Financial risk assessment method and device based on machine learning, equipment and medium

Abstract

The invention relates to a financial risk assessment method and device based on machine learning, equipment and a medium. The method comprises the following steps: acquiring user dynamic archive data and user static archive data; inputting the user static archive data into a risk tolerance grade grading machine learning model to generate user risk tolerance grade information; inputting the user dynamic archive data into a behavior risk level grading machine learning model to generate user behavior risk level information; and generating user financial risk assessment information based on the user risk tolerance level information and the user behavior risk level information. By adopting the method, intelligent static risk bearing and dynamic behavior risk fusion assessment can be realized, and the comprehensiveness and accuracy of risk assessment are improved.

Description

Technical Field

[0001] This application belongs to the field of financial risk assessment, and in particular relates to a method, apparatus, equipment and medium for financial risk assessment based on machine learning. Background Technology

[0002] Financial risk assessment, as a core risk control component of financial institutions, essentially involves comprehensively evaluating a user's creditworthiness, behavioral characteristics, and ability to fulfill obligations to identify potential issues such as credit defaults, fraudulent transactions, and compliance risks. Financial risk assessment technology is deeply intertwined with the development of financial business models, data acquisition capabilities, and data processing algorithms. Early financial risk control relied on manual review and rule engines, with risk control personnel making decisions based on fixed thresholds. However, with the widespread adoption of digital finance, these limitations have become increasingly apparent.

[0003] With the explosive growth of mobile payment, internet lending, and robo-advisory services, the static data that early risk control relied on could only reflect a user's "historical credit status," while a user's real-time actions in financial scenarios can more directly reflect their current risk intentions.

[0004] In the prior art disclosed in CN113298373A, a financial risk assessment method is disclosed. This prior art can improve the accuracy of financial risk assessment through target variable screening, sub-model fusion and contribution analysis. However, it does not distinguish between the dynamic and static characteristics of customer data and only uses a single model to uniformly assess various types of behavioral information. It cannot accurately capture the risk patterns behind different types of data, and the assessment accuracy is limited, making it difficult to meet the needs of financial business for multi-dimensional risk assessment. Summary of the Invention

[0005] Therefore, it is necessary to provide a machine learning-based financial risk assessment method, device, equipment, and medium that can accurately characterize users' risk tolerance and real-time behavioral risk, and improve the comprehensiveness, relevance, and accuracy of financial risk assessment, in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a machine learning-based financial risk assessment method, including:

[0007] Retrieve user dynamic profile data and user static profile data;

[0008] User static profile data is input into a risk tolerance level classification machine learning model to generate user risk tolerance level information;

[0009] User dynamic profile data is input into a behavioral risk level classification machine learning model to generate user behavioral risk level information;

[0010] User financial risk assessment information is generated based on user risk tolerance level information and user behavior risk level information.

[0011] In one embodiment, acquiring user dynamic profile data and user static profile data includes:

[0012] Perform a data authorization agreement check on the target user. If the data authorization agreement check result shows that the data authorization agreement is complete and correct, generate an authorization status confirmation label.

[0013] Based on the authorization status confirmation tag, the user credit information, user qualification information and user financial information of the target user are obtained, and the user credit information, user qualification information and user financial information are input into the feature parameter extraction natural language processing model to extract user credit feature parameters, user qualification feature parameters and user financial feature parameters.

[0014] Based on user credit characteristics, user qualification characteristics, and user financial characteristics, static user profile data is constructed.

[0015] Based on the authorization status confirmation tag, user financial behavior information is collected. Based on the user financial behavior information and combined with the preset financial behavior analysis knowledge graph, the financial behavior feature parameters of the user financial behavior information are extracted.

[0016] Based on the financial behavior characteristic parameters within the preset behavior analysis time window, time series data of financial behavior characteristic parameters is constructed, and user dynamic profile data is constructed based on the time series data of financial behavior characteristic parameters.

[0017] In one embodiment, target users include natural person target users and legal person target users;

[0018] User credit information includes credit information of natural persons and credit information of legal persons;

[0019] User qualification information includes qualification information for natural persons and qualification information for legal persons;

[0020] User financial information includes financial information of natural persons and financial information of legal persons.

[0021] In one embodiment, the natural person's credit information includes personal credit report information and personal guarantee information;

[0022] Corporate credit information includes corporate credit reports, contract default information, corporate bond records, and corporate guarantee information;

[0023] Information on a natural person's qualifications includes occupational information, personal identification information, and asset ownership information;

[0024] Legal entity qualification information includes enterprise certificate information and enterprise qualification certification information;

[0025] Financial information of natural persons includes personal income and expenditure information and personal asset information;

[0026] Corporate financial information includes corporate financial statements and corporate asset information.

[0027] In one embodiment, user dynamic profile data is input into a behavioral risk level classification machine learning model to generate user behavioral risk level information, including:

[0028] The time series data of financial behavior characteristic parameters are input into the time series feature extraction module of the behavioral risk level classification machine learning model to extract the time series features of financial behavior;

[0029] The time-series features of financial behavior are input into the time-series feature analysis module of the behavioral risk level classification machine learning model to generate user behavior risk level information.

[0030] In one embodiment, the machine learning-based financial risk assessment method further includes:

[0031] A temporal feature extraction module was constructed based on a long short-term memory network;

[0032] A time series feature analysis module was constructed based on the conditional random field model.

[0033] In one embodiment, the risk tolerance level classification machine learning model is a gradient boosting tree model, and the risk tolerance level classification machine learning model's first... The expression for the objective loss function of the round of iterations is:

[0034]

[0035]

[0036] In the formula, The first machine learning model for risk tolerance level classification The target loss function for each iteration, The total number of risk tolerance levels corresponding to the target risk tolerance level information. To constrain the first Round of iteration targeting the first The regularization term for the complexity of the regression tree trained on risk tolerance levels. For the first Round of iteration targeting the first Regression tree for risk tolerance level training and The penalty coefficient hyperparameter, For the first Round of iteration targeting the first The total number of leaf nodes in the regression tree for risk tolerance level training. For the first Round of iteration targeting the first The first regression tree in the risk tolerance level training The output weights of the leaf nodes For the sample size, For the first The sample belongs to the first The true label of risk tolerance level For the front The risk tolerance level classification machine learning model generated in round-by-round iterations The sample belongs to the first The sum of the predicted values ​​of risk tolerance levels, For the first The risk tolerance level classification machine learning model generated in round-by-round iterations The sample belongs to the first Predicted risk tolerance level.

[0037] Secondly, this application also provides a machine learning-based financial risk assessment device, comprising:

[0038] The user profile data acquisition module is used to acquire dynamic user profile data and static user profile data.

[0039] The risk tolerance level generation module is used to input user static profile data into the risk tolerance level classification machine learning model to generate user risk tolerance level information.

[0040] The behavior risk level generation module is used to input user dynamic profile data into the behavior risk level classification machine learning model to generate user behavior risk level information.

[0041] The user financial risk assessment module is used to generate user financial risk assessment information based on user risk tolerance level information and user behavior risk level information.

[0042] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method as described in any of the first aspects of this application.

[0043] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects of this application.

[0044] The aforementioned machine learning-based financial risk assessment methods, devices, equipment, and media, by processing user static profile data and user dynamic profile data respectively, and combining risk tolerance level classification models and behavioral risk level classification models, can achieve a two-dimensional deconstruction of user risk characteristics. Specifically, static profile data can accurately capture users' long-term stability characteristics, improving the ability to identify users' basic risk attributes, while dynamic profile data can dynamically identify users' short-term behavioral anomalies, thereby enhancing the efficiency of response to real-time risk events and improving the comprehensiveness, pertinence, and accuracy of financial risk assessment. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A flowchart illustrating a machine learning-based financial risk assessment method provided in one embodiment of this application. Figure 1 ;

[0047] Figure 2 A flowchart illustrating a machine learning-based financial risk assessment method provided in one embodiment of this application. Figure 2 ;

[0048] Figure 3 This is a schematic diagram of the structure of a machine learning-based financial risk assessment device provided in one embodiment of this application. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] In one exemplary embodiment of this application, such as Figure 1 As shown, a machine learning-based financial risk assessment method is provided. This embodiment illustrates the application of this method to a risk assessment terminal. It is understood that this method can also be applied to a risk assessment server, and further to a system including both a risk assessment terminal and a risk assessment server, and is implemented through the interaction between the risk assessment terminal and the risk assessment server. In this embodiment, the method includes the following steps:

[0051] Step S101: Obtain user dynamic profile data and user static profile data.

[0052] Specifically, the risk assessment terminal can check the data authorization agreement of the target user. If the check result shows that the agreement is complete and error-free, the risk assessment terminal can generate an authorization status confirmation tag. Based on the authorization status confirmation tag, the risk assessment terminal can obtain the target user's credit information, qualification information, and financial information. The risk assessment terminal can input the user's credit information, qualification information, and financial information into a feature parameter extraction natural language processing model to extract user credit feature parameters, qualification feature parameters, and financial feature parameters, and construct static user profile data. Based on the authorization status confirmation tag, the risk assessment terminal can collect user financial behavior information. Combining this with a preset financial behavior analysis knowledge graph, the risk assessment terminal can extract financial behavior feature parameters. According to a preset behavior analysis time window, the risk assessment terminal can construct time-series data of financial behavior feature parameters from these feature parameters, thereby generating dynamic user profile data.

[0053] Optionally, target users can include natural person target users and legal person target users.

[0054] Optionally, the risk assessment terminal can initiate a data authorization agreement check request to the target user. By connecting to the user authorization management system, it verifies whether the user has signed a complete and valid data usage authorization agreement. If the data authorization agreement check result is complete and error-free, the risk assessment terminal can generate an authorization status confirmation tag. This tag will serve as the core credential for subsequent data acquisition, ensuring that all data collection activities comply with compliance requirements.

[0055] Optionally, the risk assessment terminal can extract basic information of the target user, as well as pre-stored user credit information, user qualification information, and user financial information, from the internal core business system and / or customer management database corresponding to the authorization status confirmation tag based on the authorization status confirmation tag. The risk assessment terminal can also obtain user credit information, user qualification information, and user financial information from the external interface corresponding to the authorization status confirmation tag based on the authorization status confirmation tag.

[0056] Optionally, the risk assessment terminal can obtain various financial behavior information of the target user in a financial scenario based on the authorization status confirmation tag and the real-time log stream and transaction message queue of the business system corresponding to the authorization status confirmation tag. The risk assessment terminal can call a preset financial behavior analysis knowledge graph to perform correlation analysis and feature extraction on the collected financial behavior information to obtain the corresponding financial behavior feature parameters. The risk assessment terminal can also perform time-series integration of the financial behavior feature parameters within a preset behavior analysis time window to construct time-series data of financial behavior feature parameters. The risk assessment terminal can then construct user dynamic profile data based on the time-series data of financial behavior feature parameters.

[0057] Optionally, the financial behavior analysis knowledge graph may include user entities, behavior entities, scenario entities, and risk event entities. The financial behavior analysis knowledge graph may include relationships between entities, wherein the relationships may include, but are not limited to, "execution" relationships between user entities and behavior entities, "attribution" relationships between behavior entities and scenario entities, "trigger" relationships between behavior entities and risk event entities, and "association" relationships between user entities.

[0058] Step S102: Input the user's static profile data into the risk tolerance level classification machine learning model to generate the user's risk tolerance level information.

[0059] Specifically, the risk assessment terminal can input the completed user static profile data into a trained risk tolerance level classification machine learning model. The risk tolerance level classification machine learning model analyzes the user's long-term stable attributes and generates the user's risk tolerance level information.

[0060] Optionally, the risk tolerance level classification machine learning model can be installed on a risk assessment terminal, a risk assessment server, or a risk assessment system; there are no restrictions on this.

[0061] Optionally, user risk tolerance level information may include, but is not limited to, low risk tolerance level, lower risk tolerance level, medium risk tolerance level, higher risk tolerance level, and high risk tolerance level.

[0062] Optionally, the risk tolerance level classification machine learning model can be a gradient boosting tree model.

[0063] Step S103: Input the user dynamic profile data into the behavioral risk level classification machine learning model to generate user behavioral risk level information.

[0064] Specifically, the risk assessment terminal can input the constructed user dynamic profile data into the behavioral risk level classification machine learning model to conduct behavioral risk level classification assessment and generate user behavioral risk level information.

[0065] Optionally, the behavioral risk level classification machine learning model can be installed on a risk assessment terminal, a risk assessment server, or a risk assessment system; no specific restrictions are imposed here.

[0066] Optionally, the behavioral risk level classification machine learning model may include a time-series feature extraction module and a time-series feature analysis module.

[0067] Optionally, the temporal feature extraction module can be a machine learning model built on a long short-term memory network. The temporal feature analysis module can be a machine learning model built on a conditional random field model.

[0068] Optionally, user behavior risk level information may include, but is not limited to, low-risk level information, lower-risk level information, medium-risk level information, higher-risk level information, and high-risk level information.

[0069] Step S104: Generate user financial risk assessment information based on user risk tolerance level information and user behavior risk level information.

[0070] Specifically, the risk assessment terminal can generate user financial risk assessment information based on user risk tolerance level information and user behavioral risk level information. This user financial risk assessment information can include both user risk tolerance level information and user behavioral risk level information. Furthermore, it can include risk warning information and risk management suggestions generated when the user's behavioral risk level information exceeds their risk tolerance level information.

[0071] The aforementioned machine learning-based financial risk assessment method can construct a complete data system covering user risk attributes by using both dynamic and static user profile data. This systematically integrates static information reflecting long-term stable characteristics and dynamic behavioral data reflecting real-time risk status, thereby improving the accuracy of financial risk assessment. By constructing a risk tolerance level classification machine learning model based on the characteristics of static data, long-term risk tolerance can be accurately assessed. By constructing a behavioral risk level classification machine learning model based on the characteristics of dynamic data, real-time behavioral risks can be accurately captured. By adopting a dual-model architecture of dynamic and static data, the accuracy and stability of financial risk assessment can be improved.

[0072] In an optional embodiment of this application, please refer to Figure 1 and Figure 2Step S101, obtaining user dynamic profile data and user static profile data, may include:

[0073] Step S201: Perform a data authorization agreement check on the target user. If the data authorization agreement check result is that the data authorization agreement is complete and correct, generate an authorization status confirmation label.

[0074] Optionally, the authorization status confirmation label may include, but is not limited to, the authorization status confirmation label for natural persons' internal data, the authorization status confirmation label for natural persons' external credit information data, the authorization status confirmation label for legal persons' internal data, the authorization status confirmation label for legal persons' external qualification certification data, the authorization status confirmation label for general financial behavior dynamic data, and the authorization status confirmation label for natural persons' external financial auxiliary data.

[0075] Step S202: Based on the authorization status confirmation tag, obtain the target user's credit information, qualification information, and financial information, and input the user's credit information, qualification information, and financial information into the feature parameter extraction natural language processing model to extract user credit feature parameters, user qualification feature parameters, and user financial feature parameters.

[0076] Optionally, user credit characteristics parameters may include, but are not limited to, the number of historical defaults, the duration of overdue records, loan repayment records, credit limit, comprehensive credit score, amount of guarantee liability, frequency of credit inquiries, number of historical default records, bond repayment performance rate, and the credit risk transmission coefficient of related entities. User qualification characteristics parameters may include, but are not limited to, educational background, occupational category, years of work experience, industry experience, professional skills certification, professional qualification level, validity period of qualification certificates, occupational stability coefficient, industry access qualification level, document completeness score, clarity of asset ownership, paid-in capital ratio of enterprise registered capital, frequency of qualification certification updates, and coverage of core business licenses. User financial characteristics parameters may include, but are not limited to, income level, total assets, debt ratio, expenditure ratio, savings rate, return on investment, revenue profit margin, cash flow stability index, net fixed asset value ratio, and tax payment amount.

[0077] Step S203: Based on the user's credit characteristics, qualification characteristics, and financial characteristics, construct the user's static profile data.

[0078] Step S204: Based on the authorization status confirmation tag, collect user financial behavior information, and extract financial behavior feature parameters of user financial behavior information by combining the user financial behavior information with the preset financial behavior analysis knowledge graph.

[0079] Optionally, user financial behavior information may include, but is not limited to, information on financial product transactions, fund transfers, investment and financing decisions, insurance activities, lending, and savings.

[0080] Step S205: Based on the financial behavior characteristic parameters within the preset behavior analysis time window, construct the time series data of financial behavior characteristic parameters, and construct the user dynamic profile data based on the time series data of financial behavior characteristic parameters.

[0081] In an optional embodiment of this application, the target user can include natural person target users and legal person target users. User credit information can include natural person credit information and legal person credit information. User qualification information can include natural person qualification information and legal person qualification information. User financial information can include natural person financial information and legal person financial information.

[0082] In an optional embodiment of this application, natural person credit information may include personal credit report information and personal guarantee information. Legal person credit information may include enterprise credit report information, contract default information, enterprise bond records, and enterprise guarantee information. Natural person qualification information may include occupational information, personal identification information, and asset ownership information. Legal person qualification information may include enterprise identification information and enterprise qualification certification information. Natural person financial information may include personal income and expenditure information and personal asset information. Legal person financial information may include enterprise financial statements and enterprise asset information.

[0083] In an optional embodiment of this application, please refer to Figure 1 and Figure 2 Step S103 involves inputting user dynamic profile data into a behavioral risk level classification machine learning model to generate user behavioral risk level information, which may include:

[0084] Step S207: Input the time series data of financial behavior feature parameters into the time series feature extraction module in the behavioral risk level classification machine learning model to extract the time series features of financial behavior.

[0085] Step S208: Input the time-series features of financial behavior into the time-series feature analysis module in the behavioral risk level classification machine learning model to generate user behavior risk level information.

[0086] In an optional embodiment of this application, the machine learning-based financial risk assessment method further includes:

[0087] Specifically, the risk assessment terminal can be constructed based on a long short-term memory network to obtain a temporal feature extraction module.

[0088] Specifically, the risk assessment terminal can construct a time series feature analysis module based on the conditional random field model.

[0089] In an optional embodiment of this application, the risk tolerance level classification machine learning model can be a gradient boosting tree model, and the risk tolerance level classification machine learning model's first... The expression for the objective loss function of the round of iterations can be:

[0090]

[0091]

[0092] In the formula, The first machine learning model for risk tolerance level classification The target loss function for each iteration, The total number of risk tolerance levels corresponding to the target risk tolerance level information. To constrain the first Round of iteration targeting the first The regularization term for the complexity of the regression tree trained on risk tolerance levels. For the first Round of iteration targeting the first Regression tree for risk tolerance level training and The penalty coefficient hyperparameter, For the first Round of iteration targeting the first The total number of leaf nodes in the regression tree for risk tolerance level training. For the first Round of iteration targeting the first The first regression tree in the risk tolerance level training The output weights of the leaf nodes For the sample size, For the first The sample belongs to the first The true label of risk tolerance level For the front The risk tolerance level classification machine learning model generated in round-by-round iterations The sample belongs to the first The sum of the predicted values ​​of risk tolerance levels, For the first The risk tolerance level classification machine learning model generated in round-by-round iterations The sample belongs to the first Predicted risk tolerance level.

[0093] In one exemplary embodiment of this application, such as Figure 2As shown, a machine learning-based financial risk assessment method is provided, including:

[0094] Step S201: Perform a data authorization agreement check on the target user. If the data authorization agreement check result is that the data authorization agreement is complete and correct, generate an authorization status confirmation label.

[0095] Step S202: Based on the authorization status confirmation tag, obtain the target user's credit information, qualification information, and financial information, and input the user's credit information, qualification information, and financial information into the feature parameter extraction natural language processing model to extract user credit feature parameters, user qualification feature parameters, and user financial feature parameters.

[0096] Step S203: Based on the user's credit characteristics, qualification characteristics, and financial characteristics, construct the user's static profile data.

[0097] Step S204: Based on the authorization status confirmation tag, collect user financial behavior information, and extract financial behavior feature parameters of user financial behavior information by combining the user financial behavior information with the preset financial behavior analysis knowledge graph.

[0098] Step S205: Based on the financial behavior characteristic parameters within the preset behavior analysis time window, construct the time series data of financial behavior characteristic parameters, and construct the user dynamic profile data based on the time series data of financial behavior characteristic parameters.

[0099] Step S206: Input the user's static profile data into the risk tolerance level classification machine learning model to generate the user's risk tolerance level information.

[0100] Step S207: Input the time series data of financial behavior feature parameters into the time series feature extraction module in the behavioral risk level classification machine learning model to extract the time series features of financial behavior.

[0101] Step S208: Input the time-series features of financial behavior into the time-series feature analysis module in the behavioral risk level classification machine learning model to generate user behavior risk level information.

[0102] Step S209: Generate user financial risk assessment information based on user risk tolerance level information and user behavior risk level information.

[0103] The aforementioned machine learning-based financial risk assessment method integrates static and dynamic data to construct a comprehensive user financial risk profile, thereby improving the compliance, comprehensiveness, accuracy, and timeliness of risk assessment and providing a reliable basis for financial decision-making.

[0104] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0105] Based on the same inventive concept, this application also provides a machine learning-based financial risk assessment device for implementing the aforementioned machine learning-based financial risk assessment method. The solution provided by this device is similar to the implementation described in the above method; therefore, the specific limitations of one or more machine learning-based financial risk assessment device embodiments provided below can be found in the limitations of the machine learning-based financial risk assessment method described above, and will not be repeated here.

[0106] In one exemplary embodiment, such as Figure 3 As shown, a machine learning-based financial risk assessment device 300 is provided, comprising:

[0107] The user profile data acquisition module 301 can be used to acquire user dynamic profile data and user static profile data.

[0108] The risk tolerance level generation module 302 can be used to input user static profile data into the risk tolerance level classification machine learning model to generate user risk tolerance level information.

[0109] The behavior risk level generation module 303 can be used to input user dynamic profile data into the behavior risk level classification machine learning model to generate user behavior risk level information.

[0110] The user financial risk assessment module 304 can be used to generate user financial risk assessment information based on user risk tolerance level information and user behavior risk level information.

[0111] In an optional embodiment of this application, the user profile data acquisition module 301 can also be used for:

[0112] Perform a data authorization agreement check on the target user. If the data authorization agreement check result shows that the data authorization agreement is complete and correct, generate an authorization status confirmation label.

[0113] Based on the authorization status confirmation tag, the user's credit information, qualification information, and financial information are obtained. The user's credit information, qualification information, and financial information are then input into the feature parameter extraction natural language processing model to extract user credit feature parameters, user qualification feature parameters, and user financial feature parameters.

[0114] Based on user credit characteristics, user qualification characteristics, and user financial characteristics, static user profile data is constructed.

[0115] Based on the authorization status confirmation tag, user financial behavior information is collected. Based on the user financial behavior information and combined with the preset financial behavior analysis knowledge graph, financial behavior feature parameters of the user financial behavior information are extracted.

[0116] Based on the financial behavior characteristic parameters within the preset behavior analysis time window, time series data of financial behavior characteristic parameters is constructed, and user dynamic profile data is constructed based on the time series data of financial behavior characteristic parameters.

[0117] In an optional embodiment of this application, the behavioral risk level generation module 303 can also be used for:

[0118] The time series data of financial behavior characteristic parameters are input into the time series feature extraction module of the behavioral risk level classification machine learning model to extract the time series features of financial behavior.

[0119] The time-series features of financial behavior are input into the time-series feature analysis module of the behavioral risk level classification machine learning model to generate user behavior risk level information.

[0120] In an optional embodiment of this application, the machine learning-based financial risk assessment device 300 can also be used for:

[0121] A temporal feature extraction module was constructed based on a long short-term memory network.

[0122] A time series feature analysis module was constructed based on the conditional random field model.

[0123] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of a machine learning-based financial risk assessment method as described above.

[0124] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0125] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0126] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A machine learning-based financial risk assessment method, characterized in that, The method includes: Retrieve user dynamic profile data and user static profile data; The user's static profile data is input into the risk tolerance level classification machine learning model to generate user risk tolerance level information; The user dynamic profile data is input into the behavioral risk level classification machine learning model to generate user behavioral risk level information; User financial risk assessment information is generated based on the user's risk tolerance level information and the user's behavioral risk level information.

2. The method according to claim 1, characterized in that, The acquisition of user dynamic profile data and user static profile data includes: Perform a data authorization agreement check on the target user. If the data authorization agreement check result shows that the data authorization agreement is complete and correct, generate an authorization status confirmation label. Based on the authorization status confirmation tag, the user credit information, user qualification information and user financial information of the target user are obtained, and the user credit information, user qualification information and user financial information are input into the feature parameter extraction natural language processing model to extract user credit feature parameters, user qualification feature parameters and user financial feature parameters. The user static profile data is constructed based on the user credit feature parameters, the user qualification feature parameters, and the user financial feature parameters; Based on the authorization status confirmation tag, user financial behavior information is collected. Based on the user financial behavior information and combined with the preset financial behavior analysis knowledge graph, financial behavior feature parameters of the user financial behavior information are extracted. Based on the financial behavior feature parameters within the preset behavior analysis time window, time-series data of financial behavior feature parameters is constructed, and user dynamic profile data is constructed based on the time-series data of financial behavior feature parameters.

3. The method according to claim 2, characterized in that: The target users include natural person target users and legal person target users; The user credit information includes credit information of natural persons and credit information of legal persons; The user qualification information includes qualification information of natural persons and qualification information of legal persons; The user's financial information includes the financial information of natural persons and the financial information of legal persons.

4. The method according to claim 3, characterized in that: The natural person credit information includes personal credit report information and personal guarantee information; The corporate credit information includes corporate credit report information, contract default information, corporate bond record information, and corporate guarantee information; The natural person's qualification information includes occupational information, personal identification information, and asset ownership information; The legal entity qualification information includes enterprise certificate information and enterprise qualification certification information; The financial information of the natural person includes personal income and expenditure information and personal asset information; The corporate financial information includes corporate financial statements and corporate asset information.

5. The method according to claim 2, characterized in that, The step of inputting the user dynamic profile data into the behavioral risk level classification machine learning model to generate user behavioral risk level information includes: The time-series data of the financial behavior feature parameters are input into the time-series feature extraction module in the behavior risk level classification machine learning model to extract the financial behavior time-series features; The time-series features of the financial behavior are input into the time-series feature analysis module of the behavior risk level classification machine learning model to generate the user behavior risk level information.

6. The method according to claim 5, characterized in that, The method further includes: The temporal feature extraction module is constructed based on a long short-term memory network; The time series feature analysis module is constructed based on the conditional random field model.

7. The method according to claim 1, characterized in that, The risk tolerance level classification machine learning model is a gradient boosting tree model. The expression for the objective loss function of the round of iterations is: In the formula, The first step in the risk tolerance level classification machine learning model The target loss function in the round of iterations, The total number of risk tolerance levels corresponding to the target risk tolerance level information. To constrain the first Round of iteration targeting the first The regularization term for the complexity of the regression tree trained on risk tolerance levels. For the first Round of iteration targeting the first Regression tree for risk tolerance level training and The penalty coefficient is a hyperparameter. For the first Round of iteration targeting the first The total number of leaf nodes in the regression tree for risk tolerance level training. For the first Round of iteration targeting the first The first in the regression tree of risk tolerance level training The output weights of the leaf nodes For the sample size, For the first The sample belongs to the first The true label of risk tolerance level For the front The risk tolerance level classification machine learning model generated in the first iteration The sample belongs to the first The sum of the predicted values ​​of risk tolerance levels, For the first The risk tolerance level classification machine learning model generated in the first iteration The sample belongs to the first The predicted value of the risk tolerance level.

8. A financial risk assessment device based on machine learning, characterized in that, The device includes: The user profile data acquisition module is used to acquire dynamic user profile data and static user profile data. The risk tolerance level generation module is used to input the user static profile data into the risk tolerance level classification machine learning model to generate user risk tolerance level information. The behavior risk level generation module is used to input the user dynamic profile data into the behavior risk level classification machine learning model to generate user behavior risk level information. The user financial risk assessment module is used to generate user financial risk assessment information based on the user's risk tolerance level information and the user's behavioral risk level information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images