Financial bill risk value calculation method and device, equipment and storage medium

Abstract

The invention provides a financial bill risk value calculation method and device, equipment and a storage medium, relates to the technical field of financial bill risk assessment, and can comprehensively and deeply mine potential risks in financial bills by integrating various anomaly detection models and classification models. The anomaly detection model can accurately identify the anomaly in the data, and provides a high-quality risk data sample for the classification model, thereby ensuring the accuracy and reliability of risk probability calculation. The risk probability is predicted by using the pre-trained risk classification model, so that the calculation efficiency is greatly improved. Compared with a traditional risk assessment method, the method does not need to manually carry out tedious data analysis and judgment, and the risk identification period is remarkably shortened. And the risk probability is converted into the corresponding risk value, so that the risk assessment result is more visual and easy to understand. Risk management and decision making of related financial departments are facilitated, and comprehensive control of financial risk conditions of enterprises is facilitated.

Description

Technical Field

[0001] The present application relates to the technical field of financial bill risk assessment, and in particular to a method, device, equipment and storage medium for calculating the risk value of a financial bill. Background Art

[0002] In the context of the booming digital economy today, digital automation transformation has become an urgent need. As a key link in supporting the steady development of enterprises, the standardization and improvement of the management level of expense reimbursement management are particularly important. After years of development and practice, financial shared services have emerged as a new financial management model. This model integrates and optimizes the original accounting and financial business of the enterprise, and uses scientific and standardized unified management methods to effectively reduce operating costs, improve work efficiency, ensure information quality, and reduce management risks.

[0003] The large amount of data accumulated by the financial shared center provides a solid foundation for enterprises to build financial digital capabilities. By deeply exploring data insights, enterprises can drive business decisions more accurately, inject strong momentum into digital transformation, and thus achieve comprehensive digital transformation in the financial field. In this process, risk management plays a vital role. In particular, financial fraud risks are often difficult for enterprises to detect in a timely manner due to their high concealment. They often learn about related risks only after the regulatory authorities issue a notice. In order to avoid such risks in advance, companies and institutions urgently need to strengthen their internal audit and internal control capabilities to ensure the stability and security of corporate finances. Summary of the invention

[0004] In view of the above problems, this application provides a method for calculating the risk value of financial bills, including the following contents:

[0005] In a first aspect, the present application provides a method for calculating risk value of financial bills, the method comprising:

[0006] Obtain financial documents for risk value calculation;

[0007] The data obtained after data processing of the financial bill is input into a pre-trained risk classification model, and the risk probability of the financial bill is obtained based on the risk classification model; the pre-trained risk classification model is obtained by training based on multiple anomaly detection models and classification models, and the multiple anomaly detection models are used to perform anomaly detection on the input data to obtain risk data samples; the classification model is used to obtain the risk probability of the input data based on the risk data samples;

[0008] The risk probability is converted into a corresponding risk value.

[0009] Optionally, the method further includes:

[0010] After the anomaly detection model identifies the anomaly, the model interpretation method SHAP algorithm is used to analyze the feature sources of the anomaly.

[0011] Optionally, the method further includes:

[0012] When the pre-trained risk classification model outputs results, the SHAP algorithm is used to interpret the output of the model.

[0013] Optionally, the training process of the classification model includes:

[0014] Construct multiple unsupervised models, perform anomaly detection on sample data based on the constructed multiple unsupervised models, and obtain risk data samples;

[0015] A supervised model is trained based on the obtained risk data samples to obtain a trained risk classification model.

[0016] Optionally, the performing anomaly detection on sample data based on the constructed multiple unsupervised models includes:

[0017] The sample data were analyzed using the isolation forest model, the one-class support vector machine model and the local anomaly factor detection model;

[0018] When data is confirmed as an outlier by more than two models, it is preliminarily determined that there is an abnormal risk;

[0019] The data that is initially confirmed to have abnormal risks will be re-confirmed for risk. If the confirmation result still indicates that the data has abnormal risks, the data will be used as a risk data sample.

[0020] Optionally, converting the risk probability into a corresponding risk value includes:

[0021] Based on the scoring card model, the risk probability is processed to obtain a corresponding score; the scoring card model is used to nonlinearly convert the probability of 0-1 into a score of 0-1000.

[0022] Optionally, the method further includes:

[0023] Classify the risk value of financial bills into corresponding risk levels according to the preset precision rate and the preset recall rate;

[0024] Set a risk threshold, and send high-risk bills above the threshold to the manual review process, and low-risk bills below the threshold to the rule engine for review.

[0025] In a second aspect, the present application provides a financial note risk value calculation device, the device comprising:

[0026] An acquisition unit, used for acquiring financial documents to be subjected to risk value calculation;

[0027] A processing unit, used for inputting data obtained after data processing of the financial bill into a pre-trained risk classification model, and obtaining the risk probability of the financial bill based on the risk classification model; the pre-trained risk classification model is obtained by training based on a plurality of anomaly detection models and classification models, the plurality of anomaly detection models are used for performing anomaly detection on the input data to obtain risk data samples; the classification model is used for obtaining the risk probability of the input data based on the risk data samples;

[0028] A conversion unit is used to convert the risk probability into a corresponding risk value.

[0029] Optionally, the device further comprises a data interpretation unit for analyzing the feature sources of the abnormal points using a model interpretation method SHAP algorithm after the abnormal points are identified by the abnormal detection model.

[0030] Optionally, the device further comprises a data interpretation unit for using a SHAP algorithm to interpret the output of the model when the pre-trained risk classification model outputs a result.

[0031] Optionally, the device further includes a model training unit for training a classification model, and the training process of the classification model includes:

[0032] Construct multiple unsupervised models, perform anomaly detection on sample data based on the constructed multiple unsupervised models, and obtain risk data samples;

[0033] A supervised model is trained based on the obtained risk data samples to obtain a trained risk classification model.

[0034] Optionally, the performing anomaly detection on sample data based on the constructed multiple unsupervised models includes:

[0035] The sample data were analyzed using the isolation forest model, the one-class support vector machine model and the local anomaly factor detection model;

[0036] When data is confirmed as an outlier by more than two models, it is preliminarily determined that there is an abnormal risk;

[0037] The data that is initially confirmed to have abnormal risks will be re-confirmed for risk. If the confirmation result still indicates that the data has abnormal risks, the data will be used as a risk data sample.

[0038] Optionally, the conversion unit is specifically used to process the risk probability based on a scoring card model to obtain a corresponding score; the scoring card model is used to non-linearly convert a probability of 0-1 into a score of 0-1000.

[0039] Optionally, the device further comprises a risk level classification unit, which is used to classify the risk value of the financial bill into corresponding risk levels according to a preset precision rate and a preset recall rate;

[0040] Set a risk threshold, and send high-risk bills above the threshold to the manual review process, and low-risk bills below the threshold to the rule engine for review.

[0041] In a third aspect, the present application provides a device comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the financial note risk value calculation method introduced in any implementation of the first aspect.

[0042] In a fourth aspect, the present application provides a computer-readable storage medium, in which codes are stored. When the codes are executed, the device executing the codes implements the financial note risk value calculation method introduced in any implementation of the first aspect.

[0043] The present application provides a method for calculating the risk value of financial bills. By integrating multiple anomaly detection models and classification models, the method can comprehensively and deeply explore the potential risks in financial bills. The anomaly detection model can accurately identify anomalies in the data and provide high-quality risk data samples for the classification model, thereby ensuring the accuracy and reliability of the risk probability calculation. The use of a pre-trained risk classification model to predict the risk probability greatly improves the calculation efficiency. Compared with traditional risk assessment methods, this method does not require manual tedious data analysis and judgment, but directly automates the processing through the model, significantly shortening the risk identification cycle. The risk probability is converted into the corresponding risk value, making the risk assessment results more intuitive and easy to understand. This not only facilitates risk management and decision-making by financial-related departments, but also helps enterprises to fully control their financial risk status.

[0044] In summary, this method has significant advantages in calculating the risk value of financial bills. It not only improves the accuracy and efficiency of risk identification, but also provides enterprises with more intuitive and comprehensive risk assessment results, helping enterprises to better cope with financial risk challenges. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A flowchart of a method for calculating risk value of financial bills provided in an embodiment of the present application;

[0047] Figure 2 A flow chart of a classification model training method provided in an embodiment of the present application;

[0048] Figure 3 A schematic diagram of the structure of a financial note risk value calculation device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, 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 are within the scope of protection of this application.

[0050] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0051] Figure 1 A flowchart of a method for calculating the risk value of a financial note provided in an embodiment of the present application. Figure 1 As shown, the financial note risk value calculation method provided in the embodiment of the present application may include:

[0052] S101. Obtain financial documents for which risk value calculation is to be performed.

[0053] Financial documents are a type of written evidence used in the financial field that records the occurrence of an economic event or transaction. These documents play an important role in the daily operations, financial management, and tax processing of enterprises.

[0054] The risk value of financial instruments is a measure of the expected loss that may be caused by certain characteristics or factors of the instruments within a given period of time under normal market conditions and a specified confidence level. This risk value can help companies assess and manage potential risks associated with financial instruments.

[0055] S102: Input the data obtained after data processing of the financial bill into a pre-trained risk classification model, and obtain the risk probability of the financial bill based on the risk classification model.

[0056] Before calculating the risk value of the bill, the bill needs to be processed to obtain the corresponding data of the bill, including data preprocessing operations such as cleaning, format conversion, and extraction of key information on the original financial bill, so as to convert it into a format suitable for risk classification model input.

[0057] A pre-trained risk classification model is a machine learning model that has been trained using a large amount of historical data and can predict the risk probability based on the input data (processed financial bills). The risk classification model will calculate a probability value representing the size of the risk based on the input data through its internal algorithm. The higher the probability value, the greater the risk of the financial bill.

[0058] Specifically, the pre-trained risk classification model is trained based on multiple anomaly detection models and classification models. The multiple anomaly detection models are used to perform anomaly detection on the input data to obtain risk data samples; the classification model is used to obtain the risk probability of the input data based on the risk data samples. The anomaly detection model is specifically used to identify anomalies or abnormal behaviors in the data. In the risk assessment of financial bills, the anomaly detection model can identify bills that do not conform to normal transaction patterns, such as abnormal amounts, abnormal counterparties, etc. Based on multiple anomaly detection models, it is possible to mark data with risk values ​​in the data. The classification model is used to classify the input data into different categories. In the risk classification model, the classification model will directly output a probability value representing the size of the risk based on the detection results of the anomaly detection model and other features.

[0059] S103: Convert the risk probability into a corresponding risk value.

[0060] After obtaining the risk probability of the financial bill to be calculated for risk value through the pre-trained risk classification model, although the probability predicted by the supervised model (classification model) can reflect the fraud risk level of the bill, it is difficult to distinguish because the positive risk probabilities of a large number of bills are concentrated in dense areas. Since there are fewer fraud samples in this scenario, the data label value is generally much more than the label value of 1, that is, there is an imbalance problem in the data, which will cause the predicted positive probability to be overall smaller, and the threshold for classifying whether a case is risky or not given by the classification model is low. Therefore, it is necessary to perform derivative calculations based on the probability predicted by the model to facilitate direct application in business scenarios. In one implementation of the embodiment of the present application, based on the scoring card model, the risk probability is processed to obtain the corresponding score, and the scoring card model is used to convert the probability of 0-1 into a score of 0-1000 nonlinearly. Specifically, the skewed probability distribution can be effectively converted to a normal distribution through the scoring card model to solve the above problem. The conversion formula of the scoring card model to convert the probability value into a score is:

[0061] score=AB×ln(odds)

[0062]

[0063] Where odds represents the probability ratio of bad samples to good samples in the target value, which is the probability ratio of positive cases to non-positive cases in the present embodiment; pdo (point of double odds) represents that when odds doubles, the score will be reduced by pdo points; P0 represents the probability ratio of a specific probability ratio. In the application process, you need to set pdo, P0 and The default scores set in the common financial risk control library scorecardy are pdo=50, P0=600, According to different actual scenarios, data distribution and business requirements, different scoring card parameters can be set by yourself, and the converted score distribution only needs to be balanced and reasonable.

[0064] The above-mentioned method for calculating the risk value of financial bills in this application is explained through an embodiment. According to the above method, the risk probability of the bill to be tested can be obtained. In order to make the result more accurate, the probability value is further converted into a score value. In addition, in the field of financial risk control, the interpretability of the model occupies an important position. However, the current mainstream machine learning models generally face the challenge of being difficult to directly reveal the importance of local features. To solve this problem, the SHAP (i.e. SHapley Additive exPlanations) method is introduced in this application. SHAP accurately calculates the marginal contribution of each feature to the model prediction results, assigns a corresponding importance value to each feature, thereby achieving an effective explanation of the model output. This method can deeply analyze the composition of individual prediction results and intuitively display the dominant features and their influence in a single sample.

[0065] In this application, the SHAP method is applied to two key links. First, after the anomaly detection model identifies the anomalies, the SHAP technology can be used to trace the characteristic roots of these anomalies, providing a strong basis for the business department to determine whether these anomalies constitute an effective fraud risk. Secondly, after the supervised model is constructed, this application uses the SHAP method to output the risk characteristics of each bill, provide clear guidance for the business department, and assist in the formulation and implementation of business specifications. Specifically, after the anomaly detection model identifies the anomalies, the model interpretation method SHAP algorithm is used to analyze the characteristic sources of the anomalies. When the pre-trained risk classification model outputs the result, the SHAP algorithm is used to interpret the output of the model.

[0066] The above describes a method for calculating the risk value of a financial bill of the present application through an embodiment. The method for calculating the risk value of a financial bill uses a pre-trained risk classification model. Figure 2 The training process of the risk classification model is calculated. Figure 2 A flow chart of a training method for a classification model provided in an embodiment of the present application is shown as follows: Figure 2 As shown in Figure 1, the training process of the classification model includes:

[0067] S201. Construct a variety of unsupervised models, and perform anomaly detection on sample data based on the constructed multiple unsupervised models to obtain risk data samples.

[0068] The various unsupervised models constructed in this application include the isolation forest model, the one-class svm model and the Local outlier factor (LOF) model. The isolation forest is an unsupervised anomaly detection algorithm that can handle large-scale multidimensional data and has good robustness to high-dimensional data. Its basic principle is: using a binary search tree structure called an isolation tree (iTree) to isolate samples. Due to the small number of abnormal data and the alienation from most samples, fewer steps are required when isolated, that is, abnormal samples are more likely to fall into leaf nodes quickly. The algorithm first recursively constructs a binary tree from the training data to form an isolation forest. By calculating the path length and average path length of the sample point in each isolation tree, the abnormal score of each data point is obtained, and whether it is an abnormal data point is determined according to the set threshold; One-Class SVM distinguishes between normal and abnormal classes by constructing a hyperspherical decision boundary. The goal of OCSVM (One-Class SVM) is to learn a high-dimensional feature space through training data so that normal data is as close to the decision boundary as possible in this space, while abnormal data is far away from the decision boundary. During the training process, the algorithm learns the decision boundary by minimizing the quadratic loss function and uses regularization terms to control the model complexity and prevent overfitting. OCSVM can also solve nonlinear classification problems by mapping the input space to a high-dimensional feature space using a kernel function; the local anomaly factor algorithm uses the local neighborhood information around the data point to calculate the degree of abnormality of each data point. The algorithm considers both the local and global properties of the data set. The core idea of ​​the LOF algorithm is that anomalies usually have lower density in the neighborhood around them, while normal points have higher density in the neighborhood around them. Therefore, it is possible to determine whether a data point is an anomaly by comparing the density of other data points in its neighborhood. The specific process is as follows: for each data point, calculate its distance to all other points and sort them from near to far; then find the k nearest neighbor points of each data point and get the corresponding score through the LOF calculation formula; finally, if the LOF value is larger, the data point is more abnormal, and vice versa, if the score is smaller, the data point is more normal.

[0069] The abnormality detection of sample data based on the constructed multiple unsupervised models includes:

[0070] The sample data is analyzed using the isolation forest model, the one-class support vector machine model and the local anomaly factor detection model. When the data is confirmed as an outlier by more than two models, it is preliminarily determined that there is an abnormal risk. The data preliminarily confirmed to have an abnormal risk is reconfirmed for risk. If the confirmation result still indicates that the data is at an abnormal risk, the data is used as a risk data sample.

[0071] First, we use three advanced algorithms, the Isolation Forest Model, the One-Class Support Vector Machine Model, and the Local Anomaly Factor Detection Model, to conduct an in-depth and detailed analysis of the sample data. These models identify anomalies in the data based on different principles and mechanisms. Next, we set a judgment standard: if a piece of data is marked as an anomaly by two or more of the above models at the same time, it is preliminarily determined that the data has an abnormal risk. This standard is designed to ensure the accuracy and reliability of anomaly detection and reduce the possibility of misjudgment through cross-validation of multiple models.

[0072] However, the preliminary confirmation of abnormal risk does not mean final confirmation. In order to further improve the accuracy of detection, the data initially confirmed to have abnormal risk will be subject to more rigorous risk reconfirmation. This step may involve more complex analysis methods and manual review. Finally, if the data is still judged to have abnormal risk after reconfirmation, it will be formally classified as a risk data sample. These risk data samples will serve as an important basis for subsequent risk management and control, helping to timely discover and respond to potential financial risks.

[0073] S202: Train a supervised model based on the obtained risk data samples to obtain a trained risk classification model.

[0074] After obtaining the risk data sample data, the sample data is cleaned, including processing missing values, outliers, duplicate values, etc., to ensure data quality. Then, about a hundred valid fields are extracted from the processed sample data as features, which should be able to reflect the risk characteristics of the bill. Necessary transformations are performed on the features, such as encoding categorical variables, standardizing or normalizing numerical variables, etc., to improve model performance.

[0075] In view of the huge amount of data and high feature dimension in the application scenarios corresponding to this application, the LightGBM algorithm is selected as the classification model. LightGBM performs well in processing large-scale data, high-dimensional features and distributed computing, and can evaluate the importance of global features. Use risk data samples (labeled data sets) to train the LightGBM model to obtain a preliminary classification model. Use the LightGBM algorithm to evaluate the importance of global features and identify features that contribute more to model prediction. Screen according to feature importance and delete features with lower importance to reduce model complexity and avoid overfitting. Use the Bayesian optimization method to compare the loss function (such as logarithmic loss, cross entropy, etc.) and AUC (Area Under Curve) values ​​of the model under different parameter combinations to find the optimal parameter combination. Use the screened features and the optimal parameter combination to retrain the LightGBM model to obtain the optimal risk classification model. The model is verified by cross-validation, retention method and other methods to ensure the stability and generalization ability of the model.

[0076] In practical applications, the field-related information of the real-time bill is substituted into the optimal risk classification model to obtain the fraud prediction probability of the bill.

[0077] After converting the risk probability into the corresponding risk value, this application further divides the risk level, and converts the calculated financial bill fraud risk score into the corresponding risk level according to the preset precision and recall standards. This division is determined according to the specific requirements of the business party. For example, if the business party expects the precision rate to reach about 90%, and the bills with a score of more than 600 in the model prediction account for more than 90% of the actual fraudulent bills, 600 points can be set as the critical value between fraud and non-fraud, and the bills can be divided into different risk levels accordingly.

[0078] In addition, a risk threshold is set to distinguish between high-risk and low-risk bills. Bills above this threshold are considered high-risk and will flow directly into the manual review process to ensure in-depth review and timely handling of these potential fraud cases. Low-risk bills below the threshold will enter the rule engine review stage. The rule engine embeds multiple composite rules developed by the business side to detect abnormal conditions that may exist in the bill, such as duplicate invoice numbers and overlapping travel and accommodation expenses. If a bill triggers any of these rules, it will also be marked as high-risk and transferred to the manual review process. This dual review mechanism is designed to further reduce risks and improve the accuracy and efficiency of fraud detection.

[0079] For bills that enter the manual review process, professionals will conduct a detailed review and make a final judgment based on the actual situation. This step ensures that high-risk bills are properly handled, while reducing the possibility of misjudgment or omission of the model. Through the detailed implementation of the above steps, the risks of financial bills can be effectively managed, ensuring that high-risk cases receive timely attention and processing, while improving the overall risk prevention and control capabilities.

[0080] In order to be able to intuitively present the audit conclusions and risk sources, an intelligent risk control platform is also set up in this application. The core of this intelligent risk control platform is to intuitively present the audit conclusions and risk sources. Specifically, the data panel will display the results of the financial risk monitoring model in real time, including the risk score and risk level of each bill, as well as the corresponding early warning prompts. In this way, users can understand the current risk situation at a glance. In addition, the platform will also reveal in detail the possible risk categories and specific financial risk points. These risk points are based on the importance of the model and the SHAP (SHapleyAdditive exPlanations) value sorting, which ensures the accuracy and pertinence of risk identification. In order to display the risks more intuitively, the platform will generate corresponding risk descriptions by combining univariate data analysis with business speech, providing strong data support for financial-related departments, facilitating their risk verification and subsequent processing.

[0081] To sum up, by adding the page display function, this intelligent risk control management platform not only improves the efficiency and accuracy of risk identification, but also provides convenient risk management methods for financial-related departments, which helps enterprises to fully control financial risks and ensure the stability and security of business operations.

[0082] The above are some specific implementations of a financial bill risk value calculation method provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization.

[0083] Figure 3 A schematic diagram of a financial note risk value calculation device provided in an embodiment of the present application. Figure 3 As shown, the financial note risk value calculation device 300 provided in the embodiment of the present application includes:

[0084] An acquisition unit 310 is used to acquire a financial document to be subjected to risk value calculation;

[0085] The processing unit 320 is used to input the data obtained after data processing of the financial bill into a pre-trained risk classification model, and obtain the risk probability of the financial bill based on the risk classification model; the pre-trained risk classification model is obtained by training based on multiple anomaly detection models and classification models, and the multiple anomaly detection models are used to perform anomaly detection on the input data to obtain risk data samples; the classification model is used to obtain the risk probability of the input data based on the risk data samples;

[0086] The conversion unit 330 is used to convert the risk probability into a corresponding risk value.

[0087] In one implementation of the embodiment of the present application, the device also includes a data interpretation unit, which is used to analyze the feature source of the abnormal point using the model interpretation method SHAP algorithm after the abnormal point is identified by the abnormal detection model.

[0088] In one implementation of the embodiment of the present application, the device also includes a data interpretation unit, which is used to use the SHAP algorithm to interpret the output of the model when the pre-trained risk classification model outputs a result.

[0089] In an implementation of the embodiment of the present application, the device further includes a model training unit, which is used to train the classification model. The training process of the classification model includes:

[0090] Construct multiple unsupervised models, perform anomaly detection on sample data based on the constructed multiple unsupervised models, and obtain risk data samples;

[0091] A supervised model is trained based on the obtained risk data samples to obtain a trained risk classification model.

[0092] In one implementation of the embodiment of the present application, the abnormality detection of sample data based on the constructed multiple unsupervised models includes:

[0093] The sample data were analyzed using the isolation forest model, the one-class support vector machine model and the local anomaly factor detection model;

[0094] When data is confirmed as an outlier by more than two models, it is preliminarily determined that there is an abnormal risk;

[0095] The data that is initially confirmed to have abnormal risks will be re-confirmed for risk. If the confirmation result still indicates that the data has abnormal risks, the data will be used as a risk data sample.

[0096] In one implementation of the embodiment of the present application, the conversion unit is specifically used to process the risk probability based on a scoring card model to obtain a corresponding score; the scoring card model is used to non-linearly convert a probability of 0-1 into a score of 0-1000.

[0097] In an implementation of the embodiment of the present application, the device further includes a risk level classification unit, which is used to classify the risk value of the financial bill into corresponding risk levels according to a preset precision rate and a preset recall rate;

[0098] Set a risk threshold, and send high-risk bills above the threshold to the manual review process, and low-risk bills below the threshold to the rule engine for review.

[0099] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0100] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method described in any embodiment of the present application.

[0101] The computer storage medium stores codes, and when the codes are executed, a device executing the codes implements the method described in any embodiment of the present application.

[0102] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0103] It is understandable that in the specific implementation of the present application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved, when the above embodiments of the present application are applied to specific products or technologies, need to obtain user permission or consent, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0104] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0105] It should also be noted that the various embodiments in this specification are described in a progressive manner, and the same and similar parts between the various embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments. The device and apparatus embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0106] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for calculating risk value of financial bills, characterized in that: The method comprises: Obtain financial documents for risk value calculation; The data obtained after data processing of the financial bill is input into a pre-trained risk classification model, and the risk probability of the financial bill is obtained based on the risk classification model; the pre-trained risk classification model is obtained by training based on multiple anomaly detection models and classification models, and the multiple anomaly detection models are used to perform anomaly detection on the input data to obtain risk data samples; the classification model is used to obtain the risk probability of the input data based on the risk data samples; The risk probability is converted into a corresponding risk value.

2. The method according to claim 1, characterized in that: The method further comprises: After the anomaly detection model identifies the anomaly, the model interpretation method SHAP algorithm is used to analyze the feature sources of the anomaly.

3. The method according to claim 1, characterized in that The method further comprises: When the pre-trained risk classification model outputs results, the SHAP algorithm is used to interpret the output of the model.

4. The method according to claim 1, characterized in that: The training process of the classification model includes: Construct multiple unsupervised models, perform anomaly detection on sample data based on the constructed multiple unsupervised models, and obtain risk data samples; A supervised model is trained based on the obtained risk data samples to obtain a trained risk classification model.

5. The method according to claim 4, characterized in that The abnormality detection of sample data based on the constructed multiple unsupervised models includes: The sample data were analyzed using the isolation forest model, the one-class support vector machine model and the local anomaly factor detection model; When data is confirmed as an outlier by more than two models, it is preliminarily determined that there is an abnormal risk; The data that is initially confirmed to have abnormal risks will be re-confirmed for risk. If the confirmation result still indicates that the data has abnormal risks, the data will be used as a risk data sample.

6. The method according to claim 1, characterized in that The converting the risk probability into a corresponding risk value comprises: Based on the scoring card model, the risk probability is processed to obtain a corresponding score; the scoring card model is used to nonlinearly convert the probability of 0-1 into a score of 0-1000.

7. The method according to claim 1, characterized in that The method further comprises: Classify the risk value of financial bills into corresponding risk levels according to the preset precision rate and the preset recall rate; Set a risk threshold, and send high-risk bills above the threshold to the manual review process, and low-risk bills below the threshold to the rule engine for review.

8. A financial bill risk value calculation device, characterized in that: The device comprises: An acquisition unit, used for acquiring financial documents to be subjected to risk value calculation; A processing unit, used for inputting data obtained after data processing of the financial bill into a pre-trained risk classification model, and obtaining the risk probability of the financial bill based on the risk classification model; the pre-trained risk classification model is obtained by training based on a plurality of anomaly detection models and classification models, the plurality of anomaly detection models are used for performing anomaly detection on the input data to obtain risk data samples; the classification model is used for obtaining the risk probability of the input data based on the risk data samples; A conversion unit is used to convert the risk probability into a corresponding risk value.

9. A computing device, characterized in that The computing device includes: a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images