A method, device and equipment for allocating system resources

Abstract

The embodiments of this specification relate to the field of artificial intelligence technology, and disclose a system resource allocation method, apparatus and equipment, the method comprising obtaining a specified information set and a label set having feature data for characterizing user risk characteristics; the specified information set comprises at least a prediction sample set and three benchmark sample sets; the label set comprises risk categories corresponding to the benchmark samples; at least three sub-classifiers are constructed using at least three benchmark sample sets and the label sets; each sub-classifier is used to perform risk classification on the prediction sample, and a risk classification result of the prediction sample relative to each sub-classifier and a misclassification cost of the prediction sample are obtained; each sub-classifier is optimized according to the prediction sample set and the risk classification result of the prediction sample, and the misclassification cost of the prediction sample, to obtain each optimized sub-classifier, so as to allocate system resources to the target user based on the risk prediction result of the target user by each optimized sub-classifier.

Description

Technical Field

[0001] The present specification relates to the field of artificial intelligence technology, and in particular, to a system resource allocation method, device and equipment. Background Art

[0002] With the rapid development of big data service platform technology, the types of financial resource services and the service channels available are becoming more diverse and convenient, and user risk prediction is becoming more and more important for financial institutions. For example, for some online loan businesses with more convenient service channels, due to relatively less human intervention, if the user risk prediction is not accurate enough, there may be problems such as unreasonable resource allocation and poor user experience.

[0003] At present, the commonly used user risk prediction method is mainly a classification method based on a supervised learning model. By building a model based on data of known customer risks, the trained model is used to predict user risks for new samples to determine the risk level of the user. However, the classification method using a supervised learning model requires the use of information on known user risks. However, there are usually few samples of known user risks, and the sample data of known user risks may not accurately characterize the risk characteristics of users under new service types, thus affecting the accuracy of user risk prediction and the accuracy of resources allocated to corresponding users, which may lead to unreasonable resource allocation and reduce the user experience.

[0004] To address the above problems, no effective solution has been proposed yet. Summary of the invention

[0005] The purpose of the embodiments of this specification is to provide a system resource allocation method, device and equipment, which can improve the accuracy and efficiency of system resource allocation.

[0006] This specification provides a system resource allocation method, device and equipment which are implemented in the following ways:

[0007] A system resource allocation method is applied to a server, the method comprising: obtaining a specified information set and a label set having feature data for characterizing user risk features; the specified information set at least comprises a prediction sample set and three benchmark sample sets; the label set comprises the risk category corresponding to each benchmark sample in the benchmark sample set; at least three sub-classifiers are constructed correspondingly using at least three benchmark sample sets and the label set; the prediction samples in the prediction sample set are risk classified using the at least three sub-classifiers respectively to obtain the risk classification results of the prediction samples relative to each sub-classifier and the misclassification cost of the prediction samples; wherein the misclassification cost is used to characterize the possibility of the prediction samples being misclassified; each sub-classifier is optimized according to the prediction sample set and the risk classification results of the prediction samples relative to each sub-classifier and the misclassification cost of the prediction samples to obtain the optimized sub-classifiers, so as to allocate system resources to the target user based on the risk prediction results of the target user by the optimized sub-classifiers.

[0008] In some other embodiments of the method provided in this specification, the designated information set includes at least four benchmark sample sets; the benchmark sample set that is not used to construct the sub-classifier is used as a verification sample set; the optimized sub-classifiers are verified using the verification sample set, and the weights of the optimized sub-classifiers are determined according to the verification results; accordingly, the optimized sub-classifiers and the weights of the optimized sub-classifiers are used to predict the risk of the target user.

[0009] In some other embodiments of the method provided in the present specification, each sub-classifier is optimized according to the prediction sample set and the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples, including: for any prediction sample, counting the number of sub-classifiers corresponding to each risk classification result of the prediction sample, and marking the risk classification result with the largest number of sub-classifiers; taking the ratio of the number of sub-classifiers corresponding to the marked risk classification results to the number of all sub-classifiers as the misclassification cost of the corresponding prediction sample; accordingly, each sub-classifier is optimized according to the prediction sample set and the marked risk classification results corresponding to the prediction samples, and the misclassification cost of the prediction samples.

[0010] In some other embodiments of the method provided in this specification, each sub-classifier is optimized based on the following objective function:

[0011]

[0012] Among them, L(X n ,X u ) is the objective function, X n are at least three benchmark sample sets used to build the classifier, Xi is any benchmark sample set used to build a classifier, Y i The benchmark sample set X i The set of risk categories of each benchmark sample in , where i = 1, 2, ..., n, n ≥ 3; X u is the prediction sample set, Y u is a set of risk classification results of the markers corresponding to each prediction sample in the prediction sample set, f i represents the sub-classifier constructed based on the corresponding benchmark sample set, f i (·) represents the discriminant function of the subclassifier; R emp (·) indicates experience loss, W u Represents the set of misclassification costs of each prediction sample in the prediction sample set, R reg (·) represents the L2 regularization loss, and α and β are hyperparameters.

[0013] In some other embodiments of the method provided in this specification, determining the weights of each optimized sub-classifier according to the verification result includes:

[0014] weight i =N i / N val

[0015] Among them, weight i represents the weight of each sub-classifier, where i = 1, 2, ..., n, n ≥ 3; N i Represents the number of samples correctly classified by the i-th subclassifier, N val Indicates the number of samples in the validation sample set.

[0016] In some other embodiments of the method provided in this specification, risk prediction is performed on the target user based on the following discriminant function:

[0017]

[0018] Among them, F(x) represents the discriminant function for risk prediction of the target user, f i (·) represents the discriminant function of the i-th sub-classifier, and x is the feature data used to characterize the risk characteristics of the target user.

[0019] In some other embodiments of the method provided in this specification, the feature data includes time series aggregation features and time series historical features; wherein the time series aggregation features refer to data obtained by extracting features of the user's specified information based on different time dimensions and a time series feature extraction algorithm; the time series historical features include time series distribution data obtained by statistics of the user's specified information based on different time dimensions.

[0020] On the other hand, an embodiment of the present specification also provides a system resource allocation device, which is applied to a server, and the device includes: a data acquisition module, which is used to acquire a specified information set and a label set having feature data for characterizing user risk characteristics; the specified information set includes at least a prediction sample set and three benchmark sample sets; the label set includes the risk category corresponding to each benchmark sample in the benchmark sample set; a classifier construction module, which is used to construct at least three sub-classifiers corresponding to at least three benchmark sample sets; a classification module, which is used to use the at least three sub-classifiers to perform risk classification on the prediction samples in the prediction sample set respectively, to obtain the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples; wherein the misclassification cost is used to characterize the possibility of the prediction samples being misclassified; a classifier optimization module, which is used to optimize each sub-classifier according to the prediction sample set and the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples, to obtain the optimized sub-classifiers; a resource allocation module, which is used to allocate system resources to the target user based on the risk prediction results of the target user by each optimized sub-classifier.

[0021] In some other embodiments of the device provided in this specification, the designated information set includes at least four benchmark sample sets; the benchmark sample set that is not used to construct the sub-classifier is used as a verification sample set; the device also includes: a weight configuration module, which is used to verify the optimized sub-classifiers using the verification sample set, and determine the weights of the optimized sub-classifiers according to the verification results; accordingly, the risk prediction module is also used to use the optimized sub-classifiers and the weights of the optimized sub-classifiers to predict the risk of the target user.

[0022] On the other hand, an embodiment of the present specification also provides a system resource allocation device, which is applied to a server. The device includes at least one processor and a memory for storing processor executable instructions. When the instructions are executed by the processor, the steps of the method described in any one or more of the above embodiments are implemented.

[0023] The system resource allocation method, device and equipment provided by one or more embodiments of this specification are divided into at least four parts by non-overlapping benchmark samples corresponding to user data of known risk, wherein at least three parts are used to construct at least three sub-classifiers, and other benchmark samples are used as verification data to determine the weights of sub-classifiers in the final prediction model. During the training process, the three sub-classifiers vote on the recognition results of the prediction samples corresponding to the user data of unknown risk, and update the recognition results of the prediction samples and the misclassification costs of the prediction samples according to the results of the vote. In the process of model optimization, the recognition results and misclassification costs of the prediction samples are iteratively updated, and the recognition results and misclassification costs of the prediction samples are used to construct a cost-sensitive pseudo-empirical loss, and the three sub-classifiers are iteratively optimized by minimizing the experience loss and the cost-sensitive pseudo-empirical loss. The classification effect of each sub-classifier can also be verified using the verification data, and the weight of the sub-classifier is calculated using the results of the verification data. The final model is obtained by integrating the weights of each sub-classifier, which is conducive to reducing the misleading of the "correct minority" by the "wrong majority". Through the above method, the recognition results and misclassification costs of the predicted samples are dynamically updated during the training process, so that the model can fully learn the data distribution information contained in a variety of samples, and then a model with stronger generalization ability can be constructed. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art description. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0025] Figure 1 A schematic diagram of a flow chart of an embodiment of a system resource allocation method provided in this specification;

[0026] Figure 2 A schematic diagram of the construction process of a risk prediction model in one embodiment provided in this specification;

[0027] Figure 3 A schematic diagram of a flow chart of a system resource allocation method in an embodiment provided in this specification;

[0028] Figure 4 This is a schematic diagram of the module structure of another system resource allocation device provided in this specification. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below in conjunction with the drawings in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of the specification, not all of the embodiments. Based on one or more embodiments of the specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the embodiments of this specification.

[0030] In a scenario example provided in an embodiment of this specification, the system resource allocation method can be applied to a device for performing system resource allocation, and the device may include a server or a server cluster composed of multiple servers. For a target user, the server can extract feature data from a variety of information of the target user as the feature data of the target user, and then use a pre-configured algorithm or model to predict the risk of the target user, and obtain the risk prediction result of the target user, so as to allocate resources to the target user based on the risk prediction result. The resources may include data resources such as services and products provided or recommended to users. For example, for a loan business scenario, the resources may be the loan amount allocated to the target user. For a cloud platform data service business scenario, the resources may be system data resources allocated to the target user, etc. By accurately identifying the risk of the user, resources can be allocated to the user more accurately and reasonably, improving the user experience and effectively reducing the resource loss of the institution.

[0031] Figure 1 It is a flowchart of an embodiment of the system resource allocation method provided in this specification. Although this specification provides the method operation steps or device structure shown in the following embodiments or drawings, the method or device may include more or fewer operation steps or module units after partial merger based on routine or no creative labor. In the steps or structures where there is no necessary causal relationship logically, the execution order of these steps or the module structure of the device is not limited to the execution order or module structure shown in the embodiments or drawings of this specification. When the method or module structure is applied in an actual device, server or terminal product, it can be executed sequentially or in parallel according to the method or module structure shown in the embodiments or drawings (for example, in an environment of parallel processors or multi-threaded processing, or even including distributed processing, server cluster implementation environment). A specific implementation example Figure 1 As shown, in one embodiment of the system resource allocation method provided in this specification, the method can be applied to the data processing device, and the method may include the following steps:

[0032] S20: Obtain a specified information set having feature data for characterizing user risk characteristics and a label set; the specified information set includes at least a predicted sample set and three benchmark sample sets; the label set includes the risk category corresponding to each benchmark sample in the benchmark sample set.

[0033] The server can obtain a specified information set and a label set. The specified information set may include multiple sample data. The sample data may include feature data for characterizing user risk characteristics. Accordingly, the specified information set may be a data set composed of feature data for characterizing user risk characteristics. The feature data may be feature data extracted from business data of users stored in a business system of a financial institution. Feature extraction may be performed through feature engineering. The feature data extraction method and feature type may be set according to the actual application scenario, and are not limited here. Of course, it may also include feature data extracted from user information obtained by the server from a platform associated with a financial institution.

[0034] In some embodiments, the sample data may be a benchmark sample or a predicted sample. The benchmark sample may include sample data of risk categories of configured users. The predicted sample may refer to sample data of risk categories of unconfigured users. The risk category may be risky, non-risky, etc., or may be a risk level such as high risk, medium risk, and low risk. It can be configured according to the actual application scenario. For example, for ease of processing, a single sample data may be set to correspond to a single user. Accordingly, feature data of users of known risk categories and feature data of users of unknown risk categories may be extracted respectively to construct benchmark samples and predicted samples. After associating the feature data corresponding to the benchmark sample and the predicted sample with the user identifier, they are stored in the benchmark sample set and the predicted sample set of the specified information set, respectively. After associating the risk category corresponding to the benchmark sample with the user identifier, it is stored in the label set.

[0035] In some embodiments, the benchmark samples may be randomly divided into at least three sets. Accordingly, the specified information set may include at least a prediction sample set and three benchmark sample sets. The prediction sample set may include multiple prediction samples. The benchmark sample set may include multiple benchmark samples. The benchmark samples in each benchmark sample set may be randomly assigned.

[0036] The pre-built specified information set and label set can be stored locally or in a database. The server can extract the specified information set and label set when allocating system resources or building a prediction model. If the constructed specified information set is an information set consisting of information of users corresponding to a specified product or a specified service scenario, an information set identifier can be set for each specified information set. Accordingly, the server can obtain the specified information set and label set corresponding to the corresponding information set identifier according to the needs of the current test scenario for system resource allocation in the current test scenario. At present, most of the business data in the business system is updated at a relatively fast speed. Accordingly, the feature data and label set of the specified information set can be dynamically updated at intervals to ensure the accuracy of the information in the information set, thereby improving the accuracy of the prediction.

[0037] S22: construct at least three sub-classifiers using at least three benchmark sample sets and the label sets.

[0038] In the process of building sub-classifiers, the server can select classification algorithms as needed, such as Bayesian, support vector machine, neural network, etc. Then, based on the selected classification algorithm, each benchmark sample set is used respectively, and the risk category corresponding to each benchmark sample is obtained from the label set to build a model, and at least three classifiers are obtained accordingly.

[0039] S24: Using the at least three sub-classifiers to perform risk classification on the prediction samples in the prediction sample set, respectively, to obtain the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples; wherein the misclassification cost is used to characterize the possibility of the prediction samples being misclassified.

[0040] After obtaining the sub-classifiers, the server can use each sub-classifier to perform risk classification on each prediction sample in the prediction sample set, and obtain the risk classification results of the corresponding prediction sample relative to each sub-classifier and the misclassification cost of the corresponding prediction sample. The misclassification cost can be used to characterize the possibility that the corresponding prediction sample is misclassified.

[0041] In some embodiments, the number of constructed sub-classifiers can be set to an odd number greater than or equal to 3. For any prediction sample, each sub-classifier can be used to perform risk classification to obtain the risk classification result of the corresponding prediction sample relative to each sub-classifier. Then, the number of sub-classifiers corresponding to each risk classification result can be counted, and the risk classification result with the largest number of sub-classifiers can be marked. And the ratio of the number of sub-classifiers corresponding to the marked risk classification result to the number of all sub-classifiers is used as the misclassification cost of the corresponding prediction sample. Accordingly, each sub-classifier can be optimized according to each prediction sample, the marked risk classification result corresponding to the corresponding prediction sample, and the misclassification cost of the corresponding prediction sample.

[0042] For example, assuming that the risk category includes two categories, high risk and low risk, the number of sub-classifiers trained is three. u , if the classification results of two sub-classifiers are the same and the third one is different, then the predicted sample x u The classification results are marked by the two sub-classifiers with the same classification, and the predicted sample x u The classification results with the label are used as the learning samples for the next round. Correspondingly, the predicted sample x can be calculated based on the number of classifiers involved in labeling and the total number of classifiers. u The corresponding misclassification cost, x u The corresponding misclassification cost is If the three sub-classifiers predict the sample x u The classification results are the same, predicting sample x u and its classification results are marked by three sub-classifiers, and the predicted sample x u The classification result with the label is used as the next round of learning sample. u The corresponding misclassification cost is 1. The specific calculation method is shown in the following formula.

[0043]

[0044]

[0045] Among them, ω1 and ω2 represent different risk categories of samples, f represents the subclassifier, and f i (·) represents the discriminant function of the i-th sub-classifier, such as f1(x u ) indicates using the first sub-classifier to predict sample x u The discriminant function for predictive processing, the explanation of the discriminant function in other forms of expression in the embodiments of this specification can refer to this explanation, and will not be repeated here. u Represents x u The misclassification cost of , the discriminant function of the sub-classifier f is as follows.

[0046]

[0047] Using the classification results of the sub-classifiers with a relatively large number as the labels of the predicted samples in the next training step can further improve the efficiency of the optimization process and the accuracy of the sub-classifier classification. Furthermore, determining the misclassification cost in the above manner can also improve the efficiency and accuracy of the misclassification cost determination.

[0048] Of course, other methods can also be used to determine the label value and misclassification cost of the prediction sample used to optimize the sub-classifier. For example, the classification result may not be a definite belonging to a certain category, but may be a probability of belonging to a certain category. Then, based on the probability value, a statistical analysis is performed to determine the label value and misclassification cost of the prediction sample used to optimize the sub-classifier.

[0049] S26: Optimize each sub-classifier according to the prediction sample set and the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples to obtain optimized sub-classifiers, so as to allocate system resources to the target user based on the risk prediction results of the target user for the optimized sub-classifiers.

[0050] Each prediction sample in the prediction sample set can be processed according to the method of step S24 to obtain the risk classification result and misclassification cost of any prediction sample for each sub-classifier. Then, based on each prediction sample and the risk classification result of the corresponding prediction sample relative to each sub-classifier and the misclassification cost of the corresponding prediction sample, each sub-classifier can be optimized to obtain the optimized sub-classifiers.

[0051] In some embodiments, the following objective function may be constructed to optimize each sub-classifier:

[0052]

[0053] R emp It's a loss of experience. is the pseudo-experience loss, W u is a set of misclassification costs of predicted samples. N is the number of prediction samples, R reg is the L2 regularization loss, α and β are the hyperparameters of the model, which are used to adjust the weight of the model. Expanding the above, we can get:

[0054]

[0055] Among them, L(X n ,X u ) is the objective function, X n are at least three benchmark sample sets used to build the classifier, Xi is any benchmark sample set used to build a classifier, Y i The benchmark sample set X i The set of risk categories of each benchmark sample in , where i = 1, 2, ..., n, n ≥ 3; X u is the prediction sample set, Y u is a set of risk classification results of the markers corresponding to each prediction sample in the prediction sample set, f i represents the sub-classifier constructed based on the corresponding benchmark sample set, f i (·) represents the discriminant function of the i-th sub-classifier; R emp (·) indicates experience loss, W u Represents the set of misclassification costs of each prediction sample in the prediction sample set, R reg (·) represents the L2 regularization loss. Correspondingly, R emp (Y u ,f i (X u )) is the empirical loss corresponding to the benchmark sample, R emp (Y u ,f i (X u )) is the empirical loss corresponding to the predicted sample, R teg (f i ) represents the L2 regularization loss of the i-th subclassifier. α and β are hyperparameters.

[0056] The optimization problem of the above objective function can be solved by using the gradient descent method. During the optimization process, the risk classification results and misclassification costs of the predicted samples are updated iteratively until the preset number of iterations is reached or the difference between the values ​​of the two objective functions is less than the preset threshold. Then the iteration is stopped and the sub-classifier obtained by the last iteration is used as the optimized sub-classifier. Of course, other optimization algorithms can also be used, such as the steepest descent method, Newton method, etc.

[0057] By constructing the above objective function to optimize the sub-classifier and dynamically updating the label value and misclassification cost of the predicted sample, the classifier can effectively learn the data distribution information contained in the predicted sample while further reducing the impact of the classification bias of the predicted sample on the model, thereby improving the generalization ability of the model ultimately used for risk prediction and thus improving the accuracy of risk prediction.

[0058] The above objective function is a preferred example for illustration. In practical applications, it can be adaptively constructed according to different classification algorithms adopted by the classifier, or the above objective function can be appropriately deformed, which is not limited here.

[0059] The server can use the optimized sub-classifiers to predict the risk of the user. For example, the business data of the target user in the system can be obtained as test data, and then the test data can be preprocessed and feature extracted to obtain feature data for characterizing the user risk characteristics of the target user. Then, the feature data of the target user can be input into the optimized sub-classifiers, and the risk classification results relative to each sub-classifier can be output. Then, the risk classification results can be combined to obtain the risk prediction result of the target user.

[0060] The server may allocate resources to the target user based on the risk prediction result of the target user. The resources may include data resources such as services and products provided or recommended to the user. For example, in a loan business scenario, a user with a higher risk level may be given a lower loan amount or refused to be provided with a loan; while a user with a lower risk level may be given a higher loan amount. Alternatively, the amount of resources allocated to the user at each stage may be dynamically adjusted based on the risk prediction result of the user.

[0061] In some other embodiments, the specified information set may include at least four benchmark sample sets. The benchmark sample set that is not used to construct the sub-classifier may be used as a verification sample set. Then, the verification sample set may be used to verify the optimized sub-classifiers, and the weights of the optimized sub-classifiers may be determined based on the verification results. Accordingly, the optimized sub-classifiers and the weights of the optimized sub-classifiers may be used to predict the risk of the target user. The optimized classification model may be further verified using the feature data of users with known risks as sample data, and then the contribution of each sub-classifier to the risk prediction result of the user may be determined based on the verification results, and then the risk prediction result of the user may be determined by combining the recognition results of each sub-classifier, thereby further improving the accuracy of user analysis and identification.

[0062] In some embodiments, the weight of each sub-classifier may be determined in the following manner.

[0063] weight i =N i / N val

[0064] Among them, weight i represents the weight of each sub-classifier, where i = 1, 2, ..., n, n ≥ 3; N i Represents the number of samples correctly classified by the i-th subclassifier, N val Indicates the number of samples in the validation sample set. This method can more accurately and efficiently determine the contribution of each sub-classifier, thereby improving the accuracy of recognition. Of course, other weight determination methods can also be used, which are not limited here.

[0065] like Figure 2 As shown, through the above method, based on the optimized sub-classifiers and their corresponding weights, a risk prediction model can be constructed, and then, the risk of the target user can be predicted based on the constructed risk prediction model. In some embodiments, the risk prediction model can be constructed based on the following discriminant function to predict the risk of the target user:

[0066]

[0067] Among them, F(x) represents the discriminant function for risk prediction of the target user, f i (·) represents the discriminant function of the i-th sub-classifier, and x is the feature data used to characterize the risk characteristics of the target user.

[0068] In an implementation scenario provided in the embodiments of this specification, taking the user as an enterprise user as an example, the solution provided in the above embodiments is described as follows.

[0069] First, obtain the enterprise-related characteristic information from the data warehouse, including basic information of the legal person, legal person asset information, legal person loan information, legal person transaction information, etc., then perform data preprocessing and feature extraction, and use the basic information characteristics of the corporate legal person and the deposit and loan time series information characteristics to construct the specified information set.

[0070] Data selection. The relevant characteristics of corporate deposits can be divided into four categories: basic information of legal persons, legal person asset information, legal person loan information, and legal person transaction information. The data scope can be determined by category, thereby determining the data table involved.

[0071] Data preprocessing. Observe the data columns in the data table that involve corporate deposits and loan information. Concatenate the data columns involving deposit information in different tables according to user ID and time to form the original features. For columns with incorrect data types, convert them to the correct data type first. For example, the data type should be numeric, but a pseudo-string type is set in the data table. You can determine whether it is wrong based on the meaning of the data column name and convert the wrong ones. For columns with missing values, fill them in a certain way, such as missing values ​​of numerical features, fill them with "0", and missing values ​​of non-numeric features, fill them with "-1".

[0072] Then, feature extraction can be performed as follows.

[0073] Feature conversion: For categorical features, such as economic nature and enterprise size, they are One-Hot encoded, and some numerical features with particularly large ranges are bucketed.

[0074] In some embodiments, the feature data may include time series aggregation features and time series historical features. The time series aggregation features may refer to data obtained by extracting features from the user's specified information based on different time dimensions and time series feature extraction algorithms. The time series historical features may include time series distribution data obtained by statistics of the user's specified information based on different time dimensions. The time dimensions may include, for example, the previous month, the previous two months, the previous three months, etc., as well as the previous second month, the previous third month, the previous fourth month, etc. The time series feature extraction algorithm may include, for example, mean value, variance, standard deviation, etc. By further combining the time series feature information to construct feature data, the features of users of different churn types can be more accurately characterized, thereby improving the accuracy of system resource allocation. By performing time series feature analysis on information in the user's information that fluctuates more significantly over time, a horizontal analysis of user features can be achieved, thereby greatly improving the accuracy of user stability prediction.

[0075] In some implementations, the time series aggregation feature F agg It can be extracted in the following way:

[0076] F agg =[f(feature) time ,time=1,2,3,4,5,6,1-2,1-3,1-4,1-5,1-6]

[0077] f() takes Mean() average, Max() maximum, Min() minimum, Std() standard deviation respectively, and the time periods are the previous month, previous two months, previous three months, previous four months, previous five months, previous six months, previous second month, previous third month, previous fourth month, previous fifth month, previous sixth month. Accordingly, each deposit and loan feature derives 44-dimensional time series aggregation features.

[0078] Time series historical features F his It can be extracted in the following way:

[0079] F his =[feature time ,time=1,2,3,4,5,6]

[0080] The time periods are the first month, the second month, the third month, the fourth month, the fifth month, and the sixth month. Accordingly, each deposit and loan feature can derive a 6-dimensional time series historical feature.

[0081] By the above method, based on the feature information at different time nodes, the time series feature information is constructed, so that the model can better take into account the previous feature information when learning the features of the current time node. Then, the feature data extracted by the above method can be used as feature data to construct a specified information set. The specified information set can include a benchmark sample and a prediction sample. Then, based on the above steps S20-S28, the risk prediction model can be constructed. Figure 3 As shown, after the risk prediction model is constructed, the server can perform risk prediction on the target user based on the constructed risk prediction model, and allocate resources to the target user based on the risk prediction result.

[0082] Although there are methods for model construction that combine user data with known risks and unknown risks, such as clustering first, and then assigning risk prediction results to prediction samples with unknown risks based on the known risk benchmark samples in the cluster, the recognition results are still overly dependent on the number and quality of benchmark samples. Afterwards, when the benchmark samples and the prediction samples with risk prediction results are used together to build a risk prediction model, the accuracy of the prediction of the constructed risk prediction model is also difficult to guarantee, given that the accuracy of the previous prediction results is difficult to guarantee. At the same time, there is no difference in the contribution of benchmark samples and prediction samples to model training. In fact, the risk prediction results of prediction samples are often not very credible, which may further reduce the generalization performance of the final constructed model.

[0083] In the above embodiment of the specification, the non-overlapping benchmark samples are divided into at least four parts, wherein at least three parts are used to construct at least three sub-classifiers, and the other benchmark samples are used as verification data to determine the weights of the sub-classifiers in the final prediction model. During the training process, the three sub-classifiers vote on the recognition results of the prediction samples, and update the recognition results of the prediction samples and the misclassification costs of the prediction samples according to the results of the vote. In the process of model optimization, the recognition results and misclassification costs of the prediction samples are iteratively updated, and the recognition results and misclassification costs of the prediction samples are used to construct the cost-sensitive pseudo-empirical loss, and the three sub-classifiers are iteratively optimized by minimizing the experience loss and the cost-sensitive pseudo-empirical loss. Finally, the classification effect of each sub-classifier is verified using the verification data, and the weight of the sub-classifier is calculated using the results of the verification data. The final model is obtained by integrating the weights of each sub-classifier, which is conducive to reducing the misleading of the "correct minority" by the "wrong majority". In the above manner, the recognition results and misclassification costs of the prediction samples are dynamically updated during the training process, so that the model can fully learn the data distribution information contained in a variety of samples, and then a model with stronger generalization ability can be constructed.

[0084] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. For details, please refer to the description of the above-mentioned related processing related embodiments, and no further description is given here.

[0085] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0086] Based on the system resource allocation method described above, one or more embodiments of this specification also provide a system resource allocation device. The device may include a system, software (application), module, component, server, etc. that uses the method described in the embodiments of this specification and is combined with necessary implementation hardware. Based on the same innovative concept, the device in one or more embodiments provided in the embodiments of this specification is as described in the following embodiments. Since the implementation scheme of the device to solve the problem is similar to the method, the implementation of the specific device in the embodiments of this specification can refer to the implementation of the aforementioned method, and the repetitions will not be repeated. As used below, the term "unit" or "module" can be a combination of software and / or hardware that implements predetermined functions. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceived. Specifically, Figure 4 A schematic diagram showing the module structure of an embodiment of a system resource allocation device provided in the specification is shown in FIG. Figure 4 As shown, applied to a server, the device may include the following modules.

[0087] The data acquisition module 102 can be used to obtain a specified information set and a label set having feature data for characterizing user risk characteristics; the specified information set includes at least a predicted sample set and three benchmark sample sets; the label set includes the risk category corresponding to each benchmark sample in the benchmark sample set.

[0088] The classifier construction module 104 may be used to construct at least three sub-classifiers using at least three benchmark sample sets.

[0089] The classification module 106 can be used to perform risk classification on the prediction samples in the prediction sample set using the at least three sub-classifiers respectively, and obtain the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples; wherein the misclassification cost is used to characterize the possibility of the prediction samples being misclassified.

[0090] The classifier optimization module 108 can be used to optimize each sub-classifier according to the prediction sample set and the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification cost of the prediction samples to obtain optimized sub-classifiers.

[0091] The resource allocation module 110 may be configured to allocate system resources to the target user based on the risk prediction results of the optimized sub-classifiers for the target user.

[0092] In some other embodiments, the specified information set may include at least four benchmark sample sets; the benchmark sample sets that are not used to construct the sub-classifiers are used as verification sample sets. The device may also include a weight configuration module 112. The weight configuration module 112 may be used to verify the optimized sub-classifiers using the verification sample sets, and determine the weights of the optimized sub-classifiers based on the verification results. Accordingly, the risk prediction module 110 may also be used to perform risk prediction on the target user using the optimized sub-classifiers and the weights of the optimized sub-classifiers.

[0093] It should be noted that the above-mentioned device may also include other implementations according to the description of the method embodiment. The specific implementation methods can refer to the description of the relevant method embodiment, and will not be described one by one here.

[0094] This specification also provides a system resource allocation device, which can be applied to a separate system resource allocation system or to a variety of computer data processing systems. The system can be a separate server, or it can include a server cluster, system (including distributed system), software (application), actual operation device, logic gate circuit device, quantum computer, etc. that uses one or more of the methods or one or more embodiments of this specification and a terminal device combined with necessary implementation hardware. In some embodiments, the device may include at least one processor and a memory for storing processor executable instructions, and when the instructions are executed by the processor, the steps of the method described in any one or more of the above embodiments are implemented.

[0095] The memory may include a physical device for storing information, which is usually to digitize the information and then store it in a medium using electrical, magnetic or optical means. The storage medium may include: a device that uses electrical energy to store information, such as various memories, such as RAM, ROM, etc.; a device that uses magnetic energy to store information, such as a hard disk, a floppy disk, a magnetic tape, a magnetic core memory, a magnetic bubble memory, a USB flash drive; a device that uses optical means to store information, such as a CD or a DVD. Of course, there are other types of readable storage media, such as quantum memory, graphene memory, etc.

[0096] It should be noted that the above-mentioned device may also include other implementation methods according to the description of the method or device embodiment. The specific implementation method can refer to the description of the relevant method embodiment, which will not be described one by one here.

[0097] It should be noted that the embodiments of this specification are not limited to the situations that must comply with the standard data model / template or the embodiments of this specification. Certain industry standards or slightly modified implementation plans based on the implementation described in the custom method or the embodiment can also achieve the same, equivalent or similar, or predictable implementation effects after deformation of the above-mentioned embodiments. The embodiments obtained by using these modified or deformed data acquisition, storage, judgment, processing methods, etc. can still fall within the scope of the optional implementation plans of this specification.

[0098] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. In the description of this specification, the description of the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of this specification. In this specification, the schematic representation of the above terms does not necessarily target the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, in the absence of contradiction, a person skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0099] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A system resource allocation method, characterized in that: Applied to a server, the method comprises: Acquire a specified information set and a label set having feature data for characterizing user risk features; the specified information set includes at least a predicted sample set and three benchmark sample sets; the label set includes the risk category corresponding to each benchmark sample in the benchmark sample set; At least three sub-classifiers are constructed using at least three benchmark sample sets and the label sets; Using the at least three sub-classifiers to perform risk classification on the prediction samples in the prediction sample set, and obtaining the risk classification results of the prediction samples relative to each sub-classifier and the misclassification cost of the prediction samples; wherein the misclassification cost is used to characterize the possibility that the prediction samples are misclassified; Optimizing each sub-classifier according to the prediction sample set, the risk classification results of the prediction samples relative to each sub-classifier, and the misclassification costs of the prediction samples to obtain optimized sub-classifiers, so as to allocate system resources to the target user based on the risk prediction results of the target user by the optimized sub-classifiers; Among them, the optimization of each sub-classifier according to the prediction sample set and the risk classification results of the prediction sample relative to each sub-classifier and the misclassification cost of the prediction sample includes: for any prediction sample, counting the number of sub-classifiers corresponding to each risk classification result of the prediction sample, and marking the risk classification result with the largest number of sub-classifiers; taking the ratio of the number of sub-classifiers corresponding to the marked risk classification result to the number of all sub-classifiers as the misclassification cost of the corresponding prediction sample; accordingly, optimizing each sub-classifier according to the prediction sample set and the marked risk classification results corresponding to the prediction sample and the misclassification cost of the prediction sample; specifically, optimizing each sub-classifier based on the following objective function: Among them, L(X n ,X u ) is the objective function, X n are at least three benchmark sample sets used to build the classifier, X i is any benchmark sample set used to build a classifier, Y i The benchmark sample set X i The risk category of each benchmark sample in the set, where i = 1, 2, ... ,n,n≥3;X u is the prediction sample set, Y u is a set of risk classification results of the markers corresponding to each prediction sample in the prediction sample set, f i represents the sub-classifier constructed based on the corresponding benchmark sample set, f i ( · ) represents the discriminant function of the sub-classifier; R emp ( · ) represents the experience loss, W u Represents the set of misclassification costs of each prediction sample in the prediction sample set, R teg ( · ) represents the L2 regularization loss, and α and β are hyperparameters.

2. The method according to claim 1, characterized in that The specified information set includes at least four benchmark sample sets; the benchmark sample sets not used to construct the sub-classifiers are used as verification sample sets; Using the verification sample set to verify each optimized sub-classifier, and determining the weight of each optimized sub-classifier according to the verification result; Accordingly, the optimized sub-classifiers and the optimized weights of the sub-classifiers are used to perform risk prediction on the target user.

3. The method according to claim 2, characterized in that Determining the weights of each optimized sub-classifier according to the verification result includes: weight i =N i / N val Among them, weight i Represents the weight of each sub-classifier, where i = 1, 2, … ,n,n≥3;N i Indicates the number of samples correctly classified by the i-th subclassifier, N val Indicates the number of samples in the validation sample set.

4. The method according to claim 3, characterized in that Risk prediction of target users is performed based on the following discriminant function: Among them, F(x) represents the discriminant function for risk prediction of the target user, f i (·) represents the discriminant function of the i-th sub-classifier, and x is the feature data used to characterize the risk characteristics of the target user.

5. The method according to claim 1, characterized in that The feature data includes time series aggregation features and time series historical features; wherein the time series aggregation features refer to data obtained by extracting features from the user's specified information based on different time dimensions and a time series feature extraction algorithm; the time series historical features include time series distribution data obtained by statistics of the user's specified information based on different time dimensions.

6. A system resource allocation device, characterized in that: Applied to a server, the device comprises: A data acquisition module, used to acquire a specified information set and a label set having feature data for characterizing user risk features; the specified information set includes at least a predicted sample set and three benchmark sample sets; the label set includes the risk category corresponding to each benchmark sample in the benchmark sample set; A classifier construction module, used to construct at least three sub-classifiers using at least three benchmark sample sets; A classification module, used to perform risk classification on the prediction samples in the prediction sample set using the at least three sub-classifiers respectively, to obtain the risk classification results of the prediction samples relative to each sub-classifier and the misclassification cost of the prediction samples; wherein the misclassification cost is used to characterize the possibility that the prediction samples are misclassified; A classifier optimization module, used to optimize each sub-classifier according to the prediction sample set and the risk classification results of the prediction samples relative to each sub-classifier and the misclassification cost of the prediction samples, to obtain each optimized sub-classifier; A resource allocation module, used for allocating system resources to the target user based on the risk prediction results of the target user by each optimized sub-classifier; The classifier optimization module is specifically used to count the number of sub-classifiers corresponding to each risk classification result of any prediction sample, and mark the risk classification result with the largest number of sub-classifiers; the ratio of the number of sub-classifiers corresponding to the marked risk classification result to the number of all sub-classifiers is used as the misclassification cost of the corresponding prediction sample; accordingly, each sub-classifier is optimized according to the prediction sample set and the marked risk classification result corresponding to the prediction sample and the misclassification cost of the prediction sample; specifically, each sub-classifier is optimized based on the following objective function: Among them, L(X n ,X u ) is the objective function, X n are at least three benchmark sample sets used to build the classifier, X i is any benchmark sample set used to build a classifier, Y i The benchmark sample set X i The risk category of each benchmark sample in the set, where i = 1, 2, ... ,n,n≥3;X u is the prediction sample set, Y u is a set of risk classification results of the markers corresponding to each prediction sample in the prediction sample set, f i represents the sub-classifier constructed based on the corresponding benchmark sample set, f i ( · ) represents the discriminant function of the sub-classifier; R emp ( · ) represents the experience loss, W u Represents the set of misclassification costs of each prediction sample in the prediction sample set, R teg ( · ) represents the L2 regularization loss, and α and β are hyperparameters.

7. The device according to claim 6, characterized in that The specified information set includes at least four benchmark sample sets; the benchmark sample sets not used to construct the sub-classifiers are used as verification sample sets; the device also includes: A weight configuration module, used to verify each optimized sub-classifier using the verification sample set, and determine the weight of each optimized sub-classifier according to the verification result; Correspondingly, the risk prediction module is also used to use the optimized sub-classifiers and the optimized weights of the sub-classifiers to perform risk prediction on the target user.

8. A system resource allocation device, characterized in that: Applied to a server, the device includes at least one processor and a memory for storing processor executable instructions, and the instructions, when executed by the processor, implement the steps of the method described in any one of claims 1 to 5.

Images