Method, device and server for determining transaction risk

Abstract

The specification provides a transaction risk determination method, device and server, applied to the field of artificial intelligence technology. Based on the method, before implementation, a preset feature engineering model capable of automatically performing feature engineering processing can be obtained through reinforcement learning training based on the UCB algorithm using sample data; then the preset feature engineering model is used to process the sample data to automatically obtain sample feature data with good interpretability, and then the preset classification model can be trained using the sample feature data through machine learning; in the implementation, the preset feature engineering model can be used to automatically perform feature engineering processing on the target data of the target user to obtain corresponding target feature data; then the preset classification model is used to process the target feature data to accurately determine whether the target user has a transaction risk. Thus, the feature engineering processing can be automatically and efficiently completed, and whether the target user has a transaction risk can be accurately determined.

Description

TECHNICAL FIELD

[0001] The present specification belongs to the technical field of artificial intelligence, and particularly relates to a transaction risk determination method, device and server. BACKGROUND

[0002] When determining the transaction risk of a user by using artificial intelligence technology, it is often necessary to first perform feature engineering on the relevant data of the user to obtain feature data that is suitable for predicting the risk and has good effect.

[0003] Usually, when performing feature engineering, one method is to rely on the experience of technical personnel to manually perform feature engineering. However, this method has relatively high processing cost and relatively low processing efficiency, and is prone to introduce errors caused by human subjective factors. Another method is to rely on the convolution layer in the neural network to perform feature engineering. However, the feature data obtained based on this method often has relatively poor interpretability and is not suitable for machine learning.

[0004] At present, no effective solution has been proposed for the above problems. SUMMARY

[0005] The present specification provides a transaction risk determination method, device and server, which can automatically and efficiently complete feature engineering processing to obtain feature data with good interpretability, and accurately determine whether a target user has a transaction risk based on the above feature data.

[0006] The present specification provides a transaction risk determination method, comprising:

[0007] obtaining target data about a target user;

[0008] performing feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained by reinforcement learning based on a UCB algorithm using sample data;

[0009] processing the target feature data by using a preset classification model to obtain a target classification result; wherein the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model;

[0010] determining whether the target user has a transaction risk according to the target classification result.

[0011] In one embodiment, the target data includes attribute data of the target user and / or behavior data of the target user.

[0012] In one embodiment, before obtaining the target data about the target user, the method further comprises:

[0013] Obtaining sample data;

[0014] According to the UCB algorithm, a decision function for selecting a feature engineering operator is constructed; and a preset feature engineering operator set is configured; wherein the preset feature engineering operator set contains a plurality of feature engineering operators;

[0015] According to the decision function and the preset feature engineering operator set, the agent is controlled to perform multiple iteration processes based on the exploration and utilization mode using the sample data, to obtain a required preset feature engineering model.

[0016] In one embodiment, the feature engineering operator includes at least one of the following: a preset missing value median processing operator, a preset one-hot encoding operator for a category type feature, and a preset data normalization and conversion operator.

[0017] In one embodiment, according to the UCB algorithm, the decision function for selecting a feature engineering operator is constructed, including:

[0018] The decision function is constructed according to the following formula:

[0019]

[0020] Wherein a t is the selected feature engineering operator in the current iteration process, t is the iteration number corresponding to the current iteration process, A is the preset feature engineering operator set, B is the feature engineering operator updated in the last iteration process, Q t (a) represents the reward data generated by using the feature engineering operator selected in the last iteration process in the current iteration process, U t (a) represents the reward data generated by exploring a feature engineering operator in the current iteration process, c is a weight value, N t (a) is the number of explorations of the feature engineering operator.

[0021] In one embodiment, according to the decision function and the preset feature engineering operator set, the agent is controlled to perform multiple iteration processes based on the exploration and utilization mode using the sample data, including:

[0022] The current iteration process in the multiple iteration processes is performed in the following manner:

[0023] Obtaining the reward data generated by the feature engineering operator selected in the last iteration process, and the target feature engineering operator set in the last iteration process;

[0024] Solving the decision function by using the reward data and the sample data generated by the selected feature engineering operator in the last time, so that the agent determines the selected feature engineering operator in the current time from the preset feature engineering operator set;

[0025] According to the selected feature engineering operator in the current time, updating the target feature engineering operator set in the last time to obtain the target feature engineering operator set in the current time;

[0026] Using the target feature engineering operator set in the current time to perform feature engineering processing on the sample data to obtain the sample feature data in the current time;

[0027] According to the sample feature data in the current time, determining the reward data generated by the selected feature engineering operator in the current time.

[0028] In one embodiment, according to the sample feature data in the current time, determining the reward data generated by the selected feature engineering operator in the current time, comprises:

[0029] Using the initial classification model to process the sample feature data in the current time to obtain the classification result in the current time;

[0030] According to the classification result in the current time, determining the reward data generated by the selected feature engineering operator in the current time by cross-validation.

[0031] In one embodiment, after updating the target feature engineering operator set in the last time according to the selected feature engineering operator in the current time to obtain the target feature engineering operator set in the current time, the method further comprises:

[0032] Detecting whether the iteration number corresponding to the current iteration processing is equal to the preset iteration number threshold; and / or, detecting whether the decision function in the current iteration processing converges;

[0033] In the case of determining that the iteration number corresponding to the current iteration processing is equal to the preset iteration number threshold; and / or, determining that the decision function in the current iteration processing converges, determining to obtain the required preset feature engineering model.

[0034] In one embodiment, after determining to obtain the required preset feature engineering model, the method further comprises:

[0035] Obtaining an initial classification model;

[0036] Using the preset feature engineering model to process the sample data to obtain the sample feature data;

[0037] Using the sample feature data to perform machine learning on the initial classification model to obtain the required preset classification model.

[0038] In an embodiment, the initial classification model comprises an XGboost-based structured model.

[0039] The specification also provides a data processing method, comprising:

[0040] obtaining sample data associated with a target application scenario;

[0041] According to the UCB algorithm, a decision function for selecting a feature engineering operator is constructed; and a preset feature engineering operator set is configured; wherein the preset feature engineering operator set contains a plurality of feature engineering operators;

[0042] According to the decision function and the preset feature engineering operator set, the agent is controlled to perform multiple iteration processing based on an exploration and utilization mode using the sample data, to obtain a preset feature engineering model for the target application scenario and meeting the requirements.

[0043] The specification also provides a transaction risk determination device, comprising:

[0044] An acquisition module is configured to acquire target data about a target user;

[0045] A first processing module is configured to perform feature engineering processing on the target data of the target user using a preset feature engineering model, to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained based on the UCB algorithm through reinforcement learning using sample data;

[0046] A second processing module is configured to process the target feature data using a preset classification model, to obtain a target classification result; wherein the preset classification model is a model trained through machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data using the preset feature engineering model;

[0047] A determination module is configured to determine whether the target user has a transaction risk according to the target classification result.

[0048] The specification also provides a server comprising a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the related steps of the transaction risk determination method.

[0049] The specification also provides a computer-readable storage medium having computer instructions stored thereon, wherein the instructions are executed by a processor to implement the related steps of the transaction risk determination method.

[0050] The specification also provides a computer program product comprising a computer program which, when executed by a processor, implements the related steps of the transaction risk determination method.

[0051] Based on the transaction risk determination method, device and server provided in the specification, before implementation, a preset feature engineering model capable of automatically performing feature engineering processing can be obtained by reinforcement learning training based on the UCB algorithm using sample data; then the sample feature data with better interpretability and suitable for machine learning can be automatically obtained by processing the sample data using the preset feature engineering model; and then the preset classification model with high precision can be obtained by machine learning using the above sample feature data. In implementation, the target feature data can be obtained by automatically performing feature engineering processing on the target data of the target user using the preset feature engineering model; and then the preset classification model is used to process the target feature data to accurately determine whether the target user has a transaction risk. In this way, on the one hand, feature engineering does not need to rely on the experience of technical personnel to be performed manually, but the preset feature engineering model is directly used to automatically and efficiently complete feature engineering processing, thereby effectively reducing processing cost and improving processing efficiency. On the other hand, the target feature data obtained by processing using the above preset feature engineering model has better interpretability and is more suitable for machine learning than the feature data obtained by a neural network, and does not introduce errors caused by human subjective factors, so that the preset classification model can be used to determine whether the target user has a transaction risk based on the above feature data, thereby reducing errors during determination. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the specification, the drawings needed in the embodiments will be briefly introduced as follows. The drawings in the following description are only some embodiments described in the specification, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0053] Figure 1 is a flowchart of a transaction risk determination method provided by an embodiment of the specification;

[0054] Figure 2 is a schematic diagram of an embodiment of the transaction risk determination method provided by the specification in one scenario example;

[0055] Figure 3 is a schematic diagram of an embodiment of the transaction risk determination method provided by the specification in one scenario example;

[0056] Figure 4is a schematic diagram of an embodiment of a method for determining transaction risk provided by an embodiment of the present specification in a scenario example;

[0057] Figure 5 is a schematic diagram of an embodiment of a method for determining transaction risk provided by an embodiment of the present specification in a scenario example;

[0058] Figure 6 is a schematic diagram of a data processing method provided by an embodiment of the present specification;

[0059] Figure 7 is a schematic diagram of the structural composition of a server provided by an embodiment of the present specification;

[0060] Figure 8 is a schematic diagram of the structural composition of a transaction risk determination device provided by an embodiment of the present specification;

[0061] Figure 9 is a schematic diagram of the structural composition of a data processing device provided by an embodiment of the present specification. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the technical solutions in the present specification, the technical solutions in the present specification will be described clearly and completely in conjunction with the drawings in the embodiments of the present specification. Obviously, the described embodiments are only part of the embodiments of the present specification, not all. Based on the embodiments in the present specification, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present specification.

[0063] Referring to Figure 1 , the present specification provides a method for determining transaction risk. Wherein, the method is specifically applied to the server side. In specific implementation, the method can include the following contents:

[0064] S101: obtaining target data about a target user;

[0065] S102: performing feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein, the preset feature engineering model is a model trained by reinforcement learning based on UCB algorithm using sample data;

[0066] S103: processing the target feature data by using a preset classification model to obtain a target classification result; wherein, the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model.

[0067] S104: Based on the target classification results, determine whether the target user has transaction risks.

[0068] In some embodiments, the method for determining transaction risk described above can be specifically applied to the server side. The server may be configured with a pre-trained preset feature engineering model and a preset classification model.

[0069] Among them, see Figure 2 As shown, the aforementioned server may specifically include a backend server applied to one side of a network platform (e.g., the network platform of XX Bank), capable of data transmission, data processing, and other functions. Specifically, the server may be, for example, an electronic device with data processing, storage, and network interaction capabilities. Alternatively, the server may be a software program running on the electronic device, providing support for data processing, storage, and network interaction. In this embodiment, the number of servers is not specifically limited. The server may be a single server, several servers, or a server cluster formed by several servers.

[0070] When a target user goes to a branch to apply for related services (e.g., a loan service offered by XX Bank for users with good credit and no transaction risk), the branch staff needs to first determine whether the target user meets the conditions for applying for the service, such as whether the target user has any transaction risk.

[0071] For specific implementation, please refer to Figure 2 As shown, staff can initiate a transaction risk assessment request for the target user to the server via a terminal.

[0072] Specifically, the aforementioned terminal may include a front-end device used by staff to perform functions such as data collection and data transmission. Specifically, the terminal may be an electronic device such as a desktop computer, tablet computer, laptop computer, or smartphone. Alternatively, the terminal may be a software application that runs on the aforementioned electronic device. For example, it could be an e-banking client app installed and running on a smartphone.

[0073] The aforementioned transaction risk determination request may at least include the target user's identifier. This identifier can be understood as identifying information that indicates the target user. Examples include the target user's name, account name, user ID, etc. Of course, the user identifiers listed above are merely illustrative. In practice, depending on the specific application scenario and processing requirements, the user identifier may also include other types of identifying information. This specification does not limit this.

[0074] It should be noted that the information data related to the user in the present specification is obtained and used with the user's knowledge and consent. And the acquisition, storage, use, processing, etc. of the above information data complies with the relevant provisions of national laws and regulations.

[0075] Correspondingly, the server can receive and respond to the transaction risk determination request, obtain the target data of the target user; and utilize the transaction risk determination method provided in the present specification, call the preset feature engineering model and the preset classification model to determine whether the target user exists transaction risk by processing the target data of the target user, and obtain the corresponding risk determination result; and then feed back the risk determination result to the terminal.

[0076] The terminal shows the risk determination result to the staff. The staff determines whether the target user exists transaction risk according to the risk determination result, and further determines whether to handle the related business for the target user.

[0077] In some embodiments, the server can receive and respond to the above transaction risk determination request, and obtain the target data of the target user according to the user identifier of the target user.

[0078] In some embodiments, the target data can specifically include attribute data of the target user and / or behavior data of the target user, etc.

[0079] The attribute data of the target user specifically can include age, residence city, education, occupation, etc. of the target user. The behavior data of the target user specifically can include shopping records, transfer records, credit repayment records, etc. of the target user in a preset time period (for example, the last week, etc.).

[0080] In specific implementation, the server can query the user database of the target user according to the user identifier of the target user to obtain the attribute data of the target user; and / or query the historical transaction data record of the target user according to the user identifier of the target user to obtain the behavior data of the target user.

[0081] In some embodiments, the preset feature engineering model can be understood as a model that can automatically perform feature engineering processing on data to extract more accurate and better explanatory feature data for classification.

[0082] The preset feature engineering model can be a model trained by reinforcement learning based on UCB algorithm using sample data in advance. The specific training method of the preset feature engineering model will be described later.

[0083] Specifically, the preset feature engineering model is further configured with a target feature engineering operator set. The target feature engineering operator set includes a plurality of feature engineering operators suitable for the target application scenario (e.g., a transaction risk prediction scenario) selected in the process of reinforcement learning based on the UCB algorithm.

[0084] The UCB (Upper Confidence Bound) algorithm can also be referred to as the maximum confidence upper bound algorithm. Based on the UCB algorithm, the upper limit of a suitable confidence interval can be taken as an estimate under a certain confidence level. The UCB algorithm can be used to construct a decision function for the target application scenario, so as to filter out feature engineering operators suitable for the target application scenario through the decision function.

[0085] The reinforcement learning (RL) can be a machine learning method for describing and solving the problem of how an agent (e.g., denoted as agent) learns a strategy to maximize the reward or achieve a specific goal in the process of interaction with the environment.

[0086] In the embodiment, by introducing and using reinforcement learning, the model can focus on online learning and try to maintain a good balance between exploration and exploitation during training. Unlike conventional supervised learning and unsupervised learning, the reinforcement learning does not require any data to be given in advance, but can obtain learning information by receiving and according to the reward data (feedback or income, etc.) (e.g., denoted as reward) of the action of the environment, and then automatically update the model parameters based on the learning information, perform model training, and finally find the combination of feature engineering operators most suitable for the target application scenario to get the most reward data.

[0087] In some embodiments, referring to Figure 3 As shown, in specific implementation, the server can input the target data of the target user into the preset feature engineering model. The preset feature engineering model can use the feature engineering operators in the target feature engineering operator set to perform corresponding feature engineering processing on the target data, to obtain corresponding target feature data as the output of the preset feature engineering model.

[0088] In some embodiments, the preset classification model can be understood as a classification model capable of determining whether there is a transaction risk based on the input feature data.

[0089] The preset classification model can be obtained by using the trained preset feature engineering model through machine learning training. The specific training method of the preset classification model will be described later.

[0090] In some embodiments, referring to Figure 3 As shown, in actual implementation, the server can input the target feature data of the target user into the preset classification model, and run the preset classification model. The preset classification model obtains a probability value about the existence of transaction risk of the target user by processing the target feature data of the target user, as the target classification result output by the preset classification model.

[0091] In some embodiments, the above determining whether the target user exists transaction risk according to the target classification result, in actual implementation, can include the following contents: obtaining the probability value output by the preset classification model according to the target classification result; comparing the probability value with a preset probability threshold; obtaining a corresponding comparison result. According to the comparison result, in the case that the probability value is greater than the preset probability threshold, it is determined that the target user exists transaction risk. On the contrary, according to the comparison result, in the case that the probability value is less than or equal to the preset probability threshold, it is determined that the target user does not exist transaction risk. Wherein, the above preset probability threshold can be obtained by learning and statistics on a large amount of historical data in advance.

[0092] In some embodiments, after determining whether the target user exists transaction risk according to the target classification result, the method, in actual implementation, can further include the following contents: in the case that it is determined that the target user exists transaction risk, setting a risk label for the target user; according to the risk label, performing corresponding data processing on the target user. Specifically, for example, according to the risk label, the transaction behavior of the target user can be further monitored more finely and more targetedly; and / or, providing the target user with a business service matched with the risk label, etc.

[0093] In some embodiments, referring to Figure 4 As shown, before obtaining the target data about the target user, the method, in actual implementation, can further include the following contents:

[0094] S1: obtaining sample data;

[0095] S2: constructing a decision function for selecting feature engineering operators according to the UCB algorithm; and configuring a preset feature engineering operator set; wherein, the preset feature engineering operator set contains a plurality of feature engineering operators;

[0096] S3: according to the decision function and the preset feature engineering operator set, controlling the agent to perform multiple iteration processing based on the exploration and utilization mode by using the sample data, so as to obtain a preset feature engineering model meeting the requirements.

[0097] The sample data can be, for example, attribute data and / or behavior data of users in the past, and the like.

[0098] According to the above embodiments, a preset feature engineering model capable of automatically performing feature engineering for a target application scenario can be automatically trained by reinforcement learning according to the UCB algorithm.

[0099] In some embodiments, the preset feature engineering operator set can include a plurality of feature engineering operators related to the target application scenario. Specifically, the plurality of feature engineering operators included in the preset feature engineering operator set can be determined as follows: obtaining a plurality of feature engineering processing records under the target application scenario; and statistically summarizing the plurality of feature engineering processing records to obtain feature engineering operators involved in feature engineering processing under the target application scenario as feature engineering operators included in the preset feature engineering operator set.

[0100] In some embodiments, the feature engineering operator can include at least one of the following: a preset missing value median processing operator, a preset one-hot encoding operator for categorical features, a preset data normalization and conversion operator, and the like. Further, the feature engineering operator can also include a preset polynomial feature processing operator, a preset triangular feature processing operator, a preset grouping feature processing operator, a preset binned numerical feature processing operator, and the like.

[0101] Of course, it should be noted that the above-mentioned feature engineering operators are only illustrative. In specific implementations, the preset feature engineering operator set can also include other types of feature engineering operators according to specific application scenarios and processing requirements. The present specification does not limit this.

[0102] In some embodiments, the decision function for selecting a feature engineering operator according to the UCB algorithm includes:

[0103] The decision function is constructed according to the following formula:

[0104]

[0105] wherein a t is the selected feature engineering operator in the current iteration, t is the iteration number corresponding to the current iteration, A is the preset feature engineering operator set, B is the feature engineering operator updated in the last iteration, Q t (a) represents the reward data generated by using the feature engineering operator selected in the last iteration in the current iteration, U t(a) represents the reward data generated by exploring a feature engineering operator in the current iteration process, c is a weight value, N t (a) is the number of times a feature engineering operator is explored.

[0106] The above decision function can be specifically understood as a decision strategy function capable of achieving the current iteration process based on the exploration and exploitation (EE) mode by utilizing the reward data obtained by selecting the feature engineering operator last time and exploring the selection of the feature engineering operator this time.

[0107] Specifically, exploration (Exploration) can refer to selecting a feature engineering operator that has not been used to generate new feature data in the current iteration process. Exploitation (Exploitation) can refer to the maximum benefit (or reward data) obtained when using the feature engineering operator selected last time to generate feature data in the current iteration process.

[0108] Specifically, in the above decision function, two terms are included: Q t (a) and U t (a).

[0109] Q t (a) represents exploitation, which can be the actual distribution of current action-reward, where the action can correspond to the feature engineering operator selected last time, and the reward can correspond to the AUC value after adding or deleting the feature engineering operator last time. AUC (Area Under Curve) can be understood as the area surrounded by the coordinate axis under the ROC curve. The initial value of Q t (a) can be set to 0.

[0110] U t (a) represents exploration, which can be a measure of uncertainty for the current action, where the action can correspond to the feature engineering operator selected this time.

[0111] Based on the above decision function, for any feature engineering operator a in the preset feature engineering operator set, if the number of times N t (a) is unchanged, and as the iteration number t increases, lnt also increases accordingly; correspondingly, the value of U t (a) will become larger, and the probability of the feature engineering operator a being explored and selected will also become larger. Conversely, the same is true.

[0112] By solving the above decision function, the UCB algorithm can be used to select the most suitable feature engineering operator 'a' from the preset set of feature engineering operators by finding the confidence of the action that maximizes the current iteration. t .

[0113] In some embodiments, see Figure 5 As shown, the above-described method, based on the decision function and the preset feature engineering operator set, controls the agent to perform multiple iterative processes using the sample data based on exploration and utilization patterns. In specific implementation, this may include performing the current iteration of the multiple iterations in the following manner:

[0114] S1: Obtain the reward data generated by the previously selected feature engineering operator, and the previous set of target feature engineering operators;

[0115] S2: Using the reward data and sample data generated by the previously selected feature engineering operator, the decision function is solved to enable the agent to determine the currently selected feature engineering operator from the preset feature engineering operator set;

[0116] S3: Based on the currently selected feature engineering operator, update the previous target feature engineering operator set to obtain the current target feature engineering operator set;

[0117] S4: Use the current set of target feature engineering operators to perform feature engineering on the sample data to obtain the current sample feature data;

[0118] S5: Based on the current sample feature data, determine the reward data generated by the currently selected feature engineering operator.

[0119] In some embodiments, the feature engineering operator selected when performing the current iteration based on the above decision function is a. t .

[0120] In some embodiments, updating the previous target feature engineering operator set based on the currently selected feature engineering operator may specifically include: adding the currently selected feature engineering operator to the previous target feature engineering operator set; or deleting the currently selected feature engineering operator from the previous target feature engineering operator set.

[0121] Specifically, for example, the target feature engineering operator set of the previous time can be updated according to the currently selected feature engineering operator to obtain the target feature engineering operator set of the current time.

[0122] If a t It was already selected in the previous iteration, meaning that 'a' already existed in the previous set of target feature engineering operators.t At this time, a corresponding feature engineering operator can be added to the target feature engineering operator set of the last time to update the target feature engineering operator set of the last time to obtain the target feature engineering operator set of the current time. t At this time, a corresponding feature engineering operator can be added to the target feature engineering operator set of the last time to update the target feature engineering operator set of the last time to obtain the target feature engineering operator set of the current time.

[0123] On the contrary, if a t is not selected in the previous iteration processing, that is, a t is not in the target feature engineering operator set of the last time. t At this time, a corresponding feature engineering operator can be added to the target feature engineering operator set of the last time to update the target feature engineering operator set of the last time to obtain the target feature engineering operator set of the current time.

[0124] In addition, in each iteration processing, the sample data to be used in the current iteration processing can also be preprocessed according to the reward data of the last time, so that the current iteration processing can be better performed.

[0125] In the above manner, the agent can automatically use the sample data to perform multiple iteration processing based on the exploration and utilization mode according to the decision function and the preset feature engineering operator set until the iteration stop condition is reached to obtain the preset feature engineering model.

[0126] In the above manner, in the initial stage (for example, the first few iteration processes), the agent is relatively more inclined to exploration based on the decision function because the initial confidence is 0. In this initial stage, the agent mainly selects a feature engineering operator at random based on the decision function and obtains the related benefits. When the action selection is performed again, the uncertainty is reduced because the previous feature engineering operator has been selected at least once, and the probability of being selected again is also relatively reduced, so the agent has a greater probability of selecting from the remaining feature engineering operators in the preset feature engineering operator set.

[0127] After multiple iteration processes are performed in the above manner, the stable stage is entered. In the stable stage, utilization is mainly used, and different feature engineering operators have been selected by exploration, U t (a) values tend to be consistent, and Q t (a) of different feature engineering operators will be different. At this time, the agent is inclined to use the feature engineering operator with relatively high benefits (or reward data) based on the decision function. However, in this stage, after multiple executions, sometimes a feature engineering operator is selected multiple times, and U t (a) of the feature engineering operator decreases, resulting in the phenomenon that other feature engineering operators are selected again.

[0128] In some embodiments, the method can further include the following steps: Figure 4 As shown in the above, the reward data generated by the selected feature engineering operator in the current iteration is determined according to the sample feature data in the current iteration, and in specific implementation, the following content can be included:

[0129] S1: processing the sample feature data in the current iteration by using the initial classification model to obtain the classification result in the current iteration;

[0130] S2: determining the reward data generated by the selected feature engineering operator in the current iteration by performing cross-validation according to the classification result in the current iteration.

[0131] In specific implementation, the average AUC value can be calculated by using the ten-fold cross-validation method according to the classification result in the current iteration, as the reward data (or benefit) generated by the selected feature engineering operator in the previous iteration, for the next iteration.

[0132] In some embodiments, the initial classification model can be specifically understood as an initial model used for training the preset classification model.

[0133] In the determination of the reward data in the above manner, in addition to using the initial classification model, in specific implementation, other existing fixed classification models can also be used to replace the initial classification model to process the sample feature data obtained based on the preset feature engineering model, so as to determine the corresponding reward data.

[0134] In some embodiments, after updating the target feature engineering operator set in the last iteration according to the selected feature engineering operator in the current iteration to obtain the target feature engineering operator set in the current iteration, the method can further include the following content in specific implementation:

[0135] S1: detecting whether the iteration number corresponding to the current iteration processing is equal to the preset iteration number threshold value; and / or, detecting whether the decision function in the current iteration processing converges;

[0136] S2: in the case where it is determined that the iteration number corresponding to the current iteration processing is equal to the preset iteration number threshold value; and / or, it is determined that the decision function in the current iteration processing converges, determining that the preset feature engineering model meeting the requirements is obtained.

[0137] The preset iteration number threshold value can be obtained by statistically summarizing a large amount of historical data corresponding to the target application scenario in advance. Specifically, the preset iteration number threshold value can be set to 5.

[0138] Based on the above embodiments, it can be determined whether the iteration stopping condition is reached by detecting whether the iteration number corresponding to the current iteration processing is equal to the preset iteration number threshold; and / or, detecting whether the decision function in the current iteration processing converges. In the case of determining that the iteration stopping condition is reached, the iteration processing can be stopped; and the preset feature engineering model meeting the requirements is obtained based on the target feature engineering operator set of the current time.

[0139] In some embodiments, after determining that the preset feature engineering model meeting the requirements is obtained, the method can further include the following when implemented:

[0140] S1: obtaining an initial classification model;

[0141] S2: processing sample data by using the preset feature engineering model to obtain sample feature data;

[0142] S3: machine learning of the initial classification model by using the sample feature data to obtain the preset classification model meeting the requirements.

[0143] In some embodiments, the initial classification model can specifically include a model based on the structure of XGboost.

[0144] The XGboost is an improvement on the gradient boosting algorithm, which uses Newton method to solve the extreme value of the loss function, and expands the loss function to the second order by Taylor expansion. In addition, a regularization term is added to the loss function. The XGboost can realize machine learning under the Gradient Boosting framework, and has the advantages of high efficiency, convenience, flexibility, etc.

[0145] When implemented, a model based on the structure of XGboost can be constructed as the initial classification model. The classification model can be a binary classification model.

[0146] Based on the above embodiments, the sample feature data with better interpretability and suitable for machine learning can be obtained by processing the sample data by using the trained preset feature engineering model; and the initial classification model is machine learned by using the sample feature data, so as to efficiently train the preset classification model meeting the requirements with higher precision and better effect.

[0147] As can be seen, based on the transaction risk determination method provided by the embodiments of the present specification, before implementation, a preset feature engineering model capable of automatically performing feature engineering processing can be obtained by reinforcement learning training based on the UCB algorithm using sample data; then the sample feature data with better interpretability and suitable for machine learning is automatically obtained by processing the sample data using the preset feature engineering model; and a preset classification model with high precision is obtained by machine learning using the above sample feature data. In specific implementation, the target feature data corresponding to the target data of the target user can be obtained by automatically performing feature engineering processing on the target data of the target user using the preset feature engineering model; and the preset classification model is used to process the target feature data to accurately determine whether the target user has a transaction risk. In this way, on the one hand, feature engineering does not need to rely on the experience of technical personnel and is manually performed, but the feature engineering processing is directly and efficiently completed using the preset feature engineering model, thereby effectively reducing the processing cost and improving the processing efficiency; on the other hand, the target feature data obtained by processing using the above preset feature engineering model has better interpretability and is more suitable for machine learning than the feature data obtained by a neural network, and does not introduce errors caused by human subjective factors, thereby the preset classification model can be used to determine whether the target user has a transaction risk based on the above feature data, and the error during determination is reduced.

[0148] Referring to Figure 6 The embodiments of the present specification also provide a data processing method. In specific implementation, the method can include the following contents:

[0149] S601: Obtain sample data associated with a target application scenario;

[0150] S602: Construct a decision function for selecting a feature engineering operator according to a UCB algorithm; and configure a preset feature engineering operator set; wherein the preset feature engineering operator set includes a plurality of feature engineering operators;

[0151] S603: According to the decision function and the preset feature engineering operator set, control an agent to perform multiple iteration processing based on an exploration and utilization mode using the sample data, to obtain a preset feature engineering model for the target application scenario and meeting the requirements.

[0152] In some embodiments, the above target application scenario can specifically include a transaction risk prediction scenario, a user type classification scenario, a credit default prediction scenario, or a potential customer prediction scenario, etc. Of course, the above listed application scenarios are only illustrative. In specific implementation, according to specific conditions and processing requirements, the above data processing method can also be applied to other types of application scenarios. In this regard, the present specification is not limited.

[0153] In some embodiments, the records can be processed according to the characteristics of the target application scenario, and a preset feature engineering operator set matching the target application scenario is configured.

[0154] As can be seen from the above, based on the data processing method provided by the embodiments of the present specification, a preset feature engineering model capable of being applied to different target application scenarios and automatically and accurately implementing corresponding feature engineering processing in the target application scenario can be efficiently trained, thereby effectively reducing the overall data processing cost and improving the overall data processing efficiency.

[0155] The embodiments of the present specification also provide a server, comprising a processor and a memory for storing processor-executable instructions, wherein the processor, when implemented, can execute the following steps according to the instructions: obtaining target data about a target user; performing feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained by reinforcement learning based on a UCB algorithm using sample data; processing the target feature data by using a preset classification model to obtain a target classification result; wherein the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model; and determining whether the target user has a transaction risk according to the target classification result.

[0156] In order to be able to more accurately complete the above instructions, with reference to Figure 7 The embodiments of the present specification also provide another specific server, as shown in the figure, wherein the server comprises a network communication port 701, a processor 702 and a memory 703, and the above structures are connected by internal cables so that the structures can perform specific data interaction.

[0157] The network communication port 701 can be specifically used to obtain target data about a target user.

[0158] The processor 702 can be specifically used to perform feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained by reinforcement learning based on a UCB algorithm using sample data; process the target feature data by using a preset classification model to obtain a target classification result; wherein the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model; and determine whether the target user has a transaction risk according to the target classification result.

[0159] The memory 703 can be used to store corresponding instruction programs.

[0160] In the embodiment, the network communication port 701 can be a virtual port bound with different communication protocols, so as to send or receive different data. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for mail data communication. In addition, the network communication port can also be an entity communication interface or a communication chip. For example, it can be a wireless mobile network communication chip such as GSM, CDMA, etc.; it can also be a Wifi chip; it can also be a Bluetooth chip.

[0161] In the embodiment, the processor 702 can be implemented in any appropriate manner. For example, the processor can take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. The present specification is not limited thereto.

[0162] In the embodiment, the memory 703 can include multiple levels, and in a digital system, as long as it can save binary data, it can be a memory; in an integrated circuit, a circuit without a physical form and with a storage function is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, a TF card, etc.

[0163] The embodiment of the present specification also provides another server, including a processor and a memory for storing processor executable instructions, the processor can execute the following steps according to the instructions when implemented: obtaining sample data associated with a target application scenario; constructing a decision function for selecting a feature engineering operator according to a UCB algorithm; and configuring a preset feature engineering operator set; wherein the preset feature engineering operator set contains a plurality of feature engineering operators; according to the decision function and the preset feature engineering operator set, controlling an agent to perform multiple iteration processing based on an exploration and utilization mode using the sample data, to obtain a preset feature engineering model for the target application scenario and meeting the requirements.

[0164] The embodiment of the present specification further provides a computer storage medium based on the above transaction risk determination method, which stores computer program instructions, and when the computer program instructions are executed, the following functions are realized: obtaining target data about a target user; performing feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained by reinforcement learning based on a UCB algorithm using sample data; processing the target feature data by using a preset classification model to obtain a target classification result; wherein the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model; and determining whether the target user has a transaction risk according to the target classification result.

[0165] The embodiment of the present specification further provides a computer storage medium based on the above data processing method, which stores computer program instructions, and when the computer program instructions are executed, the following functions are realized: obtaining sample data associated with a target application scenario; constructing a decision function for selecting a feature engineering operator according to a UCB algorithm; and configuring a preset feature engineering operator set; wherein the preset feature engineering operator set contains a plurality of feature engineering operators; controlling an agent to perform multiple iteration processing based on an exploration and utilization mode by using the sample data according to the decision function and the preset feature engineering operator set, to obtain a preset feature engineering model for the target application scenario and meeting the requirements.

[0166] In the embodiment, the storage medium includes but is not limited to a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk drive (HDD) or a memory card. The memory can be used to store computer program instructions. The network communication unit can be an interface set according to the standard of the communication protocol, used for network connection communication.

[0167] In the embodiment, the functions and effects realized by the program instructions stored in the computer storage medium can be explained by comparing with other embodiments, and will not be described here.

[0168] The embodiment of the present specification provides a computer program product comprising a computer program, which, when executed by a processor, implements the following steps: obtaining target data about a target user; performing feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained by reinforcement learning based on a UCB algorithm using sample data; processing the target feature data by using a preset classification model to obtain a target classification result; wherein the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model; and determining whether the target user has a transaction risk according to the target classification result.

[0169] The embodiment of the present specification provides another computer program product comprising a computer program, which, when executed by a processor, implements the following steps: obtaining sample data associated with a target application scenario; constructing a decision function for selecting a feature engineering operator according to a UCB algorithm; and configuring a preset feature engineering operator set; wherein the preset feature engineering operator set comprises a plurality of feature engineering operators; controlling an agent to perform multiple iteration processing based on an exploration and utilization mode by using the sample data according to the decision function and the preset feature engineering operator set, to obtain a preset feature engineering model for the target application scenario and meeting the requirements.

[0170] Referring to Figure 8 The embodiment of the present specification also provides a transaction risk determination device on the software level, which specifically can include the following structure modules:

[0171] The obtaining module 801 can be specifically used for obtaining target data about a target user;

[0172] The first processing module 802 can be specifically used for performing feature engineering processing on the target data of the target user by using a preset feature engineering model to obtain corresponding target feature data; wherein the preset feature engineering model is a model trained by reinforcement learning based on a UCB algorithm using sample data;

[0173] The second processing module 803 can be specifically used for processing the target feature data by using a preset classification model to obtain a target classification result; wherein the preset classification model is a model trained by machine learning using sample feature data; the sample feature data is obtained by performing feature engineering processing on the sample data by using the preset feature engineering model;

[0174] The determining module 804, in particular, can be configured to determine whether the target user has a transaction risk according to the target classification result.

[0175] In some embodiments, the target data can specifically include attribute data of the target user and / or behavior data of the target user, etc.

[0176] In some embodiments, the apparatus can further include a first training module. Before obtaining the target data of the target user, the first training module can be configured to obtain sample data, construct a decision function for selecting a feature engineering operator according to a UCB algorithm, and configure a preset feature engineering operator set, wherein the preset feature engineering operator set contains a plurality of feature engineering operators, and control an agent to perform multiple iteration processes based on an exploration and utilization mode using the sample data according to the decision function and the preset feature engineering operator set to obtain a required preset feature engineering model.

[0177] In some embodiments, the feature engineering operator can specifically include at least one of the following: a preset missing value median processing operator, a preset one-hot encoding operator for a category type feature, a preset data normalization and conversion operator, etc.

[0178] In some embodiments, the first training module can be implemented according to the following manner to construct a decision function for selecting a feature engineering operator according to a UCB algorithm: the decision function is constructed according to the following formula:

[0179]

[0180] wherein a t is the selected feature engineering operator in the current iteration process, t is the iteration number corresponding to the current iteration process, A is the preset feature engineering operator set, B is the feature engineering operator updated in the last iteration process, Q t (a) represents the reward data generated by using the feature engineering operator selected in the last iteration process in the current iteration process, U t (a) represents the reward data generated by exploring a feature engineering operator in the current iteration process, c is a weight value, N t (a) is the number of times of exploring the feature engineering operator.

[0181] In some embodiments, the first training module, when implemented, can control the agent to perform multiple iteration processing based on the exploration and utilization mode according to the decision function and the set of preset feature engineering operators, in the following manner: performing the current iteration processing in the multiple iteration processing in the following manner: obtaining the reward data generated by the last selected feature engineering operator and the last target set of feature engineering operators; determining the current selected feature engineering operator from the set of preset feature engineering operators by solving the decision function using the reward data generated by the last selected feature engineering operator and the sample data; updating the last target set of feature engineering operators according to the current selected feature engineering operator to obtain a current target set of feature engineering operators; performing feature engineering processing on the sample data using the current target set of feature engineering operators to obtain current sample feature data; and determining the reward data generated by the current selected feature engineering operator according to the current sample feature data.

[0182] In some embodiments, the first training module, when implemented, can determine the reward data generated by the current selected feature engineering operator according to the current sample feature data in the following manner: processing the current sample feature data using the initial classification model to obtain a current classification result; and determining the reward data generated by the current selected feature engineering operator by performing cross-validation according to the current classification result.

[0183] In some embodiments, the first training module, when implemented, after updating the last target set of feature engineering operators according to the current selected feature engineering operator to obtain a current target set of feature engineering operators, can further be configured to detect whether the iteration number corresponding to the current iteration processing is equal to a preset iteration number threshold; and / or, detect whether the decision function in the current iteration processing converges; and determine that the preset feature engineering model meeting the requirements is obtained in a case where it is determined that the iteration number corresponding to the current iteration processing is equal to the preset iteration number threshold; and / or, it is determined that the decision function in the current iteration processing converges.

[0184] In some embodiments, the apparatus can further include a second training module. The second training module, when implemented after the preset feature engineering model meeting the requirements is obtained, can be configured to obtain an initial classification model; process the sample data using the preset feature engineering model to obtain sample feature data; and perform machine learning on the initial classification model using the sample feature data to obtain a preset classification model meeting the requirements.

[0185] In some embodiments, the initial classification model can specifically include a model based on the structure of XGboost.

[0186] Referring to Figure 9 The embodiment of the present specification also provides a data processing apparatus, which can specifically include the following structural modules:

[0187] The acquisition module 901 can be specifically used for acquiring sample data associated with a target application scenario;

[0188] The construction module 902 can be specifically used for constructing a decision function for selecting a feature engineering operator according to a UCB algorithm; and configuring a preset feature engineering operator set; wherein the preset feature engineering operator set contains a plurality of feature engineering operators;

[0189] The iteration module 903 can be specifically used for controlling an agent to perform multiple iteration processing based on an exploration and utilization mode by using the sample data according to the decision function and the preset feature engineering operator set, so as to obtain a preset feature engineering model that meets the requirements for the target application scenario.

[0190] It should be noted that the units, devices or modules and the like described in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described as various modules respectively described in functions. Of course, in the implementation of the present specification, the functions of each module can be implemented in the same software and / or hardware, or the modules implementing the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The above described device embodiments are only illustrative, for example, the division of the units is only a logical function division, and in actual implementation, there can be another division mode, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0191] As can be seen from the above, the transaction risk determination apparatus provided by the embodiment of the present specification can effectively reduce the processing cost and improve the processing efficiency on the one hand, because it does not need to rely on the experience of technical personnel to perform feature engineering manually, but directly uses the preset feature engineering model to automatically and efficiently complete the feature engineering processing. On the other hand, the target feature data obtained by the above-mentioned preset feature engineering model processing has better interpretability and is more suitable for machine learning than the feature data obtained by the neural network, and at the same time, it will not introduce the error influence caused by human subjective factors, so that the preset classification model can be used to determine whether the target user has transaction risk based on the above-mentioned feature data, and the error in the determination is reduced.

[0192] Although the description has been provided with reference to the method operations that are illustrated in the examples or flowcharts, additional or fewer operations can be included based on the practicality or non-creativity. The order of the steps listed in the examples is only one of the many execution orders of the steps, and does not represent the only execution order. In actual device or client product execution, the method order shown in the examples or the drawings can be executed in sequence or in parallel (for example, in a parallel processor or a multi-threaded processing environment, or even in a distributed data processing environment). The terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, product or device. Without further limitation, it does not exclude the presence of other same or equivalent elements in the process, method, product or device including the elements. The terms "first", "second" and the like are used to indicate names, and do not represent any particular order.

[0193] Those skilled in the art also know that, in addition to implementing the controller in the form of pure computer readable program code, the same function can be achieved by logically programming the method steps into the form of logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers. Therefore, such a controller can be considered as a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0194] The description can be described in the general context of computer-executable instructions, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, classes, and the like that perform particular tasks or implement particular abstract data types. The description can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including storage devices.

[0195] Those skilled in the art can clearly understand the present specification can be implemented by means of software and necessary general hardware platforms through the above description of the embodiments. Based on such understanding, the technical solutions of the present specification can essentially be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) execute the methods described in the various embodiments or some parts of the embodiments of the present specification.

[0196] The various embodiments in the present specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments. The present specification can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, small computers, large computers, distributed computing environments including any of the above systems or devices, etc.

[0197] Although the present specification is described through the embodiments, those skilled in the art know that the present specification has many modifications and changes without departing from the spirit of the present specification, and it is intended that the appended claims include these modifications and changes without departing from the spirit of the present specification.

Claims

1. A method for determining transaction risk, characterized in that, include: Obtain target data about the target users; The target user's target data is processed using a preset feature engineering model to obtain corresponding target feature data; wherein, the preset feature engineering model is a model trained by reinforcement learning based on UCB algorithm using sample data; the preset feature engineering model is also configured with a target feature engineering operator set, which contains multiple feature engineering operators that are suitable for the target application scenario determined during reinforcement learning based on UCB algorithm. The target feature data is processed using a preset classification model to obtain a target classification result; wherein, the preset classification model is a model trained by machine learning using sample feature data; and the sample feature data is obtained by performing feature engineering processing on the sample data using a preset feature engineering model. Based on the target classification results, determine whether the target user poses a transaction risk; The method further includes: acquiring sample data; constructing a decision function for selecting feature engineering operators according to the UCB algorithm; and configuring a preset set of feature engineering operators; wherein the preset set of feature engineering operators contains multiple feature engineering operators; and controlling the agent to perform multiple iterative processing based on exploration and utilization patterns using the sample data according to the decision function and the preset set of feature engineering operators to obtain a preset feature engineering model that meets the requirements. The decision function includes: Q t (a) and U t (a); Q t (a) represents utilization, which is the actual distribution of the current action-reward. Here, the action can correspond to the previously selected feature engineering operator (add or delete), and the reward corresponds to the AUC value after adding or deleting the feature engineering operator in the previous instance; U t (a) represents exploration, which is a measure of uncertainty of the current action, where the action corresponds to the currently selected addition or deletion feature engineering operator.

2. The method according to claim 1, characterized in that, The target data includes the target user's attribute data and / or the target user's behavioral data.

3. The method according to claim 1, characterized in that, The feature engineering operators include at least one of the following: a preset missing value median processing operator, a preset one-hot encoding operator for categorical features, and a preset data normalization and transformation operator.

4. The method according to claim 1, characterized in that, Based on the UCB algorithm, a decision function for selecting feature engineering operators is constructed, including: Construct the decision function according to the following formula: Among them, a t Let t be the selected feature engineering operator in the current iteration, t be the iteration number corresponding to the current iteration, A be the preset set of feature engineering operators, B be the feature engineering operator updated in the previous iteration, and Q be the feature engineering operator selected in the previous iteration. t (a) represents the reward data generated in the current iteration using the previously selected feature engineering operator, U t (a) represents the reward data generated by exploring a feature engineering operator in the current iteration, where c is the weight value and N is the weight value. t (a) represents the number of times the feature engineering operator has been explored.

5. The method according to claim 4, characterized in that, Based on the decision function and the preset feature engineering operator set, the controlled agent performs multiple iterative processes using the sample data based on exploration and utilization patterns, including: The current iteration in a multi-iteration process is processed in the following manner: Get the reward data generated by the previously selected feature engineering operator, and the previous set of target feature engineering operators; Using the reward data and sample data generated by the previously selected feature engineering operator, the decision function is solved to enable the agent to determine the currently selected feature engineering operator from the preset feature engineering operator set. Based on the currently selected feature engineering operator, update the previous target feature engineering operator set to obtain the current target feature engineering operator set; The sample data is processed by feature engineering using the current set of target feature engineering operators to obtain the current sample feature data. Based on the current sample feature data, determine the reward data generated by the currently selected feature engineering operator.

6. The method according to claim 5, characterized in that, Based on the current sample feature data, determine the reward data generated by the currently selected feature engineering operator, including: The initial classification model is used to process the current sample feature data to obtain the current classification result; Based on the current classification result, cross-validation is performed to determine the reward data generated by the currently selected feature engineering operator.

7. The method according to claim 1, characterized in that, After updating the previous set of target feature engineering operators based on the currently selected feature engineering operators to obtain the current set of target feature engineering operators, the method further includes: Check whether the number of iterations corresponding to the current iteration is equal to the preset iteration threshold; and / or check whether the decision function in the current iteration has converged; If the number of iterations corresponding to the current iteration is equal to the preset iteration threshold, and / or if the decision function in the current iteration converges, a preset feature engineering model that meets the requirements is obtained.

8. The method according to claim 1, characterized in that, After determining that a pre-defined feature engineering model that meets the requirements has been obtained, the method further includes: Obtain the initial classification model; The sample data is processed using a pre-defined feature engineering model to obtain sample feature data; Machine learning is performed on the initial classification model using sample feature data to obtain a pre-defined classification model that meets the requirements.

9. The method according to claim 8, characterized in that, The initial classification model includes a model based on the XGboost architecture.

10. A data processing method, characterized in that, include: Obtain sample data associated with the target application scenario; Based on the UCB algorithm, a decision function for selecting feature engineering operators is constructed. And configure a preset set of feature engineering operators; wherein, the preset set of feature engineering operators contains multiple feature engineering operators; Based on the decision function and the preset feature engineering operator set, the control agent uses the sample data to perform multiple iterative processes based on exploration and utilization patterns to obtain a preset feature engineering model that meets the requirements for the target application scenario. The decision function includes: Q t (a) and U t (a); Q t (a) represents utilization, which is the actual distribution of the current action-reward. Here, the action can correspond to the previously selected feature engineering operator (add or delete), and the reward corresponds to the AUC value after adding or deleting the feature engineering operator in the previous instance; U t (a) represents exploration, which is a measure of uncertainty of the current action, where the action corresponds to the feature engineering operator selected for addition or deletion in the current selection; The preset feature engineering model is also configured with a target feature engineering operator set, which includes multiple feature engineering operators that are determined to be suitable for the target application scenario during reinforcement learning based on the UCB algorithm.

11. A device for determining transaction risk, characterized in that, include: The acquisition module is used to acquire target data about the target user; The first processing module is used to perform feature engineering processing on the target data of the target user using a preset feature engineering model to obtain the corresponding target feature data; wherein, the preset feature engineering model is a model trained by reinforcement learning based on the UCB algorithm using sample data; the preset feature engineering model is also configured with a target feature engineering operator set, which contains multiple feature engineering operators that are determined to be suitable for the target application scenario during the reinforcement learning process based on the UCB algorithm. The second processing module is used to process the target feature data using a preset classification model to obtain a target classification result; wherein, the preset classification model is a model trained by machine learning using sample feature data; and the sample feature data is obtained by performing feature engineering processing on the sample data using a preset feature engineering model. The determination module is used to determine whether a target user has transaction risks based on the target classification results; The device is further configured to: acquire sample data; construct a decision function for selecting feature engineering operators according to the UCB algorithm; and configure a preset set of feature engineering operators; wherein the preset set of feature engineering operators contains multiple feature engineering operators; and control the agent to perform multiple iterative processing based on exploration and utilization patterns using the sample data according to the decision function and the preset set of feature engineering operators, so as to obtain a preset feature engineering model that meets the requirements. The decision function includes: Q t (a) and U t (a); Q t (a) represents utilization, which is the actual distribution of the current action-reward. Here, the action can correspond to the previously selected feature engineering operator (add or delete), and the reward corresponds to the AUC value after adding or deleting the feature engineering operator in the previous instance; U t (a) represents exploration, which is a measure of uncertainty of the current action, where the action corresponds to the currently selected addition or deletion feature engineering operator.

12. A server, characterized in that, It includes a processor and a memory for storing processor-executable instructions, wherein the processor, when executing the instructions, implements the steps of the method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, It stores computer instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

14. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

Images