Data processing method for AI data management platform

Abstract

The application relates to the technical field of data processing, in particular to a data processing method for an AI data management platform; the method comprises the following steps: obtaining multiple historical data of each customer in a customer cluster, taking each customer as a historical data point, clustering multiple customers in the customer cluster based on the multiple historical data corresponding to each historical data point by using mean shift clustering, and obtaining multiple clustering clusters; for any one clustering cluster, analyzing the direction consistency of historical data points in the neighborhood of an update data point and the spatial distance between the historical data points and the center of the clustering cluster, calculating the classification probability of the update data point belonging to the clustering cluster, and classifying the update data point according to the classification probability of the multiple clustering clusters corresponding to the update data point. The application has the effects of real-time and accuracy in processing sales data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a data processing method for an AI data management platform. BACKGROUND

[0002] A data management platform (DMP) collects, integrates and analyzes multi-source heterogeneous data to provide customer behavior insights and market segmentation support for customer relationship management (CRM) systems, helping sales personnel optimize customer interaction and marketing strategies. Focusing on data processing and analysis, DMP generates customer segmentation results, while CRM uses these results to achieve sales automation and customer management, and the two work together to improve sales efficiency.

[0003] The existing data processing method is mainly based on the mean shift clustering algorithm, which is a non-parametric clustering technique applied to DMP customer segmentation. Its principle is based on kernel density estimation, aiming to assign data points to corresponding clusters by iteratively computing the density center of data points. Specifically, mean shift clustering first defines a kernel function (such as Gaussian kernel) for each data point, and then estimates the local density of the data point. The algorithm determines the direction of movement of the data point to the area with higher density by calculating the gradient of the kernel function, called the "mean shift vector". In each iteration, the data point moves along the mean shift vector until it converges to the density peak (i.e. the cluster center). Finally, data points converging to the same density peak are grouped into the same cluster. For example, in DMP, by analyzing customer purchase frequency, average order value and recent purchase time, etc., mean shift clustering can divide customers into high-value customer and potential customer clusters, supporting marketing strategy formulation in CRM.

[0004] However, mean shift clustering has significant drawbacks in practical applications. The algorithm is mainly designed for static data sets, and when sales data is updated in real time (such as new orders or interaction records), the density center of the entire data set needs to be recalculated, resulting in high computational complexity and significant processing delay. For example, in a DMP containing 100,000 customer data, each new data needs to run mean shift clustering again, which may take several minutes, resulting in a trade-off between real-time and accuracy in statistical analysis of sales data. SUMMARY

[0005] To solve the problem that traditional data processing methods cannot balance the real-time and accuracy of sales data, the present application provides a data processing method for an AI data management platform.

[0006] In a first aspect, the present application provides a data processing method for an AI data management platform, which adopts the following technical solution:

[0007] The application discloses an AI data management platform data processing method, obtains multiple historical data of each customer in a customer cluster, takes each customer as a historical data point, clusters multiple customers in the customer cluster based on multiple historical data corresponding to each historical data point by using a mean shift clustering method, and obtains multiple clustering clusters; analyzes an iteration position of a data point in a mean shift clustering process, calculates direction consistency, and the direction consistency is used for representing a degree of coincidence between a moving direction in a historical data point iteration process and a direction of a clustering center;

[0008] For any historical data point, order data corresponding to the historical data point is updated to form an updated data point; a neighborhood of the updated data point is determined based on the updated data point; for any clustering cluster, direction consistency of a historical data point in a neighborhood of the updated data point and a spatial distance between the historical data point and a center of the clustering cluster are analyzed, a classification probability of the updated data point belonging to the clustering cluster is calculated, and the updated data point is classified according to the classification probability of the updated data point corresponding to multiple clustering clusters.

[0009] The method has the beneficial effects that in the method, multiple historical data of each customer are first obtained, each customer is regarded as a historical data point including multiple historical data, customers in a customer cluster are clustered by using a mean shift method, and a stable clustering structure is formed. On this basis, a drift trajectory of each historical data point in iteration calculation in a clustering process is further tracked, a direction consistency index is constructed by analyzing an iteration position change of the historical data point, and when customer data is updated to generate a new data point, the system does not need to re-cluster all data, but selects several historical data points around the updated data point as references based on a neighborhood range of the updated data point in a feature space, estimates a probability of the updated data point belonging to each clustering cluster in combination with a spatial distance and the direction consistency, and finally classifies the updated data point into a clustering cluster with the maximum classification probability. In the aspect, the newly added data point can be quickly classified without clustering global data, the consumption of calculation resources is reduced, the timeliness of system response is improved, and the accuracy of a classification result is effectively preserved.

[0010] Optionally, the step of analyzing an iteration position of a historical data point in a mean shift clustering process and calculating direction consistency used for representing a degree of coincidence between a moving direction in a historical data point iteration process and a direction of a clustering center comprises the following steps: for any iteration process of any historical data point in a mean shift clustering process, local variation of each historical data value and total variation of the historical data after clustering are obtained; a ratio of the variation to the total variation is taken as a center motion trend of the historical data in the iteration process; a variation degree of the historical data in the iteration process is obtained based on the center motion trend; a mean value of variation degrees of multiple historical data is taken as an iteration trend, the direction consistency of the historical data point is obtained based on the iteration trend corresponding to multiple iterations, and the iteration trend is proportional to the direction consistency.

[0011] The beneficial effect is that the local change amount of each historical data in each iteration of the history data point is recorded in the mean shift clustering process, and is compared with the total change amount corresponding to the historical data, a center motion trend index is constructed to represent whether the local iteration moves in the direction of the cluster center, and the index can reflect whether the iteration is converging towards the target cluster.

[0012] Optionally, for any historical data in any iteration process of the history data point, the data difference before and after the iteration of the data is taken as the local change amount of the data in the iteration.

[0013] The beneficial effect is that for any historical data, the data values before and after the iteration are obtained and the difference is taken, so that the local change amount of the iteration can be obtained.

[0014] Optionally, the step of obtaining the change degree of the iteration of the historical data based on the center motion trend includes setting a control value, and taking the sum of the normalized result of the center motion trend and the control value as the change degree.

[0015] The beneficial effect is that the control value is set to adjust the value range of the final change degree, which is convenient for subsequent calculation.

[0016] Optionally, the geometric mean of the multiple iteration trends is taken as the direction consistency of the history data point.

[0017] Optionally, the step of calculating the classification probability of the update data point belonging to the cluster cluster includes: analyzing the distance from the historical data points in the neighborhood to the cluster center of the cluster cluster and the distance from the update data point to the cluster center of the cluster cluster, obtaining the distance consistency; obtaining the trend probability based on the direction consistency of the historical data points in the neighborhood of the update data point in the cluster cluster; taking the product of the trend probability and the distance consistency as the classification probability of the update data point belonging to the cluster cluster.

[0018] The beneficial effect is that this step introduces two dimensions of "distance consistency" and "trend probability" to more carefully and dynamically model the similarity between the update data point and each cluster cluster, so as to realize more accurate and adaptive classification probability judgment than the traditional cluster center distance method.

[0019] Optionally, the step of calculating the classification probability of the update data point belonging to the cluster cluster comprises: analyzing the distance of the historical data points in the neighborhood to the cluster center of the cluster cluster and the distance of the update data point to the cluster center of the cluster cluster to obtain distance consistency; taking the ratio of the number of historical data points belonging to the cluster cluster in the neighborhood to all historical data points in the neighborhood as quantity consistency; obtaining a trend probability based on the direction consistency of the historical data points in the neighborhood of the update data point in the cluster cluster; and taking the product of the trend probability, the quantity consistency and the distance consistency as the classification probability of the update data point belonging to the cluster cluster.

[0020] Optionally, the step of obtaining distance consistency comprises: for any cluster cluster, obtaining the average Euclidean distance between the initial positions of all historical data points in the neighborhood and the cluster center, and taking the ratio of the average Euclidean distance to the Euclidean distance of the update data point to the cluster center as the distance consistency.

[0021] The beneficial effect is that the Euclidean distance of the update data point to the cluster center of the cluster cluster and the Euclidean distance of the historical data points in the neighborhood of the update data point to the cluster center of the cluster cluster are compared, thereby reflecting the relative closeness of the update data point to the cluster center of the cluster cluster, and further reflecting the possibility of the update data point belonging to the cluster cluster.

[0022] Optionally, the step of obtaining a trend probability comprises: taking the ratio of the direction consistency of the historical data points in the neighborhood of the update data point in the cluster cluster to the Euclidean distance of the historical data points to the cluster center of the cluster cluster as trend synchronism, and taking the mean of multiple trend synchronisms as the trend probability.

[0023] Optionally, the step of classifying the update data point according to the classification probabilities of the multiple cluster clusters corresponding to the update data point comprises: taking the sum of the classification probabilities of the multiple cluster clusters corresponding to the update data point as a classification basis, taking the ratio of the classification probability corresponding to any cluster cluster to the classification basis as a membership score, and classifying the update data point into the cluster cluster with the largest membership score.

[0024] The beneficial effect is that each cluster cluster corresponds to a membership score, and the larger the membership score, the more likely the update data point belongs to the cluster cluster corresponding to the cluster score, so the update data point is classified into the cluster cluster with the largest membership score.

[0025] The present application has the following technical effects:

[0026] According to the historical data of the customer, the customer is classified as a historical data point, a mean shift clustering algorithm is used in the classification process, and the running track of the historical data point is recorded in the iteration process of the historical data point, and the risk consistency is analyzed. After the update data point is generated subsequently, only the approximation of the update data point and the historical data point in the neighborhood of the update data point needs to be compared, so as to judge which cluster the update data point belongs to, and realize the rapid classification of the update data point. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a method flowchart of an AI data management platform data processing method.

[0028] Figure 2 is a method flowchart of step S2 of an AI data management platform data processing method. DETAILED DESCRIPTION

[0029] The embodiment of the application discloses an AI data management platform data processing method. In a sales management platform, the system continuously collects multiple data of customers in a sales cycle, including purchase frequency, order amount, order time interval, recent active time, consultation frequency and other dimensions. The platform hopes to build a customer clustering model according to the above data, which is used for implementing customer differentiated management, pushing strategy allocation and resource priority sorting. When new behavior data of the customer (such as a new order or consultation) is generated, the system needs to timely identify the customer cluster to which the customer belongs, so as to ensure the timeliness and accuracy of customer classification. In the method, first, historical data of the customer cluster is collected, a mean shift clustering algorithm is used to cluster the customers in the customer cluster, and in the clustering process, the drift track of each historical data point (a data point composed of multiple customer historical data, which can also be understood as a customer) is recorded, and the direction consistency of the clustering direction is analyzed. When the update data point appears subsequently, the clustering cluster to which the update data point belongs is determined based on the historical data points in the neighborhood of the update data point, so that the global clustering operation is avoided, and the data management platform considers the accuracy and timeliness.

[0030] REFERENCE Figure 1 An AI data management platform data processing method includes steps S1-S5.

[0031] S1: Obtain multiple historical data of each customer in a customer cluster; each customer is regarded as a historical data point, and multiple customers in the customer cluster are clustered based on the multiple historical data corresponding to each historical data point using a mean shift clustering algorithm, to obtain multiple clustering clusters.

[0032] The plurality of historical data of the customer in the statistical sales process includes at least two of the purchase frequency, the time of the last purchase, the peak purchase power of the customer, the single price of the customer, and the average number of consultations per month, so as to more comprehensively analyze the information of the customer and help the salesperson to efficiently sign the order.

[0033] The plurality of historical data of each customer is taken as a multi-dimensional feature vector, which is recorded as a historical data point. By combining the historical data points corresponding to a plurality of customers to form a customer feature data set, a data basis for clustering is constructed.

[0034] Based on the historical data point, a mean shift clustering algorithm is used to cluster a plurality of customers in the customer cluster. Specifically, a kernel density estimation method is used to construct a density function in the feature space, and the density gradient direction of each data point is iteratively calculated based on the density function, and the data point is constantly moved along the gradient direction until it converges to a local density peak area, that is, a class cluster center is formed. The mean shift clustering algorithm is a conventional technical means in the art, which will not be described here.

[0035] Finally, the system can obtain a plurality of class cluster centers and their corresponding customer subsets, forming a plurality of customer clustering clusters. Each clustering cluster represents a customer cluster, and the cluster has certain similarity in order data features, which can be used for subsequent personalized analysis, customer stratification, marketing strategy formulation and other application scenarios.

[0036] S2: analyze the iteration position of the data point in the mean shift clustering process, and calculate the direction consistency, which is used to represent the degree of coincidence between the moving direction of the historical data point in the iteration process and the direction of the clustering center.

[0037] Referring to Figure 2 , step S2 includes steps S21-S23.

[0038] After completing the mean shift clustering operation on the customer cluster, in order to further improve the classification accuracy of the real-time updated data point, the system analyzes the iteration position change of each historical data point in the clustering process to extract a trend index that can reflect the stability of the clustering direction of the data point, that is, the direction consistency.

[0039] Specifically, during the execution of the mean shift clustering algorithm, the system records the feature vector update of each historical data point in each iteration. Each data point will move to a high-density area according to the gradient direction calculated by the kernel density function in each iteration, until it converges to a class cluster center. By tracking the moving track of the data point from the initial position to the final clustering center, the system can evaluate the trend stability based on whether the displacement direction of the data point in each iteration is consistent with the final target direction.

[0040] S21: For any one historical data point in any iteration process of the mean shift clustering process, obtain the local change of each historical data value and the total change of the historical data after clustering is completed; and take the ratio of the local change to the total change as the center movement trend of the historical data of the data point in the iteration.

[0041] For any one historical data point after any iteration, the center movement trend of any one historical data can be expressed by the following formula: In the formula, represents the center movement trend of the i-th data in the j-th iteration process of the i-th historical data point; represents the data value of the i-th data after the j-th iteration of the i-th historical data point; represents the data value of the i-th data after the j-th iteration of the i-th historical data point; represents the data value of the i-th data after clustering is completed (i.e. after the last iteration) of the i-th historical data point; represents the data value of the i-th data before iteration of the i-th historical data point.

[0042] In the formula, represents the local change of the data value of the i-th data before and after the j-th iteration, which reflects the movement degree of the i-th data to the clustering center in the j-th iteration. represents the total movement degree of the data point before and after clustering. represents the proportion between the change of the i-th feature in the j-th iteration and the total change of the feature, which is used to reflect whether the moving direction of the i-th feature value in the j-th iteration is consistent with the converging direction. The larger the ratio is, the more consistent the direction of the iteration is with the overall trend, and the greater the contribution is; and the greater the center movement trend is.

[0043] S22: Obtain the change degree of the i-th historical data in the j-th iteration based on the center movement trend.

[0044] Set a control value, and take the sum of the normalized result of the center movement trend and the control value as the change degree.

[0045] In this embodiment, the arctangent function is used to normalize the center movement trend. Specifically, the calculation formula of the change degree can be expressed as: ​​​​​​​​​​​​

[0046] wherein, represents the first order historical data point in the first iteration process; represents the degree of change of the first represents the arc tangent function; is a control value, mainly used to control the value range of the final so that the output value of the arc tangent function is shifted upward as a whole, ensuring the positive value of the degree of change.

[0047] S23: Obtain the direction consistency of the historical data point based on the iteration trend corresponding to multiple iterations, and the iteration trend is proportional to the direction consistency.

[0048] The mean of the degrees of change corresponding to multiple historical data is taken as the iteration trend.

[0049] The geometric mean of the iteration trends of multiple iterations is taken as the direction consistency of the historical data point.

[0050] This step combines multiple iteration processes and multiple historical data to reflect the direction consistency of the data point.

[0051] S3: For any historical data point, the order data corresponding thereto is updated to form an updated data point, and the neighborhood of the updated data point is determined based on the updated data point.

[0052] In this embodiment, after the updated data point is formed, the spatial position of the updated data point is determined, a preset number of nearest historical data points are selected near the updated data point, the mean of the distances between the updated data point and the multiple historical data points is taken as the neighborhood radius, and the circular region with the updated data point as the center and the neighborhood radius as the radius is taken as the neighborhood of the updated data point.

[0053] In other embodiments, the neighborhood radius can also be directly preset, for example, the neighborhood radius is directly preset to 10.

[0054] S4: For any clustering cluster, analyze the direction consistency of the historical data points in the neighborhood of the updated data point and the spatial distance thereof from the clustering cluster center, and calculate the classification probability of the updated data point belonging to the clustering cluster.

[0055] In one embodiment, the step of calculating the classification probability of the updated data point belonging to the clustering cluster comprises: analyzing the distances of the historical data points in the neighborhood to the clustering center of the clustering cluster and the distance of the updated data point to the clustering center, obtaining distance consistency; obtaining trend probability based on the direction consistency of the historical data points in the neighborhood of the updated data point in the clustering cluster; and taking the product of the trend probability and the distance consistency as the classification probability of the updated data point belonging to the clustering cluster.

[0056] When updating a data point, multiple historical data points within its neighborhood may belong to different clusters. This means the updated data point may belong to one of these clusters. Therefore, analyzing the historical data points within the neighborhood of the updated data point allows us to determine the probability that the updated data point belongs to different clusters.

[0057] First, based on the different clusters to which different historical data points belong in the neighborhood, the clusters in the neighborhood can be divided into multiple groups, with each group corresponding to one cluster. For any cluster, the distance from the initial position of the historical data point in the neighborhood (i.e., the position before iteration) to the cluster center is analyzed, and the distance from the updated data point to the cluster center is compared to obtain the distance consistency.

[0058] In this embodiment, the average Euclidean distance between the initial position of all historical data points in the neighborhood and the cluster center is calculated. The ratio of the average Euclidean distance to the Euclidean distance from the updated data point to the cluster center is used as the distance consistency. To prevent the distance consistency from being too large and affecting subsequent calculations, a minimum value function is used to limit the range of distance consistency values.

[0059] Specifically, the formula for calculating distance consistency can be expressed as: In the formula, Indicates updating data points and the first Distance consistency between clusters; This indicates that the data point in the neighborhood of the updated data point belongs to the th _ ... The average Euclidean distance from the historical data points of each cluster to the cluster center; This represents the Euclidean distance between the updated data point and the cluster center; It is a minimum value function.

[0060] In the formula, The Euclidean distances from multiple historical data points to the cluster centers were compared with the Euclidean distances from updated data points to the cluster centers. If the Euclidean distance between the updated data point and the cluster center was less than the average Euclidean distance for the historical data points, the updated data point was more likely to belong to that cluster because it was closer to the cluster center. The larger this part is, the greater the consistency between the updated data point and the cluster.

[0061] Then, the trend probability is obtained based on the directional consistency of historical data points located in the neighborhood of the updated data point in the cluster.

[0062] In this embodiment, the ratio of the directional consistency of historical data points in the neighborhood of the updated data point in the cluster to the Euclidean distance from the historical data point to the cluster center is used as the trend synchronicity, and the mean of multiple trend synchronicities is used as the trend probability.

[0063] In another embodiment, the step of calculating the classification probability of the update data point belonging to the cluster cluster comprises: obtaining distance consistency by analyzing the distance of the historical data points in the neighborhood to the cluster center of the cluster cluster and the distance of the update data point to the cluster center of the cluster cluster; obtaining quantity consistency by taking the ratio of the number of the historical data points belonging to the cluster cluster in the neighborhood to the number of all the historical data points in the neighborhood; obtaining trend probability based on the direction consistency of the historical data points in the cluster cluster located in the neighborhood of the update data point; and taking the product of the trend probability, the quantity consistency and the distance consistency as the classification probability of the update data point belonging to the cluster cluster.

[0064] The steps of obtaining the distance consistency and the trend probability are the same as the above steps of obtaining the distance consistency and the trend probability, and will not be repeated here.

[0065] Meanwhile, in the embodiment, for any cluster cluster corresponding to the update data point, the number of the historical data points belonging to the cluster cluster in the neighborhood of the update data point is counted, and the number of the data points in the neighborhood of the update data point (20 in the embodiment) is counted; the ratio of the number of the historical data points belonging to the cluster cluster in the neighborhood to the number of all the historical data points in the neighborhood is taken as quantity consistency. The product of the trend probability, the quantity consistency and the distance consistency is taken as the classification probability of the update data point belonging to the cluster cluster. In the method, if the historical data points in the neighborhood of the update data point mostly belong to a cluster cluster, then the update data point also has a high probability of belonging to the cluster cluster, and thus the number of the historical data points in the neighborhood of the update data point belonging to a cluster cluster can also reflect the possibility of the update data point belonging to the cluster cluster to some extent, thereby further improving the accuracy of the classification of the update data point.

[0066] In the method, the quantity consistency is combined, thereby more comprehensively reflecting the classification probability between the update data point and the cluster cluster.

[0067] S5: classifying the update data point according to the classification probabilities of the multiple cluster clusters corresponding to the update data point.

[0068] After obtaining the classification probabilities, the classification probabilities are normalized. Specifically, the sum of the classification probabilities of the multiple cluster clusters corresponding to the update data point is taken as a classification basis, the ratio of the classification probability of any cluster cluster to the classification basis is taken as a normalization result of the classification probability, and the normalization result is taken as the attribution score of the update data point belonging to the cluster cluster. Then, by comparing the attribution scores of different cluster clusters, the update data point is divided into the cluster cluster with the maximum attribution score.

[0069] In the subsequent process of classifying the update data point, the attribution score of the update data point is taken as the direction consistency, so as to classify the newly generated data point in real time.

[0070] The classification processing of the data points is performed according to the attribution score of the updated data points in real time, the classification of the updated data points is performed by learning the classification mode of the data points close in Euclidean distance, the redundant calculation amount of the conventional mean shift clustering is saved, the accurate classification is performed on the updated data points by learning the classification of the data points close thereto, and the fast and accurate classification of the real-time data points is realized.

[0071] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application, so: any equivalent changes made on the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A data processing method for an AI data management platform, characterized in that, Obtaining a plurality of historical data of each customer in the customer cluster, taking each customer as a historical data point, clustering the plurality of customers in the customer cluster based on the plurality of historical data corresponding to each historical data point using mean shift clustering to obtain a plurality of clustering clusters; analyzing the iteration position of the historical data point in the mean shift clustering process, and calculating the direction consistency, the direction consistency being used to represent the degree of coincidence between the moving direction of the historical data point in the iteration process and the direction of the clustering center, comprising: for any iteration process of any historical data point in the mean shift clustering process, obtaining the local change of each historical data value and the total change of the historical data after clustering; taking the ratio of the local change and the total change as the center motion trend of the historical data of the data point in the iteration; obtaining the change degree of the historical data in the iteration based on the center motion trend; taking the mean value of the change degree of the plurality of historical data as the iteration trend, and obtaining the direction consistency of the historical data point based on the iteration trend corresponding to a plurality of iterations, the iteration trend being proportional to the direction consistency; For any historical data point, the order data corresponding to the historical data point is updated to form an updated data point, and a neighborhood of the updated data point is determined based on the updated data point; for any clustering cluster, analyzing the direction consistency of the historical data points in the neighborhood of the updated data point and the spatial distance between the historical data points in the neighborhood of the updated data point and the center of the clustering cluster, calculating the classification probability of the updated data point belonging to the clustering cluster, and classifying the updated data point according to the classification probability of the plurality of clustering clusters corresponding to the updated data point.

2. The data processing method for an AI data management platform according to claim 1, wherein, For any iteration process of any historical data of the historical data point, the data difference before and after the iteration of the historical data is taken as the local change of the historical data in the iteration.

3. The data processing method for an AI data management platform according to claim 1, wherein, The step of obtaining the change degree of the historical data in the iteration based on the center motion trend comprises: setting a control value, and taking the sum of the normalized result of the center motion trend and the control value as the change degree.

4. The data processing method for an AI data management platform according to claim 1, characterized in that, The geometric mean of the plurality of iteration trends is taken as the direction consistency of the historical data point.

5. The data processing method for an AI data management platform according to claim 1, characterized in that, The step of calculating the classification probability of the updated data point belonging to the clustering cluster comprises: analyzing the distance between the historical data points in the neighborhood and the center of the clustering cluster and the distance between the updated data point and the center of the clustering cluster, obtaining the distance consistency; obtaining the trend probability based on the direction consistency of the historical data points in the neighborhood of the updated data point in the clustering cluster; and taking the product of the trend probability and the distance consistency as the classification probability of the updated data point belonging to the clustering cluster.

6. The data processing method for an AI data management platform according to claim 1, wherein, The step of calculating the classification probability of the updated data point belonging to the clustering cluster comprises: analyzing the distance between the historical data points in the neighborhood and the center of the clustering cluster and the distance between the updated data point and the center of the clustering cluster, obtaining the distance consistency; taking the ratio of the number of historical data points belonging to the clustering cluster in the neighborhood and all historical data points in the neighborhood as the quantity consistency; obtaining the trend probability based on the direction consistency of the historical data points in the neighborhood of the updated data point in the clustering cluster; and taking the product of the trend probability, the quantity consistency and the distance consistency as the classification probability of the updated data point belonging to the clustering cluster.

7. The data processing method for an AI data management platform according to claim 6, characterized in that, The step of obtaining the distance consistency comprises: obtaining, for any one of the clustering clusters, an average Euclidean distance between initial positions of all historical data points in the neighborhood and the clustering cluster center, and taking a ratio of the average Euclidean distance to a Euclidean distance of the updating data point to the clustering cluster center as the distance consistency. 8.The data processing method for an AI data management platform according to claim 6, wherein, The step of obtaining the trend probability comprises: taking a ratio of the direction consistency of the historical data points in the neighborhood of the updating data point to the clustering cluster center to a Euclidean distance of the historical data points to the clustering cluster center as the trend consistency, and taking a mean value of a plurality of trend consistencies as the trend probability.

9. The data processing method for an AI data management platform according to claim 1, wherein, The step of classifying the updating data point according to the classification probabilities of the plurality of clustering clusters corresponding to the updating data point comprises: taking a sum of the classification probabilities of the plurality of clustering clusters corresponding to the updating data point as a classification basis, taking a ratio of the classification probability of any one of the clustering clusters to the classification basis as a belonging score, and classifying the updating data point into a clustering cluster with the largest belonging score.

Images