Information processing method, electronic device and computer program product

Abstract

The invention relates to an information processing method, electronic equipment and a computer program product. The method comprises the following steps: acquiring a risk credit training set and risk credit configuration information; configuring a tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT; performing machine learning on the risk credit training set through a target TPOT to obtain at least one candidate task flow; determining a target task flow from the at least one candidate task flow; and constructing a risk credit model based on the target task flow. By adopting the method, the information processing efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of AI (Artificial Intelligence) technology, and in particular to an information processing method, an electronic device, and a computer program product. Background Art

[0002] With the development of the economy and society, users can obtain risk credit from banks or financial institutions. For banks or financial institutions, it is a common practice to use risk credit models to understand users' credit and repayment conditions in order to provide credit limits more accurately.

[0003] For risk credit models, machine learning is usually used for modeling. In traditional machine learning, feature engineering is a very critical, time-consuming and labor-intensive step. It usually includes the following main tasks: Feature selection: From a large number of original features, manually select feature subsets that have a greater impact on the model prediction results. This requires a deep understanding of the business background and a large number of attempts at various feature combinations; Feature conversion: Perform various mathematical transformations on the original features, such as standardization, normalization, and discretization, to better adapt to the model assumptions, which requires a deep understanding of data characteristics; Feature combination: Based on domain knowledge, manually construct new composite features to explore the interactive relationship between features, which requires a lot of trial and error and iteration.

[0004] In other words, the traditional information processing method for modeling has the problem of low efficiency. Summary of the invention

[0005] The embodiments of the present application provide an information processing method, device, electronic device, computer-readable storage medium, and computer program product, which can improve information processing efficiency.

[0006] In a first aspect, the present application provides an information processing method, comprising:

[0007] Obtain risky credit training set and risky credit configuration information;

[0008] configuring a tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT;

[0009] By using the target TPOT, machine learning is performed on the risk credit training set to obtain at least one candidate task flow;

[0010] Determine a target task flow from the at least one candidate task flow;

[0011] Based on the target task flow, a risk credit model is constructed.

[0012] In one embodiment, the step of performing machine learning on the risk credit training set through the target TPOT to obtain at least one candidate task flow includes:

[0013] Through the target TPOT, various feature engineering components are combined to obtain the first task flow;

[0014] For each first task flow, performing machine learning on the risk credit training set to obtain a comprehensive score of the first task flow;

[0015] If the comprehensive score of the first task flow is greater than the score threshold, the first task flow is used as a candidate task flow.

[0016] In one embodiment, the risk credit training set includes sample features and target results; and for each first task flow, performing machine learning on the risk credit training set to obtain a comprehensive score of the first task flow includes:

[0017] For each first task flow, obtaining a correspondence between the sample feature and the target result, and predicting the sample feature based on the correspondence to obtain a prediction result;

[0018] Determining a first score based on the degree of difference between the predicted result and the target result;

[0019] Based on the sample features and the first score, a comprehensive score of the first task flow is determined.

[0020] In one embodiment, determining the comprehensive score of the first task flow based on the sample feature and the first score includes:

[0021] Determining a risk differentiation capability of the first task flow based on the sample characteristics and the first score;

[0022] determining a target monotonicity between the sample feature and the first score;

[0023] determining a target stability of the first score on the sample characteristics;

[0024] Based on at least one of the risk differentiation capability, the target monotonicity, and the target stability, a comprehensive score of the first task flow is determined.

[0025] In one embodiment, the step of performing machine learning on the risk credit training set through the target TPOT to obtain at least one candidate task flow includes:

[0026] Through the target TPOT, the parameters within the target search range are obtained;

[0027] According to the parameters within the target search range, machine learning is performed on the risk credit training set to obtain at least one candidate task flow.

[0028] In one embodiment, constructing a risk credit model based on the target task flow includes:

[0029] In response to the modification instruction, modify the target task flow to obtain a new target task flow;

[0030] The new target task flow is trained using the risk credit training set to obtain a risk credit model.

[0031] In one of the embodiments, after determining the target task flow from the at least one candidate task flow, the method further includes:

[0032] Converting the target task flow into visualization features;

[0033] According to the visualization features, evaluating the interpretability of the target task flow;

[0034] If the interpretability of the target task flow reaches a preset interpretability level, the step of constructing a risk credit model based on the target task flow is performed.

[0035] In one of the embodiments, after determining the target task flow from the at least one candidate task flow, the method further includes:

[0036] Get the risk credit validation set;

[0037] By using the target task flow, learning the risk credit verification set, and determining a second score;

[0038] Based on the second score, performing a model effect evaluation on the target task flow;

[0039] If the model effect evaluation is passed, the step of constructing a risk credit model based on the target task flow is executed.

[0040] In a second aspect, the present application further provides an information processing device, comprising:

[0041] A data acquisition module is used to obtain risk credit training sets and risk credit configuration information;

[0042] A TPOT configuration module, configured to configure the tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT;

[0043] A target task flow determination module is used to perform machine learning on the risk credit training set through a target TPOT to obtain at least one candidate task flow; and determine a target task flow from the at least one candidate task flow;

[0044] The risk credit model building module is used to build a risk credit model based on the target task flow.

[0045] In a third aspect, the present application further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0046] Obtain risky credit training set and risky credit configuration information;

[0047] configuring a tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT;

[0048] By using the target TPOT, machine learning is performed on the risk credit training set to obtain at least one candidate task flow;

[0049] Determine a target task flow from the at least one candidate task flow;

[0050] Based on the target task flow, a risk credit model is constructed.

[0051] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0052] Obtain risky credit training set and risky credit configuration information;

[0053] configuring a tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT;

[0054] By using the target TPOT, machine learning is performed on the risk credit training set to obtain at least one candidate task flow;

[0055] Determine a target task flow from the at least one candidate task flow;

[0056] Based on the target task flow, a risk credit model is constructed.

[0057] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0058] Obtain risky credit training set and risky credit configuration information;

[0059] configuring a tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT;

[0060] By using the target TPOT, machine learning is performed on the risk credit training set to obtain at least one candidate task flow;

[0061] Determine a target task flow from the at least one candidate task flow;

[0062] Based on the target task flow, a risk credit model is constructed.

[0063] The above-mentioned information processing method, device, electronic device, computer-readable storage medium and computer program product obtain risk credit training set and risk credit configuration information, configure the tree-based pipeline optimization tool TPOT based on the risk credit configuration information, and obtain the target TPOT. Then, through the target TPOT, it is possible to automatically perform machine learning on the risk credit training set, obtain at least one candidate task flow more quickly, and determine the target task flow from at least one candidate task flow, and then based on the target task flow, build a risk credit model more quickly, thereby improving the modeling and information processing efficiency of the risk credit model. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0065] Figure 1 is a flowchart of an information processing method in an embodiment;

[0066] Figure 2 is a schematic diagram of TPOT in one embodiment;

[0067] Figure 3 A schematic diagram of TPOT automated machine learning in one embodiment;

[0068] Figure 4 is a schematic diagram of a flow chart of a genetic algorithm in one embodiment;

[0069] Figure 5 is a structural block diagram of an information processing device in one embodiment;

[0070] Figure 6 FIG. 4 is a diagram showing the internal structure of an electronic device in one embodiment. DETAILED DESCRIPTION

[0071] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0072] In one embodiment, Figure 1 As shown, an information processing method is provided. This embodiment uses the method applied to an electronic device as an example. The electronic device can be a terminal or a server. It can be understood that the method can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0073] In this embodiment, the information processing method includes the following steps:

[0074] Step S102, obtaining risky credit training set and risky credit configuration information.

[0075] The risk credit training set is the training data used to train the risk credit model, that is, the training data for determining the learning parameters such as the weight and bias of the risk credit model. The risk credit configuration information is the relevant information used to configure the tree-based pipeline optimization tool TPOT, and then build the risk credit model.

[0076] The risk credit configuration information may include at least one of an evaluation function and a parameter configuration file, and may also include genetic algorithm (GA) parameters such as population size, number of iterations, mutation rate, crossover rate, etc. Genetic algorithm is a heuristic optimization algorithm based on natural selection and heredity, which simulates the evolution process of organisms in nature to find the optimal solution to the problem.

[0077] Optionally, the electronic device obtains the original risk credit data, and preprocesses the original risk credit data to obtain a risk credit training set. The preprocessing may include preprocessing steps such as data cleaning, feature selection, and normalization, but is not limited thereto.

[0078] Optionally, the electronic device preprocesses the risk credit original data to obtain preprocessed data, and divides the preprocessed data into a risk credit training set and a risk credit verification set according to time.

[0079] Optionally, the electronic device obtains risk credit configuration information input by the user.

[0080] Step S104, configuring the tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT.

[0081] Among them, the Tree-based Pipeline Optimization Tool (TPOT) is an automated machine learning framework that is used to automatically train the target task flow to build a risk credit model. Target TPOT is an automated machine learning framework for training the task flow corresponding to the risk credit model.

[0082] like Figure 2 As shown in the figure, TPOT is an automated machine learning framework developed based on sklearn. It supports multiple classification algorithms such as Decision Tree, Random Forest, Logistic Regression, SVM, XGBoost, etc., which can meet the modeling needs of most risk control scenarios. It also provides multiple feature engineering components such as preprocessors, classifiers, and feature selection. The preprocessors include standard scaler, robust scaler, polynomial feature generator, and random principal component analysis. The feature selectors include recursive feature elimination (RFE), best k features (SelectKBest), best n percentile features (SelectPercentile), and remove features with variance below the threshold (VarianceThreshold). The classifiers include decision tree classifier (DecisionTree), random forest classifier (RandomForest), gradient boosting classifier (GBDT), support vector machine (SVM), logistic regression (LR), and ensemble learning (Ensemble Learning). Features can be scaled using sample mean, variance, median, and quartiles, and interactive features can be generated through polynomial combinations of numerical features. In addition, feature selectors are provided for feature screening based on multiple rules such as RFE and KBest.

[0083] TPOT uses a genetic algorithm (GA) to automatically combine preprocessors, decomposers (classifiers), feature selectors, models, and their parameters to build the optimal machine learning task flow. Figure 3 As shown, the electronic device inputs the raw data into TPOT, and the raw data undergoes feature selection, data cleaning, feature preprocessing, model selection, parameter optimization, model validation, and feature construction for automated machine learning, which can build the optimal machine learning task flow.

[0084] Optionally, the electronic device creates a tree-based pipeline optimization tool TPOT to be configured, inputs the risk credit configuration information into the TPOT, configures the TPOT based on the risk credit configuration information, and obtains a target TPOT.

[0085] Optionally, the electronic device configures genetic algorithm parameters in the TPOT, and obtains at least one of an evaluation function and a parameter configuration file to obtain a target TPOT.

[0086] Step S106: Perform machine learning on the risk credit training set through the target TPOT to obtain at least one candidate task flow.

[0087] Step S108: determining a target task flow from at least one candidate task flow.

[0088] Among them, machine learning is a multidisciplinary cross-disciplinary major that covers knowledge of probability theory, statistics, approximate theory and complex algorithm knowledge. It uses computers as tools and is committed to simulating human learning methods in real time, and divides existing content into knowledge structures to effectively improve learning efficiency.

[0089] It can be understood that TPOT automatically combines preprocessors, decomposers, feature selectors, models and their parameters through genetic algorithms, and intelligently explores thousands of possible pipelines to find the best pipeline, which is the optimal target task flow, and the target task flow is the pipeline with the best processing efficiency and performance. The first task flow is the initial task flow obtained by combining various feature engineering components by the target TPOT. Feature engineering components include preprocessors, decomposers, feature selectors, etc., which can form task flows and model.

[0090] Optionally, the electronic device uses a genetic algorithm to perform machine learning on a risk credit training set through a target TPOT to determine a target task flow.

[0091] Optionally, the electronic device uses multi-threading to perform machine learning on the risk credit training set through target TPOT to determine the target task flow and further improve the efficiency of the overall modeling.

[0092] Optionally, the electronic device periodically performs machine learning on the risk credit training set through the target TPOT at preset intervals, that is, automatically iterates the risk credit model at regular intervals, thereby solving the problem of tight development cycle time.

[0093] Optionally, the electronic device performs machine learning on the risk credit training set through the target TPOT to obtain the first task flow; if the first task flow meets the target condition, the first task flow is used as a candidate task flow. The target condition can be set as needed. Exemplarily, the target condition can be that the comprehensive score of the first task flow is greater than the score threshold, or that the monotonicity corresponding to the first task flow meets the monotonicity rule, etc., but is not limited thereto.

[0094] Optionally, if the electronic device obtains one candidate task flow, the candidate task flow is determined as the target task flow; if the electronic device obtains at least two candidate task flows, the target task flow is determined from the at least two candidate task flows.

[0095] Optionally, the electronic device obtains the comprehensive score of each candidate task flow, and determines the target task flow based on the comprehensive scores of each candidate task flow. The electronic device may determine the candidate task flow with the highest comprehensive score as the target task flow, or may determine the candidate task flow with the second highest comprehensive score as the target task flow, which is not limited here.

[0096] Optionally, the electronic device may determine the candidate task flow obtained at the last moment as the target task flow from the at least two candidate task flows. Optionally, the electronic device may randomly determine the target task flow from the at least two candidate task flows.

[0097] Step S120: constructing a risk credit model based on the target task flow.

[0098] Among them, the risk credit model is used to provide relevant information on risk credit, such as for risk prediction of users and determination of credit limits.

[0099] Optionally, the electronic device obtains the target task flow output by the target TPOT, and trains the target task flow using the risk credit training set to obtain the risk credit model. The target task flow output by the TPOT may be a Python code.

[0100] Optionally, constructing a risk credit model based on the target task flow includes: modifying the target task flow in response to a modification instruction to obtain a new target task flow; and training the new target task flow using a risk credit training set to obtain a risk credit model.

[0101] It is understandable that after the target TPOT outputs the target task flow, it is necessary to retrain and model according to the target task flow, which may involve some parts that require code modification. For example, during the remodeling process, some features may be evaluated by the model personnel as having poor interpretability. This requires modifying the code in the target task flow, eliminating the variable and then remodeling. After the modeling is completed, the modeler needs to change the code to save the model as the required type.

[0102] Optionally, the electronic device modifies the code of the target task flow in response to the modification instruction of the target task flow to obtain a new target task flow, and retrains the new target task flow using the risk credit training set to obtain a risk credit model. The new target task flow includes the optimal features, model, parameters, etc.

[0103] Optionally, after the risk credit model is accepted, the electronic device deploys the risk credit model to various platform applications for use.

[0104] The above information processing method obtains the risk credit training set and the risk credit configuration information, configures the tree-based pipeline optimization tool TPOT based on the risk credit configuration information, and obtains the target TPOT. Then, through the target TPOT, the risk credit training set can be automatically machine-learned, and at least one candidate task flow can be obtained more quickly, and the target task flow can be determined from at least one candidate task flow, and then based on the target task flow, the risk credit model can be built more quickly, thereby improving the modeling and information processing efficiency of the risk credit model; reducing the workload of manual feature engineering and improving the efficiency of modeling; compared with traditional modeling methods, more expert experience is required, and the above information processing method based on TPOT can lower the technical threshold, and the modeler can complete multiple steps such as feature screening, feature processing, model selection, and parameter adjustment based on TPOT, thereby improving the efficiency of modeling.

[0105] Moreover, as can be seen from the following Table 1, the above information processing method can effectively prevent the overfitting problem. On the same OOT sample, the KS effect is 45.91% higher than the XGBoost risk model established by traditional methods, and 3.85% higher than the KS effect of the traditional TPOT model, which significantly improves the prediction accuracy and Lift ranking of the risk credit model, providing a more reliable decision-making basis for risk management. The automated design of the risk credit model is realized through TPOT, which improves the modeling performance of the risk credit model.

[0106] Table 1:

[0107]

[0108] In one embodiment, machine learning is performed on a risk credit training set through a target TPOT to obtain at least one candidate task flow, including: combining various feature engineering components through the target TPOT to obtain a first task flow; for each first task flow, machine learning is performed on the risk credit training set to obtain a comprehensive score of the first task flow; if the comprehensive score of the first task flow is greater than a score threshold, the first task flow is used as a candidate task flow.

[0109] The comprehensive score of the first task flow is used to comprehensively evaluate the model corresponding to the first task flow. The comprehensive score of the first task flow is positively correlated with the machine learning ability of the first task flow, and the comprehensive score of the first task flow is positively correlated with the performance of the model corresponding to the first task flow. The score threshold can be set as needed.

[0110] Optionally, for each first task flow, the electronic device performs machine learning on the risk credit training set, and performs score evaluation on the first task flow through the evaluation function in the target TPOT to obtain a comprehensive score of the first task flow.

[0111] If the comprehensive score of the first task flow is less than or equal to the score threshold, it means that the machine learning ability of the first task flow is low and the model performance corresponding to the first task flow is low, then a target penalty value is returned, and the target TPOT is re-executed to combine various feature engineering components to obtain the first task flow. The target penalty value is a larger penalty value that can be set as needed to indicate that the model corresponding to the first task flow does not meet the requirements.

[0112] In this embodiment, through the target TPOT, the various feature engineering components are combined to obtain the first task flow; for each first task flow, the risk credit training set is machine-learned to obtain the comprehensive score of the first task flow. If the comprehensive score of the first task flow is greater than the score threshold, it means that the model corresponding to the first task flow meets the requirements, that is, the model performance is high, then the first task flow is used as a candidate task flow, and then the accurate target task flow can be obtained, and information processing can be performed more accurately, which significantly improves the prediction accuracy of the risk credit model. Since the customized evaluation function makes requirements on the stability of the model, on the OOT sample, the attenuation of the model performance is much lower than that of the model directly constructed by TPOT, so that it can better adapt to the dynamically changing risk environment. In addition, this embodiment also improves the interpretability of the risk credit model, which helps to deeply understand the risk factors and their impact. In the risk scenario, the customized evaluation function restricts the method and results of TPOT feature generation, which can ensure that the input model features or the transformed input model features are monotonic with the target under equal frequency binning.

[0113] In one embodiment, a risk credit training set includes sample features and target results; for each first task flow, machine learning is performed on the risk credit training set to obtain a comprehensive score of the first task flow, including: for each first task flow, obtaining a correspondence between sample features and target results, and predicting the sample features based on the correspondence to obtain a predicted result; determining a first score based on the degree of difference between the predicted result and the target result; and determining a comprehensive score of the first task flow based on the sample features and the first score.

[0114] Among them, the target result is the learning result corresponding to the sample feature, that is, the Y label of the model. In supervised learning, the model (first task flow) needs the target result (label) to learn. Taking the risk model as an example, the target result is that the user is overdue. Then the model (first task flow) learns the sample features of overdue users in the sample, such as historical overdue, low income, high debt, and debts in multiple places. Sample features, so as to understand that these sample features are related to overdue (target result), and then these sample features can be used to predict the result (whether the user is overdue).

[0115] The sample feature is the input variable of the model corresponding to the first task flow, and the first score is the output variable of the model corresponding to the first task flow. For each first task flow, the correspondence between the sample feature and the target result is obtained, and the correspondence is learned. The sample feature is predicted based on the correspondence, that is, when the sample feature is known and the target result is unknown, the prediction result of the sample feature is obtained.

[0116] Optionally, the difference between the predicted result and the target result is negatively correlated with the first score, that is, the smaller the difference between the predicted result and the target result, the closer the predicted result of the sample feature by the model corresponding to the first task flow is to the target result, and a larger first score can be determined. The first score is used to evaluate the machine learning capability of the model corresponding to the first task flow.

[0117] Optionally, the electronic device obtains an evaluation function in the target TPOT, and determines a comprehensive score of the first task flow based on the sample feature and the first score through the evaluation function.

[0118] Optionally, based on the sample characteristics and the first score, a comprehensive score of the first task flow is determined, including: based on the sample characteristics and the first score, determining the risk differentiation capability of the first task flow; determining the target monotonicity between the sample characteristics and the first score; determining the target stability of the first score on the sample characteristics; based on at least one of monotonicity and stability, determining the comprehensive score of the first task flow.

[0119] Among them, the risk differentiation ability (Kolmogorov-Smirnov, KS) is the difference (maximum value) of the cumulative distribution between good and bad samples. The greater the cumulative difference between good and bad samples, the stronger the risk differentiation ability of the model corresponding to the first task flow, and the larger the KS indicator.

[0120] Monotonicity refers to whether the relationship between a feature and a model score is monotonically increasing or monotonically decreasing. In order to make the risk credit model more interpretable, the sample features used by the model corresponding to the first task flow generally need to maintain monotonicity with the first score, that is, when other variables remain unchanged, the larger the value of the sample feature, the more constant the trend of the first score change. Stability refers to the degree of overall migration of features or model scores in different time periods (such as different months). General businesses tend to use stable features and stable first scores.

[0121] Target monotonicity is used to characterize whether there is monotonicity between the sample features and the first score. Target stability is used to characterize whether the first score is stable in terms of the sample features.

[0122] Optionally, the electronic device inputs the sample features and the first score into an evaluation function, determines the risk differentiation ability of the first task flow, the target monotonicity between the sample features and the first score, and the target stability of the first score on the sample features through the evaluation function, and determines the comprehensive score of the first task flow based on the risk differentiation ability, target monotonicity and target stability.

[0123] Optionally, the electronic device determines the comprehensive score of the first task flow through the following evaluation function:

[0124] Score = F(KS, PSI, Monotonicity Rule)

[0125] Among them, Score is the comprehensive score of the first task flow, F( ) is the evaluation function, KS is the risk differentiation ability, PSI (Population Stability Index) is the target stability, and Monotonicity Rule is the monotonicity rule.

[0126] It is understandable that in this evaluation function, since KS and PSI are both numerical values, they can be dimensioned for mathematical calculations, and the monotonicity rule is a rule rather than a numerical value, so the sample features and whether the first score meets the target monotonicity of the monotonicity rule are mapped to a numerical value, and then the comprehensive score of the first task flow is obtained. In the risk scenario, the customized evaluation function restricts the method and results of TPOT feature generation, which can ensure that the input features or transformed input features are monotonic with the target under equal frequency binning, and also ensure the robustness of the model to meet the needs of risk control scenarios.

[0127] like Figure 4 As shown, the electronic device uses a genetic algorithm to determine the best target task flow, which is as follows: the electronic device initializes the population and obtains the first generation of the population G=0, and the population is the first task flow; the fitness of the population is calculated, that is, the comprehensive score of the first task flow is determined by a custom evaluation function, and the evaluation function comprehensively considers feature monotonicity, model stability (PSI) and model effect (KS). If the comprehensive score of the first task flow is greater than the score threshold, the first task flow is used as a candidate task flow; selection operations, crossover operations and mutation operations are performed on each feature engineering component to generate the next generation population, that is, G=G+1; it is detected whether the number of generations of the current population is greater than the preset threshold GEN, if not, the fitness of the current population is calculated, if it is, G>GEN, then the target task flow is output, and the target task flow includes the optimal features, models and parameters.

[0128] In this embodiment, for each first task flow, the electronic device learns based on the correspondence between sample features and target results, determines the first score by the degree of difference between the predicted results and the target results, so as to determine the learning ability of the model corresponding to the first task flow, and further more accurately determine the comprehensive score of the first task flow.

[0129] Furthermore, the electronic device determines the risk differentiation ability of the first task flow, the target monotonicity between the sample characteristics and the first score, and the target stability. Based on at least one of the risk differentiation ability, the target monotonicity and the target stability, the comprehensive score of the first task flow can be determined more accurately.

[0130] In one embodiment, machine learning is performed on the risk credit training set through the target TPOT to determine the target task flow, including: obtaining parameters within the target search range through the target TPOT; and performing machine learning on the risk credit training set according to the parameters within the target search range to obtain at least one candidate task flow.

[0131] The target search range can be set as needed or based on the preset search range rules. The config_dict parameter is the preset search range rule. The config_dict parameter is a text in a nested dictionary format. The first-level key is the path and name of the operator (e.g. sklearn.naive_bayes.MultinomialNB), and the second-level key is the parameter name corresponding to the operator (e.g. fit_prior). The second-level key should point to the parameter value list of the parameter, such as 'fit_prior': [True, False]. By setting the config_dict parameter, TPOT will search and optimize the model type and its corresponding parameter space within the target search range, thereby improving the efficiency of the global search.

[0132] Parameters include feature screening logic, feature processing methods, and model type. Exemplarily, the parameters within the target search range may include classifiers with better interpretability, model type parameters within a specified parameter range, etc. Classifiers with better interpretability include logistic regression, decision trees, XGBoost, etc. For each model type, parameters such as max_depth, min_samples_split, and min_samples_leaf are configured to specify the parameter range. After limiting the model type, the target task flow output by TPOT is a model with better interpretability, which helps the business side understand the meaning of the features.

[0133] Optionally, the target TPOT is used to obtain parameters within a target search range in a parameter configuration file. The parameter configuration file includes a target search range for the parameters.

[0134] In this embodiment, the target TPOT is used to obtain parameters within the target search range, and machine learning is performed on the risk credit training set according to the parameters within the target search range. This can avoid searching a large number of parameters and task flows outside the target search range, thereby more quickly obtaining at least one candidate task flow, improving information processing efficiency, and saving computer resources required for modeling.

[0135] The speed of model training is 5 times faster than that of directly using TPOT. The parameters within the target search range are limited for risk control scenarios, which reduces the overall search space of TPOT and also enhances the generalization ability of the model. In a test training sample with 100 features and 1000 samples, the time spent on the above information processing is one-fifth of that of traditional TPOT.

[0136] In addition, after limiting parameters such as model types through the target search range, the models output by TPOT will include models with better interpretability, which will help the business understand the meaning of features.

[0137] In one embodiment, after machine learning is performed on the risk credit training set through the target TPOT to obtain at least one candidate task flow, it also includes: converting the target task flow into visualization features; evaluating the interpretability of the target task flow based on the visualization features; if the interpretability of the target task flow reaches a preset interpretability level, executing the step of building a risk credit model based on the target task flow.

[0138] Among them, visualization features are intuitive features. For example, visualization features such as model lift graphs, AUC graphs, feature importance ranking graphs, etc. can help modelers understand and evaluate risk credit models. Interpretability refers to the degree to which people can understand the reasons for decisions. For the business side, interpretability means that through the model architecture, it is possible to understand which variables have a positive or negative effect on the score. Risk credit models with good interpretability can be predicted based on some intuitive and easy-to-understand rules or conditions. The preset degree of interpretability can be set as needed.

[0139] Optionally, the electronic device uses visualization technology to convert the target task flow into visualization features; the interpretability of the target task flow can be evaluated based on each visualization feature; if the interpretability of the target task flow reaches a preset interpretability level, indicating that the target task flow has a high interpretability, then a risk credit model is constructed based on the target task flow; if the interpretability of the target task flow does not reach a preset interpretability level, indicating that the target task flow has a high interpretability, then the execution is returned to perform machine learning on the risk credit training set through the target TPOT to determine the steps of the target task flow.

[0140] The electronic device constructs a PDP graph or a tree graph according to various visualization features, and evaluates the interpretability of the target task flow through the PDP graph or the tree graph.

[0141] In this embodiment, the electronic device converts the target task flow into visualization features. Based on the visualization features, the interpretability of the target task flow can be evaluated. If the interpretability of the target task flow reaches a preset interpretability level, a risk credit model is constructed based on the target task flow, which improves the interpretability of the risk credit model and helps to deeply understand the risk factors and their impacts.

[0142] In one embodiment, after machine learning is performed on the risk credit training set through the target TPOT to obtain at least one candidate task flow, the method further includes: evaluating the model effect of the target task flow; if the model effect evaluation passes, executing the step of building a risk credit model based on the target task flow.

[0143] Optionally, the electronic device obtains the performance indicators of the target task flow, performs a model effect evaluation on the target task flow based on each performance indicator, and obtains the target performance of the target task flow; if the target performance is greater than the preset performance, indicating that the model effect evaluation has passed, then the step of building a risk credit model based on the target task flow is executed; if the target performance is less than or equal to the preset performance, indicating that the model effect evaluation has not passed, then the step of returning to execute the target TPOT and performing machine learning on the risk credit training set to determine the step of the target task flow. Among them, the preset performance can be set as needed.

[0144] The performance indicators of the target task flow can include KS, Lift, AUC, PSI and other indicators. Among them, KS (Kolmogorov-Smirnov): the difference (maximum value) of the cumulative distribution between good and bad samples, which is used to evaluate the risk differentiation ability of the model. Lift is an indicator for evaluating the performance of classification models, and is usually used to evaluate the performance of models in marketing, advertising and other scenarios. It refers to the increase in the target conversion rate relative to random selection after using the prediction model. AUC (Area Under the Curve), that is, the area under the Receiver Operating Characteristic Curve, is an indicator used to evaluate the performance of a binary classification model. PSI is used to evaluate the stability of the model corresponding to the target task flow.

[0145] Optionally, a model effect evaluation is performed on the target task flow, including: obtaining a risk credit verification set; learning the risk credit verification set through the target task flow to determine a second score; and evaluating the model effect of the target task flow based on the second score.

[0146] The risk credit validation set is used to evaluate the data of the model performance effect corresponding to the target task flow. The second score is used to evaluate the model effect of the target task flow.

[0147] Optionally, the electronic device learns the risk credit verification set through the target task flow, outputs a second score, and evaluates the model effect of the target task flow based on the difference between the second score and the actual target score. The actual target score is the actual score corresponding to the risk credit verification set. The difference is negatively correlated with the model effect of the target task flow, that is, the smaller the difference, the better the model effect of the target task flow.

[0148] In this embodiment, the electronic device performs a model effect evaluation on the target task flow. If the model effect evaluation passes, a risk credit model can be constructed more accurately based on the target task flow to obtain a risk credit model with better model effect.

[0149] Furthermore, the electronic device obtains a risk credit verification set, learns the risk credit verification set through the target task flow, and determines a second score. Then, based on the second score, the model effect of the target task flow can be evaluated more accurately.

[0150] In one embodiment, an information processing method is also provided, which is applied to an electronic device, and the information processing method includes the following steps:

[0151] Step A1, obtaining a risk credit training set and risk credit configuration information; the risk credit training set includes sample features and target results.

[0152] Step A2, configuring the tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT.

[0153] Step A3, obtaining parameters within the target search range through the target TPOT; obtaining each feature engineering component according to the parameters within the target search range, and combining each feature engineering component to obtain the first task flow.

[0154] Step A4, for each first task flow, obtain the correspondence between the sample features and the target results, and predict the sample features based on the correspondence to obtain the predicted results; determine the first score based on the degree of difference between the predicted results and the target results.

[0155] Step A5, based on the sample characteristics and the first score, determine the risk differentiation ability of the first task flow; determine the target monotonicity between the sample characteristics and the first score; determine the target stability of the first score on the sample characteristics; based on at least one of the risk differentiation ability, target monotonicity and target stability, determine the comprehensive score of the first task flow; if the comprehensive score of the first task flow is greater than the score threshold, then use the first task flow as a candidate task flow.

[0156] Step A6: determining a target task flow from at least one candidate task flow.

[0157] Step A7: convert the target task flow into visualization features; and evaluate the interpretability of the target task flow based on the visualization features.

[0158] Step A8, obtaining a risk credit verification set; learning the risk credit verification set through the target task flow to determine a second score; and evaluating the model effect of the target task flow based on the second score.

[0159] Step A9: If the interpretability of the target task flow reaches the preset interpretability level and the model effect evaluation is passed, the target task flow is modified in response to the modification instruction to obtain a new target task flow; the new target task flow is trained using the risk credit training set to obtain a risk credit model.

[0160] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0161] Based on the same inventive concept, the embodiment of the present application also provides an information processing device for implementing the above-mentioned information processing method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in the one or more information processing device embodiments provided below can refer to the limitations on the information processing method above, and will not be repeated here.

[0162] In an exemplary embodiment, Figure 5 As shown, an information processing device is provided, including: a data acquisition module 502, a TPOT configuration module 504, a target task flow determination module 506 and a risk credit model construction module 508, wherein:

[0163] The data acquisition module 502 is used to acquire risky credit training sets and risky credit configuration information.

[0164] The TPOT configuration module 504 is used to configure the tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT.

[0165] The target task flow determination module 506 is used to perform machine learning on the risk credit training set through the target TPOT to obtain at least one candidate task flow; and determine the target task flow from the at least one candidate task flow.

[0166] The risk credit model building module 508 is used to build a risk credit model based on the target task flow.

[0167] The above-mentioned information processing device obtains a risk credit training set and risk credit configuration information, configures a tree-based pipeline optimization tool TPOT based on the risk credit configuration information, and obtains a target TPOT. Then, through the target TPOT, it is possible to automatically perform machine learning on the risk credit training set, obtain at least one candidate task flow more quickly, and determine a target task flow from at least one candidate task flow, and then, based on the target task flow, build a risk credit model more quickly, thereby improving information processing efficiency.

[0168] In one embodiment, the target task flow determination module 506 is further used to combine various feature engineering components through target TPOT to obtain a first task flow; for each first task flow, machine learning is performed on the risk credit training set to obtain a comprehensive score of the first task flow; if the comprehensive score of the first task flow is greater than the score threshold, the first task flow is used as a candidate task flow.

[0169] In one embodiment, the risk credit training set includes sample features and target results; the above-mentioned target task flow determination module 506 is also used to obtain the correspondence between the sample features and the target results for each first task flow, and predict the sample features based on the correspondence to obtain the predicted results; determine the first score based on the degree of difference between the predicted results and the target results; determine the comprehensive score of the first task flow based on the sample features and the first score.

[0170] In one embodiment, the above-mentioned target task flow determination module 506 is also used to determine the risk differentiation ability of the first task flow based on the sample characteristics and the first score; determine the target monotonicity between the sample characteristics and the first score; determine the target stability of the first score on the sample characteristics; and determine the comprehensive score of the first task flow based on at least one of the risk differentiation ability, target monotonicity and target stability.

[0171] In one embodiment, the target task flow determination module 506 is further used to obtain parameters within the target search range through the target TPOT; perform machine learning on the risk credit training set according to the parameters within the target search range to obtain at least one candidate task flow.

[0172] In one embodiment, the risk credit model building module 508 is further used to modify the target task flow in response to a modification instruction to obtain a new target task flow; and train the new target task flow using a risk credit training set to obtain a risk credit model.

[0173] In one embodiment, the above-mentioned device also includes a target task flow evaluation module, which is used to convert the target task flow into visualization features; based on the visualization features, the interpretability of the target task flow is evaluated; if the interpretability of the target task flow reaches a preset interpretability level, the above-mentioned risk credit model construction module 508 is used to construct a risk credit model based on the target task flow.

[0174] In one embodiment, the target task flow evaluation module is further used to perform model effect evaluation on the target task flow; if the model effect evaluation passes, the risk credit model construction module 508 is used to construct a risk credit model based on the target task flow.

[0175] In one embodiment, the target task flow evaluation module is also used to obtain a risk credit verification set; through the target task flow, the risk credit verification set is learned to determine a second score; based on the second score, the model effect of the target task flow is evaluated; if the model effect evaluation passes, the risk credit model construction module 508 is used to construct a risk credit model based on the target task flow.

[0176] Each module in the above information processing device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in an electronic device in the form of hardware, or can be stored in a memory in an electronic device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0177] In an exemplary embodiment, an electronic device is provided. The electronic device may be a server, and its internal structure diagram may be as shown in FIG. Figure 6As shown. The electronic device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store data such as risk credit training sets and risk credit configuration information. The input / output interface of the electronic device is used to exchange information between the processor and an external device. The communication interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an information processing method is implemented.

[0178] Those skilled in the art will understand that Figure 6 The structure shown in the figure is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0179] In one embodiment, an electronic device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0180] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0181] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0183] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0184] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0185] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An information processing method, characterized in that: The method comprises: Obtain risky credit training set and risky credit configuration information; configuring a tree-based pipeline optimization tool TPOT based on the risk credit configuration information to obtain a target TPOT; By using the target TPOT, machine learning is performed on the risk credit training set to obtain at least one candidate task flow; Determine a target task flow from the at least one candidate task flow; Based on the target task flow, a risk credit model is constructed.

2. The method according to claim 1, characterized in that The step of performing machine learning on the risk credit training set through the target TPOT to obtain at least one candidate task flow includes: Through the target TPOT, various feature engineering components are combined to obtain the first task flow; For each first task flow, performing machine learning on the risk credit training set to obtain a comprehensive score of the first task flow; If the comprehensive score of the first task flow is greater than the score threshold, the first task flow is used as a candidate task flow.

3. The method according to claim 2, characterized in that The risk credit training set includes sample characteristics and target results; The step of performing machine learning on the risk credit training set for each first task flow to obtain a comprehensive score of the first task flow includes: For each first task flow, obtaining a correspondence between the sample feature and the target result, and predicting the sample feature based on the correspondence to obtain a prediction result; Determining a first score based on the degree of difference between the predicted result and the target result; Based on the sample features and the first score, a comprehensive score of the first task flow is determined.

4. The method according to claim 3, characterized in that The step of determining a comprehensive score of the first task flow based on the sample feature and the first score includes: Determining a risk differentiation capability of the first task flow based on the sample characteristics and the first score; determining a target monotonicity between the sample feature and the first score; determining a target stability of the first score on the sample characteristics; Based on at least one of the risk differentiation capability, the target monotonicity, and the target stability, a comprehensive score of the first task flow is determined.

5. The method according to claim 1, characterized in that The step of performing machine learning on the risk credit training set through the target TPOT to obtain at least one candidate task flow includes: Through the target TPOT, the parameters within the target search range are obtained; According to the parameters within the target search range, machine learning is performed on the risk credit training set to obtain at least one candidate task flow.

6. The method according to claim 1, characterized in that The step of constructing a risk credit model based on the target task flow includes: In response to the modification instruction, modify the target task flow to obtain a new target task flow; The new target task flow is trained using the risk credit training set to obtain a risk credit model.

7. The method according to claim 1, characterized in that After determining the target task flow from the at least one candidate task flow, the method further includes: Converting the target task flow into visualization features; According to the visualization features, evaluating the interpretability of the target task flow; If the interpretability of the target task flow reaches a preset interpretability level, the step of constructing a risk credit model based on the target task flow is performed.

8. The method according to claim 1, characterized in that After determining the target task flow from the at least one candidate task flow, the method further includes: Get the risk credit validation set; By using the target task flow, learning the risk credit verification set, and determining a second score; Based on the second score, performing a model effect evaluation on the target task flow; If the model effect evaluation is passed, the step of constructing a risk credit model based on the target task flow is executed.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

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

Images