the collection and feature extraction of account attribute data to ensure that the second evaluation model (such as the LightGBM model) can comprehensively consider the static attributes and credit history of the account.

[0093] As an optional approach, after inputting the second and third feature data into the second assessment model and obtaining the account risk assessment result output by the second assessment model, the method further includes:

[0094] If the risk assessment indicates that the risk type is high, the account is prohibited from triggering transaction events;

[0095] If the risk assessment results indicate that the risk type is medium risk, add transaction restrictions to the transaction event when the account triggers a transaction event;

[0096] If the risk assessment results indicate that the risk type is low, the account is allowed to trigger transaction events.

[0097] Optionally, in this embodiment, when the model's predicted score is higher than a preset high-risk threshold, the account is marked as high-risk. These accounts may pose a higher risk and require strict monitoring and management. Accounts with model predicted scores within the medium-risk threshold range require certain transaction restrictions or additional verification processes compared to low-risk accounts to balance risk and service experience. If the model's predicted score is lower than the medium-risk threshold, the account is considered low-risk; these accounts experience the fewest restrictions during transactions and enjoy a smoother service experience.

[0098] Optionally, in this embodiment, when a medium-risk account makes a transaction, the system may add additional audit conditions or operational constraints, such as requiring secondary confirmation, adding identity verification steps, or limiting the transaction amount or frequency, in order to reduce potential risks.

[0099] After the second assessment model generates the risk assessment results, the system analyzes the results to determine the risk type of the account, namely, high risk, medium risk, or low risk.

[0100] For accounts identified as high-risk, the system immediately takes measures to prevent the account from triggering any trading events until manual review or risk mitigation strategies are implemented.

[0101] If an account is classified as medium risk, the system will automatically impose transaction restrictions when the account triggers a transaction event, such as limiting transaction amounts or requiring additional identity verification, to ensure transaction security.

[0102] For low-risk accounts, the system will allow them to trigger trading events freely without additional trading restrictions, thus providing a better customer experience.

[0103] The embodiments provided in this application ensure that risk assessment results can be quickly translated into specific risk control actions, and reasonable management measures can be taken for accounts with different risk levels, thereby maximizing the protection of the rights and interests of normal trading users and achieving dual optimization of risk management and user experience.

[0104] As an optional approach, the second and third feature data are input into the second assessment model to obtain the account risk assessment results output by the second assessment model, including:

[0105] The second and third feature data are input into multiple second assessment models to obtain multiple risk types output by multiple second assessment models;

[0106] Obtain the first average risk type corresponding to multiple risk types;

[0107] From multiple risk types, remove the risk types that differ more than the expected degree of difference from the first average risk type, and determine the second average risk type corresponding to the remaining risk types as the risk assessment result.

[0108] Optionally, in this embodiment, multiple second evaluation models refer to multiple pre-trained LightGBM models that receive the same or different combinations of features to enhance the robustness and prediction accuracy of the models.

[0109] Optionally, in this embodiment, the second feature data (account attribute information) and the third feature data (time-dependent features of transaction behavior) are input into multiple pre-trained second evaluation models (such as multiple LightGBM model instances), and each model outputs a risk type prediction result based on its learned weights and structure.

[0110] Collect all risk types output by the second assessment model, calculate their arithmetic mean as a preliminary average risk type judgment, and provide a benchmark for subsequent exclusion of abnormal results.

[0111] Based on the first average risk type, a threshold for the expected degree of difference is set. Risk types whose predicted results differ from the average risk type by more than this threshold are considered abnormal and excluded, aiming to reduce errors caused by model overfitting or abnormal predictions of specific samples.

[0112] After removing outliers, the remaining risk types are averaged again to generate a second average risk type, which serves as the final risk assessment result and is provided to the risk decision engine for subsequent risk control decisions.

[0113] The embodiments provided in this application fully utilize the concept of ensemble learning. By working collaboratively and fusing the results of multiple models, the accuracy of risk assessment is significantly improved, and the risk misjudgment caused by the prediction bias of a single model is reduced. This provides strong technical support for financial institutions in controlling account transaction risks and ensures the scientific nature and effectiveness of risk control decisions.

[0114] This invention applies the aforementioned account risk assessment method to the scenario of handling account transaction risks in the financial field. In this scenario, traditional risk assessment models typically require manual feature extraction, which is not only time-consuming and labor-intensive but may also miss some important features. Traditional risk assessment methods struggle to cope with the challenges posed by large-scale and diverse data. When faced with nonlinear relationships and high-dimensional data, traditional methods are prone to overfitting or underfitting, resulting in poor model generalization ability.

[0115] To address the shortcomings of existing solutions, this invention overcomes the deficiencies of the prior art by providing an account transaction risk assessment method based on a LightGBM+LSTM fusion model. This method automatically learns features, reducing the workload of manual feature engineering. Simultaneously, the fusion model can handle complex data structures and capture nonlinear relationships between data. Through this deep learning fusion model, the accuracy of risk assessment in real-time account transactions can be improved, reducing the false positive rate. Deep learning models are better suited to large-scale data and can extract useful information from it. Furthermore, the real-time inference mechanism enables real-time monitoring of account transactions and rapid identification of abnormal transaction behavior, which helps banks take timely measures to reduce potential economic losses.

[0116] Optionally, this embodiment uses a stacking model combining LightGBM and LSTM (Long Short-Term Memory) to assess the risk of real-time account transactions. This combination leverages the efficiency of LightGBM and the ability of LSTM to process sequential data, enabling the model to capture both static customer characteristics and dynamic characteristics that change over time.

[0117] This embodiment establishes a LightGBM-based customer credit risk assessment method model based on gradient boosting decision tree theory, efficiently classifying and predicting risk levels. For the temporal features present in the customer risk assessment process, this paper establishes an LSTM prediction model to deeply mine these features. To comprehensively explore data features, this paper fuses LightGBM and LSTM models, combining the advantages of both to establish a stacking-based customer credit risk assessment method model, and improves the fused model with an algorithm based on individual learner accuracy weighting. Overall, the three models exhibit good delay prediction capabilities, with the improved stacking fusion model showing the best prediction accuracy and generalization ability.

[0118] Specifically, the steps for real-time account transaction risk assessment based on LightGBM+LSTM are as follows:

[0119] Data collection and preprocessing:

[0120] Collect static information data, including the account holder's basic personal information (age, gender, marital status, etc.), financial information (income, debt, deposits, etc.), and credit history (credit score, historical loan records, etc.).

[0121] Collect historical credit data: the user's past credit history, such as credit report score, repayment history of the bank account, number of overdue payments, etc.

[0122] Collect dynamic behavioral data: such as time-series data of account transaction records (spending amount, repayment date, number of overdue days, etc.), which need to be arranged in chronological order.

[0123] Collect relationship network data: real-time or historical transaction behavior logs of users, including transaction time, amount, location, merchant type, APP operation records, etc.

[0124] Collect third-party data: Information from external data sources that are compliant with regulations (such as operator data, e-commerce behavior data, and social security payment records).

[0125] Feature engineering is performed on the collected data:

[0126] Encode non-numerical features (e.g., one-hot encoding, label encoding); standardize or normalize numerical features. Extract time-series features, such as average, maximum, minimum, and standard deviation of consumption amounts; and calculate consumption trends (e.g., monthly growth / decline rate).

[0127] Building the model architecture:

[0128] LSTM model: Input dynamic time series data, use LSTM layers to capture long-term dependencies in the time series. The output of the LSTM layer can be the state of the last time step or the state of the entire sequence.

[0129] The LightGBM model combines static features and the output of an LSTM layer as input to construct a LightGBM model. LightGBM is an efficient gradient boosting framework that can handle large-scale datasets and has good predictive performance.

[0130] Model training and parameter tuning:

[0131] The model is trained using the training dataset. For the LSTM part, the loss function is set to cross-entropy loss and the Adam optimizer. For the LightGBM part, hyperparameters (such as learning rate, tree depth, number of leaf nodes, etc.) are tuned to optimize model performance.

[0132] Cross-validation is used to evaluate the model's generalization ability, and necessary hyperparameter tuning is performed.

[0133] Model evaluation:

[0134] Evaluate the model's performance on independent test datasets, using metrics such as accuracy, precision, recall, and F1 score to measure its quality. Visualize the confusion matrix to understand the model's performance across different categories.

[0135] Model Deployment and Monitoring:

[0136] Deploy the trained model to the production environment for real-time or batch assessment of customer credit risk. Regularly monitor model performance, collect feedback data, and update the model as needed.

[0137] It should be noted that for the LightGBM model, feature importance can be used to explain which factors have the greatest impact on credit risk. For the LSTM model, although the interpretability is poor, some understanding can be gained by visualizing the changes in the hidden states.

[0138] This approach combines LightGBM's powerful classification capabilities with LSTM's advantages in processing time-series data, enabling the construction of a comprehensive and effective real-time account transaction risk assessment system.

[0139] Optionally, in this embodiment, a schematic diagram of a model training method based on the stacking algorithm is shown below. Figure 3 As shown, the stacking method mainly consists of a primary learner in the first layer and a meta learner in the second layer. Stacking first trains the original features using the primary learner, then uses the predicted values ​​output by the primary learner as the input features of the meta learner, and uses the original feature labels as new labels to form a new data feature, which is then further trained in the meta learner.

[0140] In the stacking model fusion process, to prevent overfitting caused by repeated learning on the training set, K-fold cross-validation is used to train the above models. For example, a schematic diagram of a cross-validation process (K=5 folds) is shown below. Figure 4 As shown, the dataset is first divided into training and test sets. Then, the training set is randomly divided into five parts without replacement, with one part used for prediction and the remaining four parts used for training the learner model. Next, the first-layer learner predicts the output of each prediction dataset in the training set, resulting in five predicted values. The algorithm then combines these newly generated predicted values ​​using a simple average to form a new feature dataset. Finally, the meta-learner trains based on the new feature dataset and the newly generated test set, outputting the final result. Model 1 can be, but is not limited to, the LightGBM model, and Model 2 can be, but is not limited to, the LSTM model.

[0141] The stacking technology that integrates LightGBM and LSTM can bring multiple benefits to customer credit risk assessment. LSTM excels at processing time-series data and can capture dynamic customer behavior patterns; while LightGBM performs exceptionally well when dealing with large amounts of static features. Combining the two can fully utilize various types of feature information, improving the overall predictive accuracy of the model. By more accurately identifying high-risk customers, financial institutions can take more targeted risk management measures.

[0142] Accurate risk assessment can reduce unnecessary waste of credit resources, enabling financial institutions to reduce lending to high-risk customers while increasing credit support for low-risk customers. Automated and intelligent risk assessment processes reduce the workload of manual review and improve business processing speed.

[0143] More detailed risk assessments allow for customized financial products and services to be offered to clients with different risk appetites. Automated assessment systems enable faster approval processes and improve customer satisfaction.

[0144] In summary, the technical solution that integrates LightGBM and LSTM can improve prediction accuracy, optimize resource allocation, enhance competitiveness, improve user experience, strengthen regulatory compliance and risk management, and promote data-driven decision-making.

[0145] Example 2

[0146] The account risk assessment device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in the above embodiment one. The specific implementation method and beneficial effects can be referred to the aforementioned method embodiment, and will not be repeated here.

[0147] Figure 5 This is a schematic diagram of an optional account risk assessment device according to an embodiment of the present invention, such as... Figure 5 As shown, the risk assessment device for this account may include:

[0148] The acquisition unit 502 is used to acquire first feature data and second feature data associated with the account to be evaluated, wherein the first feature data is used to indicate the account's historical transaction behavior and the second feature data is used to indicate the account's account attributes.

[0149] The first evaluation unit 504 is used to input the first feature data into the first evaluation model to obtain the third feature data output by the first evaluation model. The first evaluation model is a pre-trained neural network model used to capture the temporal dependencies of the input data, and the third feature data is used to indicate the temporal dependencies of the first feature data.

[0150] The second evaluation unit 506 is used to input the second feature data and the third feature data into the second evaluation model to obtain the risk evaluation result of the account output by the second evaluation model. The second evaluation model is a pre-trained neural network model used to predict the risk type of the account based on the input data. The risk evaluation result is used to indicate the risk type of the account.

[0151] As an optional solution, the device also includes:

[0152] The first acquisition module is used to acquire N datasets for training the first and second evaluation models before acquiring the first feature data and the second feature data associated with the account to be evaluated, where N is a positive integer greater than 2.

[0153] The partitioning module is used to divide N datasets into one initial test dataset and N-1 initial training datasets before obtaining the first and second feature data of the account associations to be evaluated.

[0154] The first determining module is used to determine one first test dataset and N-2 initial training datasets sequentially from N-1 initial training datasets before obtaining the first feature data and second feature data of the account association to be evaluated, so as to obtain N-1 first test datasets and N-1 sets of first training datasets. Among them, one set of first training datasets and one first test dataset corresponding to one set of first training datasets constitute N-1 initial training datasets.

[0155] The verification module is used to verify each first test dataset corresponding to each first training dataset using the trained first evaluation model after training the first evaluation model with each set of first training datasets before obtaining the first feature data and second feature data associated with the account to be evaluated, to obtain N-1 first verification results, and to verify the initial test dataset to obtain N-1 second verification results.

[0156] The training module is used to train the second evaluation model based on N-1 first verification results and N-1 second verification results before obtaining the first feature data and second feature data of the account association to be evaluated.

[0157] As an optional approach, the first training module includes:

[0158] The first determining submodule is used to determine the average verification result corresponding to N-1 second verification results as the second test dataset;

[0159] The second determination submodule is used to determine the set corresponding to N-1 first verification results as the second training dataset;

[0160] The training submodule is used to train the second evaluation model using the second training dataset.

[0161] The validation submodule is used to validate the second test dataset using the trained second evaluation model.

[0162] As an optional solution, the acquisition unit 502 is characterized by comprising:

[0163] The second acquisition module is used to acquire the account's historical transaction data. The historical transaction data includes transaction information corresponding to the transaction operations triggered by the account. The transaction information includes the operation subject information, operation time information, operation evaluation information, and operation resource information of the transaction operation.

[0164] The first extraction module is used to extract features from historical transaction data to obtain the first feature data.

[0165] As an optional solution, the acquisition unit 502 includes:

[0166] The third acquisition module is used to acquire account attribute data, including historical attribute data such as account subject information, account rating information, and account resource information.

[0167] The second extraction module is used to extract features from historical attribute data to obtain second feature data.

[0168] As an optional solution, the device also includes:

[0169] The first control module is used to prevent the account from triggering transaction events after inputting the second feature data and the third feature data into the second evaluation model and obtaining the risk assessment result of the account output by the second evaluation model.

[0170] The second control module is used to add transaction restriction conditions to the transaction event when the account triggers a transaction event, after inputting the second feature data and the third feature data into the second evaluation model and obtaining the risk assessment result of the account output by the second evaluation model, if the risk assessment result indicates that the risk type is medium risk.

[0171] The third control module is used to allow the account to trigger a transaction event after the second feature data and the third feature data are input into the second evaluation model and the risk assessment result of the account is output by the second evaluation model, provided that the risk assessment result indicates that the risk type is low.

[0172] As an optional solution, the second evaluation unit 506 includes:

[0173] The input module is used to input the second feature data and the third feature data into multiple second assessment models to obtain multiple risk types output by the multiple second assessment models;

[0174] The fourth acquisition module is used to acquire the first average risk type corresponding to multiple risk types;

[0175] The second determining module is used to remove risk types from multiple risk types that differ from the first average risk type by a greater than expected degree of difference, and to determine the second average risk type corresponding to the remaining risk types as the risk assessment result.

[0176] The risk assessment device for the aforementioned account may also include a processor and a memory. The aforementioned acquisition unit 502, first assessment unit 504, and second assessment unit 506 are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0177] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and account risk assessment can be implemented by adjusting kernel parameters.

[0178] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0179] Example 3

[0180] Embodiments of this application may provide an electronic device. Figure 6 This is a structural block diagram of an electronic device for performing an account risk assessment method according to an embodiment of this application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (Only one is shown) Processor 602, memory 604, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0181] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the account risk assessment method and apparatus in this application embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned account risk assessment method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0182] The processor can invoke information and applications stored in the memory via the transmission device to perform the following steps: in response to a first backup request triggered by a first resource of a first shared client, verifying the resource type of the first resource; if the resource type of the first resource is a shared resource type, determining the target source resource corresponding to the first resource on the shared server associated with the first shared client, wherein the shared server is used to share at least one source resource with at least two clients, the at least one source resource includes the target source resource, and the at least two clients include the first shared client; and copying the target source resource from the shared server to the backup server.

[0183] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 6 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 6 The different configurations shown.

[0184] Those skilled in the art will understand that all or part of the steps in the risk assessment methods for various accounts in the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0185] Example 4

[0186] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the account risk assessment method provided in Embodiment 1.

[0187] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the account risk assessment method of any one of the above embodiments.

[0188] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0189] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the account risk assessment method in various embodiments of this application.

[0190] This application also provides a computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the account risk assessment method in various embodiments of this application.

[0191] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0192] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0193] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0194] The units described as separate components may or may not be physically separate. Similarly, the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0195] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0196] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0197] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for risk assessment of an account, characterized in that, include: Obtain first feature data and second feature data associated with the account to be evaluated, wherein the first feature data is used to indicate the account's historical transaction behavior and the second feature data is used to indicate the account's account attributes; The first feature data is input into the first evaluation model to obtain the third feature data output by the first evaluation model. The first evaluation model is a pre-trained neural network model used to capture the temporal dependencies of the input data, and the third feature data is used to indicate the temporal dependencies of the first feature data. The second feature data and the third feature data are input into the second evaluation model to obtain the risk assessment result of the account output by the second evaluation model. The second evaluation model is a pre-trained neural network model used to predict the risk type of the account based on the input data. The risk assessment result is used to indicate the risk type of the account.

2. The method according to claim 1, characterized in that, Before obtaining the first and second feature data associated with the account to be evaluated, the method further includes: Obtain N datasets for training the first evaluation model and the second evaluation model, where N is a positive integer greater than 2; The N datasets are divided into one initial test dataset and N-1 initial training datasets; One first test dataset and N-2 initial training datasets are determined sequentially from the N-1 initial training datasets to obtain N-1 first test datasets and N-1 sets of first training datasets. Among them, a set of first training datasets and a first test dataset corresponding to the set of first training datasets constitute the N-1 initial training datasets. After training the first evaluation model using each set of first training datasets, the trained first evaluation model is used to validate each of the first test datasets corresponding to each set of first training datasets to obtain N-1 first validation results, and the initial test dataset is validated to obtain N-1 second validation results. The second evaluation model is trained based on the N-1 first verification results and the N-1 second verification results.

3. The method according to claim 2, characterized in that, The step of training the second evaluation model based on the N-1 first verification results and the N-1 second verification results includes: The average verification result corresponding to the N-1 second verification results is determined as the second test dataset; The set corresponding to the N-1 first verification results is determined as the second training dataset; The second evaluation model is trained using the second training dataset; The trained second evaluation model was used to validate the second test dataset.

4. The method according to any one of claims 1 to 3, characterized in that, The acquisition of the first and second feature data associated with the account to be evaluated includes: Obtain the historical transaction data of the account, wherein the historical transaction data includes transaction information corresponding to the transaction operations triggered by the account, and the transaction information includes the operation subject information, operation time information, operation evaluation information and operation resource information of the transaction operation; The historical transaction data is used to extract features to obtain the first feature data.

5. The method according to any one of claims 1 to 3, characterized in that, The acquisition of the first and second feature data associated with the account to be evaluated includes: Obtain the account attribute data of the account, wherein the historical attribute data includes the account subject information, account rating information and account resource information of the account; The historical attribute data is subjected to feature extraction to obtain the second feature data.

6. The method according to any one of claims 1 to 3, characterized in that, After inputting the second feature data and the third feature data into the second assessment model to obtain the risk assessment result of the account output by the second assessment model, the method further includes: If the risk assessment result indicates that the risk type is high-risk, the account is prohibited from triggering transaction events; If the risk assessment result indicates that the risk type is medium risk, when the account triggers the transaction event, transaction restrictions are added to the transaction event. If the risk assessment result indicates that the risk type is low, the account is allowed to trigger the transaction event.

7. The method according to any one of claims 1 to 3, characterized in that, The step of inputting the second feature data and the third feature data into the second assessment model to obtain the risk assessment result of the account output by the second assessment model includes: The second feature data and the third feature data are input into multiple second assessment models to obtain multiple risk types output by multiple second assessment models; Obtain the first average risk type corresponding to the multiple risk types; From the multiple risk types, remove the risk types that differ from the first average risk type by a greater than expected degree, and determine the second average risk type corresponding to the remaining risk types as the risk assessment result.

8. A risk assessment device for an account, characterized in that, include: The acquisition unit is used to acquire first feature data and second feature data associated with the account to be evaluated, wherein the first feature data is used to indicate the account's historical transaction behavior and the second feature data is used to indicate the account's account attributes. The first evaluation unit is used to input the first feature data into the first evaluation model to obtain the third feature data output by the first evaluation model. The first evaluation model is a pre-trained neural network model used to capture the temporal dependencies of the input data, and the third feature data is used to indicate the temporal dependencies of the first feature data. The second assessment unit is used to input the second feature data and the third feature data into the second assessment model to obtain the risk assessment result of the account output by the second assessment model. The second assessment model is a pre-trained neural network model used to predict the risk type of the account based on the input data. The risk assessment result is used to indicate the risk type of the account.

9. An electronic device, characterized in that, The method includes one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

Images