Feature combination screening method and device based on three-section funnel type and medium

Abstract

The invention relates to the technical field of data mining, and discloses a three-section funnel type-based feature combination screening method and device and a medium, and the method comprises the steps: carrying out the feature combination of original risk control features according to a combination operator, and carrying out the distributed storage of the combined features, calculating a correlation coefficient value between each combined feature after distributed storage and the business target, and performing first funnel screening on the combined features according to the correlation coefficient value; freezing a tree structure of the constructed service tree model, and splitting the frozen service tree model one by one by using each screened combination feature; calculating an optimization index of the split service tree model, and performing second funnel screening on each screened combination feature according to the optimization index; and calculating an importance value of each combination feature after the second screening by using the business tree model, and performing third funnel screening according to the importance value to obtain a target feature combination conforming to the business target. According to the invention, the efficiency of feature screening can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a feature combination screening method, equipment and medium based on a three-stage funnel. Background Art

[0002] With the rapid development of Internet finance, financial services such as mobile payment and wealth management have become an indispensable part of people's daily lives. In financial credit business, an intelligent risk control system is the cornerstone of the Internet financial field, and how to mine valuable information from data is particularly important. Putting as much feature information as possible into model training can allow the model to mine more information, which is conducive to improving the accuracy of the final result. However, as more features are added, the mutual influence between features and the complexity of the model are also increasing, which greatly hinders the improvement of model effects. Therefore, it is necessary to screen a large number of features to achieve the same business indicator effect as data mining experts.

[0003] The existing feature screening is to directly calculate the importance of each feature, and then implement feature screening based on importance. In practical applications, with the growth of business and the increase of feature collection, data mining experts usually need to face thousands of features, and there are also the feature magnitudes brought by the combination of these features in pairs or other combinations. Simply calculating the importance of each feature will result in too much calculation, making the feature screening inefficient. Summary of the invention

[0004] The present invention provides a feature combination screening method, equipment and medium based on a three-stage funnel, the main purpose of which is to solve the problem of low efficiency in feature screening.

[0005] To achieve the above object, the present invention provides a feature combination screening method based on a three-stage funnel, comprising: Acquire original risk control features corresponding to the risk control business scenario, and perform feature combination on the original risk control features according to a pre-generated combination operator to obtain a combined feature set; Distributed storage is performed on the combined feature set, a correlation coefficient value between each combined feature in the distributed stored combined feature set and a preset business goal is calculated, and a first funnel screening is performed on the combined features according to the correlation coefficient value to obtain a first screened combined feature; Constructing a business tree model according to the original risk control features, freezing the tree structure of the business tree model, adding each combined feature in the first screening combined features to the frozen business tree model one by one, and splitting the frozen business tree model with the added combined features; Calculating an optimization index of the split service tree model, and performing a second funnel screening on each combination feature in the first screening combination features according to the optimization index to obtain a second screening combination feature; The business tree model is used to calculate the importance value of each combination feature in the second screening combination feature, and the second screening combination feature is subjected to a third funnel screening according to the importance value to obtain a target feature combination that meets the business goal.

[0006] Optionally, combining the original risk control features according to the pre-generated combination operator to obtain the combined features includes: Dividing the original risk control features into continuous features and discrete features; Divide the pre-generated combinatorial operators into continuous feature operators and discrete feature operators; Combining the continuous features according to the operator quantity of the continuous feature operator and the continuous feature operator to obtain a continuous combination feature; Combining the discrete features according to the number of operators of the discrete feature operators and the discrete feature operators to obtain discrete combination features; The continuous combination features and the discrete combination features are collected into a combination feature set.

[0007] Optionally, the distributed storage of the combined feature set includes: Extracting a single feature corresponding to each combined feature in the combined feature set, and storing the single feature in a preset database list; Extract the single feature corresponding to the preset business requirement data; The combined feature of the business demand data is determined according to the single feature, and the combined feature is stored in a predefined storage location.

[0008] Optionally, the calculating of the correlation coefficient value between each combined feature in the distributed stored combined feature set and a preset business goal includes: Extract the feature value corresponding to each combined feature in the combined feature set after distributed storage; Determine target values ​​based on preset business goals; The correlation coefficient values ​​between the characteristic values ​​and the target values ​​are calculated one by one.

[0009] Optionally, splitting the frozen business tree model with the combined feature added includes: Dividing the combined features into target numerical features and target category features; Inputting the target numerical feature into the frozen service tree model, and splitting the target numerical feature in the frozen service tree model by a preset splitting threshold; Merge the split target numerical features into a numerical feature split set; Inputting the target category features into the frozen business tree model, and splitting the target category features in the frozen business tree model by a preset splitting category; The split target category features are merged into the category feature split set.

[0010] Optionally, the calculating of the optimization index of the split service tree model includes: Counting the number of true positive examples and the number of false negative examples of the split service tree model, and calculating the true positive rate according to the number of true positive examples and the number of false negative examples; Counting the number of false positive examples and the number of true negative examples of the business tree model after the split, and calculating the false positive rate according to the number of false positive examples and the number of true negative examples; Generate a characteristic curve according to the true positive rate and the false positive rate; The curve integral corresponding to the characteristic curve is calculated, and the integral value of the curve integral is determined as the optimization index of the classified business tree model.

[0011] Optionally, performing a second funnel screening on each combination feature in the first screening combination features according to the optimization index to obtain a second screening combination feature includes: Determine a target combination feature in the first screening combination feature according to a leaf node in the business tree model; Determining the optimization index of the business tree model as the optimization index corresponding to the target combination feature; When the optimization index corresponding to the target combination feature is a positive index, saving the target combination feature; When the optimization index corresponding to the target combination feature is a negative index, deleting the target combination feature; The target combination features after aggregation and storage are the second screening combination features.

[0012] Optionally, the calculating the importance value of each combination feature in the second screening combination feature by using the service tree model includes: Counting the number of splits of each combined feature in the second screening combined features in the business tree model, and determining the participation degree of each combined feature according to the number of splits; The participation degree is normalized, and the normalized participation degree is determined as the importance value of each combination feature in the second screening combination feature.

[0013] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned feature combination screening method based on the three-stage funnel type.

[0014] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the above-mentioned feature combination screening method based on the three-stage funnel type.

[0015] The embodiment of the present invention combines the original risk control features in N orders through a set of defined standard combination operators to form massive combination features, and performs distributed storage on the massive combination features. When calculation is required, the data of a single feature is taken for storage, thereby greatly reducing the storage requirements; and then the three-stage funnel is used to filter the features that are effective for the business, thereby achieving the effect of automatically improving the business model. In the same business scenario, the MOB6 (MOB6 is an indicator used in the financial industry to describe the sixth consecutive overdue period of a customer's loan or credit card account, mainly used to assess the customer's credit risk level) overdue ratio model is relatively improved by 1900% compared with the data mining expert, and the final business indicator AUC is relatively improved by 2%. Therefore, the feature combination screening method, device and medium based on the three-stage funnel type proposed in the present invention can solve the problem of low efficiency when performing feature screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a feature combination screening method based on a three-stage funnel type provided in one embodiment of the present invention; Figure 2 A schematic diagram of a feature combination provided by an embodiment of the present invention; Figure 3 A schematic diagram of distributed storage provided by an embodiment of the present invention; Figure 4 A schematic diagram of a correlation coefficient value provided by an embodiment of the present invention; Figure 5 A schematic diagram of frozen model improvement provided by an embodiment of the present invention; Figure 6 A schematic diagram of feature importance analysis provided by an embodiment of the present invention; Figure 7A schematic diagram of a feature screening process provided by an embodiment of the present invention; Figure 8 A schematic diagram of the structure of an electronic device for implementing the three-stage funnel-based feature combination screening method provided in one embodiment of the present invention.

[0017] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0018] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0019] The embodiment of the present application provides a feature combination screening method based on a three-stage funnel type. The execution subject of the feature combination screening method based on the three-stage funnel type includes but is not limited to at least one of the electronic devices that can be configured to execute the method provided in the embodiment of the present application, such as a server, a terminal, etc. In other words, the feature combination screening method based on the three-stage funnel type can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0020] Reference Figure 1 FIG. 1 is a flow chart of a feature combination screening method based on a three-stage funnel type according to an embodiment of the present invention. In this embodiment, the feature combination screening method based on a three-stage funnel type includes: S1. Obtain original risk control features corresponding to the risk control business scenario, and perform feature combination on the original risk control features according to a pre-generated combination operator to obtain a combined feature set.

[0021] In the embodiment of the present invention, the risk control business scenario refers to the scenario constructed for specific risk control needs and goals in the risk control process, such as credit risk scenario and anti-fraud scenario, and the original risk control feature refers to the basic information of users, user transaction data and user overdue bills involved in the risk control business scenario, wherein the basic information of users includes but is not limited to name, age and gender; the user transaction data includes but is not limited to transaction time, transaction amount and transaction frequency; the user overdue bill includes but is not limited to expected number of times and expected number of days, wherein the original risk control features corresponding to the risk control business scenario can be obtained from the pre-stored storage area through computer statements with data capture function (such as Java statements, Python statements, etc.), wherein the storage area includes but is not limited to database and blockchain. Furthermore, it is necessary to combine the existing risk control features and mine the effective business feature information from the massive combination features, and the current combination of thousands of original features, such as two-by-two combination or three-by-three combination, will make the magnitude of the features reach more than 100 million features, therefore, it is necessary to design a set of standard combination operators to standardize how to combine the features.

[0022] In an embodiment of the present invention, the combined features refer to operating and transforming the original risk control features to generate new features, so as to better capture the potential patterns in the data, that is, new features generated by performing mathematical operations, logical operations or other transformations on one or more original features.

[0023] In the embodiment of the present invention, the combining of the original risk control features according to the pre-generated combination operator to obtain the combined features includes: Dividing the original risk control features into continuous features and discrete features; Divide the pre-generated combinatorial operators into continuous feature operators and discrete feature operators; Combining the continuous features according to the operator quantity of the continuous feature operator and the continuous feature operator to obtain a continuous combination feature; Combining the discrete features according to the number of operators of the discrete feature operators and the discrete feature operators to obtain discrete combination features; The continuous combination features and the discrete combination features are collected into a combination feature set.

[0024] In detail, the original risk control features include continuous and discrete features, where continuous features refer to numerical features, such as age, income, credit score, etc.; discrete features refer to features that tend to have finite values, such as (male / female), region (urban / rural), etc., and then the generated combination operators are divided into continuous feature operators and discrete feature operators. The continuous feature operators include single feature operators (such as: log, sqrt) and multi-feature operators (such as: +, -), and the discrete feature operators include logical symbols (such as and, or) and functions (such as group by). Then, the original risk control features are combined according to the continuous feature operators and discrete feature operators, and the combination between features is controlled by order.

[0025] Specifically, the continuous features are combined by using the pre-generated continuous feature operators. For example, for features A and B, new features such as A+B and A / B can be generated, and then all the generated continuous combination features are collected. The discrete features are combined by using the pre-generated discrete feature operators. For example, for features C and D, new features such as C and D and C orD can be generated, and then all the generated discrete combination features are collected. Finally, all the generated continuous combination features and discrete combination features are collected into a combination feature set, such as Figure 2 As shown in the figure, it is a schematic diagram of feature combination. The combination of features is determined according to the number of combination operators. For example, the first-order combination is itself, that is, a combination operator, indicating ; The second-order combination is a combination of two features, that is, two combination operators, indicating ; The third-order combination is a combination of three features, that is, three combination operators, indicating ; The 4th-order combination is a combination of four features, that is, four combination operators, representing func (A, B, C).

[0026] Furthermore, the combined features will generate massive amounts of data, which is obviously difficult to store and compute in a single-machine environment and requires distributed storage. Therefore, the combined massive feature data needs to be distributedly stored to improve the efficiency of subsequent feature screening.

[0027] S2. Distribute and store the combined feature set, calculate the correlation coefficient value between each combined feature in the distributed stored combined feature set and the preset business goal, and perform a first funnel screening on the combined features according to the correlation coefficient value to obtain a first screened combined feature.

[0028] In an embodiment of the present invention, the combined feature set is stored in a distributed manner, that is, single feature storage and combined feature storage are adopted. That is, the combined data is not directly stored, but the single feature data is taken for storage when calculation is needed, thereby greatly reducing the storage requirements.

[0029] In the embodiment of the present invention, the distributed storage of the combined feature set includes: Extracting a single feature corresponding to each combined feature in the combined feature set, and storing the single feature in a preset database list; Extract the single feature corresponding to the preset business requirement data; Determine the combined feature of the business demand data according to the single feature, and store the combined feature in a predefined storage location. In detail, a single feature corresponding to each combined feature in the combined feature set is extracted, that is, each combined feature is decomposed into multiple separate features, and the extracted single features are stored in a database list corresponding to the feature field, and the combined features corresponding to the single features are stored in a predefined storage location, that is, each combined feature in the combined feature set is also a column feature, and in the matrix, each column represents digital data corresponding to a combined feature, that is, each combined feature is stored according to the column dimension.

[0030] For example, assuming that the data stored on a unit machine is massive and difficult to store, single feature storage and combined feature storage are used, that is, the combined data is not directly stored, but the single feature data is taken for storage when calculation is needed. In the data storage, the basic features of the data are stored separately. For example, in an e-commerce user data set, the user's age, gender, number of purchases, consumption amount, etc. are all different single features. Each single feature will be stored in the corresponding database field or data file column. The combined feature storage does not pre-calculate or pre-store the combined features generated by combining multiple single features, but stores the formulas for generating these combined features. For example, in an e-commerce scenario, if you want to obtain the combined feature of user consumption intensity, it may be defined as consumption amount / number of purchases. At this time, the consumption intensity of each user will not be calculated in advance and stored, but the formula for calculating this consumption intensity will be stored. When the business needs to use the consumption intensity data, it will obtain data from the place where the single features (consumption amount and number of purchases) are stored in real time according to this formula, and calculate the result, such as Figure 3 As shown in the figure, it is a schematic diagram of distributed storage, where each combination formula is stored separately in one location according to the column dimension, such as Stored in a column, Stored in a column, Stored in one column, thus improving the efficiency of subsequent feature screening.

[0031] Furthermore, since a large number of combined features will be generated in the process of combining features, it is obviously impossible to calculate if the traditional tree model feature importance is used for screening. Therefore, it is necessary to quickly screen the features first, eliminate some irrelevant features, and then perform effectiveness screening to obtain the effective features themselves. The first step is to eliminate the correlation of the features.

[0032] In an embodiment of the present invention, during the screening process, correlation elimination is first performed on a large number of derived features, that is, the first-level funnel correlation elimination. That is, this stage is faced with a large number of features, and a method with fast calculation speed is required to quickly eliminate invalid features. Therefore, this method of selecting the correlation between features and business objectives is used for fast calculation elimination, and the Pearson correlation coefficient is used for fast calculation elimination. The correlation coefficient value is a statistical indicator used to measure the strength and direction of the linear relationship between two variables.

[0033] In the embodiment of the present invention, the calculation of the correlation coefficient value between each combined feature in the distributed stored combined feature set and the preset business goal includes: Extract the feature value corresponding to each combined feature in the combined feature set after distributed storage; Determine target values ​​based on preset business goals; The correlation coefficient values ​​between the characteristic values ​​and the target values ​​are calculated one by one.

[0034] In detail, the Pearson correlation coefficient is used to calculate the Pearson correlation coefficient between each combined feature and the business goal, where the business goal can be configured differently according to different scenarios. For example, the corresponding business goal in the anti-fraud business scenario is to determine whether it is anti-fraud, 0 means no anti-fraud, and 1 means anti-fraud. For example, the corresponding business goal in the user overdue business scenario is to determine whether the user is overdue, 0 means no overdue, and 1 means overdue. The business goal is a sample label constructed based on the business scenario. , and then according to the eigenvalue and sample labels Calculate the correlation coefficient value between each combined feature and the business goal.

[0035] Specifically, the correlation coefficient calculation formula is: ,in, express and The correlation coefficient of express and The covariance of represents the expected value, express The variance (standard deviation) of express The variance of express The mean of express The mean of express and The expected value of the product of express The expected value of express The expected value of express The expected value of the square of express The expected value of the square of express The square of the expected value of express The square of the expected value of represents the eigenvalue corresponding to the combined feature, It indicates the target value corresponding to the business goal, and 0.8-1.0 indicates extremely strong correlation; 0.6-0.8 indicates strong correlation; 0.4-0.6 indicates moderate correlation; 0.2-0.4 indicates weak correlation; and 0.0-0.2 indicates extremely weak correlation or no correlation.

[0036] Further, the combination features are screened by a first funnel according to the correlation coefficient value to obtain the first screening combination features, that is, the combination features with the calculated correlation coefficient value less than 0.2 are eliminated, and the combination features with the correlation coefficient value greater than 0.2 are retained, so that the retained combination features are aggregated into the first screening combination features, such as Figure 4 As shown in the figure, it is a schematic diagram of the correlation coefficient value. The corresponding data of the combined feature f1 is 11231. The value is calculated to obtain the combined feature f1 and The correlation coefficient between them is 1.0, and the combined feature f2 and The correlation coefficient between them is 0.37, and the combined feature f3 and If the correlation coefficient value between them is 0.1, the combination feature f3 is eliminated, and the combination features f1 and f2 are retained.

[0037] Furthermore, even after the first layer of elimination, the number of features is still huge. Therefore, a second layer of funnel is needed for filtering. Therefore, the lifting method is used, that is, the process needs to consider both the calculation speed and the effect.

[0038] S3. Build a business tree model according to the original risk control features, freeze the tree structure of the business tree model, add each combination feature in the first screening combination feature to the frozen business tree model one by one, and split the frozen business tree model with the added combination features.

[0039] In an embodiment of the present invention, the business tree model refers to a model generated based on a business scenario and is used for risk control. It helps decision makers understand complex business processes and decision rules through a tree structure, and facilitates the identification of key factors and their impact on the final decision.

[0040] In detail, first, the feature data corresponding to the original risk control features are cleaned, that is, missing values, outliers and duplicate data are processed to ensure the integrity and accuracy of the data. Then the business tree structure is defined, and the goal of building the business tree model is clearly defined, such as identifying high-risk users. Then, the feature data is selected as the nodes of the tree according to the business logic, and the data is divided according to the feature value of each node. If the credit score is >700, it is divided into "low risk", and the credit score ≤ 700 and the income is >5000, it is divided into "medium risk", otherwise it is "high risk". Then, the feature data corresponding to the original risk control features are divided into training sets and test sets. The training set is used to build the business tree model, and the test set is used to adjust the parameters to improve the model performance.

[0041] Furthermore, a business tree model is trained based on the original online business data, and then the business tree model is frozen, that is, the tree structure is not changed subsequently, and the decision conditions of all nodes in the business tree model, the connection relationship between nodes, and the results of leaf nodes are fixed. No matter what data is input subsequently, the business tree model is processed according to the tree results of the frozen business tree model. Therefore, each combination feature in the first screening combination feature is added one by one to the frozen business tree model, that is, for each combination feature, the business tree model is added, and then only the combination feature is continued to be trained, so as to obtain the evaluation optimization index (such as AUC) in the training process. The other parts of the business tree model training are frozen, and only a single combination feature can participate in the training, so the calculation speed is fast, and because it is directly related to the business indicator, it also takes into account certain effects.

[0042] In an embodiment of the present invention, in the process of performing the second-level funnel to screen the combined features, both the calculation speed and the effect need to be considered. Therefore, the business tree model generated based on the original business data needs to be frozen, and the lifting method needs to be used to evaluate the effect of each combined feature.

[0043] In the embodiment of the present invention, the step of splitting the frozen service tree model to which the combined feature is added includes: Dividing the combined features into target numerical features and target category features; Inputting the target numerical feature into the frozen service tree model, and splitting the target numerical feature in the frozen service tree model by a preset splitting threshold; Merge the split target numerical features into a numerical feature split set; Inputting the target category features into the frozen business tree model, and splitting the target category features in the frozen business tree model by a preset splitting category; The split target category features are merged into the category feature split set.

[0044] In detail, each combined feature in the first filtered combined features after the first layer of filtering is divided into a target numerical feature and a target category feature, that is, the target numerical feature is continuous data, and the target category feature is discrete data. Then, for each target numerical feature, it is input into the frozen business data model one by one, that is, the target numerical feature is input into the business tree model one by one. According to business needs or historical data, a preset splitting threshold is set. For example, income greater than 5,000 yuan and less than or equal to 5,000 yuan can be used as a splitting threshold. The frozen business tree is split using the splitting threshold. This process will generate new child nodes, representing the classification results generated according to different feature values, that is, merged into a numerical feature split set, and the numerical features with the same classification result are merged into one set; for each target category feature, it is input into the frozen business tree model one by one for splitting, that is, the target category features are input into the business tree model one by one to determine the splitting rules of different categories. For example, gender can be divided into "male" and "female", and regions can be classified according to different geographical locations. Splits can be performed according to the set categories to generate new child nodes, representing decision results of different categories, that is, merged into category feature split sets, and category features with the same classification results are merged into one set.

[0045] Specifically, the frozen business tree model can be updated according to the new feature combination, and the final result will provide a more refined classification, that is, only continue to train each combination feature, and the optimization index of each combination feature during the training process, such as Figure 5 As shown, it is a schematic diagram of frozen model improvement. First, a business tree model is built based on the original feature sample data, and then the business tree model is frozen. Each combined feature is input one by one. The combined feature f1 is input into the frozen business tree model, and the optimization index obtained is -0.67. The combined feature f2 is input into the frozen business tree model, and the optimization index obtained is 0.75. The combined feature f3 is input into the frozen business tree model, and the optimization index obtained is 0.75.

[0046] Furthermore, a second funnel screening is performed on the combined features based on the optimization index, so that the magnitude of the screened features is within a controllable range.

[0047] S4. Calculate the optimization index of the split business tree model, and perform a second funnel screening on each combination feature in the first screening combination features according to the optimization index to obtain a second screening combination feature.

[0048] In the embodiment of the present invention, the optimization index refers to a numerical index used to evaluate the model effect in the process of training the business tree model for each combined feature, which helps analyze the performance of the model in a specific task for comparison, adjustment and improvement.

[0049] In the embodiment of the present invention, the optimization index of the business tree model after the calculation of the split includes: Counting the number of true positive examples and the number of false negative examples of the split service tree model, and calculating the true positive rate according to the number of true positive examples and the number of false negative examples; Counting the number of false positive examples and the number of true negative examples of the business tree model after the split, and calculating the false positive rate according to the number of false positive examples and the number of true negative examples; Generate a characteristic curve according to the true positive rate and the false positive rate; The curve integral corresponding to the characteristic curve is calculated, and the integral value of the curve integral is determined as the optimization index of the classified business tree model.

[0050] In detail, firstly, the instance data of each combination feature predicted as positive by the business tree model during the training process, i.e., the true positive (TP) and the number of instances predicted as negative by the business tree model but actually positive, i.e., the false negative (FN), are counted, and the number of instances incorrectly predicted as positive by the model but actually negative, i.e., the false positive (FP) and the number of instances correctly predicted as negative by the model, i.e., the true negative (TN), are counted, and the false positive rate is FP / (FP + TN), and then a characteristic curve ROC (Receiver Operating Characteristic Curve) is drawn with the false positive rate as the X-axis and the true positive rate as the Y-axis, that is, the corresponding relationship between the false positive rate and the true positive rate corresponding to each combination feature is plotted as coordinate points, so as to generate a characteristic curve according to the coordinate points, and then the AUC (Area Under the curve) is calculated according to the area under the curve corresponding to the characteristic curve. Curve, area under the curve), that is, the area under the curve is calculated by a numerical integration method (such as the trapezoidal rule), that is, the curve integral of the characteristic curve is calculated, and the integral value corresponding to the curve integral is determined as the optimization index of the classified business tree model, so as to measure the classification performance of the business tree model according to the calculated optimization index (such as AUC). The value range of AUC is between 0 and 1. The closer the value is to 1, the better the model performance.

[0051] Furthermore, the combined features after the first-level funnel screening are screened for the second time according to the optimization index to ensure that the order of magnitude of the remaining features is usually within a controllable range.

[0052] In the embodiment of the present invention, the second screening combination features refers to further screening and optimization of the feature combination screened out in the first layer to improve the performance and accuracy of the model.

[0053] In the embodiment of the present invention, performing a second funnel screening on each combination feature in the first screening combination feature according to the optimization index to obtain a second screening combination feature includes: Determine a target combination feature in the first screening combination feature according to a leaf node in the business tree model; Determining the optimization index of the business tree model as the optimization index corresponding to the target combination feature; When the optimization index corresponding to the target combination feature is a positive index, saving the target combination feature; When the optimization index corresponding to the target combination feature is a negative index, deleting the target combination feature; The target combination features after aggregation and storage are the second screening combination features.

[0054] In detail, a business tree model is trained based on the original online business data, and then the business tree model is frozen, that is, no subsequent changes are made to the tree structure. Next, each combination feature is added to the tree model, and then only the combination feature is continued to be trained to see the optimization index of the training process, that is, the business node of the business tree model is determined as the target combination feature to be added, and the optimization index corresponding to the combination feature is determined according to the added target combination feature. If the optimization index is positive, it indicates that this combination feature is effective for the business, that is, the combination feature is retained. If it is negative, it indicates that this combination feature is invalid for the business, that is, the combination feature is eliminated, and then all the retained combination features are gathered to form a second screening combination feature, where the optimization index includes but is not limited to AUC (Area Under the Curve).

[0055] For example, in Figure 5 In the figure, the optimization index corresponding to the combination feature f1 is -0.67, the optimization index corresponding to the combination feature f2 is 0.75, and the optimization index corresponding to the combination feature f3 is 0.75. If the optimization index corresponding to the initial business tree model is 0.7, the optimization index corresponding to the combination feature f1 is negative, and the combination feature f1 is eliminated. The optimization indexes corresponding to the combination features f2 and f3 are positive, and the combination features f2 and f3 are retained. Compared with the original business tree model, the combination feature f2 is improved by 0.05, and the combination feature f3 is also improved by 0.05 compared with the original business tree model.

[0056] Furthermore, if the feature magnitude of the second screening combination features obtained after the second funnel screening is within a controllable range, it is necessary to use a business tree model related to the business objectives to perform feature importance analysis to obtain relatively important feature data.

[0057] S5. Calculate the importance value of each combination feature in the second screening combination feature using the business tree model, and perform a third funnel screening on the second screening combination feature according to the importance value to obtain a target feature combination that meets the business goal.

[0058] In an embodiment of the present invention, the importance value is an indicator used to measure the contribution of a feature in the business tree model, indicating the influence of a certain feature on the model decision, usually expressed in numerical form. The higher the value, the more important the feature is in the business tree model.

[0059] In the embodiment of the present invention, the calculating the importance value of each combination feature in the second screening combination feature by using the service tree model includes: Counting the number of splits of each combined feature in the second screening combined features in the business tree model, and determining the participation degree of each combined feature according to the number of splits; The participation degree is normalized, and the normalized participation degree is determined as the importance value of each combination feature in the second screening combination feature.

[0060] Specifically, during the model training process, the number of splits of each combined feature in the business tree model is recorded, and the number of splits is determined as the participation of each combined feature. In order to compare the importance of different features, the participation needs to be normalized, and the participation of each feature is divided by the sum of the participations of all features, so that the normalized participation is determined as the importance value of each combined feature in the second screening combined feature, such as Figure 6 As shown, it is a schematic diagram of feature importance analysis. The importance of each combined feature is determined based on the feature sample data. For example, the importance value corresponding to the combined feature f1 is 0.8, the importance value corresponding to the combined feature f2 is 0.6, and the importance value corresponding to the combined feature f3 is 0.01.

[0061] Specifically, for the importance value corresponding to each combined feature calculated, all combined features are sorted according to the importance value in descending order to obtain a feature sequence corresponding to the combined feature, and the first k features are selected in the feature sequence as the target feature combination that meets the business goal.

[0062] Furthermore, if Figure 7 As shown in the figure, it is a schematic diagram of the feature screening process, including the feature combination process and the feature screening process. For example, there are 1000 original features, that is, N=1000, and the additive combination method can be used to obtain About 500,000 derived features are obtained based on the screening strategy, and finally about dozens of features are obtained, so that the order of magnitude of the features finally screened is within a controllable range.

[0063] The embodiment of the present invention combines the original risk control features in N stages through a set of defined standard combination operators to form massive combination features, and performs distributed storage on the massive combination features. When calculation is required, the data of a single feature is taken for storage, thereby greatly reducing the storage requirements; and then the three-stage funnel is used to filter the features that are effective for the business, thereby achieving the effect of automatically improving the business model. In the same business scenario MOB6 overdue ratio model, the efficiency is relatively improved by 1900% compared with the data mining expert, and the final business indicator AUC is relatively improved by 2%. Therefore, the feature combination screening method, device and medium based on the three-stage funnel type proposed in the present invention can solve the problem of low efficiency when performing feature screening.

[0064] like Figure 8 , is a schematic diagram of the structure of an electronic device for implementing a feature combination screening method based on a three-stage funnel type provided by an embodiment of the present invention.

[0065] The electronic device may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a feature combination screening program based on a three-stage funnel type.

[0066] In some embodiments, the processor 10 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions and processes data of the electronic device by running or executing programs or modules stored in the memory 11 (for example, executing a feature combination screening program based on a three-stage funnel type, etc.), and calling data stored in the memory 11.

[0067] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the memory 11 may also be an external storage device of an electronic device, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 11 may also include both an internal storage unit of the electronic device and an external storage device. The memory 11 may not only be used to store application software and various types of data installed in the electronic device, such as the code of a feature combination screening program based on a three-stage funnel type, but may also be used to temporarily store data that has been output or is to be output.

[0068] The communication bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0069] The communication interface 13 is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface may also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device and to display a visual user interface.

[0070] The figure only shows an electronic device with components. Those skilled in the art will understand that the structure shown in the figure does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0071] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0072] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0073] The feature combination screening program based on the three-stage funnel type stored in the memory 11 of the electronic device is a combination of multiple instructions. When running in the processor 10, it can achieve: Acquire original risk control features corresponding to the risk control business scenario, and perform feature combination on the original risk control features according to a pre-generated combination operator to obtain a combined feature set; Distributed storage is performed on the combined feature set, a correlation coefficient value between each combined feature in the distributed stored combined feature set and a preset business goal is calculated, and a first funnel screening is performed on the combined features according to the correlation coefficient value to obtain a first screened combined feature; Constructing a business tree model according to the original risk control features, freezing the tree structure of the business tree model, adding each combined feature in the first screening combined features to the frozen business tree model one by one, and splitting the frozen business tree model with the added combined features; Calculating an optimization index of the split service tree model, and performing a second funnel screening on each combination feature in the first screening combination features according to the optimization index to obtain a second screening combination feature; The business tree model is used to calculate the importance value of each combination feature in the second screening combination feature, and the second screening combination feature is subjected to a third funnel screening according to the importance value to obtain a target feature combination that meets the business goal.

[0074] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, which will not be repeated here.

[0075] Furthermore, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0076] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement: Acquire original risk control features corresponding to the risk control business scenario, and perform feature combination on the original risk control features according to a pre-generated combination operator to obtain a combined feature set; Distributed storage is performed on the combined feature set, a correlation coefficient value between each combined feature in the distributed stored combined feature set and a preset business goal is calculated, and a first funnel screening is performed on the combined features according to the correlation coefficient value to obtain a first screened combined feature; Constructing a business tree model according to the original risk control features, freezing the tree structure of the business tree model, adding each combined feature in the first screening combined features to the frozen business tree model one by one, and splitting the frozen business tree model with the added combined features; Calculating an optimization index of the split service tree model, and performing a second funnel screening on each combination feature in the first screening combination features according to the optimization index to obtain a second screening combination feature; The business tree model is used to calculate the importance value of each combination feature in the second screening combination feature, and the second screening combination feature is subjected to a third funnel screening according to the importance value to obtain a target feature combination that meets the business goal.

[0077] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, media and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0078] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0079] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0080] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0081] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is not limited only according to the above description, and it is intended that all changes within the meaning and scope of equivalent elements within the scope of protection are included in the present invention.

[0082] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0083] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A feature combination screening method based on a three-stage funnel, characterized in that: The method comprises: Acquire original risk control features corresponding to the risk control business scenario, and perform feature combination on the original risk control features according to a pre-generated combination operator to obtain a combined feature set; Distributed storage is performed on the combined feature set, a correlation coefficient value between each combined feature in the distributed stored combined feature set and a preset business goal is calculated, and a first funnel screening is performed on the combined features according to the correlation coefficient value to obtain a first screened combined feature; Constructing a business tree model according to the original risk control features, freezing the tree structure of the business tree model, adding each combined feature in the first screening combined features to the frozen business tree model one by one, and splitting the frozen business tree model with the added combined features; Calculating an optimization index of the split service tree model, and performing a second funnel screening on each combination feature in the first screening combination features according to the optimization index to obtain a second screening combination feature; The business tree model is used to calculate the importance value of each combination feature in the second screening combination feature, and the second screening combination feature is subjected to a third funnel screening according to the importance value to obtain a target feature combination that meets the business goal.

2. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The combining of the original risk control features according to the pre-generated combination operator to obtain a combined feature set includes: Dividing the original risk control features into continuous features and discrete features; Divide the pre-generated combinatorial operators into continuous feature operators and discrete feature operators; Combining the continuous features according to the operator quantity of the continuous feature operator and the continuous feature operator to obtain a continuous combination feature; Combining the discrete features according to the number of operators of the discrete feature operators and the discrete feature operators to obtain discrete combination features; The continuous combination features and the discrete combination features are collected into a combination feature set.

3. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The distributed storage of the combined feature set includes: Extracting a single feature corresponding to each combined feature in the combined feature set, and storing the single feature in a preset database list; Extract the single feature corresponding to the preset business requirement data; Determine the combined feature of the business demand data according to the single feature, and store the combined feature in a predefined storage location.

4. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The calculation of the correlation coefficient value between each combined feature in the distributed stored combined feature set and the preset business goal includes: Extract the feature value corresponding to each combined feature in the combined feature set after distributed storage; Determine target values ​​based on preset business goals; The correlation coefficient values ​​between the characteristic values ​​and the target values ​​are calculated one by one.

5. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The step of splitting the frozen business tree model to which the combined feature is added includes: Dividing each of the combined features into a target numerical feature and a target category feature; Inputting the target numerical feature into the frozen service tree model, and splitting the target numerical feature in the frozen service tree model by a preset splitting threshold; Merge the split target numerical features into a numerical feature split set; Inputting the target category features into the frozen business tree model, and splitting the target category features in the frozen business tree model by a preset splitting category; The split target category features are merged into the category feature split set.

6. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The optimization index of the business tree model after the calculation of the split includes: Counting the number of true positive examples and the number of false negative examples of the split service tree model, and calculating the true positive rate according to the number of true positive examples and the number of false negative examples; Counting the number of false positive examples and the number of true negative examples of the business tree model after the split, and calculating the false positive rate according to the number of false positive examples and the number of true negative examples; Generate a characteristic curve according to the true positive rate and the false positive rate; The curve integral corresponding to the characteristic curve is calculated, and the integral value of the curve integral is determined as the optimization index of the classified business tree model.

7. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The performing a second funnel screening on each combination feature in the first screening combination feature according to the optimization index to obtain a second screening combination feature includes: Determine a target combination feature in the first screening combination feature according to a leaf node in the business tree model; Determining the optimization index of the business tree model as the optimization index corresponding to the target combination feature; When the optimization index corresponding to the target combination feature is a positive index, saving the target combination feature; When the optimization index corresponding to the target combination feature is a negative index, deleting the target combination feature; The target combination features after aggregation and storage are the second screening combination features.

8. The feature combination screening method based on the three-stage funnel type according to claim 1, characterized in that: The calculating the importance value of each combination feature in the second screening combination feature by using the service tree model includes: Counting the number of splits of each combined feature in the second screening combined features in the business tree model, and determining the participation degree of each combined feature according to the number of splits; The participation degree is normalized, and the normalized participation degree is determined as the importance value of each combination feature in the second screening combination feature.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the feature combination screening method based on the three-stage funnel type as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the feature combination screening method based on the three-stage funnel type as described in any one of claims 1 to 8 is implemented.

Images