A model training method, a risk control execution method, a device, and an electronic device

Abstract

The specification discloses a model training method, a risk control execution method, a device and electronic equipment. The training method comprises: obtaining time series data of a first type of monitoring item corresponding to a first historical sample set of a target activity, the first type of monitoring item being a monitoring item that has been determined to be affected by the target activity; performing similarity clustering on the time series data of the first type of monitoring item corresponding to the first historical sample set to obtain a plurality of clusters; obtaining time series data of a second type of monitoring item corresponding to a second historical sample set of the target activity, the second type of monitoring item being different from the first type of monitoring item, the samples of the second historical sample set being labeled with training labels, the training labels being used to indicate whether the samples are affected by the target activity; constructing training features of the second historical sample set based on the cluster identifiers of the clusters to which the time series data of the second type of monitoring item corresponding to the second historical sample set belong and the corresponding similarity distances; and training a prediction model based on the training labels and the training features of the second historical sample set.

Description

Technical Field

[0001] This document belongs to the field of artificial intelligence technology, and in particular relates to a model training method, a risk control execution method, a device, and electronic equipment. Background Technology

[0002] Nowadays, various e-commerce platforms frequently launch activities related to their own businesses. With the development of information technology, the architecture of these e-commerce platforms is becoming increasingly complex. Each activity involves more and more applications, interfaces, and remote call programs, and these are becoming increasingly subtle, making it difficult to manually analyze which are affected. If the platform lacks assessment measures for the impact of these activities, significant problems could result in substantial losses for both the platform and its users.

[0003] Therefore, the technical problem to be solved in this application is how to accurately analyze the nodes in the platform that are potentially affected by the activity, and then formulate corresponding risk control measures. Summary of the Invention

[0004] This specification provides a model training method, risk control execution method, device, and electronic equipment that can accurately analyze potential nodes in a business platform that may be affected by activities, thereby enabling the formulation and execution of corresponding risk control measures in advance to avoid losses to the platform and users due to sudden problems in the activity chain.

[0005] To solve the above-mentioned technical problems, the embodiments in this specification are implemented as follows:

[0006] Firstly, a model training method is proposed, including:

[0007] Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity;

[0008] Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters;

[0009] Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity, wherein the second type of monitoring item is different from the first type of monitoring item, and the samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity;

[0010] Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance;

[0011] Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0012] Secondly, a risk control implementation method is proposed, including:

[0013] Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity;

[0014] The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0015] The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0016] If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed.

[0017] Thirdly, a model training device is proposed, comprising:

[0018] The first acquisition module acquires time-series data of a first type of monitoring item corresponding to a first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity;

[0019] The first clustering calculation module performs similarity clustering on the time-series data of the first historical sample set corresponding to the first type of monitoring items to obtain multiple clusters;

[0020] The second acquisition module acquires time-series data of the second type of monitoring items corresponding to the second historical sample set of the target activity. The second type of monitoring items is different from the first type of monitoring items. The samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity.

[0021] The feature construction module determines the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and constructs the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance.

[0022] The model training module trains a prediction model based on the training labels and training features of the second historical sample set, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0023] Fourthly, an electronic device is proposed, comprising: a processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform the following operations:

[0024] Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity;

[0025] Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters;

[0026] Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity, wherein the second type of monitoring item is different from the first type of monitoring item, and the samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity;

[0027] Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance;

[0028] Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0029] Fifthly, a computer-readable storage medium is provided that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations:

[0030] Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity;

[0031] Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters;

[0032] Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity, wherein the second type of monitoring item is different from the first type of monitoring item, and the samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity;

[0033] Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance;

[0034] Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0035] Sixthly, a risk control execution device is proposed, comprising:

[0036] The third acquisition module acquires the time series data of the target monitoring item corresponding to the historical sample set to be predicted of the target activity;

[0037] The second clustering calculation module determines the cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters, and constructs the input features of the historical sample set to be predicted based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0038] The model prediction module inputs the input features of the historical sample set to be predicted into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0039] If the prediction result indicates that the target monitoring item is affected by the target activity, the risk control execution module executes risk control decisions related to the target activity based on the monitoring data of the target monitoring item corresponding to the target activity.

[0040] In a seventh aspect, an electronic device is proposed, comprising: a processor; and a memory arranged to store computer-executable instructions, which, when executed, cause the processor to perform the following operations:

[0041] Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity;

[0042] The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0043] The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0044] If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed.

[0045] Eighthly, a computer-readable storage medium is provided that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations:

[0046] Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity;

[0047] The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0048] The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0049] If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed.

[0050] Based on the embodiments of this specification, a predictive model capable of measuring whether a monitoring item in a business platform is affected by an activity can be trained. During training, a small number of first and second historical sample sets are prepared in advance. First, similarity clustering is performed on the time-series data of the first type of monitoring items identified as being affected by the activity, based on the first historical sample set, resulting in multiple clusters as reference standards. Then, the time-series data of other second type of monitoring items corresponding to the second historical sample set is used as the training set to determine the cluster to which the time-series data in the training set belongs, and training features are constructed based on the cluster identifiers and similarity distances of the corresponding clusters. Finally, the predictive model is trained in a supervised manner based on the training features and pre-labeled training labels, enabling the predictive model to analyze whether a target monitoring item is affected by an activity when predicting whether such an item is affected, based on the cluster identifiers and corresponding similarity distances of the time-series data corresponding to the target monitoring item during the activity. Furthermore, in practical applications, after the predictive model determines that a target monitored item is affected by an activity, risk control decisions related to the activity can be executed based on the monitoring data of the corresponding activity. This allows for timely prevention of losses to the business platform and users due to problems with the target monitored item. In this solution, the training features of the predictive model consist of cluster identifiers and similarity distances between time-series data and a reference standard. Therefore, classifying and analyzing the potential impact of activities on the business platform from a time-series perspective is independent of specific business characteristics and can be applied to different business scenarios, demonstrating excellent versatility and transferability. Attached Figure Description

[0051] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and their descriptions, serving to explain this specification and do not constitute an undue limitation thereof.

[0052] In the picture:

[0053] Figure 1 This is a flowchart illustrating the base model training method provided in the embodiments of this specification.

[0054] Figure 2 This diagram illustrates the clustering calculation of time series data using the base model training method provided in the embodiments of this specification.

[0055] Figure 3 This is a schematic diagram illustrating the construction of training features for the base model training method provided in the embodiments of this specification.

[0056] Figure 4 This is a flowchart illustrating the risk control execution method provided in the embodiments of this specification.

[0057] Figure 5 This is a schematic diagram of the architecture corresponding to the risk control execution method of the embodiments in this specification implemented on the business platform.

[0058] Figure 6 This is a schematic diagram of the structure of the model training device provided in the embodiments of this specification.

[0059] Figure 7 This is a schematic diagram of the structure of the risk control execution device provided in the embodiments of this specification.

[0060] Figure 8 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this specification. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this document clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments in this document, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this document.

[0062] As mentioned earlier, e-commerce platforms nowadays frequently conduct activities related to their own business. With the development of information technology, the architecture of these e-commerce platforms is becoming increasingly complex, and each activity involves more and more applications, interfaces, and remote call programs, which are becoming increasingly obscure, making it difficult to manually analyze which are affected. If the platform lacks assessment measures for the impact of activities across the entire chain, a major problem could result in significant losses for both the platform and its users.

[0063] Therefore, this specification aims to propose an artificial intelligence solution based on deep learning technology and model analysis of the impact of activities. This solution can accurately identify potential nodes in the business platform that are affected by activities, and then formulate risk control measures related to the activities for the affected nodes, so as to achieve the purpose of automated early warning.

[0064] Specifically, Figure 1 This is a flowchart of a model training method provided in one embodiment of this specification. Figure 2 The method shown can be performed by the apparatus described below, and specifically includes the following steps:

[0065] S102, Obtain the time series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity.

[0066] In this embodiment, data points can be set up on high-value nodes in the business platform, such as applications, interfaces, remote procedure calls, and data packets, to serve as monitoring items. By summarizing the time-series data monitored by these data points before and after the target activity is carried out, a historical sample of the target activity can be obtained.

[0067] It should be understood that the time-series data monitored by the event tracking points reflects the changes brought about before and after the target activity. If the changes are significant, it indicates that the monitoring corresponding to the event tracking point is affected by the target activity. It should be noted that the time-series data corresponding to different monitoring items depends on the specific risk control requirements of the business platform.

[0068] Here, we'll use a specific target interface within a business platform as an example. Assuming the target activity is a new activity launched by the business platform, to test whether the target interface can handle the traffic pressure of the activity, we should set up monitoring points on the target interface to track traffic-related time-series data. Correspondingly, if the traffic fluctuations of the target interface change significantly before and after the target activity, it indicates that the target interface is a relatively important node in the target activity. For the business platform, the bandwidth resources allocated to this target interface can be used to make traffic load-related risk control decisions. For example, if the target interface is overloaded during the target activity, traffic resources can be increased.

[0069] Alternatively, consider a target application within a business platform. Suppose a target activity conducted by the business platform is implemented within a target application. To monitor user feedback on this activity, tracking points can be set up on the target application to monitor relevant time-series data. Correspondingly, if the target application's rating fluctuates significantly before and after the activity, it indicates that the target application is a crucial platform for the activity. For the business platform, user reviews of this target application can inform risk control decisions related to activity quality. For instance, if the user rating of the target application drops beyond a preset threshold during the activity, the platform can proactively improve the service quality of the activity.

[0070] Specifically, the first type of monitoring item is the monitoring item in the business platform that has been qualitatively affected by the target activity. In this embodiment, the first type of monitoring item serves as a reference benchmark for monitoring items affected by the target activity. It belongs to prior knowledge and can be obtained through human analysis or mined based on other artificial intelligence technologies. This document does not make any specific limitations here.

[0071] S104. Similarity clustering is performed on the time series data of the first type of monitoring item corresponding to the first historical sample set to obtain multiple clusters.

[0072] As mentioned earlier, this embodiment uses the first type of monitoring items as a reference benchmark for those affected by the target activity. After clustering the time-series data of the first type of monitoring items in the historical sample set, each cluster is equivalent to a classification of the known monitoring items affected by the target activity.

[0073] Typically, there is a certain correlation in data fluctuations between nodes in an activity chain. For monitoring items whose impact from the target activity is unknown, the similarity distance between their corresponding time-series data and these clusters can serve as a reference for analyzing whether the target activity has an impact.

[0074] The following section introduces methods for similarity clustering.

[0075] Specifically, this embodiment pre-sets multiple feature dimensions related to risk control decisions for the target activity. For example, in a risk control decision-making scenario involving resource allocation, different types of resources such as processors, memory, and loans can be used as feature dimensions, and then sub-time series data can be extracted according to each feature dimension.

[0076] Here, if different feature dimensions correspond to different time slot length requirements, it is necessary to first perform time-series splitting on the time-series data of the first historical sample set corresponding to the first type of monitoring item according to the time-series length requirements of multiple feature dimensions, so as to ensure that sub-time-series data with time-series length matching the feature dimensions are obtained. Then, feature extraction is performed on the sub-time-series data that match the time-series length based on multiple feature dimensions.

[0077] After feature extraction is completed, a feature matrix composed of the feature extraction results is constructed for the time series data of the first type of monitoring item corresponding to the first historical sample set.

[0078] Finally, similarity clustering is performed on the constructed feature matrix based on the similarity metric function.

[0079] For ease of understanding, combined with Figure 2 The following is an example illustration.

[0080] In this embodiment, the initial n time series data corresponding to the first type of monitoring item in the first historical sample set can be regarded as a time series matrix n*1*s, where n represents the number of samples in the first historical sample set, 1 represents the current feature dimension, and s represents the time series length.

[0081] It should be understood that the initial time series data is composed of a superposition of multiple dimensions of features such as activity characteristics, daily fluctuation characteristics, and application habits, and cannot reflect the feature information under a specific feature dimension.

[0082] To this end, the Empirical Mode Decomposition (EMD) algorithm is further employed to split each initial time series data according to the time series length corresponding to the set m feature dimensions, thereby obtaining the corresponding sub-time series. For example... Figure 1 In the process, the initial time series data 1 is split into sub-time series feature 1.1, sub-time series feature 1.2, ..., sub-time series feature 1.m; the initial time series data n is split into sub-time series feature n.1, sub-time series feature n.2, ..., and sub-time series feature nm.

[0083] It should be understood that after obtaining multi-dimensional features based on the EMD algorithm, the time series matrix changes from n*1*s to n*m*s. Here, m represents the feature dimension.

[0084] Here, I(n) is defined as the initial input signal of the time-series data, then I(n) can be expressed by the following equation:

[0085]

[0086] Where n represents the nth time series data in the first historical sample set, m represents the index of the mth feature dimension corresponding to the time series data, and IMF m (n) represents the eigenmode function of multiple features, Res m (n) represents the residual.

[0087] Subsequently, the split sub-time series data undergo corresponding feature extraction to obtain local feature information at different time scales. For example, feature extraction is performed on sub-time series feature 1.1 based on feature dimension 1, on sub-time series feature 1.2 based on feature dimension 2, and so on, on sub-time series feature 1.m based on feature dimension m; similarly, feature extraction is performed on sub-time series feature n.1 based on feature dimension 1, on sub-time series feature n.2 based on feature dimension 2, and so on, on sub-time series feature nm based on feature dimension m.

[0088] Various feature IMFs were obtained through the EMD algorithm. m After obtaining the time series matrix n*m*s of (n), further filtering and selection can be performed to extract the actual required activity features.

[0089] The screening is based on the assumption that the activities are sudden events and will not exhibit periodic fluctuations. Specifically, Fourier transform algorithms can be used to further filter out IMFs with periodic fluctuations. m (n) Temporal features, and only the remaining temporal feature matrix is ​​used for subsequent analysis and decision-making. At the same time, in order to ensure the consistency of matrix dimensions, it is necessary to determine the proportion of each feature dimension appearing in the first historical sample set, and filter out the feature extraction results corresponding to feature dimensions with a proportion lower than a preset threshold.

[0090] For example, the filtering constraints for the Fourier transform are:

[0091]

[0092] Based on the above formula, feature extraction results corresponding to feature dimensions with a proportion less than 0.5 can be removed. The resulting time series matrix is ​​n*h*s, where h < m.

[0093] After feature extraction, more accurate activity feature information is obtained, effectively filtering out irrelevant noise data. Subsequently, to further reduce computational complexity, this embodiment can employ a k-means clustering model and perform clustering calculations based on the L2-norm similarity metric.

[0094] The similarity metric function for L2-norm is: dist = ||x 2 -y 2 || . x and x' represent the two endpoints for which the similarity distance needs to be calculated.

[0095] After obtaining the clustering model, the center time-series data of each cluster can be easily calculated. The center time-series data refers to the data with the shortest average similarity distance to the time-series data of other monitoring items within its cluster. Based on this similarity distance model, monitoring items whose impact from the target activity has not yet been determined can be analyzed using the clustering model. This involves calculating the similarity between the time-series data of the target activity and the center time-series data of each cluster, thereby determining whether the unanalyzed monitoring items are strongly correlated with the target activity. If there is a strong correlation, it indicates that the unanalyzed monitoring items are affected by the target activity.

[0096] S106, Obtain the time series data of the second type of monitoring item corresponding to the second historical sample set of the target activity. The second type of monitoring item is different from the first type of monitoring item. The samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity.

[0097] In this embodiment, the second historical sample set is used as the training set for the prediction model.

[0098] Here, a supervised training method is used to train the prediction model. The ultimate goal of the training is to enable the prediction model to predict whether the temporal features of a certain monitored item are affected by the target activity. Therefore, it belongs to the binary classification problem.

[0099] In binary classification prediction, the corresponding training labels only need to classify the time-series data of the second type of monitoring item in the training set into "positive samples" and "negative samples". Here, "positive samples" refer to time-series data that are affected by the target activity, serving as positive examples for training the prediction model; "negative samples" refer to time-series data that are not affected by the target activity, serving as negative examples for training the prediction model.

[0100] It should be noted that the second historical sample set is pre-prepared data, and its corresponding training labels can be obtained manually. This article does not specify the method of obtaining the second historical sample set.

[0101] S108, determine the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier and the corresponding similarity distance.

[0102] As mentioned above, in this embodiment, the similarity distance between the monitoring items whose status as being unaffected by the target activity and the clusters obtained in S104 is used as a reference for analyzing whether the target activity has an impact.

[0103] Therefore, a cluster identifier can be configured for each cluster obtained in S104. This cluster identifier is only used to distinguish between clusters and has no practical meaning. Then, based on the similarity distance between the time series data of the second type of monitoring item corresponding to the second historical sample set and each cluster, the cluster with the smallest similarity distance is taken as the cluster to which it belongs.

[0104] Here, in this embodiment, the cluster identifier and the corresponding similarity distance of the belonging cluster are used as two basic features of the training set for training the prediction model. These are combined with features of the time-series data of the second historical sample set in other dimensions, such as mean, standard deviation, maximum, minimum, kurtosis, and skewness in statistics, to construct the training features corresponding to the second historical sample set. The statistical features can be automatically extracted using the tsfresh algorithm package, which is specifically designed for feature engineering of time-series data; however, this paper does not specify a particular feature extraction method.

[0105] It should be understood and referenced. Figure 3 As shown, the training features in this embodiment have at least two dimensions: one is the cluster identifier of the cluster to which it belongs, and the other is the similarity distance with the cluster to which it belongs. Here, the training features are regarded as n*2 matrix data. Based on n*2, if features of other dimensions are introduced, such as k statistical features extracted by the tsfresh algorithm package, then the training features are regarded as n*(k+2) matrix data.

[0106] S110, Based on the training labels and training features of the second historical sample set, a prediction model is trained. The prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0107] In this embodiment, the training labels serve as the expected output of the prediction model. The training process involves inputting the training features "n*(k+2)" from the second historical sample set into the prediction model, which then provides a prediction of whether the target activity has an impact. During the training phase, because the prediction model's predictive ability is limited, its output predictions are not necessarily accurate and may differ from the results labeled with the ground truth training labels. This error is the training loss of the prediction model. This step adjusts the parameters of the prediction model by reducing the training loss, thereby gradually bringing the prediction results output by the prediction model closer to the ground truth results labeled with the training labels, thus achieving the training objective.

[0108] As an example, the essence of a predictive model is to calculate the following formula:

[0109]

[0110] Where, x i w represents the training feature of the nth time series data. i b represents the weight corresponding to the nth time series data.i This represents the deviation corresponding to the nth time series data.

[0111] The objective function for calculating the training loss is:

[0112]

[0113] Training involves finding the optimal solution for y using the objective function, and determining w. i .

[0114] Based on the embodiments of this specification, a predictive model capable of measuring whether a monitoring item in a business platform is affected by an activity can be trained. During training, a small number of first and second historical sample sets are prepared in advance. First, similarity clustering is performed on the time-series data of the first type of monitoring items identified as being affected by the activity, based on the first historical sample set, resulting in multiple clusters as reference standards. Then, the time-series data of other second type of monitoring items corresponding to the second historical sample set is used as the training set to determine the cluster to which the time-series data in the training set belongs, and training features are constructed based on the cluster identifiers and similarity distances of the corresponding clusters. Finally, the predictive model is trained in a supervised manner based on the training features and pre-labeled training labels, enabling the predictive model to analyze whether a target monitoring item is affected by an activity when predicting whether such an item is affected, based on the cluster identifiers and corresponding similarity distances of the time-series data corresponding to the target monitoring item during the activity.

[0115] In the entire solution, the training features of the prediction model are composed of cluster identifiers and similarity distances of time series data relative to the reference standard. Therefore, it classifies and analyzes the potential impact of activities on the business platform from a time series perspective, without relying on the business characteristics under specific business scenarios. Thus, it can be applied to different business scenarios, and the solution has good versatility and transferability.

[0116] Figure 4 This is a flowchart of a risk control execution method provided in one embodiment of this specification. Figure 4 The method shown can be performed by the apparatus described below, and specifically includes the following steps:

[0117] S402, Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted of the target activity.

[0118] In this embodiment, the target monitoring item refers to the monitoring item in the business platform that needs to be determined whether it is affected by the target activity, such as the application, interface, remote call program and data packet of the business platform.

[0119] S404, determine the cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters, and construct the input features of the historical sample set to be predicted based on the cluster identifier and the corresponding similarity distance.

[0120] Among them, multiple clusters are obtained by similarity clustering of time series data corresponding to the first type of monitoring items in the first historical sample set of the target activity. The first type of monitoring items refers to the monitoring items in the business platform that have been identified as being affected by the target activity.

[0121] It should be understood that the multiple clusters mentioned in this step are obtained based on similarity clustering in S104, which will not be elaborated on here. It should be noted that the input features constructed in this step should have the same feature dimension as the training features used in the second historical sample set during the training phase of the prediction model.

[0122] Suppose that the training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which it belongs, the similarity distance with the cluster to which it belongs, and the statistical mean and standard deviation. Then the training features should also be constructed based on the cluster identifier of the historical sample set to be predicted, the similarity distance with the cluster to which it belongs, and the statistical mean and standard deviation.

[0123] S406, input the input features of the historical sample set to be predicted into the prediction model to obtain the prediction result of whether the target monitoring item is affected by the target activity.

[0124] The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifiers of the time series data of the second type of monitoring item corresponding to the second historical sample set in multiple clusters and the corresponding similarity distances. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0125] S408, if the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target activity corresponding to the target monitoring item, a risk control decision related to the target activity is executed.

[0126] Here, an example is given to illustrate the application of the risk control execution method of the embodiments in this specification.

[0127] Suppose a target activity in the business platform is transitioning from version 1 to version 2. Version 2 adds new features compared to version 1.

[0128] Since the activity chain for version 1 is already defined, the monitoring items in the current activity chain of version 1 are defined as first-category monitoring items. For risk control, during the period when the target activity is version 1, the business platform will obtain monitoring data from the monitoring items in the current activity chain of version 1, and then execute corresponding risk control decisions based on the monitoring data.

[0129] When the target activity is iterated to the second version, new features are added in the second version. Therefore, it is necessary to add monitoring items corresponding to the new features to the existing activity chain to ensure that the business platform's risk control can cover the new features.

[0130] Here, a first historical sample set and a second historical sample set can be extracted from the historical samples of the first version target activity recorded by the business system. Then, based on the time-series data of the first type of monitoring item (which has been determined to be affected by the first version target activity) in the current activity chain corresponding to the first historical sample set, clustering calculations are performed to obtain multiple clusters.

[0131] Next, a small number of historical samples of the second version of the target activity recorded by the business system are manually extracted as the second historical sample set, and training labels are labeled on the second historical sample set. After the labeling is completed, based on the time-series data of the second type of monitoring item outside the current activity chain corresponding to the second historical sample set, the cluster to which it belongs and the relative degree distance with the cluster to which it belongs are calculated, and then training features are constructed. Then, the prediction model is trained based on the training labels and training features of the second historical sample set.

[0132] Once the prediction model is trained, to analyze whether a specific monitoring item (not belonging to the current activity chain) is affected by the second version of the target activity, time-series data of that monitoring item during the second version's target activity can be extracted from the business platform. Based on this time-series data, the similarity distance with each cluster is calculated. Then, the cluster with the smallest similarity distance is identified as the cluster to which the monitoring item belongs, and input features are constructed based on the cluster identifier and the corresponding similarity distance. Finally, after inputting these features into the model, the corresponding prediction result is obtained.

[0133] If the prediction results indicate that the target monitoring item is affected by the second version of the target activity, the target monitoring item can be added to the current activity chain. Subsequently, the business platform will obtain monitoring data from the target monitoring item for risk control decisions on the target activity.

[0134] like Figure 5 As shown, when the above solution is applied to the specific implementation of the business platform, the embodiments of this specification can construct an activity impact surface analysis system, which is specifically divided into an offline analysis framework and an online analysis framework.

[0135] Here, the database tables of the Open Data Processing Service (ODPS) can be responsible for recording all historical samples of the target activity.

[0136] For the offline analysis framework, during the training phase, time-series data from the first and second historical sample sets are extracted from the ODPS database. This data is then input into pre-prepared clustering and prediction models to train the prediction model. The trained clustering and prediction models can then be deployed in the business platform's Operation Support Systems (OSS) to provide offline prediction capabilities. Furthermore, the outputs of the clustering and prediction models can be stored in ODPS data tables, allowing technical personnel to perceive differences in model accuracy across different iterations.

[0137] For the online analysis framework, the time-series data of unanalyzed monitoring items corresponding to historical samples are still retrieved from the ODPS database and fed into the corresponding algorithm service provided by OSS. Specifically, a clustering model first calculates the cluster to which the monitored time-series data belongs and its similarity distance to other clusters. Then, input features are constructed based on the cluster identifier and similarity distance to other clusters. These trained features are then fed into the prediction model to identify all monitoring items affected by the target activity and to aggregate the relevant impact area.

[0138] After determining the scope of impact, the system compares the similarities and differences in the monitoring timelines of the monitored items within this scope between the current and historical activities, and provides an evaluation result, which is stored in the ODPS data table for technical personnel to filter. The filtering results of the technical personnel are also simultaneously added to the ODPS data table as feedback annotation information, serving as sample data for the next model training, thus achieving a closed loop of human feedback -> model training.

[0139] For the system described in this specification, the prediction model no longer needs to rely on manual parameter tuning by developers. It can adaptively optimize algorithm parameters based on human feedback and automatically acquire features of various activity scenarios to make decisions.

[0140] and Figure 1 Corresponding to the method shown, embodiments of this specification also provide a model training apparatus. Figure 6 This is a schematic diagram of the structure of a model training device 600 provided in one embodiment of this specification, including:

[0141] The first acquisition module 610 acquires time-series data of a first type of monitoring item corresponding to a first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity.

[0142] The first clustering calculation module 620 performs similarity clustering on the time-series data of the first historical sample set corresponding to the first type of monitoring items to obtain multiple clusters.

[0143] The second acquisition module 630 acquires time-series data of the second type of monitoring items corresponding to the second historical sample set of the target activity. The second type of monitoring items is different from the first type of monitoring items. The samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity.

[0144] The feature construction module 640 determines the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and constructs training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance.

[0145] The model training module 650 trains a prediction model based on the training labels and training features of the second historical sample set, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0146] The apparatus described in this specification can train a predictive model capable of measuring whether a monitoring item in a business platform is affected by an activity. During training, a small number of first and second historical sample sets are prepared in advance. First, similarity clustering is performed on the time-series data of the first type of monitoring items in the first historical sample set that have been determined to be affected by the activity, resulting in multiple clusters as reference standards. Then, the time-series data of other second-type monitoring items in the second historical sample set is used as the training set to determine the cluster to which the time-series data in the training set belongs, and training features are constructed based on the cluster identifiers and similarity distances of the corresponding clusters. Finally, the predictive model is trained in a supervised manner based on the training features and pre-labeled training labels, enabling the predictive model to analyze whether a target monitoring item is affected by an activity when predicting whether such an activity has affected it, based on the cluster identifiers and corresponding similarity distances of the time-series data corresponding to the target monitoring item in the activity.

[0147] Optionally, the first clustering calculation module 620 performs similarity clustering on the time-series data of the first historical sample set corresponding to the first type of monitoring item, including: splitting the time-series data of the first historical sample set corresponding to the first type of monitoring item into time-series segments based on pre-set time-series length requirements of multiple feature dimensions to obtain sub-time-series data whose time-series length matches the multiple feature dimensions, wherein the multiple feature dimensions are associated with the target activity; extracting features from the sub-time-series data that match the time-series length based on the multiple feature dimensions; constructing a feature matrix composed of feature extraction results for the time-series data of the first historical sample set corresponding to the first type of monitoring item; and performing similarity clustering on the constructed feature matrix based on a similarity metric function.

[0148] Optionally, before constructing the feature matrix corresponding to the first type of monitoring item in the first historical sample set based on the feature extraction results, the feature construction module 640 further performs the following: based on the obtained feature extraction results, determining the proportion of each feature dimension appearing in the first historical sample set, and filtering out the feature extraction results corresponding to feature dimensions with a proportion lower than a preset threshold.

[0149] Optionally, the first type of monitoring items and / or the second type of monitoring items include at least one of the applications, interfaces, remote call programs, and data packets of the business platform.

[0150] Optionally, the feature construction module 640 constructs training features for the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance, including: constructing training features for the second historical sample set based on the cluster identifier of the belonging cluster, the similarity distance to the belonging cluster, and the statistical indicators of the time-series data of the second type of monitoring item corresponding to the second historical sample set.

[0151] Obviously, the model training device in the embodiments of this specification can achieve... Figure 1 The steps and functions in the illustrated embodiments will not be described in detail here.

[0152] and Figure 4 Corresponding to the method shown, this specification also provides a risk control execution device in its embodiments. Figure 7 This is a schematic diagram of the structure of a risk control execution device 700 provided in one embodiment of this specification, including:

[0153] The third acquisition module 710 acquires the time series data of the target monitoring item corresponding to the historical sample set to be predicted of the target activity.

[0154] The second clustering calculation module 720 determines the cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters, and constructs the input features of the historical sample set to be predicted based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0155] The model prediction module 730 inputs the input features of the historical sample set to be predicted into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0156] The risk control execution module 740, if the prediction result indicates that the target monitoring item is affected by the target activity, then executes risk control decisions related to the target activity based on the monitoring data of the target monitoring item corresponding to the target activity.

[0157] Optionally, the samples in the first historical sample set correspond to the target activity in the first version, and the samples in the historical sample set to be predicted correspond to the target activity in the second version, where the second version is later than the first version. The target monitoring item is a monitoring item to be evaluated as to whether the target activity has been affected after iterating from the first version to the second version.

[0158] Obviously, the risk control execution device in the embodiments of this specification can achieve... Figure 4 The steps and functions in the illustrated embodiments will not be described in detail here.

[0159] Figure 8 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this specification. Please refer to it. Figure 8 At the hardware level, the electronic device includes a processor, and optionally also includes an internal bus, a network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the electronic device may also include other hardware required for other business operations.

[0160] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0161] Memory is used to store programs. Specifically, programs may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0162] The processor reads the corresponding computer program from non-volatile memory into main memory and then runs it, forming the aforementioned model training device at the logical level. The processor executes the program stored in memory and specifically performs the following operations:

[0163] Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity.

[0164] Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters.

[0165] Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity. The second type of monitoring item is different from the first type of monitoring item. The samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity.

[0166] Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier and the corresponding similarity distance.

[0167] Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0168] Alternatively, the processor reads the corresponding computer program from non-volatile memory into memory and then runs it, forming the aforementioned risk control execution device at the logical level. The processor executes the program stored in memory and specifically performs the following operations:

[0169] Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity;

[0170] The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0171] The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0172] If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed.

[0173] The above is as described in this instruction manual. Figure 1 or Figure 4The methods disclosed in the illustrated embodiments can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above methods can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The methods, steps, and logic block diagrams disclosed in one or more embodiments of this specification can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in one or more embodiments of this specification can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0174] Of course, in addition to software implementation, the electronic device described in this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0175] This specification also provides an embodiment of a computer-readable storage medium that stores one or more programs.

[0176] The program includes instructions that, when executed by a portable electronic device comprising multiple applications, enable the portable electronic device to perform... Figure 1 The method of the illustrated embodiment is specifically used to perform the following operations:

[0177] Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity.

[0178] Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters.

[0179] Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity. The second type of monitoring item is different from the first type of monitoring item. The samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity.

[0180] Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier and the corresponding similarity distance.

[0181] Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity.

[0182] Alternatively, the one or more programs include instructions that, when executed by a portable electronic device including multiple applications, enable the portable electronic device to perform... Figure 4 The method of the illustrated embodiment is specifically used to perform the following operations:

[0183] Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity;

[0184] The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity.

[0185] The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item.

[0186] If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed.

[0187] In summary, the above description is merely a preferred embodiment of this specification and is not intended to limit the scope of protection of this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of this specification should be included within the scope of protection of one or more embodiments of this specification.

[0188] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0189] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0190] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0191] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

Claims

1. A model training method, comprising: Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity; Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters; Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity, wherein the second type of monitoring item is different from the first type of monitoring item, and the samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity; Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance; Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

2. The method according to claim 1, Similarity clustering is performed on the time-series data corresponding to the first type of monitoring items in the first historical sample set, including: Based on the time series length requirements of multiple pre-set feature dimensions, the time series data of the first type of monitoring item corresponding to the first historical sample set is split into time series data to obtain sub-time series data whose time series length matches the multiple feature dimensions, wherein the multiple feature dimensions are associated with the target activity. Based on the aforementioned multiple feature dimensions, feature extraction is performed on sub-time series data that match the time series length; For the time-series data of the first type of monitoring item corresponding to the first historical sample set, a feature matrix composed of feature extraction results is constructed; Based on the similarity metric function, similarity clustering is performed on the constructed feature matrix.

3. The method according to claim 2, Before constructing the feature matrix of the first historical sample set corresponding to the first type of monitoring item based on the feature extraction results, the method further includes: Based on the obtained feature extraction results, the proportion of each feature dimension appearing in the first historical sample set is determined, and the feature extraction results corresponding to feature dimensions with a proportion lower than a preset threshold are filtered out.

4. The method according to claim 1, The first type of monitoring items and / or the second type of monitoring items include at least one of the following: application, interface, remote call program, and data packet of the business platform.

5. The method according to claim 1, Based on the cluster identifier and the corresponding similarity distance, the training features of the second historical sample set are constructed, including: Based on the cluster identifier of the belonging cluster, the similarity distance with the belonging cluster, and the statistical indicators of the time series data of the second type of monitoring item corresponding to the second historical sample set, the training features of the second historical sample set are constructed.

6. A risk control execution method, comprising: Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity; The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity. The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item. If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

7. The method according to claim 6, The samples in the first historical sample set correspond to the target activity in the first version, and the samples in the historical sample set to be predicted correspond to the target activity in the second version, which is later than the first version. The target monitoring item is a monitoring item to be evaluated as to whether the target activity has been affected after iterating from the first version to the second version.

8. A model training device, comprising: The first acquisition module acquires time-series data of a first type of monitoring item corresponding to a first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity; The first clustering calculation module performs similarity clustering on the time-series data of the first historical sample set corresponding to the first type of monitoring items to obtain multiple clusters; The second acquisition module acquires time-series data of the second type of monitoring items corresponding to the second historical sample set of the target activity. The second type of monitoring items is different from the first type of monitoring items. The samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity. The feature construction module determines the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and constructs the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance. The model training module trains a prediction model based on the training labels and training features of the second historical sample set, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

9. An electronic device, comprising: processor; and memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the following operations: Obtain time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, wherein the first type of monitoring item has been determined to be affected by the target activity; Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters; Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity, wherein the second type of monitoring item is different from the first type of monitoring item, and the samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity; Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance; Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

10. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations: Obtain the time-series data of the first type of monitoring item corresponding to the first historical sample set of the target activity, where, The first type of monitoring item has been determined to be affected by the target activity; Similarity clustering is performed on the time-series data of the first type of monitoring items corresponding to the first historical sample set to obtain multiple clusters; Obtain time-series data of the second type of monitoring item corresponding to the second historical sample set of the target activity, wherein the second type of monitoring item is different from the first type of monitoring item, and the samples in the second historical sample set are labeled with training labels, which are used to indicate whether they are affected by the target activity; Determine the cluster to which the time-series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters, and construct the training features of the second historical sample set based on the cluster identifier of the belonging cluster and the corresponding similarity distance; Based on the training labels and training features of the second historical sample set, a prediction model is trained, wherein the prediction model is used to predict whether the monitoring items of the activity link are affected by the target activity; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

11. A risk control execution device, comprising: The third acquisition module acquires the time series data of the target monitoring item corresponding to the historical sample set to be predicted of the target activity; The second clustering calculation module determines the cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters, and constructs the input features of the historical sample set to be predicted based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity. The model prediction module inputs the input features of the historical sample set to be predicted into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item. The risk control execution module, if the prediction result indicates that the target monitoring item is affected by the target activity, then executes risk control decisions related to the target activity based on the monitoring data of the target monitoring item corresponding to the target activity; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

12. An electronic device, comprising: processor; and memory configured to store computer-executable instructions, which, when executed, cause the processor to perform the following operations: Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity; The cluster to which the time series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of the time series data of the first historical sample set of the target activity corresponding to the first type of monitoring item, and the first type of monitoring item has been determined to be affected by the target activity. The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item. If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

13. A computer-readable storage medium storing one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform the following operations: Obtain the time series data of the target monitoring item corresponding to the historical sample set to be predicted for the target activity; The cluster to which the time-series data of the target monitoring item corresponding to the historical sample set to be predicted belongs in multiple clusters is determined, and the input features of the historical sample set to be predicted are constructed based on the cluster identifier and the corresponding similarity distance. The multiple clusters are obtained by similarity clustering of time-series data corresponding to the first type of monitoring items in the first historical sample set of the target activity, and the first type of monitoring items have been determined to be affected by the target activity; The input features of the historical sample set to be predicted are input into the prediction model to obtain a prediction result of whether the target monitoring item is affected by the target activity. The prediction model is trained based on the training labels and training features of the second historical sample set of the target activity. The training features of the second historical sample set are constructed based on the cluster identifier of the cluster to which the time series data of the second type of monitoring item corresponding to the second historical sample set belongs in the multiple clusters and the corresponding similarity distance. The target monitoring item and the second type of monitoring item are different from the first type of monitoring item. If the prediction result indicates that the target monitoring item is affected by the target activity, then based on the monitoring data of the target monitoring item corresponding to the target activity, risk control decisions related to the target activity are executed; The target activity is the business activity of the e-commerce platform; the first type of monitoring item and the second type of monitoring item include at least one of the e-commerce platform's applications, interfaces, and remote call programs; whether any monitoring item is affected by the target activity is determined based on the changes in the time-series data corresponding to that monitoring item relative to the target activity before and after it is carried out.

Images