Abnormal value adjustment method and device, electronic equipment and storage medium

Abstract

The invention provides an abnormal value adjustment method and device, electronic equipment and a storage medium, and relates to the technical field of financial risk control. In the application, loan application data and a risk control policy document are acquired, policy constraint conditions are extracted from the risk control policy document, user features in the loan application data are classified based on the policy constraint conditions to obtain a first user feature subset and a second user feature subset, and the user features in the loan application data are classified based on a first constraint sub-condition and a second constraint sub-condition. Performing abnormal value judgment on the first user feature subset and the second user feature subset to obtain at least one first abnormal sample and at least one second abnormal sample, and performing abnormal value adjustment on the at least one first abnormal sample and the at least one second abnormal sample based on the feature types to obtain the abnormal value of the at least one first abnormal sample and the at least one second abnormal sample. And obtaining an abnormal sample set and an adjusted abnormal sample, and adding the adjusted abnormal sample to a normal sample set. By means of the method, the accuracy of abnormal value judgment and adjustment is improved.

Description

Technical Field

[0001] This application relates to the field of financial risk control technology, and in particular to an outlier adjustment method, device, electronic device and storage medium. Background Technology

[0002] In the field of financial risk control technology, financial institutions typically train a risk prediction model in advance to conduct credit assessments of customers. Based on the risk value output by the risk control model, they predict the risk of customers and provide effective data support for financial business, such as credit approval, credit limit management, and risk quota setting.

[0003] In related technologies, risk prediction models are trained based on customers' financial credit data. Since financial credit data contains multi-dimensional values, outliers in these values ​​can easily interfere with model training, causing the training model to overfit the training samples and resulting in low generalization ability and prediction accuracy of the final risk prediction model.

[0004] Therefore, in order to remove outliers from multi-dimensional values, the statistical quantile method is usually used to process these outliers. The statistical quantile method will delete values ​​that exceed a set multiple in multi-dimensional values, but it cannot distinguish between policy constraints and real anomalies. For example, the policy constraint is to reject loans for people over 60 years old, while the real anomaly is an age of 300 years old. According to the policy constraint, samples that violate the policy will be deleted, such as only retaining data with an age distribution of 18-60 years old. This will destroy the original distribution of features, making the variable distribution uneven and reducing the risk prediction model's ability to learn about overall risk. Alternatively, high-quality anomaly label data can be obtained to train the risk prediction model. However, the acquisition cost of high-quality anomaly label data is high and it is difficult to adapt to the dynamic changes in risk control policies, thus affecting business efficiency.

[0005] In summary, the above methods of handling outliers will delete valid data, distort data distribution, and fail to respond quickly to changes in risk control policies, resulting in low accuracy in outlier removal. Summary of the Invention

[0006] This application provides an outlier adjustment method, apparatus, electronic device, and storage medium to ensure the accuracy of outlier determination, adjustment, and analysis.

[0007] In a first aspect, embodiments of this application provide an outlier adjustment method, the method comprising: Obtain loan application data and risk control policy documents; the loan application data includes multiple user characteristics for each user. Extract policy constraints from risk control policy documents, and classify user characteristics in loan application data based on policy constraints to obtain a first subset of user characteristics and a second subset of user characteristics; Determine the first and second constraint sub-conditions in the policy constraints, and based on the first and second constraint sub-conditions, perform outlier determination on the first and second user feature subsets respectively to obtain at least one first outlier sample and at least one second outlier sample. Based on the feature types corresponding to multiple user features, at least one first abnormal sample and at least one second abnormal sample are adjusted for outlier values ​​to obtain an abnormal sample set and adjusted abnormal samples. The adjusted abnormal samples are then added to the normal sample set.

[0008] In one optional implementation, user characteristics in loan application data are classified based on policy constraints to obtain a first subset of user characteristics and a second subset of user characteristics, including: For each user characteristic in the loan application data, perform the following operations respectively: The system checks whether the first user characteristic in the loan application data meets the policy constraints; where the first user characteristic can be any one of the various user characteristics. If the first user characteristic meets the policy constraints, then the first user characteristic will be classified into the first user characteristic subset; If the first user characteristic does not meet the policy constraints, then the first user characteristic will be classified into the second user characteristic subset.

[0009] In one alternative implementation, classifying the first user feature into a subset of the second user features includes: The multiple feature values ​​included in the feature value set corresponding to the first user feature are sorted in order to obtain the first feature value corresponding to the first quartile and the second feature value corresponding to the third quartile. Determine the difference between the first eigenvalue and the second eigenvalue, and define the difference as the interquartile range difference; A first outlier threshold is calculated based on the difference between the first feature value and the interquartile range, and a second outlier threshold is calculated based on the difference between the second feature value and the interquartile range; wherein the first outlier threshold is less than the second outlier threshold. Feature values ​​that do not fall within the outlier range between the first and second outlier thresholds are classified into the second user feature subset.

[0010] In one optional implementation, outlier adjustment is performed on at least one first outlier sample and at least one second outlier sample based on the feature types corresponding to each of the multiple user features, resulting in an outlier sample set and adjusted outlier samples, including: Based on the feature type corresponding to each first abnormal sample, retain the abnormal boundary value in each first abnormal sample to obtain the adjusted abnormal sample; Based on the feature type corresponding to each second abnormal sample, the outliers in each second abnormal sample are subjected to one-sided or two-sided tail reduction processing to obtain the adjusted abnormal samples.

[0011] In one alternative implementation, the feature types include: hard policy constraint type and soft policy constraint type; wherein the adjustment frequency of the hard policy constraint type is lower than that of the soft policy constraint type.

[0012] In one alternative implementation, it also includes: Receive adjustment requests corresponding to policy constraints; The target policy constraints are parsed from the adjustment request, and the differences between the policy constraints and the target policy constraints corresponding to the adjustment request are compared to obtain the updated constraints. The outlier threshold is adjusted based on the updated constraints to obtain the target outlier threshold.

[0013] In one optional implementation, based on the feature types corresponding to multiple user features, at least one abnormal sample undergoes outlier adjustment to obtain an abnormal sample set and adjusted abnormal samples. After adding the adjusted abnormal samples to the normal sample set, the implementation further includes: Label the abnormal sample set as negative samples and the normal sample set as positive samples; Positive and negative samples are input into the risk prediction model for feature extraction, and multiple loan risk features are output. Users in the abnormal sample set and the normal sample set are relabeled based on multiple loan risk characteristics to obtain the target abnormal sample set and the target normal sample set.

[0014] Secondly, embodiments of this application also provide an outlier adjustment device, the device comprising: The acquisition module is used to acquire loan application data and risk control policy documents; the loan application data includes multiple user characteristics for each user. The classification module is used to extract policy constraints from risk control policy documents, classify user characteristics in loan application data based on policy constraints, and obtain a first subset of user characteristics and a second subset of user characteristics. The determination module is used to determine the first and second constraint sub-conditions in the policy constraints, and based on the first and second constraint sub-conditions, to determine outliers in the first and second user feature subsets respectively, so as to obtain at least one first outlier sample and at least one second outlier sample. The adjustment module is used to adjust the outlier values ​​of at least one first outlier sample and at least one second outlier sample based on the feature types corresponding to multiple user features, to obtain an outlier sample set and the adjusted outlier samples, and to add the adjusted outlier samples to the normal sample set.

[0015] In one optional implementation, the classification module is specifically used to perform the following operations for each user feature in the loan application data: detect whether the first user feature in the loan application data meets the policy constraints; wherein, the first user feature is any one of the user features, if the first user feature meets the policy constraints, then the first user feature is classified into the first user feature subset, if the first user feature does not meet the policy constraints, then the first user feature is classified into the second user feature subset.

[0016] In one optional implementation, the classification module is further configured to sequentially sort the multiple feature values ​​included in the feature value set corresponding to the first user feature to obtain the first feature value corresponding to the first quartile and the second feature value corresponding to the third quartile, determine the difference between the first feature value and the second feature value, define the difference as the interquartile range difference, calculate the first outlier threshold based on the first feature value and the interquartile range difference, and calculate the second outlier threshold based on the second feature value and the interquartile range difference; wherein, if the first outlier threshold is less than the second outlier threshold, feature values ​​that do not belong to the outlier range between the first outlier threshold and the second outlier threshold are classified into the second user feature subset.

[0017] In one optional implementation, the adjustment module is specifically used to retain the abnormal boundary value in each first abnormal sample based on the feature type corresponding to each first abnormal sample to obtain the adjusted abnormal sample, and to perform one-sided or two-sided tail reduction processing on the abnormal value in each second abnormal sample based on the feature type corresponding to each second abnormal sample to obtain the adjusted abnormal sample.

[0018] In one optional implementation, the adjustment module is further configured to receive adjustment requests corresponding to policy constraints, parse the target policy constraints from the adjustment requests, compare the differences between the policy constraints and the target policy constraints corresponding to the adjustment requests to obtain updated constraints, and adjust the outlier threshold based on the updated constraints to obtain the target outlier threshold.

[0019] In one alternative implementation, the adjustment module is further configured to mark the abnormal sample set as negative samples and the normal sample set as positive samples, input the positive and negative samples into the risk prediction model for feature extraction, output multiple loan risk features, and re-label the users in the abnormal and normal sample sets based on the multiple loan risk features to obtain the target abnormal sample set and the target normal sample set.

[0020] Thirdly, embodiments of this application provide an electronic device, including: Memory, used to store computer programs; When the processor executes the computer program stored in the memory, it implements the above-described steps of the outlier adjustment method.

[0021] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described outlier adjustment method steps.

[0022] The beneficial effects of this application are as follows: In the outlier adjustment method provided in this application embodiment, loan application data is initially screened based on policy constraints in the risk control policy document to remove obviously non-compliant data. Subsequently, outlier determination and adjustment are performed on the first and second user feature subsets based on the first and second constraint sub-conditions in the policy constraints, respectively, to accurately identify outliers. Furthermore, a risk prediction model is used to learn from the normal and outlier sample sets to distinguish between policy constraints and the user's actual risk, ensuring improved identification of marginal users. When the risk control policy document is updated, the outlier threshold can be dynamically adjusted to obtain the target outlier threshold, ensuring the accuracy of the target outlier threshold and meeting the current risk control business's requirements for accuracy and adaptability.

[0023] Furthermore, other features and advantages of this application will be set forth in the following description and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described herein are used to provide a further understanding of this application, constitute a part of this application, and do not constitute an improper limitation of this application. In the accompanying drawings: Figure 1 This is a schematic diagram of an application scenario provided by an embodiment of this application; Figure 2 A schematic diagram illustrating the implementation process of an outlier adjustment method provided in this application embodiment; Figure 3 This application provides a schematic diagram of the process for determining outliers from loan application data using IQR, as illustrated in the embodiments of this application. Figure 4 This is a flowchart illustrating the process of reclassifying abnormal and normal sample sets based on loan risk characteristics, as provided in an embodiment of this application. Figure 5 This is a flowchart illustrating the outlier handling process under the risk control policy document provided in this application embodiment. Figure 6 This is a schematic diagram of an outlier adjustment device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The specific operational methods in the method embodiments can also be applied to the device embodiments or system embodiments. It should be noted that in the description of this application, "multiple" is understood as "at least two". "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. A connected to B can represent: A and B directly connected, and A and B connected through C. Furthermore, in the description of this application, terms such as "first" and "second" are used only for distinguishing the purpose of description and should not be construed as indicating or implying relative importance or order.

[0026] The following explanations of some terms used in the embodiments of this application are provided to facilitate understanding by those skilled in the art.

[0027] (1) Policy hard constraint type: User characteristics that are explicitly prohibited in risk control policy documents, such as age.

[0028] (2) Soft constraint type: User characteristics in risk control policy documents that can be adjusted as the financial market changes or the current risk control business changes.

[0029] Based on the above explanations of terms and related terminology, the design concept of the embodiments of this application will be briefly introduced below: This application embodiment classifies loan application data according to policy constraints, thereby achieving preliminary screening of outliers in the loan application data, obtaining a first user feature subset and a second user feature subset, and then performing outlier determination and adjustment on the first user feature subset and the second user feature subset based on different feature types, thereby meeting the adaptability and accuracy requirements of current risk control business.

[0030] In related technologies, statistical quantiles are commonly used to process outliers. However, statistical quantiles cannot distinguish between policy constraints and genuine anomalies. Policy constraints will delete samples that violate the policy, disrupting the original distribution of features and making the variable distribution uneven. This reduces the risk prediction model's ability to learn about overall risk. In addition, high-quality anomaly label data can be used to train the risk prediction model. However, obtaining high-quality anomaly label data is costly and difficult to adapt to the dynamic changes in risk control policies. Therefore, how to improve the accuracy of outlier removal has become the main problem to be solved.

[0031] In view of this, in order to solve or improve the above problems, this application proposes an outlier adjustment method, which may specifically include: First, loan application data and risk control policy documents are obtained. The loan application data includes multiple user characteristics corresponding to each user. Next, policy constraints are extracted from the risk control policy documents. Based on the policy constraints, the user characteristics in the loan application data are classified to obtain a first user characteristic subset and a second user characteristic subset. Then, the first constraint sub-condition and the second constraint sub-condition in the policy constraints are determined. Based on the first constraint sub-condition and the second constraint sub-condition, outlier determination is performed on the first user characteristic subset and the second user characteristic subset respectively to obtain at least one first outlier sample and at least one second outlier sample. Finally, based on the feature types corresponding to the multiple user characteristics, outlier adjustment is performed on at least one first outlier sample and at least one second outlier sample to obtain an outlier sample set and adjusted outlier samples. The adjusted outlier samples are added to the normal sample set.

[0032] In particular, the preferred embodiments of this application will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments of this application and the features in the embodiments can be combined with each other without conflict.

[0033] See Figure 1The diagram illustrates an application scenario applicable to an embodiment of this application. This scenario may include a terminal device (101a, 101b) and a server 102. The terminal device (101a, 101b) and the server 102 can interact via a communication network. The communication network can employ wireless communication or wired communication methods. For example, the terminal device (101a, 101b) can access the network and communicate with the server 102 via cellular mobile communication technology. This cellular mobile communication technology may include, for example, 5G (5th generation mobile networks) or next-generation mobile communication technology. Optionally, the terminal device (101a, 101b) can access the network and communicate with the server 102 via short-range wireless communication. This short-range wireless communication method may include, for example, Wi-Fi (wireless fidelity) technology.

[0034] This application embodiment does not limit the number of communication devices involved in the above application scenarios. For example, the above system architecture may include more terminal devices, fewer terminal devices, or other network devices. Figure 1 As shown, only terminal devices (101a, 101b) and server 102 are described as examples. The following is a brief introduction to the above communication devices and their respective functions.

[0035] A terminal device (101a, 101b) is a device that can provide voice and / or data connectivity to a user, and may be a device that supports wired and / or wireless connections.

[0036] For example, terminal devices (101a, 101b) may include, but are not limited to: mobile phones, tablets, laptops, handheld computers, mobile internet devices (MID), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminal devices in industrial control, wireless terminal devices in autonomous driving, wireless terminal devices in smart grids, wireless terminal devices in transportation safety, wireless terminal devices in smart cities, or wireless terminal devices in smart homes, etc.

[0037] Furthermore, the terminal devices (101a, 101b) may have related client software installed. This client software can be, for example, an application (APP), a browser, a short video app, or a webpage, a mini-program, etc. It should be noted that the terminal devices (101a, 101b) in this embodiment can enable the aforementioned client software related to loan application data to send loan application data to the server 102 for subsequent outlier removal and other methods.

[0038] Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0039] It is worth mentioning that, in this embodiment of the application, the server 102 can be used to iteratively train the risk prediction model, which is used to extract the user's loan risk characteristics, thereby realizing precise dynamic adjustment of the current risk control business and precise adjustment of outliers.

[0040] The outlier adjustment method provided by the exemplary embodiments of this application will be described below in conjunction with the above application scenarios and with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way in this respect.

[0041] See Figure 2 The diagram shown illustrates the implementation flow of an outlier adjustment method provided in this application. Taking a server as an example, the specific implementation flow of this method is as follows: Step S21: Obtain loan application data and risk control policy documents.

[0042] To ensure accurate and dynamic adjustments to current risk control operations, the server needs to obtain users' initial loan application data and initial risk control policy documents issued by third-party institutions. These initial risk control policy documents record the policy constraints for users. Because these documents are constantly changing with financial market adjustments, and because users' initial loan application data may contain irregularities or outliers, the server needs to check whether the initial risk control policy documents are the latest version. Furthermore, for current risk control operations, the server also needs to check the validity of the initial risk control policy documents to ensure they comply with currently effective laws and regulations.

[0043] Specifically, to ensure that the loan application data processed by the server is standardized, it is necessary to perform integrity verification on the user's initial loan application data to prevent loan failure due to incomplete initial application data. Integrity verification can be achieved through mandatory field verification. In this embodiment, mandatory fields include: ID card number, bank card number, income certificate, loan purpose, etc. Mandatory fields can be added or deleted according to changes in current risk control business. When a mandatory field is missing, the server can mark the missing field to prompt business personnel or users, making it easier for users to supplement the missing field, thereby improving the efficiency of supplementing missing fields.

[0044] In the current risk control process, initial loan application data may contain data with abnormal formats. The server needs to perform format validation on the initial loan application data. Format validation is achieved by verifying the character types and fixed encoding rules in the initial loan application data, thus obtaining the loan application data. The loan application data records multiple user characteristics for each user, such as age, monthly income, and app usage time. When the character type of a required field does not conform to the preset character type, the server determines that the initial loan application data has an abnormal format. For example, an ID number consists of 18 consecutive digits or 17 consecutive digits plus a letter. In this case, the ID number is 11a201547852395412. 11a201547852395412 does not conform to either 18 consecutive digits or 17 consecutive digits plus a letter. Therefore, the server confirms that 11a201547852395412 is data with an abnormal format.

[0045] Furthermore, the fixed encoding rule in this application embodiment can be the encoding rule of the issuing bank for the issued bank card number. Different issuing banks have different encoding rules for bank card numbers. The encoding rule can be the total length of the bank card number, the bank identification code, etc. For example, the default issuing banks of the server are issuing bank A, issuing bank B, and issuing bank C. Issuing bank A, issuing bank B, and issuing bank C each have their own corresponding bank identification codes. At this time, the bank card number in the initial loan application data belongs to overseas bank D. This bank card number does not conform to the fixed encoding rule set by the server. That is, overseas bank D does not belong to the default issuing bank of the server. Therefore, the server will determine that the bank card number in the initial loan application data does not conform to the fixed encoding rule.

[0046] In this embodiment, the server matches the currently received initial risk control policy document with historically input risk control policy documents. When a historical risk control policy document is matched, it determines that the currently received initial risk control policy document is not the latest version, and then determines the latest version of the historical risk control policy document as the risk control policy document. When no historical risk control policy document is matched, it determines that the currently received initial risk control policy document is the latest version, thereby ensuring the accuracy of the risk control policy document and thus ensuring the accuracy of identifying outliers from loan application data.

[0047] In addition, for the current risk control documents, it is also necessary to check the validity of the risk control policy documents. The server needs to perform a semantic comparison between the content explicitly prohibited in the risk control policy documents and the current regulations. When the content explicitly prohibited is consistent with the semantics of the current regulations, the risk control policy documents are deemed valid. When the content explicitly prohibited is inconsistent with the semantics of the current regulations, the inconsistent content needs to be marked in the risk control policy documents. This helps business personnel to identify the marked content in the risk control policy documents and ensures the accuracy of the risk control policy documents.

[0048] The aforementioned current regulations can be the Consumer Rights Protection Law. These regulations can be added to or deleted based on actual risk control operations, which will not be discussed further here.

[0049] By using the methods described above, initial loan application data and initial risk control policy documents are obtained, and these data and documents are verified to obtain loan application data and risk control policy documents. This ensures the standardization of the loan application data and risk control policy documents, thereby improving the efficiency of outlier identification.

[0050] Step S22: Extract policy constraints from the risk control policy document, and classify user characteristics in the loan application data based on the policy constraints to obtain a first subset of user characteristics and a second subset of user characteristics.

[0051] Since the risk control policy document records the policy constraints for users, it is necessary to perform semantic parsing on the risk control policy document to extract multi-dimensional user characteristics and the corresponding feature values ​​for each user characteristic. Based on each user characteristic and its corresponding feature values, policy constraints are generated. For example, the policy constraints generated based on user characteristics and their corresponding feature values ​​are shown in Table 1 below. Table 1 Examples of User Characteristics and Their Corresponding Feature Values ​​and Policy Constraints

[0052] The user characteristics, corresponding feature values, and policy constraints in Table 1 above can all be adjusted according to actual risk control business, and will not be elaborated on here.

[0053] After determining the policy constraints, we check whether each user characteristic in the loan application data meets the policy constraints. Here, we take the first user characteristic as an example. The first user characteristic can be any one of multiple user characteristics. We perform the following operations on the first user characteristic: When the first user characteristic meets the policy constraints, the first user characteristic is classified into the first user characteristic subset; when the user characteristic does not meet the policy constraints, the first user characteristic is classified into the second user characteristic subset.

[0054] To identify the user's true needs and asset quality, in one optional implementation, during step S22, refer to... Figure 3 As shown, this is a second subset of user features obtained by identifying outliers from loan application data through IQR. Taking the server as an example, the specific steps of this method are as follows: S31, sort the multiple feature values ​​included in the feature value set corresponding to the first user feature in order to obtain the first feature value corresponding to the first quartile and the second feature value corresponding to the third quartile.

[0055] Specifically, the first quartile is the value at the first 25% position of the sequentially sorted feature value set, and the value at the 25% position is determined as the first feature value; and the third quartile is the value at the last 75% position of the sequentially sorted feature value set, and the value at the last 75% position is determined as the second feature value.

[0056] S32, determine the difference between the first eigenvalue and the second eigenvalue, and define the difference as the interquartile range difference.

[0057] The interquartile range difference is calculated using the first and second eigenvalues. This interquartile range difference can be used to measure the dispersion of the middle 50% of the eigenvalues. The larger the interquartile range difference, the greater the dispersion of the middle part of the eigenvalues.

[0058] S33, calculate the first outlier threshold based on the difference between the first feature value and the interquartile range, and calculate the second outlier threshold based on the difference between the second feature value and the interquartile range.

[0059] Calculate the difference between the first eigenvalue and the interquartile range difference using a set coefficient multiple parameter. The set coefficient can be 1.5 to obtain the first outlier threshold. Calculate the sum between the second eigenvalue and the interquartile range difference using a set coefficient multiple parameter to obtain the second outlier threshold. The first outlier threshold can be a lower limit value, and the second outlier threshold can be an upper limit value. That is, the first outlier threshold is less than the second outlier threshold.

[0060] S34, classify the feature values ​​that do not belong to the outlier range between the first outlier threshold and the second outlier threshold into the second user feature subset.

[0061] Feature values ​​between the first outlier threshold and the second outlier threshold are added to the normal sample set, while feature values ​​that do not fall within the outlier range between the first outlier threshold and the second outlier threshold are classified into the second user feature subset. This ensures the accuracy of the normal sample set and the outlier sample set, which is beneficial for subsequent dynamic outlier adjustment.

[0062] For example, if the user characteristic in loan application data is the loan amount, the IQR method needs to calculate the first quartile (Q1) of the loan amount, the third quartile (Q3) of the loan amount, and finally, calculate the interquartile range (Q3 - Q1). The upper limit of outliers in the loan amount is calculated as Q3 + 1.5 * IQR, and the lower limit is calculated as Q1 - 1.5 * IQR. Therefore, the effective range for the loan amount is [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR]. When the loan amount in the loan application data exceeds the range [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR], the feature value corresponding to the loan amount that does not belong to the outlier range of [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR] is identified as an outlier.

[0063] Based on the above method, IQR can effectively filter out the influence of extreme feature values ​​in loan application data. As a result, the user features in the normal sample set can better reflect the main situation of the user's loan application, avoiding interference from abnormal data in subsequent analysis. The accurate normal sample set and abnormal sample set provide a basis for subsequent outlier analysis.

[0064] In one possible design, considering that in actual risk control operations, there may be "extreme feature value anomalies" caused by data entry errors, server failures, etc., for example, the loan amount in most loan application data is between 100,000 and 500,000. In this case, the server detects that a user's loan amount is 10 million. The 10 million may be due to the user entering an extra 0 when inputting the loan amount. Therefore, this embodiment of the application can use Z-Score to determine outliers from the loan application data, thereby obtaining a second subset of user features. The execution subject is still the server as an example, and the specific steps of the method are as follows: The Z-score is calculated for each user feature, where Z = (x - mean) / standard deviation, and x is the feature value of each user feature, with a pre-set threshold. Data points exceeding the Z-Score are classified into the second user feature subset. Since the calculation process of the mean and standard deviation is a well-known technique to those skilled in the art, it will not be explained in detail here.

[0065] By using the above method, loan application data is classified based on policy constraints, and outliers are identified from user characteristics that do not meet the policy constraints based on IQR and / or Z-Score. This ensures the accuracy of the identified outliers and helps to improve the accuracy of outlier adjustment in loan application data.

[0066] Step S33: Determine the first and second constraint sub-conditions in the policy constraints. Based on the first and second constraint sub-conditions, perform outlier determination on the first and second user feature subsets respectively to obtain at least one first outlier sample and at least one second outlier sample.

[0067] In order to achieve multi-level risk control, this application implements a first constraint sub-condition and a second constraint sub-condition in the policy constraint conditions that need to be determined. The strictly restrictive conditions in the policy constraint conditions can be determined as the first constraint sub-condition, and the flexible conditions can be determined as the second constraint sub-condition. The second constraint sub-condition can be dynamically adjusted according to the actual business needs of the financial market.

[0068] Based on the first constraint sub-condition, at least one abnormal boundary value is determined from the first user feature subset, and the at least one abnormal boundary value is determined as at least one first abnormal sample. Based on the second constraint sub-condition, at least one abnormal value is determined from the second user feature subset, and the at least one abnormal value is determined as at least one second abnormal sample.

[0069] For example, if the risk control policy document stipulates that being under 18 years old or over 60 years old violates the policy constraint, then at least one abnormal boundary value is determined in the first user feature subset, namely 18 and 60, and 18 and 60 are the first abnormal samples; if the policy constraint is that the monthly income is not less than 3,000 yuan, then the monthly income in the second user feature subset is 2,999 yuan, and 2,999 yuan is determined as the second abnormal sample.

[0070] Using the above method, outlier determination is performed on the first and second user feature subsets based on the first and second constraint sub-conditions in the policy constraints, respectively, thereby identifying invalid and non-compliant data, i.e., identifying outliers and avoiding server decision bias caused by sample confusion.

[0071] Step S4: Based on the feature types corresponding to each of the multiple user features, adjust the outlier values ​​of at least one first outlier sample and at least one second outlier sample to obtain an outlier sample set and adjusted outlier samples. Add the adjusted outlier samples to the normal sample set.

[0072] User features can be of hard policy constraint type or soft policy constraint type. The adjustment frequency of hard policy constraint type is lower than that of soft policy constraint type. If the user feature of at least one anomalous sample corresponds to the hard policy constraint type, then at least one first anomalous sample is subjected to anomalous boundary value retention processing to obtain the adjusted anomalous sample and anomalous sample set. If the user feature of at least one anomalous sample corresponds to the soft policy constraint type, then at least one second anomalous sample is subjected to smoothing processing. The smoothing processing method can be one-sided tail reduction or two-sided tail reduction to obtain the adjusted anomalous sample and anomalous sample set.

[0073] One-sided tail reduction processes only the extreme values ​​at one end, while two-sided tail reduction replaces feature values ​​below the lower limit with the lower limit value and feature values ​​above the upper limit with the upper limit value.

[0074] For example, if the range of monthly income is (3000, 50000), one-sided shrinking can only process feature values ​​with monthly income below 3000, or only process feature values ​​with monthly income above 50000. Two-sided shrinking can adjust feature values ​​with monthly income below 3000 to 3000, and at the same time, adjust feature values ​​with monthly income above 50000 to 50000.

[0075] Based on the above description, samples with an age of 18 or a monthly income of 2999 are identified as the adjusted abnormal sample set. The adjusted abnormal sample set is then added to the normal sample set to ensure that the normal sample set contains more business users, meeting the promotion needs of the financial market and thus ensuring that the adaptability and accuracy of the current risk control business can be met.

[0076] In one possible design, when the risk control policy document is updated, the server receives the adjustment request corresponding to the policy constraints, extracts the target policy constraints from the adjustment request, compares the differences between the target policy constraints and the policy constraints to obtain the updated constraints, and adjusts the outlier threshold based on the updated constraints to obtain the target outlier threshold. This enables a rapid response to the updated risk control policy document, ensures the accuracy of the target outlier threshold, and improves the accuracy of outlier judgment, adjustment, and analysis of loan application data.

[0077] In one alternative implementation, users who are "stuck at the boundaries of risk control policy documents, superficially compliant but actually high-risk" are categorized into the normal sample set, while users who "minorly violate policies but are actually low-risk" are categorized into the abnormal sample set. This wastes valuable customer resources and makes it impossible to distinguish between policy constraints and users' true risk characteristics. Therefore, after executing step S4, refer to [the relevant documentation / reference]. Figure 4 As shown, this method reclassifies abnormal and normal sample sets based on loan risk characteristics. Using a server as the execution entity, the specific steps of this method are as follows: S41, mark the abnormal sample set as negative samples, and the normal sample set as positive samples.

[0078] S42 inputs positive and negative samples into the risk prediction model for feature extraction, and outputs multiple loan risk features.

[0079] By inputting positive and negative samples into the risk prediction model for learning, the risk prediction model can extract features from normal and abnormal sample sets to obtain multiple loan risk features, thereby freeing the loan risk data from the influence of policy constraints.

[0080] S43, based on multiple loan risk characteristics, relabel users in the abnormal sample set and the normal sample set to obtain the target abnormal sample set and the target normal sample set.

[0081] Based on the loan risk characteristics, the abnormal sample set and the normal sample set are relabeled to obtain the target normal sample set and the target abnormal sample set. At this time, the risk prediction model classifies users based on loan risk characteristics, rather than risk control policy documents. Therefore, it ensures the risk prediction model's ability to distinguish between policy constraints and users' true risk characteristics.

[0082] It should be noted that the normal and abnormal sample sets do not directly contribute to the loss calculation, but are used to train an adversarial component, thereby improving the overfitting of the risk prediction model. At the same time, it avoids the risk prediction model training being dominated by the target abnormal sample set, thus improving the robustness of the risk prediction model.

[0083] The risk prediction model can be a generative adversarial network (GAN) model. Since the GAN model is a well-known technology in the art, it will not be elaborated on here.

[0084] For example, when the risk prediction model identifies a marginal user with a monthly income of 3,100 yuan, it uses the user's "number of delinquencies in the past 6 months" as a loan risk characteristic. If the number of delinquencies in the past 6 months exceeds the expected threshold, it means that the marginal user has too many delinquencies, i.e., the user's income meets the policy constraints, and the user is also identified as high-risk. In the second outlier detection process, such marginal users are assigned to the target outlier sample set. In the second output target outlier sample set, low-risk users who were mistakenly rejected are eliminated, and high-risk users who were missed are added. This not only improves the ability to identify marginal users, but also makes the target outlier sample set and the target normal sample set more consistent with the user's actual risk situation, thus improving the accuracy of the target outlier sample set and the target normal sample set.

[0085] During the retraining process of the aforementioned risk prediction model, in addition to outputting the target abnormal sample set and the target normal sample set, it can also output an outlier removal report and a variable distribution comparison chart. The outlier removal report includes: outliers removed based on policy constraints, and outliers removed using the IQR method or the Z-Score method. The variable distribution comparison chart can show the sample data distribution corresponding to the risk control policy document and the updated risk control policy document, or the impact of changes in policy constraints on the population distribution. The embodiments of this application can adjust the content of the variable distribution comparison chart according to the actual needs of the current risk control business. Based on the outlier removal report and the variable distribution comparison chart, outliers can be analyzed more accurately, improving the accuracy of outlier analysis and facilitating accurate subsequent analysis of the current risk control business.

[0086] The flowchart illustrating outlier handling in the risk control policy document in this application embodiment is as follows: Figure 5 As shown, loan application data is input into the server, and risk control policy documents are pre-stored. The risk control policy documents are parsed to obtain policy constraints. Based on the policy constraints, the loan application data is filtered to obtain a first subset of user features and a second subset of user features. Outlier determination is performed on the first subset of user features and the second subset of user features to obtain a first subset of outlier samples and a second subset of outlier samples. Then, outlier adjustment is performed on the first subset of outlier samples based on the type of hard policy constraints, and on the second subset of outlier samples based on the type of soft policy constraints, resulting in a final set of outlier samples and adjusted outlier samples. The adjusted outlier samples are added to the normal sample set. The normal sample set and the outlier sample set are labeled, with the outlier sample set labeled as negative samples and the normal sample set labeled as positive samples. The loan risk characteristics of the positive and negative samples are learned based on the risk prediction model, thereby reclassifying users based on loan risk characteristics to obtain a target set of outlier samples and a target set of normal samples. Finally, the target set of normal samples is output.

[0087] Based on the above description, loan application data is initially screened according to the policy constraints in the risk control policy document to remove obviously non-compliant data. Subsequently, outlier determination and adjustment are performed on the first and second user feature subsets based on the first and second constraint sub-conditions in the policy constraints, respectively, to accurately identify outliers. Furthermore, the risk prediction model learns from the normal and outlier sample sets to distinguish between policy constraints and the user's actual risk, ensuring improved identification of marginal users. When the risk control policy document is updated, the outlier threshold can be dynamically adjusted to obtain the target outlier threshold, ensuring the accuracy of the target outlier threshold and improving the accuracy of outlier determination, adjustment, and analysis, thus meeting the current risk control business's requirements for accuracy and adaptability.

[0088] Furthermore, based on the same technical concept, embodiments of this application provide an outlier adjustment device, which is used to implement the above-described method flow of embodiments of this application. See also... Figure 6 As shown, the outlier adjustment device 600 includes: an acquisition module 601, a classification module 602, a judgment module 603, and an adjustment module 604, wherein: The acquisition module 601 is used to acquire loan application data and risk control policy documents; the loan application data includes multiple user characteristics for each user. The classification module 602 is used to extract policy constraints from risk control policy documents, classify user characteristics in loan application data based on policy constraints, and obtain a first subset of user characteristics and a second subset of user characteristics. The determination module 603 is used to determine the first constraint sub-condition and the second constraint sub-condition in the policy constraint conditions, and based on the first constraint sub-condition and the second constraint sub-condition, to determine the outliers of the first user feature subset and the second user feature subset respectively, and to obtain at least one first outlier sample and at least one second outlier sample. The adjustment module 604 is used to adjust the outlier values ​​of at least one first abnormal sample and at least one second abnormal sample based on the feature types corresponding to the multiple user features, to obtain an abnormal sample set and the adjusted abnormal samples, and to add the adjusted abnormal samples to the normal sample set.

[0089] In one optional implementation, the classification module 602 is specifically used to perform the following operations for each user feature in the loan application data: detect whether the first user feature in the loan application data meets the policy constraints; wherein, the first user feature is any one of the user features; if the first user feature meets the policy constraints, then the first user feature is classified into the first user feature subset; if the first user feature does not meet the policy constraints, then the first user feature is classified into the second user feature subset.

[0090] In one optional implementation, the classification module 602 is further configured to sort the multiple feature values ​​included in the feature value set corresponding to the first user feature in order to obtain the first feature value corresponding to the first quartile and the second feature value corresponding to the third quartile, determine the difference between the first feature value and the second feature value, define the difference as the interquartile range difference, calculate the first outlier threshold based on the first feature value and the interquartile range difference, and calculate the second outlier threshold based on the second feature value and the interquartile range difference; wherein, the first outlier threshold is less than the second outlier threshold, and feature values ​​that do not belong to the outlier range between the first outlier threshold and the second outlier threshold are classified into the second user feature subset.

[0091] In one optional implementation, the adjustment module 604 is specifically used to retain the abnormal boundary value in each first abnormal sample based on the feature type corresponding to each first abnormal sample to obtain the adjusted abnormal sample, and to perform one-sided or two-sided tail reduction processing on the abnormal value in each second abnormal sample based on the feature type corresponding to each second abnormal sample to obtain the adjusted abnormal sample.

[0092] In one optional implementation, the adjustment module 604 is further configured to receive an adjustment request corresponding to the policy constraint, parse the target policy constraint from the adjustment request, compare the policy constraint with the target policy constraint corresponding to the adjustment request to obtain the updated constraint, and adjust the outlier threshold based on the updated constraint to obtain the target outlier threshold.

[0093] In one alternative implementation, the adjustment module 604 is further configured to mark the abnormal sample set as negative samples and the normal sample set as positive samples, input the positive and negative samples into the risk prediction model for feature extraction, output multiple loan risk features, and re-label the users in the abnormal and normal sample sets based on the multiple loan risk features to obtain the target abnormal sample set and the target normal sample set.

[0094] Based on the description of the method and apparatus embodiments above, an exemplary embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method according to an embodiment of the present invention.

[0095] This application also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this application.

[0096] This application also provides a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this application.

[0097] See Figure 7 The diagram shown below illustrates the structure of an electronic device 700 that can serve as a server or client in this application, and is an example of a hardware device that can be applied to various aspects of this application. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0098] like Figure 7 As shown, the electronic device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. The RAM 703 may also store various programs and data required for the operation of the device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0099] Multiple components in electronic device 700 are connected to I / O interface 705, including: input unit 706, output unit 707, storage unit 708, and communication unit 709. Input unit 706 can be any type of device capable of inputting information to electronic device 700. Input unit 706 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 707 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 708 may include, but is not limited to, disk and optical disk. Communication unit 709 allows electronic device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers and / or chipsets, such as Bluetooth devices, WiFi devices, worldwide interoperability for microwave access (WiMax) devices, cellular communication devices, and / or the like.

[0100] The computing unit 701 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above. For example, in some embodiments, the outlier adjustment method described above can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 700 via ROM 702 and / or communication unit 709. In some embodiments, the computing unit 701 can be configured to perform the outlier adjustment method described above by any other suitable means (e.g., by means of firmware).

[0101] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0102] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM) or flash memory, optical fibers, compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0103] As used in this application, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device, PLD) used to provide machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal used to provide machine instructions and / or data to a programmable processor.

[0104] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0105] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0106] Computer systems may include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on respective computers and having a client-server relationship with each other. Obviously, those skilled in the art can make various modifications and variations to this application without departing from its spirit and scope. Therefore, this application is also intended to include such modifications and variations if they fall within the scope of the claims and their equivalents.

Claims

1. An outlier adjustment method characterized by, The method comprises: obtaining loan application data and a risk control policy document, wherein the loan application data comprises a plurality of user features corresponding to each user; extracting policy constraint conditions from the risk control policy document, classifying the user features in the loan application data based on the policy constraint conditions, and obtaining a first user feature subset and a second user feature subset; determining a first constraint sub-condition and a second constraint sub-condition in the policy constraint conditions, respectively determining outliers in the first user feature subset and the second user feature subset based on the first constraint sub-condition and the second constraint sub-condition, and obtaining at least one first abnormal sample and at least one second abnormal sample; based on the respective feature types of the plurality of user features, adjusting outliers in the at least one first abnormal sample and the at least one second abnormal sample, obtaining an abnormal sample set and an adjusted abnormal sample, and adding the adjusted abnormal sample to a normal sample set.

2. The method of claim 1, wherein, The method comprises: for each user feature in the loan application data, the following operations are performed: detecting whether the first user feature in the loan application data meets the policy constraint condition; wherein the first user feature is any one of the user features; if the first user feature meets the policy constraint condition, the first user feature is classified into the first user feature subset; if the first user feature does not meet the policy constraint condition, the first user feature is classified into the second user feature subset.

3. The method of claim 2, wherein, The method comprises: sequentially sorting a plurality of feature values included in a feature value set corresponding to the first user feature to obtain a first feature value corresponding to a first quartile and a second feature value corresponding to a third quartile; determining a difference value between the first feature value and the second feature value, and determining the difference value as a quartile range difference value; based on the first feature value and the quartile range difference value, calculating a first outlier threshold value, and based on the second feature value and the quartile range difference value, calculating a second outlier threshold value; wherein the first outlier threshold value is less than the second outlier threshold value; classifying feature values that do not belong to an outlier value range between the first outlier threshold value and the second outlier threshold value into the second user feature subset.

4. The method of claim 1, wherein, The method comprises: based on the feature type corresponding to each first abnormal sample, retaining an abnormal boundary value in each first abnormal sample to obtain the adjusted abnormal sample; based on the feature type corresponding to each second abnormal sample, performing one-sided tail shrinkage or double-sided tail shrinkage processing on the outliers in each second abnormal sample to obtain the adjusted abnormal sample.

5. The method of claim 1, wherein, The feature type includes a policy hard constraint type and a policy soft constraint type, wherein the adjustment frequency of the policy hard constraint type is lower than the adjustment frequency of the policy soft constraint type.

6. The method according to any one of claims 1 to 5, wherein Further comprising: receiving an adjustment request corresponding to the policy constraint condition; parsing a target policy constraint condition from the adjustment request, and performing a difference comparison between the policy constraint condition and the target policy constraint condition corresponding to the adjustment request to obtain an updated constraint condition; adjusting an abnormal value threshold based on the updated constraint condition to obtain a target abnormal value threshold.

7. The method according to any one of claims 1 to 5, wherein The abnormal value adjustment of the at least one abnormal sample based on the feature type corresponding to each of the plurality of user features, to obtain an abnormal sample set and an adjusted abnormal sample, further comprising: marking the abnormal sample set as a negative sample, and marking the normal sample set as a positive sample; inputting the positive sample and the negative sample into a risk prediction model for feature extraction, and outputting a plurality of loan risk features; relabeling the users in the abnormal sample set and the normal sample set based on the plurality of loan risk features to obtain a target abnormal sample set and a target normal sample set.

8. An outlier adjustment apparatus characterized by comprising: comprising: an acquisition module configured to acquire loan application data and a risk control policy document; wherein the loan application data comprises a plurality of user features corresponding to each user; a classification module configured to extract policy constraint conditions from the risk control policy document, and classify user features in the loan application data based on the policy constraint conditions to obtain a first user feature subset and a second user feature subset; a determination module configured to determine a first constraint sub-condition and a second constraint sub-condition in the policy constraint conditions, and perform abnormal value determination on the first user feature subset and the second user feature subset based on the first constraint sub-condition and the second constraint sub-condition, respectively, to obtain at least one first abnormal sample and at least one second abnormal sample; an adjustment module configured to perform abnormal value adjustment on the at least one first abnormal sample and the at least one second abnormal sample based on a feature type corresponding to each of the plurality of user features, to obtain an abnormal sample set and an adjusted abnormal sample, and add the adjusted abnormal sample to a normal sample set.

9. An electronic device, comprising: comprising: a memory configured to store a computer program; a processor configured to execute the computer program stored in the memory to implement the method steps of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer program is stored in the computer readable storage medium, and the computer program is executed by the processor to implement the method steps of any one of claims 1-7.

Images