Polynomial label generation method and device for risk profiling

Abstract

This invention provides a multinomial label generation method and apparatus for risk profile construction, characterized by the following steps: S1-S3, dividing all features into multiple sub-feature sets, and selecting features from each sub-feature set to construct a filtered feature set based on importance value, relevance value, and feature selection quantity; S4, performing multinomial cross-feature calculation on the features in the filtered feature set to obtain a multinomial feature set; S5, calculating the importance value, relevance value, and feature selection quantity of each feature in the multinomial feature set; S6, selecting multiple features from the multinomial feature set based on importance value, relevance value, and feature selection quantity to construct a multinomial filtered feature set; and S7, calculating the WOE value of each feature corresponding to each target in the multinomial filtered feature set. In summary, this method can describe the multidimensional risk status of a target with fine granularity.

Description

Technical Field

[0001] This invention belongs to the field of risk profiling, specifically relating to a polynomial tag generation method and apparatus for constructing risk profiles. Background Technology

[0002] Risk profiling is used across multiple domains to characterize an entity's risk status by analyzing historical data and multi-dimensional features, thereby aiding decision-making. Existing risk profiling technologies primarily rely on rule-driven methods, which construct risk assessment logic and generate corresponding risk labels based on domain expert knowledge and existing rules. These methods offer certain advantages in practice, such as clear logic and ease of interpretation, but they also have significant limitations, specifically as follows:

[0003] 1) Insufficient representational power of original features: In risk scenarios, original features may be insufficient to fully describe the complex changes in risk, especially when there are interactive effects or nonlinear relationships between multiple features. Most existing methods are based on single features or simple combinations of features, which cannot accurately capture these complex relationships, thus limiting the representational power of risk profiling.

[0004] 2) Limitations of Feature Selection and Dimensionality Reduction: Traditional methods often use feature selection techniques based on statistical indicators or simple algorithms, which may not be able to effectively extract important features related to risk. In addition, redundant and noisy features in high-dimensional feature spaces can easily lead to high computational costs and large model complexity, while reducing the efficiency and accuracy of risk label generation.

[0005] 3) Insufficient representation of multidimensional feature relationships: Existing technologies typically consider only a single feature or a simple combination of features when generating feature labels, failing to fully utilize the interaction relationships between features. For example, linear models often ignore nonlinear feature relationships, while deep learning models, although capable of capturing certain nonlinear relationships, lack targeted optimization, and their performance depends on large-scale data and complex computations. Summary of the Invention

[0006] This invention is made to solve the above-mentioned problems, and its purpose is to provide a polynomial tag generation method and apparatus for risk profile construction.

[0007] This invention provides a multinomial label generation method for risk profile construction, used to generate label results related to the risk profile for each target based on risk-related data. The risk-related data includes the values ​​of multiple different features corresponding to each target, and risk judgment labels corresponding to each target. The method includes the following steps: Step S1, dividing all features into multiple categories to generate sub-feature sets corresponding to each category, each sub-feature set including at least one feature and all values ​​corresponding to that feature; Step S2, for each sub-feature set, calculating the importance value, relevance value, and feature selection quantity for each feature based on the number of features and their corresponding values; Step S3, selecting multiple features from each sub-feature set based on the importance value, relevance value, and feature selection quantity to construct a selection feature set; Step S4, ... Step S5: Perform multinomial cross-feature calculation on the features in the selected feature set to obtain a multinomial feature set; Step S6: Calculate the importance value, relevance value, and feature selection quantity of each feature based on the number of features and their corresponding values ​​in the multinomial feature set; Step S7: Select multiple features from the multinomial feature set based on the importance value, relevance value, and feature selection quantity to construct a multinomial selection feature set; Step S8: Perform binning calculation on the multinomial selection feature set to obtain the binning value of each feature corresponding to each target in the multinomial selection feature set, and calculate the WOE value of each feature corresponding to each target in the multinomial selection feature set based on all binning values ​​and risk judgment labels. The label result of the target is the target's features in the multinomial selection feature set and its corresponding WOE value.

[0008] The polynomial label generation method for risk profile construction provided by this invention may also have the following feature: wherein, in step S4, the calculation expression of the polynomial feature set is: X polyfea =ploy(X selected ), x = [χ1, χ2, ..., χ n ], where X polyfea Let X be the set of polynomial characteristics. selected To filter the feature set, χ n Let x be the nth feature in the feature set.

[0009] The polynomial label generation method for risk profile construction provided by this invention may also have the following feature: wherein, in step S3, the calculation expression for filtering the feature set is: X j,selected =X j,corr ∪X j,lgbm In the formula X selectedTo filter the feature set, m is the total number of sub-feature sets, X j,selected Let X be the selection set corresponding to the sub-feature set j. j,corr To filter the feature set corresponding to the relevance of sub-feature set j, X j,lgbm To filter the feature set corresponding to the importance of sub-feature set j, sort all features of sub-feature set j in descending order of relevance value, select the feature with the number of features to be filtered from the beginning of the sequence, and construct the relevance-filtered feature set. To filter the feature set by sorting all features of sub-feature set j in descending order of importance value, select the feature with the number of features to be filtered from the beginning of the sequence.

[0010] The polynomial label generation method for risk profile construction provided by this invention may also have the following feature: wherein, in step S2, the expression for calculating the correlation value is: In the formula Let i be the correlation value of feature i in feature set j. Let i be the value of feature i corresponding to the k-th target in feature set j. Let y be the mean of feature i in feature set j. k The risk assessment label corresponding to the k-th target. Let n be the average of the risk assessment labels for all targets, and n be the total number of targets.

[0011] The multinomial label generation method for risk profile construction provided by this invention may also have the following feature: wherein, in step S2, the corresponding importance value is calculated based on the information gain of the feature at the node of the tree in the LightGBM model, and the expression for calculating the importance value is: In the formula Let be the importance value of feature i in feature set j, p be the total number of features in feature set j, M be the total number of trees in the LightGBM model, and N be the total number of trees in the model. m Let m be the set of nodes of the m-th tree. For the indicator function, ΔL m,n Let i be the information gain of feature i at node n in the m-th tree, when node n uses feature i for splitting. When node n does not use feature i for splitting

[0012] The polynomial label generation method for risk profile construction provided by this invention may also have the following feature: wherein, in step S2, the expression for calculating the feature selection quantity is: In the formula, select_len j Let be the feature selection quantity corresponding to feature set j, and p be the total number of features in feature set j. This is a floor operation.

[0013] The polynomial label generation method for risk profile construction provided by this invention may also have the following features: Step S7 includes the following sub-steps: Step S7-1, generating multiple candidate split thresholds for the polynomial screening feature set, and calculating the split value corresponding to each candidate split threshold; Step S7-2, taking the candidate split threshold corresponding to the largest split value as the optimal split point; Step S7-3, repeating steps S7-1 to S7-2 until the number of optimal split points reaches a preset value; Step S7-4, calculating the binning boundary based on all optimal split points; Step S7-5, calculating the bins value and bin0 value of each feature corresponding to each target in the polynomial screening feature set as binning values ​​based on the binning boundary; Step S7-6, calculating the woe value of each feature corresponding to each target in the polynomial screening feature set based on all binning values.

[0014] The polynomial label generation method for risk profile construction provided by this invention may also have the following feature: wherein the calculation expression for the split value is: H(D) ps )=-∑ c∈C p(c)log2p(c), In the formula IG(D ps ,θ) is the splitting value of the candidate splitting threshold θ corresponding to the polynomial screening feature set, and |D| is the number of samples in D. v represents the value of feature i corresponding to target k in the multinomial filtering feature set. k For target k, y k Let C be the risk assessment label corresponding to target k, C be the total number of features in the multinomial screening feature set, p(c) be the probability of feature c in the multinomial screening feature set, and the expression for calculating the binning boundary is: In the formula for The set of bin boundary values, To sort all optimal split points in ascending order, the expressions for calculating the bins and bin0 values ​​are as follows: In the formula The value of feature i corresponding to target k in the multinomial-selected feature set. for The left boundary value of the u-th bin interval.

[0015] The multinomial label generation method for risk profile construction provided by this invention may also have the following feature: In steps S7-6, based on the bin0 value of the feature corresponding to the target, the values ​​of all feature corresponding to the target are divided into multiple groups, where the values ​​of all feature corresponding to the target in each group have the same bin0 value. The risk judgment label is a binary classification label, and the expression for calculating the WOE value is: In the formula Let Woe be the value of feature i corresponding to target k in the polynomial-selected feature set, and N be the value of Woe. bad Let N be the total number of risk judgment labels corresponding to all values ​​of feature i in the group containing the target k in the polynomial screening feature set, where the label is of the bad category. good Let N be the total number of risk judgment labels corresponding to all values ​​of feature i in the group containing the target k in the polynomial screening feature set, and all of which are classified as "good". total Let i be the total number of values ​​of feature i corresponding to target k in the polynomial selection feature set.

[0016] This invention also provides a multinomial label generation device for risk profile construction, used to generate label results related to the risk profile for each target based on risk-related data. The risk-related data includes values ​​of multiple different features corresponding to each target, and risk judgment labels corresponding to each target. The device comprises: a subset partitioning module, which divides all features into multiple categories and generates sub-feature sets corresponding to each category, each sub-feature set including at least one feature and all values ​​corresponding to that feature; a filtering module, which calculates the importance value, relevance value, and feature filtering quantity of each feature based on the number of features and their corresponding values ​​in the input feature set; and a filtering set generation module, which selects from each sub-feature set based on the importance value, relevance value, and feature filtering quantity corresponding to each sub-feature set. The process involves several steps: First, a selection feature set is constructed by taking multiple features. Second, a multinomial cross-feature module performs multinomial cross-feature calculations on the features in the selection feature set to obtain a multinomial feature set. Third, a multinomial selection set generation module selects multiple features from the multinomial feature set based on their importance, relevance, and selection quantity to construct the multinomial selection feature set. Fourth, a WOE calculation module performs binning calculations on the multinomial selection feature set to obtain the binning values ​​of each feature corresponding to each target in the multinomial selection feature set. Finally, based on all binning values ​​and risk assessment labels corresponding to the multinomial selection feature set, the WOE values ​​of each feature corresponding to each target in the multinomial selection feature set are calculated. The label result for a target is the target's features in the multinomial selection feature set and its corresponding WOE value.

[0017] The role and effect of invention

[0018] According to the polynomial label generation method and apparatus for risk profile construction of the present invention, by dividing features into sub-feature sets and calculating the corresponding importance values, relevance values, and feature selection factors, multiple features are selected from each sub-feature set to construct a selection feature set. Then, polynomial cross-feature calculation is performed on the features in the selection feature set to obtain a polynomial feature set. Subsequently, the corresponding importance values, relevance values, and feature selection factors are calculated on the polynomial feature set to further select features and construct the polynomial selection feature set. Finally, the WOE value of each target is calculated for the features in the polynomial selection feature set to obtain the label result. Therefore, the polynomial label generation method and apparatus for risk profile construction of the present invention can describe the multidimensional risk status of targets with fine granularity. Attached Figure Description

[0019] Figure 1 This is a block diagram of a polynomial tag generation device for risk profile construction in an embodiment of the present invention;

[0020] Figure 2 This is a schematic diagram of the process by which the woe calculation module generates woe values ​​in an embodiment of the present invention;

[0021] Figure 3 This is a flowchart illustrating the polynomial tag generation method for risk profile construction in an embodiment of the present invention.

[0022] Figure 4 This is a radar diagram illustrating a high-risk user profile in an embodiment of the present invention;

[0023] Figure 5 This is a radar schematic diagram of a low-risk user profile in an embodiment of the present invention;

[0024] Figure 6 This is a schematic diagram of the scorecard visualization in an embodiment of the present invention. Detailed Implementation

[0025] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the polynomial tag generation method and apparatus for risk profile construction of the present invention.

[0026] This embodiment provides a multinomial label generation device for risk profile construction, used to generate label results related to the risk profile for each target based on risk-related data. The risk-related data includes the values ​​of multiple different features corresponding to each target, and risk assessment labels for each target. The risk assessment labels are binary classification labels.

[0027] In this embodiment, the multinomial tag generation device for risk profiling is applied in the financial field, and the risk-related data is financial transaction data D = [VXY]. Where V = {v1,...,v...} n} represents a set of trading accounts, where each trading account is the target. For each v i Each ∈V has a binary label indicating whether it is illegal, i.e., a separator label y. i ∈Y, where n is the number of trading accounts, and the feature space constructed from the perspectives of statistical calculation, complex rules, expert knowledge, and intelligent algorithms is X, where X={X1,X2,…,X…} d}, X i ={χ i,k |k=1,2,…,n}, representing the feature χ i The feature set, where χ i,k χ represents the feature corresponding to the k-th transaction account. i The value of . In other embodiments, the polynomial label generation device for risk profile construction can be applied to other fields, such as food safety, where the target is a specific food and the risk-related data is production data related to the specified food, such as the manufacturer.

[0028] Figure 1 This is a block diagram of a polynomial tag generation device for risk profile construction in an embodiment of the present invention.

[0029] like Figure 1 As shown, the polynomial tag generation device 100 for risk profile construction includes a subset partitioning module 11, a filtering module 12, a filtering set generation module 13, a polynomial cross module 14, a polynomial filtering set generation module 15, a WOE calculation module 16, and a control module 17 that controls the operation of the above modules.

[0030] Subset partitioning module 11 divides all features into multiple categories and generates sub-feature sets corresponding to each category. Each sub-feature set includes at least one feature and all values ​​corresponding to that feature.

[0031] In this embodiment, the subset partitioning module 11 divides information into m categories, thus obtaining the sub-feature set X. 1 ,X 2 ,…,X m And satisfy

[0032] The filtering module 12 calculates the importance value, relevance value, and feature filtering quantity of each feature based on the number of features and the corresponding values ​​in the input feature set.

[0033] The filtering module 12 includes a LightGBM model. Based on the information gain of features at the nodes of the LightGBM model's tree, the filtering module 12 calculates the corresponding importance value. The expression for calculating the importance value is as follows:

[0034]

[0035] In the formula Let be the importance value of feature i in feature set j, p be the total number of features in feature set j, M be the total number of trees in the LightGBM model, and N be the total number of trees in the model. m Let m be the set of nodes of the m-th tree. For the indicator function, ΔL m,n Let i be the information gain of feature i at node n in the m-th tree. When node n splits using feature i, When node n does not use feature i for splitting

[0036] In this embodiment, the loss function corresponding to the LightGBM model is the binary cross-entropy loss L, and its calculation expression is as follows:

[0037]

[0038] In the formula, N is the total number of samples, and y i Let i be the risk assessment label for the i-th sample. Let be the predicted probability of the LightGBM model for the i-th sample.

[0039] The expression for calculating the correlation value is:

[0040]

[0041] In the formula Let i be the correlation value of feature i in feature set j. Let i be the value of feature i corresponding to the k-th target in feature set j. Let y be the mean of feature i in feature set j. k The risk assessment label corresponding to the k-th target. Let n be the average of the risk assessment labels for all targets, and n be the total number of targets.

[0042] The expression for calculating the feature selection quantity is:

[0043]

[0044] In the formula, select_len j Let be the feature selection quantity corresponding to feature set j, and p be the total number of features in feature set j. This is a floor operation.

[0045] The filter set generation module 13 selects multiple features from each sub-feature set based on the importance value, relevance value, and feature selection quantity of the corresponding sub-feature set to construct a filter feature set.

[0046] The calculation expression for selecting the feature set is as follows:

[0047]

[0048] X j,selected =X j,corr ∪X j,lgbm ,

[0049] In the formula X selected To filter the feature set, m is the total number of sub-feature sets, X j,selected Let X be the selection set corresponding to the sub-feature set j. j,corr To filter the feature set corresponding to the relevance of sub-feature set j, X j,lgbm Filter the feature set corresponding to the importance of sub-feature set j.

[0050] Sort all features in sub-feature set j according to their relevance values ​​from largest to smallest. Then, select the feature with the highest number of features selected from the beginning of this sequence to construct the relevance-selected feature set. Relevance-selected feature set X j,corr The calculation expression is:

[0051] X j,corr =argsort(|correlations) j |)[:select_len j ],

[0052]

[0053] In the formula, argsort(|correlations) j |) This sorts the relevance values ​​of the sub-feature set j in descending order by absolute value; that is, in this embodiment, the sorting is based on the magnitude of the absolute value of the relevance values. [:select_len j ] represents the feature selection quantity corresponding to the sub-feature set j.

[0054] Sort all features in sub-feature set j according to their importance value from largest to smallest. Then, select the feature with the highest importance value from the beginning of this sequence to construct the importance-selected feature set. Importance-selected feature set X j,lgbm The calculation expression is:

[0055] X j,lgbm =argsort(|lgbm_importance j|)[:select_len j ],

[0056]

[0057] In the formula, argsort(|lgbm_importance) j |) sorts the sub-feature set j in descending order of importance values ​​by absolute value.

[0058] The polynomial cross module 14 performs polynomial cross feature calculation on the features in the selected feature set to obtain a polynomial feature set.

[0059] The expression for calculating the polynomial characteristic set is as follows:

[0060] X polyfea =poly(X) selected ),

[0061]

[0062] x=[χ1,χ2,…,χ n ],

[0063] In the formula X polyfea Let X be the set of polynomial characteristics. selected To filter the feature set, χ n Let x be the nth feature in the feature set.

[0064] The polynomial filtering set generation module 15 selects multiple features from the polynomial feature set according to their corresponding importance values, relevance values, and feature selection quantities to construct a polynomial filtering feature set. In this embodiment, the specific processing procedure of the polynomial filtering set generation module 15 in generating the polynomial filtering feature set based on the polynomial feature set is the same as the processing procedure of the filtering set generation module 13 in generating the corresponding filtering set based on the sub-feature set, and will not be repeated here.

[0065] The WOE calculation module 16 performs binning calculations on the multinomial screening feature set to obtain the binning value of each feature corresponding to each target in the multinomial screening feature set. Based on all the binning values ​​and risk assessment labels corresponding to the multinomial screening feature set, it calculates the WOE value of each feature corresponding to each target in the multinomial screening feature set. The target label result consists of each feature of that target in the multinomial screening feature set and its corresponding WOE value.

[0066] Figure 2 This is a schematic diagram of the process by which the woe calculation module generates woe values ​​in an embodiment of the present invention.

[0067] like Figure 2As shown, the process of generating woe values ​​by the woe calculation module 16 includes the following steps:

[0068] Step S7-1: For the polynomial-filtered feature set, generate multiple candidate split thresholds and calculate the split value corresponding to each candidate split threshold. In this embodiment, the candidate split threshold is obtained by averaging the sorted adjacent samples.

[0069] The expression for calculating the splitting value is as follows:

[0070]

[0071] H(D ps )=-∑ c∈C p(c)log2p(c),

[0072]

[0073] In the formula IG(D ps Let θ be the splitting value of the candidate splitting threshold θ corresponding to the polynomial screening feature set, and |D| be the number of samples in D. A sample includes the values ​​of all features corresponding to a target in set D. v represents the value of feature i corresponding to target k in the multinomial filtering feature set. k For target k, y k Let C be the risk assessment label corresponding to target k, C be the total number of features in the multinomial screening feature set, and p(c) be the probability of feature c in the multinomial screening feature set.

[0074] Step S7-2: Select the candidate split threshold corresponding to the largest split value as the optimal split point.

[0075] Step S7-3: Repeat steps S7-1 to S7-2 until the number of optimal split points reaches a preset value. In this embodiment, iterating through two layers generates at most four optimal split points.

[0076] Step S7-4: Calculate the bin boundaries based on all optimal split points.

[0077] The calculation expression for the bin boundary is as follows:

[0078]

[0079] In the formula for The set of bin boundary values, To sort all optimal split points in ascending order, negative infinity (-∞) and positive infinity (∞) are used in this embodiment to ensure that the final bin boundaries cover all values.

[0080] Step S7-5: Based on the binning boundaries, calculate the bins and bin0 values ​​of each feature corresponding to each target in the polynomial filtering feature set as binning values.

[0081] The expressions for calculating the bins and bin0 values ​​are as follows:

[0082]

[0083]

[0084] In the formula The value of feature i corresponding to target k in the multinomial-selected feature set. for The left boundary value of the u-th bin interval.

[0085] Step S7-6: Based on all binning values, calculate the WOE value of each feature corresponding to each target in the multinomial screening feature set.

[0086] Specifically, the woe calculation module 16 divides the values ​​of all target-corresponding features into multiple groups based on the bin0 value of the feature corresponding to the target. The feature values ​​corresponding to all targets in the group have the same bin0 value, that is, if The features corresponding to the target

[0087] The expression for calculating the woe value is:

[0088]

[0089] In the formula Let Woe be the value of feature i corresponding to target k in the polynomial-selected feature set, and N be the value of Woe. bad Let N be the total number of risk judgment labels corresponding to all values ​​of feature i in the group containing the target k in the polynomial screening feature set, where the label is of the bad category. good Let N be the total number of risk judgment labels corresponding to all values ​​of feature i in the group containing the target k in the polynomial screening feature set, and all of which are classified as "good". total This represents the total number of values ​​of feature i corresponding to target k in the multinomial screening feature set. In this embodiment, the risk judgment label is 0 or 1, where 0 represents the good category and 1 represents the bad category. In other embodiments, other forms can be used to replace 0 and 1 to distinguish between good and bad categories.

[0090] Control module 17 stores the control program that controls the operation of each module.

[0091] The following description, in conjunction with the accompanying drawings, explains the process of using a polynomial label generation device 100 for risk profile construction to generate polynomial labels.

[0092] Figure 3 This is a flowchart illustrating the polynomial tag generation method for risk profile construction in an embodiment of the present invention.

[0093] like Figure 3 As shown, the polynomial label generation method for risk profile construction includes the following steps:

[0094] Step S1: Use the subset partitioning module 11 to divide all features into multiple categories and generate a sub-feature set corresponding to each category. The sub-feature set includes at least one feature and all values ​​corresponding to that feature.

[0095] Step S2: Using the filtering module 12, for each sub-feature set, the importance value, relevance value, and feature filtering quantity of each feature are calculated based on the number of features in the sub-feature set and the corresponding values ​​of the features.

[0096] Step S3: The filtering set generation module 13 selects multiple features from each sub-feature set based on importance value, relevance value and feature filtering quantity to construct a filtering feature set.

[0097] Step S4: Use the polynomial cross module 14 to perform polynomial cross feature calculation on the features in the selected feature set to obtain the polynomial feature set.

[0098] Step S5: The filtering module 12 calculates the importance value, relevance value, and feature filtering quantity of each feature based on the number of features and the corresponding values ​​in the polynomial feature set.

[0099] Step S6: The multinomial filtering set generation module 15 selects multiple features from the multinomial feature set based on importance value, relevance value and feature filtering quantity to construct a multinomial filtering feature set.

[0100] Step S7: Use the WOE calculation module 16 to perform binning calculation on the multinomial screening feature set to obtain the binning value of each feature corresponding to each target in the multinomial screening feature set. Based on all the binning values ​​and risk judgment labels corresponding to the multinomial screening feature set, calculate the WOE value of each feature corresponding to each target in the multinomial screening feature set.

[0101] In this embodiment, a radar map of the user profile of the target can be constructed based on the target's tag results.

[0102] Figure 4 This is a radar diagram illustrating a high-risk user profile in an embodiment of the present invention.

[0103] Figure 5 This is a radar diagram illustrating a low-risk user profile in an embodiment of the present invention.

[0104] like Figure 4 and Figure 5 As shown, the labeling results include eight features (polynomial labels) and their corresponding values: ploy_fea_1, ploy_fea_2, ploy_fea_3, ploy_fea_4, ploy_fea_5, ploy_fea_6, ploy_fea_7, and ploy_fea_8. It is evident that abnormal subjects (high-risk users) occupy a larger area, while normal subjects (low-risk users) occupy a smaller area, thus intuitively demonstrating the effective multi-dimensional risk hierarchy portrayal of the subject's risk by the labeling results.

[0105] This embodiment extracts feature coefficients and intercepts based on existing scoring models, sets a baseline score and a score PDO for each doubling of odds, and calculates factor weights and offsets to create a scorecard containing features, coefficients, sample scores, and predicted probabilities. The scores are divided into several equidistant scoring intervals, and the number and proportion of positive and negative samples within each interval are calculated to visually reflect the distribution of positive and negative samples across the intervals.

[0106] Figure 6 This is a schematic diagram of the scorecard visualization in an embodiment of the present invention.

[0107] like Figure 6 As shown, the horizontal axis represents the sample score interval values ​​(score_bins), the vertical axis of the histogram represents the number of samples (Count) within that score interval, and the vertical axis of the line graph represents the bad rate within that score interval. This demonstrates that the scorecard based on the label results intuitively shows how each module distinguishes between positive and abnormal samples, achieving a strong correlation between the abnormal subject and the risk score.

[0108] In this embodiment, some key model labels were selected and grouped using binning values. The number and proportion of bad samples within each group were calculated, and the results are shown in the table below:

[0109]

[0110] The first column in the table above represents the different polynomial labels χ. polyfea1 , χ polyfea2 and χ polyfea3The second to sixth columns represent the woe value, bin range, number of bad samples, total number of samples, and percentage of bad samples, respectively. This shows that in the given example key polynomial label groupings, the optimal label consistently achieves a bad sample ratio of 98.43%, demonstrating the precise representation of bad samples by polynomial labels within a specific dimension.

[0111] The role and effect of the embodiments

[0112] According to the polynomial label generation method and apparatus for risk profile construction involved in this embodiment, features are divided into sub-feature sets, and corresponding importance values, relevance values, and feature selection factors are calculated. Multiple features are selected from each sub-feature set to construct a selection feature set. Then, polynomial cross-feature calculation is performed on the features in the selection feature set to obtain a polynomial feature set. Subsequently, the corresponding importance values, relevance values, and feature selection factors are calculated on the polynomial feature set to select features and construct the polynomial selection feature set. Finally, the WOE value of each target is calculated for the features in the polynomial selection feature set to obtain the label result. In summary, this method enhances the capture of nonlinear relationships and the characterization of feature interaction relationships by generating polynomial selection feature sets through polynomial cross-calculation and selection. It effectively reduces dimensionality while retaining key risk-related information, providing a fine-grained description of the multidimensional risk state of the target.

[0113] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A multinomial label generation method for risk profile construction, used to generate label results related to the risk profile for each target based on risk-related data, wherein the risk-related data includes values ​​of multiple different features corresponding to each target, and risk judgment labels corresponding to each target, characterized in that, Includes the following steps: Step S1: Divide all the features into multiple categories and generate a sub-feature set corresponding to each category. The sub-feature set includes at least one feature and all the values ​​corresponding to that feature. Step S2: For each of the sub-feature sets, calculate the importance value, relevance value, and feature selection quantity of each feature based on the number of features in the sub-feature set and the value corresponding to the feature; Step S3: From each of the sub-feature sets, select multiple features based on the importance value, the relevance value, and the feature selection quantity to construct a selection feature set; Step S4: Perform polynomial cross feature calculation on the features in the selected feature set to obtain a polynomial feature set; Step S5: Based on the number of features in the polynomial feature set and the value corresponding to the feature, calculate the importance value, the relevance value, and the feature selection quantity for each feature; Step S6: From the polynomial feature set, select multiple features based on the importance value, the relevance value, and the feature selection quantity to construct a polynomial selection feature set; Step S7: Perform binning calculation on the polynomial screening feature set to obtain the binning value of each feature corresponding to each target in the polynomial screening feature set. Based on all the binning values ​​corresponding to the polynomial screening feature set and the risk judgment label, calculate the WOE value of each feature corresponding to each target in the polynomial screening feature set. Wherein, the label result of the target is each of the features of the target in the polynomial filtering feature set and its corresponding WOE value. The multinomial label generation method for risk profiling is applied in the financial field, where the risk-related data is financial transaction data D = [VXY], where V = {v1,...,v...} n } represents a set of trading accounts, where each trading account is the target. For each v i Each ∈V has a binary label indicating whether it is illegal, i.e., a separator label y. i ∈Y, where n is the number of trading accounts, and the feature space is X, where X={X1,X2,…,X…} d }, X i ={x i,k |k=1,2,…,n}, representing the feature x i The feature set of x, where x i,k Let x represent the feature corresponding to the k-th transaction account. i The value of .

2. The polynomial tag generation method for risk profile construction according to claim 1, characterized in that: in, In step S4, the calculation expression for the polynomial feature set is: X polyfea =poly(X selected ), x=[x1,x2,…,x n ], In the formula X polyfea Let X be the set of polynomial characteristics. selected To filter the feature set, x n Let x be the nth feature in the feature set.

3. The polynomial tag generation method for risk profile construction according to claim 1, characterized in that: in, In step S3, the calculation expression for the filter feature set is: X j , selected =X j , corr ∪X j , lgbm , In the formula X selected To filter the feature set, m is the total number of sub-feature sets, X j,selected Let X be the selection set corresponding to the sub-feature set j. j,corr To filter the feature set corresponding to the relevance of sub-feature set j, X j,lgbm To select the feature set corresponding to the importance of sub-feature set j, Sort all features in sub-feature set j according to their relevance values ​​from largest to smallest, and select the number of features selected from the beginning of the sequence to construct the relevance filtering feature set. Sort all features in sub-feature set j in descending order of importance value, and select the number of features selected from the beginning of the sequence to construct the importance filtering feature set.

4. The polynomial tag generation method for risk profile construction according to claim 1, characterized in that: in, In step S2, the expression for calculating the correlation value is: In the formula Let i be the correlation value of feature i in feature set j. Let i be the value of feature i corresponding to the k-th target in feature set j. Let y be the mean of feature i in feature set j. k The risk assessment label corresponding to the k-th target. The average of the risk assessment labels for all targets is denoted by n, where n is the total number of targets.

5. The polynomial tag generation method for risk profile construction according to claim 1, characterized in that: in, In step S2, the corresponding importance value is calculated based on the information gain of the feature at the node of the tree in the LightGBM model. The expression for calculating the importance value is: In the formula Let be the importance value of feature i in feature set j, p be the total number of features in feature set j, M be the total number of trees in the LightGBM model, and N be the total number of trees in the model. m Let m be the set of nodes of the m-th tree. For the indicator function, ΔL m,n Let i be the information gain of feature i at node n in the m-th tree. When node n is split using feature i When node n does not use feature i for splitting 6. The polynomial tag generation method for risk profile construction according to claim 1, characterized in that: in, In step S2, the expression for calculating the feature selection quantity is: In the formula, select_len j Let be the feature selection quantity corresponding to feature set j, and p be the total number of features in feature set j. This is a floor operation.

7. The polynomial tag generation method for risk profile construction according to claim 1, Its features are: Step S7 includes the following sub-steps: Step S7-1: For the polynomial screening feature set, generate multiple candidate split thresholds and calculate the split value corresponding to each candidate split threshold; Step S7-2: The candidate splitting threshold corresponding to the largest splitting value is taken as the optimal splitting point; Step S7-3: Repeat steps S7-1 to S7-2 until the number of optimal split points reaches a preset value; Step S7-4: Calculate the bin boundaries based on all the optimal split points; Step S7-5: Based on the binning boundary, calculate the bins value and bin0 value of each feature corresponding to each target in the polynomial screening feature set as the binning value; Step S7-6: Based on all the binning values, calculate the woe value of each feature corresponding to each target in the polynomial screening feature set.

8. The polynomial tag generation method for risk profile construction according to claim 7, characterized in that: in, The expression for calculating the splitting value is: H(D ps )=-Σ c∈C p(c)log2p(c), In the formula IG(D) ps ,θ) is the splitting value of the candidate splitting threshold θ corresponding to the polynomial screening feature set, and |D| is the number of samples in D. v represents the value of feature i corresponding to target k in the multinomial filtering feature set. k For target k, y k Let C be the risk assessment label corresponding to target k, C be the total number of features in the multinomial screening feature set, and p(c) be the probability of feature c in the multinomial screening feature set. The calculation expression for the bin boundary is: In the formula for The set of bin boundary values, To sort all optimal split points in ascending order. The calculation expressions for the bins value and the bin0 value are as follows: In the formula The value of feature i corresponding to target k in the multinomial-selected feature set. for The left boundary value of the u-th bin interval.

9. The polynomial tag generation method for risk profile construction according to claim 7, characterized in that: in, In steps S7-6, based on the bin0 value of the feature corresponding to the target, the values ​​of all the features corresponding to the target are divided into multiple groups. The value of the feature corresponding to all the targets in the group has the same bin0 value. The risk assessment label is a binary label. The expression for calculating the woe value is: In the formula Let Woe be the value of feature i corresponding to target k in the polynomial-selected feature set, and N be the value of Woe. bad Let N be the total number of risk judgment labels corresponding to all values ​​of feature i in the group containing the target k in the polynomial screening feature set, where the label is of the bad category. goad Let N be the total number of risk judgment labels corresponding to all values ​​of feature i in the group containing the target k in the polynomial screening feature set, and all of which are classified as "good". total Let i be the total number of values ​​of feature i corresponding to target k in the polynomial selection feature set.

10. A multinomial label generation device for constructing risk profiles, used to generate label results related to the risk profile for each target based on risk-related data, wherein the risk-related data includes values ​​of multiple different features corresponding to each target, and risk judgment labels corresponding to each target, characterized in that, include: The subset partitioning module divides all the features into multiple categories and generates a sub-feature set corresponding to each category. The sub-feature set includes at least one feature and all the values ​​corresponding to that feature. The filtering module calculates the importance value, relevance value, and feature filtering quantity of each feature based on the number of features in the input feature set and the corresponding values ​​of the features. The filter set generation module selects multiple features from each of the sub-feature sets according to the importance value, the relevance value, and the feature selection quantity corresponding to the sub-feature sets to construct a filter feature set; The polynomial cross module performs polynomial cross feature calculation on the features in the selected feature set to obtain a polynomial feature set; The polynomial filtering set generation module selects multiple features from the polynomial feature set according to the corresponding importance value, the relevance value and the feature filtering quantity to construct a polynomial filtering feature set; The WOE calculation module performs binning calculations on the polynomial screening feature set to obtain the binning value of each feature corresponding to each target in the polynomial screening feature set. Based on all the binning values ​​corresponding to the polynomial screening feature set and the risk assessment label, it calculates the WOE value of each feature corresponding to each target in the polynomial screening feature set. Wherein, the label result of the target is each of the features of the target in the polynomial filtering feature set and its corresponding WOE value. The multinomial tag generation device for risk profiling is applied in the financial field. The risk-related data is financial transaction data D = [VXY], where V = {v1,...,v...} n } represents a set of trading accounts, where each trading account is the target. For each v i Each ∈V has a binary label indicating whether it is illegal, i.e., a separator label y. i ∈Y, where n is the number of trading accounts, and the feature space is X, where X={X1,X2,…,X…} d }, X i ={x i,k |k=1,2,…,n}, representing the feature x i The feature set of x, where x i,k Let x represent the feature corresponding to the k-th transaction account. i The value of .

Images