Data processing method, device and server

Abstract

The present application provides a data processing method, device and server, which relate to data processing technology. The method includes: obtaining a prediction request, and obtaining business data corresponding to both a user identifier and a business scenario identifier. Clustering the business data according to a preset clustering algorithm to obtain multiple clusters. According to a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm, data prediction is performed on multiple clusters to determine the business behavior information at the target moment; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier. The method of the present application has the advantages of high robustness and high real-time performance when applied in the fields of precision marketing, anti-money laundering, anti-fraud, etc., and solves the technical problem that it is difficult to determine the business behavior information of the classification body at the preset moment.

Description

Technical Field

[0001] The present application relates to data processing technology, and in particular to a data processing method, device and server. Background Art

[0002] At present, bank transaction data is huge and complex, containing high-quality information in multiple dimensions. Transaction information usually includes information such as time, location, merchant, product, and consumer, and bank transaction data needs to be analyzed.

[0003] In the prior art, cluster analysis of bank transaction data is performed based on the K-MEANS algorithm.

[0004] However, in the existing technology, since only a simple clustering analysis can be performed on the existing bank transaction data based on the K-MEANS algorithm, the clustering results of the classification body that have occurred are obtained. If the clustering results that have occurred are dangerous behaviors, it will be impossible to prevent them in time before they occur, and thus it will be impossible to effectively monitor the market. Summary of the Invention

[0005] The present application provides a data processing method, device and server to solve the technical problem of high difficulty in determining the business behavior information of a classification body at a preset time.

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

[0007] Obtaining a prediction request, wherein the prediction request includes a user identifier and a business scenario identifier, and obtaining business data corresponding to both the user identifier and the business scenario identifier;

[0008] Clustering the business data according to a preset clustering algorithm to obtain a plurality of clusters; wherein the preset clustering algorithm is used to iteratively classify the business data, and each cluster includes a plurality of business data;

[0009] According to a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm, data prediction is performed on the multiple clusters to determine business behavior information at the target moment; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier.

[0010] Furthermore, clustering the business data according to a preset clustering algorithm to obtain multiple clusters includes:

[0011] Determine a plurality of initial centroids in the business data, wherein each business data is a sample point;

[0012] Generate a cluster corresponding to each of the multiple initial centroids; wherein each cluster corresponds to a cluster category;

[0013] Sequentially traverse the sample points in the business data except the multiple initial centroids, respectively, and determine the distance between each of the sample points and the initial centroid, and determine the sample point and the initial centroid corresponding to the minimum distance;

[0014] According to the cluster category corresponding to the cluster where each initial centroid is located, determine the cluster category to which the sample point corresponding to the minimum distance belongs, and classify the sample point into the cluster corresponding to the corresponding cluster category;

[0015] Based on a preset formula, determine the new centroid of each cluster;

[0016] Based on the new centroid in each cluster, the sample points in the business data except the multiple initial centroids are traversed in sequence, and the distance between each of the initial centroids is determined, and the sample points and the initial centroid corresponding to the minimum distance are determined. When it is determined that the new centroid in each cluster remains unchanged, the traversal is stopped and multiple clusters are obtained.

[0017] Furthermore, performing data prediction on the multiple clusters according to a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine the business behavior information at the target moment includes:

[0018] Determining the next target moment after the current moment according to a preset time interval in the multi-Bernoulli filter algorithm;

[0019] According to a preset multi-Bernoulli filtering algorithm, data prediction is performed on the multiple clusters to determine the business behavior information at the target moment.

[0020] Furthermore, the method further comprises:

[0021] If it is determined that the business behavior information at the target moment does not meet the preset business threshold, a prompt message is issued.

[0022] In a second aspect, the present application provides a data processing device, comprising:

[0023] an acquiring unit, configured to acquire a prediction request, wherein the prediction request includes a user identifier and a business scenario identifier, and acquire business data corresponding to both the user identifier and the business scenario identifier;

[0024] a clustering unit, configured to cluster the business data according to a preset clustering algorithm to obtain a plurality of clusters; wherein the preset clustering algorithm is configured to iteratively classify the business data, and each cluster includes a plurality of business data;

[0025] A prediction unit is configured to perform data prediction on the multiple clusters based on a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine business behavior information at a target moment; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier.

[0026] Furthermore, the clustering unit includes:

[0027] A first determination module is configured to determine a plurality of initial centroids in the business data, wherein each business data is a sample point;

[0028] A generating module, configured to generate a cluster corresponding to each of the plurality of initial centroids; wherein each cluster corresponds to a cluster category;

[0029] A second determining module is configured to sequentially traverse the distance between each of the sample points in the business data except the multiple initial centroids and each of the initial centroids, and determine the sample point and the initial centroid corresponding to the minimum distance;

[0030] A third determination module is used to determine the cluster category to which the sample point corresponding to the minimum distance belongs based on the cluster category corresponding to the cluster where each initial centroid is located, and classify the sample point into the cluster corresponding to the corresponding cluster category;

[0031] a fourth determination module, configured to determine a new centroid in each cluster based on a preset formula;

[0032] The acquisition module is used to perform a step of traversing the sample points in the business data except the multiple initial centroids based on the new centroid in each cluster, and determining the distance between each sample point and each initial centroid, and determining the sample point and the initial centroid corresponding to the minimum distance, until it is determined that the new centroid in each cluster remains unchanged, stopping the traversal and obtaining multiple clusters.

[0033] Furthermore, the prediction unit includes:

[0034] a fifth determination module, configured to determine a next target time after the current time according to a preset time interval in the multi-Bernoulli filter algorithm;

[0035] The prediction module is used to perform data prediction on the multiple clusters according to a preset multi-Bernoulli filtering algorithm to determine the business behavior information at the target moment.

[0036] Furthermore, the device further comprises:

[0037] The prompt unit is configured to issue a prompt message if it is determined that the business behavior information at the target moment does not meet a preset business threshold.

[0038] In a third aspect, the present application provides a server comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method described in the first aspect is implemented.

[0039] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the method described in the first aspect.

[0040] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which implements the method described in the first aspect when executed by a processor.

[0041] The present application provides a data processing method, device, and server for obtaining a prediction request, wherein the prediction request includes a user identifier and a business scenario identifier, and obtaining business data corresponding to both the user identifier and the business scenario identifier. The business data is clustered according to a preset clustering algorithm to obtain multiple clusters. The preset clustering algorithm is used to iteratively classify the business data, with each cluster containing multiple business data. Data prediction is performed on the multiple clusters based on a preset multi-Bernoulli filtering algorithm and a preset time interval within the multi-Bernoulli filtering algorithm to determine business behavior information at a target time. The preset multi-Bernoulli filtering algorithm is used to predict historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier. In this solution, business data corresponding to both the user identifier and the business scenario identifier is obtained, clustered according to a preset clustering algorithm, and divided into multiple groups, i.e., multiple clusters are obtained. Data prediction is then performed on the multiple clusters based on the preset multi-Bernoulli filtering algorithm and a preset time interval within the multi-Bernoulli filtering algorithm to determine business behavior information at a target time, thereby determining the business behavior of the user corresponding to the user identifier. Therefore, first, business data is clustered in multiple dimensions based on a clustering algorithm, and complex bank transaction data is clustered into multi-dimensional clusters. Second, clustering targets are predicted based on a multi-Bernoulli filtering algorithm, and changes in clustered data are determined in real time, with high real-time and robustness. Third, the clustering algorithm is combined with the multi-Bernoulli filtering algorithm to extract specified feature values ​​from bank transaction data for real-time clustering and locking. Through the combination of the above three points, banks can ultimately achieve high-quality clustering and locking processing of complex data, thereby effectively monitoring the market. Applications in precision marketing, anti-money laundering, anti-fraud, and other fields have the advantages of high robustness and high real-time performance, solving the technical problem of the high difficulty in determining the business behavior information of the classifier at a preset time. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0043] Figure 1 A flowchart of a data processing method provided in an embodiment of the present application;

[0044] Figure 2 A flowchart of another data processing method provided in an embodiment of the present application;

[0045] Figure 3 A flowchart of another data processing method provided in an embodiment of the present application;

[0046] Figure 4 A flowchart of another data processing method provided in an embodiment of the present application;

[0047] Figure 5 A schematic diagram of another data processing method provided in an embodiment of the present application;

[0048] Figure 6 A schematic diagram of the structure of a data processing device provided in an embodiment of the present application;

[0049] Figure 7 A schematic diagram of the structure of another data processing device provided in an embodiment of the present application;

[0050] Figure 8 A schematic diagram of the structure of a server provided in an embodiment of the present application.

[0051] The above drawings illustrate specific embodiments of the present disclosure, which will be described in more detail below. These drawings and textual descriptions are not intended to limit the scope of the present disclosure in any way, but rather to illustrate the concepts of the present disclosure to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0052] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the present disclosure.

[0053] The K-means algorithm, with its long history and simple implementation, has been applied to numerous complex data mining scenarios. James MacQueen first coined the term "K-means" in his 1967 paper, "Some Methods for Classifying and Analyzing Multivariate Observations." The algorithm essentially classifies data so that the data within each class has characteristics that are more similar than those in other groups. In this application, it is primarily used to classify bank data, transforming complex and large-scale data into multiple random objects that can be used to determine trajectories.

[0054] The multi-target multi-Bernoulli filter algorithm, proposed by Mahler, originates from the theory of random finite sets (RFS). Compared to the complex data correlation and high time cost of multi-target tracking algorithms developed using data association, the tracking algorithm based on the RFS framework has minimal data correlation, enabling fast, efficient, and robust trajectory determination for a variable number of multi-target tracking systems. However, in practical engineering applications, multidimensional integration problems inevitably arise in the multi-target state space, resulting in a significant computational overhead, or even insolvability. To address this issue, Mahler proposed the Probability Hypothesis Density (PHD) filter algorithm after research. Compared to the RFS theory, which directly calculates the joint posterior probability distribution of all multiple targets in a set, the PHD algorithm uses a simple recursive form to estimate the first-order statistics of the random set variables, namely the probability hypothesis density, and then extracts the state vector of each individual target from this first-order statistic. However, many researchers have found that when using the PF PHD filtering framework, extracting the target's estimated state involves computationally intensive calculations such as clustering, which imposes a high computational cost on the algorithm and reduces its overall stability. Subsequently, Mahler et al. proposed Multi-Target Multi-Bernoulli (MeMBer) filtering. Ba-Tuong Vo et al. improved and standardized the Bayesian method in multi-target systems and studied a multi-Bernoulli filtering algorithm that can be applied to practical engineering projects. The study found that the multi-Bernoulli filtering algorithm can accurately determine the trajectory of an unknown number of targets without the need for complex data association.

[0055] In one example, bank transaction data is vast and complex, containing high-quality, multi-dimensional information. This information typically includes time, location, merchant, product, and consumer information, necessitating analysis. However, existing techniques rely solely on clustering data using the K-MEANS algorithm, yielding clustering results for existing entities. If these clusters indicate dangerous behavior, it's impossible to prevent it before it occurs, hindering effective market monitoring.

[0056] The present application provides a data processing method, device, and server, which aim to solve the above technical problems in the prior art.

[0057] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0058] Figure 1 A flow chart of a data processing method provided in an embodiment of the present application is shown as follows: Figure 1 As shown, the method includes:

[0059] 101. Obtain a prediction request, where the prediction request includes a user identifier and a business scenario identifier, and obtain business data corresponding to both the user identifier and the business scenario identifier.

[0060] By way of example, the execution entity of this embodiment may be a server. First, the server obtains a prediction request. Specifically, a user clicks a "Start Prediction" button on a display screen corresponding to the server, triggering the generation of a prediction request, which the server then obtains. Alternatively, the server receives a prediction request from another server, where the prediction request includes a user identifier and a business scenario identifier. Based on the user identifier and business scenario identifier in the prediction request, the server can obtain business data corresponding to both the user identifier and the business scenario identifier.

[0061] 102. Cluster the business data according to a preset clustering algorithm to obtain multiple clusters; wherein the preset clustering algorithm is used to iteratively classify the business data, and each cluster includes multiple business data.

[0062] For example, the preset clustering algorithms include the K-MEANS clustering algorithm, etc. The K-MEANS clustering algorithm is a typical unsupervised learning algorithm that mainly automatically groups similar samples into one category. The essence of the algorithm is: a constant K is determined in advance, which means the final number of cluster categories. First, an initial point is randomly selected as the centroid, and the sample points are classified into the most similar class by calculating the similarity (i.e., Euclidean distance) between each sample and the centroid; then, the centroid (class center) of each class is recalculated, and this process is repeated until the centroid no longer changes, and finally the category to which each sample belongs and the centroid of each class are determined.

[0063] The K-MEANS clustering algorithm treats business data as sample points and iteratively classifies multiple sample points to obtain multiple clusters, each of which includes multiple sample points. Therefore, after the server obtains multiple business data, it can cluster the business data according to the preset clustering algorithm to obtain multiple clusters.

[0064] 103. Based on a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm, data prediction is performed on multiple clusters to determine business behavior information at a target moment; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier.

[0065] For example, the pre-set multi-Bernoulli filtering algorithm: Multi-objective Multi-Bernoulli (MeMBer) recursion is based on the premise that the multi-objective random finite set is estimated by the multi-Bernoulli random finite set in each time period, based on the following modeling assumptions:

[0066] 1. The evolution and derivation of each goal is independent;

[0067] 2. The new generation of the target follows the multi-Bernoulli RFS and exists independently of the target;

[0068] 3. The clutter follows Poisson RFS and cannot be too dense. It also generates measurement values ​​for independent targets.

[0069] In this step, the server can determine the current moment, determine the next target moment after the current moment based on the preset time interval in the multi-Bernoulli filtering algorithm, and finally perform data prediction on multiple clusters based on the preset multi-Bernoulli filtering algorithm to determine the business behavior information at the target moment. The business behavior information represents the business behavior of the user corresponding to the user identifier at the target moment.

[0070] For example, Figure 2 A flow chart of another data processing method provided in an embodiment of the present application is shown as follows: Figure 2 As shown, according to Figure 2It can be seen that after cluster 1 obtains the clustering result at time T, it can continue to predict cluster 1 according to the multi-Bernoulli filtering algorithm to obtain the business behavior information of cluster 1 at time T+1; the prediction process of cluster 2 refers to cluster 1 and will not be repeated here.

[0071] In an embodiment of the present application, a prediction request is obtained, wherein the prediction request includes a user identifier and a business scenario identifier, and business data corresponding to both the user identifier and the business scenario identifier is obtained. The business data is clustered according to a preset clustering algorithm to obtain multiple clusters; wherein the preset clustering algorithm is used to iteratively classify the business data, with each cluster including multiple business data. Data prediction is performed on the multiple clusters according to a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine business behavior information at a target time; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier. In this solution, business data corresponding to both the user identifier and the business scenario identifier is obtained, clustered according to a preset clustering algorithm, and divided into multiple groups, i.e., multiple clusters are obtained. Data prediction is then performed on the multiple clusters according to the preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine business behavior information at a target time, thereby determining the business behavior of the user corresponding to the user identifier. Therefore, first, business data is clustered in multiple dimensions based on a clustering algorithm, and complex bank transaction data is clustered into multi-dimensional clusters. Second, clustering targets are predicted based on a multi-Bernoulli filtering algorithm, and changes in clustered data are determined in real time, with high real-time and robustness. Third, the clustering algorithm is combined with the multi-Bernoulli filtering algorithm to extract specified feature values ​​from bank transaction data for real-time clustering and locking. Through the combination of the above three points, banks can ultimately achieve high-quality clustering and locking processing of complex data, thereby effectively monitoring the market. Applications in precision marketing, anti-money laundering, anti-fraud, and other fields have the advantages of high robustness and high real-time performance, solving the technical problem of the high difficulty in determining the business behavior information of the classifier at a preset time.

[0072] Figure 3 A flow chart of another data processing method provided in an embodiment of the present application is shown as follows: Figure 3 As shown, the method includes:

[0073] 201. Obtain a prediction request, where the prediction request includes a user identifier and a business scenario identifier, and obtain business data corresponding to both the user identifier and the business scenario identifier.

[0074] Exemplarily, when the server obtains the prediction request, it obtains the business data corresponding to both the user identifier and the business scenario identifier based on the user identifier and the business scenario identifier in the prediction request.

[0075] Since each business data is a sample point, the sample set D = {x1, x2, ...x m}, the cluster tree is k, the maximum number of iterations is N, and the output is the cluster partition C = {C1, C2, ...C k}, that is, there are k clusters in total.

[0076] 202. Determine multiple initial centroids in the business data; each business data is a sample point.

[0077] For example, for the K-Means algorithm, the first thing to pay attention to is the selection of the k value. An appropriate k value is selected based on prior experience with the data, for example, based on multiple business types in the business scenario. The server determines k initial centroids in the business data.

[0078] The location of the k initial mass centers has a great impact on the final clustering results and running time. Therefore, it is important to select appropriate k mass centers, and the mass centers should not be too close. After determining the number of k, it is necessary to select k initial mass centers. The initial mass centers can be randomly selected. For example, k sample points are randomly selected from the sample set D as the initial mass center vectors: {μ1, μ2, ..., μ k}, the initial centroid vector is the initial centroid.

[0079] 203. Generate a cluster corresponding to each of the multiple initial centroids; wherein each cluster corresponds to a cluster category.

[0080] For example, for the number of iterations n=1, 2, ..., N, the cluster partition C is initialized as Generate a cluster corresponding to each initial centroid in the multiple initial centroids, that is, each initialized cluster contains an initial centroid, wherein each cluster corresponds to a cluster category.

[0081] 204. Sequentially traverse the sample points in the business data except for the multiple initial centroids, calculate the distance between each sample point and each initial centroid, and determine the sample point and the initial centroid corresponding to the minimum distance.

[0082] For example, for i=1, 2...m, calculate the sample point x in the sample set i and each initial centroid vector μ j The distance d between (j=1, 2, ... k) ij :

[0083]

[0084] Determine the sample point x i and each initial centroid vector μ j The distance dij After that, multiple distances are compared to determine the minimum distance and the sample point x corresponding to the minimum distance. i with the initial centroid.

[0085] 205. According to the cluster category corresponding to the cluster where each initial centroid is located, determine the cluster category to which the sample point corresponding to the minimum distance belongs, and classify the sample point into the cluster corresponding to the corresponding cluster category.

[0086] For example, since each cluster where the initial centroid is located corresponds to a cluster category, the server can determine the sample point x corresponding to the minimum distance i Cluster category And update the cluster category Furthermore, according to The sample point x i Classify into corresponding cluster categories corresponding clusters.

[0087] 206. Based on the preset formula, determine the new centroid of each cluster.

[0088] For example, based on a preset formula, for j=1, 2, ..., k, a new centroid is recalculated for all sample points in each cluster Cj. The preset formula is as follows:

[0089]

[0090] 207. Based on the new centroid in each cluster, perform a step of traversing the sample points in the business data except the multiple initial centroids, and determining the distance between each sample point and each initial centroid, and determining the sample point and the initial centroid corresponding to the minimum distance, until it is determined that the new centroid in each cluster remains unchanged, stop traversing and obtain multiple clusters.

[0091] For example, based on the new centroid in each cluster, steps 204 to 206 are iteratively performed until it is determined that the new centroid in each cluster remains unchanged, then the traversal is stopped and the cluster partition C = {C1, C2, ... C k}.

[0092] For example, Figure 4 A flow chart of another data processing method provided in an embodiment of the present application is shown as follows: Figure 4 As shown, after clustering from step (a) to step (f), the final two clusters can be obtained.

[0093] 208. Determine the next target time after the current time according to the preset time interval in the multi-Bernoulli filter algorithm.

[0094] Exemplarily, the server may determine the current moment, and determine the next target moment after the current moment according to a preset time interval in the multi-Bernoulli filter algorithm.

[0095] 209. According to the preset multi-Bernoulli filtering algorithm, data prediction is performed on multiple clusters to determine the business behavior information at the target time.

[0096] For example, the multi-objective multi-Bernoulli (MeMBer) recursion is summarized in Propositions 1 and 2 as follows. The original equation can be found in the literature ([1] Mahler R. Statistical multisource-multitarget information fusion [M]. Norwood, MA: Artech House, 2007.). The Bernoulli filter algorithm is described below. The algorithm is mainly divided into two parts: prediction and update.

[0097] Proposition 1 (Multi-target Multi-Bernoulli (MeMBer) prediction): Assume that at time k-1, the posterior probability density function of the multi-target can be represented by a set of multi-Bernoulli parameters:

[0098]

[0099] Then, the multi-target probability density after prediction is also a multi-Bernoulli, that is:

[0100]

[0101] Among them, the following formula (2):

[0102]

[0103]

[0104] f k|k-1 (·|ζ) = the transition density of a single target at time k given the previous state ζ;

[0105] p S，k (ζ) = probability of the target existing at time k, given the previous state ζ;

[0106]

[0107] In essence, the multi-Bernoulli parameter set for the predicted multi-target probability density is composed of the multi-Bernoulli parameter set of the surviving target and the number of new targets. The total number of parameters of the hypothetical trajectory target after prediction is .

[0108] Proposition 2 (Multi-objective Multi-Bernoulli (MeMBer) Update): Assume that at time k, the predicted multi-objective probability density is a Multi-Bernoulli form:

[0109]

[0110] Then, the posterior multi-target probability density can be estimated by the following multi-Bernoulli:

[0111]

[0112] in,

[0113]

[0114]

[0115]

[0116]

[0117] Ψ k，Z (x) = g k (z|x)p D，k (x)

[0118] Z k = the set of measurements at time k,

[0119] g x (#|x)=the measurement likelihood of a single target given the state x at time k,

[0120] p D，k (x) = the probability of target detection when the state x is known at time k,

[0121] K k (z) = Poisson clutter probability density at time k.

[0122] From the above, we can see that it is implicitly assumed that p D，k and It is impossible to be equal to 1. In fact, for the updated multi-target probability density π k The multi-Bernoulli parameter set of is composed of two parts: the multi-Bernoulli parameter set of missed targets and the multi-Bernoulli parameter set of measurement updates. The total number of posterior hypothesis trajectories is M k =M k|k-1 +|Z k |.

[0123] Here, we use Figure 5 To show the state transition of multiple targets over time during the prediction update process, such as Figure 5 As shown, Figure 5 A schematic diagram of a scenario of another data processing method provided in an embodiment of the present application.

[0124] If the prediction step of the multi-objective multi-Bernoulli recursion is accurate, then based on the estimate of PGFl for the posterior multi-objective state at time k, the data update step is:

[0125]

[0126] in,

[0127]

[0128]

[0129]

[0130] q D，k =1-p D，k

[0131] This is a reasonable approximation when the clutter is not very dense. For the conclusion behind this approximation, we can refer to Mahler's original derivation in [1].

[0132] However, we can see that in formula (8), the first product is a multi-Bernoulli, but the second product is not. In fact, each factor in the second product may not even be a PGFl of a random finite set. Nevertheless, we still need to find a Bernoulli approximation so that the second product in formula (8) can be approximated by a multi-Bernoulli.

[0133] In the multi-Bernoulli update approximation, Mahler simply sets h in formula (11) to 1, that is:

[0134]

[0135] Substituting formula (10) into the equation, we get the following Bernoulli approximation:

[0136] G U，k [h; z] ≈ 1 - r U，k (z)+r U，k (z) <p U，k (·; z), h> (13)

[0137] r U，k (z) and p U，k (·; z) comes from formulas (6) and (7).

[0138] Based on the above explanation of the principles of the Multi-Bernoulli filter algorithm, it can be determined that the Multi-Bernoulli filter algorithm is suitable for predicting data. Therefore, in this step, data prediction is performed on multiple clusters using the preset Multi-Bernoulli filter algorithm to determine the business behavior information at the target time. The business behavior information represents the business behavior of the user corresponding to the user ID at the target time.

[0139] 210. If it is determined that the business behavior information at the target time does not meet the preset business threshold, a prompt message is issued.

[0140] For example, if it is determined that the business behavior information at the target moment does not meet the preset business threshold, a prompt message is issued to provide a timely alarm.

[0141] In an embodiment of the present application, a prediction request is obtained, wherein the prediction request includes a user identifier and a business scenario identifier, and business data corresponding to both the user identifier and the business scenario identifier is obtained. Multiple initial centroids are determined in the business data, wherein each business data point is a sample point. A cluster corresponding to each of the multiple initial centroids is generated, wherein each cluster corresponds to a cluster category. Sample points in the business data other than the multiple initial centroids are sequentially traversed, and the distance between each initial centroid and the sample point corresponding to the minimum distance from the initial centroid is determined. Based on the cluster category corresponding to the cluster in which each initial centroid resides, the cluster category to which the sample point corresponding to the minimum distance belongs is determined, and the sample point is assigned to the cluster corresponding to the corresponding cluster category. A new centroid is determined in each cluster based on a preset formula. Based on the new centroid in each cluster, the steps of sequentially traversing the distance between each sample point in the business data other than the multiple initial centroids and the initial centroid and determining the minimum distance between the sample point and the initial centroid are repeated until the new centroid in each cluster is determined to be unchanged, whereupon the traversal stops and the multiple clusters are obtained. Based on the preset time interval in the Multi-Bernoulli filter algorithm, the next target time after the current time is determined. Based on the preset Multi-Bernoulli filter algorithm, data is predicted for multiple clusters to determine business behavior information at the target time. If the business behavior information at the target time does not meet the preset business threshold, a prompt message is issued. Therefore, first, the clustering algorithm is used to perform multi-dimensional clustering of business data, clustering complex bank transaction data into multi-dimensional clusters. Second, the Multi-Bernoulli filter algorithm is used to predict cluster targets and determine changes in cluster data in real time, achieving high real-time and robust performance. Third, the clustering algorithm is combined with the Multi-Bernoulli filter algorithm to extract specified feature values ​​from bank transaction data and perform real-time clustering. Through the combination of these three points, banks can ultimately achieve high-quality clustering and locking of complex data, thereby effectively monitoring the market. This system has the advantages of high robustness and real-time performance in applications such as precision marketing, anti-money laundering, and anti-fraud, and solves the technical problem of the difficulty in determining business behavior information of a classifier at a preset time. In addition, the purpose of this application is to make full use of rich business data and the connections between business data to build data clusters, and predict the business behavior information of the clustering target at the target moment, so as to concretize the abstract and complex data, and make effective early warnings for sudden changes in customer portraits, providing banks with a solid data foundation and scientific support in market monitoring and prediction, precision marketing, and improving service experience.

[0142] Figure 6 A structural diagram of a data processing device provided in an embodiment of the present application is shown in FIG. Figure 6 As shown, the device includes:

[0143] The acquiring unit 31 is configured to acquire a prediction request, wherein the prediction request includes a user identifier and a business scenario identifier, and acquire business data corresponding to both the user identifier and the business scenario identifier.

[0144] The clustering unit 32 is configured to cluster the business data according to a preset clustering algorithm to obtain a plurality of clusters; wherein the preset clustering algorithm is configured to iteratively classify the business data, and each cluster includes a plurality of business data.

[0145] The prediction unit 33 is used to perform data prediction on multiple clusters based on a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine the business behavior information at the target time; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier.

[0146] The device of this embodiment can execute the technical solution in the above method. Its specific implementation process and technical principles are the same and will not be repeated here.

[0147] Figure 7 A structural diagram of another data processing device provided in an embodiment of the present application, Figure 6 Based on the embodiment shown, Figure 7 As shown, the clustering unit 32 includes:

[0148] The first determination module 321 is used to determine multiple initial centroids in the business data, wherein each business data is a sample point.

[0149] The generating module 322 is configured to generate a cluster corresponding to each of the multiple initial centroids; wherein each cluster corresponds to a cluster category.

[0150] The second determining module 323 is used to sequentially traverse the distance between each sample point and each initial centroid in the business data except for the multiple initial centroids, and determine the sample point and the initial centroid corresponding to the minimum distance.

[0151] The third determining module 324 is configured to determine the cluster category to which the sample point corresponding to the minimum distance belongs based on the cluster category corresponding to the cluster where each initial centroid is located, and classify the sample point into the cluster corresponding to the corresponding cluster category.

[0152] The fourth determination module 325 is configured to determine a new centroid in each cluster based on a preset formula.

[0153] The acquisition module 326 is used to perform a step of traversing the sample points in the business data except the multiple initial centroids based on the new centroid in each cluster, and determining the distance between each sample point and each initial centroid, and determining the sample point and the initial centroid corresponding to the minimum distance, until it is determined that the new centroid in each cluster remains unchanged, stopping the traversal and obtaining multiple clusters.

[0154] In one example, the prediction unit 33 includes:

[0155] The fifth determination module 331 is configured to determine the next target time after the current time according to a preset time interval in the multi-Bernoulli filter algorithm.

[0156] The prediction module 332 is used to perform data prediction on multiple clusters based on a preset multi-Bernoulli filtering algorithm to determine the business behavior information at the target moment.

[0157] In one example, the apparatus further includes:

[0158] The prompt unit 41 is configured to issue a prompt message if it is determined that the service behavior information at the target moment does not meet a preset service threshold.

[0159] The device of this embodiment can execute the technical solution in the above method. Its specific implementation process and technical principles are the same and will not be repeated here.

[0160] Figure 8 A schematic diagram of the structure of a server provided in an embodiment of the present application is shown in FIG. Figure 8 As shown, the server includes: a memory 51 and a processor 52.

[0161] The memory 51 stores computer programs that can be executed on the processor 52 .

[0162] The processor 52 is configured to execute the method provided in the above embodiments.

[0163] The server further includes a receiver 53 and a transmitter 54. The receiver 53 is used to receive instructions and data sent by an external device, and the transmitter 54 is used to send instructions and data to the external device.

[0164] An embodiment of the present application also provides a non-temporary computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of the server, enables the server to execute the method provided in the above embodiment.

[0165] An embodiment of the present application also provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of the server can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the server executes the solution provided by any of the above embodiments.

[0166] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0167] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A data processing method, characterized in that: include: Obtaining a prediction request, wherein the prediction request includes a user identifier and a business scenario identifier, and obtaining business data corresponding to both the user identifier and the business scenario identifier; Clustering the business data according to a preset clustering algorithm to obtain a plurality of clusters; wherein the preset clustering algorithm is used to iteratively classify the business data, and each cluster includes a plurality of business data; performing data prediction on the multiple clusters based on a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine business behavior information at a target time; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents the business behavior of the user corresponding to the user identifier; The performing data prediction on the plurality of clusters according to a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine the business behavior information at a target time includes: Determining the next target moment after the current moment according to a preset time interval in the multi-Bernoulli filter algorithm; According to a preset multi-Bernoulli filtering algorithm, data prediction is performed on the multiple clusters to determine the business behavior information at the target moment.

2. The method according to claim 1, characterized in that Clustering the business data according to a preset clustering algorithm to obtain multiple clusters includes: Determine a plurality of initial centroids in the business data, wherein each business data is a sample point; Generate a cluster corresponding to each of the multiple initial centroids; wherein each cluster corresponds to a cluster category; Sequentially traverse the sample points in the business data except the multiple initial centroids, respectively, and determine the distance between each of the sample points and the initial centroid, and determine the sample point and the initial centroid corresponding to the minimum distance; According to the cluster category corresponding to the cluster where each initial centroid is located, determine the cluster category to which the sample point corresponding to the minimum distance belongs, and classify the sample point into the cluster corresponding to the corresponding cluster category; Based on a preset formula, determine the new centroid of each cluster; Based on the new centroid in each cluster, the sample points in the business data except the multiple initial centroids are traversed in sequence, and the distance between each of the initial centroids is determined, and the sample points and the initial centroid corresponding to the minimum distance are determined. When it is determined that the new centroid in each cluster remains unchanged, the traversal is stopped and multiple clusters are obtained.

3. The method according to any one of claims 1-2, characterized in that The method further comprises: If it is determined that the business behavior information at the target moment does not meet the preset business threshold, a prompt message is issued.

4. A data processing device, characterized in that: include: an acquiring unit, configured to acquire a prediction request, wherein the prediction request includes a user identifier and a business scenario identifier, and acquire business data corresponding to both the user identifier and the business scenario identifier; a clustering unit, configured to cluster the business data according to a preset clustering algorithm to obtain a plurality of clusters; wherein the preset clustering algorithm is configured to iteratively classify the business data, and each cluster includes a plurality of business data; a prediction unit, configured to perform data prediction on the plurality of clusters based on a preset multi-Bernoulli filtering algorithm and a preset time interval in the multi-Bernoulli filtering algorithm to determine business behavior information at a target moment; wherein the preset multi-Bernoulli filtering algorithm is used to perform data prediction on historical data, and the business behavior information represents business behavior of a user corresponding to the user identifier; The prediction unit includes: a fifth determination module, configured to determine a next target time after the current time according to a preset time interval in the multi-Bernoulli filter algorithm; The prediction module is used to perform data prediction on the multiple clusters according to a preset multi-Bernoulli filtering algorithm to determine the business behavior information at the target moment.

5. The device according to claim 4, characterized in that The clustering unit comprises: A first determination module is configured to determine a plurality of initial centroids in the business data, wherein each business data is a sample point; A generating module, configured to generate a cluster corresponding to each of the plurality of initial centroids; wherein each cluster corresponds to a cluster category; A second determining module is configured to sequentially traverse the distance between each of the sample points in the business data except the multiple initial centroids and each of the initial centroids, and determine the sample point and the initial centroid corresponding to the minimum distance; A third determination module is used to determine the cluster category to which the sample point corresponding to the minimum distance belongs based on the cluster category corresponding to the cluster where each initial centroid is located, and classify the sample point into the cluster corresponding to the corresponding cluster category; a fourth determination module, configured to determine a new centroid in each cluster based on a preset formula; The acquisition module is used to perform a step of traversing the sample points in the business data except the multiple initial centroids based on the new centroid in each cluster, and determining the distance between each sample point and each initial centroid, and determining the sample point and the initial centroid corresponding to the minimum distance, until it is determined that the new centroid in each cluster remains unchanged, stopping the traversal and obtaining multiple clusters.

6. A server, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method according to any one of claims 1 to 3 is implemented.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 3 when executed by a processor.

8. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 3 when executed by a processor.

Images