E-commerce risk control method and device based on data missing values and storage medium

Abstract

The application relates to the field of intelligent risk control, and discloses an e-commerce risk control method and device based on data missing values and a storage medium. The method comprises the following steps: receiving e-commerce feature data, calculating a missing rate of the e-commerce feature data to generate a missing rate; based on parameterized likelihood ratio test, judging whether the missing rate corresponding to the target feature is completely random missing; when it is not completely random missing, judging whether the missing rate corresponding to the target feature is random missing according to logistic regression analysis; when it is not random missing, performing category screening processing on the missing rate corresponding to the target feature according to a screening algorithm to generate a screening result, and performing strategy division processing on the combination of the feature category corresponding to the target feature and the screening result according to a strategy rule to generate a risk control result. In the embodiment of the application, the potential risks in the missing values are identified, and the identification rate of potential risk behaviors is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent risk control, and in particular to an e-commerce risk control method, device and storage medium based on missing data values. Background Technology

[0002] Currently, missing value handling technologies in the cross-border e-commerce risk control field are mainly based on general processing logic. The mainstream approach is to directly delete features with high missing rates and use the mean or mode to imput all missing values. Some solutions simply distinguish between continuous and discrete variables but do not customize strategies based on business scenarios.

[0003] The lack of analysis on the risks associated with missing values ​​and the uniform processing of these values ​​result in the neglect of implicit business risk signals. For example, the absence of non-random occupational or income characteristics may itself represent high risk, and direct imputation will lose this risk information. Missing values ​​themselves can also constitute risk control factors, but current e-commerce risk control lacks the ability to analyze the reasons for missing values ​​to address this issue. A new technology is needed to solve this problem. Summary of the Invention

[0004] The main objective of this invention is to solve the technical problem of the lack of analysis of the causes of missing values ​​in existing e-commerce risk control systems.

[0005] The first aspect of this invention provides an e-commerce risk control method based on missing data values, comprising the following steps: Receive e-commerce feature data, calculate the missing rate of the e-commerce feature data, generate multiple missing rates corresponding to feature data, and filter out target features based on the missing rates corresponding to multiple feature data. Based on a preset parameterized likelihood ratio test, it is determined whether the missing rate corresponding to the target feature is completely random missing. When the feature is completely randomly missing, the target feature is divided into two categories according to the preset strategy rules, and the combination of the target feature and the completely randomly missing feature is processed to generate a risk control result. When the missing values ​​are not completely random, the missing rate corresponding to the target feature is determined to be random based on the preset logistic regression analysis. When there is random missing information, the target feature is divided into two categories according to the preset strategy rules, and the combination of the target feature and the random missing information is processed to generate a risk control result. When the missing data is not random, the missing rate corresponding to the target feature is filtered according to a preset filtering algorithm to generate a filtering result. The combination of the feature type corresponding to the target feature and the filtering result is divided according to a preset strategy rule to generate a risk control result. The filtering result includes: non-random missing data.

[0006] Optionally, in a first implementation of the first aspect of the present invention, the step of calculating the missing rate of the e-commerce feature data and generating multiple feature data corresponding to missing rates includes: ; Among them, R miss j Let x be the missing rate of feature j, N be the total number of samples with feature j, and II() be an exponential function. ij Let be the sample with the i-th feature j.

[0007] Optionally, in a second implementation of the first aspect of the present invention, the step of determining whether the missing rate corresponding to the target feature is completely random missing based on a preset parameterized likelihood ratio test includes: Initial parameters are generated based on the sample mean and sample covariance corresponding to the target features as initial estimates. Based on the initial parameters, calculate the expected value of the target feature; With the goal of maximizing the expected value, the initial parameters are updated until a preset stopping rule is met, and the final parameters are generated. Based on the final parameters, substitute them into the preset chi-square formula and the preset degree of freedom formula to calculate the chi-square value and the degree of freedom value; Based on the chi-square value and the degree of freedom value, determine whether the missing rate corresponding to the target feature is completely random missing.

[0008] Optionally, in a third implementation of the first aspect of the present invention, the chi-square formula includes: ; Among them, X 2 Here, n is the chi-square value, G is the number of groups with different missing patterns, and n is the number of groups with different missing patterns. g Let y be the sample size of the g-th group. g - Let u be the sample mean vector of the g-th group. g ^ The mean term of the final parameter in the g-th group. Let g be the covariance term of the final parameter in the g-th group.

[0009] Optionally, in a fourth implementation of the first aspect of the present invention, the degree-of-freedom formula includes: ; Where df is the degrees of freedom, G is the number of groups with different deletion patterns, and p is the total number of research variables. g is the number of observed variables in the g-th missing pattern.

[0010] Optionally, in the fifth implementation of the first aspect of the present invention, the step of determining whether the missing rate corresponding to the target feature is random missing based on a preset logistic regression analysis includes the following steps: Create indicator variables with missing values ​​based on target features; Based on a preset Lasso logistic regression, the indicator variable is subjected to missing factor detection processing to generate detection results; Based on the detection results, it is determined whether the missing rate corresponding to the target feature is random missing.

[0011] Optionally, in a sixth implementation of the first aspect of the present invention, the step of performing category filtering processing on the missing rate corresponding to the target feature according to a preset filtering algorithm to generate filtering results includes the following steps: Create indicator variables with missing values ​​based on target features; Based on the preset information evaluation formula and preset risk control labels, the information value of the indicator variables is calculated to generate information value; Based on the preset chi-square test algorithm, the correlation between the indicator variable and the preset risk control label is calculated, and a chi-square value is generated; Based on the chi-square value and the information value, a screening result is generated.

[0012] Optionally, in a seventh implementation of the first aspect of the present invention, generating the screening result based on the chi-square value and the information value includes the following steps: When the chi-square value is less than a preset first threshold and the information value is greater than a preset second threshold, the missing rate corresponding to the target feature is confirmed to be non-random missing. When the chi-square value is not less than a preset first threshold or the information value is not greater than a preset second threshold, the missing rate corresponding to the target feature is confirmed to be completely random missing.

[0013] A second aspect of the present invention provides an e-commerce risk control device based on missing data values, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the e-commerce risk control device based on missing data values ​​to execute the aforementioned e-commerce risk control method based on missing data values.

[0014] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described e-commerce risk control method based on missing data values.

[0015] In this embodiment of the invention, by calculating the missing rate of e-commerce feature data and utilizing the requirements corresponding to the missing rates of different features, target features requiring risk control analysis are identified. Then, parametric likelihood ratio tests, logistic regression analysis, and information value screening are performed on the missing rates of the target features to classify the causes of missing values ​​into three categories: completely random missing, random missing, and non-random missing. Based on the risk control strategy, the business attributes of the target features and the causes of missing values ​​are combined to classify the risk and generate risk control results. This solution can accurately identify target features with high risk, realize the identification of potential risks in missing values, significantly improve the identification rate of potential risky behaviors, and solve the technical problem in existing e-commerce risk control that lacks analysis of the causes of missing values ​​for risk control. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an embodiment of the e-commerce risk control method based on missing data values ​​in this invention. Figure 2 This is a schematic diagram of a specific embodiment of the 102 steps of the e-commerce risk control method based on missing data values ​​in this invention. Figure 3 This is a schematic diagram of a specific embodiment of the 103 steps of the e-commerce risk control method based on missing data values ​​in this invention. Figure 4 This is a schematic diagram of a specific embodiment of the 104 steps of the e-commerce risk control method based on missing data values ​​in this invention. Figure 5 This is a schematic diagram of an embodiment of an e-commerce risk control device based on missing data values ​​in this invention. Detailed Implementation

[0017] This invention provides an e-commerce risk control method, device, and storage medium based on missing data values.

[0018] The embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0019] In the description of the embodiments disclosed in this invention, the term "comprising" and similar terms should be understood as open-ended inclusion, i.e., "including but not limited to". The term "based on" should be understood as "at least partially based on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0020] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 An embodiment of the e-commerce risk control method based on missing data values ​​in this invention includes the following steps: 101. Receive e-commerce feature data, calculate the missing rate of the e-commerce feature data, generate multiple missing rates corresponding to feature data, and filter out target features based on the missing rates corresponding to multiple feature data. In this embodiment, e-commerce feature data is received, and based on the core logic of cross-border e-commerce risk control, multi-dimensional structured indicators are collected from the perspective of a three-dimensional verification system encompassing e-commerce flow, capital flow, and information flow, including individual dimensions (income, education, etc.) and store dimensions (operating data). Subsequently, data preprocessing is performed to remove invalid samples with inconsistent formats, missing key identifiers in fields, or business logic contradictions such as a severe mismatch between income and business scale. Valid feature variables and observation samples that conform to the cross-border e-commerce risk control business rules are then selected, providing a high-quality data foundation for subsequent missing value mechanism testing.

[0021] The missing rate is calculated for each dimension of the e-commerce feature data, generating a corresponding missing rate for each feature. Missing rate thresholds are set for different feature dimensions; for example, a missing rate >10% for income features and >30% for occupation features are considered high missing rates. This quickly filters out low-value, redundant features with low missing rates, avoiding the waste of computational resources on invalid data and selecting target features relevant to the analytical scenario.

[0022] Specifically, step 101, "calculating the missing rate of the e-commerce feature data and generating multiple missing rates corresponding to the feature data," includes the following specific implementation methods: ; Among them, R miss j Let x be the missing rate of feature j, N be the total number of samples with feature j, and II() be an exponential function. ij Let be the sample with the i-th feature j.

[0023] It should be noted that the II() indicator function is set when x ij When x equals null, the value of II() is 1; when x... ij The value of II() is 0 when it is not equal to null.

[0024] 102. Based on the preset parameterized likelihood ratio test, determine whether the missing rate corresponding to the target feature is completely random missing; In this embodiment, the causes of missing features are classified based on business logic, and the missing mechanisms are determined to be three types: completely random missing, random missing, and non-random missing. These three types of missing features are defined as follows: 1. Completely random missing values ​​are defined as P(M|X) obs ,X miss ) = P(M), where X obs X is the observed value. miss This represents the actual missing data situation, and the probability of missing data is independent of any variable.

[0025] 2. Random missing data is defined as P(M|X) obs ,X miss )=P(M|X obs The probability of missing values ​​depends only on other observed variables.

[0026] 3. Non-random missing data is defined as P(M|X) obs ,X miss )=P(M|X miss The probability of a missing variable depends on the missing variable itself.

[0027] First, a statistical determination is made based on a pre-defined parametric likelihood ratio test. The core assumption of this test is that the analyzed variables follow a multivariate normal distribution. The difference in the mean vectors of different missing pattern groups is used to test whether the data meets the definition of complete randomness. The null hypothesis H0 is that the data is completely random, that is, the probability of missing data is independent of any variable. The alternative hypothesis H1 is that the data does not meet the definition of complete randomness, that is, it belongs to random missing data or non-random missing data.

[0028] The core of this test is to construct a quadratic statistic to quantify the Mahalanobis distance between the sample mean of each missing pattern group and the expected mean of H0. Specifically, by comparing the difference between the expected mean vector estimated under the H0 hypothesis (where all data originate from the same population) and the observed sample mean vector under the H1 hypothesis (which allows different missing pattern groups to have different mean vectors), the test ultimately determines whether the missing rate corresponding to the target feature is completely random.

[0029] For details, please refer to Figure 2 , Figure 2 This is a schematic diagram of a specific embodiment of step 102 of the e-commerce risk control method based on missing data values ​​in this invention. Step 102 includes the following specific implementation methods: 1021. Initial parameters are generated based on the sample mean and sample covariance corresponding to the target features as initial estimates; 1022. Calculate the expected value of the target feature based on the initial parameters; 1023. With the goal of maximizing the expected value, update the initial parameters until the preset stopping rule is met, and generate the final parameters; 1024. Based on the final parameters, substitute them into the preset chi-square formula and the preset degree of freedom formula to calculate the chi-square value and the degree of freedom value; 1025. Based on the chi-square value and the degree of freedom value, determine whether the missing rate corresponding to the target feature is completely random missing.

[0030] In steps 1021-1025, under the assumption of H0, the population mean vector µˆ and covariance matrix are estimated using the expectation-maximization algorithm. A strict convergence criterion is set for the iterative process to balance accuracy and efficiency. First, the sample mean and sample covariance corresponding to the target feature are initialized as initial estimates, i.e., θ(0) = (µˆ(0), Σˆ(0)), where θ(0) are the initial parameters, µˆ(0) are the sample mean, and Σˆ(0) are the sample covariance. Using the sample mean and covariance calculated based on observable data as initial estimates provides good numerical stability and helps improve the algorithm's convergence efficiency.

[0031] Calculate the expectation E of the missing data under the observed data conditions based on the current parameter estimate, and then maximize the log-likelihood function of the complete data under this expectation to update the parameter θ. (t+1) The iteration is terminated and the final parameters are output when any of the following conditions are met: (1) The relative change in the mean vector or covariance matrix is ​​<10 -5 ; (2) The relative change in the log-likelihood function is <10 -5 ; (3) The number of iterations reaches 5000, which is the maximum number of iterations; (4) The log-likelihood no longer increases monotonically.

[0032] Substitute the final parameters into the chi-square formula and the preset degrees of freedom formula to calculate the chi-square value and the degrees of freedom value.

[0033] By checking the chi-square value and degrees of freedom in the distribution table and determining whether the missing rate of the target feature is completely random, the final determination can be made.

[0034] Specifically, the chi-square formula includes: ; Among them, X 2 Here, n is the chi-square value, G is the number of groups with different missing patterns, and n is the number of groups with different missing patterns. g Let y be the sample size of the g-th group. g- Let u be the sample mean vector of the g-th group. g ^ The mean term of the final parameter in the g-th group. Let g be the covariance term of the final parameter in the g-th group.

[0035] It should be noted that u g ^ and These are the subvectors and subcovariance matrices of the final parameters obtained through iterative estimation under the H0 assumption, respectively, on this set of observed variables.

[0036] Specifically, the formula for the degrees of freedom includes: ; Where df is the degrees of freedom, G is the number of groups with different deletion patterns, and p is the total number of research variables. g is the number of observed variables in the g-th missing pattern.

[0037] It should be noted that the formula for degrees of freedom requires that the sample size of the g-th group be n. g >p g To ensure that Σˆg is positive definite and invertible. This test requires a total sample size n ≥ 50 to ensure the effectiveness of the chi-square distribution approximation.

[0038] The significance level α for the p-value in the chi-square test is set to 0.05 (a standard statistical threshold). For special scenarios with high reliability requirements, it can be adjusted to 0.01 or 0.10 according to actual needs. If the p-value obtained from the chi-square test is greater than α, there is insufficient statistical evidence to reject the null hypothesis H0, and the missing data mechanism satisfies completely random missing data. This type of missing data is caused by random factors and has minimal interference with subsequent analysis. Simple methods such as mean imputation, median imputation, or direct deletion can be used to handle it. Conversely, if the p-value is ≤ α, the null hypothesis H0 is rejected, the data does not satisfy completely random missing data, and further analysis proceeds to step 104.

[0039] 103. When the feature is completely randomly missing, the target feature is divided into two categories according to the preset strategy rules, and the combination of the target feature and the completely randomly missing feature is processed to generate a risk control result. In this embodiment, the strategy rules will apply risk control hit strategies to the feature types corresponding to the target features and completely random missing features, and generate relevant risk control results. The strategy rules can be set according to needs, which will not be elaborated here, but can be referred to step 106.

[0040] 104. When the missing data is not completely random, the missing rate corresponding to the target feature is determined to be random based on the preset logistic regression analysis. In this embodiment, logistic regression analysis is used to examine whether the missing data can be explained by other observed variables, providing a statistical basis for determining random missing data. The logistic regression analysis employs Lasso logistic regression to perform model fit and significance tests: if the overall model test (likelihood ratio test / chi-square test) result is p < 0.05, or at least one independent variable coefficient is significantly different from 0, it indicates that the probability of missing target features can be effectively predicted by other observed variables. This result provides strong statistical evidence that the missing data mechanism is random, and subsequent modeling methods such as multiple imputation and regression imputation can be used for missing data processing.

[0041] For details, please refer to Figure 3 , Figure 3 This is a schematic diagram of a specific embodiment of the 104 steps of the e-commerce risk control method based on missing data values ​​in this invention. The 104 steps include the following specific implementation methods: 1041. Create indicator variables with binary missing values ​​based on target features; 1042. Based on the preset Lasso logistic regression, perform missing factor detection processing on the indicator variable to generate detection results; 1043. Based on the detection results, determine whether the missing rate corresponding to the target feature is random missing.

[0042] In steps 1041-1043, a binary missing indicator variable Mx is created for the missing feature X: ; Subsequently, using Mx as the dependent variable, and other fully observed variables in the dataset (missing rate of 0) or those with missing rates below a very low threshold and for which initial imputation has been completed, to prevent overfitting due to high-dimensional features, the model employs L1 regularization (Lasso logistic regression) for automatic feature selection, retaining only predictors that have significant explanatory power for the missing state Mx, and constructing a regression model:

[0043] Where β0, β1, ..., β k Let Z1, Z2, ..., Zk be the coefficients of the independent variable, and M be the independent variable. x It is the dependent variable.

[0044] Finally, the model is tested for fit and significance: if the overall model test (likelihood ratio test / chi-square test) result is p<0.05, or at least one independent variable coefficient is significantly different from 0, it indicates that the probability of missing target features can be effectively predicted by other observed variables. This result provides strong statistical evidence that the missing mechanism is random, and subsequent modeling methods such as multiple imputation and regression imputation can be used for missing feature processing.

[0045] 105. When there is random missing information, the target feature is divided into two categories according to the preset strategy rules, and the combination of the target feature and the random missing information is processed to generate a risk control result. In this embodiment, the strategy rules will apply risk control hit strategies to the feature types and random missing features corresponding to the target features, and generate relevant risk control results. The strategy rules can be set according to requirements, which will not be elaborated here, but can be referred to step 106.

[0046] 106. When the missing data is not random, the missing rate corresponding to the target feature is filtered according to a preset filtering algorithm to generate a filtering result. The combination of the feature type corresponding to the target feature and the filtering result is divided according to a preset strategy rule to generate a risk control result. The filtering result includes: non-random missing data.

[0047] In this embodiment, when the missing rate corresponding to the target feature is neither random nor completely random (i.e., the missing pattern cannot be fully explained by other observed variables), it should be highly suspected that it is non-random missing. Since non-random missing involves the unobserved value itself, its determination must follow a two-step principle: initial screening based on business logic and data risk verification. A rationality analysis of the potential causes of feature missing is conducted, incorporating domain knowledge. In cross-border e-commerce risk control scenarios, typical business types with non-random missing include: 1. Personal risk characteristics: Applicants with high income (tax avoidance or high net worth) or very low income (weak repayment ability) are more likely to conceal their income, and those with poor credit are more likely to not fill in the historical debt field.

[0048] 2. Business risk characteristics: Sellers with deteriorating business conditions and tight cash flow are more likely to lack store transaction records or information on major payment accounts for the past six months.

[0049] Based on the preset screening algorithm, the missing rate corresponding to the target feature is filtered by type to generate screening results. When the screening result is non-random missing, the strategy rules will apply risk control hit strategies to the feature type corresponding to the target feature and non-random missing, generating relevant risk control results. The strategy rules can be set according to requirements; see Table 1 for specific setting examples.

[0050] Table 1. Strategy Rules Table

[0051] For details, please refer to Figure 4 , Figure 4 This is a schematic diagram of a specific embodiment of the 106 steps of the e-commerce risk control method based on missing data values ​​in this invention. Step 106, "based on a preset filtering algorithm, performs category filtering processing on the missing rate corresponding to the target feature to generate filtering results," includes the following specific implementation methods: 1061. Create indicator variables with missing binary values ​​based on target features; 1062. Based on the preset information evaluation formula and preset risk control labels, calculate the information value of the indicator variables to generate information value; 1063. Calculate the correlation between the indicator variable and the preset risk control label according to the preset chi-square test algorithm, and generate a chi-square value; 1064. Based on the chi-square value and the information value, generate the screening results.

[0052] In steps 1061-1064, a binary missing indicator variable Mx is generated for the target feature X to be verified, where M... x =1 indicates missing, M x =0 indicates an observation.

[0053] Using the risk control label Y as the benchmark, Y=1 represents a bad sample / overdue, and Y=0 represents a good sample. A minimum smoothing term ε=10 is introduced. -5 To avoid the denominator being zero, calculate M. x The information value IV.

[0054] For an index i ∈ {0, 1}, index i = 1 indicates that the target feature is missing, and i = 0 indicates that the target feature is observed. Calculate the smoothed conditional percentage of the missing variable with value i in both the bad sample group and the good sample group:

[0055]

[0056] Wherein, N(M) x =i, Y=1) represents the number of bad samples where the variable is missing only when it takes the value i, N(M x P(Y=1, Y=0) represents the number of good samples where the missing indicator variable takes the value i, N(Y=1) represents the total number of bad samples, and N(Y=0) represents the total number of good samples. i Bad P is the conditional probability of a bad sample. i Good The conditional probability of a good sample.

[0057] Further calculate the weighted evidence value:

[0058] Then, the value of the information is calculated further: ; Where IV stands for Information Value, and WOE (Word of Value) i The weighted evidence value, index i∈{0,1} represents the risk control label, i=1 indicates missing features, i=0 indicates observed features, M xIt is an indicator variable.

[0059] The chi-square test algorithm is used to calculate the indicator variable M. x The correlation with preset risk control labels is used to generate a chi-square value (P-value). When the p-value is less than a preset first threshold and the information value (IV) is greater than a preset second threshold, it is determined that there is a significant risk correlation between the missing state of the target feature and the distribution of bad samples, confirming that the missing mechanism corresponding to the target feature is non-random missing.

[0060] When the p-value is not less than a preset first threshold, or the information value (IV) is not greater than a preset second threshold, it is determined that there is no significant risk association between the missing state of the target feature and the distribution of bad samples, and it cannot be confirmed as a risk-related non-random missing feature. In this case, according to the preset strategy rules, the missing state corresponding to the target feature can be downgraded to low-risk missing feature processing, and subsequent processing can be carried out using methods such as mean imputation, median imputation, independent binning, or retention of missing label.

[0061] Specifically, step 1064 includes the following specific implementation methods: 10641. When the chi-square value is less than a preset first threshold and the information value is greater than a preset second threshold, the missing rate corresponding to the target feature is confirmed to be non-random missing. 10642. When the chi-square value is not less than a preset first threshold or the information value is not greater than a preset second threshold, the missing rate corresponding to the target feature is confirmed to be completely random missing.

[0062] In steps 10641-10642, when the information value (IV) is greater than 0.02 and the chi-square value (P) is less than 0.05, the missing rate corresponding to the target feature is confirmed to be non-random missing.

[0063] If the information value (IV) is not greater than 0.02 or the chi-square value (P) is not less than 0.05, it cannot be definitively determined whether the missing information is random or completely random. Since the test has rejected the strictly non-random missing information hypothesis, theoretically the missing information is not completely random; however, given that the missing pattern is uninterpretable and has no risk correlation, in practice, it can be downgraded to confirming the missing rate corresponding to the target feature as completely random missing. Missing values ​​can be imputed using the mean, median, or deletion methods to balance computational efficiency and model stability.

[0064] In this embodiment of the invention, by calculating the missing rate of e-commerce feature data and utilizing the requirements corresponding to the missing rates of different features, target features requiring risk control analysis are identified. Then, parametric likelihood ratio tests, logistic regression analysis, and information value screening are performed on the missing rates of the target features to classify the causes of missing values ​​into three categories: completely random missing, random missing, and non-random missing. Based on the risk control strategy, the business attributes of the target features and the causes of missing values ​​are combined to classify strategic risks and generate risk control results. This solution can accurately identify target features with high risk, realize the identification of potential risks in missing values, significantly improve the identification rate of potential risky behaviors, and solve the technical problem in existing e-commerce risk control that lacks analysis of the causes of missing values ​​for risk control.

[0065] Figure 5 This is a schematic diagram of the structure of an e-commerce risk control device 500 based on missing data values, provided by an embodiment of the present invention. This e-commerce risk control device 500 based on missing data values ​​can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 510 and memory 520, and one or more storage media 530 for storing application programs 533 or data 532. The memory 520 and storage media 530 can be temporary or persistent storage. The program stored in the storage media 530 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the e-commerce risk control device 500 based on missing data values. Furthermore, the processor 510 may be configured to communicate with the storage media 530 and execute the series of instruction operations in the storage media 530 on the e-commerce risk control device 500 based on missing data values.

[0066] The e-commerce risk control device 500 based on missing data values ​​may also include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input / output interfaces 560, and / or one or more operating systems 531, such as Windows Server, Mac OS X, Unix, Linux, Free BSD, etc. Those skilled in the art will understand that... Figure 5 The illustrated e-commerce risk control device structure based on missing data values ​​does not constitute a limitation on e-commerce risk control devices based on missing data values. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0067] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the e-commerce risk control method based on missing data values.

[0068] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0069] Furthermore, although the operations are described in a specific order, this should be understood as requiring that such operations be performed in the specific order shown or in sequential order, or requiring that all illustrated operations be performed to achieve the desired result. In certain environments, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0070] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. An e-commerce risk control method based on missing data values, characterized in that, Including the following steps: Receive e-commerce feature data, calculate the missing rate of the e-commerce feature data, generate multiple missing rates corresponding to feature data, and filter out target features based on the missing rates corresponding to multiple feature data. Based on a preset parameterized likelihood ratio test, it is determined whether the missing rate corresponding to the target feature is completely random missing. When the feature is completely randomly missing, the target feature is divided into two categories according to the preset strategy rules, and the combination of the target feature and the completely randomly missing feature is processed to generate a risk control result. When the missing values ​​are not completely random, the missing rate corresponding to the target feature is determined to be random based on the preset logistic regression analysis. When there is random missing information, the target feature is divided into two categories according to the preset strategy rules, and the combination of the target feature and the random missing information is processed to generate a risk control result. When the missing data is not random, the missing rate corresponding to the target feature is filtered according to a preset filtering algorithm to generate a filtering result. The combination of the feature type corresponding to the target feature and the filtering result is divided according to a preset strategy rule to generate a risk control result. The filtering result includes: non-random missing data.

2. The e-commerce risk control method based on missing data values ​​according to claim 1, characterized in that, The step of calculating the missing rate of the e-commerce feature data and generating multiple missing rates corresponding to the feature data includes: ； wherein R miss j is the missing rate of the j-th feature, N is the total number of samples of the j-th feature, II() is an exponential function, x ij is the sample of the i-th j-th feature.

3. The e-commerce risk control method based on missing data values ​​according to claim 1, characterized in that, The step of determining whether the missing rate corresponding to the target feature is completely random missing based on the preset parameterized likelihood ratio test includes: Initial parameters are generated based on the sample mean and sample covariance corresponding to the target features as initial estimates. Based on the initial parameters, calculate the expected value of the target feature; With the goal of maximizing the expected value, the initial parameters are updated until a preset stopping rule is met, and the final parameters are generated. Based on the final parameters, substitute them into the preset chi-square formula and the preset degree of freedom formula to calculate the chi-square value and the degree of freedom value; Based on the chi-square value and the degree of freedom value, determine whether the missing rate corresponding to the target feature is completely random missing.

4. The e-commerce risk control method based on missing data values ​​according to claim 3, characterized in that, The chi-square formula includes: ； Among them, X 2 Here, n is the chi-square value, G is the number of groups with different missing patterns, and n is the number of groups with different missing patterns. g Let y be the sample size of the g-th group. g - Let u be the sample mean vector of the g-th group. g ^ The mean term of the final parameter in the g-th group. Let g be the covariance term of the final parameter in the g-th group.

5. The e-commerce risk control method based on missing data values ​​according to claim 3, characterized in that, The formula for the degrees of freedom includes: ； Where df is the degrees of freedom, G is the number of groups with different deletion patterns, and p is the total number of research variables. g is the number of observed variables in the g-th missing pattern.

6. The e-commerce risk control method based on missing data values ​​according to claim 1, characterized in that, The step of determining whether the missing rate corresponding to the target feature is random missing based on preset logistic regression analysis includes the following steps: Create indicator variables with missing values ​​based on target features; Based on a preset Lasso logistic regression, the indicator variable is subjected to missing factor detection processing to generate detection results; Based on the detection results, it is determined whether the missing rate corresponding to the target feature is random missing.

7. The e-commerce risk control method based on missing data values ​​according to claim 1, characterized in that, The step of performing category filtering on the missing rate corresponding to the target feature according to a preset filtering algorithm to generate filtering results includes the following steps: Create indicator variables with missing values ​​based on target features; Based on the preset information evaluation formula and preset risk control labels, the information value of the indicator variables is calculated to generate information value; Based on the preset chi-square test algorithm, the correlation between the indicator variable and the preset risk control label is calculated, and a chi-square value is generated; Based on the chi-square value and the information value, a screening result is generated.

8. The e-commerce risk control method based on missing data values ​​according to claim 7, characterized in that, The process of generating the screening results based on the chi-square value and the information value includes the following steps: When the chi-square value is less than a preset first threshold and the information value is greater than a preset second threshold, the missing rate corresponding to the target feature is confirmed to be non-random missing. When the chi-square value is not less than a preset first threshold or the information value is not greater than a preset second threshold, the missing rate corresponding to the target feature is confirmed to be completely random missing.

9. An e-commerce risk control device based on missing data values, characterized in that, The e-commerce risk control device based on missing data values ​​includes: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a line. The at least one processor invokes the instructions in the memory to cause the e-commerce risk control device based on missing data values ​​to execute the e-commerce risk control method based on missing data values ​​as described in any one of claims 1-8.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the e-commerce risk control method based on missing data values ​​as described in any one of claims 1-8.

Images