Risk assessment method and device, storage medium and electronic equipment

Abstract

The invention discloses a risk assessment method and device, a storage medium and electronic equipment. The method comprises the steps that multiple features of a to-be-evaluated object are acquired, and the multiple features are used for evaluating the risk level of the to-be-evaluated object; at least one feature meeting a feature expectation condition is determined from the multiple features, and the feature expectation condition is used for indicating that the feature effectiveness degree of the at least one feature is larger than an expectation threshold value; and inputting the at least one feature into a risk assessment model, and performing risk assessment processing to obtain a risk coefficient of the to-be-assessed object output by the risk assessment model, the risk coefficient being used for indicating a risk level, and the risk assessment model being obtained by training by using a data set. Through the risk assessment method and device, the problem of relatively low accuracy of risk assessment in related technologies is solved.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and specifically, to a risk assessment method, device, storage medium and electronic device. Background Art

[0002] In the traditional credit risk control system, risk assessment relies on the empirical rules of business experts and basic data statistics. These methods have limitations in dealing with the effectiveness of features and complex interactions between features. For example, they may be imprecise due to subjectivity and may become invalid over time, resulting in low accuracy of risk assessment.

[0003] Currently, no effective solution has been proposed to address the problem of low accuracy of risk assessment in related technologies. Summary of the invention

[0004] The main purpose of this application is to provide a risk assessment method, device, storage medium and electronic device to solve the problem of low accuracy of risk assessment in related technologies.

[0005] In order to achieve the above-mentioned purpose, according to one aspect of the present application, a risk assessment method is provided. The method comprises: obtaining multiple features of an object to be assessed, wherein the multiple features are used to assess the risk level of the object to be assessed; determining at least one feature that meets the feature expectation condition from the multiple features, wherein the feature expectation condition is used to indicate that the feature effectiveness of at least one feature is greater than the expected threshold; inputting at least one feature into a risk assessment model, performing risk assessment processing, and obtaining a risk coefficient of the object to be assessed output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using a data set.

[0006] In order to achieve the above-mentioned purpose, according to another aspect of the present application, a risk assessment device is provided. The device includes: an acquisition unit, which is used to acquire multiple features of an object to be assessed, wherein the multiple features are used to assess the risk level of the object to be assessed; a determination unit, which is used to determine at least one feature that meets the feature expectation condition from the multiple features, wherein the feature expectation condition is used to indicate that the feature effectiveness of at least one feature is greater than the expected threshold; an assessment unit, which is used to input at least one feature into a risk assessment model, perform risk assessment processing, and obtain a risk coefficient of the object to be assessed output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using a data set.

[0007] In the embodiment of the present application, by calculating the feature effectiveness of the features, effective features with strong risk prediction capabilities are screened, thereby avoiding the noise impact caused by invalid features after the features are input into the risk assessment model, and improving the prediction accuracy of the risk assessment model. Through feature screening and feature interactive learning, the prediction ability and interpretability of the risk assessment model are effectively improved, providing a more accurate risk quantification technology for credit business, thereby achieving the technical effect of improving the accuracy of risk assessment, and solving the technical problem of low accuracy of risk assessment in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0009] Figure 1 A hardware structure block diagram of a computer terminal for implementing a risk assessment method is shown;

[0010] Figure 2 is a flow chart of a risk assessment method provided according to an embodiment of the present application;

[0011] Figure 3 is a schematic diagram of a risk assessment method provided according to an embodiment of the present application;

[0012] Figure 4 is a schematic diagram of a risk assessment method provided according to an embodiment of the present application;

[0013] Figure 5 is a schematic diagram of a risk assessment method provided according to an embodiment of the present application;

[0014] Figure 6 is a schematic diagram of a risk assessment device provided according to an embodiment of the present application;

[0015] Figure 7 It is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0017] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0018] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0019] It should be noted that the information and data collected in this application are authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0020] Example 1

[0021] According to an embodiment of the present application, a method embodiment of risk assessment is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0022] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a risk assessment method. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also 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 the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations shown.

[0023] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0024] 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 risk assessment method in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, to implement the above-mentioned risk assessment method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0025] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 10. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0026] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0027] Under the above operating environment, this application provides Figure 2 The risk assessment method shown includes:

[0028] S201, obtaining multiple features of the object to be evaluated, wherein the multiple features are used to evaluate the risk level of the object to be evaluated;

[0029] S202, determining at least one feature that meets a feature expectation condition from a plurality of features, wherein the feature expectation condition is used to indicate that a feature effectiveness degree of the at least one feature is greater than an expected threshold;

[0030] S203, inputting at least one feature into a risk assessment model, performing risk assessment processing, and obtaining a risk coefficient of the object to be assessed output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using a data set.

[0031] Optionally, in this embodiment, the object to be evaluated may be, but is not limited to, an individual or enterprise applying for a credit card, loan or other financial product. The attributes and behavior data of these objects will be used for risk level evaluation.

[0032] Optionally, in this embodiment, the multiple features cover a set of data points used to describe the attributes and behaviors of the object to be evaluated. These features may include basic customer information (such as gender, age, education), loan information (such as loan amount, repayment history), and external data (such as personal credit report).

[0033] Optionally, in this embodiment, the feature expectation condition defines a criterion for screening features. In this embodiment, the feature expectation condition may be based on, but not limited to, IV (Information Value) values, which are used to measure the contribution of features to the prediction results. If the feature effectiveness is greater than the expected threshold, it means that the feature has sufficient influence on risk prediction.

[0034] Optionally, in this embodiment, the risk assessment model may be, but is not limited to, a model constructed by the xDeepFM algorithm, which is used to quantify the credit risk of the object to be assessed. The model can automatically learn the explicit and implicit interactions between features, thereby improving the accuracy and timeliness of the prediction.

[0035] Optionally, in this embodiment, the risk level is different risk levels divided according to the risk coefficient, such as high risk, medium risk or low risk, which is used to guide credit decisions.

[0036] Optionally, in this embodiment, the risk coefficient is a quantitative result output by the model, representing the risk level of a specific object to be evaluated. The level of the risk coefficient is directly related to the classification of risk levels.

[0037] Optionally, in this embodiment, multiple features of the object to be evaluated are obtained. This may involve, but is not limited to, collecting comprehensive information related to the object to be evaluated from bank databases and third-party data sources, including but not limited to basic personal information, loan history, external credit records, etc. These features will be used for subsequent analysis and risk assessment.

[0038] Optionally, in this embodiment, at least one feature that meets the feature expectation condition is determined from multiple features. In this process, feature screening can be achieved by, but is not limited to, calculating the IV value, which measures the contribution of each feature to distinguishing high-risk objects from low-risk objects. For example, assuming that the expected threshold is set to 0.1, it means that only features with an IV value greater than 0.1 will be selected for input into the model for risk assessment, thereby ensuring that the features input into the model have sufficient predictive power.

[0039] Optionally, in this embodiment, at least one feature is input into the risk assessment model, and risk assessment processing is performed to obtain the risk coefficient of the object to be assessed output by the risk assessment model. The risk assessment model integrates the outputs of various parts to generate a risk coefficient for accurately representing the credit risk of the object to be assessed. The size of the coefficient reflects the risk level of the object.

[0040] Through the embodiments provided by the present application, by calculating the feature effectiveness of the features, effective features with strong risk prediction capabilities are screened, thereby avoiding the noise impact caused by invalid features after the features are input into the risk assessment model, and improving the prediction accuracy of the risk assessment model. Through feature screening and feature interactive learning, the prediction ability and interpretability of the risk assessment model are effectively improved, providing a more accurate risk quantification technology for credit business, thereby achieving the technical effect of improving the accuracy of risk assessment.

[0041] As an optional solution, determining at least one feature that meets the feature expectation condition from the multiple features includes:

[0042] Calculate the feature effectiveness of each of the multiple features in turn;

[0043] Among them, the calculation method of the feature effectiveness of the i-th feature includes:

[0044] Use the i-th feature to perform risk assessment on multiple objects to obtain the i-th risk assessment result;

[0045] Obtaining a first proportion of objects whose risk levels indicated by the i-th risk assessment result are less than a preset level among the multiple objects, and obtaining a second proportion of objects whose risk levels indicated by the i-th risk assessment result are greater than or equal to the preset level among the multiple objects;

[0046] The difference between the first ratio and the second ratio is multiplied by the preset evidence weight to obtain the feature effectiveness of the i-th feature.

[0047] Optionally, in this embodiment, the feature effectiveness refers to the predictive value of a single feature in the risk assessment model, that is, to what extent the feature can distinguish objects of different risk levels.

[0048] Optionally, in this embodiment, the first ratio and the second ratio respectively refer to the ratio of objects with risk levels lower than a preset level and objects with risk levels greater than or equal to a preset level among all customers, which are used to calculate the effectiveness of the feature.

[0049] Optionally, in this embodiment, determining at least one feature that meets the expected condition of the feature from multiple features includes: calculating the feature effectiveness of each of the multiple features in turn; in this step, evaluating the predictive value of all collected features one by one to ensure that the features input to the model have a substantial contribution to the risk assessment.

[0050] Optionally, in this embodiment, the calculation method of the feature effectiveness of the ith feature includes: using the ith feature to perform risk assessment processing on multiple objects to obtain the ith risk assessment result. The calculation process of the feature effectiveness first uses the ith feature to predict a series of objects to be evaluated, obtains the risk assessment result related to the feature, obtains the first proportion of objects with a risk level less than the preset level indicated by the ith risk assessment result among multiple objects, and obtains the second proportion of objects with a risk level greater than or equal to the preset level indicated by the ith risk assessment result among multiple objects; next, by comparing the risk assessment result with the preset risk level, calculate the proportion of objects with a risk level lower than the preset level and a risk level greater than or equal to the preset level, which is used for subsequent calculation of the feature effectiveness. Use the ratio difference between the first ratio and the second ratio, multiply by the preset evidence weight, to obtain the feature effectiveness of the ith feature. Finally, calculate according to the ratio difference and the preset evidence weight to obtain the feature effectiveness of the ith feature, which is used to determine whether the feature meets the expected condition of the feature.

[0051] Through the embodiments provided in this application, the role of quantitative features in risk assessment is used to achieve automatic screening of feature effectiveness. Through this process, the risk assessment model can better identify and assess credit risks and improve the accuracy of risk assessment.

[0052] As an optional solution, before using the ratio difference between the first ratio and the second ratio to multiply by a preset evidence weight to obtain the feature effectiveness of the i-th feature, the method further includes:

[0053] Obtaining a ratio quotient between the first ratio and the second ratio;

[0054] The natural logarithm of the ratio quotient is determined as the preset weight of evidence.

[0055] Optionally, in this embodiment, the ratio quotient value refers to the quotient of the first ratio (the proportion of low-risk customers) and the second ratio (the proportion of high-risk and above customers) in the feature effectiveness assessment, reflecting the ability of the feature to distinguish customers of different risk levels. The preset weight of evidence (WOE) is an indicator used to measure the strength of the association between the feature value and the risk level in the risk assessment model, and is calculated as the natural logarithm of the ratio quotient value. The natural logarithm is a logarithmic function with the mathematical constant e as the base, which is usually used to transform and standardize data in statistics and data processing.

[0056] Optionally, in this embodiment, before calculating the effectiveness of the feature, the ratio quotient between the first ratio and the second ratio is first obtained, that is, the quotient of the proportion of low-risk customers and the proportion of high-risk and above customers. This ratio quotient intuitively reflects the ability of the feature to distinguish customers of different risk levels. Then, the ratio quotient is converted into a natural logarithm to obtain a preset weight of evidence (WOE). This conversion step standardizes the quotient, making the comparison and combination of different features more reasonable, and avoiding the unbalanced impact that may be caused by directly using the ratio quotient.

[0057] It is understandable that the preset weight of evidence (WOE) plays an important regulatory role in the calculation of feature effectiveness. It is multiplied with the proportional difference to jointly determine the predictive value of the feature. This means that the feature not only needs to be able to distinguish between high-risk and low-risk objects, but also this distinguishing ability needs to be weighted by the preset weight of evidence (WOE) to ensure that the model's reliance on the feature is based on the quantitative results of its actual distinguishing ability. This design effectively balances the predictive power of the feature and the complexity of the model, avoiding the model's excessive reliance on certain features, thereby improving the generalization ability and robustness of the model.

[0058] Through the embodiments provided in this application, by introducing the ratio quotient and converting it into the preset weight of evidence (WOE), a more accurate basis is provided for calculating the effectiveness of the feature. This process not only takes into account the ability of the feature to distinguish the risk level, but also standardizes the distinguishing ability through natural logarithm transformation to ensure that the model's evaluation of the feature is comprehensive and reasonable. Through this mechanism, features that make substantial contributions to credit risk assessment can be screened out more accurately, improving the accuracy and efficiency of model predictions and reducing the risk of model overfitting.

[0059] As an optional solution, at least one feature is input into the risk assessment model to perform risk assessment processing, and the risk coefficient of the object to be assessed outputted by the risk assessment model includes:

[0060] Obtain a prediction vector output by the compression interaction network module, obtain a first prediction value output by the linear module, and obtain a second prediction value output by the deep neural network module, wherein the prediction vector is used to indicate deep information of a combination of at least one feature, the first prediction value is used to indicate a linear relationship of at least one feature, and the second prediction value is used to indicate a nonlinear relationship of at least one feature,

[0061] The risk assessment model includes a compressed interaction network module, a linear module and a deep neural network module, the compressed interaction network module is used to learn explicit high-order interactions between features, the linear module is used to capture the linear relationship of at least one feature, and the deep neural network module is used to capture the nonlinear relationship of at least one feature;

[0062] The prediction vector, the first prediction value and the second prediction value are combined to obtain a risk coefficient.

[0063] Optionally, in this embodiment, the compressed interaction network (CIN) module is a network architecture for learning explicit high-order interactions between features. In this embodiment, the CIN module can automatically capture the deep information of feature cross-combinations without manual intervention, avoiding the dimensionality curse and reducing the complexity of feature engineering.

[0064] Optionally, in this embodiment, the linear module is part of the risk assessment model and is used to capture the linear relationship between features. It processes features by one-hot encoding and other methods, which can intuitively reflect the direct impact of each feature on risk assessment and enhance the interpretability of the model.

[0065] Optionally, in this embodiment, a deep neural network (DNN) module: another part of the risk assessment model, is used to capture the nonlinear relationship between features. The DNN module can learn complex nonlinear combinations between features, improve the model's ability to handle nonlinear features, and thus improve the accuracy of the overall prediction.

[0066] Optionally, in this embodiment, the prediction vector is a vector output by the CIN module, which contains deep information of the feature combination and reflects the impact of explicit high-order interactions between features on risk prediction. The first prediction value and the second prediction value refer to the values ​​output by the linear module and the DNN module, respectively. The first prediction value reflects the contribution of the linear relationship of the features to risk prediction, while the second prediction value covers the prediction value of the nonlinear relationship of the features.

[0067] Optionally, in this embodiment, data merging is used to indicate the process of integrating the prediction vector, the first prediction value and the second prediction value into a single prediction result (risk coefficient). Through data merging, the model can integrate linear and nonlinear feature relationships to provide a more comprehensive risk assessment.

[0068] It is understandable that the risk assessment model captures the deep information of explicit high-order feature interactions through the CIN module, avoiding the limitations of artificial feature combinations and the problem of over-reliance on expert rules, and improving the degree of automation of feature interactions. At the same time, the linear module and DNN module process the linear and nonlinear relationships of features respectively, ensuring that the model can fully capture various association patterns in the data and improve the accuracy and generalization ability of predictions.

[0069] Finally, through the data merging step, the model integrates the output results of the CIN, linear module and DNN module into a risk coefficient, which combines the impact of various relationships of features on risk assessment and provides a more accurate risk quantification technology for the business. This integration method not only improves the prediction effect of the model, but also maintains the interpretability of the model, enabling business personnel to understand the logic behind the model's decision-making, and enhancing the practicality and credibility of the model.

[0070] Through the embodiments provided in this application, the problems of feature relationship learning and risk quantification in the field of credit risk control are effectively solved. Through multi-level feature relationship capture and data merging, accurate assessment of credit risk is achieved, the credit decision-making process is optimized, and business risks are reduced, providing strong technical support for the digital transformation and risk management of credit business.

[0071] As an optional solution, before inputting at least one feature into the risk assessment model, performing risk assessment processing, and obtaining the risk coefficient of the object to be assessed output by the risk assessment model, the method further includes:

[0072] Preprocessing the data in the data set to obtain preprocessed data, wherein the preprocessed data includes different types of data;

[0073] Perform feature extraction on the preprocessed data to obtain a feature set;

[0074] Normalize the feature set to obtain normalized features;

[0075] The normalized features are used to train the deep self-learning model to obtain the risk assessment model.

[0076] Optionally, in this embodiment, preprocessing is a series of cleaning, conversion and normalization operations performed before data is input into the model to ensure the quality and applicability of the data and provide a reliable data foundation for model training.

[0077] Optionally, in this embodiment, normalization processing is used to adjust the numerical range of the features so that they fall on the same scale to avoid the influence of numerical differences on the model training results.

[0078] Optionally, in this embodiment, a systematic data preprocessing and feature engineering process is provided for the construction of the risk assessment model. First, through the preprocessing operation, the raw data is cleaned and transformed to ensure data quality and provide a solid foundation for subsequent analysis. Next, the feature extraction step obtains key information related to credit risk from the preprocessed data and constructs a comprehensive feature set. The subsequent normalization process further optimizes the feature set, making it suitable for input into the deep self-learning model for training.

[0079] By using normalized features to train the deep self-learning model, the resulting risk assessment model has the ability to automatically learn and optimize features. It can not only capture the linear and nonlinear relationships between features, but also use the CIN layer to effectively learn the explicit high-order interactions between features, greatly improving the prediction accuracy of the model.

[0080] Through the embodiments provided in this application, the deficiencies of traditional models in data preprocessing and feature engineering are solved, the efficiency of model training and the accuracy of prediction results are ensured, strong technical support is provided for intelligent risk control and decision-making of credit business, and the digitalization and intelligentization of credit business are promoted.

[0081] As an optional solution, before preprocessing the data in the data set to obtain the preprocessed data, the method further includes:

[0082] Acquire data associated with multiple objects, wherein the data associated with each object includes basic identity information and transaction information of each object;

[0083] According to business rules, risk assessment is performed on the data associated with each object to obtain the risk level associated with each object;

[0084] The data associated with each object and the risk level associated with each object are combined to obtain a data set.

[0085] Optionally, in this embodiment, basic identity information includes but is not limited to the subject's name, age, gender, ID number, mobile phone number, education level, marital status, etc. This information provides the subject's personal attributes and background.

[0086] Optionally, in this embodiment, the transaction information covers the transaction records of the subject in banks or other financial institutions, including transaction amounts, transaction frequencies, repayment records, overdue conditions, etc., reflecting the behavior patterns and credit performance of the subject.

[0087] Optionally, in this embodiment, the business rules refer to guiding principles for preliminarily identifying risk levels, which may be rules based on expert experience, or industry standards or internal policies of a bank.

[0088] Optionally, in this embodiment, first, by collecting basic identity information and transaction information, a comprehensive data file containing personal attributes and credit performance is established for each object. Then, according to the preset business rules, this data is preliminarily evaluated and divided into different risk levels. This preliminary evaluation provides a type of label for model training, which facilitates the model to learn the association between different features and risk levels. Finally, the data and risk level information of all objects are integrated to form a complete data set required for training.

[0089] Through the embodiments provided in this application, through this data set construction process, the comprehensiveness and accuracy of the model training data are ensured, providing rich learning materials for the deep risk assessment model. This technical solution solves the key problem in data collection and preprocessing, that is, how to effectively collect and evaluate customer data, laying a solid foundation for subsequent feature engineering and model training, and helping to improve the efficiency of the entire risk control process and the prediction accuracy of the model.

[0090] As an optional solution, after inputting at least one feature into the risk assessment model, performing risk assessment processing, and obtaining the risk coefficient of the object to be assessed output by the risk assessment model, the method further includes:

[0091] According to the risk coefficient, the object to be assessed is classified to obtain the risk level of the object to be assessed;

[0092] Based on the risk level, control operations associated with the risk level are formulated for the assessed object, where the higher the risk level, the more restrictive the control operations are.

[0093] Optionally, in this embodiment, the objects to be evaluated are classified and their risk levels are determined according to the risk coefficients output by the model. This classification process is based on a preset risk level standard, and the risk coefficients are mapped to specific risk levels. For example, objects to be evaluated whose risk coefficients exceed a certain threshold will be classified as high risk, and vice versa.

[0094] Optionally, in this embodiment, it is clearly stated that the control operations associated with the object to be evaluated are formulated according to the risk level of the object to be evaluated. The restrictiveness of the control operation is proportional to the risk level. The object to be evaluated with a higher risk level will face more stringent control measures, which may include raising the credit threshold, increasing the review frequency, limiting the loan amount, etc. In this way, a more cautious credit strategy can be adopted for high-risk customers, thereby effectively controlling the potential risks in the credit business.

[0095] Through the embodiments provided in this application, by combining the risk coefficient output by the model with the actual management and control operations, the direct application of risk assessment results is achieved, providing a direct basis for the decision-making of bank credit business. The management and control operations based on risk levels effectively solve the problem of insufficient processing of the correlation between risk levels and management and control operations in traditional risk control models, and achieve a balance between risk management and business development through automation and data-driven methods.

[0096] As an optional solution, the above risk assessment method is applied to the training and use scenario of a risk control model based on the xDeepFM algorithm. Optionally, in this scenario, the mainstream risk control model is basically a combination of the LR (LogisticRegression) model and the "expert rule". The LR model is equivalent to a linear regression model equipped with a sigmoid function. It is a classification algorithm that judges the model effect by calculating the loss function and optimizes the parameters on this basis. Because it is linear, it has strong interpretability, but it also requires a lot of effort in user grouping, feature binning conversion, etc. to increase the fitting ability. The LR model has been widely used in industrial problems today. It has good robustness to small noise in the data and can also solve the problem of multicollinearity. It only needs to be combined with regularization. However, in the face of the current increase in bad debt rates, the requirements for model accuracy have to be improved, and the LR algorithm is not strong enough to handle nonlinear features and needs to be converted; to improve model accuracy, it is inevitable to increase feature dimensions, and when the feature space is large, the performance of LR is not very friendly.

[0097] Optionally, in this scenario, a method for training and using a risk control model based on the xDeepFM algorithm includes: Figure 3 The data acquisition module 302, data processing module 304, feature engineering module 306, and deep self-learning module 308 are shown.

[0098] Specifically, the data acquisition module 302 is responsible for the collection and sampling of basic data while ensuring its validity; since the basic source data is large in volume and generally sparse, the data processing module is for artificial noise reduction, including but not limited to adding expert rules, preprocessing of basic data and feature importance screening; after the data processing module performs ETL operations on the data, feature engineering is required to increase the dimensions of large wide tables of data in order to more comprehensively cover the attributes in specific scenarios, thereby increasing the upper limit of model training, which is the top priority of the entire model; finally, the data is randomly split into training sets and test sets in proportion, and put into the deep self-learning module, waiting for the final output results, and then making business decisions.

[0099] Optionally, bank credit card users can be used as research samples, and historical user risk analysis data stored in the credit card center can be simply sampled. However, after sampling, the sample set must be checked again to ensure that it conforms to the actual law, that is, the number of risky users is much smaller than that of normal users. Secondly, since banks have a variety of service scenarios and sample sources are also diverse, it is necessary to design a set of benchmarks to screen samples so that the constructed model is universal and achieves the expected effect, mainly in the following aspects:

[0100] Special sample processing: Special months are taken out separately as test sets or modeled separately, such as samples during the financial crisis, the period of bankruptcies, or the epidemic.

[0101] Timeliness: When screening samples, the sample date should not be too early. It is best to be within the past two years, and the sample size in each month should be relatively balanced.

[0102] Representativeness: Samples need to come from stable and compliant businesses that reflect the mainstream business scenarios in their respective market segments. They should also avoid restrictions on a single region and exclude samples of partners whose businesses are mainly local.

[0103] Stability: The sample size distribution of each partner's samples in each province should be consistent with the actual population distribution, and the bad sample rate in each month should be relatively stable.

[0104] The sample size is relatively balanced: the sample sizes corresponding to different scenarios should not differ significantly.

[0105] After sampling according to the following principle, the total samples are randomly divided into training set samples and test set samples in a ratio of 8:2.

[0106] Interview business experts in the corresponding scenarios to understand how to distinguish fraudulent customers from normal customers through manual review without relevant models. The input of business experts can help model developers quickly do a layer of feature screening and feature combination work, so as to improve modeling efficiency and ensure the lower limit of model effect, reduce the cost of modeling trial and error; at the same time, it can effectively avoid overfitting of the model. The following are two common rules.

[0107] Whether the information is consistent: When reviewing loan applications, bank risk control personnel need to pay special attention to whether the user's application information is consistent with the data on the central bank's credit report; if there is an inconsistency, it means that the user may be hiding real information.

[0108] Repayment behavior in different time periods: The credit risk control approval personnel in the financial industry have a popular rule of "no three or six consecutive defaults within half a year", that is, users cannot have six loan defaults within half a year, and cannot have three consecutive months of loan defaults.

[0109] Retrieve basic customer information from the bank database, including but not limited to gender, age, region, education level, marital status, unit type and other attributes; loan-related information, such as product type, pre-approved credit amount, basic rating, loan card number, application period, etc.; retrieve external (third-party) data, such as personal credit information related to the central bank.

[0110] The data processing module 304 determines the importance of the feature to the indicator by calculating IV (Information Value). The calculation formula is as follows:

[0111]

[0112] Among them, IV i is the IV value of the i-th feature, Bad i is the number of objects that do not meet the feature expectation conditions, Bad t is the total number of all objects, Good i is the number of objects that meet the feature expectation conditions, Good t is the total number of all objects.

[0113] After calculation, consider the attributes with IV < 0.02 as invalid attributes, those with 0.02 < IV < 0.1 as weak-effect features, those with 0.1 < IV < 0.3 as medium-effect features, those with 0.3 < IV < 0.5 as strong-effect features, and those with IV > 0.5 as extremely strong-effect features. Select the attributes with IV values above 0.1 to reduce the impact of invalid features on the model and achieve the purpose of improving efficiency.

[0114] The data processing module 304 is also used to perform data preprocessing operations. If an attribute has a high blank rate or a large number of invalid values, has a low coverage rate, and its importance is relatively low, the variable can be directly deleted. For data that conforms to a uniform distribution, fill it with the mean value of the variable; for data with a skewed distribution, use the median for filling. If the variable is discrete and has few different values, it can be converted into dummy variables. For example, for the gender SEX variable, which has three different values: male, female, and NA, this column can be converted into IS_SEX_MALE, IS_SEX_FEMALE, and IS_SEX_NA. If a variable has more than a dozen different values, according to the frequency of each value, group the values with smaller frequencies into a category 'other' to reduce the dimension. This approach can maximize the retention of variable information.

[0115] The data processing module 304 is also used for data normalization processing. The dimensions of different features in the data may be inconsistent, and the differences between values may be large. Without processing, it may affect the results of data analysis. Therefore, it is necessary to scale the data according to a certain ratio so that it falls within a specific range for comprehensive analysis. The maximum-minimum normalization is selected, that is, the data is mapped to the [0, 1] interval, and the formula is as follows:

[0116]

[0117] The feature engineering module 306 is used to extract and construct features. Common basic attributes include name, age, gender, ID card, mobile phone number, education level, date of birth, income, industry, occupation, etc. Among them, indicators such as age, education level, and income do not require special processing; the scope of occupation or industry is relatively wide and the distribution is relatively scattered. Therefore, it is best not to directly include such features in the model. In the face of this situation, recent sample applications and post-loan samples should be taken to determine which industries belong to high-risk customer groups and achieve the purpose of effective interception. For the customer address information, the home address, work address, residential address, and current GPS location can be obtained through the APP on the user's mobile device. The address information is split and grouped by province, the post-loan overdue levels of each province in the recent period are calculated and sorted, and the regions with the top rankings are selected; in addition, the frequency of obtaining GPS location by the APP is increased, all GPS locations of the customer within a short period of time are recorded and de-duplicated, and the distances between addresses are calculated. The basic data information such as the maximum and minimum values and standard deviation are used as derivative features to be included in the model. For customers with long-term "loan-rushing" behaviors, they often choose to use others' ID cards to apply for loans. Such customers also belong to high-risk groups. Therefore, the phone numbers and bank card numbers corresponding to a single device ID, the device ID, bank cards, and ID card numbers corresponding to a single phone number, and the phone numbers and device ID numbers corresponding to a single bank card can be selected as new features to be included in the model; in addition, the APP names and package names in the user's mobile device are obtained as appropriate, and the quantity and proportion of some lending-related APPs are statistically calculated. The quantity of such APPs and their proportion in all APPs can often reflect the user's lending intention. Chinese APP names contain words such as "borrow", "spend", "credit", "loan", "gold", etc. At the same time, more attention should be paid to collecting and updating information on certain high-risk APPs in order to effectively identify customer groups with obvious debt repayment pressure in a timely manner.

[0118] The deep self-learning module 308 is used to train and learn the model based on the xDeepFM model. xDeepFM proposes a brand-new CIN (Compressed Intereaction Network) - Compressed Interaction Network.

[0119] As Figure 3 shown, the schematic diagram of the overall structure of CIN. The input into CIN is a matrix obtained by horizontally concatenating the column vectors of all attribute fields (m is the number of vectors, D is the dimension), and the form of the feature combination matrix obtained in each subsequent layer is (H i , D), and the H i in each layer is no longer equal, but the dimension D remains unchanged, which is the same as the input layer. The calculation method of each vector in the k-th layer is as follows:

[0120] where 1 ≤ h ≤ Hk , W k,h represents the weight matrix of the hth vector of the kth layer, Denotes the Hadamard product.

[0121] As can be seen from the figure above, CIN first takes the X of the previous layer k-1 The Hth k-1 vector, and the X of the input layer 0 Perform Hadamard multiplication on the m vectors in pairs to obtain H k-1 *m vectors, then weighted sum; the difference between each layer is W k,h This weight matrix is ​​a hyperparameter that can be adjusted.

[0122] Then, if Figure 4 As shown, CIN has done a feature map process, that is, using Hk+1 m*H k The convolution kernel generates the state of the next hidden layer, which is basically the same as the popular CNN, but the difference lies in the design of the convolution kernel. CIN calculates the intermediate result - tensor Z k+1 , and then use the weight matrix W k,i Following the tensor Z k+1 We multiply and sum the dimensions D layer by layer to get the i-th vector of the k+1 layer. Finally, we can see that the original three-dimensional tensor is compressed into a two-dimensional matrix. This is the origin of “Compressed” in CIN.

[0123] Back to Figure 3 The intermediate steps show different depths (b). The convolution matrix is ​​sumpooled. The purpose of this is to superimpose all input features, thereby achieving the effect of interacting features of different orders. The final output is a vector.

[0124] When in use, this embodiment recommends not performing any manual interactive combination processing on the features, but rather determining the best parameters suitable for the current model through Grid-Search.

[0125] Figure 5 The following is a schematic diagram of the overall framework of the xDeepFM model. The above mainly introduces CIN, which is the core of this model and is used to learn high-order explicit feature interaction patterns. It is only part of the final output layer of the model. The other two parts are the linear module and the DNN module.

[0126] The leftmost one is LinearModule, which directly uses the vector obtained by one-hot encoding the original field as the input of the model.

[0127] The one on the far right is DNNModule, whose purpose is to learn high-order implicit feature interaction patterns.

[0128] Finally, the outputs of the three modules are combined to obtain a predicted value, which is the risk factor for a certain customer in the credit risk control scenario mentioned in this patent:

[0129]

[0130] For high risk, For medium risk, For low risk, the risk factor standard can be adjusted as appropriate according to the business scenario, and the indicators can also be refined.

[0131] Through the embodiments provided in this application, the xDeepFM model can effectively reduce labor costs, eliminate the need for manual feature interaction, and reduce human errors; all learned features are self-learned, so the model parameters can be updated in a timely manner to achieve the best effect at the time, and the impact of lag caused by historical data is avoided to a large extent; more importantly, for business personnel, this model is no longer a box, it is interpretable, more friendly to business personnel, and has stronger generalization capabilities.

[0132] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0133] Example 2

[0134] The embodiment of the present application also provides a risk assessment device. It should be noted that the risk assessment device of the embodiment of the present application can be used to execute the risk assessment method provided by the embodiment of the present application. The risk assessment device provided by the embodiment of the present application is introduced below.

[0135] According to an embodiment of the present application, a device for implementing the above risk assessment method is also provided, such as Figure 6 As shown, the device comprises:

[0136] An acquisition unit 602 is used to acquire multiple features of the object to be evaluated, wherein the multiple features are used to evaluate the risk level of the object to be evaluated;

[0137] A determining unit 604, configured to determine at least one feature that meets a feature expectation condition from the plurality of features, wherein the feature expectation condition is used to indicate that a feature effectiveness degree of the at least one feature is greater than an expected threshold;

[0138] The evaluation unit 606 is used to input at least one feature into the risk assessment model, perform risk assessment processing, and obtain the risk coefficient of the object to be evaluated output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using the data set.

[0139] As an optional solution, the determining unit 604 includes:

[0140] A calculation module, used to calculate the feature effectiveness of each of the multiple features in turn;

[0141] Among them, the calculation method of the feature effectiveness of the i-th feature includes:

[0142] Use the i-th feature to perform risk assessment on multiple objects to obtain the i-th risk assessment result;

[0143] Obtaining a first proportion of objects whose risk levels indicated by the i-th risk assessment result are less than a preset level among the multiple objects, and obtaining a second proportion of objects whose risk levels indicated by the i-th risk assessment result are greater than or equal to the preset level among the multiple objects;

[0144] The difference between the first ratio and the second ratio is multiplied by the preset evidence weight to obtain the feature effectiveness of the i-th feature.

[0145] As an optional solution, the device also includes:

[0146] A first acquisition module is used to obtain a ratio quotient between the first ratio and the second ratio before multiplying the ratio difference between the first ratio and the second ratio by a preset evidence weight to obtain a feature validity degree of the i-th feature;

[0147] The determination module is used to determine the natural logarithm of the ratio quotient as the preset evidence weight before using the ratio difference between the first ratio and the second ratio to multiply by the preset evidence weight to obtain the feature effectiveness of the i-th feature.

[0148] As an optional solution, the evaluation unit 606 includes:

[0149] The second acquisition module is used to obtain the prediction vector output by the compression interaction network module, obtain the first prediction value output by the linear module, and obtain the second prediction value output by the deep neural network module, wherein the prediction vector is used to indicate the deep information of the combination of at least one feature, the first prediction value is used to indicate the linear relationship of at least one feature, and the second prediction value is used to indicate the nonlinear relationship of at least one feature.

[0150] The risk assessment model includes a compressed interaction network module, a linear module and a deep neural network module, the compressed interaction network module is used to learn explicit high-order interactions between features, the linear module is used to capture the linear relationship of at least one feature, and the deep neural network module is used to capture the nonlinear relationship of at least one feature;

[0151] The merging module is used to merge the prediction vector, the first prediction value and the second prediction value to obtain the risk coefficient.

[0152] As an optional solution, the device also includes:

[0153] A preprocessing module is used to preprocess the data in the data set to obtain preprocessed data before inputting at least one feature into the risk assessment model to perform risk assessment processing and obtain the risk coefficient of the object to be assessed output by the risk assessment model, wherein the preprocessed data includes different types of data;

[0154] A feature extraction module is used to extract features from pre-processed data to obtain a feature set before inputting at least one feature into a risk assessment model to perform risk assessment processing and obtain a risk coefficient of the object to be assessed output by the risk assessment model;

[0155] A normalization module is used to normalize the feature set to obtain normalized features before inputting at least one feature into the risk assessment model to perform risk assessment processing and obtain the risk coefficient of the object to be assessed output by the risk assessment model;

[0156] The training module is used to train the deep self-learning model using the normalized features to obtain the risk assessment model before inputting at least one feature into the risk assessment model to perform risk assessment processing and obtain the risk coefficient of the object to be assessed output by the risk assessment model.

[0157] As an optional solution, the device also includes:

[0158] A third acquisition module is used to obtain data associated with multiple objects before preprocessing the data in the data set to obtain the preprocessed data, wherein the data associated with each object includes basic identity information and transaction information of each object;

[0159] The assessment module is used to pre-process the data in the data set, and before obtaining the pre-processed data, perform risk assessment on the data associated with each object according to business rules to obtain the risk level associated with each object;

[0160] The combination module is used to combine the data associated with each object and the risk level associated with each object to obtain the data set before preprocessing the data in the data set to obtain the preprocessed data.

[0161] As an optional solution, the device also includes:

[0162] A classification module is used to input at least one feature into the risk assessment model, perform risk assessment processing, obtain the risk coefficient of the object to be assessed output by the risk assessment model, and then classify the object to be assessed according to the risk coefficient to obtain the risk level of the object to be assessed;

[0163] The management and control module is used to input at least one feature into the risk assessment model, perform risk assessment processing, and obtain the risk coefficient of the object to be assessed output by the risk assessment model. Then, according to the risk level, formulate management and control operations associated with the risk level for the object to be assessed, wherein the higher the risk level, the more restrictive the management and control operations are.

[0164] It should be noted that the above-mentioned modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n), and the above-mentioned modules can also be run as part of the device in the computer terminal 10 provided in Example 1.

[0165] Example 3

[0166] An embodiment of the present application may provide an electronic device, Figure 7 is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 7 As shown, the electronic device may include: one or more ( Figure 7 (only one is shown) processor 702, memory 704, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0167] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a 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 a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0168] The processor may call the information and application program stored in the memory through the transmission device to perform the following steps:

[0169] Acquire multiple features of the object to be evaluated, wherein the multiple features are used to evaluate the risk level of the object to be evaluated;

[0170] Determine at least one feature that meets a feature expectation condition from the plurality of features, wherein the feature expectation condition is used to indicate that a feature effectiveness degree of the at least one feature is greater than an expected threshold;

[0171] At least one feature is input into the risk assessment model, and risk assessment processing is performed to obtain a risk coefficient of the object to be assessed output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using a data set.

[0172] It can be understood by those skilled in the art that Figure 7 The structure shown is for illustration only, and the electronic device may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (Mobile Internet Devices, MID), a PAD, or other terminal devices. Figure 7 The structure of the electronic device is not limited. Figure 7 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 7 Different configurations shown.

[0173] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0174] Example 4

[0175] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the risk assessment method provided in the first embodiment.

[0176] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0177] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the risk assessment method.

[0178] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0179] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0180] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0181] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0182] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0183] If the integrated unit is implemented in the form of 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 application, 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, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0184] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A risk assessment method, characterized in that: include: Acquire multiple features of the object to be evaluated, wherein the multiple features are used to evaluate the risk level of the object to be evaluated; Determine at least one feature that meets a feature expectation condition from the multiple features, wherein the feature expectation condition is used to indicate that a feature effectiveness degree of the at least one feature is greater than an expected threshold; The at least one feature is input into a risk assessment model to perform risk assessment processing, and a risk coefficient of the object to be assessed is obtained as output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using a data set.

2. The method according to claim 1, characterized in that The determining of at least one feature that meets the feature expectation condition from the multiple features includes: Calculating the feature effectiveness of each of the multiple features in turn; Among them, the calculation method of the feature effectiveness of the i-th feature includes: Using the i-th feature to perform risk assessment processing on multiple objects to obtain an i-th risk assessment result; Obtaining a first ratio of objects whose risk levels indicated by the i-th risk assessment result are lower than a preset level among the multiple objects, and obtaining a second ratio of objects whose risk levels indicated by the i-th risk assessment result are greater than or equal to the preset level among the multiple objects; The difference between the first ratio and the second ratio is multiplied by a preset weight of evidence to obtain the feature effectiveness of the i-th feature.

3. The method according to claim 2, characterized in that Before obtaining the feature effectiveness of the i-th feature by multiplying the ratio difference between the first ratio and the second ratio by a preset evidence weight, the method further includes: Obtaining a ratio quotient between the first ratio and the second ratio; The natural logarithm of the ratio quotient is determined as the preset evidence weight.

4. The method according to claim 1, characterized in that: The step of inputting the at least one feature into a risk assessment model to perform risk assessment processing and obtaining the risk coefficient of the object to be assessed output by the risk assessment model comprises: Obtaining a prediction vector output by the compression interaction network module, obtaining a first prediction value output by the linear module, and obtaining a second prediction value output by the deep neural network module, wherein the prediction vector is used to indicate deep information of the combination of the at least one feature, the first prediction value is used to indicate a linear relationship of the at least one feature, and the second prediction value is used to indicate a nonlinear relationship of the at least one feature, The risk assessment model includes the compressed interaction network module, the linear module and the deep neural network module, the compressed interaction network module is used to learn explicit high-order interactions between features, the linear module is used to capture the linear relationship of the at least one feature, and the deep neural network module is used to capture the nonlinear relationship of the at least one feature; The prediction vector, the first prediction value and the second prediction value are combined to obtain the risk coefficient.

5. The method according to claim 1, characterized in that Before inputting the at least one feature into the risk assessment model, performing risk assessment processing, and obtaining the risk coefficient of the object to be assessed output by the risk assessment model, the method further includes: Preprocessing the data in the data set to obtain preprocessed data, wherein the preprocessed data includes different types of data; Performing feature extraction on the preprocessed data to obtain a feature set; Normalizing the feature set to obtain normalized features; The normalized features are used to train the deep self-learning model to obtain the risk assessment model.

6. The method according to claim 5, characterized in that Before preprocessing the data in the data set to obtain the preprocessed data, the method further includes: Acquire data associated with multiple objects, wherein the data associated with each object includes basic identity information and transaction information of each object; According to business rules, risk assessment is performed on the data associated with each object to obtain the risk level associated with each object; The data associated with each object and the risk level associated with each object are combined to obtain the data set.

7. The method according to any one of claims 1 to 6, characterized in that: After inputting the at least one feature into the risk assessment model, performing risk assessment processing, and obtaining the risk coefficient of the object to be assessed output by the risk assessment model, the method further includes: Classifying the object to be evaluated according to the risk coefficient to obtain the risk level of the object to be evaluated; According to the risk level, a control operation associated with the risk level is formulated for the object to be assessed, wherein the higher the risk level, the more restrictive the control operation is.

8. A risk assessment device, characterized in that: include: An acquisition unit, used to acquire multiple features of the object to be evaluated, wherein the multiple features are used to evaluate the risk level of the object to be evaluated; a determining unit, configured to determine at least one feature that meets a feature expectation condition from among the plurality of features, wherein the feature expectation condition is used to indicate that a feature effectiveness degree of the at least one feature is greater than an expected threshold; The evaluation unit is used to input at least one feature into the risk assessment model, perform risk assessment processing, and obtain the risk coefficient of the object to be evaluated output by the risk assessment model, wherein the risk coefficient is used to indicate the risk level, and the risk assessment model is obtained by training using the data set.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 7 when running.

Images