Object attribute information processing method, device, computer equipment, and storage medium

Abstract

The present application relates to the field of big data technology, and in particular to an object attribute information processing method, device, computer equipment, storage medium and computer program product. The method comprises: based on the object attribute information of multiple sample objects, determining the sample feature sequence, default risk level, and sample default label corresponding to each sample object. For each sub-feature in the feature sequence, the feature risk probability corresponding to the corresponding sub-feature is determined based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label. Based on each feature risk probability, each sample feature sequence, and each sample default label, a correlation analysis is performed to obtain a target feature vector corresponding to each sample object, and each target feature vector is used to train a default risk prediction model. In this way, the accuracy of the default risk prediction model is greatly improved.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a method, apparatus, computer equipment, storage medium and computer program product for processing object attribute information. Background Art

[0002] With the development of big data technology, in order to predict the default risk of objects in the financial field, it is necessary to train the default risk prediction model. In the process of training the default risk prediction model, it is often necessary to quantify and standardize the risk-related information of the object, and use the processed information for the training of the default risk prediction model. The risk-related information can be the age, income, occupation and other category characteristics of the object.

[0003] In the related art, during the information processing process, each category feature is directly used for the training of the default risk prediction model. For example, the occupational feature of the object is directly used as a variable for model training. The occupational feature can only reflect the degree of influence of the position on the risk of the object.

[0004] However, for each category feature, the single-dimensional category feature is relatively simple in reflecting the risk performance of the object, and it is difficult to accurately reflect the performance of the object risk, and there is a problem of low accuracy in the prediction of the default risk prediction model. Summary of the invention

[0005] Based on this, it is necessary to provide an object attribute information processing method, apparatus, computer equipment, computer-readable storage medium and computer program product to address the above technical issues.

[0006] In a first aspect, the present application provides a method for processing object attribute information. The method comprises:

[0007] Obtain object attribute information of multiple sample objects;

[0008] Based on the object attribute information of the multiple sample objects, determine the sample feature sequence, default risk level, and sample default label corresponding to each sample object; wherein the sample feature sequence corresponds to attribute feature components corresponding to multiple attribute dimensions, and each attribute feature includes multiple sub-features;

[0009] For each sub-feature, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs;

[0010] Based on each characteristic risk probability, each sample characteristic sequence, and each sample default label, correlation analysis is performed to obtain a target characteristic vector corresponding to each sample object, and each of the target characteristic vectors is used to train a default risk prediction model.

[0011] In a second aspect, the present application also provides an object attribute information processing device. The device comprises:

[0012] An acquisition module, used to acquire object attribute information of multiple sample objects;

[0013] A first determination module is used to determine the sample feature sequence, default risk level, and sample default label corresponding to each sample object based on the object attribute information of the multiple sample objects; wherein the sample feature sequence corresponds to attribute feature components corresponding to multiple attribute dimensions, and each attribute feature includes multiple sub-features;

[0014] A second determination module is used to determine, for each sub-feature, a characteristic risk probability corresponding to the corresponding sub-feature based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs;

[0015] The analysis module is used to perform correlation analysis based on each feature risk probability, each sample feature sequence, and each sample default label to obtain a target feature vector corresponding to each sample object. Each of the target feature vectors is used to train a default risk prediction model.

[0016] In a third aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, any of the above-mentioned object attribute information processing methods is implemented.

[0017] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, any of the above-mentioned object attribute information processing methods is implemented.

[0018] In a fifth aspect, the present application further provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements any of the above-mentioned object attribute information processing methods.

[0019] The object attribute information processing method, device, computer equipment, storage medium and computer program product obtain object attribute information of multiple sample objects, and determine the sample feature sequence, default risk level and sample default label corresponding to each sample object based on the object attribute information of the multiple sample objects, wherein the sample feature sequence corresponds to the attribute feature composition corresponding to multiple attribute dimensions, and each attribute feature contains multiple sub-features. For each sub-feature, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs. In this way, based on the characteristic risk probability corresponding to each sub-feature, the default risk corresponding to each default risk level in each sub-feature can be accurately reflected. Based on each characteristic risk probability, each sample feature sequence, and each sample default label, correlation analysis is performed to obtain the target feature vector corresponding to each sample object, and each of the target feature vectors is used to train the default risk prediction model. Therefore, each target feature vector carries the characteristic risk probabilities corresponding to multiple default risk levels, that is, the target feature vector of a single dimension is expanded to a target feature vector carrying multiple default risk dimensions. In this way, training the default risk prediction model based on the target feature vectors of multiple default risk dimensions can greatly improve the prediction accuracy of the default risk prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is an application environment diagram of an object attribute information processing method in an embodiment;

[0021] Figure 2 is a flowchart of a method for processing object attribute information in one embodiment;

[0022] Figure 3 A schematic diagram of a code for obtaining a data set in one embodiment;

[0023] Figure 4 A schematic diagram of a step of determining a characteristic risk probability in one embodiment;

[0024] Figure 5 A schematic diagram of a flow chart of a step of determining a target feature vector in one embodiment;

[0025] Figure 6 A schematic diagram of a flow chart of a step of determining a first intermediate vector in one embodiment;

[0026] Figure 7 A schematic diagram of a flow chart of a step of determining a second intermediate vector in one embodiment;

[0027] Figure 8 is a flowchart of a method for processing object attribute information in another embodiment;

[0028] Fig. 9 is a structural block diagram of an object attribute information device in one embodiment;

[0029] Fig.10 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0031] The object attribute information processing method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers. The terminal 102 and the server 104 can be used alone to execute the object attribute information processing method, or they can be used together to execute the object attribute information processing method. Taking the server alone executing the object attribute information processing method as an example, the server 104 obtains the object attribute information of multiple sample objects, and based on the object attribute information of the multiple sample objects, determines the sample feature sequence, default risk level, and sample default label corresponding to each sample object. Among them, the sample feature sequence corresponds to the attribute feature composition corresponding to multiple attribute dimensions, and each attribute feature contains multiple sub-features. For each sub-feature, the server 104 determines the feature risk probability corresponding to the corresponding sub-feature based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs. The server 104 performs correlation analysis based on each characteristic risk probability, each sample characteristic sequence, and each sample default label to obtain a target characteristic vector corresponding to each sample object, and each of the target characteristic vectors is used to train the default risk prediction model. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, etc. The server 104 can be implemented as an independent server or a server cluster consisting of multiple servers.

[0032] In one embodiment, Figure 2 As shown, a method for processing object attribute information is provided, and the method is applied to a computer device as an example for explanation. The computer device may be Figure 1 The terminal or server in the embodiment of the present invention, wherein the object attribute information processing method comprises the following steps:

[0033] Step S202: Obtain object attribute information of multiple sample objects.

[0034] The sample object is a user holding a credit card, or a user in other fields. The object attribute information includes characteristic information of the sample object's own attributes, the term information of the sample object's account, and the default behavior information of the sample object. The characteristic information of the sample object's own attributes can be the user's age, occupation, credit card overdraft information, etc.

[0035] Specifically, the computer device extracts object attribute information of multiple sample objects from the historical sample data. For example, the computer device obtains the historical sample data and the identities of the multiple sample objects, and extracts the historical sample data through SQL (Structured Query Language) based on the identities of the multiple sample objects to obtain the object attribute information of the multiple sample objects. SQL is a programming language used to access data and query, update and manage relational database systems.

[0036] Step S204, based on the object attribute information of the multiple sample objects, determine the sample feature sequence, default risk level, and sample default label corresponding to each sample object; wherein the sample feature sequence corresponds to attribute feature components corresponding to multiple attribute dimensions, and each attribute feature contains multiple sub-features.

[0037] Among them, the object attribute information includes the characteristic information of the sample object's own attributes, the term information of the sample object's account, and the default behavior information of the sample object. The sample feature sequence is used to characterize the feature set of the sample object's own attributes, such as the age and occupation of the sample object, the default risk level is used to characterize the degree of the sample object's default risk behavior, and the sample default label is used to characterize the default result of the sample object. The default risk level is divided into five categories, namely "normal", "concern", "secondary", "doubtful", and "loss".

[0038] Specifically, the computer device determines the sample feature sequence corresponding to each sample object based on the object attribute information of multiple sample objects through the feature information of each object attribute information in the object attribute information, determines the default risk level corresponding to each sample object through the term information of the sample object account in the object attribute information, and determines the sample default label corresponding to each sample object through the default behavior information of the sample object in the object attribute information.

[0039] The sample feature sequence is composed of attribute features corresponding to multiple attribute dimensions, and each attribute feature contains multiple sub-features. The attribute feature can be occupation, age, income, and when the attribute feature is occupation, the sub-feature can be doctor, teacher, lawyer, etc.

[0040] Step S206, for each sub-feature, based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs, determine the feature risk probability corresponding to the corresponding sub-feature.

[0041] Among them, the sub-features are various categories belonging to the same attribute feature. For example, when the attribute feature is occupation, the sub-features are doctor, teacher, lawyer, etc.

[0042] Specifically, the computer device determines multiple sub-features in each attribute feature based on the acquired sample feature sequence. For each sub-feature, the computer device determines the feature risk probability corresponding to the corresponding sub-feature based on the default risk level of the sample object corresponding to the sub-feature and the sample default label to which the corresponding sample object belongs.

[0043] For example, based on 100 sample data in the data set, each sample object corresponds to a sample data, and the sample data contains sub-features, attribute features of sub-features, default risk level, and sample default label. The computer device obtains 100 sample feature sequences as X = (X1, ..., X t ), X t Represents attribute feature t, and for each sample object, this attribute feature is a determined sub-feature. If X1 is an attribute feature of an occupation, then X1 contains six sub-features in 100 sample data, namely 20 for doctors, 10 for teachers, 15 for teachers, 20 for lawyers, 5 for nurses, and 20 for architects. Among them, the sub-features of sample objects No. 1, ..., and No. 20 are doctors. Based on the sample data of sample objects No. 1 to No. 20, the default risk level and sample default label of the 20 sample objects are determined, and based on the default risk level and sample default label of sample objects No. 1 to No. 20, the characteristic risk probability of the doctor sub-feature is determined. It should be noted that each default risk level corresponds to each characteristic risk probability. For example, the default risk level of "normal" corresponds to a characteristic risk probability, and each of the five risk levels corresponds to a characteristic risk probability.

[0044] Step S208, based on each characteristic risk probability, each sample characteristic sequence, and each sample default label, correlation analysis is performed to obtain a target characteristic vector corresponding to each sample object, and each of the target characteristic vectors is used to train a default risk prediction model.

[0045] Among them, correlation analysis refers to analyzing two or more correlated variable elements to measure the degree of correlation between the two variable factors. The target feature vector is a collection of feature data associated with multiple default risk dimensions. The feature data can reflect the correlation between the default risk level and the default behavior, and can reflect the correlation between the attribute characteristics and the default behavior.

[0046] Specifically, the computer device determines the numerical type of each attribute feature based on each sample feature sequence, and obtains the classification results of the numerical attribute features and the non-numerical attribute features. For each classification result, a correlation analysis is performed based on each feature risk probability and each sample default label to obtain the correlation results corresponding to each classification result. Based on each correlation result, the default risk dimension in each attribute feature is determined. Based on the default risk dimension in each attribute feature, the computer device determines the target feature vector corresponding to each sample object. Among them, each of the target feature vectors is used to train the default risk prediction model.

[0047] For example, the existing sample feature sequence X=(X1,…,X5), which are occupation, age, gender, provident fund, and marital status, respectively. Among them, occupation, gender, and marital status are all non-numerical attribute features, and age and provident fund are all numerical attribute features. For each type of attribute feature, a correlation analysis is performed based on each feature risk probability and each sample default label to obtain the correlation results of each attribute feature. For example, for the attribute feature of occupation, the feature risk probability corresponding to the default risk level of "secondary" has a weak correlation with the sample default label, and the other four default risk levels have a strong correlation. The default risk dimensions of the occupation attribute feature are determined to be "normal", "concern", "suspicious", and "loss". For each sample object, the part of the target feature vector corresponding to the occupation attribute feature obtains the feature risk probability corresponding to the default risk dimensions of "normal", "concern", "suspicious", and "loss". For example, for sample object No. 1, the sub-feature is doctor. Among them, the doctor sub-feature has five feature risk probabilities, that is, the feature risk probabilities corresponding to the default risk dimensions of "normal", "concern", "secondary", "suspicious", and "loss" are A, B, C, D, and E respectively. When the target feature vector of sample object No. 1 is used for training the default risk prediction model, the part of the vector corresponding to the occupational attribute characteristics used for training is determined to be A, B, D, and E.

[0048] In the above object attribute information processing method, by obtaining the object attribute information of multiple sample objects, based on the object attribute information of the multiple sample objects, the sample feature sequence, the default risk level, and the sample default label corresponding to each sample object are determined, wherein the sample feature sequence corresponds to the attribute feature composition corresponding to multiple attribute dimensions, and each attribute feature contains multiple sub-features. For each sub-feature, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the default risk level to which the sample object corresponding to the corresponding sub-feature belongs and the sample default label to which the corresponding sample object belongs. In this way, based on the characteristic risk probability corresponding to each sub-feature, the default risk situation corresponding to each default risk level in each sub-feature can be accurately reflected. Based on each characteristic risk probability, each sample feature sequence, and each sample default label, correlation analysis is performed to obtain the target feature vector corresponding to each sample object, and each of the target feature vectors is used to train the default risk prediction model. Therefore, each target feature vector carries the characteristic risk probabilities corresponding to multiple default risk levels, that is, the target feature vector of a single dimension is expanded to a target feature vector carrying multiple default risk dimensions. In this way, training the default risk prediction model based on the target feature vectors of multiple default risk dimensions can greatly improve the prediction accuracy of the default risk prediction model.

[0049] In one embodiment, the method of determining the sample feature sequence, default risk level, and sample default label corresponding to each sample object based on the object attribute information of the multiple sample objects includes: extracting the object attribute information of each sample object to obtain the sample feature sequence, overdraft term, and default result in a predetermined time period corresponding to each sample object. Based on the overdraft term corresponding to each sample object, the default risk level corresponding to each sample object is determined. Based on the default result corresponding to each sample object in a predetermined time period, the sample default label corresponding to each sample object is determined, and the sample default label includes a white label that does not meet the default condition and a black label that meets the default condition.

[0050] Among them, the overdraft period is a time interval classification mark divided according to the length of the account overdraft, and the sample default labels are divided into black labels and white labels. The default risk level is divided into five categories, namely "normal", "concern", "substandard", "doubtful" and "loss".

[0051] Specifically, the computer device extracts the object attribute information of each sample object to obtain the sample feature sequence, overdraft period, and default result in a predetermined time period corresponding to each sample object. The computer device obtains the mapping relationship between the overdraft period and the default risk level, and determines the default risk level corresponding to each sample object based on the overdraft period and the mapping relationship corresponding to each sample object. The computer device determines whether the default result meets the default condition based on the default result corresponding to each sample object in the predetermined time period, and determines the sample default label corresponding to each sample object based on the judgment result. If the default condition is not met, the sample default label is a white label, and if the default condition is met, the sample default label is a black label.

[0052] Among them, the mapping relationship between the overdraft period and the default risk level is: if the overdraft period is less than 1 month, it is represented as Z 0-1 , then the default risk level is "normal". If the overdraft period is 1 to 2 months, it is characterized by Z 2-4 , the default risk level is "concern". If the overdraft period is 3 to 5 months, it is characterized as Z 5-7 , then the default risk level is "secondary". If the overdraft period is 6 to 8 months, it is characterized as Z 8-10 , the default risk level is "doubtful". If the overdraft period is 9 to 12 months, it is characterized as Z 11-12 , then the default risk level is "suspicious". The predetermined time period may be 1 year.

[0053] In this embodiment, the object attribute information of each sample object is extracted to obtain the sample feature sequence, overdraft term, and default result in a predetermined time period corresponding to each sample object. Based on the overdraft term corresponding to each sample object, the default risk level corresponding to each sample object is determined. In this way, the overdraft term of each sample object is converted into a default risk level, which can concretize the default risk situation of each sample object. Based on the default result of each sample object in a predetermined time period, the sample default label corresponding to each sample object is determined. In this way, based on data processing of the object attribute information of multiple sample objects, multiple attribute feature information related to each sample object, the default risk situation carrying the default risk situation information, and the sample default label carrying the actual default situation information in the predetermined time period are obtained, which can greatly improve the credibility of the target feature vector.

[0054] In one embodiment, if Figure 3 As shown, the computer device obtains historical sample data based on the table storing relevant data, and obtains the identity identifiers of multiple sample objects (corresponding to Figure 3The computer device extracts the object attribute information of each sample object through SQL (Structured Query Language) based on the identity identification of multiple sample objects, and obtains the sample feature sequence, overdraft term segment, and default result in a predetermined time period corresponding to each sample object. The computer device obtains the mapping relationship between the overdraft term segment and the default risk level, and determines the default risk level corresponding to each sample object based on the overdraft term and the mapping relationship corresponding to each sample object. The computer device determines whether the default result meets the default condition based on the default result corresponding to each sample object in the predetermined time period, and determines the sample default label corresponding to each sample object based on the judgment result. The computer device performs data cleaning on multiple sample feature sequences, default risk levels, and sample default labels, and obtains multiple sample feature sequences after cleaning, multiple default risk levels after cleaning, and multiple sample default labels after cleaning. Among them, the multiple sample feature sequences after cleaning, the multiple default risk levels after cleaning, and the multiple sample default labels after cleaning constitute a data set.

[0055] In this embodiment, based on the extraction of sample feature sequences of multiple sample objects from historical sample data, the corresponding default risk level is determined based on the overdraft period corresponding to each sample object, and the corresponding sample default label is determined based on the default result of the predetermined time period corresponding to each sample object. In this way, based on data processing of the object attribute information of multiple sample objects, multiple attribute feature information related to each sample object, the default risk situation carrying the default risk situation information, and the sample default label carrying the actual default situation information within the predetermined time period are obtained, which can greatly improve the credibility of the target feature vector. In addition, by performing data cleaning on the sample feature sequence, the default risk level, and the sample default label, the credibility of the target feature vector is further improved.

[0056] In one embodiment, the sample default label includes a white label that does not meet the default condition, and for each sub-feature, based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs, the characteristic risk probability corresponding to the corresponding sub-feature is determined, including: for each sub-feature, based on the default risk level of the sample object corresponding to the corresponding sub-feature, the number of levels belonging to the same default risk level in the corresponding sample object is determined. For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the number of white labels to which the sample object corresponding to the corresponding sub-feature belongs and the number of levels of the corresponding default risk level. Wherein, each characteristic risk probability corresponds to each default risk level one by one.

[0057] Specifically, for each sub-feature, the computer device determines the number of levels belonging to the same default risk level in the corresponding sample object based on the default risk level to which the sample object corresponding to the corresponding sub-feature belongs. For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the computer device determines the number of white labels to which the sample object corresponding to the corresponding sub-feature belongs. For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the computer device divides the number of white labels of the corresponding default risk level by the number of levels of the corresponding default risk level to obtain the characteristic risk probability corresponding to the corresponding sub-feature. Among them, each characteristic risk probability corresponds to each default risk level one by one.

[0058] For example, among 100 sample objects, there are six sub-features in the occupational attribute feature, namely doctor, teacher, teacher, lawyer, nurse, and architect. Among them, there are 20 doctor sub-features. Among the 20 sample data of doctor sub-features, there are 2 at the "normal" default risk level, 5 at the "concern" default risk level, 8 at the "sub-level" default risk level, 2 at the "suspicious" default risk level, and 3 at the "loss" default risk level. If the two sample objects at the "normal" default risk level are A and B, respectively. The computer device obtains the sample default labels of A and B, and determines the number of sample default labels that are white labels. If the sample default label of A is a white label and the sample default label of B is a black label, then the characteristic risk probability of the doctor sub-feature at the "normal" default risk level is 50%. Among them, the characteristic risk probabilities corresponding to its characteristic default risk levels are all calculated using the calculation method of the characteristic risk probabilities of the "normal" default risk level. In this way, the sub-features of each sample object can be expanded to carry multiple default risk dimensions, such as Figure 4 As shown, the sub-features of each sample object (i.e. corresponding Figure 4 The one-dimensional information, feature x) can obtain the information corresponding to the five default risk levels (i.e., corresponding to dimension information 1 (x1) to dimension information (x5) in the figure).

[0059] It should be noted that the number of characteristic risk probabilities of each sub-characteristic is the same as the number of default risk levels.

[0060] In this embodiment, for each sub-feature, the number of levels belonging to the same default risk level in the corresponding sample object is determined based on the default risk level to which the sample object corresponding to the corresponding sub-feature belongs. For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the number of white labels to which the sample object corresponding to the corresponding sub-feature belongs and the number of levels of the corresponding default risk level. Moreover, each characteristic risk probability corresponds to each default risk level one by one. In this way, based on the characteristic risk probability corresponding to each sub-feature, the default risk situation corresponding to each default risk level in each sub-feature can be accurately reflected.

[0061] In one embodiment, Figure 5 As shown, the characteristic risk probability corresponding to the sub-feature includes the characteristic risk probability corresponding to each default risk level. Based on each characteristic risk probability, each sample feature sequence, and each sample default label, correlation analysis is performed to obtain a target feature vector corresponding to each sample object, including:

[0062] Step S502: for each sample object, determine a plurality of feature risk probabilities of each sub-feature of the corresponding sample object at a plurality of default risk levels.

[0063] Specifically, for each sample object, each sub-feature of each sample object is determined, and based on each sub-feature of each sample object and the characteristic risk probabilities corresponding to each default risk level, multiple characteristic risk probabilities of each sub-feature of the corresponding sample object under multiple default risk levels are determined.

[0064] Step S504, if the attribute feature is of numerical type, then for each numerical attribute feature, based on multiple feature risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute value of the numerical attribute feature, and the sample default label of each sample object, determine the first matrix to be processed corresponding to the corresponding numerical attribute feature.

[0065] Specifically, the computer device determines multiple attribute features based on each sample feature sequence. The computer device determines the numerical type of each attribute feature. If the attribute feature dimension is numerical, then for each numerical attribute feature, the characteristic risk probability corresponding to the numerical attribute feature in each sample object, the attribute value of the numerical attribute feature in each sample object, and the sample default label of each sample object are obtained. For each numerical attribute feature, based on the multiple characteristic risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute values ​​of the numerical attribute features, and the sample default labels of each sample object, the first matrix to be processed corresponding to the corresponding numerical attribute feature is determined. The rows of each of the first matrices to be processed are determined by multiple sample objects, and the columns are determined by the columns corresponding to multiple default risk levels, the columns where the corresponding numerical attribute features are located, and the columns where the multiple sample default labels are located.

[0066] For example, based on 100 sample data in a data set, that is, the sample data of the data set corresponds to the sample objects one by one, that is, sample objects No. 1 to sample objects No. 100. Taking age as an example, the characteristic risk probability (P a11 , P a12 , P a13 , P a14 , P a15 ), the characteristic risk probability corresponding to the age sub-characteristic 2 (21 to 40 years old) (P a21 , P a22 , P a23 , P a24 , P a25 ), the characteristic risk probability corresponding to the age sub-characteristic 3 (41 years old value 60 years old) (P a31 , P a32 , P a33 , P a34 , P a35 ), the characteristic risk probability corresponding to the age sub-characteristic (greater than 61 years old) (P a41 , P a42 , P a43 , P a44 , P a45). In the first matrix to be processed corresponding to age, the first to fifth columns represent the characteristic risk probabilities corresponding to each default risk level, the sixth column is the age value of each sample object, and the seventh column is the sample default label of each sample object. For example, for the first row where sample object No. 1 (age is 30 years old, i.e., the corresponding sub-feature 2) is located, the first column of the first row to the fifth column of the first row represent the characteristic risk probabilities corresponding to the five default risk levels under the condition of age sub-feature 2. The sixth column of the first row is the age value of sample object No. 1, i.e., 30, and the seventh column of the first row is the sample default label of sample object No. 1. For the second row where sample object No. 2 (age is 20 years old, i.e., the corresponding sub-feature 1) is located, the first column of the second row to the fifth column of the second row represent the characteristic risk probabilities corresponding to the five default risk levels under the condition of age sub-feature 1. The sixth column of the second row is the age value of sample object No. 2, i.e., 20, and the seventh column of the second row is the sample default label of sample object No. 2.

[0067] Step S506: For each first matrix to be processed, covariance calculation is performed on each first matrix to be processed, and the corresponding first matrix to be processed is screened to determine the first intermediate vector corresponding to each numerical attribute feature.

[0068] Specifically, the computer device obtains the first matrix to be processed corresponding to each numerical attribute feature. For each first matrix to be processed, each feature risk probability in the corresponding first matrix to be processed is used as the first variable of the first matrix to be processed, and the attribute value of the numerical attribute feature in the corresponding first matrix to be processed is used as the second variable of the matrix to be processed, and the sample default label of each sample object is used as the third variable of the first matrix to be processed. For each first matrix to be processed, the computer device calculates the covariance of each first variable of the corresponding first matrix to be processed with the third variable, and screens each first variable in the corresponding first matrix to be processed to obtain the screened first variable corresponding to each first matrix to be processed. For each matrix to be processed, the computer device calculates the covariance of the second variable of the corresponding first matrix to be processed with the third variable, and screens the second variable in the corresponding first matrix to be processed to obtain the screening result of the second variable corresponding to each first matrix to be processed. The computer device determines the first intermediate vector corresponding to each numerical attribute feature based on the screened first variable corresponding to each first matrix to be processed and the screening result of the second variable corresponding to each first matrix to be processed. Among them, each numerical attribute feature corresponds to a first matrix to be processed.

[0069] For example, based on 100 sample data in a data set, that is, the sample data of the data set corresponds to the sample object one by one. Among them, from sample object No. 1 to sample object No. 100, each numerical attribute feature corresponds to a first to-be-processed matrix. Taking the numerical attribute feature as age as an example, the first to-be-processed matrix of age is defined as the first to-be-processed matrix corresponding to age. In the first to-be-processed matrix of age, the number of rows is 100 rows, the first to the fifth columns represent the characteristic risk probability corresponding to each default risk level, the sixth column is the age value of each sample object, and the seventh column is the sample default label of each sample object. Among them, in the first to the fifth columns, the default risk level corresponding to each column is used as the first variable in the first to-be-processed matrix of age, that is, the corresponding matrix contains 5 first variables, the sixth column is used as the second variable in the first to-be-processed matrix of age, and the seventh column is used as the third variable in the first to-be-processed matrix of age. For the first to-be-processed matrix of age, the computer device calculates the covariance of each first variable with the third variable to obtain the first variable corresponding to the first to-be-processed matrix of age after screening. The computer device calculates the covariance of the second variable and the third variable to obtain a screening result of the second variable corresponding to the first matrix to be processed of age. The computer device determines a first intermediate vector corresponding to the numerical attribute feature of age based on the screening result of the first variable corresponding to the first matrix to be processed of age and the second variable corresponding to the first matrix to be processed of age.

[0070] Step S508, if the attribute feature is non-numeric, then for each non-numeric attribute feature, based on multiple feature risk probabilities corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object, determine the second matrix to be processed corresponding to the corresponding non-numeric attribute feature.

[0071] Specifically, the computer device determines multiple attribute features based on each sample feature sequence. The computer device determines the numerical type of each attribute feature. If the attribute feature dimension is non-numeric, then for each non-numeric attribute feature, the characteristic risk probability corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object are obtained. For each non-numeric attribute feature, based on the multiple characteristic risk probabilities corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object, the second matrix to be processed corresponding to the corresponding non-numeric attribute feature is determined. The rows of each second matrix to be processed are determined by multiple sample objects, and the columns are determined by the columns corresponding to the multiple default risk levels and the columns where the multiple sample default labels are located.

[0072] For example, based on 100 sample data in the data set, that is, the sample data of the data set corresponds to the sample objects one by one. Among them, from sample object No. 1 to sample object No. 100. Taking the non-numerical attribute feature as occupation as an example, the risk feature probability corresponding to the doctor sub-feature 1 is (P b11 , P b12 , P b13 , P b14 , P b15 ), the risk characteristic probability corresponding to teacher sub-characteristic 2 is (P b21 , P b22 , P b23 , P b24 , P b25 ) etc. In the first matrix to be processed corresponding to the occupation, the first to fifth columns represent the characteristic risk probabilities corresponding to each default risk level, and the sixth column is the sample default label of each sample object. For example, for the first row where sample object No. 1 (occupation is doctor, i.e. corresponding sub-feature 1), the first column of the first row to the fifth column of the first row represent the characteristic risk probabilities corresponding to the five default risk levels under the doctor sub-feature 1. The sixth column of the first row is the sample default label of sample object No. 1.

[0073] Step S510: For each second matrix to be processed, covariance calculation is performed on each second matrix to be processed, and the corresponding second matrix to be processed is screened to determine the second intermediate vector corresponding to each non-numerical attribute feature.

[0074] Specifically, the computer device obtains the second matrix to be processed corresponding to each non-numerical attribute feature. For each second matrix to be processed, each characteristic risk probability in the corresponding second matrix to be processed is used as the first variable of the second matrix to be processed, and the sample default label of each sample object is used as the third variable of the second matrix to be processed. For each second matrix to be processed, the computer device calculates the covariance of each first variable of the corresponding second matrix to be processed and the third variable, and screens each first variable in the corresponding second matrix to be processed to obtain the screened first variable corresponding to each second matrix to be processed. The computer device uses the screened first variable corresponding to each second matrix to be processed as the second intermediate vector corresponding to each non-numerical attribute feature. Among them, each non-numerical attribute feature corresponds to a second matrix to be processed.

[0075] For example, based on 100 sample data in a data set, that is, the sample data of the data set corresponds to the sample object one by one. From sample object No. 1 to sample object No. 100, each non-numerical attribute feature corresponds to a second to-be-processed matrix. Taking the non-numerical attribute feature as occupation as an example, the occupation second to-be-processed matrix is ​​defined as the second to-be-processed matrix corresponding to the occupation. In the occupation second to-be-processed matrix, the number of rows is 100 rows, and the 1st to 5th columns represent the characteristic risk probability corresponding to each default risk level, and the 6th column is the sample default label of each sample object. Among them, in the 1st to 5th columns, the default risk level corresponding to each column is used as the first variable in the occupation second to-be-processed matrix, that is, the corresponding matrix contains 5 first variables, and the 6th column is used as the third variable in the occupation second to-be-processed matrix. For the occupation second to-be-processed matrix, the computer device calculates the covariance of each first variable with the third variable to obtain the first variable corresponding to the occupation second to-be-processed matrix. The computer device uses the first variable corresponding to the occupation second to-be-processed matrix as the second intermediate vector corresponding to the occupation non-numerical attribute feature. Each non-numerical attribute feature corresponds to a second matrix to be processed.

[0076] Step S512: Combine the first intermediate vector and the second intermediate vector to obtain target feature vectors corresponding to each sample object.

[0077] Specifically, the computer device obtains the first intermediate vector corresponding to each numerical attribute feature and the second intermediate variable corresponding to each non-numerical attribute feature, and combines the first intermediate vector and the second intermediate vector to obtain the target feature vector corresponding to each sample object.

[0078] In this embodiment, for each sample object, multiple characteristic risk probabilities of each sub-feature of the corresponding sample object under multiple default risk levels are determined. If the attribute feature is a numerical type, then for each numerical attribute feature, based on the multiple characteristic risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute value of the numerical attribute feature, and the sample default label of each sample object, the first matrix to be processed corresponding to the corresponding numerical attribute feature is determined. For each first matrix to be processed, the covariance calculation is performed on each first matrix to be processed, the corresponding first matrix to be processed is screened, and the first intermediate vector corresponding to each numerical attribute feature is determined. In this way, based on the first intermediate vectors corresponding to each numerical attribute feature, multiple characteristic risk probabilities in each numerical attribute feature of each sample object can be screened to obtain characteristic risk probabilities with strong correlation. If the attribute feature is a non-numerical type, then for each non-numerical attribute feature, based on the multiple characteristic risk probabilities corresponding to the non-numerical attribute feature in each sample object, and the sample default label of each sample object, the second matrix to be processed corresponding to the corresponding non-numerical attribute feature is determined. For each second matrix to be processed, covariance calculation is performed on each second matrix to be processed, and the corresponding second matrix to be processed is screened to determine the second intermediate vector corresponding to each non-numerical attribute feature. In this way, based on the second intermediate vectors corresponding to each non-numerical feature, multiple feature risk probabilities in each non-numerical attribute feature of each sample object can be screened to obtain feature risk probabilities with strong correlation. The first intermediate vector and the second intermediate vector are combined to obtain target feature vectors corresponding to each sample object. In this way, by distinguishing the attribute features into numerical and non-numerical types, the corresponding correlation analysis can be performed in combination with the characteristics of the specific type of data, which greatly improves the correctness of the correlation analysis results. In addition, by combining the correlation analysis of different types of attribute features, the credibility of the target feature vector is greatly increased.

[0079] In one embodiment, Figure 6 As shown, for each first matrix to be processed, covariance calculation is performed on each first matrix to be processed, and the corresponding first matrix to be processed is screened to determine the first intermediate vector corresponding to each numerical attribute feature, including:

[0080] Step S602: for each first processed matrix corresponding to a numerical attribute feature, based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding first processed matrix, determine the first variance corresponding to each default risk level in the corresponding first processed matrix.

[0081] Specifically, for each first matrix to be processed corresponding to a numerical attribute feature, the computer device uses each feature risk probability in the corresponding first matrix to be processed as a first variable of the first matrix to be processed, wherein each first variable in each first matrix to be processed corresponds to each default risk level one by one.

[0082] For each first matrix to be processed, a plurality of first variables corresponding to the same default risk level are averaged, and based on the result of the average calculation, the first variance corresponding to each default risk level in the corresponding first matrix to be processed is determined. In the first matrix to be processed, the number of first variables with the same default risk level is consistent with the number of sample objects.

[0083] For example, based on 100 sample data in the data set, taking age as an example of a numerical attribute feature, the age first processing matrix is ​​defined as the first processing matrix corresponding to age. In columns 1 to 5, the default risk level corresponding to each column is used as the first variable in the age first processing matrix, that is, the corresponding matrix contains 5 first variables. Among them, for the "normal" default risk level in the age first processing matrix, there are 100 first variables corresponding to the "normal" default risk level. The computer device calculates the mean based on the 100 first variables of the "normal" default risk level, and determines the first variance of the "normal" default risk level in the age first processing matrix based on the mean result.

[0084] Step S604: for each first matrix to be processed corresponding to the numerical attribute feature, based on the attribute values ​​of the plurality of sample objects in the first matrix to be processed, determine the second variance corresponding to the attribute values ​​in the first matrix to be processed.

[0085] Specifically, for each first matrix to be processed corresponding to a numerical attribute feature, the computer device uses the attribute value of the numerical attribute feature in the corresponding matrix to be processed as the second variable of the first matrix to be processed. The number of second variables of each first matrix to be processed is consistent with the number of sample objects. For each first matrix to be processed, the mean of the multiple second variables is calculated, and based on the result of the mean calculation, the second variance corresponding to the attribute value in the corresponding first matrix to be processed is determined.

[0086] For example, based on 100 sample data in the data set, taking age as an example of a numerical attribute feature, the age first processing matrix is ​​defined as the first processing matrix corresponding to age, and the sixth column is the age value of each sample object, and the sixth column is used as the second variable in the age first processing matrix. For the age first processing matrix, the computer device will perform mean calculation based on the 100 second variables, and determine the second variance corresponding to the age attribute value in the age first processing matrix based on the mean result.

[0087] Step S606: determining a third party difference corresponding to the default label based on the sample default labels corresponding to the plurality of sample objects.

[0088] Specifically, the computer device obtains sample default labels corresponding to multiple sample objects, wherein the white label that does not meet the default condition is set to a value of 0, and the black label that meets the default condition is set to a value of 1. The computer device calculates the mean value based on the values ​​of the sample default labels of the multiple sample objects, and determines the third difference corresponding to the default label based on the mean result.

[0089] Step S608: For each default risk level, perform covariance calculation on the corresponding first variance and the third variance to obtain a first covariance result corresponding to the corresponding default risk level, and perform covariance calculation based on the second variance and the third variance to obtain a second covariance result.

[0090] Specifically, the computer device obtains the first variance corresponding to each first matrix to be processed. For each default risk level of each first matrix to be processed, the computer device performs covariance calculation on the corresponding first variance and the third variance to obtain the first covariance result corresponding to the corresponding default risk level in the corresponding first matrix to be processed. The computer device obtains the second variance of each first matrix to be processed, and for each first matrix to be processed, the computer device performs covariance calculation on the second variance of the corresponding first matrix to be processed and the third variance to obtain the second covariance result.

[0091] Step S610, based on the multiple first covariance results, multiple default risk levels in each numerical attribute feature are screened to determine a first intermediate default risk level, and based on the second covariance results, multiple numerical attribute features in each numerical attribute feature are screened to obtain a first intermediate numerical attribute feature.

[0092] Specifically, the computer device obtains multiple first covariance results corresponding to each of the first matrices to be processed. For each first matrix to be processed, each first covariance result in the corresponding first matrix to be processed is compared with a threshold value. If the first covariance result is greater than or equal to the threshold value, the screening condition is met. The computer device uses the first covariance of the corresponding first matrix to be processed that meets the screening condition as the target first covariance. The computer device uses the multiple default risk levels corresponding to the multiple target first covariances in the corresponding first matrix to be processed as the first intermediate default risk level of the corresponding first matrix to be processed. For each first matrix to be processed, the second covariance result in the corresponding first matrix to be processed is compared with the threshold value. If the second covariance result is greater than or equal to the threshold value, the screening condition is met. If the second covariance result in the corresponding first matrix to be processed meets the screening condition, the numerical attribute feature in the corresponding first matrix to be processed is used as the first intermediate numerical attribute feature in the corresponding first matrix to be processed. Among them, the threshold is used to characterize the strength of the correlation. If the screening condition is met, it is determined that the correlation is strong.

[0093] Step S612: Based on the first intermediate default risk level and the first intermediate numerical attribute feature, determine a first intermediate vector corresponding to the numerical attribute feature in each sample object.

[0094] Specifically, for each first matrix to be processed, the computer device obtains each first intermediate default risk level and each first intermediate numerical attribute feature in the corresponding first matrix to be processed, and combines each first intermediate default risk level and each first intermediate numerical attribute feature in the corresponding first matrix to be processed to determine the first intermediate vector corresponding to the numerical attribute feature in each sample object.

[0095] For example, taking the age first processing matrix as an example, if the first intermediate default risk level in the age first processing matrix is ​​"normal" and "concern", and the first intermediate numerical attribute feature is age. Then the first intermediate vector corresponding to the age attribute feature in each sample object is (x1, x2, x3). Among them, x1 is the probability of the "normal" default risk level corresponding to the age sub-feature of the sample object, x2 is the probability of the "concern" default risk level corresponding to the age sub-feature of the sample object, and x3 is the numerical value of the age of the sample object.

[0096] In this embodiment, for each first matrix to be processed corresponding to a numerical attribute feature, the first variance corresponding to each default risk level in the corresponding first matrix to be processed is determined based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding first matrix to be processed. For each first matrix to be processed corresponding to a numerical attribute feature, the second variance corresponding to the attribute value in the corresponding first matrix to be processed is determined based on the attribute values ​​of multiple sample objects in the corresponding first matrix to be processed. Based on the sample default labels corresponding to the multiple sample objects, the third variance corresponding to the default label is determined. For each default risk level, the corresponding first variance and the third variance are covariance calculated to obtain the first covariance result corresponding to the corresponding default risk level, and the covariance calculation is performed based on the second variance and the third variance to obtain the second covariance result. Based on the multiple first covariance results, multiple default risk levels in each numerical attribute feature are screened to determine the first intermediate default risk level, and based on the second covariance result, multiple numerical attribute features in each numerical attribute feature are screened to obtain the first intermediate numerical attribute feature. Based on the first intermediate default risk level and the first intermediate numerical attribute feature, a first intermediate vector corresponding to the numerical attribute feature in each sample object is determined. In this way, based on the first intermediate vectors corresponding to each numerical attribute feature, multiple characteristic risk probabilities in each numerical attribute feature of each sample object can be screened to obtain characteristic risk probabilities with strong correlation.

[0097] In one embodiment, Figure 7 As shown, for each second matrix to be processed, covariance calculation is performed on each second matrix to be processed, and the corresponding second matrix to be processed is screened to determine the second intermediate vector corresponding to each non-numerical attribute feature, including:

[0098] Step S702: for each second matrix to be processed corresponding to the non-numerical attribute feature, based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding second matrix to be processed, determine the fourth variance corresponding to each default risk level in the corresponding second matrix to be processed.

[0099] Specifically, for each second matrix to be processed corresponding to a non-numerical attribute feature, the computer device uses each feature risk probability in the corresponding second matrix to be processed as the first variable of the second matrix to be processed. Wherein, each first variable in each second matrix to be processed corresponds to each default risk level one by one. For each second matrix to be processed, the mean of multiple first variables corresponding to the same default risk level is calculated, and based on the result of the mean calculation, the fourth variance corresponding to each default risk level in the corresponding second matrix to be processed is determined. Wherein, in the second matrix to be processed, the number of first variables with the same default risk level is consistent with the number of sample objects.

[0100] For example, based on 100 sample data in the data set, taking occupation as an example of non-numerical attribute feature, the occupation second to-be-processed matrix is ​​defined as the second to-be-processed matrix corresponding to the occupation. In columns 1 to 5, the default risk level corresponding to each column is used as the first variable in the occupation second to-be-processed matrix, that is, the corresponding matrix contains 5 first variables. Among them, for the "normal" default risk level in the occupation second to-be-processed matrix, there are 100 first variables corresponding to the "normal" default risk level in the occupation second matrix. The computer device calculates the mean based on the 100 first variables of the "normal" default risk level in the occupation second matrix, and determines the fourth variance of the "normal" default risk level in the occupation second to-be-processed matrix based on the mean result.

[0101] Step S704: determining a fifth variance corresponding to the default label based on the sample default labels corresponding to the plurality of sample objects.

[0102] Specifically, the computer device obtains sample default labels corresponding to multiple sample objects, wherein a white label that does not meet the default condition is set to a value of 0, and a black label that meets the default condition is set to a value of 1. The computer device calculates the mean based on the values ​​of the sample default labels of the multiple sample objects, and determines the fifth variance corresponding to the default label based on the mean result.

[0103] Step S706: For each default risk level, the corresponding fourth variance and the fifth variance are subjected to covariance calculation to obtain a third covariance result corresponding to the corresponding default risk level.

[0104] Specifically, the computer device obtains the fourth variance corresponding to each second matrix to be processed. For each default risk level in each second matrix to be processed, the computer device performs covariance calculation on the corresponding fourth variance and fifth variance to obtain the third covariance result corresponding to the corresponding default risk level in the corresponding second matrix to be processed.

[0105] Step S708: screening multiple default risk levels in each non-numerical attribute feature based on multiple third covariance results to obtain a second intermediate default risk level.

[0106] Specifically, the computer device obtains multiple third covariance results corresponding to each second matrix to be processed. For each second matrix to be processed, each third covariance result in the corresponding second matrix to be processed is compared with the threshold value. If the third covariance result is greater than or equal to the threshold value, the screening condition is met. The computer device uses the third covariance of the corresponding second matrix to be processed that meets the screening condition as the target third covariance. The computer device uses the multiple default risk levels corresponding to the multiple target third covariances in the corresponding second matrix to be processed as the second intermediate default risk level in the corresponding second matrix to be processed.

[0107] Step S710: Based on the second intermediate default risk level, determine a second intermediate vector corresponding to each sample object whose attribute feature is non-numeric.

[0108] Specifically, for each second matrix to be processed, the computer device obtains each second intermediate default risk level in the corresponding second matrix to be processed, and uses each second intermediate default level in the corresponding second matrix to be processed as the second intermediate vector corresponding to the non-numeric attribute feature in each sample object.

[0109] For example, taking the second occupational matrix to be processed as an example, if the second default risk level in the occupational matrix to be processed is "secondary" or "suspicious", then the first intermediate vector corresponding to the occupational attribute characteristics in each sample object is (x1, x2). Among them, x1 is the probability of the "secondary" default risk level corresponding to the occupational sub-characteristic of the sample object, and x2 is the probability of the "suspicious" default risk level corresponding to the age sub-characteristic of the sample object.

[0110] In this embodiment, for the second matrix to be processed corresponding to each non-numerical attribute feature, the fourth variance corresponding to each default risk level in the corresponding second matrix to be processed is determined based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding second matrix to be processed. Based on the sample default labels corresponding to the multiple sample objects, the fifth variance corresponding to the default label is determined. For each default risk level, the corresponding fourth variance and the fifth variance are covariance calculated to obtain the third covariance result corresponding to the corresponding default risk level. Based on the multiple third covariance results, multiple default risk levels in each non-numerical attribute feature are screened to obtain the second intermediate default risk level. Based on the second intermediate default risk level, the second intermediate vector corresponding to the non-numerical attribute feature in each sample object is determined. In this way, based on the second intermediate vectors corresponding to each non-numerical feature, multiple characteristic risk probabilities in each non-numerical attribute feature of each sample object can be screened to obtain a characteristic risk probability with strong correlation.

[0111] In order to facilitate a clearer understanding of the technical solution of the present application, a more detailed description is provided. Figure 8 As shown, first, prepare the data set D. Specifically, the computer device obtains historical sample data based on the table storing relevant data, and obtains the identity identifiers of multiple sample objects (corresponding to Figure 3 The computer device obtains the mapping relationship between the overdraft period and the default risk level, and determines the default risk level corresponding to each sample object based on the overdraft period and the mapping relationship corresponding to each sample object. The computer device determines whether the default result meets the default condition based on the default result corresponding to each sample object in the predetermined time period, and determines the sample default label corresponding to each sample object based on the judgment result. The computer device performs data cleaning on the multiple sample feature sequences, the default risk level, and the sample default label, and obtains the cleaned multiple sample feature sequences, the cleaned multiple default risk levels, and the cleaned multiple sample default labels. The cleaned multiple sample feature sequences, the cleaned multiple default risk levels, and the cleaned multiple sample default labels constitute the data set D. The data cleaning is to determine whether the data corresponding to the attribute feature is empty and whether the value of the data in numerical form is abnormal. For example, if the information of the marital status attribute feature of sample object 1 is empty, then the marital status of sample object 1 can be determined and filled based on empirical estimates of other historical data in sample object 1. For data format, whether the data is abnormal, for example, for the provident fund attribute feature, the mean and standard deviation are calculated based on the provident funds of multiple sample objects, and the cleaning range of the provident fund attribute feature is determined based on the mean and standard deviation (for example, the cleaning range is higher than the difference between the mean and 3 times the standard deviation, and lower than the sum of the mean and 3 times the standard deviation). Based on this cleaning range, the provident fund values ​​in multiple sample objects are cleaned to remove abnormal data.

[0112] Then, based on the determined data set D, the computer device performs feature multi-dimensional information processing. Specifically, for each sub-feature, the computer device determines the number of levels belonging to the same default risk level in the corresponding sample object based on the default risk level to which the sample object corresponding to the corresponding sub-feature belongs. For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the computer device determines the number of white labels to which the sample object corresponding to the corresponding sub-feature belongs. For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the computer device divides the number of white labels of the corresponding default risk level by the number of levels of the corresponding default risk level to obtain the feature risk probability corresponding to the corresponding sub-feature. Among them, each feature risk probability corresponds to each default risk level one by one.

[0113] Then, the correlation analysis between each default risk level and the sample default label is performed. Specifically, for each sample object, based on each sub-feature of each sample object and the feature risk probability corresponding to each default risk level, multiple feature risk probabilities of each sub-feature of the corresponding sample object under multiple default risk levels are determined, and the numerical type of each attribute feature is judged.

[0114] If the attribute feature dimension is numerical, then for each numerical attribute feature, obtain the characteristic risk probability corresponding to the numerical attribute feature in each sample object, the attribute value of the numerical attribute feature in each sample object, and the sample default label of each sample object. For each numerical attribute feature, based on the multiple characteristic risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute values ​​of the numerical attribute feature, and the sample default labels of each sample object, determine the first matrix to be processed corresponding to the corresponding numerical attribute feature. The rows of each first matrix to be processed are determined by multiple sample objects, and the columns are determined by the columns corresponding to multiple default risk levels, the columns where the corresponding numerical attribute features are located, and the columns where the multiple sample default labels are located. For each first matrix to be processed corresponding to the numerical attribute feature, the computer device uses each characteristic risk probability in the corresponding first matrix to be processed as the first variable of the first matrix to be processed. Among them, each first variable in each first matrix to be processed corresponds to each default risk level one by one. For each first matrix to be processed, the multiple first variables corresponding to the same default risk level are averaged, and based on the result of the average calculation, the first variance corresponding to each default risk level in the corresponding first matrix to be processed is determined. Wherein, in the first matrix to be processed, the number of first variables of the same default risk level is consistent with the number of sample objects. For each first matrix to be processed corresponding to the numerical attribute feature, the computer device uses the attribute value of the numerical attribute feature in the corresponding matrix to be processed as the second variable of the first matrix to be processed. Wherein, the number of second variables of each first matrix to be processed is consistent with the number of sample objects. For each first matrix to be processed, the mean of multiple second variables is calculated, and based on the result of the mean calculation, the second variance corresponding to the attribute value in the corresponding first matrix to be processed is determined. The computer device sets the white label that does not meet the default condition to the value 0, and sets the black label that meets the default condition to the value 1, and performs mean calculation based on the values ​​of the sample default labels of multiple sample objects, and determines the third variance corresponding to the default label based on the mean result. For each default risk level of each first matrix to be processed, the computer device performs covariance calculation on the corresponding first variance and the third variance, and obtains the first covariance result corresponding to the corresponding default risk level in the corresponding first matrix to be processed. For each first matrix to be processed, the computer device performs covariance calculation on the second variance of the corresponding first matrix to be processed and the third variance, and obtains the second covariance result. For each first matrix to be processed, the computer device compares each first covariance result in the corresponding first matrix to be processed with the threshold value, and if the first covariance result is greater than or equal to the threshold value, the screening condition is met. The computer device uses the first covariance of the corresponding first matrix to be processed that meets the screening condition as the target first covariance.The computer device uses the multiple default risk levels corresponding to the multiple target first covariances in the corresponding first matrix to be processed as the first intermediate default risk level of the corresponding first matrix to be processed. For each first matrix to be processed, the second covariance result in the corresponding first matrix to be processed is compared with the threshold value. If the second covariance result is greater than or equal to the threshold value, the screening condition is met. If the second covariance result in the corresponding first matrix to be processed meets the screening condition, the numerical attribute feature in the corresponding first matrix to be processed is used as the first intermediate numerical attribute feature in the corresponding first matrix to be processed. Among them, the threshold is used to characterize the strength of the correlation. If the screening condition is met, it is determined that the correlation is strong. For each first matrix to be processed, the computer device combines each first intermediate default risk level and the first intermediate numerical attribute feature in the corresponding first matrix to be processed, and determines the first intermediate vector corresponding to the numerical attribute feature in each sample object.

[0115] If the attribute feature dimension is non-numeric, then for each non-numeric attribute feature, determine the characteristic risk probability corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object. For each non-numeric attribute feature, determine the second matrix to be processed corresponding to the corresponding non-numeric attribute feature based on the multiple characteristic risk probabilities corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object. The rows of each second matrix to be processed are determined by multiple sample objects, and the columns are determined by the columns corresponding to the multiple default risk levels and the columns where the multiple sample default labels are located. For each second matrix to be processed corresponding to the non-numeric attribute feature, the computer device uses each characteristic risk probability in the corresponding second matrix to be processed as the first variable of the second matrix to be processed. Wherein, each first variable in each second matrix to be processed corresponds to each default risk level one by one. For each second matrix to be processed, the mean of multiple first variables corresponding to the same default risk level is calculated, and based on the result of the mean calculation, the fourth variance corresponding to each default risk level in the corresponding second matrix to be processed is determined. Wherein, in the second matrix to be processed, the number of first variables with the same default risk level is consistent with the number of sample objects. The computer device sets the white label that does not meet the default condition to a value of 0, sets the black label that meets the default condition to a value of 1, and performs mean calculation based on the values ​​of the sample default labels of multiple sample objects, and determines the fifth variance corresponding to the default label based on the mean result. For each default risk level in each second matrix to be processed, the computer device performs covariance calculation on the corresponding fourth variance and fifth variance to obtain the third covariance result corresponding to the corresponding default risk level in the corresponding second matrix to be processed. For each second matrix to be processed, the computer device compares each third covariance result in the corresponding second matrix to be processed with the threshold value. If the third covariance result is greater than or equal to the threshold value, the screening condition is met. The computer device uses the third covariance of the corresponding second matrix to be processed that meets the screening condition as the target third covariance. The computer device uses the multiple default risk levels corresponding to the multiple target third covariances in the corresponding second matrix to be processed as the second intermediate default risk level in the corresponding second matrix to be processed. For each second matrix to be processed, the computer device uses each second intermediate default level in the corresponding second matrix to be processed as the second intermediate vector corresponding to the non-numeric attribute feature in each sample object. The computer device obtains the first intermediate vector corresponding to each numerical attribute feature and the second intermediate variable corresponding to each non-numerical attribute feature. The computer device combines the first intermediate vector and the second intermediate vector to obtain the target feature vector corresponding to each sample object.

[0116] Finally, the target vector is reduced in dimension through the PCA (Principal Component Analysis) dimensionality reduction technique to obtain the standardized target feature vector.

[0117] In addition, in order to further verify the validity of the target feature vector, the Random Forest Classifier algorithm is used to classify the credit quality (i.e., the assessment of default risk) of credit card users (sample objects). Under the condition that other conditions remain unchanged, the scheme of this application and the scheme of this application are adopted, and the corresponding relevant data of the samples are processed, and the precision and recall rates of the two during verification and testing are statistically analyzed. As shown in Table 1 and Table 2 below:

[0118] Table 1 Validation set - precision and recall

[0119]

[0120] Table 2 Test set - precision and recall

[0121]

[0122] As shown in Tables 1 and 2 above, after the default risk prediction model is trained using the solution of the present application, both the precision and recall of the default risk prediction model are improved.

[0123] In this embodiment, based on the object attribute information of the multiple sample objects, the sample feature sequence, the default risk level, and the sample default label corresponding to each sample object are determined, wherein the sample feature sequence corresponds to the attribute feature composition corresponding to multiple attribute dimensions, and each attribute feature includes multiple sub-features. For each sub-feature, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the default risk level to which the sample object corresponding to the corresponding sub-feature belongs and the sample default label to which the corresponding sample object belongs. In this way, based on the characteristic risk probability corresponding to each sub-feature, the default risk situation corresponding to each default risk level in each sub-feature can be accurately reflected. Based on each characteristic risk probability, each sample feature sequence, and each sample default label, correlation analysis is performed to obtain the target feature vector corresponding to each sample object, and each of the target feature vectors is used to train the default risk prediction model. Therefore, each target feature vector carries the characteristic risk probabilities corresponding to multiple default risk levels, that is, the target feature vector of a single dimension is expanded to a target feature vector carrying multiple default risk dimensions. In this way, training the default risk prediction model based on the target feature vectors of multiple default risk dimensions can greatly improve the accuracy of the default risk prediction model prediction. In addition, reducing the dimension of the target feature vector through PCA technology can avoid the problem of multicollinearity, and the target feature vector after dimension reduction (i.e., standardization) can further improve the accuracy of the default risk prediction model.

[0124] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0125] Based on the same inventive concept, the embodiment of the present application also provides an object attribute information processing device for implementing the object attribute information processing method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the one or more object attribute information processing device embodiments provided below can refer to the limitations of the object attribute information processing method above, and will not be repeated here.

[0126] In one embodiment, Fig. 9 As shown, an object attribute information processing device is provided, the device 900 includes: an acquisition module 902, a first determination module 904, a second determination module 906 and an analysis module 908, wherein:

[0127] The acquisition module 902 is used to acquire object attribute information of multiple sample objects.

[0128] The first determination module 904 is used to determine the sample feature sequence, default risk level, and sample default label corresponding to each sample object based on the object attribute information of the multiple sample objects; wherein the sample feature sequence corresponds to attribute feature components corresponding to multiple attribute dimensions, and each attribute feature contains multiple sub-features.

[0129] The second determination module 906 is used to determine, for each sub-feature, the feature risk probability corresponding to the corresponding sub-feature based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs.

[0130] The analysis module 908 is used to perform correlation analysis based on each feature risk probability, each sample feature sequence, and each sample default label to obtain a target feature vector corresponding to each sample object. Each of the target feature vectors is used to train the default risk prediction model.

[0131] In one embodiment, the first determination module 904 is used to extract the object attribute information of each sample object to obtain the sample feature sequence, overdraft term period, and default result in a predetermined time period corresponding to each sample object. Based on the overdraft term period corresponding to each sample object, the default risk level corresponding to each sample object is determined. Based on the default result corresponding to each sample object in a predetermined time period, the sample default label corresponding to each sample object is determined, and the sample default label includes a white label that does not meet the default condition and a black label that meets the default condition.

[0132] In one embodiment, the second determination module 906 is used to determine, for each sub-feature, the number of levels of the corresponding sample objects belonging to the same default risk level based on the default risk level to which the sample objects corresponding to the corresponding sub-feature belong. For each default risk level to which the sample objects corresponding to the corresponding sub-feature belong, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the number of white labels to which the sample objects corresponding to the corresponding sub-feature belong and the number of levels of the corresponding default risk level. Each characteristic risk probability corresponds to each default risk level one by one.

[0133] In one embodiment, the analysis module 908 is used to determine, for each sample object, multiple characteristic risk probabilities of each sub-feature of the corresponding sample object under multiple default risk levels. If the attribute feature is a numerical type, then for each numerical attribute feature, based on the multiple characteristic risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute value of the numerical attribute feature, and the sample default label of each sample object, determine the first matrix to be processed corresponding to the corresponding numerical attribute feature. For each first matrix to be processed, perform covariance calculation on each first matrix to be processed, screen the corresponding first matrix to be processed, and determine the first intermediate vector corresponding to each numerical attribute feature. If the attribute feature is a non-numerical type, then for each non-numerical attribute feature, based on the multiple characteristic risk probabilities corresponding to the non-numerical attribute feature in each sample object, and the sample default label of each sample object, determine the second matrix to be processed corresponding to the corresponding non-numerical attribute feature. For each second matrix to be processed, perform covariance calculation on each second matrix to be processed, screen the corresponding second matrix to be processed, and determine the second intermediate vector corresponding to each non-numerical attribute feature. The first intermediate vector and the second intermediate vector are combined to obtain target feature vectors corresponding to each sample object.

[0134] In one embodiment, the analysis module 908 is used to determine the first variance corresponding to each default risk level in the corresponding first matrix to be processed for each numerical attribute feature, based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding first matrix to be processed. For each first matrix to be processed corresponding to the numerical attribute feature, the second variance corresponding to the attribute value in the corresponding first matrix to be processed is determined based on the attribute values ​​of multiple sample objects in the corresponding first matrix to be processed. Based on the sample default labels corresponding to the multiple sample objects, the third variance corresponding to the default label is determined. For each default risk level, the corresponding first variance and the third variance are covariance calculated to obtain the first covariance result corresponding to the corresponding default risk level, and the covariance is calculated based on the second variance and the third variance to obtain the second covariance result. Based on the multiple first covariance results, multiple default risk levels in each numerical attribute feature are screened to determine the first intermediate default risk level, and based on the second covariance result, multiple numerical attribute features in each numerical attribute feature are screened to obtain the first intermediate numerical attribute feature. Based on the first intermediate default risk level and the first intermediate numerical attribute feature, a first intermediate vector corresponding to the numerical attribute feature in each sample object is determined.

[0135] In one embodiment, the analysis module 908 is used to determine the fourth variance corresponding to each default risk level in the corresponding second matrix to be processed for each non-numeric attribute feature, based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding second matrix to be processed. Based on the sample default labels corresponding to the multiple sample objects, the corresponding fifth variance of the default label is determined. For each default risk level, the corresponding fourth variance and the fifth variance are covariance calculated to obtain the third covariance result corresponding to the corresponding default risk level. Based on the multiple third covariance results, multiple default risk levels in each non-numeric attribute feature are screened to obtain a second intermediate default risk level. Based on the second intermediate default risk level, the second intermediate vector corresponding to the non-numeric attribute feature in each sample object is determined.

[0136] Each module in the above-mentioned object attribute information processing device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module above.

[0137] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store object attribute information processing data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an object attribute information processing method is implemented.

[0138] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0139] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0140] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0141] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0142] 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.

[0143] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0144] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0145] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for processing object attribute information, characterized in that: The method comprises: Obtain object attribute information of multiple sample objects; Based on the object attribute information of the multiple sample objects, determine the sample feature sequence, default risk level, and sample default label corresponding to each sample object; wherein the sample feature sequence is composed of attribute features corresponding to multiple attribute dimensions, and each attribute feature includes multiple sub-features, and the sample default label includes a white label that does not meet the default condition; For each sub-feature, the feature risk probability corresponding to the corresponding sub-feature is determined based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs, including: For each sub-feature, based on the default risk level of the sample object corresponding to the corresponding sub-feature, the number of levels of the corresponding sample objects belonging to the same default risk level is determined; For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the number of white labels to which the sample object corresponding to the corresponding sub-feature belongs and the number of levels of the corresponding default risk level; wherein each characteristic risk probability corresponds to each default risk level one by one, and the characteristic risk probability corresponding to the sub-feature includes the characteristic risk probability corresponding to each default risk level respectively; Based on each characteristic risk probability, each sample characteristic sequence, and each sample default label, correlation analysis is performed to obtain a target characteristic vector corresponding to each sample object, each of which is used to train a default risk prediction model, including: For each sample object, determining multiple feature risk probabilities of each sub-feature of the corresponding sample object under multiple default risk levels; If the attribute feature is of numerical type, then for each numerical attribute feature, based on the multiple feature risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute values ​​of the numerical attribute feature, and the sample default labels of each sample object, determine the first matrix to be processed corresponding to the corresponding numerical attribute feature; for each first matrix to be processed, perform covariance calculation on each first matrix to be processed, screen the corresponding first matrix to be processed, and determine the first intermediate vector corresponding to each numerical attribute feature; If the attribute feature is non-numeric, then for each non-numeric attribute feature, based on multiple feature risk probabilities corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object, determine the second matrix to be processed corresponding to the corresponding non-numeric attribute feature; for each second matrix to be processed, perform covariance calculation on each second matrix to be processed, screen the corresponding second matrix to be processed, and determine the second intermediate vector corresponding to each non-numeric attribute feature; The first intermediate vector and the second intermediate vector are combined to obtain target feature vectors corresponding to each sample object.

2. The method according to claim 1, characterized in that The determining, based on the object attribute information of the plurality of sample objects, a sample feature sequence, a default risk level, and a sample default label corresponding to each sample object includes: Extracting object attribute information of each sample object to obtain the sample feature sequence, overdraft period, and default result in a predetermined time period corresponding to each sample object; Determining the default risk level corresponding to each sample object based on the overdraft period corresponding to each sample object; Based on the default results of each sample object in a predetermined time period, a sample default label corresponding to each sample object is determined, and the sample default label includes a white label that does not meet the default condition and a black label that meets the default condition.

3. The method according to claim 1, characterized in that For each first matrix to be processed, covariance calculation is performed on each first matrix to be processed, and the corresponding first matrix to be processed is screened to determine the first intermediate vector corresponding to each numerical attribute feature, including: For each first matrix to be processed corresponding to each numerical attribute feature, based on the characteristic risk probability of the same default risk level of multiple sample objects in the corresponding first matrix to be processed, determine the first variance corresponding to each default risk level in the corresponding first matrix to be processed; For each first matrix to be processed corresponding to the numerical attribute feature, based on the attribute values ​​of the plurality of sample objects in the corresponding first matrix to be processed, determine the second variance corresponding to the attribute values ​​in the corresponding first matrix to be processed; Based on the sample default labels corresponding to the multiple sample objects, determining a third party difference corresponding to the default label; For each default risk level, performing covariance calculation on the corresponding first variance and the third variance to obtain a first covariance result corresponding to the corresponding default risk level, and performing covariance calculation based on the second variance and the third variance to obtain a second covariance result; Based on the multiple first covariance results, multiple default risk levels in each numerical attribute feature are screened to determine a first intermediate default risk level, and based on the second covariance results, multiple numerical attribute features in each numerical attribute feature are screened to obtain a first intermediate numerical attribute feature; Based on the first intermediate default risk level and the first intermediate numerical attribute feature, a first intermediate vector corresponding to the numerical attribute feature in each sample object is determined.

4. The method according to claim 1, characterized in that: For each second matrix to be processed, covariance calculation is performed on each second matrix to be processed, and the corresponding second matrix to be processed is screened to determine the second intermediate vector corresponding to each non-numerical attribute feature, including: For each second matrix to be processed corresponding to the non-numerical attribute feature, based on the characteristic risk probabilities of the same default risk level of multiple sample objects in the corresponding second matrix to be processed, determine the fourth variance corresponding to each default risk level in the corresponding second matrix to be processed; Determining a fifth variance corresponding to the default label based on the sample default labels corresponding to the plurality of sample objects; For each default risk level, the corresponding fourth variance and the fifth variance are subjected to covariance calculation to obtain a third covariance result corresponding to the corresponding default risk level; Based on multiple third covariance results, multiple default risk levels in each non-numerical attribute feature are screened to obtain a second intermediate default risk level; Based on the second intermediate default risk level, a second intermediate vector corresponding to each sample object having a non-numeric attribute feature is determined.

5. An object attribute information processing device, characterized in that: The device comprises: An acquisition module, used to acquire object attribute information of multiple sample objects; A first determination module is used to determine, based on the object attribute information of the multiple sample objects, a sample feature sequence, a default risk level, and a sample default label corresponding to each sample object; wherein the sample feature sequence is composed of attribute features corresponding to multiple attribute dimensions, and each attribute feature includes multiple sub-features, and the sample default label includes a white label that does not meet the default condition; The second determination module is used to determine, for each sub-feature, the feature risk probability corresponding to the corresponding sub-feature based on the default risk level of the sample object corresponding to the corresponding sub-feature and the sample default label to which the corresponding sample object belongs, including: For each sub-feature, based on the default risk level of the sample object corresponding to the corresponding sub-feature, the number of levels of the corresponding sample objects belonging to the same default risk level is determined; For each default risk level to which the sample object corresponding to the corresponding sub-feature belongs, the characteristic risk probability corresponding to the corresponding sub-feature is determined based on the number of white labels to which the sample object corresponding to the corresponding sub-feature belongs and the number of levels of the corresponding default risk level; wherein each characteristic risk probability corresponds to each default risk level one by one, and the characteristic risk probability corresponding to the sub-feature includes the characteristic risk probability corresponding to each default risk level respectively; The analysis module is used to perform correlation analysis based on each characteristic risk probability, each sample characteristic sequence, and each sample default label to obtain a target characteristic vector corresponding to each sample object, each of which is used to train a default risk prediction model, including: For each sample object, determining multiple feature risk probabilities of each sub-feature of the corresponding sample object under multiple default risk levels; If the attribute feature is of numerical type, then for each numerical attribute feature, based on the multiple feature risk probabilities corresponding to the numerical attribute feature in each sample object, the attribute values ​​of the numerical attribute feature, and the sample default labels of each sample object, determine the first matrix to be processed corresponding to the corresponding numerical attribute feature; for each first matrix to be processed, perform covariance calculation on each first matrix to be processed, screen the corresponding first matrix to be processed, and determine the first intermediate vector corresponding to each numerical attribute feature; If the attribute feature is non-numeric, then for each non-numeric attribute feature, based on multiple feature risk probabilities corresponding to the non-numeric attribute feature in each sample object and the sample default label of each sample object, determine the second matrix to be processed corresponding to the corresponding non-numeric attribute feature; for each second matrix to be processed, perform covariance calculation on each second matrix to be processed, screen the corresponding second matrix to be processed, and determine the second intermediate vector corresponding to each non-numeric attribute feature; The first intermediate vector and the second intermediate vector are combined to obtain target feature vectors corresponding to each sample object.

6. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images